doi
stringlengths 28
28
| title
stringlengths 19
311
| abstract
stringlengths 217
5.08k
| plain language summary
stringlengths 115
4.83k
| article
stringlengths 3.87k
161k
|
|---|---|---|---|---|
10.1371/journal.pntd.0003469 | Characterization of Plasmodium ovale curtisi and P. ovale wallikeri in Western Kenya Utilizing a Novel Species-specific Real-time PCR Assay | Plasmodium ovale is comprised of two genetically distinct subspecies, P. ovale curtisi and P. ovale wallikeri. Although P. ovale subspecies are similar based on morphology and geographical distribution, allelic differences indicate that P. ovale curtisi and P. ovale wallikeri are genetically divergent. Additionally, potential clinical and latency duration differences between P. ovale curtisi and P. ovale wallikeri demonstrate the need for investigation into the contribution of this neglected malaria parasite to the global malaria burden.
In order to detect all P. ovale subspecies simultaneously, we developed an inclusive P. ovale-specific real-time PCR assay based on conserved regions between P. ovale curtisi and P. ovale wallikeri in the reticulocyte binding protein 2 (rbp2) gene. Additionally, we characterized the P. ovale subspecies prevalence from 22 asymptomatic malaria infections using multilocus genotyping to discriminate P. ovale curtisi and P. ovale wallikeri.
Our P. ovale rbp2 qPCR assay validation experiments demonstrated a linear dynamic range from 6.25 rbp2 plasmid copies/microliter to 100,000 rbp2 plasmid copies/microliter and a limit of detection of 1.5 rbp2 plasmid copies/microliter. Specificity experiments showed the ability of the rbp2 qPCR assay to detect low-levels of P. ovale in the presence of additional malaria parasite species, including P. falciparum, P. vivax, and P. malariae. We identified P. ovale curtisi and P. ovale wallikeri in Western Kenya by DNA sequencing of the tryptophan-rich antigen gene, the small subunit ribosomal RNA gene, and the rbp2 gene.
Our novel P. ovale rbp2 qPCR assay detects P. ovale curtisi and P. ovale wallikeri simultaneously and can be utilized to characterize the prevalence, distribution, and burden of P. ovale in malaria endemic regions. Using multilocus genotyping, we also provided the first description of the prevalence of P. ovale curtisi and P. ovale wallikeri in Western Kenya, a region holoendemic for malaria transmission.
| Humans can be infected with five malaria parasite species: Plasmodium falciparum, P. vivax, P. malariae, P. knowlesi, and P. ovale. Although the vast majority of malaria morbidity and mortality worldwide can be attributed to P. falciparum, non-falciparum malaria parasites can also cause clinical disease. Researchers use nucleic acid based detection methods, such a polymerase chain reaction (PCR), to detect low-density malaria parasitemias that can evade microscopic detection. P. ovale was recently identified to exist as two subspecies, P. ovale curtisi and P. ovale wallikeri, that look identical but differ genetically. In this study, we developed a novel real-time PCR (qPCR) assay to detect all P. ovale parasites, based on a conserved gene between P. ovale curtisi and P. ovale wallikeri. We also used DNA sequencing to differentiate between P. ovale curtisi and P. ovale wallikeri from a small sample of P. ovale asymptomatic infections in Western Kenya. Through the use of our novel rbp2 qPCR assay, we aim to characterize the prevalence of P. ovale in future epidemiological studies in order to better understand this neglected malaria parasite species.
| Plasmodium ovale, the causative agent of benign tertian malaria, was identified as a distinct malaria parasite species in 1922 based on its characteristic oval morphology in infected erythrocytes [1]. P. ovale rarely causes severe disease in humans living in malaria endemic regions, but can cause serious clinical disease in naive travelers [2–9]. The actual prevalence and clinical relevance of P. ovale is likely underestimated for the following reasons. First, P. ovale is often found as a mixed infection with other malaria parasite species [10–12]. This can confound microscopic identification of P. ovale due to difficulties in differentiating P. ovale from other morphologically similar malaria parasites, such as P. vivax. Second, the characteristic low-level parasitemia of P. ovale infection further complicates microscopic detection due to the difficulty in finding and identifying low numbers of P. ovale parasites [13]. Finally, malaria Rapid Diagnostic Tests (RDTs) show a reduced ability to detect P. ovale compared to other human malaria parasites, resulting in false negative cases [14–16]. However, the use of extremely sensitive molecular detection methods, such as polymerase chain reaction (PCR), have revealed a higher prevalence of P. ovale and expanded the geographical distribution of this malaria parasite compared to what was previously identified based on microscopy [10, 17–20].
Recent findings demonstrated that P. ovale exists as two genetically distinct sympatric subspecies, P. ovale curtisi and P. ovale wallikeri[21–24]. Morphological differences between the two P. ovale subspecies have not been identified, thereby limiting the use of microscopy to differentiate P. ovale curtisi and P. ovale wallikeri. As recent studies suggest potential clinical and latency duration differences between the two P. ovale subspecies, [25, 26], a discriminatory assay to differentiate P. ovale curtisi and P. ovale wallikeri is clinically relevant. Additionally, initial P. ovale-specific assays developed by our group and others were unknowingly designed based on gene sequences specific to only one subspecies, thereby failing to detect the other P. ovale subspecies. PCR assays that target conserved genetic regions between the two subspecies are, therefore, necessary to determine the true P. ovale prevalence and distribution [27–30].
Small-subunit ribosomal RNA (ssrRNA) genes are common targets for malaria parasite species-specific assays based on nucleotide polymorphisms that facilitate specific detection of the species of interest [28, 29, 31]. Although rRNA based PCR assays have proven useful for the detection of low-level parasitemias of a single malaria parasite species, Demas et al. demonstrated that alternative gene targets may be more sensitive for species-specific detection in the context of mixed species infections [32]. A quality control program to determine the ability of 10 different laboratories to detect malaria parasite species based on rRNA PCR revealed detection of P. ovale to be the most difficult, with a detection rate of 70% [33]. Additionally, allelic diversity within the P. ovale ssrRNA alleles may further limit the ability of rRNA specific PCR assays to detect P. ovale infections [34]. Due to these difficulties in the detection of P. ovale, we designed a novel P. ovale-specific assay based on a gene found only in P. ovale curtisi and P. ovale wallikeri and not present in other human malaria parasite species. This approach reduces aberrant amplification of non-target malaria species and allows for the detection of low-level P. ovale infections in the presence of high parasitemias of other malaria parasite species, such as P. falciparum.
Several epidemiology surveys of exant malaria species have established the endemicity of P. ovale in Western Kenya based on microscopic identification, entomological studies, and nucleic acid detection methods [13, 35–38]. Clinical cases due to P. ovale relapse in non-immune individuals after traveling to Western Kenya have also been reported, including a single case of a returned traveller with P. ovale curtisi infection [25, 39]. However, the lack of data on the prevalence and distribution of P. ovale curtisi and P. ovale wallikeri in Western Kenya represents a critical gap in our understanding of the true malaria epidemiology in this region that could impact both patient treatment and malaria control strategies.
In this study, we developed a novel, highly specific, real-time PCR (qPCR) assay to detect all P. ovale subspecies simultaneously based on a conserved region of the P. ovale-specific reticulocyte binding protein 2 (rbp2)gene. This inclusive P. ovale rbp2 qPCR assay was characterized and validated to determine the sensitivity, limit of detection, limit of quantification, specificity, repeatability, and reproducibility. In addition, the occurrence of both P. ovale subspecies (P. ovale curtisi and P. ovale wallikeri) was documented in Western Kenya using multilocus genotyping. Our P. ovale species-specific assay can be utilized to better characterize the presence, parasitemia, geographical distribution, and the contribution of this malaria parasite species to mixed species infections and to clinical disease in malaria endemic regions.
Anonymized human whole blood samples were collected with signed informed consent under approved protocols (Walter Reed Army Institute of Research Human Use and Review Committee Protocols #1720 and 1306, Kenya Medical Research Institute (KEMRI) SSC#2008 and 1111). Clinically healthy (asymptomatic) adult individuals in Nyanza Province, Kenya were screened (active detection) with the Parascreen Pan/Pf ® malaria Rapid Diagnostic Test (Zephyr Biomedicals, Verna, Goa, India) for the presence/absence of malaria parasites from March through September of 2008. Thin and thick smears were examined subsequently by up to 5 expert microscopists in the Malaria Diagnostic Centre (MDC), Kisumu, Kenya for malaria species designation and estimation of quantitative parasitemia [40]. Samples identified as positive for P. ovale (n = 22) via microscopy, in which all were mixed infections with other malaria species, were targeted for DNA extraction and PCR based analysis. DNA was extracted from 200 microliters of whole blood using the QIAamp DNA Minikit (Qiagen, Venlo, Netherlands) following the manufacturer’s protocol. DNA was eluted in 200 microliters of Buffer EB and samples were stored at −20°C until time of use. A human-specific RNaseP based qPCR assay was performed for each sample in duplicate to confirm successful nucleic acid extraction [41].
Tryptophan-rich antigen (tra) gene. The P. ovale-specific tryptophan-rich antigen (tra) gene was recently identified as a target to discriminate between P. ovale subspecies based on DNA sequence length and single nucleotide polymorphisms (SNPs) [22, 23, 30]. We utilized the PoTRA fwd3 and PoTRA rev3 primers reported in Oguike et al. 2011 for PCR analysis [23]. Primers (Table 1) were synthesized by Integrated DNA Technologies (IDT, Coralville, IA, USA) and purified by standard desalting methods. Each PCR assay consisted of 1X Sigma JumpStart REDTaq ReadyMix (20 mM Tris-HCl, 100 mM KCl, 4 mM MgCl2, 0.4 mM of each dNTP, 0.03 unit/μl of Taq DNA polymerase, Sigma, Balcatta, WA, USA), 8.75 picomoles of each primer, and one microliter of template with a final volume of 25 microliters. PCR cycling conditions were: initial denaturation for 2 minutes at 95°C followed by 45 cycles of 95°C for 30 seconds, 58°C for 45 seconds, 72°C for 1 min and a final extension at 75°C for 5 minutes. All conventional PCRs were performed on a DNA Engine PTC-200 Thermal Cycler (MJ Research, Waltham, MA, USA).
Reticulocyte binding protein 2 (rbp2) gene. The reticulocyte binding protein 2 (rbp2) gene was utilized by Oguike et al. 2011 to differentiate between P. ovale subspecies using qPCR melt curve profiles based on six SNPs present within a 120 base pair fragment. We designed a novel set of primers (Table 1, IDT) using Primer Express software (Life Technologies, version 3.0; Frederick, MD, USA) to amplify a smaller, 74 base pair region of the rbp2 gene for assay development. Our primers (PoRBP2f and PoRBP2r) are located within conserved DNA sequences of the P. ovale subspecies to ensure detection and amplification of both P. ovale subspecies. The amplicon also contains a single SNP to distinguish P. ovale subspecies by DNA sequencing. Fig. 1 shows the single SNP in the rbp2 amplicon at position 431, in which P. ovale curtisi contains an adenine and P. ovale wallikeri contains a thymine. Primer BLAST was utilized to ensure our primers were specific for P. ovale and would not amplify non- P. ovale malaria parasite DNA or human DNA. PCRs consisted of 1X Sigma JumpStart REDTaq ReadyMix Reaction Mix, 25 picomoles of each primer, and one microliter of template, with a final volume of 25 microliters. PCR cycling conditions were as follows: initial denaturation at 95°C for 2 minutes followed by 40 cycles of 95°C for 30 seconds, 55°C for 30 seconds, 72°C for 30 seconds, and a final extension at 72°C for 10 minutes.
Small subunit ribosomal RNA (ssrRNA) gene. We utilized P. ovale-specific primers (Table 1, IDT) reported by Fuehrer et al. 2012 (rOVA1WC and rOVA2WC) to further characterize P. ovale positive samples based on differences within the small subunit ribosomal RNA (ssrRNA) gene [29]. PCRs consisted of 1X Sigma JumpStart REDTaq ReadyMix Reaction Mix, 25 picomoles of each primer, one microliter of template, and a final volume of 25 microliters. PCR cycling conditions were as follows: initial denaturation at 95°C for 4 minutes followed by 35 cycles of 94°C for 1 minute, 58°C for 2 minutes, 72°C for 2 minutes, and a final extension at 72°C for 5 minutes.
DNA sequencing. PCR products were visualized on 0.7% agarose gels stained with ethidium bromide. PCR products were cloned into the pCR 2.1-TOPO TA vector (Life Technologies) based on manufacturer’s guidelines. Plasmid purification was performed using the QIAprep Spin Miniprep kit (Qiagen) and used as template for sequencing reactions. PCR products were sequenced using the M13 Forward (−20) Primer (Life Technologies) at the Biomedical Instrumentation Center at the Uniformed Services University or GENEWIZ Inc (Germantown, MD, USA) using the ABI 3500XL Genetic Analyzer and the ABI 3730XL DNA Analyzer, respectively. Sequencing facility was chosen based on temporal availability. DNA sequences were aligned and analyzed with previously published sequences using SeqMan software (DNAStar Lasergene Version 8.1.5, Madison, WI, USA). Reference sequences utilized for DNA alignments are shown in Table 2.
Primer Express software (Life Technologies, version 3.0) was utilized to design a hydrolysis probe (Table 1) for use with our rbp2 primers on the ABI 7500 fast real-time PCR (qPCR) platform (Life Technologies). An alignment of the P. ovale rbp2 DNA sequences was constructed using the Clustal Omega Program provided by the European Molecular Biology Laboratory—European Bioinformatics Institute (EMBL-EBI) [42, 43]. We utilized the Jalview output tool to visualize the DNA sequence alignment (Fig. 1)[44]. Primers and probe were designed in order to amplify a conserved region within the rbp2 gene to ensure detection of both P. ovale subspecies by our qPCR assay at the same time. In silico analyses were performed to ensure primers and probe were specific to P. ovale and would not amplify genes of other malaria parasites or human DNA. Each qPCR reaction consisted of the following: 1X TaqMan Fast Universal PCR Master Mix, No AmpErase UNG (Life Technologies, Cat No. 4364103), 5 picomoles of each primer and probe, and one microliter of template in a final volume of 20 microliters. Real-time PCR was performed utilizing fast thermal cycling conditions (95°C for 20 seconds, followed by 40–60 cycles of 95°C for 3 seconds and 60°C for 30 seconds). Analysis of qPCR results was performed using ABI 7500 Fast Real-Time PCR Systems Software (Life Technologies, Version 2.0.5). Basic statistical analyses (means, standard deviations, coefficient of variation), generation of standard curve graphs, calculation of slopes, and coefficient of correlation were performed in Microsoft Excel or GraphPad Prism (GraphPad Prism Software Version 6, La Jolla, CA, USA).
Plasmid standard curve. We cloned the 74 base pair rbp2 amplicon into the pCR 2.1-TOPO TA vector (Life Technologies) following manufacturer’s guidelines and eluted the rbp2 plasmid in PCR grade water. The approximate rbp2 amplicon copy number per microliter was determined based on spectrophotometer (Nanodrop 2000c) concentration in nanograms per microliter. Plasmids with the rbp2 amplicon (rbp2 plasmid) were diluted in water to generate a ten-fold serial dilution from 100,000 rbp2 copies per microliter to 0.1 rbp2 copies per microliter. The resulting non-linearized ten-fold serial dilution series was utilized as a standard curve in subsequent validation experiments including determination of the linear dynamic range, specificity, reproducibility, repeatability, and limit of detection. The effect of the conformation of the rbp2 plasmid on standard curve linearity was analyzed by linearizing the rbp2 plasmid using the NotI restriction enzyme (New England BioLabs Inc, Ipswich, MA, USA) according to the manufacturer’s protocol. Rbp2 plasmid linearization was confirmed by gel electrophoresis on a 0.7% agarose gel stained with ethidium bromide. Linearized rbp2 plasmid was purified using the Qiagen PCR Purification Kit following the manufacturer’s protocol. The approximate rbp2 copy number per microliter of the linearized rbp2 plasmid was determined and diluted in water to generate a ten-fold serial dilution (100,000 to 0.1 copies per microliter). The rbp2 standard curve PCR efficiency and coefficient of correlation (R2) were determined and the Pearson product-moment correlation was used to compare the linearized and non-linearized rbp2 plasmid standard curves (GraphPad Prism).
Validation experiments. Real-time PCR efficiency was determined using a standard curve of 10-fold serial dilutions of the non-linearized rbp2 plasmid. Efficiency (E) was calculated using the following formula: E = 10(− 1/ slope) −1. Rbp2 plasmid standard curve samples were run at least in duplicate and the mean quantification cycle (Cq) value was utilized to generate the standard curve. The limit of detection was defined as the concentration of rbp2 plasmid in copies per microliter that gave a positive signal in at least one replicate well in two separate qPCR experiments. Limit of quantification was defined as the range of rbp2 plasmid concentrations that maintained linearity and therefore could be used to quantify P. ovale concentration from test samples.
Specificity was analyzed using DNA template from non- P. ovale malaria parasite species and uninfected human DNA. Genomic DNAs from P. falciparum strains 3D7 (WRAIR), FCR3CSA (ATCC/BEI Resources, MR4, Manassas, Virginia), Dd2 (ATCC/BEI Resources, MR4), and NF54 (ATCC/BEI Resources, MR4) were utilized as template to assess specificity. P. vivax genomic DNA was extracted from frozen whole parasites (kind gift of Dr. J. Prachumsri, Mahidol University, Bangkok, Thailand). Since pure P. malariae positive samples were unavailable, we utilized three samples collected as part of the blood collection protocol in Kenya that were positive for P. malariae as well as P. falciparum by microscopy and PCR, but were negative for P. ovale. The P. malariae parasitemias ranged from approximately 30 to 2400 parasites per microliter. Additionally, genomic DNAs from P. knowlesi, P. simiovale, P. fragile, and P. cynomolgi (ATCC/BEI Resources, MR4), were also utilized as templates. Specificity was further analyzed by performing spiking experiments in which a known concentration of rbp2 plasmid was added to template containing P. falciparum 3D7 DNA (10,000 parasites per microliter) or P. vivax DNA (517 parasites per microliter). One-way analysis of variance (ANOVA) was used to determine differences in Cq values for spiking experiments (GraphPad Prism).
Within-run repeatability was defined as the variation of Cq values within a single run and was analyzed by calculating the percent coefficient of variation (%CV) of Cq values in replicate wells. Between-run repeatability was defined as the variation of Cq values in separate qPCR runs and was determined by calculating the percent coefficient of variation (%CV) of mean Cq values based on six separate qPCR experiments. Reproducibility was evaluated by comparing the assay performance by a technician at the USAMRU-K laboratory in Kisumu, Kenya and the Uniformed Services University in Bethesda, Maryland, USA.
Quantification comparison: Microscopy versus rbp2 qPCR. Parasitemias were determined for P. ovale positive blood films based using standard microscopic methods at the Malaria Diagnostic Centre, affiliated with both USAMRU-K and KEMRI, in Kisumu, Kenya. DNA was extracted from microscopy-positive P. ovale samples and tested using the P. ovale-specific rbp2 qPCR assay. Approximate rbp2 copy number per microliter was determined based on the rbp2 plasmid standard curve. Parasitemias as determined by expert microscopy (parasites per microliter) were compared to rbp2 copy number per microliter as determined by the P. ovale-specific qPCR in order to examine potential correlation between rbp2 plasmid copy number and microscopic parasitemias.
Human-specific RNaseP qPCR. A previously described qPCR assay based on the human-specific RNaseP gene was performed to confirm the presence of nucleic after DNA extraction [41]. The human RNaseP gene was detected from all 22 samples (Average Cq = 29.12, Cq Range = 28.2–32.87, standard deviation = 1.02), indicating extraction methods yielded DNA suitable for subsequent PCR experiments.
Tryptophan-rich antigen (tra) gene. Alignments of tra gene sequences revealed nine samples (40.9%) positive for P. ovale curtisi type 1, two samples (9.1%) positive for P. ovale curtisi type 2, six samples (27.3%) positive for P. ovale wallikeri type 1, and three samples (13.6%) positive for P. ovale wallikeri type 2 (Table 3). Previously published GenBank accession numbers were utilized as reference sequences for alignment and are shown in Table 2. Representative P. ovale curtisi type 1, P. ovale curtisi type 2, P. ovale wallikeri type 1, and P. ovale wallikeri type 2 tra DNA sequences were deposited under GenBank accession numbers KM494978-KM494981, respectively, and are identical to the reference sequences. As shown in Table 4, unique polymorphisms within the tra gene were also detected and confirmed by at least two separate sequencing reactions for 5 samples: Po05, Po12, Po20, Po06, and Po07 (Accession numbers KM494982-KM494986, respectively). Samples Po12 and Po20 contained an 18 base pair insertion between nucleotide positions 171 and 172 (based on P. ovale wallikeri type 1 HM594180 reference sequence), which represents a short sequence repeated throughout the tra gene. Two samples, Po9 and Po18, failed to amplify with the tra primers despite multiple PCR attempts.
Reticulocyte binding protein 2 (rbp2) gene. DNA sequences of the rbp2 gene were obtained for all 22 P. ovale samples (Table 3). S1 Table contains the 74 pair rbp2 amplicon for both P. ovale curtisi and P. ovale wallikeri. These sequences were not eligible for submission as the minimum length requirement for GenBank is 200 nucleotides. P. ovale subspecies results based on rbp2 gene sequences agreed with subspecies results based on the tra gene sequences. Thirteen (59%) of the P. ovale samples were positive for P. ovale curtisi and 9 (41%) were positive for P. ovale wallikeri. None of our samples failed to amplify with the rbp2 primers.
Small subunit rRNA (ssrRNA) gene. Nineteen of the 22 P. ovale positive samples were detected by the ssrRNA gene assay (Table 3). P. ovale curtisi and P. ovale wallikeri ssrRNA sequences were approximately 99% identical to previously published sequences at this locus. Representative P. ovale curtisi and P. ovale wallikeri ssrRNA sequences were deposited in GenBank as KM494987 and KM494988, respectively. P. ovale subspecies results based ssrRNA gene sequences agreed with subspecies results based on tra and rbp2 gene sequences. Three samples, Po9, Po11, and Po18, failed to amplify using the ssrRNA primers despite a second attempt using an additional microliter of template DNA.
Plasmid standard curve analysis of rbp2 qPCR assay. Since all 22 P. ovale microscopy positive samples were successfully amplified and sequenced using the rbp2 primers, we developed an rbp2 based qPCR assay to detect all P. ovale subspecies simultaneously in a single assay. Efficiency of the rbp2 qPCR assay was analyzed using the non-linearized rbp2 plasmid 10-fold serial dilution standard curve. Efficiency ranged from 90%–99% for six consecutive qPCR experiments with a coefficient of correlation (R2) greater than 0.99. A representative qPCR amplification plot and standard curve are shown in Fig. 2 and 3, respectively. All 22 P. ovale samples identified as P. ovale positive by expert microscopy were detected using our rbp2 qPCR assay. There was no difference in PCR efficiency or R2 value based on the conformation (linearized vs. non-linearized) of the rbp2 plasmid standard curve (Pearson product-moment correlation = 0.998, P<0.001).
Limit of quantification and limit of detection. The linear dynamic range of the rbp2 qPCR assay was determined to be between 6.25 copies per microliter and 100,000 copies per microliter based on serial dilutions of the rbp2 plasmid. Two-fold serial dilutions of known concentrations of the rbp2 plasmid were performed in at least duplicate to determine the limit of detection. Dilutions containing 1.5 copies per microliter of the rbp2 plasmid were detected by at least one replicate well in two separate qPCR experiments.
Specificity. In order to test the specificity of our rbp2 assay for P. ovale, we performed qPCR using DNA isolated from cultured P. falciparum 3D7 (10,000 parasites per microliter) and P. vivax DNA (517 parasites per microliter). Based on a series of ten separate qPCR experiments, DNA from P. falciparum and P. vivax were uniformly negative. To ensure no background from other P. falciparum strains, we tested genomic DNAs from strains Dd2, NF54, and FCR3CSA, which were also not detected by our assay. We tested DNA from P. knowlesi, P. fragile, and P. cynomolgi and found DNA from these malaria parasite species were undetectable by our rbp2 qPCR assay. As we were unable to obtain pure P. malariae samples, we examined DNA samples isolated from the blood of individuals co-infected with both P. malariae and P. falciparum. These P. falciparum and P. malariae co-infected samples were also negative, indicating that our rbp2 qPCR assay does not detect P. malariae DNA. Two different control DNA samples from malaria uninfected human blood were also uniformly negative. All specificity experiments were carried out to 60 cycles in an attempt to capture non-specific amplification, which was never seen, although the standard curve and the P. ovale-containing field samples amplified appropriately.
Spiking experiments, in which P. falciparum DNA or P. vivax DNA was added to the rbp2 plasmid standard curve samples and subsequently utilized as template for the rbp2 qPCR did not significantly alter the Cq values compared to when the standard curve plasmid samples were run alone (ANOVA, P = 0.9993, Fig. 4).
Interestingly, our rbp2 qPCR assay detected P. simiovale genomic DNA isolated from filter paper. DNA sequencing utilizing the rbp2 primers revealed that the 74 base pair rbp2 region in P. simiovale is identical to that in P. ovale curtisi. Subsequent attempts using additional primers to sequence the full-length rbp2 gene of P. simiovale were not successful. As these additional primers successfully amplified P. ovale positive samples, the inability to amplify the full-length P. simiovale rbp2 gene is likely due to sequence polymorphisms between P. ovale and P. simiovale in the primer binding regions.
Repeatability. Within-run repeatability of the rbp2 plasmid standard curve Cq values was high, with the percent coefficient of variation (%CV) of dilution series replicates between 0.00–2.23% (Table 5). Results were also repeatable between runs, with the percent coefficient of variation (%CV) between 1.17–3.43% (Table 5). Repeatability was determined using results from six separate consecutive qPCR experiments.
Reproducibility. Analysis of the efficiency of the rbp2 assay was performed independently at the USAMRU-K laboratory. A known concentration of non-linearized rbp2 plasmid was diluted in PCR grade water to generate a 10-fold dilution standard curve for PCR efficiency analysis. The assay was performed with the same P. ovale-specific primers and probe utilized in validation experiments in a final volume of 50 microliters of Life Technologies TaqMan Fast Master Mix for the USAMRU-K ABI 7500. Despite slight variations in qPCR set up and cycling conditions, the Kenya laboratory obtained a PCR efficiency of 93.6% with an R2 >0.99 for the standard curve analysis. These results are identical to the PCR efficiencies and R2 values obtained at USU. The USAMRU-K laboratory also performed specificity experiments and demonstrated no amplification from P. falciparum DNA, DNA from uninfected human blood, or from negative template controls.
Quantification comparison: Microscopy versus rbp2 qPCR. Quantitative parasitemia determined by expert microscopy (parasites per microliter) was compared to the rbp2 copy number per microliter based on the rbp2 plasmid standard curve (Fig. 5). A modest correlation was determined (R2 = 0.6595). This lack of a strong correlation is not surprising, as all P. ovale parasitemias were low, ranging from 16–3800 parasites/μl, and such low-level parasitemias are notoriously difficult to quantify accurately by microscopy [40, 45–47]. Additionally, the samples utilized for comparison were mixed malaria species infections, mainly with P. falciparum. Mixed species infections create further difficulties for the accurate quantification of P. ovale-specific parasitemia based on light microscopy, but single-species P. ovale infected samples were not available.
Based on multilocus genotyping using the rbp2, ssrRNA, and tra genes, we detected both P. ovale curtisi and P. ovale wallikeri in approximately equal frequencies in a small sample set from Western Kenya, a region in which P. ovale subspecies characterization had not been previously performed. The presence of both P. ovale subspecies in Western Kenya is in agreement with other studies in sub-Saharan Africa and P. ovale endemic regions that describe the sympatric distribution of P. ovale curtisi and P. ovale wallikeri [23, 27, 48]. We also identified additional allelic diversity within the tra gene in P. ovale samples from Kenya (Table 4) compared to what was previously identified in P. ovale samples from other malaria endemic regions [23]. This allelic diversity at the P. ovale tra gene is consistent with reports of other tra variants identified by DNA sequencing, however our tra sequences are unique from previously published tra gene sequences [30].
Our new inclusive P. ovale-specific qPCR assay is based on rbp2, a gene that contains conserved regions between P. ovale curtisi and P. ovale wallikeri but that is absent from other human malaria parasite species. The rbp2 qPCR assay described herein allows simultaneous detection of both P. ovale subspecies using a single set of primers and probe. All 22 samples were detected and sequenced using our rbp2 primers, highlighting the utility of these primers for P. ovale identification. P. ovale subspecies differentiation by DNA sequencing of the 74 base pair rbp2 sequence region was in absolute agreement with tra and ssrRNA DNA sequencing results. This again emphasizes the utility of the PoRBP2fwd1 and PoRBP2rev1 primers for P. ovale subspecies discrimination based on a single SNP at position 431 (Fig. 1) located between these primers. In agreement with other previous studies, these data demonstrate perfect dimorphism between P. ovale curtisi and P. ovale wallikeri, providing further support for the separation of the two P. ovale subspecies [21–24, 48–50]. As we begin to understand potential clinical, pathological, and biological differences between the two P. ovale subspecies, molecular methods to distinguish P. ovale curtisi and P. ovale wallikeri will aid in these research efforts. Additionally, as genomic data and full genome sequences become available for P. ovale curtisi and P. ovale wallikeri, phylogenetic analyses to determine the evolutionary relatedness between these and other malaria species will likely further our understanding of these newly characterized but poorly understood human parasites.
Using the rbp2 plasmid as a standard curve, the linear dynamic range of our assay was determined to be between 6.25 copies of rbp2 per microliter to 100,000 copies of rbp2 per microliter. The lower, non-linear but still clearly positive limit of detection of our assay was determined to be 1.5 copies of rbp2 per microliter, confirming this assay’s capacity to detect low-level parasitemias. P. ovale parasitemias are characteristically lower than other malaria species, so we limited the testing of our upper dynamic range to 100,000 rbp2 copies per microliter, as higher copy numbers would likely be epidemiologically and clinically irrelevant. We used the rbp2 plasmid to determine the linear dynamic range and limit of detection because of difficulties obtaining pure P. ovale infected samples from malaria endemic regions and the inability to culture P. ovale parasites. The paucity of published genomic information for P. ovale also hinders the determination of copy number of P. ovale-specific genes, such as the rbp2, tra, and ssrRNA genes, utilized in this study. Thus, we are further limited in our attempts to appropriately correlate rbp2 copy number and P. ovale parasitemias. Despite these limitations, we demonstrate the utility of our P. ovale-specific assay to detect low-levels of the rbp2 plasmid and to detect low P. ovale parasitemias (as low as 16 parasites per microliter) from human blood samples collected in Western Kenya. Our study was also limited by only testing samples collected in Western Kenya and additional validation is therefore needed to confirm the ability of the rbp2 qPCR assay to detect total P. ovale from other malaria endemic regions. As the 22 samples included in this study were identified as P. ovale by microscopy, further studies are needed to test the P. ovale rbp2 qPCR assay with submicroscopic and asymptomatic P. ovale infections with a range of parasitemias.
Repeatability and reproducibility of qPCR assays are important components of assay validation as they indicate the assay’s capacity to provide consistent and reliable results in different environments. Different users under modified laboratory conditions performed this assay successfully, with high PCR efficiency and equivalent quantification.
Specificity experiments showed no cross reactivity of our assay with P. falciparum, P. vivax, P. malariae, P. cynomolgi, P. knowlesi, P. fragile, and DNA from uninfected human blood even when qPCR was performed for 60 cycles. The complete lack of background amplification from human and other malaria parasite DNA, verifies the exquisite specificity of the assay. Further, assay performance was unchanged in the presence of DNA from other malaria parasite species. This is of particular importance for P. ovale, as this malaria species is often found as a co-infection with other malaria species. Interestingly, our rbp2 qPCR assay also detected DNA obtained from P. simiovale. As P. simiovale rbp2 sequence information is not available, we attempted to amplify the full-length P. simiovale rbp2 gene using additional primers based on the P. ovale rbp2 gene. However, we were unable to amplify the full P. simiovale rbp2 gene, suggesting the P. ovale and P. simiovale rbp2 genes may be similar but not identical. These results warrant further investigation of the P. simiovale rbp2 and additional specificity experiments of other P.ovale assays that may also unknowingly detect P. simiovale.
Of the 22 samples identified as P. ovale positive by expert microscopy, three samples (Po9, Po11, Po18) failed to amplify at two of the three loci tested despite multiple attempts (Table 3). However, the rbp2 gene was successfully amplified for all 22 samples as was a human-specific RNaseP endogenous control. These data, along with the parasitemia data from multiple expert microscopists, indicate that the 22 samples were P. ovale positive and that DNA template quality was unlikely to be the cause of the failed amplifications at the tra and ssrRNA loci. The inability to successfully amplify at all three loci could be explained by several reasons including: sequence polymorphisms, template degradation, low P. ovale density, and inter-laboratory variability due to reagents, equipment, and personnel. Additional investigation into potential reasons for the failure to amplify at all loci was prevented due to limited sample volume.
The limited correlation between microscopy and rbp2 qPCR results (Fig. 5) is not surprising as parasitemia calculations for P. ovale human samples at low parasitemias are notoriously difficult, particularly in co-infected samples [45]. Our P. ovale positive samples from Western Kenya are all co-infected with either P. falciparum or P. malariae, thus likely complicating the microscopy quantitation further. Variation between parasitemia and rbp2 copy number could also be explained by the P. ovale parasite stage. For example, a P. ovale ring stage counts as a single parasite by microscopy and DNA extracted from a P. ovale ring stage parasite represents one genome. However, a P. ovale schizont is counted as a single parasite by microscopy but DNA extracted from a P. ovale schizont may contain up to 14 genomes. This is a limitation of our study, as any relationship between P. ovale parasitemia and rbp2 copy number based on qPCR would depend on the parasite stages observed under the microscope and present in the blood sample obtained for DNA extraction.
Utilizing a plasmid standard curve for qPCR assays provides an efficient method for standardizing assays that does not require culturing organisms or using human samples. However, recent studies have highlighted important concerns regarding the plasmid template conformation that could lead to quantification bias of plasmid template by qPCR [51, 52]. After linearizing our template plasmid to compare with a non-linearized plasmid standard curve, we found no difference in Cq value, R2, slope, or PCR efficiency with the rbp2 qPCR assay. This is in agreement with another recent study, which also found no difference in plasmid standard curve based on the plasmid confirmation (linearized versus non-linearized) [53]. These results indicate that the effect of plasmid conformation on standard curve quantification may be assay specific. In addition to plasmid conformation, several additional quality control factors were optimized, including plasmid isolation methods, purification, storage, and developing appropriate laboratory protocols to minimize freeze-thawing, handling, and contamination.
Conventional PCR assays targeting the multi-copy small subunit ssrRNA genes are sensitive methods to detect and differentiate malaria species [54]. Initial P. ovale-specific ssrRNA PCR protocols showed limited capability to detect both P. ovale subspecies and have since been adapted to target conserved regions between the two subspecies. [29, 55, 56]. Although ssrRNA conventional PCR protocols have shown high sensitivity and specificity for malaria detection, we aimed to develop a novel P. ovale-specific assay based on a gene target that is found only in P. ovale and is absent from other malaria species infecting humans. We believe this approach enhances the specificity of our P. ovale-specific assay and eliminates the potential for nonspecific amplification of non- P. ovale species. Additionally, allelic variation within the ssrRNA genes of P. ovale curtisi and P. ovale wallikeri may limit the ability of ssrRNA based assays to capture all P. ovale infections due to sequence polymorphisms [34]. We found no allelic variation in the primer and probe-binding regions of the rbp2 gene from 22 P. ovale positive samples, indicating the potential utility of rbp2 for P. ovale subspecies detection.
While nested PCR is often utilized to enhance sensitivity for malaria PCR detection, we chose a single step qPCR protocol, as a nested PCR approach requires additional labor and cost to perform the second PCR. Nested PCR also increases the risk of laboratory contamination of PCR product and requires separate laboratory space to minimize the risk of contamination. Our P. ovale-specific qPCR assay maintains high sensitivity while also minimizing the additional cost, labor, designated laboratory space, and potential PCR product contamination that can be associated with nested PCR protocols.
Our P. ovale-specific qPCR assay provides several advantages for our future epidemiological studies of this neglected, and clinically relevant, malaria parasite species. First, fast qPCR conditions allow for a reaction to be completed in less than 1 hour. Second, the qPCR platform bypasses the need for gel electrophoresis, reducing the risk of amplicon contamination of the laboratory. Third, the use of a hydrolysis probe increases specificity compared to double stand DNA (dsDNA) based qPCR product detection. Our P. ovale-specific rbp2 qPCR assay can be utilized to better characterize the presence, parasitemia, geographical distribution, and the contribution of P. ovale to mixed-species infections and to clinical disease in malaria endemic regions.
|
10.1371/journal.pgen.1003722 | A Role for CF1A 3′ End Processing Complex in Promoter-Associated Transcription | The Cleavage Factor 1A (CF1A) complex, which is required for the termination of transcription in budding yeast, occupies the 3′ end of transcriptionally active genes. We recently demonstrated that CF1A subunits also crosslink to the 5′ end of genes during transcription. The presence of CF1A complex at the promoter suggested its possible involvement in the initiation/reinitiation of transcription. To check this possibility, we performed transcription run-on assay, RNAP II-density ChIP and strand-specific RT-PCR analysis in a mutant of CF1A subunit Clp1. As expected, RNAP II read through the termination signal in the temperature-sensitive mutant of clp1 at elevated temperature. The transcription readthrough phenotype was accompanied by a decrease in the density of RNAP II in the vicinity of the promoter region. With the exception of TFIIB and TFIIF, the recruitment of the general transcription factors onto the promoter, however, remained unaffected in the clp1 mutant. These results suggest that the CF1A complex affects the recruitment of RNAP II onto the promoter for reinitiation of transcription. Simultaneously, an increase in synthesis of promoter-initiated divergent antisense transcript was observed in the clp1 mutant, thereby implicating CF1A complex in providing directionality to the promoter-bound polymerase. Chromosome Conformation Capture (3C) analysis revealed a physical interaction of the promoter and terminator regions of a gene in the presence of a functional CF1A complex. Gene looping was completely abolished in the clp1 mutant. On the basis of these results, we propose that the CF1A-dependent recruitment of RNAP II onto the promoter for reinitiation and the regulation of directionality of promoter-associated transcription are accomplished through gene looping.
| The termination of transcription requires two major multisubunit complexes in budding yeast. These termination complexes are localized at the 3′ end of genes. Recent studies have found the termination factors occupying the 5′ end of genes as well. In this study, we investigate the physiological role of a termination factor at the 5′ end of a gene. Our results show that the CF1 termination complex affects the recruitment of the transcription enzyme RNAP II onto the promoter for reinitiation of transcription. The complex also affects the directionality of transcription of the promoter-bound polymerase. We also found that the looped gene conformation was disrupted in the absence of a functional termination complex. The overall conclusion of these results is that the terminator-bound factors contact the 5′ end of genes due to gene looping, and affect both the recruitment of the polymerase at the promoter for reinitiation, and directionality of the promoter-initiated transcription. Thus, the role of termination factors is not restricted to the 3′ end of the gene, but they are also involved in promoter-associated transcription.
| The process of transcription can be divided into three principal steps; initiation, elongation and termination [1]. The accomplishment of each of these steps during the RNAP II-mediated transcription cycle requires a number of accessory factors. The initiation of transcription requires gene specific transcription factors as well as general transcription factors (GTFs): TFIID, TFIIB, TFIIA, TFIIF, TFIIE, TFIIH and Mediator complex, that assemble on the promoter to form the preinitiation complex [2]–[5]. The termination of transcription, which is intimately linked to the cleavage and polyadenylation of precursor mRNA, exhibits a similar requirement for a group of termination factors organized into two macromolecular complexes called Cleavage-Polyadenylation-Factor (CPF) complex and Cleavage Factor-1 (CF1) complex in yeast [6]–[10]. The initiation and termination factors have been remarkably conserved during evolution. The generally accepted view is that the initiation factors operate exclusively at the 5′ end of a gene and are committed to starting the transcription cycle, while termination factors have a dedicated role in ending the transcription cycle at the 3′ end of a gene. A number of recently published reports, however, challenge this dogma. It is evident that at least some initiation factors are also necessary for termination, and the termination factors likewise may have a role in the initiation or reinitiation step of the transcription cycle [10]–[14].
An increasing amount of biochemical, genetic and functional evidence suggest the existence of a network of complex interactions between initiation and termination factors. The general transcription factor TFIIB, for example, exhibits multiple genetic and physical interactions with the factors operating at the 3′ end of genes [15]–[18]. These studies suggested a plausible role for TFIIB in the termination process. Accordingly, it was recently demonstrated that TFIIB is indeed actively engaged in termination of transcription in mammals and flies [17], [19]. Yeast Mediator subunit Srb5, which has a well-established function in the initiation of transcription, likewise, crosslinks to the 3′ end of genes and participates in the termination process [20]. TFIID is another promoter-bound protein that contacts the factors operating at the 3′ end of genes. Biochemical analysis of mammalian TFIID has revealed its reciprocal interaction with the CPSF 3′ end processing complex [21]. The TFIID-CPSF interaction is evolutionarily conserved. A recent proteomic analysis of yeast TFIID complex identified multiple interactions of TFIID subunit TAF150 with the components of the CPF 3′ end processing complex, which is the yeast homologue of CPSF complex [22]–[24].
Like initiation factors, an array of termination factors also crosstalk with the 5′ end of genes. The foremost among them is Ssu72, which was discovered as a protein of unknown function that genetically interacts with TFIIB [15]. Later on, yeast proteomic analysis identified Ssu72 as a component of the CPF 3′ end processing complex [25]–[27]. Ssu72 crosslinks to the 5′ end of genes, and interacts with several promoter-bound factors [22], . Pta1, which is a subunit of CPF complex, and Rat1 are other terminator-bound factors that physically interact with the 5′ end of genes and the associated initiation factors [30], [35]. Besides CPF complex, CF1 complex is also required for both the cleavage-polyadenylation of mRNA as well as termination of transcription. At least three subunits of this complex (Rna14, Rna15 and Pcf11) associate with both ends of a transcriptionally engaged gene [18], [36]. CF1A subunits exhibit genetic and physical interaction with several promoter-bound factors that include both the general transcription factors and gene specific factors [16], [18], [33], [37]–[41]. Furthermore, CF1A subunits are also required for juxtaposition of the promoter and terminator regions to form a looped gene structure [18]. The well-orchestrated interaction of the distal ends of a gene strongly suggests that the termination and initiation steps of transcription may operate in a cooperative manner.
The presence of termination factors on the promoter region could influence the events taking place at the 5′ end of genes. One possible role of the termination factors at the 5′ end could be to regulate initiation or reinitiation of transcription. It was recently demonstrated that proper termination of transcription is required for efficient execution of the transcription cycle in mammalian cells [42]. In that study, a termination defect adversely affected the recruitment of the general transcription factors onto the promoter of the same gene leading to a decrease in initiation of transcription. In a related study, a decrease in the density of RNAP II at the promoter region was observed in the termination-defective Ssu72-C15S mutant [43]. One possible interpretation of these results is that proper termination is important for the recruitment of polymerase at the promoter for reinitiation. It is conceivable that the physical proximity of the promoter and terminator regions, which results in a looped gene conformation, facilitates a direct transfer of the released polymerase from the 3′ end to the juxtaposed promoter [30]. This would help bypass the rate-limiting step of recruitment of polymerase on the promoter, leading to enhanced transcription of the gene. A transfer of polymerase molecules from the terminator to the promoter has, indeed, been shown for RNAP III-transcribed genes [44]. We propose that a similar termination-reinitiation coupling is taking place during RNAP II-mediated transcription as well. Another possible function of termination factors at the 5′ end of genes could be in providing directionality to the promoter-bound RNAP II to transcribe the sense strand. Genome wide analysis of human and yeast systems revealed the unexpected finding that RNAP II tends to transcribe both in the sense as well as anti-sense direction from the promoter region [45]–[48]. The promoter initiated anti-sense transcription, however, is aborted, thereby favoring productive elongation of the sense transcript. What confers directionality to the promoter-bound polymerase remains unclear. A recent study carried out in budding yeast demonstrated that the termination factors inhibit transcription of the promoter-initiated anti-sense transcripts, thereby providing directionality to the promoter-bound polymerase [49].
Here we demonstrate the role of CF1A complex in the promoter-associated transcription in budding yeast. In a mutant of Clp1 subunit of the CF1A complex, recruitment of the whole CF1A complex at the 3′ end of genes was compromised, leading to a termination defect. The termination defect coincided with a decrease in the recruitment of RNAP II on the promoter indicating a possible initiation defect. Since there was no significant decrease in the recruitment of the general transcription factors onto the 5′ end of a gene in the clp1 mutant, these results strongly suggest a novel role for the CF1A complex in reinitiation of transcription. We further found a role for CF1A complex in the inhibition of promoter-initiated anti-sense transcription. Thus, CF1A complex may have an additional function in providing directionality to bivalent yeast promoters. The CF1A-dependent promoter-based events coincide with the gene assuming a looped conformation, thereby suggesting a possible role of gene looping in reinitiation of transcription in the sense direction.
CF1A is a hexameric complex comprised of two subunits each of Rna14 and Rna15, and one subunit each of Pcf11 and Clp1 [50]. The Rna14, Rna15 and Pcf11 subunits have been studied extensively due to the availability of conditional mutant alleles. In contrast, little is known about the physiological role of Clp1. Recent studies, however, have implicated Clp1 both in the 3′ end processing of precursor mRNA and in the termination of transcription [51]–[53]. Structural analysis using mutants revealed that Clp1 makes a direct physical contact with the Pcf11 subunit of CF1A complex as well as with the Ssu72 and Ysh1 subunits of CPF complex [50]–[52].
To further analyze the role of Clp1 in transcription, we used a temperature-sensitive mutant of the factor called clp1-769-5 [54]. Western blot analysis revealed that the Clp1 protein almost completely disappeared from the mutant cells following the temperature shift to 37°C, but there was only a marginal change in the signal for other CF1A subunits at the elevated temperature (Figure S1). We examined the transcription of INO1 and CHA1 in the mutant clp1 strain in cells grown at the permissive (25°C) and non-permissive (37°C) temperatures. We chose INO1 and CHA1 for our study because their regulation is relatively well understood and their transcriptional state can be regulated by simply changing the growth conditions. Furthermore, CHA1 is relatively isolated in the yeast genome and therefore is a good candidate to study upstream and downstream transcription by transcription run-on (TRO) assay. RT-PCR was carried out using primers A and B as shown in Figure 1A and 1D in the mutant and wild type strains at 25°C and 37°C. RT-PCR analysis revealed that the transcript level of both INO1 and CHA1 decreased by about 4–8 fold upon shifting the mutant cells to 37°C (Figure 1B and 1E, lane 4; Figure 1C and 1F). No such decrease in transcript level was observed upon shifting the wild type cells to elevated temperature (Figure 1B and 1E, lane 4; Figure 1C and 1F). Thus, Clp1 is essential for optimal transcription of both INO1 and CHA1 in yeast. Since there was no appreciable decrease in the amount of CF1A subunits Rna14, Rna15 and Pcf11 in the mutant cells at the elevated temperature (Figure S1), we next checked if CF1A complex is recruited at the 3′ end of genes in the mutant cells. ChIP analysis revealed that the recruitment of Rna14, Pcf11 and Rna15 at the 3′ end of INO1 and CHA1 exhibited a decline following the temperature shift to 37°C (Figure S2, B and D, lanes 4, 12 and 20). No such decrease in the recruitment of CF1A subunits was observed in the wild type cells at elevated temperature (Figure S3, B and D, lanes 4, 12 and 20). The overall conclusion of these results is that the normal expression of INO1 and CHA1 is dependent on Clp1, and that the recruitment of a functional CF1A complex at the 3′ end of these two genes occurs in a Clp1-dependent manner.
To understand the role of Clp1 in the transcription cycle, we performed transcription run-on (TRO) analysis of CHA1 in the wild type and temperature-sensitive clp1-769-5 strains during different transcriptional states of the gene. The transcription of CHA1 is regulated by the nitrogen source in the growth medium. The gene is maintained in a transcriptionally repressed state in a medium containing ammonium sulfate as the nitrogen source, and is stimulated upon shifting cells to a medium containing serine and threonine [55]. The position of transcriptionally active RNAP II was monitored at the positions A to I as shown in Figure 2A. The TRO analysis found transcriptionally active RNAP II being almost uniformly distributed between the promoter and the terminator regions of CHA1 in the wild type strain during induced transcription (Figure 2B, lanes 3–7 and 13–17; Figure 2D). In the clp1-769-5 mutant, however, the polymerase read through the termination signal into the downstream region at elevated temperature (Figure 2C, lanes 38 and 39; Figure 2E). No such transcription readthrough was observed in the mutant strain at the permissive temperature (Figure 2C, lanes 28 and 29; Figure 2E) or in the wild type cells at 37°C (Figure 2B, lanes 18 and 19; Figure 2D). Strand-specific RT-PCR analysis corroborated the presence of sense transcripts downstream of the termination signal of CHA1 in the clp1 mutant at elevated temperature (Figure 3B, region Z). No such readthrough transcripts were observed in the isogenic wild type strain under identical conditions (Figure 3C, region Z). Strand-specific RT-PCR analysis was carried out using primers shown in Figure 3A and described in the figure legend. These results confirmed the role of Clp1 in the termination of transcription in budding yeast.
Recently, we demonstrated crosslinking of Rna14, Rna15 and Pcf11 subunits of CF1A complex to the distal ends of genes in a transcription-dependent manner [18]. Here we show that the Clp1 subunit also localizes to both the 5′ and 3′ ends of transcriptionally active INO1 and CHA1 (Figure S4, B and E, lanes 1 and 4; Figure S4, C and F). The CF1A complex, being a cleavage-polyadenylation factor, is expected to bind to the 3′ end of genes. It was, however, intriguing to find the entire CF1A complex occupying the 5′ end of genes as well. A clue regarding the role of the CF1A complex at the 5′ end of genes came when we observed that the transcription readthrough phenotype of the mutant strain at the elevated temperature was accompanied by a decrease in the TRO signal in the promoter-proximal coding region (Figure 2C, lane 33). This result strongly suggested a role for Clp1 in the initiation/reinitiation of transcription. To determine if the observed decrease in TRO signal near the 5′ end of CHA1 in the mutant was due to a failure to recruit RNAP II onto the promoter or due to a post-recruitment defect, we performed RNAP II density ChIP during the transcriptionally activated state of INO1 and CHA1 in clp1-769-5 strain at permissive and non-permissive temperatures. RNAP II ChIP was performed using primer pairs A, B, C, D, E, and F as indicated in Figure 4A and 4D. Our results show that there was indeed a decrease in the density of RNAP II at the promoter region of both INO1 and CHA1 at elevated temperature (Figure 4B, lanes 1, 2 and Figure 4C, regions A, B; Figure 4E, lane 1 and Figure 4F, region A). There was no such decrease in the polymerase density at the promoter region of genes in the wild type cells at 37°C (Figure S5, B and E lanes 1 and 2; Figure S5, C and F). The RNAP II-ChIP experiment revealed a nearly 2-fold decrease in the polymerase signal at the 5′ end of CHA1 in the mutant at 37°C (Figure 4F, region A). In contrast, TRO assay showed an at least 5-fold decrease in the polymerase signal near the promoter region of CHA1 under identical conditions (Figure 2E, region C). This discrepancy could be attributed to the presence of transcriptionally inactive paused polymerase near the 5′ end of CHA1 that can be detected by ChIP assay, but not by TRO assay. The overall conclusion of both the TRO and RNAP II-density ChIP results is that there is clearly a decrease in the amount of polymerase at the 5′ end of a gene in the clp1 mutant at elevated temperature. A plausible interpretation of these results is that a functional CF1A complex facilitates the recruitment of RNAP II onto the promoter during transcription.
Next we asked if CF1A-dependent recruitment of RNAP II on the promoter occurs during the initiation or reinitiation of transcription. During initiation of transcription, TFIID, TFIIB, TFIIA, TFIIF, RNAP II, TFIIE and TFIIH are recruited onto the promoter in that order to form the preinitiation complex (PIC) [56], [57]. The recruitment of RNAP II occurs subsequent to the formation of a TFIID-TFIIB-TFIIA complex on the promoter. This is followed by the binding of TFIIE and TFIIH to form the PIC. Following initiation of transcription, RNAP II along with TFIIF is released from the complex for elongation [58]. Simultaneously, TFIIB is also released from the complex, while the rest of the general transcription factors are left behind on the promoter forming a ‘scaffold’ that is used as a loading dock for the re-entry of RNAP II for reinitiation of transcription during subsequent transcription cycles. The composition of protein factors on the promoter, therefore, can distinguish an ‘initiation complex’ from the ‘reinitiation scaffold’ [14]. The initiation complex will contain all general transcription factors along with RNAP II, while the reinitiation scaffold will have general transcription factors with the exception of TFIIB and TFIIF and no RNAP II. Thus, to determine if CF1A-dependent recruitment of RNAP II was occurring during the initiation or reinitiation of transcription, we examined the promoter occupancy of INO1 and CHA1 for TFIID, TFIIB, TFIIF, TFIIE and TFIIH in clp1-769-5 strain at the permissive and non-permissive temperatures by ChIP assay using primer pairs indicated in Figures 5A and 5C. Our results demonstrate that TFIID, TFIIB, TFIIF, TFIIE and TFIIH occupied the promoter region of both genes in the mutant at 25°C as well as 37°C (Figure 5B and 5D, region A black bar). Similar results were observed in the isogenic wild type strain (Figure S6). TFIIB also occupied the terminator region of both genes at 25°C (Figure 5B and 5D, region D grey bar for TFIIB-ChIP panel). The presence of TFIIB at the 3′ end of genes is linked to CF1A-dependent gene looping [18]. A decrease in TFIIB signal near the 3′ end of both INO1 and CHA1 was observed in the clp1 mutant at 37°C (Figure 5B and 5D, region D grey bar for TFIIB-ChIP panel). This is in accord with the observed decrease in the TFIIB occupancy of the terminator region of transcriptionally active genes in the mutants of CF1A subunits [18]. A 25% decrease in the crosslinking of TFIIB and TFIIF to the promoter region of both INO1 and CHA1 was also observed in the mutant following the temperature shift to 37°C (Figure 5B and 5D, region A grey bar). This is in agreement with the reported release of TFIIB and TFIIF from the promoter following initiation of transcription [58]. There was no appreciable change in the promoter occupancy of the rest of the general transcription factors following a shift to elevated temperature, despite a decrease in the promoter-bound RNAP II signal. These results suggest that it is the reinitiation of transcription that is adversely affected in the clp1-769-5 cells at elevated temperature. The overall conclusion of these results is that a functional CF1A complex is required for the recruitment of polymerase to the promoter for reinitiation of transcription. The possibility of CF1A complex being required for the recruitment of TFIIB and TFIIF for reinitiation cannot be ruled out.
During the transcription cycle, RNAP II in the promoter-bound initiation complex transcribes in the sense direction, producing mRNA. Genome wide analysis of transcribing polymerases has identified RNAP II molecules in the region just upstream of the transcription start site in most eukaryotic genes [45]–[48]. These upstream polymerases are involved in divergent anti-sense transcription, producing non-coding RNA (ncRNA). These promoter-initiated, anti-sense ncRNAs are capped, non-adenylated, heterogeneous in size and often belong to a class of RNA called CUTs (cryptic unstable transcripts) that are rapidly degraded by the RNA surveillance mechanism of the cell [59], [60]. Having already implicated CF1A complex in the sense-transcription of mRNA, we next asked if CF1A complex has a role in the regulation of divergent, anti-sense transcription of ncRNA. To address the issue, we performed strand-specific RT-PCR for CHA1 in wild type and clp1-769-5 mutant as described in [61]. In wild type cells, we could not detect promoter-initiated anti-sense transcripts under any condition (Figure 3C, region W). In the clp1-769-5 mutant also, no appreciable divergent anti-sense RNA could be detected at 25°C (Figure 3B, region W, black bar). At the elevated temperature, however, a 5-fold increase in the signal for promoter-associated anti-sense transcripts was observed in the mutant strain (Figure 3B, region W, grey bar). These results were corroborated by TRO assay, which detected the presence of transcriptionally engaged polymerase in the region upstream of CHA1 in the mutant strain at 37°C (Figure 2C, lane 31; Figure 2E region A).
The increase in the level of divergent anti-sense transcripts initiating from the 5′ end of the gene in the mutant could be attributed either to the stabilization of the transcripts or to the synthesis of promoter-initiated anti-sense transcripts in the mutant. Since TRO assay detected the presence of transcriptionally active RNAP II just upstream of the promoter of CHA1 in the clp1 mutant at elevated temperature, it is reasonable to conclude that the observed anti-sense transcripts were not the consequence of stabilization of RNA, but the result of divergent anti-sense transcription initiating from the 5′ end of the gene. These results raise the possibility of a role for the CF1A complex in limiting the transcription of promoter-associated anti-sense ncRNA, thereby favoring transcription of mRNA in the sense direction. We therefore propose that the CF1A complex may have an additional role in providing directionality to otherwise bidirectional yeast promoters. Our results are in agreement with a recent report that showed an increase in promoter-initiated divergent anti-sense transcription in termination-defective mutants [49].
Thus, in the absence of a functional CF1A complex in the clp1-769-5 mutant, the promoter-associated downstream transcription of mRNA in the sense direction as well as the divergent upstream transcription of anti-sense RNA, exhibited an aberrant pattern.
A logical interpretation of the results described above is that the CF1A complex is not merely contacting the 5′ end of transcriptionally active genes, but is also influencing early events in the transcription cycle. Next we asked how the CF1A complex is recruited to the 5′ end of a gene. The binding of CF1A complex to the 5′ end could be independent of its recruitment at the 3′ end of a gene. Alternatively, gene looping, which is the transcription-dependent interaction of the promoter and the terminator regions of a gene, may facilitate positioning of the terminator-bound CF1A complex at the 5′ end of a gene [30]. We have earlier demonstrated the role of CF1A subunits Rna14, Rna15 and Pcf11 in gene looping [18]. To corroborate the role of CF1A complex in gene loop formation, we performed 3C analysis of INO1 and CHA1 in the clp1-769-5 mutant at the permissive and non-permissive temperatures. Gene looping was monitored by the P1-T1 primer pair shown in Figure 6A and 6D, by the method described in [62]. A distinct P1-T1 PCR signal was obtained for both INO1 and CHA1 when the mutant cells were grown at 25°C (Figure 6B and 6E, lane 1; Figure 6C and 6F, black bar). The P1-T1 looping signal decreased by about 4–6 fold following transfer of cells to 37°C (Figure 6B and 6E, lane 2; Figure 6C and 6F, grey bar). These results confirmed that a functional CF1A complex is indispensable for gene loop formation in budding yeast.
The CF1A complex, which is known to localize and operate at the 3′ end of RNAP II-transcribed genes in yeast, also contacts the 5′ end of genes. The promoter occupancy of the CF1A complex coincides with the gene assuming a looped conformation. We recently purified a holo-TFIIB complex that contained all the CF1 subunits and the general transcription factor TFIIB [18]. We showed that the holo-TFIIB complex mediates gene loop formation by simultaneously contacting the distal ends of a gene. Accordingly, gene looping was not observed in mutants of the Rna14, Rna15 and Pcf11 subunits of CF1 complex. Here we show that gene looping is abolished in the clp1 mutant as well. Whether the presence of CF1A at the 5′ end is the cause or the effect of gene looping is still unclear, but it is quite evident that the CF1A subunits at the 5′ end of a gene affect early events during the transcription cycle. The CF1A-dependent gene loop juxtaposes the terminator region of a gene with its cognate promoter. This arrangement may facilitate binding of the RNAP II released from the terminator at the end of a transcription cycle to the promoter for starting the next round of transcription. Accordingly, we observed a 2-fold decrease in the RNAP II density at the promoter in the absence of a functional CF1A complex. Since the promoter occupancy of the general transcription factors, with the exception of TFIIB and TFIIF, remained unaltered in the clp1 mutant, we propose that the CF1A complex, by virtue of its role in gene looping, affects reinitiation rather than initiation of transcription. The possibility of CF1A subunits playing a role in the initiation, however, still cannot be ruled out. A similar study carried out in a mammalian system found termination factors affecting initiation rather than reinitiation of transcription [42]. The mechanism of termination-dependent initiation, however, was not clear in that study. Here we propose that the CF1A-dependent gene looping may account for the termination-reinitiation link.
Since a majority of eukaryotic promoters are intrinsically bidirectional, there should be some mechanism in the cell to favor transcription of mRNA in the sense direction, over the anti-sense transcription of ncRNA [48]. We found that CF1A complex, while facilitating reinitiation in the sense direction, has an additional function in restricting transcription of the promoter-associated anti-sense RNA. The divergent, anti-sense transcription of ncRNA is widely believed to be terminated by the Nrd1-dependent pathway in yeast [63]. The CF1A complex, in general, is associated with the termination of mRNA synthesis by the poly(A)-dependent pathway [8], [9]. Our results suggest that CF1A complex may be involved in the termination of anti-sense ncRNA synthesis as well. These results are in agreement with a recent report that demonstrated crosslinking of mammalian termination factors Xrn2 and TTF2 to the 5′ end of genes and their involvement in limiting promoter-initiated anti-sense transcription [64]. The regulation of transcriptional directionality by Ssu72, which is a subunit of the CPF 3′ end processing complex in yeast, further corroborates our results [49]. The limiting of promoter-initiated anti-sense transcription may direct the polymerase to move in the sense direction, thereby producing mRNA. Thus, CF1A complex may be involved in providing directionality to bivalent promoters.
Based on these results we propose a model of transcription by RNAP II (Figure 7). The transcription-dependent promoter-terminator interaction places CF1A complex in the vicinity of the promoter. The promoter-bound CF1A affects transcription at two levels. First, CF1A-dependent termination releases RNAP II molecules from the 3′ end of gene near the promoter, thereby facilitating the recruitment of RNAP II to the promoter for reinitiation. Secondly, it provides directionality to the bidirectional promoter, thereby promoting the synthesis of mRNA over anti-sense ncRNA. Whether the CF1A complex limits promoter-initiated anti-sense transcripts by virtue of its termination activity needs further investigation. The net result is an upregulation of mRNA synthesis in the presence of a functional CF1A complex. Although a role for gene looping in facilitating transfer of polymerase from the terminator to the promoter for reinitiation has previously been hypothesized, this is the first instance where gene looping has actually been shown to help reinitiation of transcription.
Yeast strains used in this study are listed in supplemental Table S1. SAM53, which contained the Myc-tag at the carboxy-terminal of Clp1 in BY4733 strain background, was constructed by transforming the parental strain with the PCR product amplified from pFA6-13Myc-TRP1. The temperature-sensitive mutant clp1-769-5 was kindly provided by Dr. Philip Hieter. Strains NAH20, NAH21, NAH22, NAH31, NAH32 and NAH33 were derived from the temperature sensitive clp1-769-5 strain by adding either the Myc or the Tap-tag at the carboxy terminus of an initiation factor or a termination factor. Strains NAH20 (Myc-tagged TFIIB), NAH21 (Myc-tagged Rna14), NAH22 (Myc-tagged Pcf11) and NAH25 (Myc-tagged Rna15), which contained the Myc-tag at the carboxy-terminus of the indicated factor, were constructed by transforming the clp1-769-5 strain with the PCR product amplified from pFA6-13Myc-KanMX6. For TAP-tagging of the general transcription factors, first the temperature-sensitive clp1-769-5 strain was made trp1− by replacing TRP1 with a KanMX cassette that was PCR amplified from pUG6. Next a TAP-tag was inserted at the carboxy-terminus of TFIIH subunit Ccl1 (NAH31), TFIIF subunit Tfg2 (NAH33) and TFIIE subunit Tfa2 (NAH32) by transforming the clp1-769-5-(trp1) strain with the TAP-cassette amplified from plasmid pBS1479.
Cultures were started by inoculating 5 ml of YP-dextrose medium with colonies from a freshly streaked plate, and grown at 25°C overnight with constant shaking. Next morning, overnight grown cultures were diluted (1∶100 dilution for the temperature-sensitive strains, and 0.5∶100 dilution for the wild type strains) to an appropriate volume and grown to OD600∼0.4. The dilution was done in the appropriate synthetic complete-drop out medium. Induction was done for 2 hrs at 25°C before shifting the cells to 37°C for another 2 hours for the deactivation step. Usually, this takes the cells to OD600 of about 0.7–0.8. At this stage, the cells are ready for processing for RT-PCR, 3C, ChIP, or TRO assays.
Transcription run-on (TRO) assay was performed by the modification of protocols described in Birse et al., 1997 and Hirayoshi and Lis, 1999 [65], [66]. For CHA1, WT and clp1-769-5 cells were grown in 100 ml of synthetic complete medium containing ammonium sulfate until A600 reached 0.4. Cells were centrifuged and resuspended in 100 ml of synthetic media containing serine and threonine (1 g/l each) and induced for 2 hours at 25°C. 50 ml of the cultures were centrifuged and resuspended in 50 ml of pre-warmed (37°C) serine and threonine containing medium and deactivation was done at 37°C for 120 minutes. The cell pellet obtained from 50 ml of liquid culture was washed with 10 ml cold TMN buffer (10 mM Tris-HCl pH 7.5, 5 mM MgCl2, 100 mM NaCl) and resuspended in 940 µl of DEPC (Diethylpyrocarbonate)-treated cold water. To the cell suspension, 60 µl of 10% sarkosyl was added and incubation performed on ice for 25 min to permeabilize the cells. Permeabilized cells were recovered by a low-speed centrifugation (1.2×g, 6 minutes) and directly used in the run on transcription assay. Elongation of transcripts initiated in vivo was resumed by resuspending cells in 120 µl of 2.5× reaction buffer (50 mM Tris-HCl pH 7.5, 500 mM KCl, 80 mM MgCl2, 5 mM DTT), 45 µl of NTPs/RNase inhibitor mix (10 mM each of CTP, ATP, and GTP and 300 units of RNase Inhibitor), and 7 µl of [α-32P]-UTP (3000 Ci/mmol, 10 µCi/µl). The reaction mix was incubated at 30°C for 2 minutes to allow transcript elongation. The reaction was stopped by adding 1 ml of cold TMN buffer and quickly spun at low speed. The recovered pellet was resuspended in 350 µl of Trizol. About 250 µl of acid-washed glass beads were added and the cells were lysed by vigorous shaking for 5 minutes on an agitator at room temperature. After lysis, tubes were spun for 5 minutes at 13800×g. To the recovered supernatant, 700 µl of Trizol and 200 µl of Chloroform were added and the samples were vigorously shaken on a vortexer, left on the bench for 5 minutes, and centrifuged at high speed for 10 minutes.
To isolate RNA, the supernatant was extracted twice with phenol/chloroform (pH 4.2). Labeled RNA was precipitated by adding 0.1 volumes of 10 M LiCl, 0.1 volumes of yeast tRNA (80 mg/ml) and 2.5 volumes of 100% ethanol. The mix was incubated at −20°C for 20 minutes followed by centrifugation at maximum speed for 15 minutes. The RNA pellet was resuspended in 60 µl of DEPC-treated water and denatured by adding 5 µl of 2 M NaOH followed by incubation on ice for 5 minutes. The NaOH was then neutralized by adding 12 µl of sodium acetate/acetic acid mix (0.3 M sodium acetate pH 5.2 and 0.5 µl of glacial acetic acid) and boiling the contents for 5 minutes.
In parallel, DNA probes of about 200–300 bp each in length, spanning the desired regions of the CHA1 gene, including the upstream and downstream regions, were obtained by PCR amplification (See Fig. 1A for the position of probes). 10 µg of probe was denatured by boiling in 0.1 N NaOH and 1 mM EDTA for 10 minutes to form single stranded DNA. The heat-denatured probes were then slot-blotted on a ZETA-probe GT membrane (BIO-RAD), according to manufacturer's instructions. Adsorbed DNA was crosslinked to the membrane by baking at 80°C in a vacuum oven for 30 minutes. The membrane was then prehybridized with 10 ml of hybridization solution (0.5M potassium phosphate pH 7.2, 7% SDS) at 55°C for at least 30 minutes. The denatured RNA in hybridization solution from the step described above was added to the prehybridized membrane. Labeled RNA was allowed to hybridize to the probe for 18–24 hours at 55°C in a rotator. After hybridization, the membrane was washed twice with 20 ml of a solution containing 0.1% SDS and 1XSSC for 7 minutes at 55°C, and twice with 20 ml of a solution containing 0.1% SDS and 0.1XSSC for 7 minutes at 55°C. After drying, the membrane was exposed to X-ray film overnight in an autoradiography cassette and the films were developed in a Kodak M35A X-OMAT system. All TRO signals were quantified using the GEL LOGIC 200 (KODAK) system and normalized with respect to the 18S control.
ChIP was performed as described in [16]. Primers used for ChIP-PCR are described in supplemental Table S2 and indicated in Figures 4A, 4D, 5A and 5C. RNAP II ChIP was performed using anti-Rpb3 antibodies obtained from Santa Cruz (Cat# sc-101614). For ChIP analysis of CF1 subunits Clp1, Rna14, Rna15 and Pcf11, a Myc-tag was inserted at the carboxy-terminus of each subunit, and ChIP was performed using anti-Myc antibodies obtained from Upstate Biotechnology (Cat# 06-549). ChIP of TFIID was performed using anti-TBP antibodies obtained from Santa Cruz (Cat# sc-33736). ChIP analysis of TFIIB was carried out using anti-Myc antibodies in a strain with a C-terminus Myc-tagged TFIIB. For ChIP of TFIIF, TFIIE and TFIIH, strains were constructed with a TAP-tag inserted at the carboxy-terminus of Tfg2, Tfa2, and Ccl1 subunits respectively, and ChIP was performed using IgG-Sepharose beads.
Crosslinking, cell lysis and isolation of chromatin was done as described in [16]. Chromatin preparation obtained above was sheared by sonication (15 pulses of 20 seconds each with 1 minute cooling after each pulse). Sonication was performed at the 25% duty cycle in a Branson digital sonifier. Following sonication, samples were centrifuged at 14,000 rpm for 15 minutes in a refrigerated microfuge. The pellet was discarded and the supernatant was used in subsequent steps. The amount of sonicated chromatin to be used for immunoprecipitation depended on the quality of the antibody and the amount protein (antigen) present in the chromatin preparation. Approximately 5–10 µg of appropriate antibody (the amount of antibody added need to be optimized for each antibody preparation) was added to the chromatin preparation and allowed to bind for 4 hours at 4°C with gentle shaking. The antigen-antibody complex was adsorbed onto 20 µl of Protein A-Sepharose beads (the beads should be pre-washed with FA-lysis buffer) for 1 hour with gentle shaking at 4°C.
The beads were washed successively with 1 ml each of FA-lysis buffer (two times), FA-lysis buffer containing 500 mM NaCl (two times), ChIP wash buffer (10 mM Tris-HCl pH 8.0, 250 mM LiCl2, 0.5% tergitol, 0.5% sodium deoxycholate and 1 mM EDTA) and TE buffer. All the washing steps were performed at room temperature. The beads were resuspended in 250 µl of ChIP elution buffer (50 mM Tris-HCl of pH 8.0, 10 mM EDTA and 1% SDS); incubated at 65°C for 20 minutes; briefly spun; and the supernatant was collected and incubated with 10 µg of DNase-free RNase for 15 minutes at 37°C. 20 µg proteinase K and 2.5 µl 10% SDS were added and the crosslinks were reversed by overnight incubation at 65°C. Samples were extracted with phenol-chloroform at least two times followed by ethanol precipitation of DNA using glycogen as a carrier. The DNA pellet was resuspended in 50 µl TE and used as template for PCR. Chromatin immunoprecipitated DNA was PCR amplified (30 cycles) by appropriate primer pairs, and subjected to quantification and statistical analysis as described below. Each experiment was repeated with at least two independently grown cultures.
3C experiments were performed exactly as described previously [62]. The primers used for 3C analysis are shown in supplemental Table S2. A 50 ml cell culture was grown as described above. Cells were formaldehyde crosslinked for 15 minutes at 25°C. The crosslinked crude chromatin was digested with restriction endonuclease(s) (Alu1 for INO1; NlaIV and Alu1 for CHA1). After restriction digestion, the reaction volume was diluted by 7.5 fold to minimize intermolecular ligation in the next step. Ligation reactions were performed at room temperature for 90 minutes. The crosslinks were reversed by incubating at 65°C overnight. DNA was extracted with phenol-chloroform followed by ethanol precipitation. 300 ng of DNA was used as template in the PCR using the P1-T1 divergent primer pair as indicated in Figures 6A and 6D. Control PCR products were generated using a convergent primer pair (F2-R1). PCR and detection of products were performed exactly as described in [62]. Each experiment was performed with at least four independently grown cultures. The P1-T1 PCR signals are normalized with respect to F2-R1 PCR signals.
Isolation of total RNA and transcription analysis was performed by RT-PCR using oligo-dT primer at the reverse transcription step as described previously (El Kaderi et al 2009). The RT-PCR primers are shown in supplemental Table S2. A minus-RT control (without reverse transcriptase) was always performed to ensure that the RT-PCR signal was not coming from contaminating DNA. The RT-PCR results were normalized with respect to the 18S rRNA control that is transcribed by RNAP I and requires a different set of transcription factors.
Strand-specific RT-PCR was performed to distinguish between sense and anti-sense transcripts. Total RNA for this procedure was extracted using Trizol reagent. The cell pellet was resuspended in 500 µl of Trizol. Acid-washed glass beads (about 250 µl) were added to the cell suspension. Cells were lysed by vigorous shaking for 10 minutes on an agitator at 4°C. Whole cell lysate was recovered by puncturing the bottom of the tube with a 22-guage needle, placing it on the top of a 15 ml pre-chilled centrifuge tube and centrifuging at 300×g for 2 minutes. The filtrate was transferred into a chilled 1.5 ml microfuge tube and 500 µl more Trizol reagent was added. After adding 200 µl of chloroform, tubes were vigorously agitated and left on the bench for 5 minutes. The tubes were then centrifuged at high speed for 10 minutes. The supernatant was extracted two times with an equal volume of phenol/chloroform (pH 4.3), followed by an extraction with chloroform only. RNA was precipitated using 0.1 volumes 10 M LiCl and 3 volumes cold ethanol in the presence of glycogen as a carrier. The precipitated RNA was collected by centrifugation at 14220×g on a table-top centrifuge for 15 minutes. The air-dried RNA pellet was resuspended in 50 µl of DEPC-treated water and the concentration was estimated using a spectrophotometer.
Strand specific RT-PCR was now performed as described in [16]. 1 µg of RNA was used to make cDNA using strand-specific primers for CHA1 as shown in Figure 2A. Primers As, Bs, Cs and Ds were used to reverse-transcribe sense mRNA, while Aas, Bas and Cas primers were used for reverse transcription of anti-sense transcripts. This was followed by PCR amplification of cDNA for regions W, X, Y and Z using primer pairs Aas-Bs, Aas-As, Bas-Cs and Cas-Ds respectively. A minus-RT control (without reverse transcriptase) was always performed to ensure that the strand-specific RT-PCR signal was not due to contaminating DNA in the RNA preparation. RT-PCR results were normalized with respect to the 18S rRNA control that is transcribed by RNAP I and requires a different set of transcription factors.
The quantification was performed as described in [62]. In ChIP, 3C and RT-PCR experiments described above, PCR products were fractionated on a 1.5% agarose gel and visualized by ethidium bromide using the Gel Logic 200 system. The net intensity of the bands was calculated using the Kodak 1D software. Using the scaled net intensities, a minimum of eight trials were analyzed under the Univariate ANOVA model in the SPSS statistical software to verify that there was no significant gel interaction (P<0.05). Each trial was also duplicated to ensure that there was no significant trial interaction (P<0.05). Scaled net intensities were then used to generate ratio data comparing the experimental test with that of the control PCR, which was then used to generate the mean and standard deviation as shown in the graphs. For all the quantification graphs, the error bars represent one unit of standard deviation based on at least eight independent PCRs from four separate IPs or 3C reactions or reverse-transcribed RNA samples from two independently grown cultures. For TRO, quantification was done with four independent replicates.
|
10.1371/journal.pntd.0006069 | Polyclonal antibodies for the detection of Trypanosoma cruzi circulating antigens | Detection of Trypanosoma cruzi antigens in clinical samples is considered an important diagnostic tool for Chagas disease. The production and use of polyclonal antibodies may contribute to an increase in the sensitivity of immunodiagnosis of Chagas disease.
Polyclonal antibodies were raised in alpacas, rabbits, and hens immunized with trypomastigote excreted-secreted antigen, membrane proteins, trypomastigote lysate antigen and recombinant 1F8 to produce polyclonal antibodies. Western blot analysis was performed to determine specificity of the developed antibodies. An antigen capture ELISA of circulating antigens in serum, plasma and urine samples was developed using IgY polyclonal antibodies against T. cruzi membrane antigens (capture antibody) and IgG from alpaca raised against TESA. A total of 33 serum, 23 plasma and 9 urine samples were analyzed using the developed test. Among serum samples, compared to serology, the antigen capture ELISA tested positive in 55% of samples. All plasma samples from serology positive subjects were positive in the antigen capture ELISA. All urine positive samples had corresponding plasma samples that were also positive when tested by the antigen capture ELISA.
Polyclonal antibodies are useful for detection of circulating antigens in both the plasma and urine of infected individuals. Detection of antigens is direct evidence of the presence of the parasite, and could be a better surrogate of current infection status.
| Current diagnosis of Chagas disease is still cumbersome. Diagnosis is based on antibody detection with at least two assays of distinct mechanisms. If a discrepancy exists, a third assay must be performed. However, detection of antibodies is not indicative of current infection. Molecular-based techniques such as qPCR have been used for diagnosis and as a gold standard in the demonstration of therapeutic failure, but availability of genomic material depends on the presence of parasites in the bloodstream. Detection of parasite-derived antigens represents a better alternative for diagnosis, as several proteins are secreted/excreted by the parasites and may be detected in blood and in the urine of infected individuals. This study describes the development of polyclonal antibodies raised against different Trypanosoma cruzi antigens and their applicability for the diagnosis of Chagas disease using the widely-used ELISA format.
| Chagas disease, caused by the protozoan parasite Trypanosoma cruzi, is endemic to many parts of the Americas [1–3]. This parasite infects a wide variety of wild and domestic mammals including humans [2]. The disease is transmitted by insect vectors (members of the Triatominae family) with metacyclic trypomastigotes present in their feces. Parasite trypomastigotes gain access to the tissue and circulatory systems at the bite site, through wounds caused by scratching the bite site, or through the mucous membranes [4]. Other transmission routes include congenital transmission, blood transfusions, organ transplantation, oral transmission through the consumption of food contaminated with feces from infected insects, and accidental laboratory exposure [1, 3, 4].
It has been estimated that Chagas disease affects approximately 8 million people and may cause about 12,000 deaths each year (45,000 in the 1980s and 23,000 in the 1990s) [2,5]. Bolivia is the country with highest endemicity with a prevalence of up to 80–90% in rural areas [6,7].
Two main phases can be distinguished in Chagas disease, the acute and chronic phase, each with different characteristics. The acute phase occurs at the beginning of the infection and is characterized by patent parasitemia with most of the patients asymptomatic. Approximately 75% of acute cases are in children under 10 years old [3,8]. In most cases, patients are asymptomatic (95%), however when the inoculation site is the conjunctiva mucous membrane, the characteristic Romaña’s sign, an eyelid edema, may appear [9,10].
The chronic phase can appear years to decades following the acute phase. This phase is characterized by a very low parasitemia and the most suitable methods for diagnosis are immunological assays [11]. Unlike the acute phase, about one third of infected patients will develop chronic phase symptoms. These symptoms mainly include heart diseases such as cardiomyopathy which is associated with heart insufficiency, sometimes leading to mortality including sudden death in some patients [10, 12]. The disease may also affect the gastrointestinal system causing mega colon or mega esophagus [9, 13].
Diagnosis of Chagas disease is based on clinical and laboratory assessment. Most laboratory assays are dependent on the detection of antibodies. Two positive results, preferably based on serological methods of distinct mechanisms (e.g., whole-parasite lysate and recombinant antigens), are required for an individual to be considered Chagas disease positive [3, 14]. As with most of the of serological assays, these tests are not indicative of current infection when used alone and may cross-react with other parasitic diseases such as leishmaniasis and malaria, depending on the antigen used in the assay [15].
Immunological diagnosis is based on the use of parasite derived antigens. Trypomastigote excreted-secreted antigen (TESA) is used in Western blot (TESA-Blot) for both acute and chronic phase diagnosis, generating a characteristic pattern of bands depending on the strain of T. cruzi used. TESA-blot is highly specific and sensitive; sera of infected individuals identify protein bands of 130–200 kDa in the acute phase, while sera from individuals in the chronic phase identify protein bands of 150–160 kDa [16]. However, Western blot is not very economical to produce, requiring special training and a sophisticated laboratory.
While antibody detection is indirect evidence of infection, detection of any antigenic fraction of the parasite is considered the equivalent of finding the whole parasite. Antigen detection may even occur prior to development antibodies at detectable levels [17]. Several reports have shown the presence of different proteins of T. cruzi in the urine of infected animals and humans [18–20]. These proteins have been used to develop monoclonal and polyclonal specific antibodies to be used for antigen detection both in urine and serum samples [17,19,21]. Due to the variety of circulating and excreted antigens, targeting a specific protein decreases sensitivity because the presence of antigens is variable and depends on different factors such as the phase of the disease [21,22], renal injury [20] among others.
The antigen 1F8 is a recombinant protein with a molecular weight of 24–25 kDa derived from a protein found in the flagellum of T. cruzi [23]. This calcium binding protein is used as antigen in an ELISA for the diagnosis of both acute and chronic Chagas disease [24] with high sensitivity and specificity; but detection of the presence of this protein in sera or urine samples has not been performed.
The use of antigens of the parasite for the production of polyclonal antibodies designated for diagnosis or for therapy has been a very important tool for research. IgY is a type of immunoglobulin, and the major one in birds, with a molecular weight of 180 kDa. This is much larger than the IgG of most mammals, often about 159 kDa [25]. One of the most important characteristics of IgY antibodies is that they are able to recognize different epitopes than the antibodies raised in the mammals usually do [26]. In addition, IgY does not activate the complement system, providing a great advantage when used as capture antibody in immunoassays [27]. Antibodies from camelids (IgG2 and IgG3) because of their lower size probably recognize inaccessible epitopes that may not be recognized by mammalian antibodies [28–30].
To produce polyclonal antibodies, we used three different animals: alpacas, rabbits, and hens and immunized them with different T. cruzi antigens. The resulting antibodies were then used to develop an antigen capture ELISA and tested for their ability to discriminate and identify individuals infected with T. cruzi the agent of Chagas disease.
Human sera, plasma, and urine samples were archived samples obtained from previous studies. The Human Ethics Committee of the Universidad Peruana Cayetano Heredia approved the use of these archived samples.
The Animal Ethics Committee of the Universidad Peruana Cayetano Heredia approved the protocols for the use of animals for antibody production, approval Code 61549—Cons-CIEA-029-2014.
The Animal Ethics Committee of the Universidad Peruana Cayetano Heredia is registered in the Office of Laboratory Animal Welfare, Department of Health and Human Services, National Institutes of Health (NIH—USA) and follows its rules and laws.
The use of animals in this study was performed following the Deontological Code of the Medical Veterinary College of Peru; The Care and Use of Experimental Animals. Canadian Council on Animal Care 1980 and the Australian code of practice for the care and use of animals for scientific purposes 1997.
In order to produce polyclonal antibodies alpacas, rabbits, and hens were immunized using four different antigens: trypomastigote lysate antigen (TLA), trypomastigote membrane proteins (TMP), trypomastigote excretory-secretory antigen (TESA), and a commercial recombinant 1F8 T. cruzi antigen (Genway Biotech Inc, CA-USA). TLA and TMP were obtained from T. cruzi Y strain trypomastigotes. Parasites were washed three times using cold PBS at 2,000 x g for 10 min before antigen preparation.
For TLA the parasite pellet was resuspended in 2 ml PBS, frozen and thawed three times using a dry ice-ethanol bath, and sonicated (Misonix, Sonicator 3000) for 4 cycles at 30 s ON, 60 s OFF (Output Power: 3). The suspension was centrifuge at 13,500 x g for 20 min. The supernatant was recovered and used as antigen.
TMP were extracted using a modify protocol [31]. The parasite pellet was resuspended in 200 μl of 10 mM Tris-HCI, pH 7.4, 140 mM NaC1, and 2.0% Triton X-114 (Triton X-114 buffer), the tube was incubated 90 min at 4°C, and centrifuged at 10,000 x g for 15 min at 4°C. The detergent phase, found at bottom of the tube, was mixed with an equal volume of Triton X-114 buffer, incubated 1 min at 37°C, and centrifuged for15 min at 10,000 x g at room temperature. The detergent phase was used as antigen.
The TESA antigen was harvested from T. cruzi Y strain growth in LLC-MK2 cells as previously described [15]. After the sodium dodecyl sulfate-polyacrylamide gel electrophoresis (SDS-PAGE), and transfer to nitrocellulose, the protein band (150–160 kDa) was excised from the nitrocellulose paper using the reaction of chronic positive sera as a reference. The paper was digested by incubating the excised paper with 250 μl of dimethyl sulfoxide (DMSO) per 20 mm2 of nitrocellulose for 1h at room temperature on a rocker mixer. After the incubation, 0.05 M carbonate/bicarbonate buffer (pH = 9.6) was added drop by drop until a volume equivalent to the DMSO used for extracting the antigen from the nitrocellulose paper was added. The mixture was then centrifuged at 4°C for 10 min at 10,000 x g and the pellet washed by centrifugation at 4°C for 10 min at 10,000 x g with 0.1 volumes of phosphate buffered saline (PBS, pH = 7.4). The pellet was re-suspended in an equal volume of PBS. After production, all antigens were stored at -80°C (no more than one month) until used.
Alpacas were initially immunized using complete Freund’s adjuvant and boosted with incomplete Freund’s adjuvant (Sigma—Aldrich, MO-USA). Rabbits and chickens were immunized using one volume of antigen and one volume of Sigma Adjuvant System (SAS, S6322—Sigma—Aldrich). Three, two years old female alpacas (Huacaya breed), eight hens (New Hampshire breed) and eight, two months old rabbits (New Zealand breed) were immunized according to the scheme showed in Table 1.
Alpaca immunization was carried out according to previous references [32–34]. The blood from alpacas and rabbits were collected and centrifuged at 1,100 x g for 10 min to obtain the sera. The sera were stored at -20°C until used. The collected eggs were maintained at 4°C prior to antibody purification.
The HiTrap Protein A columns (GE Healthcare Life Sciences, PA, USA) was initially used to separate IgG3 and IgG1 from alpaca. The non-bound fraction was further purified using HiTrap Protein G (GE Healthcare Life Sciences) to separate the IgG2. Rabbit IgG was purified by affinity chromatography using HiTrap Protein A column (GE Healthcare Life Sciences) following manufacturer's instructions.
After two months of immunization, all the eggs were pooled according to antigen and time of collection. Egg yolk was separated from the white and proteins were extracted as described before with some modifications [35,36]. Briefly, for one yolk (about 15 ml volume), 30 ml of PBS were added and mixed carefully for 5 min. Then, 15 ml of chloroform were added and mixed again. The mixture was refrigerated at 4°C for 1–2 h then centrifuged at 1,100 x g for 10 min at room temperature. The upper phase was collected and dialyzed overnight with PBS then concentrated using an Amicon YM 100 filter unit (Amicon, Millipore, Darmstadt, Germany). To remove lipid residues and proteins completely, the concentrated samples were further purified with the Pierce Chicken IgY purification kit (Thermo Fisher Scientific) following manufacturer’s instructions. Finally, IgY antibodies were purified using the HiTrap IgY Purification columns (General Electric, Uppsala, Sweden) following manufacturer's instructions.
To verify the purity of antibodies, SDS-PAGE under non-reducing conditions was performed. Alpaca IgG1, IgG2, and IgG3, and rabbit IgG samples were diluted 1:10 using sample buffer 1 [100mM Tris-HCl (pH = 6.8), 4% (w/v) dodecyl sulfate (SDS), 0.2% (w/v) bromophenol blue, and 20% (v/v) glycerol)]. IgY samples were diluted 1:4 using a sample buffer 2 [62.6mM Tris-HCl (pH = 6.8), 2% (v/v) SDS, 25% (v/v) glycerol)]. After electrophoresis, the gels were stained using 0.25% Coomassie blue.
To corroborate that the antibodies were reacting with the target parts of the parasite an indirect immunofluorescence assay (IFA) was performed using T. cruzi Y strain epimastigotes. Briefly, epimastigotes were harvested from LIT cultures (liver infusion tryptose medium) and washed by centrifugation at 1,100 x g for 10 min using 1% formalin in PBS and resuspended in PBS to a final concentration of 103 epimastigotes/ml. A total of 20 μl/well of epimastigote suspension was fixed on poly-L-lysine pretreated slides. Fixed epimastigotes were then incubated with each of the purified antibodies at 37°C for 45 min, washed three times with PBS, and incubated with Goat Anti-Llama IgG H&L (FITC), Goat Anti-Rabbit IgG H&L (FITC) or Goat Anti-Chicken IgY H&L (FITC) diluted in PBS, 0.002% Evans blue, and incubated at 37°C for 30 min. Slides were observed at 400X under an immunofluorescence microscope.
Three milliliters of urine sample were lyophilized and reconstituted in 300 μl of PBS (pH 7.2).
Serum or plasma samples were pretreated as previously described [37]. Briefly, 50 μl of sample was diluted with 60 μl of PBS, 0.05% Tween 20, 1.0% milk– 0.2% Bovine Serum Albumin (BSA) and heated at 56°C for 30 min.
Initially different combinations of the developed polyclonal antibodies were used to standardize an antigen capture ELISA using TLA spiked urine or sera samples. The final protocol consisted in sensitizing a Nunc Maxisorp 96 well plate (Nunc Nalgene, Rochester, NY), overnight at 4°C, with anti-membrane IgY (4 μg/ml) in carbonate-bicarbonate buffer (pH = 9.6, capture antibody). The plate was washed three times with PBS, 0.05% Tween 20, blocked with PBS, 0.05% Tween 20, 6% semi-skimmed milk, 1% BSA for 2 h at room temperature. Following washing, 100 μl/well of pretreated samples of either urine, serum or plasma samples pretreated were added and the plate was incubated at 37°C for 1 h. The plate was washed again and 100 μl of detection antibody (alpaca anti-TESA IgG) at 4 μg/ml in PBS, 0.05%Tween 20, 1% milk, 0.2% BSA was added and incubated at 37°C for 1 h. After the final wash, goat-anti-llama peroxidase conjugate (Bethyl, Laboratories Inc.) was added at 1: 7500 in PBS, 0.05%Tween 20, 1% milk, 0.2% BSA and incubated at 37°C for 30 min. After washing, the plate was developed using OPD (Sigma FAST Sigma-Aldrich) as substrate for 15 min. The reaction was stopped using 2 M H2SO4 and the plate was read at 490 nm using the VERSA Max ELISA plate reader (Molecular Devices, LLC, Sunnyvale, CA).
A sample was considered positive if the absorbance (optical density, OD) obtained in the ELISA was higher than the cut-off value. The cut-off value was determined using the mean plus two standard deviations of the absorbance obtained from all samples negative by serology and qPCR including samples from the volunteers.
Diagnosis of Chagas disease in human samples was based on serological assays. ELISA was performed using Chagatek Wiener Recombinante v3.0 ELISA (Wiener laboratories, Rosario, Argentina). Western blot analysis was performed using same TESA used for antibody production. Indirect hemagglutination assay (IHA) was performed using the Chagas Polychaco kit (Lemos Laboratories, Buenos Aires, Argentina). Real time PCR (qPCR) was performed using primers and TaqMan probes targeting the nuclear satellite DNA of T. cruzi as described previously [38–40].
Serum (n = 30), plasma (n = 23) and urine samples (n = 6) were archived samples from HIV positive adults. None of the subjects received or was receiving treatment at the moment of enrollment. A sample was considered positive or negative for Chagas disease by serology if they tested positive or negative to all of the following assays: ELISA, TESA blot and IHA, respectively.
All serum samples were collected in Santa Cruz, Bolivia; 18 were positive and 12 negative to Chagas disease by serology.
Plasma samples and urine samples were collected in Cochabamba, Bolivia. Among the plasma samples, 20 were positive and 3 were negative for Chagas disease by serology. Four Chagas-positive and 2 Chagas-negative individuals provided urine samples.
Three serum samples and respective urine samples, obtained from healthy adult volunteers from Lima, Peru (non-endemic for Chagas disease), were included in the analysis. All samples from these volunteers tested negative in all serology assays and with qPCR.
The qPCR was performed using clot samples and phenol chloroform extraction [41]. A sample with a quantification cycle (Cq) equal or greater than 40 was considered qPCR negative.
Cross-reactivity of the Ag-ELISA was determined using serum and plasma samples from adults positive for malaria (n = 5), toxoplasmosis (n = 5) and leishmaniasis (n = 4). Malaria serum samples were positive for Plasmodium vivax by both thick blood smear and PCR. Toxoplasmosis plasma samples were from HIV-positive individuals who tested positive for Toxoplasma gondii by ELISA (ELISA-IBL international, Hamburg, Germany). Leishmaniasis samples (2 plasma and 2 serum) were from adult subjects infected with mucocutaneous leishmaniasis diagnosed by ELISA and PCR.
The antibodies purified from alpaca, when analyzed by SDS- PAGE and Coomassie blue stained, showed that the isotypes IgG3 and IgG2 purified have a molecular weight ranging between 100–120 kDa while the IgG1 fraction showed a molecular weight of 170 kDa. The IgG2 fraction was not completely purified, it showed traces of IgG1 (Fig 1A). The IgG purified from rabbits showed a major band at 180 kDa (Fig 1B) while the IgY of chicken showed a major band at 190 kDa (Fig 1C).
When TLA was used as antigen for western blot and tested with IgG3 from alpaca immunized with TLA, two proteins bands of 40 and 50 kDa were recognized. This pattern of bands was similar to pattern of protein bands recognize by sera from chronic patients. When IgG3 from alpaca was evaluated against membrane antigen, by western blot, it recognized a major band of 50 kDa; the sera of Chagas chronic subjects also recognized this protein band (Fig 2A). The IgG purified from rabbits and IgY purified from eggs immunized with 1F8 antigen detected a band of 25.8 kDa; which did the pre-immune antibodies not recognize (S1A and S1B Fig).
When western blot was performed using TESA, the IgG3 from alpaca, the IgG from rabbit, and the IgY from eggs immunized with TESA recognized a band of 150-160kDa concordant with the SAPA (Shed Acute Phase Antigen) pattern detected by the serum from Chagas acute patients. This protein band also coincides with one of the intense protein bands recognized by the serum of Chagas chronic patients. In addition, the IgG from rabbit and IgY from eggs detected a cross-reacting protein band of 200 kDa also recognized by their pre-immune antibodies (Fig 2B and 2C).
By western blot analysis the evaluation of IgG purified from rabbit and IgY from hens immunized with the membrane antigen, showed similar band patterns of both IgG and IgY pre-immune and post-immune (Fig 2B and 2C). In the IFA, the pre-immune and post-immune IgG from rabbit immunized with the membrane antigen shown similar fluorescent patterns (S2A Fig). In contrast, only post-immune IgY showed fluorescence in the IFA (S2B Fig). Because of these results IgG from rabbits immunized with membrane were not used for further assays.
Among the serum samples the antigen capture ELISA (Ag-ELISA) identified as positive 10 out of the 18 serology positive samples, 9 of these samples were also qPCR positive samples, while all 15 samples negative by both serology and qPCR yield a negative result on the Ag-ELISA (Table 2). Among serum samples, and considering the serology as gold standard, the sensitivity and specificity of the Ag-ELISA were 56% (10/18) and 100% (15/15) respectively.
None of the samples tested for cross-reactivity tested positive in the Ag-ELISA.
Of the 23 plasma samples only three were negative both by serology and qPCR. The Ag-ELISA was positive for all the serology positive samples (n = 20), four of the serology positive samples were qPCR negative but Ag-ELISA positive (Table 2). Overall, among the plasma samples the sensitivity and specificity of the Ag-ELISA, compared to serology were 100%.
All the urine samples that tested positive to the Ag-ELISA also tested positive on the Ag-ELISA performed in their respective plasma samples (Table 3). Three samples were urine samples collected from volunteers and sera instead of plasma was analyzed; both sera and urine tested negative to the Ag-ELISA.
Antigen detection limit was determined by two-fold dilution of the TLA antigen from 1 μg/ml to 0.975 ng/ml; the detection limit was 3.9 ng/ml.
Chagas disease remains an important health problem worldwide. Diagnosis is based on antibody detection by several methods, however antibody detection is not necessarily indicative of current infection. Thus, antigen detection might be a better determinate of current Chagas infection. To our knowledge, this study represents the first study that demonstrated the usefulness of polyclonal antibodies for the diagnosis of Chagas disease in an ELISA format. We have developed polyclonal antibodies against a variety of T. cruzi antigens in alpacas, rabbits, and hens (eggs), with adequate sensitivity and specificity as to be employed for antigen detection in clinical samples obtained from infected individuals. A combination of chicken IgY developed against T. cruzi membrane antigens (capture antibody) with alpaca polyclonal antibodies developed against excretory/secretory antigens (detection antibody) in a sandwich ELISA format was useful for the detection of circulating antigens in sera or plasma samples as well as excreted antigens present in the urine samples of infected individuals.
Development and uses of polyclonal antibodies in rabbits and hens have been widely described before with inherent variations depending on the nature of the antigen and the dose and route of administration [42, 43]. Although only two animals (a pair of hens or rabbits) were used to produce antibodies against each antigen, the immune response of each animal was similar as demonstrated by Western blot analysis. Rabbits did not produce a good antibody response when immunized with T. cruzi membrane antigens, since the Western blot and IFA analysis showed that the pre-immune response was similar to post-immune response. Probably the lack of immune response to these antigens was inherent to the rabbits and not due to the antigen preparation, since both hens and alpacas produced good antibody response. Moreover, the antibodies produced in hens (eggs) against membrane antigens were used in this study to detect T. cruzi antigens in clinical samples with high sensitivity and specificity.
The use of heavy chain antibodies (modified into nano-antibodies) has been recently explored against Trypanosomes [44, 45] suggesting that alpacas may be capable of generating an adequate immune response against complex mixtures of T. cruzi antigens. Here we have shown the usefulness of alpaca antibodies for the detection of circulating antigens in different clinical samples.
Several studies describe the use of antigen detection for the diagnosis of Chagas disease [18, 20, 46–49], but all are oriented to the detection of T. cruzi antigens in urine samples; and all demonstrate the presence of these antigens by Western blot and complicated pre-analytical handling of urine samples. Although Western blot is a standard technique that is widely used is most developed settings, it is still difficult to access in Chagas endemic settings. ELISA based diagnosis is more accessible and useful in these settings. The methodology presented here is simple, accessible and does not require the use of sophisticated laboratory equipment. Although urine samples were lyophilized in this study for practical reasons, ethanol precipitated of antigens [18] perform similarly. They do not differ in the final result from lyophilized antigens, as tested with TLA spiked urine samples and with the positive control samples. Thus, ethanol precipitation might be a good alternative to lyophilization. Only nine urine samples were analyzed in this study. Although there was a perfect correlation between Ag-ELISA in plasma and urine samples, those results need to be corroborated with a larger number of samples.
Lower levels of antigen were detected via the Ag-ELISA in serum samples as compared to the corresponding plasma samples, even in those samples that were qPCR positive. The reason for this low performance is unclear, although it is probable that the pre-treatment technique (heating of samples) might not liberate the T. cruzi antigens from the immune complexes associated to this disease [29, 50]. Immune complexes may be trapped within the fibrin clot, lowering the availability of antigen. Alternatively, since the test was performed with archived samples, the antigens may have degraded during the storage process. Further analysis is required to clarify this issue.
A high background was observed on the ELISA technique described here. The cut-off value (mean plus two standard deviations) was the highest for urine samples, and the lowest for serum samples. The high cut-off values may be a consequence of the nature of the antibodies and probably could be improved with the use of better blocking agents and techniques.
We have developed polyclonal antibodies that are useful for the detection of circulating antigens in serum, plasma, and urine samples of human subjects infected with Chagas disease using a simple ELISA technique. The antigen detection strategy described here is a promising methodology for the diagnosis of Chagas disease and, because it detects antigens, may be a good surrogate of current infection. Pretreatment of samples is straightforward, and the ELISA is a simple and more accessible technique than the currently described strategies oriented to the detection of T. cruzi antigens in urine samples.
|
10.1371/journal.pcbi.1005213 | Antioxidant Properties of Kynurenines: Density Functional Theory Calculations | Kynurenines, the main products of tryptophan catabolism, possess both prooxidant and anioxidant effects. Having multiple neuroactive properties, kynurenines are implicated in the development of neurological and cognitive disorders, such as Alzheimer's, Parkinson's, and Huntington's diseases. Autoxidation of 3-hydroxykynurenine (3HOK) and its derivatives, 3-hydroxyanthranilic acid (3HAA) and xanthommatin (XAN), leads to the hyperproduction of reactive oxygen species (ROS) which damage cell structures. At the same time, 3HOK and 3HAA have been shown to be powerful ROS scavengers. Their ability to quench free radicals is believed to result from the presence of the aromatic hydroxyl group which is able to easily abstract an electron and H-atom. In this study, the redox properties for kynurenines and several natural and synthetic antioxidants have been calculated at different levels of density functional theory in the gas phase and water solution. Hydroxyl bond dissociation enthalpy (BDE) and ionization potential (IP) for 3HOK and 3HAA appear to be lower than for xanthurenic acid (XAA), several phenolic antioxidants, and ascorbic acid. BDE and IP for the compounds with aromatic hydroxyl group are lower than for their precursors without hydroxyl group. The reaction rate for H donation to *O-atom of phenoxyl radical (Ph-O*) and methyl peroxy radical (Met-OO*) decreases in the following rankings: 3HOK ~ 3HAA > XAAOXO > XAAENOL. The enthalpy absolute value for Met-OO* addition to the aromatic ring of the antioxidant radical increases in the following rankings: 3HAA* < 3HOK* < XAAOXO* < XAAENOL*. Thus, the high free radical scavenging activity of 3HAA and 3HOK can be explained by the easiness of H-atom abstraction and transfer to O-atom of the free radical, rather than by Met-OO* addition to the kynurenine radical.
| Kynurenines, the tryptophan metabolites with multiple biological activities, regulate the production of reactive oxygen species (ROS) during several neurodegenerative diseases. Many experiments show that kynurenines can be both prooxidants and antioxidants depending on their concentration, mode of action, and cell redox potential. However, there is lack of computational studies of kynurenines properties which could help us better understand the biophysical mechanism of their antioxidant activity. We performed the computations of kynurenines' hydrogen and electron donating power, both in the gas phase and in water solution. We found that aromatic hydroxyl group facilitates hydrogen and electron abstraction by kynurenines, in agreement with experimental data and computations earlier performed for phenolic antioxidants. We revealed the correlations of kynurenines' antioxidant power with their electronic structure, charge, and surroundings. We also found that 3-hydroxykynurenine and 3-hydroxyanthranilic acid can fastly quench free radicals by hydrogen atom donation. Hence both of them are potent antioxidants. The therapeutic strategy may be to inhibit their oxidative dimerization leading to ROS production.
| The kynurenine pathway (KP), the primary route of tryptophan degradation in mammalian cells, includes kynurenine (KYN), kynurenic acid (KYNA), 3-hydroxykynurenine (3HOK), 3-hydroxyanthranilic acid (3HAA), quinolinic acid (QUIN), and other metabolites collectively called kynurenines (Fig 1).
There are multiple mechanisms of kynurenines' action on nervous system. QUIN and KYNA, the ligands of ionotropic glutamate receptors [5,6], modulate neurodegenerative processes in the brain [7]. The autoxidation of 3HOK and 3HAA leads to the hyperproduction of reactive oxygen species (ROS) which damage cellular lipids, proteins, and DNA [8–10]. Kynurenine 3-monooxygenase (KMO), an enzyme producing 3HOK from KYN, has been linked to the pathophysiology of HD by a mechanism involving ROS [11]. Accumulation of 3HOK in the central nervous system of Drosophila cardinal mutant leads to the progressive memory loss [12]. Since 3HOK is capable of auto-condensation, the eyes of this mutant, as well as the color of mammalian lens cataract [13] progressively get brown on ageing. The Drosophila eye color mutants are started to be envisioned as a therapeutic tools for HD [14].
At the same time, both 3HOK and 3HAA were shown to be powerful antioxidants scavenging peroxyl radicals [15,16]. Xanthurenic acid (XAA), a product of KYNA hydroxylation, has similar antioxidative properties, but its rate of interaction with free radicals is slower [15]. Tryptophan and its catabolites without aromatic hydroxyl group, such as kynurenine (KYN), KYNA, and anthranilic acid (AA) have no effect on peroxy-mediated oxidation. Thus, phenolic hydroxyl group is important for antioxidant activity of kynurenines. Antioxidants are supposed to beneficially interfere with diseases-related oxidative stress, however, the interplay of endogenous and exogenous antioxidants with the overall redox system is far from clear [17].
Phenolic compounds suppress lipid peroxidation due to their ability to react with free radicals at a faster rate than with the substrate [18,19]. There are two main pathways of phenolic antioxidants quenching free radicals: electron transfer and H-atom transfer. H-atom easily abstracted from the aromatic OH-group interacts with peroxyl radical ROO* produced during lipid peroxidation and breaks the chain reaction:
Ar−OH+ROO*→Ar−O*+ROOH
(1)
There are two pathways of hydrogen transfer: hydrogen atom transfer (HAT) and proton-coupled electron transfer (PCET) [20]. HAT is preferable when electron density of singly occupied molecular orbital (SOMO) in the transition structure (TS) lies along the same line as the O…H…O bond and H is transferred between the oxygens as a whole particle. PCET is preferable when SOMO is orthogonal to O…H…O bond, as in phenoxyl-phenol complex, and proton is transferred between oxygen σ lone pairs forming hydrogen bonds with them, while the electron is transferred between oxygen π-orbitals.
Also, phenolic antioxidant radicals are able to quench peroxyl radical via its addition to the aromatic ring at ortho- or para-position. In order to trap the radical and not to react with hydrocarbon R-H substrate, an antioxidant should have less value for the homolytic O-H bond dissociation enthalpy (BDE) than ROO-H and R-H. Moreover, antioxidant radical should be kinetically stable to prevent its reaction with substrate [21,19]. Thus, the antioxidant power is not an absolute property of Ar-OH, but depends on the substrate which should be protected.
The toxicity of 3HOK depends mainly on the products of its oxidative dimerization, such as hydrogen peroxide, xanthommatin (XAN), 4,6-dihydroxyquinolinequinonecarboxylic acid (DHQCA), their active free radical forms, and o-aminoquinone [22]. Ommochromes XAN and dihydroxanthommatine (DXAN), the brown eye pigments, easily transform into each other under physiological conditions [23,24]. DXAN is synthesized from 3HOK by phenoxazinone synthetase (PHS)–the process disturbed by the cardinal mutation [23]. PHS catalyzes two consecutive abstractions of H-atoms from the hydroxyl group of o-aminophenols, 3HOK or 3HAA, followed by their non-enzymatic condensation to phenoxazinone [25]. The formation of ommochromes can also result from non-enzymatic oxidation of 3HOK [26]. High concentration of 3HOK catabolite hydrogen peroxide induces apoptotic cell death in neuronal cell cultures [27]. 3HOK and 3HAA generate superoxide anion and hydrogen peroxide in the presence of copper–the process leading to the formation of a quinoneimine structure [28]. Both amino and hydroxyl aromatic groups are important for lowering 3HOK and 3HAA oxidation potential. Initially, they can be two-electron donors with antioxidant activity, but their quinoneimine products are highly reactive and damage cell structures. Pro- and antioxidant power of o-aminophenols depends on the whole activity of the redox systems in cell [29].
Other kynurenine metabolites also possess pro- and antioxidant activity [4]. In particular, KYNA is able to scavenge hydroxyl radicals, superoxide anion radicals, and peroxynitrite, decreasing lipid peroxidation and ROS formation [30]. QUIN affects the ROS level only together with iron ions; the pro- and antioxidant effects of QUIN are concentration-dependent [31]. Free radicals scavenging mechanisms shown for non-o-aminophenol kynurenines include electron transfer, metal ion chelation, destruction of carbon skeleton, and radical addition to the aromatic ring [4].
Whereas experimental data regarding chemical and physiological properties of kynurenines are abundant and diverse, there are few computational studies on kynurenines. Quantum chemical calculations could provide a better understanding of the mechanisms of kynurenines' antioxidant activity. In this study, the redox properties of kynurenines and several synthetic phenolic antioxidants were investigated computationally using density functional theory (DFT) approach. The validity of B3LYP methods to model phenolic antioxidants and free radical reactions has already been proved [18,32]. The methodology was similar to that of [18,33]: the energies of frontier highest occupied and lowest unoccupied molecular orbitals (EHOMO, ELUMO), phenolic O-H bond dissociation enthalpy (BDE), and ionization potential (IP) were calculated and compared for structures fully optimized in the gas phase. We also studied the influence of water solvation on the chemical properties of antioxidants. Finally, we modeled the kinetic behavior of hydroxykynurenines interaction with phenoxyl and peroxyl radicals.
Optimal geometries for kynurenines and synthetic antioxidants with substituted phenolic groups were calculated at different levels (Table 1). Six compounds with experimentally known BDE values are used as standards for the estimation of the validity of computational methods. Despite high diversity of chemical structures, Pearson correlation coefficient R is high for level II (0.870 and 0.867 for BDE and BDECOR, respectively; p<0.05) and III (0.865, 0.863; p<0.05), being less for level I (S1 Table) (0.717, 0.710; p > 0.1). Total spin <S2> shows small spin contamination ranging from 0.75 to 0.80 for all free radicals (II, III), being abnormally high for some radicals calculated at level I. Thus, (I) computational data were omitted from further analysis. BDE and BDECOR (II, III) for phenol are greatly higher than the experimental value and the value previously calculated at the same level of theory; the cause is explained in Methods section. With the exception of phenol, the correlation of BDE/BDECOR with experimental values is very strong (R = 0.959 and 0.974 for level II and III, respectively; p < 0.05, n = 5). The goal of this study was not the precise calculations of energy values, but rather the comparison of such values for different antioxidants. Thus, DFT calculations at level II or III can be used to predict the relative antioxidant power of the studied compounds.
The values of EHOMO, ELUMO, BDE, and BDECOR at level II are highly correlated with those calculated at level III (R = 0.98–1.00) In general, level III gives a slightly higher BDE/BDECOR than level II. BDE/BDECOR calculated at B3LYP and HCTH/407 levels of DFT (basis set II) are strongly correlated (R = 0.999 and 0.998, respectively; p < 0.05, n = 16, without phenol). HCTH/407 gives slightly lower values of BDE than B3LYP (S1 Table; ΔE = 2.252±0.751 kcal/mol). For phenol, BDEHCTH/407 is 84.702 kcal/mol, which is much closer to the experimental value. Thus, both functionals can be used to estimate BDEs for kynurenines and phenolic antioxidants.
The rankings for O-H homolytical BDECOR are nearly the same at levels II and III. O-H bond is the strongest in water and the weakest in negatively charged 3HAACO2-. BDECOR values for 3HOK and 3HAA are close to that for 2-aminophenol, their structural precursor. 2-aminophenol is an antioxidant with a large decrease in the O-H BDE compared to phenol [38]. L-3HOK and D-3HOK optical isomers have almost equal BDE values. 3HOK and 3HAA are characterized by the decreased energies of H abstraction compared to phenol and its derivatives DIBP and DTBP, both native and modified by propenoic acid (DIBA, DTBA). Total energy for XAA oxo form is lower by 7.4 kcal/mol than for enol form (level III); therefore, we used the oxo form in the majority of calculations. XAAOXO is close to phenolic antioxidants in its H donating properties.
B3LYP, as well as most DFT methods, is known to give EHOMO and ELUMO in a very poor agreement with experiment, significantly underestimating H-L gap. Using of tuned range-separated hybrid functionals can solve this problem [39,40]. We have computed EHOMO, ELUMO, H-L gap, and IP for five compounds optimized at level III, B3LYP (L-3HOK, 3-HAA, XAAOXO, 2-aminophenol, and DTBP) using tuned LC-BLYP range-separated functional. Indeed, LC-BLYP gives significantly higher absolute values for H-L gap (ΔELC-BLYP–B3LYP = -69.849±8.135 kcal/mol), and IP values are close to -EHOMO (ΔE-HOMO–IP = -3.333 ±2.929 kcal/mol). The optimal range-separation parameter μ values are close for four aminophenols (~0.20) and differ from that for phenolic antioxidant DTBP. At the same time, EHOMO, ELUMO, and H-L gap values calculated by III(LC-BLYP) and III(B3LYP) are highly correlated (R = 0.911, 0.989 and 0.955, respectively), thereby the rankings for electron donating power are virtually the same in both cases. XAAOXO has the highest values for IP and -EHOMO, whereas L-3HOK is the best electron donor among the uncharged hydroxykynurenines.
There is a moderate negative correlation between EHOMO and BDE/BDECOR (levels II, III) (S2 Table). The correlation between H-L gap and BDE/BDECOR is even stronger. Hence the ability of O-H bond homolytical dissociation tends to increase along with the lowering of EHOMO and H-L gap absolute values. The correlation between ELUMO and BDE/BDECOR is not significant at all levels. For the compounds with an ionized group, such as 3HAACO2-, XAACO2-, and L3HKNH3+, EHOMO significantly differ from those of uncharged compounds. 3HAACO2- and XAAOXO/CO2- are more powerful H donors than the uncharged forms. On the contrary, protonation of NH2 group in L-3HOK phenolic ring significantly complicates O-H dissociation. This is in agreement with the fact that electron-donating groups reduce O-H BDE, thus enhancing antioxidant activity, whereas electron-withdrawing substitutions raise it [41,42,18]. DIBP and DTBP, the substances with skeletal isomerism, have similar BDEs. However, for each of them, BDE is closer to that of its propenoic derivative than to BDE of its isomer. Hence the side chain isomerism significantly affects the H- donating properties of phenolic group. DXAN has the least stable O-H bond among the uncharged compounds, making it a potent anioxidant with the high H donating ability.
In order to check the possible effect of the basis set superposition error (BSSE) on BDE, BSSE correction was performed for phenol, DIBP, and DXAN with a small, intermediate, and large hydrocarbon moiety of a radical. The values of BSSE (III) are -0.722, -0.832 and -1.45 kcal/mol, respectively, being small and similar in all compounds. The decrease of BDE for bulky antioxidants cannot be explained by the growth of BSSE.
The geometry of frontier molecular orbitals and spin-orbits was calculated at level III for hydroxykynurenines and their precursors, as well as for their derivatives without an aromatic hydroxyl group. The highest occupied molecular orbital (HOMO) of phenolic antioxidants and kynurenines is localized mainly on the phenolic ring. HOMO is divided into two parts: the first part occupies phenolic OH group and three approximate C atoms, and the second part occupies the opposite two or three C atoms (Fig 2). HOMO also occupies unsaturated and polar groups attached to the phenolic ring, such as the aromatic NH2 group of L-3HOK and 3HAA, which HOMO's and spin-orbit's geometry is virtually the same as that for 2-aminophenol. Together with the aromatic rings, OH and NH2 groups form a π-conjugated systems known to decrease IP [43]. DXAN has the largest conjugated system allocated mainly to phenoxazinone structure, which possibly facilitates H-atom and electron abstraction. For the ionized compounds, HOMO is moved from the aromatic hydroxyl group to the charged group.
BDE, IP and frontier orbital energies for compounds optimized at level III were calculated at level IV (B3LYP/6-311+(O)+G(d)) in the gas phase and water solution (Table 2, Fig 3A and 3B). A moderate negative correlation between BDE and EHOMO/H-L gap values was observed, as well as for levels I-III, both in the gas phase and in water solution (S2 Table, second part). A strong correlation between adaibatic IP and -EHOMO or the so called vertical IP can be seen. This is in accordance with Koopmans' theorem, applicable in high approximation for outer valence Kohn-Sham orbitals [44]. The difference between IP and -EHOMO is 36.7±3.3 kcal/mol in the gas phase, in agreement with the fact that B3LYP underestimates the absolute values for EHOMO [39]. In water solution, IP becomes slightly lower than the negative of EHOMO. BDE is positively correlated with IP; thereby the electron and H donating capacities of the studied compounds are interrelated.
BDE of aromatic antioxidants is strongly correlated with standard deviation of Mulliken spin density (δSD) on radical (RGAS = 0.879, RWATER = 0.917; p < 0.05, n = 14) and spin density (SD) on radical O* atom (RGAS = 0.910, RWATER = 0.951; p = 0.05, n = 14) after H abstraction. BDEs for phenol and DTBP were not considered due to significant deviations from experimental values. Also, BDE is strongly correlated with SD on radical CPARA aromatic atom after H abstraction (RGAS = 0.849, RWATER = 0.876; p < 0.05, n = 13, without KYNAENOL which has N instead CPARA).
There is no significant correlation between IP and δSD for kynurenine radicals after single electron abstraction. Thus, electron delocalization on kynurenines seems to be more important for H-atom donation activity than for the electron donation activity.
OH group bonded to the aromatic ring significantly increases the ability of kynurenines to donate H-atom and electron. L-KYN C3-H BDE is much higher than L-3HOK O3-H BDE: the difference is 40.4 kcal/mol in the gas phase and 40.3 kcal/mol in water solution (level V: B3LYP/6-311++(d,p)). IP value is lower for XAAOXO, L-3HOK, and 3HAA than for KYNAOXO, L-KYN, and AA, respectively (ΔIPGAS = 6.4±2.9 kcal/mol, ΔIPWATER = 5.5±0.6 kcal/mol). However, the relative IP rankings are the same for compounds with and without OH group: KYNA has higher IP compared to L-KYN and AA, as well as XAA compared to L-3HOK and 3HAA. This is in agreement with experimental data: the electrochemical potential Epa for kynurenines has been shown to decrease in the following rankings: KYNA > KYN > AA > 3HOK > 3HAA [29]. There is a strong positive correlation between experimental Epa and IP calculated for non-ionized compounds in water solution at level IV (R = 0.924; p < 0.05, n = 5). For compounds with an ionized CO2 group, the correlation is not significant (R = 0.804; p > 0.1, n = 5), probably, due to the small sampling. QUIN is the least powerful electron donor among the uncharged kynurenines. IPGAS for XAN is close to that for 3HAA, hence XAN easily abstracts electron, but not H-atom. This possibly makes it a prooxidant with toxic effects [22].
The electron-donating substituents are known to decrease IP and to increase the antioxidant activity [33]. In general, compounds with ionized CO2 group have lower IP and higher EHOMO values than their neutral forms (ΔIPGAS = -88.5±11.5 kcal/mol, ΔIPWATER = -8.7±9.5 kcal/mol, without KYNAOXO; ΔEHOMO/GAS = 88.6±8.4 kcal/mol; ΔEHOMO/WATER = 7.4±1.6 kcal/mol). KYNAOXO in water solution has lower IP than KYNAOXO/CO2. Probably, geometry optimization of charged compounds in the gas phase leads to some distortions in KYNAOXO/CO2 structure. In water solution, IP becomes lower for the majority of compounds and higher for the anionic forms. This seems to result from a high dielectric capacity which decreases electrostatic interactions, stabilizes anions, and diminishes electron attraction to cations and neutral molecules. The change of gas–water -EHOMO and -ELUMO is correlated with the change of IP. BDE for XAAOXO/CO2- becomes higher than that for 3HAA and only 2 kcal/mol less than that for XAAOXO. The same trend is observed for 3HAA/3HAACO2- and L-3HOK/L-3HOKNH3+. Thus, water solution significantly diminishes the influence of charged groups on BDEs and IPs.
There are different pathways for ROS inactivation by antioxidants [19]. Most likely, kynurenines quench radicals by donating aromatic hydroxyl H-atom to radical *O-group. We have computationally studied the kinetics of this process for the complexes of four hydroxykynurenines, 3HAA, L-3HOK, XAAOXO, and XAAENOL, with phenoxyl radical (Ph-O*) and methyl peroxy radical (Met-OO*). BDE difference for Met-OO* and buthyl peroxy radical is less than 0.7 kcal/mol (levels II, III), hence Met-OO* can be used instead of the radicals with long aliphatic chain to simplify calculations. Ph-O*–DTBP and Ph-O*–DTBA complexes have been also calculated, as well as Met-OO* complex with XAA in ionized form. TSs for reaction pathways were located at level II. Reagent and product complex structures are in good agreement with the results of IRC calculations (RMSD = 0.026±0.015 Å for all complexes and 0.019±0.008 Å for kynurenines' complexes).
The values for the reaction rate and height of activation barrier were calculated in the gas phase and water solution (Table 3). k(T) values are significantly higher than those experimentally shown for phenolic compounds with BDE values of 70–80 kcal/mol, which are about 104−107 M-1s-1 [19]. This fits the fact that B3LYP underestimates the reaction barrier heights, whereas functional XYG3 is almost as accurate, as the highly precise CCSD(T) method [45].
It is rather difficult to calculate the exact value of the reaction rate, as multiple factors should be considered, and appropriate DFT level should be used [46]. However, the location of TS point calculated for 3HAA–Met-OO* and XAAOXO−Met-OO* by B3LYP (level II) is similar to that calculated by XYG3 (level V) (S4 Fig). Thus, even B3LYP with the relatively small basis set II correctly describes the geometry of TS structure. ΔETS-R (XYG3) for XAAOXO complex is higher than that for 3HAA complex.
For both Ph-O* and Met-OO*, k(T) increases in the following rankings: XAAENOL < XAAOXO <3HAA ~ L-3HOK (Table 3, Fig 3(C)); the same rankings applies to -ΔEP-R.
The structures of radical complexes with 3HAA, L-3HOK, and XAAOXO are very similar (Fig 4, Table 4). For kynurenines in complex with Ph-O*, aromatic rings of reagents and products form the plane angle of ~50–70°. The geometry of Ph-O*–DTBP is very different: aromatic rings are nearly perpendicular in reagent and product complexes. Ph-O*–DTBP and Ph-O*–DTBA complexes are rather similar. For 3HAA, L-3HOK, and XAAOXO in complex with Met-OO*, the radical rotates in space along with the attachment of H-atom, so O-O* and C-O bonds in Met-OOH become nearly perpendicular to those in reagent complexes. The direction of Met-OO* rotation is different in complexes with XAAENOL and XAAOXO/CO2. In all cases, O…H…O bond significantly shortens upon the TS formation.
The influence of solvent and partial charges' distribution on antioxidant activity depends on whether HAT or PCET is the dominant mechanism of H transfer. The increase in H-atom charge in the TS compared to the parent antioxidant is specific for PCET [20]. The interaction of phenolic antioxidants with tert-buthyl-peroxy radical is known to occur via PCET [47]. PCET-TS is stabilized by the enhanced spin density (SD) and electron density on radical O2 and O3 atoms. Thereby (O3+O2-O1) negative charge and Δ(O3+O2-O1)TS-R negative charge correlate with the reaction rate [47]. In our study, positive charge on H-atom increases in all TSs (ΔQ(H) > 0), and the negative charge on O atoms in the gas phase moves towards the free radical (Δ(dQ)TS-R < 0) (Table 5). In the gas phase, there is a strong correlation between ΔETS-R and SD on O1 and CPARA atoms of antioxidants. Hence the high SD on these atoms decreases the reaction rate.
For Ph-O* complexes in water solution, the decrease of the negative charge on radical O atoms (Δ(dQ)TS-R) correlates with the growth of ΔETS-R, as it is typical for PCET. There is a negative correlation between ΔETS-R and ETS-SOMO. Thus, ESOMO may serve to predict the reaction rate, as shown by Nikolic [47].
The geometry of TS SOMO and spin-orbit on O1 and O2 differs from both classical σ- and π-orbitals: p-orbitals on O atoms form a sharp angle projected to plane passing through H atom perpendicular to O…H…O bond (Fig 5). In Ph-O*–kynurenines' complexes, p-orbitals are nearly parallel to this plane and perpendicular to O…H…O bond. In Ph-O*–DTBP complex, O1 and O2 protrude parts of the electron clouds towards H, and in Met-OO*–kynurenines' complexes, the angle between O…H…O and O1 p-orbital is close to 45°. Hence kynurenine's SOMO in Ph-O* complexes is closer to π-orbital than in Met-OO* complexes.
Partial charges Q and ΔQ on H are higher, and the negative charge displacement to O1 is lower for Ph-O* complexes than for Met-OO* complexes. Thus, PCET seems to be more preferable for kynurenines' reaction with Ph-O* than for their reaction with Met-OO*. HAT may also occur in both cases, however, SOMO geometry and charges distribution character indicate that it is not the chief mechanism of H transfer for the studied complexes.
Another possible way of free radical quenching is its addition to the aromatic ring of the antioxidant radical. We have modeled the products of Met-OO* addition to the aromatic ring of phenoxyl and kynurenines radicals at para-position relative to O* atom (Fig 6, Table 6). The orientation of side chain Met-OO group varies, being closer for the different forms of XAA than for the different antioxidants.
In the gas phase, all reactions are thermodynamically favorable (ΔEP-R/COR < 0), in contrast to H abstraction. In water solution, radical addition to 3HAA and L-3HOK radicals is slightly unfavorable. The rankings of -ΔEP-R and -ΔEP-R/COR are the same at all levels: 3HAA* < L-3HOK* < XAAOXO/CO2-* < XAAOXO* < DTBP* < XAAENOL* < Ph-O*. It is reverse to the rankings of -ΔEP-R and k(T) for H-atom donation: the affinity to Met-OO* is minimal for 3HAA* and maximal for phenoxyl radical. XAAOXO/CO2-* is less active than XAAOXO* and more active than L-3HOK*, both in the gas phase and in water solution. Thus, high Met-OO* scavenging activity of L-3HOK and 3-HAA is unlikely to be explained by Met-OO* addition to the aromatic rings of kynurenine radicals.
The antioxidant power of a substance depends not only on its chemical properties, but also on its ability to penetrate into the surroundings where it displays its antioxidant activity. To inhibit lipid peroxidation, a substance should have high lipophilicity. It can be measured as a logP value, where P is the octanol-water partition coefficient [48]. We used the Molinspiration method of logP calculation reported to be robust and precious.
Among the antioxidants studied, substituted phenols, such as DTBP and DTBA, have maximal lipophilicity, whereas the kynurenines' ions have higher water solubility compared to ASC (Table 7). Lipophilicity decreases in the following rankings: AA > 3HAA> XAAOXO > QUIN > L-3HOK; the rankings are the same for kynurenines' carboxylic anions. Hence XAA should penetrate through lipid bilayer better than 3HOK and 3HAA. This fact does not fit with the lower rate of XAA reaction with peroxy radicals, which is rather explained by the higher rate of H donation.
Topological polar surface area (TPSA) is a molecular descriptor numerically close to PSA which can be used for the prediction of passive transport through membranes in intestines and blood-brain barrier [49]. Drugs that penetrate the brain by passive absorption typically have PSA < 70 Å2, while the most non-CNS active drugs have much larger PSA values up to 120 Å2 [50]. In our study, TPSA is less than 120 Å2 for all compounds except L-3HOK, so they are more or less capable of penetrating passively through the plasma membrane. TPSA is minimal for phenolic antioxidants, which should be easily absorbed in intestines and penetrate into the brain. It is significantly higher for 3HOK (> 120 Å2) than for 3HAA (< 90 Å2), while XAA has the intermediate TPSA. Hence in the case of absence of specific carriers 3HAA should more actively penetrate through lipid bilayer than 3HOK.
A great need in studies of biochemical properties of KP is promoted by the fact that this very pathway plays an overwhelming role in physiology and pathology. Dysregulation of this pathway, resulting in hyper- or hypofunction of active metabolites, is associated with neurodegeneration and other disorders, such as depression and schizophrenia [51], diabetes mellitus [52,53], attention-deficit hyperactivity disorder [54], and cataract [13]. Some KP metabolites are neuroactive, while others are molecules with prooxidant and antioxidant properties [3]. Therefore, it is necessary to understand the molecular and biophysical mechanisms of kynurenines' activity to elaborate the strategy of disorders' prevention and therapy.
In this study, we investigated the antioxidant activity of kynurenines, namely their ability to donate electron and H atom. The hydroxyl group BDE and adiabatic IP are the most important determinants for the radical scavenging activity of substituted phenols [55]. According to our data, the antioxidant properties of 3HOK and 3HAA are determined by their 2-aminophenolic moiety. For the uncharged hydroxykynurenines, BDE and IP are maximal for KYNAENOL and miminal for DXAN, both in the gas phase and in water solution. 3HOK and 3HAA have lower BDE and IP than XAA, ascorbic acid, and some phenolic antioxidants, such as DTBP and DTBA. Aromatic OH group diminishes IP values for 3HOK, 3HAA, and XAA relative to KYN, AA, and KYNA. Our results confirm the correlation between BDE and IP also shown by Borges [56]. Negatively charged carboxylic group significantly diminishes BDE and IP values, while positively charged amino group enhances them. This phenomenon can be explained by electron-donating and withdrawing effects of substituents [18]. The effects of charged groups are significantly more pronounced in the gas phase than in water solution. Basis set and the type of density functional had a little effect on the rankings for BDE values of kynurenines.
Adiabatic IP strongly correlates with -EHOMO, confirming that Koopmans' theorem can be used to calculate IP at DFT level [44]. However, B3LYP significantly underestimates the absolute values of EHOMO. We have used the tuned LC-BLYP range-separated functional to compute EHOMO, ELUMO, H-L gap, and IP for several antioxidants, including three hydroxykynurenines. The tuned LC-BLYP gives significantly higher absolute values for EHOMO and H-L gap, also, there is a minor difference between -EHOMO and adiabatic IP. However, the rankings for kynurenines' EHOMO values are the same as for those calculated using B3LYP. Also, BDEs calculated with B3LYP and HCTH/407 are highly correlated. B3LYP is significantly faster than the high quality functionals partly based on perturbation theory, such as XYG3. B3LYP contains less empirical parameters than HCTH/407, thereby it seems to be more universal. B3LYP has been successfully used to model both thermodynamic and kinetic properties of free radicals [18,32,33]. Thus, we used B3LYP in the majority of our calculations. At the same time, using LC-BLYP and other range-separated functionals may be favorable to predict the exact values for frontier orbital energies.
High radical stability and even spin distribution are among the factors predisposing low BDE and IP values [55]. Conjugated bonds system facilitates electron delocalization after HAT or single electron transfer (SET). Standard deviation of SD in kynurenine radicals is correlated with BDE; however, there is no significant correlation between SD and IP. BDE for kynurenines is correlated with SD on O* and CPARA atoms. SD on antioxidant O* atom in TS complex with Met-OO* is also strongly correlated with the height of the activation barrier. The same is true for OPARA atom in phenolic antioxidants [47].
The rate of H-atom donation to phenoxyl and methyl peroxy radicals is correlated with BDE: 3HAA and 3HOK are more active radical scavengers than XAA. Likewise, for phenolic compounds donating H to hydroxyl radical, the rate constant is negatively correlated with O-H bond straight, IP, and SET enthalpy [57]. The rankings for free energies of radical addition to kynurenine radicals in para-position relative to OH group is reverse: 3HAA* radical has the lowest affinity to Met-OO*. Thus, high antioxidant activity of 3HAA and 3HOK relative to XAA [15] rather can be explained by their lower BDEs and higher rates of H-atom donation to peroxy radical. PCET seems to be the chief mechanism for H donation by kynurenines to phenoxyl radical and, probably, to Met-OO* radical. Studies of oxidations of O-H bond usually invoke stepwise oxidation, where the positive and negative charges are transferred separately. Here, there may be a complex dependence of BDE and k(T) on solvent, biochemical surroundings, and pH [58].
Low BDE and IP values are not sufficient for Ant-OH to be a powerful antioxidant without toxic side effects. Some of the necessary conditions include: 1) O2 should not abstract H from Ant-OH; 2) Ant-OH should react with ROO* much faster than ROO* with R-H; 3) Ant-O* should not abstract H from R-H at an appreciable rate; 4) Ant* should not react with O2 to produce AOO*; 5) Ant-OH and its products should not be toxic [21]. Though we did not concentrate on the study of these properties, our calculations performed for 3HAA at level II have shown that:
Hence O2 can abstract H from 3HAA, but the reaction is dramatically slower than H abstraction by Met-OO* (2.2 x 1011 M-1s-1; Table 3). The aliphatic ethane interacts with Met-OO* faster than with 3HAA, and 3HAA interacts with Met-OO* dramatically faster than with ethane. Only the reaction of 3HAA with Met-OO* is thermodynamically favorable, so O2 or free radicals must be in high concentration to hinder the protective action of 3HAA. Therefore, it seems to be a potent antioxidant, as well as 3HOK.
At the same time, the ability of 3HAA and 3HOK to form dimers leads to the production of toxic free radicals which damage the cell [23,26,22]. 3HAA undergoes three successive one-electron oxidative reactions: 1. conversion to semiquinoneimine (hydroxyl H abstraction); 2. conversion to quinoneimine (amine H abstraction); 3. two quinoneimine molecules condensation to cinnabarinic acid [59,60]. The rate of 3HAA oxidation increases exponentially with increasing pH [59]. This corresponds to our data that BDE decreases and k(T) for H-atom donation increases for compounds with ionized carboxylic group. 3HOK autoxidation is similar to that of 3HAA [29]. BDE for 3HOK N-H is significantly higher than for 3HOK O-H, being 99.5 and 77.2 kcal/mol, respectively (level III). Both of them are smaller for semiquinoneimine, becoming 91.8 and 69.4 kcal/mol after the other H-atom abstraction. Hence the first stage of oxidation facilitates the second one. o-Quinoneimine which is synthesized at the second stage may be responsible for the prooxidant effects of 3HOK and 3HAA [29]. The enzymatic oxidation of o-aminophenols leads to the concomitant reduction of oxygen to water [25]. Non-enzymatic oxidation produces the toxic reactive forms of oxygen. Therefore, it may be therapeutically important to enhance the antioxidant power of hydroxykynurenines by inhibiting their non-enzymatic dimerization and/or stimulating the enzymatic dimerization.
The inhibition of tryptophan 2,3-dioxygenase, the key enzyme of KP, is neuroprotective in Drosophila huntingtin (htt) mutant. Feeding flies by 3HOK alone, in the absence of mutant HTT, did not cause neurodegeneration [14]. Thus, the high level of 3HOK is toxic, yet, it may be not sufficient for neurodegeneration which also requires the additional factors, such as the lack of neuroprotectant KYNA. Both 3HOK and 3HAA inhibit the spontaneous lipid peroxidation in the brain [61]. The dual redox activity of 3HOK makes it prooxidant at low concentrations (5–20 μM) and antioxidant at higher concentrations (100 μM) in the rat striatum slices. 3HOK seems to be a redox modulatory molecule which stimulate the increase in glutathione reductase and glutathione S-transferase activities [62]. Interferon-γ induces TRP degradation along the KYN pathway in mononuclear blood cells and inhibits the oxidation of low density lipoprotein (LDL). 3HAA inhibits LDL oxidation in submicromolar concentrations, probably being a catalyst for the other antioxidants [63]. It is a highly efficient coantioxidant for plasma lipid peroxidation which can be initiated by α-tocopherol radical (α-TO*).
3HAA in low concentration (5μM) inhibits α-TO* production and accumulation of lipid peroxides. 3HOK inhibitory efficacy is the same as for 3HAA, but AA lacking the phenolic group can not reduce α-TO* [16]. This fits with low BDE for kynurenines having phenolic group. Monocytes in human blood can release 3HAA in concentration up to 30 μM [63]. Thus, 3HOK and 3HAA display the antioxidant activity under physiological conditions, not only in the brain, but also in blood plasma regulating the process of atherogenesis. In contrast to 3HAA, the antioxidant properties of XAA are in relation to its ability to chelate the transition metals which induce LDL oxidation [64]. Thus, kynurenines' action on redox conditions and physiological processes depends on their level in organism. The lack of kynurenines in Drosophila mutant vermilion, as well as the excess of 3HOK in cardinal, leads to the progressive loss of 3 h memory performance under conditioned courtship suppression paradigm [65,12]. The lack of kynurenines redox activity might partially cause these effects.
Not only BDE and IP define the antioxidant power of substances, but also their ability to pass through the biological barriers, mainly the lipid bilayers. The lipophilicity of kynurenines is low, compared to phenolic antioxidants, due to their polar and charged groups. Hence they should rather act in water environment than in membrane. Besides, their surfaces are quite large, that should hamper their penetration through intestine and blood-brain barriers. Indeed, 3HAA, KYNA, and QUIN poorly cross the blood-brain barrier by passive diffusion, but KYN and 3HOK are taken up into the brain by a large neutral amino acid carrier [66]. AA easily penetrates into the brain by passive diffusion that can be explained by its high logP and low TPSA values. Kynurenine pathway enzymes in the brain are preferentially localized in astrocytes and microglia; however, the cerebral pathway is driven mainly by blood-borne KYN [2]. Thus, 3HOK, KYN, and AA may play an important role in the brain both as prooxidants and antioxidants.
In our study, we did not consider several important factors affecting the antioxidant power of kynurenines, such as: 1) thermodynamics and kinetics of 3HOK and 3HAA dimerization; 2) energy and rate of proton abstraction from antioxidant OH group; 3) interaction between solvent and OH group of kynurenines; 4) hydrogen bond formation between functional kynurenine groups; 5) steric effects of side-chain groups on free energy and rate of kynurenine interaction with radicals, etc. Other functional groups of kynurenines can also donate H-atom, such as the 3HOK aromatic NH2 group, which BDE was shown to be significantly higher than that for OH group. It would be interesting to evaluate the activity of kynurenines and phenolic antioxidants in their native surroundings, such as lipid bilayer, affecting the dielectric capacity and hydrophobic interactions. Consideration of these factors is a task for the future.
The structures of 2-aminophenol, anthranilic acid (AA), ascorbic acid (ASC), kynurenic acid (KYNA), L-kynurenine (L-KYN), D-3-hydroxykynurenine (D-3HOK), L-3-hydroxykynurenine (L-3HOK), quinolinic acid (QUIN), 2,6-di-tert-butylphenyl-4-hydroxymetylphenol, xanthommatin (XAN), dihydroxanthommatin (DXAN), and xanthurenic acid (XAA) were taken from PubChem Compound database [67]. The structures of phenol, 2,6-di-isobutylphenol (DIBP), 2,6-di-tert-butylphenol (DTBP), β-(4-hydroxy-3,5-di-isobutylphenyl) propenoic acid (DIBA), β-(4-hydroxy-3,5-di-tert-butylphenyl) propenoic acid (DTBA), and 3-hydroxyanthranilic acid (3HAA) were constructed on the base of PubChem structures using Vega ZZ 3.0.3 [68]. The systematic conformational search of low-energy geometry for the constructed structures was performed using Avogadro [69].
The ionic forms for kynurenines with α-carboxylic group (total charge -1) were modeled as well as the uncharged forms. The major forms for 3HAA and XAA at physiological pH (7.4) are the forms with the ionized carboxylic group, while KYN and 3HOK are mainly in zwitterionic form with ionized α-amino and α-carboxylic group [70].
All quantum chemical calculations were performed using Firefly 8.1.0 partially based on the GAMESS (US) [71] source code. Firefly 8.1.0 was kindly provided by Alex A. Granovsky [72]. The geometries of molecular structures with neutral total charges were fully optimized using density functional theory (DFT) at 6-31G(d) level (I), B3LYP/6-31G(d) level (II), and B3LYP/6-311G(d,p) level (III) [73–75]. B3LYP1 version of B3LYP was used. Highly parameterized functional HCTH/407 [76] was also used to calculate BDE values for compounds at level II. Closed shell configurations were calculated with restricted Hartree-Fock or DFT methods; open shell configurations were calculated with unrestricted Hartree-Fock or DFT methods. All closed shell molecules were calculated in a singlet state, whereas doublet state was used for free radicals. The symmetry point group was set as C1 for all compounds. Hessian matrix, vibrational frequencies, and thermal corrections to the enthalpy were calculated with the same methods. The enthalpies and free energies were obtained from the vibrational frequency calculations at 298.15 K, using unscaled frequencies. In order to calculate the adiabatic ionization potential (IP), cation-radical forms of corresponding molecules were fully optimized at level III. BSSE correction [77] was performed for several compounds at level III. The nature of all stationary points was determined by evaluating the vibrational frequencies. Standard deviation of Mulliken spin density (δSD) was used as an estimator of electron delocalization on the radicals.
The following energy parameters were estimated:
where ECAT is the energy of cation radical after single electron abstraction (or the neutral form for compounds with ionized carboxylic group), EW is the total energy of the whole molecule.
BDEs for methane, water, phenol, 2-aminophenol, water-soluble antioxidant ASC (uncharged form), and phenolic antioxidant DTBP were used as standards and reference points to estimate the relative activities of antioxidants. For symmetric phenol and DTBP radicals, there were significant deviations of BDE from the experimental values. To exclude the possible artefacts, BDE was also calculated for several structural analogues of DTBP–DTBA, DIBP, and DIBA, which are believed to have similar BDE values.
Zwitterionic form is not stable in the gas phase, therefore, the optimization of KYN and 3HOK in the neutral form was performed. To check the influence of positively charged group on BDEs and IPs, calculations were performed for L-3HOK with aromatic NH3+ group (total charge +1). 3HAA and AA cations with ionized carboxylic group are not stable in the gas phase, therefore, their optimization was performed in water solution at level IV (see below), without cavitation, dispersion and repulsion free energies.
NWChem software [78] was used to calculate EHOMO, ELUMO, H-L gap, and IP with the help of the tuned range-separated hybrid functional LC-BLYP for five compounds optimized with B3LYP (III). The tuning of optimal range-separation parameter μ was done as in [40]: the single point energies were calculated using basis set III for antioxidant's cation, anion, and neutral form for different values of μ ranging from 0.05 to 0.9 with increments of 0.05, and then the optimal parameter was obtained by minimizing the following function:
J2(μ)=[EμHOMO(N)+IPμ(N)]2+[EμHOMO(N+1)+IPμ(N+1)]2
(6)
where N is the number of electrons in antioxidant.
For XAAOXO, J2(0.1) and J2 (0.15) were obtained by spline interpolation due to the problem with DFT convergence. The curves for J2(μ) are shown in S5 Fig; the minimum of each curve (optimal μ, see Table 1) was obtained by spline interpolation.
Firefly 8.1.0 and Gaussian 98 [79] give almost equal values of total energy for phenol and significantly different values for phenoxyl radical. Gaussian 98 uses Harris functional for the initial orbital guess by default instead of extended Huckel calculations used by Firefly. Harris functional is a nonself-consistent approximation to Kohn-Sham density functional theory [80], hence electron correlations should be partially taken into account. Nevertheless, the differences between Gaussian 98 (S1 Table) and Firefly (Table 1) BDE values are generally small (0.092±0.03 kcal/mol; p < 0.05, n = 16, without phenol).
For antioxidants in complex with phenoxyl radical (Ph-O*) and methyl peroxy radical (Met-OO*), transition structures (TSs) and corresponding local minima were optimized at level II. Intrinsic reaction coordinates (IRC) calculations [81] were performed for all TS species at the same level to confirm that anticipated reagent (R) and product (P) are connected to TS on potential energy surface. The products of the Met-OO* addition to antioxidant radical in para-position were optimized at level II. ΔECOR is corrected reaction activation energy:
ΔETS−R/COR=ΔETS−R+ΔGTS−R
(7)
ΔETS−R=ETS–ER
(8)
ΔGTS−R=GTS–GR
(9)
ΔEP−R/COR=ΔEP−R+ΔGP−R
(10)
ΔEP−R=EP–ER
(11)
ΔGP−R=GP–GR
(12)
where ΔGTS-R and ΔGP-R signify thermal correction to free energy at 298.15 K; ETS, ER, and EP are total energies of TS, R, and P; GTS, GR, and GP are thermal contributions to free energies of TS, R, and P.
The rate of reaction (M-1s-1) between antioxidant and radical was calculated as in [32,46] using conventional TS theory:
k(T)=Ix(kBT/h)x[exp(−ΔETS−R/COR/RT)]x24.3xA(T)
(13)
where I is the reaction pathway degeneracy (equal to 1 for the all compounds), kB is Boltzmann's constant, h is Planck's constant, 24.3 is a multiplier used to convert the units from 1 atmosphere standard state to 1 M standard state, and A(T) is a temperature-dependent factor which corresponds to quantum mechanical tunneling, approximated by the Wigner method [82]:
A(T)=1+(1/24)x(1.44νi/T)2
(14)
where νi is the imaginary frequency (cm-1) whose vibrational motion determines the direction of the reaction.
Atom coordinates of the optimized structures are given in S1 Dataset.
For the structures optimized at level II (TSs) and III (all other structures), the single point energy calculations were performed at B3LYP/6-311+(O)+G(d) level with diffuse sp functions added only to O atoms (level IV) or at B3LYP/6-311++G(d,p) level (V), both in the gas phase and in water solution at 298.15 K using dielectric polarizable continuum model (DPCM) [83]. Due to the DFT convergence problem, the calculations were performed at level V only for five compounds (L-KYN, L-3HOK, 3HAA, KYNA, and AA). Pearson correlation coefficient R for IPs calculated at levels IV and V is 0.999. Hence the lack of H(p) polarization functions and C, N(sp) diffuse functions at level IV did not change the rankings of IP values for different kynurenines. Single point energy calculations were performed for 3HAA–Met-OO* and XAAOXO−Met-OO* IRC at level V in the gas phase using XYG3 functional [45].
The values of the total free energy in solvent were used to calculate ΔE for the compounds in water solution. Since the value of the thermal correction to BDE (III) was very similar for different compounds (-6.645±0.260 kcal/mol, p < 0.05, n = 16), it was not considered. Also, the value for the thermal correction to antioxidant IP (III) was small (-0.20±0.25 kcal/mol, p < 0.05, n = 21) and was not considered. For the TSs, the values of ΔGTS-R and νi obtained at level II were used to calculate the values of ΔECOR and k(T) at level IV.
Statistical analyses were performed using Social Science Statistics online resource [84]. Illustrations were prepared with the help of MaSK 1.3.0. [85] and VMD [86]. The lipophilicity (logP) of compounds was calculated using the Molinspiration server [87].
|
10.1371/journal.pgen.1003417 | High-Resolution Mapping of H1 Linker Histone Variants in Embryonic Stem Cells | H1 linker histones facilitate higher-order chromatin folding and are essential for mammalian development. To achieve high-resolution mapping of H1 variants H1d and H1c in embryonic stem cells (ESCs), we have established a knock-in system and shown that the N-terminally tagged H1 proteins are functionally interchangeable to their endogenous counterparts in vivo. H1d and H1c are depleted from GC- and gene-rich regions and active promoters, inversely correlated with H3K4me3, but positively correlated with H3K9me3 and associated with characteristic sequence features. Surprisingly, both H1d and H1c are significantly enriched at major satellites, which display increased nucleosome spacing compared with bulk chromatin. While also depleted at active promoters and enriched at major satellites, overexpressed H10 displays differential binding patterns in specific repetitive sequences compared with H1d and H1c. Depletion of H1c, H1d, and H1e causes pericentric chromocenter clustering and de-repression of major satellites. These results integrate the localization of an understudied type of chromatin proteins, namely the H1 variants, into the epigenome map of mouse ESCs, and we identify significant changes at pericentric heterochromatin upon depletion of this epigenetic mark.
| Embryonic stem cells (ESCs) possess unique chromatin and epigenetic signatures, which are important in defining the identity and genome plasticity of pluripotent stem cells. Although ESC epigenomes have been extensively characterized, the genome localization of histone H1 variants, the chromatin structural proteins facilitating higher-order chromatin folding, remains elusive. Linker histone H1 is essential for mammalian development and regulates the expression of specific genes in ESCs. Here, by using a knock-in system coupled with ChIP–seq, we first achieve the high resolution mapping of two H1 variants on a genome-wide scale in mouse ESCs. Our study reveals the correlations of this underexplored histone family with other epigenetic marks and genome attributes. Surprisingly, we identify a dramatic enrichment of H1d and H1c at major satellite sequences. H10, mapped using an overexpressing ESC line, shows similar features at active promoters but differential binding at repetitive sequences compared with H1d and H1c. Furthermore, using mutant ESCs that are deficient for multiple H1 variants, we demonstrate the role of H1 in chromocenter clustering and transcriptional repression of major satellites. Thus, these results connect this important repressive mark with the well understood ESC epigenome and identify novel functions of H1 in mammalian genome organization.
| In all eukaryotes, nuclear DNA is packaged into chromatin by its association with histones [1]. The nucleosome, the building block of chromatin, consists of an octamer of four core histones (H2A, H2B, H3 and H4) wrapped by 147 bp of DNA [2]. Linker histone H1 binds to DNA entering and exiting nucleosome core particles as well as the linker DNA between nucleosomes, facilitating folding of chromatin into higher order structure [1], [3]–[5].
The H1 histone family is the most divergent group among the highly conserved histone proteins. To date, 11 different H1 variants have been characterized in mammals, including somatic H1 variants (H1a to H1e), the replacement H1 (H10), germ cell specific H1s (H1t, H1T2, HILS1 and H1oo), as well as the recently characterized variant H1x [6]. Deletion of three major somatic H1 variants (H1c, H1d and H1e) together leads to a 50% reduction of the total H1 level and embryonic lethality at midgestation, demonstrating that H1 level is critical for mammalian development [7]. H1 variants are conserved from mouse to human, and differ in their biochemical properties and expression patterns during development and malignant transformation [8]–[11]. Although none of the H1 variants tested is essential for mouse development [12]–[15], they have been shown to regulate specific gene expression in various cell types [6], [16]–[18]. However, the mechanisms by which H1 variants modulate chromatin structure and gene expression remain under-explored. Mapping of the precise genomic localizations of different H1 variants in vivo is likely to provide significant insights, but has been challenging due to the lack of high quality antibodies that could accurately distinguish different H1 variants.
Pluripotent embryonic stem cells (ESCs) can differentiate into cells of all three germ layers, offering great potential in regenerative medicine. The epigenome is suggested to play a critical role in stem cell fate determination, and genome-wide mapping studies have revealed that ESCs have characteristic epigenetic landscapes that differ from differentiated cells [19], [20]. However, despite significant efforts to characterize the chromatin features of human and mouse ESCs, both by individual labs [19], [21]–[23] and by large consortia (ENCODE [24], Roadmap Epigenomics [25]), the landscapes of linker histone H1 variants have not been described on a genome-wide scale.
In this study, we have achieved high resolution mapping of H1d, H1c and H10 in ESCs by chromatin immunoprecipitation followed by massive parallel sequencing (ChIP-seq). H1d and H1c are among the most abundant linker histones in mouse ESCs, accounting respectively for 32.6% and 16.4% of total H1, whereas the differentiation associated H1, H10, accounts for 2% of H1 in undifferentiated ESCs [26], [27]. These three variants differ significantly in terms of their residence time on chromatin and their ability to promote chromatin condensation in vitro [28], [29]. They also display different expression patterns during mammalian development and in exponentially growing cells vs. quiescent cells [8], [10], [30]. Here, we have generated FLAG-tagged H1d knock-in ESCs, Myc-tagged H1c knock-in ESCs, as well as FLAG-tagged H10 overexpressing ESCs, designated as respective H1dFLAG, H1cMyc, and fH10 cells. We demonstrate that tagged H1 variants maintain the biochemical properties of the endogenous H1s in vivo and that FLAG-H1d can substitute for H1d during mouse development. High resolution mapping reveals that H1d and H1c occupancies are highly correlated, both enriched at AT-rich regions, but also possess different binding specificity. Both H1d and H1c largely co-localize with H3K9me3, but show an inverse correlation with GC% or H3K4me3. Importantly, we discover that H1d and H1c are highly enriched at major satellite elements, which display a longer nucleosome repeat length than bulk chromatin in ESCs. Finally, we show that H1 depletion leads to chromocenter clustering and increased expression of major satellites independent of multiple epigenetic marks at these regions.
Efforts to generate high resolution genome-wide maps of H1 variants were hampered by the lack of H1 variant specific antibodies of sufficient quality for ChIP-seq. Here, we established knock-in mouse ESC lines in which H1d or H1c variant was N-terminally tagged with an epitope (FLAG or Myc) for which highly specific antibodies exist. An H1dFLAG cell line was created by inserting the FLAG tag coding sequence at the endogenous H1d locus through homologous recombination (Figure 1A). H1c/H1e double knockout mice develop normally, yet H1c/H1d/H1e triple knockout (H1 TKO) mice are embryonic lethal [7]. Thus, ESCs with H1dFLAG allele in H1c+/−H1e+/− background could be used to produce H1c−/−H1dFLAG/FLAGH1e−/− mice to determine whether FLAG-tagged H1d (FLAG-H1d) functions equivalently to endogenous H1d by assessing if the tagged H1d can rescue the embryonic lethality of H1 TKO mutants. Toward this end, we generated both H1c+/−H1d+/FLAGH1e+/− (“H1dFLAG”) and H1c+/−H1dFLAG/−H1e+/− (“H1d-trans”) ESC lines by transfection of the FLAG-H1d targeting vector (Figure 1A) into the cis triply targeted H1c+/−H1d+/−H1e+/− ESCs established previously [7]. ESC clones with either cis or trans configuration of the H1dFLAG allele with the H1c and H1e KO allele were identified and verified by Southern blotting (Figure 1B). As expected, FLAG-H1d was located in the nuclei of the H1dFLAG cells (data not shown). Analysis of histone extracts of chromatin prepared from cis-targeted H1dFLAG cells by HPLC and immunoblotting indicated that FLAG-H1d was associated with chromatin and eluted in the same fraction as the endogenous H1d, suggesting that FLAG-H1d has the same hydrophobicity as the endogenous H1d (Figure 1C and 1D). The ratio of somatic H1 variants, H1 a–e, to nucleosome (H1/nuc) of H1dFLAG cells was nearly identical to that of H1c+/−H1d+/+H1e+/− (cehet) cells, indicating a similar expression level of FLAG-H1d as the endogenous H1d (Figure 1E). As expected, the protein level of differentiation associated H10 variant was minimal in undifferentiated ESCs. We injected cis-targeted H1dFLAG cells into mouse blastocysts and produced chimeric mice which gave germline transmission of the H1dFLAG allele. H1c+/−H1d+/FLAGH1e+/− mice were intercrossed to generate H1c−/−H1dFLAG/FLAGH1e−/− homozygous mice (designated as H1dFLAG/FLAG mice) (Figure S1Ai). These homozygotes were viable, fertile and developed normally as H1c/H1e double null (ceKO) mice, demonstrating that FLAG-H1d can substitute for the endogenous H1d to fully rescue the lethal phenotype of H1 TKO mutants. HPLC, mass spectrometry and immunoblotting demonstrated that H1dFLAG/FLAG mice had full replacement of H1d by FLAG-H1d (Figure S1Aii and S1Aiii) and that the H1/nuc ratio of spleen chromatin from H1dFLAG/FLAG mice was 0.7, comparable to that of ceKO mice (Figure S1Aiv). Taken together, these results demonstrate that FLAG-H1d maintains the expression level and properties of the endogenous H1d in vivo.
Using a similar knock-in strategy, we generated H1c+/MycH1d+/−H1e+/− ESCs (H1cMyc) by transfection of the H1cMyc targeting construct into the cis triply targeted H1c+/−H1d+/−H1e+/− ESCs and selected ESC clones that underwent homologous recombination at H1c locus (Figure S1Bi and S1Bii). Similar to FLAG-H1d, the N-terminally Myc tagged H1c (Myc-H1c) colocalized with Hoechst stained nuclear regions in H1cMyc cells (data not shown), and Myc-H1c was eluted in the same fraction as the endogenous H1c protein from HPLC analysis (Figure S1Biii and S1Biv). H1cMyc cells had a H1/nuc ratio of 0.38, comparable to the ratio of 0.36 in cehet cells (Figure 1E, Figure S1Biii), indicating that like FLAG-H1d, Myc-H1c has the same expression level and biochemical properties as the endogenous H1c.
To achieve high resolution mapping of H1d and H1c variants in mouse ESC genome, we performed ChIP-seq in cis-targeted H1dFLAG and H1cMyc ESCs using anti-FLAG and anti-Myc antibodies, respectively. In each ChIP-seq library, approximately 80–90% of reads were mappable to the mouse genome (mm9) using the Bowtie aligner [31] (Table S1). While sonicated chromatin input control libraries on average had 65% vs. 22% of reads mapped to unique positions and multiple positions respectively, the H1c ChIP-seq libraries had 44% vs. 45% mapped to unique vs. multiple positions, suggesting that a higher proportion of H1c resides on repetitive sequences. Similarly, an overrepresentation of multi-match sequence reads (39% of mapped reads) occurred in H1d ChIP-seq libraries. A survey of sequencing signal intensities indicated that H1d and H1c were generally depleted from gene rich regions with the deepest dips around transcription start sites of active genes (examples shown in Figure 2A and Figure S2A). ChIP-seq with the anti-FLAG antibody in control ESCs not containing FLAG-H1d generated minimal random background signals (data not shown), and examination of H1c (anti-Myc) signals showed no enrichment at c-Myc target genes, such as Oct4, Nanog and Sox2 [32] (Figure S2A), indicating no cross-reactivity for these antibodies. To compare H1 occupancy with other histone marks, we performed ChIP-seq of an active histone mark, H3K4me3, and two repressive histone marks, H3K9me3 and H3K27me3, in murine ESCs. Visual examination of the track files revealed that H1 dips often coincided with H3K9me3 dips or H3K4me3 peaks and that H1 displayed higher signals at gene poor regions with high AT% (low GC%) (Figure 2A). H3K27me3, enriched at Hox gene clusters (Figure S2B) as expected, did not show obvious pattern correlation with H1 (Figure 2A and Figure S2A). These observations suggest possible correlations of H1d and H1c with H3K9me3, H3K4me3, gene distribution and GC content in vivo.
We next investigated the relationship between H1 occupancy and gene expression levels at a 10 kb region centered around transcription start sites (TSSs) as well as a 10 kb region centered around transcription termination sites (TTSs) using GenPlay software [33]. Such metagene analysis revealed that H1 signals were always lower than chromatin input control within these regions (IP-IN<0) (Figure 2B), suggesting a general depletion of H1 at gene containing regions. Both H1d and H1c were especially depleted around the TSSs with dips much deeper at highly active genes than at silent genes (Figure 2B). Interestingly, except at TSSs and promoters, H1 signals remained largely constant throughout the gene encompassing regions and the signal intensity was higher at the silent genes than that at active genes, suggesting that H1 is underrepresented at surrounding regions of active genes as well (Figure 2B). Indeed, for genes highly depleted of H1 variants at promoters, the signal values of H1s, although gradually increased toward distal regions, remained diminished up to 200 kb from TSS (Figure S3), suggesting that H1s are depleted from broad domains at these regions in the genome. H3K4me3 is known to be peaked around TSS of active genes [34], [35], and metagene H3K4me3 curves displayed an opposite pattern to that of H1 (Figure 2B), further indicating that H1 is absent at active promoters. H3K9me3 exhibited a very similar distribution pattern to that of H1d and H1c, whereas H3K27me3 did not show similar profiles to that of H1 variants (Figure 2B). Metagene analysis of H1 and histone marks on genes finely partitioned by expression levels (each group with 20% of genes) over a 10 kb region (−5 kb to +5 kb of TSS) further corroborated their distinctive patterns at TSSs as a function of gene expression (Figure 2C).
To better define the correlation of H1 occupancy with histone marks around TSSs and promoters, metagene analysis of H1 signals was performed for genes partitioned into 5 groups according to their levels of H3K9me3, H3K4me3, or H3K27me3, which displayed characteristic profiles around TSS (Figure 2D, 2E, 2F and Figure S2C). H1 signals displayed positive and negative correlations with respective H3K9me3 and H3K4me3 signals, having the deepest dip for promoters and TSSs with the lowest H3K9me3 levels (Figure 2D) or highest H3K4me3 signals (Figure 2E). On the other hand, H1 signals showed no correlation with H3K27me3 levels and no difference among the 5 groups of genes partitioned according to H3K27me3 levels (Figure 2F). Interestingly, H1 was also depleted at the promoters of genes bound by H3K4me3 and H3K27me3 bivalent marks [21] but not at H3K4me3-free promoters, regardless of the presence or absence of H3K27me3 (Figure 2G).
Although most H1d and H1c signals appeared universally distributed, we identified regions enriched for H1 binding using SICER [36] and GenPlay software. Identified H1d and H1c enriched regions often formed broad domains (examples shown in Figure S4A). Annotation of H1d- and H1c- rich regions using CEAS [37], a software designed to characterize both sharp and broad ChIP-seq enrichment, indicated that, similar to H3K9me3, both H1d and H1c “peaks” were over-represented in distal intergenic regions and under-represented at promoters and 5′UTR, which were highly enriched with H3K4me3 peaks as reported previously (Figure S4B and [34]).
We next performed genome-wide correlation analysis to determine if the similarity and/or contrast of H1 variants with GC% and histone marks at TSSs also extend to a genome-wide scale. Indeed, the distribution of H1d and H1c were highly correlated throughout the genome (R = 0.7866) (Figure 3A), and both variants were negatively correlated with GC% (R = −0.4182 and −0.4140 for respective H1d and H1c), indicating that H1d and H1c were enriched or depleted at similar regions. Both H1d and H1c were correlated negatively with H3K4me3 (R = −0.2640 and −0.3317 respectively), but positively with H3K9me3 (R = 0.5732, 0.5790) (Figure 3B), suggesting their enrichment at heterochromatin. On the other hand, these two variants showed no obvious correlation with H3K27me3 (R = −0.08 for both variants) (Figure 3B). Correlation analysis of sequencing signals on enriched or depleted regions gave similar coefficients as the respective genome-wide coefficients (data not shown). It is interesting to note that the coefficients of H1 vs. H3K4me3 on sex chromosomes were dramatically different from those of autosomes (Figure S5A). This result echoes the previous finding that sex-chromosome genes are overrepresented among genes with altered expression levels by triple H1 deletion in ESCs [26], suggesting that H1 may play a role in regulating higher order chromatin structures of sex chromosomes.
To gain a comprehensive view of the DNA features of H1d- and H1c- rich regions, we selected the regions highly enriched for H1 variants and histone marks, and performed cross-comparison of genome attributes using the statistical analysis software EpiGRAPH [38]. Such analysis (Figure 3C and Figure S5B) revealed that: a) H1d/H1c common peaks (regions highly enriched for both H1d and H1c) appeared similar to H3K9me3 peaks in genome attributes, except for satellite DNA which was relatively overrepresented in H1 peak regions; b) H1d/H1c common peaks were enriched at AT-rich sequences, satellite DNA, and chromosome G-bands but were absent from GC-rich regions, and genes or exons when compared with H3K4me3 or H3K27me3 peaks; c) comparison of H1d/H1c common peaks with H1d/H1c unique peaks (regions highly enriched for H1d or H1c but not both) showed similar features as the comparison of H1d/H1c common peaks with H3K4me3 or H3K27me3 peaks; d) comparison of H1d vs. H1c specific peaks indicated that H1d unique peaks were relatively enriched at GC-rich sequences and LINEs, whereas H1c unique peaks were more enriched at AT-rich sequences, Giemsa positive regions and satellite DNA; e) the overrepresentation analyses between H1d (or H1c) unique peak regions and histone mark peak regions exhibited similar features as comparisons using H1 common peaks. These results define common and unique features for H1d and H1c enriched regions.
The EpiGRAPH overrepresentation analysis indicated that peak regions of H1d and H1c were enriched for satellite repeats. Indeed, examination of the top ranked H1 peak regions with especially high binding signals revealed that these regions overlap perfectly with major satellite sequences (examples shown in Figure 4A). This finding and the above observation of overrepresentation of multi-match sequence reads in H1 ChIP-seq libraries prompted us to perform a thorough mapping study of sequence reads to a database of repetitive sequences. We aligned sequence reads of H1d, H1c, H3K9me3, H3K27me3 and H3K4me3 ChIP-seq libraries to Repbase Update, a comprehensive database of repetitive elements from diverse eukaryotic organisms [39]–[41]. We found that both H1d and H1c were significantly enriched at repetitive sequences, with H1d and H1c ChIP-seq libraries having on average percent mapped repeats respective 2.3-, and 2.8-fold of that of chromatin input-seq libraries (Figure 4B). H3K9me3, H3K27me3 and H3K4me3 ChIP-seq libraries had an average respective percent mapped repeats 1.4-, 0.7-, and 0.9- fold compared with input controls (Figure 4B), suggesting an overrepresentation of H3K9me3, yet not as dramatic as H1d and H1c, at repetitive sequences.
Importantly, we found that the increased proportion of total reads of H1 libraries mapped to repetitive sequences was predominantly caused by overrepresentation on the major satellite sequences on which the levels of H1d and H1c occupancy were enriched on average 4.0- and 5.6-fold compared with the chromatin input control (Figure 4B). This level of H1 enrichment appeared to be specific to major satellites because we did not observe H1d and H1c enrichment among other abundant repeats, except for a moderate increase of H1d and H1c occupancy at minor satellites. qChIP-PCR results confirmed the preferential binding of these two H1 variants to major satellites (Figure S6). Sequencing results showed that H1d and H1c levels on most of other less abundant classes of repetitive elements, such as L1, IAP LTR retrotransposons, SINE, non-LTR retrotransposons, and DNA transposons, were similar or lower compared with the input control (Figure 4B and Figure S7). H3K4me3 was highly enriched at 5′end of a subset of LINE L1 sequence (Figure S7), consistent with the abundant expression of L1 detected in multiple cell types [42]–[44], whereas H3K9me3 was enriched at major satellite repeats and LTR transposons, such as IAP particles, with similar levels as previously reported [34], [45] (Figure 4B and Figure S7). Enrichment of H1 variants at major satellites was also confirmed by calculating the normalized “IP-IN” signals at major satellite regions in mouse genome mm9 assembly (July 2007) annotated by RepeatMasker (http://repeatmasker.org) (Figure S8). Analysis of ChIP-seq libraries of FLAG-H1d in H1d-trans ESCs, which had similar levels of FLAG-H1d and total H1/nuc ratio as the cis H1dFLAG ESCs, also showed similar level of enrichment at major satellites as H1dFLAG ESCs (Figure S9).
The level of H1 has been shown to be a determinant of nucleosome repeat length (NRL) with a higher level of H1 correlating with a longer NRL [46], [47]. To validate the enrichment of H1 variants at major satellites and to investigate its impact on the local chromatin structure at these regions, we measured the NRL of bulk chromatin and that of the pericentromeric (major satellites) and centromeric (minor satellites) regions with a time-course micrococcal nuclease (MNase) digestion assay. Southern blotting images revealed that chromatin at major satellites was more resistant to MNase digestion than bulk chromatin and minor satellites (Figure 5A). Consistent with previous studies [26], the bulk chromatin of mouse ESCs displayed a NRL of ∼187 bp (Figure 5B). However, the NRL at major satellites had a value of 200 bp, which was ∼13 bp and ∼8 bp longer than the NRLs of respective bulk chromatin and minor satellites in ESCs (Figure 5B). These results suggest that the enrichment of H1d and H1c at major satellite repeats may contribute to the increase of NRL in the pericentromeric region compared with bulk ESC chromatin. Analysis of H1c/H1d/H1e triple knockout (H1 TKO) ESCs established previously, which have an H1/nuc ratio of 0.25 in bulk chromatin compared with that of 0.46 in WT ESCs [26], indicated that H1 depletion caused a proportional decrease of NRLs in bulk chromatin, major satellites and minor satellites (Figure S10). Consistently, qChIP analysis using a pan-H1 antibody showed total H1 levels were reduced at major and minor satellites by H1 depletion (Figure S10D).
Major satellite repeats at pericentric heterochromatin from different chromosomes tend to cluster together and form the chromocenter, a nuclear compartment that plays an important role in structural maintenance of the chromosomes [48], [49]. Several chromatin proteins such as MeCP2, MBD2, DNMT3a, DNMT3b, and UHRF1 have been shown to contribute to chromocenter clustering [50]–[52], however, the role of H1 in chromocenter formation has not been studied to date. Since both H1d and H1c are markedly enriched at major satellites, we set out to determine the effects of H1 depletion on chromocenter clustering in WT and H1 TKO ESCs by fluorescence in situ hybridization (FISH) using a major satellite specific probe. The chromocenter numbers in H1 TKO ESCs (median = 8, n = 160) were significantly lower than WT cells (median = 17, n = 206) (Figure 6), and the size of chromocenters in H1 TKO ESCs on average was bigger than that in WT ESCs (Figure S11), demonstrating a previously unnoticed defect in the pericentromeric chromatin structure caused by H1 depletion. Analysis of “rescue” (RES) cells established previously [53] showed that overexpressing H1d in H1 TKO cells effectively restored the size and the numbers of chromocenters to the levels comparable to WT cells (Figure 6 and Figure S11). Similarly, H1dFLAG and H1cMyc cells displayed normal chromocenter clustering as WT ESCs (Figure S15). These results indicate that the increased chromocenter clustering is likely due to the dramatic decrease of total H1 levels in H1 TKO ESCs.
Pervasive transcription of repetitive sequences contributes to genome regulation, and aberrant regulation of the expression of satellite sequences interferes with heterochromatin assembly and chromosome segregation [49], [54]–[56]. To further examine the effects of H1 depletion on major satellites, we analyzed several repetitive sequences for expression and epigenetic marks in WT and H1 TKO ESCs. Quantitative reverse transcription-PCR (qRT-PCR) analysis showed that the expression levels of major satellites were 3.5-fold higher in H1 TKO ESCs than in WT ESCs, whereas the expression levels of minor satellites and LINE L1 were not significantly changed (Figure 7A). Such de-repression of major satellites by H1 depletion was dramatically curbed in RES cells (Figure 7A) as well as in H1dFLAG and H1cMyc ESCs (Figure S16), indicating that the levels of H1s have a direct impact on transcriptional regulation of major satellites. Notably, the levels of multiple epigenetic marks, such as repressive marks H3K9me3, H3K27me3, and H4K20me3, the active mark H3K4me3, as well as DNA methylation all remained unchanged at the analyzed repeats in H1 TKO ESCs compared with WT ESCs (Figure 7B and 7C). The lack of significant changes in the histone marks and DNA methylation at these repetitive sequences suggests that the increase in expression levels at major satellites may be due to an effect of local chromatin decondensation caused by H1 depletion in H1 TKO ESCs.
We note that the level of H10, the replacement H1 variant, was increased significantly in TKO ESCs compared with that in undifferentiated WT ESCs where H10 was minimal [26], [53]. To examine if the increased chromocenter clustering and expression of major satellites in H1 TKO ESCs could be attributed to an increase in H10 levels, we generated “fH10” cells by over-expressing FLAG-H10 in WT ESCs, and selected cell lines that expressed FLAG-H10 at a similar level to that of H10 in H1 TKO ESCs (Figure S12 and [26]). As expected, FLAG-H10 was eluted in the same fraction as endogenous H10. ChIP-seq of H10 in fH10 cells with an anti-FLAG antibody indicated that, despite its different biochemical properties and unique expression patterns [6], [8], [57], H10 shared similar distribution features to that of H1d and H1c in ESCs, including depletion at active promoters and enrichment at major satellites (Figure S3, Figure S8, Figure S13, and Figure S14). Similar to H1d and H1c, H10 also displayed overall positive correlation with H3K9me3 and inverse correlations with GC% and H3K4me3, although the level of correlation was to a lesser extent (data not shown). Furthermore, H10 enriched regions were significantly under-represented in gene regions but over-represented in distal intergenic regions with 80.1% of H10 peaks located in these regions (data not shown). Beside major satellites, H10 also appeared to be enriched at minor satellites and, to a lesser extent, at LINE L1 elements as determined by ChIP-seq and ChIP-PCR (Figure S14B and S14C), suggesting differential binding preferences of H10 compared with H1d and H1c.
Analysis of fH10 ESCs by FISH and qRT-PCR indicated that the chromocenter numbers were not reduced compared with WT ESCs (Figure S15) and that expression of major satellites remained at low levels (Figure S16), excluding the possibility of H10 upregulation being responsible for chromocenter clustering and upregulation of major satellite transcription in H1 TKO ESCs.
Collectively, these results demonstrate increased chromocenter clustering and major satellite transcription by H1 depletion, and suggest important roles of the dominant H1 variants in ESCs in maintaining pericentric chromatin properties.
H1 Linker histones are abundant chromatin binding proteins that facilitate the formation of higher order chromatin structures [1], [2]. The existence of multiple mammalian H1 variants which are differentially regulated during development presumably offers additional levels of modulation on chromatin structure and function. Despite many efforts, the in vivo localization and function of individual H1 variants in genome organization remain elusive. Chromatin plays critical roles in stem cell fate determination and reprogramming, and the epigenome of ESCs has been intensively studied. However, the genome-wide maps of one group of the major chromatin proteins, H1 variants, have not been established. Here, we have filled both gaps by generating high resolution maps of three H1 variants in mouse ESCs, identified unique H1 binding features, and discovered an unusual enrichment and function of H1 variants at major satellites.
We have established a knock-in system to stringently test the functions of the tagged H1s and to facilitate the generation of high resolution maps of H1 variants in ESCs by ChIP-seq. Our results demonstrate that, when tagged at the N-terminus, the short FLAG and Myc tags, with respective 8 and 13 amino acids, do not alter the biochemical and cellular properties of H1 proteins in vivo. The strategy of homologous recombination ensures that the expression of tagged H1 variants is comparable to that of their endogenous counterparts. FLAG-H1d fully rescues the lethal phenotype of H1d deletion on H1c/H1e double knockout genetic background, further demonstrating the functional equivalence of the tagged H1 and the respective endogenous H1 variant in vivo. Although Myc-H1c was not tested in mice, it is anticipated to mimic the endogenous H1c based on all the other assays performed. These data provide a technical demonstration on how highly similar protein variants can be analyzed differentially and on a genomic scale using in vivo validated knock-in mice.
On the H1 genome-wide maps we have generated here, H1d and H1c are highly correlated and display similar binding patterns in the ESC genome. Both variants are enriched at AT-rich regions, gene deserts and major satellites, but are depleted at GC-rich, gene-rich regions and especially at active promoters. Thus, despite their differences in compacting DNA in vitro and the expression patterns during development [8], [10], [28], H1d and H1c are quite similar in overall distribution in the genome, which we surmise contributes to the redundancy among the major somatic H1s as suggested from previous studies of single or double H1 variants knockout mice [7], [14]. Nevertheless, analyses of the regions that are uniquely enriched for H1d or H1c reveal some differences in sequence features (Figure 3C and Figure S5B). H1c has a higher enrichment at major satellites than H1d but is relatively depleted from LINE sequences (Figure 3C and Figure 4B). In addition, H1c enriched regions have a higher proportion in gene bodies and proximal regions compared with H1d peak distribution (Figure S4B). These differences may account for an additional level of modulation and fine-tuning of genome function by the presence of multiple H1 variants in mammals.
H10, the H1 variant associated with differentiation, has unique expression pattern and biochemical properties. It is highly basic, expressed in differentiated cell types, and more similar to histone H5 in avian red blood cells than any other somatic variants [53], [57]. However, overexpressed H10 (in fH10 cells) shares the distinctive features of H1d and H1c in ESCs in genome-wide occupancy. It is worth noting, though, that endogenous H10 proteins are present at very low levels in undifferentiated WT ESCs and the genome-wide localization of H10 in ESCs may differ significantly from its binding patterns in differentiated cells. It would be interesting to systematically determine the genome-wide maps of histone variants in different cell types, particularly in light of a recent study reporting a distribution pattern change of H1.5 in cellular differentiation [58]. The cell lines and mouse models generated in this study will greatly facilitate these future studies.
The prevalent H1 variants binding with local troughs at active promoters we observed here in the mouse ESC genome is reminiscent of the previous results when ChIP-chip and a pan-H1 antibody were used to map H1 on a portion of the human genome in MCF-7 cells [59] or when DamID method was used to map H1 in Drosophila cells [60]. The depletion of H1 at TSSs of active genes observed in three systems suggests that this feature is common to all H1s and evolutionarily conserved. However, our study differs from the two previous studies and offers more opportunities for high resolution and in-depth analysis because the knock-in system generated in this study allows for robust and highly specific mapping of H1 variants and deep-sequencing covers the entire genome including the repetitive genome. Furthermore, we have found that the depletion of H1 at active genes is not restricted to regions around the TSS, but also expands to the entire gene encompassing domain (Figure 2B and 2C). Such phenomena suggests that a wide-spread change in higher order chromatin structure may be associated with gene expression and that gene-rich domains may adopt an overall decondensed chromatin structure with less H1 occupancy.
Correlation analyses indicate that H1d and H1c are inversely correlated with GC content, H3K4me3 mark, but positively correlated with H3K9me3 mark across the mouse ESC genome (Figure 3B). Our finding that the common peaks of H1d and H1c are enriched with AT-rich DNA sequences in vivo resonates with the previous observation that H1 is preferentially associated with scaffold associated regions (SAR) [61], which are also AT-rich sequences [62]. This binding feature may reflect a higher affinity of H1 to AT-tracts observed in in vitro studies [63], [64]. The GC content has been suggested to be an intrinsic factor for nucleosome occupancy [65], and our data suggest that it may also have an impact on H1 binding. It is also noteworthy that, compared with gene expression levels, H3K4me3 and H3K9me3 correlate better with H1 levels at TSS. For example, we did not observe dips of H1d and H1c around promoters of 40% genes when partitioned by H3K4me3 or H3K9me3 signals, whereas a small H1 signal dip exists even for the 20% genes with lowest expression values (Figure 2C, 2D, and 2E). It is possible that the steady state level of RNA messages (expression) may not faithfully reflect the active/inactive state of the promoters which may correlate better with the status of histone marks. It has been reported that promoters of many genes with low expression have high H3K4me3 levels [21], and we surmise that H1 may be absent from these gene promoters as well.
The co-localization of H1d and H1c with H3K9me3 suggests that these two variants are enriched at heterochromatin and may facilitate the maintenance of constitutive heterochromatin structure. Such association may be mediated through HP1, the heterochromatin protein binding to H3K9me3 and H3K9 methyltransferase Suv39h and facilitating spreading of heterochromatin marks [66]–[68]. Indeed, H1 has been shown to interact in vitro with HP1α [69], [70]. On the other hand, localization of HP1 is impaired in H1 depleted Drosophila [71], suggesting that H1 may also contribute to the proper targeting of HP1.
Surprisingly, we found that, at major satellite sequences, H1d and H1c signals are dramatically overrepresented, and this accounts for almost all the increased proportion of H1 sequence reads at repetitive sequences. The levels of H1d and H1c at major satellites are much higher than H3K9me3 (Figure 4B), a repressive histone mark also enriched at these repeats [34]. The overrepresentation of H1 at major satellites in ESCs is also supported by a longer NRL, which suggests a higher local H1 level than bulk chromatin and minor satellites. Consistent with previous observations [49], [51], we find that major satellites are more resistant to MNase digestion than bulk chromatin and minor satellites in ESCs (Figure 5), suggesting that pericentromeric regions may adopt special higher order chromatin structure as indicated by sucrose sedimentation assay [72]. High resolution mapping in this study identifies major satellites as the dominant preferential binding sites for H1 variants in ESCs, suggesting that H1 may play an important role in mediating the formation of distinct chromatin structure at pericetromeric regions. This is further supported by the effects of H1 depletion on chromocenter clustering and expression of major satellites. We note that a higher NRL in major satellites than bulk chromatin is also present in H1 TKO ESCs (Figure S10), suggesting a possible enrichment of the remaining H1 variants at major satellite sequences in H1 TKO ESCs. Consistently, we find that overexpressed H10 also appear to preferentially accumulate at satellite sequences in ESCs (Figure S14).
The enrichment of H1 at major satellites could not be solely attributed to the relatively high affinity of H1c and H1d to AT-rich sequences. Major and minor satellites sequences contain approximately 65% of A and T, with a ratio of A∶T being respective 2.6∶1 and 1.8∶1. This could result in major satellites having more A-tracts to which H1 might have a higher affinity. Phased nucleosome positioning observed at the major satellites [73], [74] could also contribute to the preferential binding of H1 at this region because different nucleosome positioning patterns have been shown to differentially affect H1 binding in vitro [75].
Mouse major satellites, constituting the pericentromere [76], [77] necessary for chromosome structure and function, are shown to form clusters/chromocenters, exhibit distinct heterochromatin features and adopt a more stable and condensed chromatin conformation than the bulk chromatin [49], [72]. Our findings of the preferential binding of H1 at major satellites and chromocenter clustering (reduced number of chromocenters) in H1 TKO ESCs suggest that H1 contributes to and may be required for the proper formation of pericentric heterochromatin. The rescue of the clustering effects by overexpressing H1d in H1 TKO ESCs or in H1dFLAG and H1cMyc cells compared with H1 TKO ESCs indicates that the total H1 level, rather than a specific H1 variant, is a key determining factor of chromocenter clustering. This conclusion is further supported by our finding that overexpressing H10 level to 3.5 fold of that of endogenous H10 in WT ESCs has little effect on chromocenter numbers or major satellite expression. In vitro studies have shown highly cooperative binding of H1 globular domain to DNA [78], a property which we speculate could contribute to increased chromocenter clustering in the face of marked reduction of H1 levels in H1 TKO ESCs. A larger nucleosome spacing (200 bp) (Figure 5) together with a higher local H1 level at major satellites could be important for efficient compaction of pericentromeric chromatin because nucleosome arrays with a NRL of 197 bp are able to form 30 nm fiber structure in vitro in the presence of linker histones whereas arrays with a short NRL are only able to form thinner and less compact structures [5].
The effects of H1 on major satellites are not restricted to chromatin structure and heterochromatin formation. Loss of H1c, H1d and H1e causes a dramatic increase in transcripts from major satellites, but does not change the levels of the repressive epigenetic marks, H3K9me3, H4K20me3, H3K27me3, or DNA methylation at these sequences. This suggests that the increase in expression of major satellites in H1 TKO ESCs is not mediated by loss of these repressive epigenetic marks, but rather caused by reduced binding of H1 per se or the potential decondensation of local chromatin structure. The phenomenon of changes in chromocenter organization independent of H3K9me3 is reminiscent of results from deletion of UHRF1 [52], a histone binding protein or overexpression of MeCP2 in mouse myoblasts [50]. Chromocenter organization is likely to be independent of H3K9me3 pathway because double deletion of Suv39h1 and Suv39h2 has minimal effects on the number and size of chromocenters in mouse cells [79], [80]. The expression changes in major satellites in H1 TKO ESCs are also not due to potential changes in cell cycle since H1 TKO ESCs have similar growth rate [26] and cell cycle profiles (data not shown) to WT ESCs. The reduction in expression levels of major satellites detected in RES cells compared with H1 TKO cells further supports that the drastic decrease in H1 levels causes de-repression of major satellites. Noncoding major satellite transcripts have been shown to be important for proper chromocenter formation [81], thus we speculate that the increased levels of major satellite transcripts contribute to chromocenter clustering in H1 TKO cells. In light of previous findings that ESCs null for DNA methyltransferases displayed chromocenter clustering [51], similar to what we observed in H1 TKO ESCs, we surmise that H1 and DNA methylation may act cooperatively in the proper maintenance of chromocenter structure.
In summary, we report high resolution maps of two abundant somatic H1 variants and the replacement H1 variant in mouse ESCs, connecting this important yet under-explored repressive mark with the well-studied ESC epigenome. The enrichment and effects of H1d, H1c and H10 on major satellites highlight an important role of these H1 variants in the maintenance of chromosome architecture and function. The cell lines and mouse strains we generated using the knock-in system also provide valuable tools for studying H1 variant specific functions both in vitro and in vivo. Genome-wide distribution studies of other H1 variants as well as in differentiated cell types are likely to lead to a better understanding of the role of H1 and higher order chromatin folding in gene expression and chromatin function.
The H1dFLAG knock-in targeting vector containing H1d 5′ and 3′ homology regions flanking the N-terminal FLAG-tagged H1d and the SV40-Blasticidin resistant gene was transfected into ESCs as described previously [14]. 200 ESC clones resistant to 20 µg/ml Blasticidin (Life Technologies) and 2 µM gancyclovir (Sigma-Aldrich) were picked, and 5 clones with homologous recombination were identified by Southern blotting using the probe shown in Figure 1A. Two cis-targeted clones were injected into C57BL/6 recipient blastocysts to produce chimeric mice, which gave germline transmission. H1c+/−H1d+/FLAGH1e+/− mice were intercrossed to generate H1c−/−H1dFLAG/FLAGH1e−/− (H1dFLAG/FLAG) mice. All animal work was performed according to procedures approved by the Institutional Animal Care and Use Committee (IACUC) at Georgia Institute of Technology.
Nuclei and chromatin of ESCs and mouse tissues were prepared and analyzed according to protocols described previously [82], [83]. Histones were extracted from chromatin with 0.2 N sulfuric acid and 50–100 µg of total histone preparations were injected into a C18 reverse phase column (Vydac) on an ÄKTA UPC10 system (GE Healthcare). The effluent was monitored at 214 nm (A214), and the peak areas were recorded and analyzed with ÄKTA UNICORN 5.11 software. The A214 values of the H1 and H2B peaks were adjusted by the number of peptide bonds in each H1 variant and H2B. The H1/nucleosome ratio was determined by dividing the A214 of all H1 peaks by half of the A214 of the H2B peak. Fractions corresponding to different H1 variants from HPLC analysis were collected, lyophilized and analyzed with silver staining, Coomassie staining and Western blotting.
The following antibodies were used in this study: anti-FLAG (Sigma-Aldrich F3165), anti-DYKDDDDK tag (Cell Signaling #2368), anti-Myc-tag (Cell Signaling #2272), anti-H3K4me3 (Millipore 07-473), anti-H3K9me3 (Abcam 8898), anti-H3K27me3 (Millipore 07-449), anti-H4K20me3 (Millipore 07-463), anti-H10 (Santa Cruz 56695), anti-H1 (Milipore 05-457) and IgG (Millipore 12-370).
ChIP assays were performed as described previously [26] with the following modifications: 20 µl of Dynabeads Protein G (Life Technologies) were incubated with 2 µg of antibody for 4 hours, followed by incubation with 40 µg of sonicated soluble chromatin overnight at 4°C. Dynabeads were washed, immunoprecipitates were eluted, and DNA-protein complexes were incubated overnight at 65°C to reverse crosslinks. DNA was purified with a DNA Isolation column (Qiagen). Input control DNA was prepared from reverse-crosslinked soluble chromatin prior to immunoprecipitation. Quantitatitve PCR on ChIP samples for major satellites, minor satellites, LINE L1, IAP LTR and Hprt was performed with primers published previously [45], [84].
The libraries for massive parallel sequencing were prepared with the ChIP-seq Sample Preparation Kit (Illumina) according to the manufacturer's instructions. Briefly, 10 ng of immunoprecipiated DNA or input DNA were end repaired, 3′ adenylated and ligated with adapter oligos supplied by the manufacturer. DNA fragments within the range of 120∼500 bp were purified following gel electrophoresis and amplified with primers provided by the manufacturer. Library DNA was subsequently purified with a Qiagen DNA Isolation column, quantified and submitted for sequencing.
Sequencing was performed with Illumina Genome Analyzer II and Illumina HiSeq 2000 systems, and raw sequence reads containing more than 30% of ‘N’ were removed and adaptor sequences were trimmed. Clean sequences were aligned against mouse genome, mm9 (UCSC website), and 2,669 categories of mammalian repeats from RepBase version 14.07 [39], [40] using Bowtie aligner software (http://bowtie-bio.sourceforge.net/index.shtml). The first 40 bp (for alignment to mm9) or the first 35 bp (for alignment to RepBase) of the reads were used as seed sequences with up to two mismatches allowed for the alignment, and aligned number of reads were scored. Reads with multiple alignment positions were mapped randomly to one of the possible position. Reads for each ChIP-seq or input-seq library aligned to mm9 were normalized to 10 million reads, and IP-IN signals were calculated in each 100 bp sliding window by subtraction of normalized read counts per 10 million mappable reads of ChIP-seq library by that of its corresponding input-seq library using GenPlay software (http://genplay.einstein.yu.edu/wiki/index.php/Documentation) [33]). Percentage of reads for each repeat mapped to RepBase was calculated by dividing reads mapped to the respective repeat by the total reads in the library, and the fold enrichment for the respective repeat was subsequently calculated as the ratio of the percent of reads of ChIP-seq library to that of the input-seq library. Read length and read counts of each library are listed in Table S1. Representative ChIP-seq libraries with the most sequencing reads mapped to mm9 were utilized for genome browser visualization and metagene analysis, and all replicate ChIP-seq libraries were included in repetitive sequence analysis. Sequencing data have been deposited in NCBI's Gene Expression Omnibus database and assigned GEO Series accession number GSE46134.
The sum of signals (IP-IN) for each 1000 bp window (normalized to 10 million reads) was used to calculate the correlation coefficients of H1 variants with GC% and different histone markers. Genome-wide and chromosome-wide correlation coefficients were calculated, and the scatter-plots were generated using Matlab.
Significantly enriched regions were identified using GenPlay or SICER v1.1 [36] at the following parameter settings: window size = 200, gap size = 600, E-value = 1000, an effective genome size of 80% of the entire mouse genome, and q-value (FDR) = 0.001. In order to optimize the gap size for H1 variants, the gap size was varied from 0 to 3 times the window size (0, 200, 400, 600) and the best value was chosen according to the criteria as previously described [36]. Distribution of peak regions relative to gene regions was analyzed by CEAS [37]. Top 10% of enriched regions for each ChIP-seq library were selected to identify the overrepresented features using EpiGRAPH (http://epigraph.mpi-inf.mpg.de/WebGRAPH/) [38]. 2214 H1d/H1c common peaks, 1939 H1d unique peaks, 433 H1c unique peaks, 1891 H3K9me3 peaks, 4778 H3K27me3 peaks, and 3446 H3K4me3 peaks were analyzed by EpiGRAPH.
ESC nuclei were extracted and MNase digestion was performed as described previously [26]. Briefly, 2.5×106 nuclei were resuspended in 200 µl of MNase digestion buffer (0.32 M sucrose, 50 mM Tris-HCl pH 7.5, 4 mM MgCl2, 1 mM CaCl2, 0.1 mM PMSF) and digested at 37°C with 20 units of micrococcal nuclease (MNase) (Worthington) for time course analysis or 2 units of MNase (Worthington) for 5 min in analysis shown in Figure S10A. Nuclei were lysed and DNA was subsequently purified and analyzed by electrophoresis. Southern blotting was performed using major or minor satellite specific probes as described previously [26]. The NRL at each time point was calculated using the regression line generated with size (bp) of polynucleosomes [7], [26], and the values at time “0” were extrapolated as described previously [72].
FISH was performed as described previously [85]. The major satellite probe was biotin-labeled, denatured and hybridized to the slides overnight. The nuclei were incubated with FITC-Avidin for 1 hour, and counterstained with DAPI. Signals were detected with an Olympus Epifluorescence Microscope (Olympus, Inc.) equipped with an Olympus QCLR3 cooled digital camera. The experiments were repeated three times, and the number of chromocenters for each cell line was counted by three researchers as blind tests. Statistical analysis was performed using a Mann-Whitney U nonparametric test. Areas of chromocenters were quantitated using AxioVision software V4.8.2.0 and presented as pixel2. The conversion factor of pixel/micron was 18.7 pixels per micrometer.
1 µg of total RNA extracted from ESCs was treated with RNase free DNaseI (Sigma-Aldrich) and reverse transcribed using a SuperScript first-strand cDNA synthesis kit with random hexamers (Life Technologies). Triplicate PCR reactions using the iQ SYBR Green Supermix (Bio-Rad) were analyzed in a MyIQ Real-Time PCR Detection System (Bio-Rad). All samples were typically analyzed in two independent experiments. Relative expression units were calculated by subtracting the mock reverse-transcribed signals (RT−) from reverse transcribed signals (RT+) and normalizing the adjusted values with signals of the housekeeping gene GAPDH. The qRT-PCR primers for repetitive sequences are the same as in qChIP, and the primers for GAPDH are as described previously [53].
1 µg of DNA extracted from ESCs was treated with the CpGenome DNA modification Kit (Millipore) according to the manufacturer's manual. 20 ng of treated DNA was used in each PCR reaction as previously described [26]. The primers used to generate PCR products from the bisulfite-converted DNA are specific for the converted DNA sequence of the analyzed regions. The PCR products were subsequently cloned using the TOPO TA Cloning kit (Life Technologies), and colonies containing the converted DNA inserts were picked. DNA inserts were sequenced and analyzed with BiQ Analyzer [86]. Primers for major and minor satellites were as previously described [87].
|
10.1371/journal.pntd.0001630 | Deficient Regulatory T Cell Activity and Low Frequency of IL-17-Producing T Cells Correlate with the Extent of Cardiomyopathy in Human Chagas' Disease | Myocardium damage during Chagas' disease results from the immunological imbalance between pro- and production of anti-inflammatory cytokines and has been explained based on the Th1–Th2 dichotomy and regulatory T cell activity. Recently, we demonstrated that IL-17 produced during experimental T. cruzi infection regulates Th1 cells differentiation and parasite induced myocarditis. Here, we investigated the role of IL-17 and regulatory T cell during human Chagas' disease.
First, we observed CD4+IL-17+ T cells in culture of peripheral blood mononuclear cells (PBMC) from Chagas' disease patients and we evaluated Th1, Th2, Th17 cytokine profile production in the PBMC cells from Chagas' disease patients (cardiomyopathy-free, and with mild, moderate or severe cardiomyopathy) cultured with T. cruzi antigen. Cultures of PBMC from patients with moderate and severe cardiomyopathy produced high levels of TNF-α, IFN-γ and low levels of IL-10, when compared to mild cardiomyopathy or cardiomyopathy-free patients. Flow cytometry analysis showed higher CD4+IL-17+ cells in PBMC cultured from patients without or with mild cardiomyopathy, in comparison to patients with moderate or severe cardiomyopathy. We then analyzed the presence and function of regulatory T cells in all patients. All groups of Chagas' disease patients presented the same frequency of CD4+CD25+ regulatory T cells. However, CD4+CD25+ T cells from patients with mild cardiomyopathy or cardiomyopathy-free showed higher suppressive activity than those with moderate and severe cardiomyopathy. IFN-γ levels during chronic Chagas' disease are inversely correlated to the LVEF (P = 0.007, r = −0.614), while regulatory T cell activity is directly correlated with LVEF (P = 0.022, r = 0.500).
These results indicate that reduced production of the cytokines IL-10 and IL-17 in association with high levels of IFN-γ and TNF-α is correlated with the severity of the Chagas' disease cardiomyopathy, and the immunological imbalance observed may be causally related with deficient suppressor activity of regulatory T cells that controls myocardial inflammation.
| Dilated cardiomyopathy is one of the clinical forms of Chagas' disease (CD) after the infection caused by the parasite Trypanosoma cruzi. Even though strategies adopted in most Latin-American countries in the last decades towards vector control have been effective in reducing the incidence of CD, active transmission is maintained in some regions, and secondary prevention approaches are still required for the infected patients, mostly because the specific anti-parasitic medications are toxic and perhaps of limited efficacy in chronically infected individuals. Moreover, there are no markers to predict the risk of developing dilated cardiomyopathy in asymptomatic, chronically infected patients, although the failure in the mechanisms that control the immune response can be involved in the development of Chagas' heart disease. In this study we show that preserved activity of regulatory T cells and the production of the cytokine IL-17 are connected with a more benign evolution of the disease, which brings a new understanding on the mechanisms associated with progression of CD.
| At the present time, about 7.7 million people are infected and 28 million are at risk of being infected with Trypanosoma cruzi in Central and South America [1]–[3]. This hemoflagellate protozoan is the etiological agent of Chagas' disease. Most of the infected individuals remain asymptomatic during chronic infection (60–70%), characterizing the indeterminate form of the disease. Conversely, 30–40% of chronically infected patients progress to cardiac and/or digestive pathologic involvement [4], [5], and prognostic markers for heart disease progression are required.
A balanced immune response during T. cruzi infection is critical to control the parasite burden in heart and digestive tissues [6], [7]. Production of pro-inflammatory cytokines is required for activation of the effector T lymphocytes responses and is associated with the pathogenesis of Chagas' disease cardiomyopathy (CC), while regulatory cytokines (mainly IL-10) are related to protection [8], [9]. Peripheral blood mononuclear cells (PBMC) from patients with CC produce more IFN-γ, TNF-α and IL-6, and less IL-4 and IL-10, compared to individuals with the indeterminate form of the disease [1], [3], [7], [10]–[14]. However, other studies failed to demonstrate any correlation between production of Th1 and Th2 cytokines profile and the clinical stages of Chagas' disease [15], being that further investigations to elucidate such mechanisms are necessary, one aim of this work.
Regulatory T cells (Treg) are an important source of regulatory cytokines and are involved in the control of the local inflammatory response and in avoiding extensive tissue destruction. However, their presence in the site of infections is frequently regarded as an inducer of parasite persistence [16]. Treg are able to migrate to the site of cardiac inflammation triggered by T. cruzi, and to suppress the effector function of CD4 and CD8 T cells during infectious processes [17]. They suppress the proliferation of effector T cells (CD4+CD25−) when co-cultured, and can also inhibit the activation of auto-reactive T cells through the expression of co-inhibitory molecules (CTLA-4) and the production of suppressive cytokines (IL-10, TGF-β, IL-35) [12], [18], [19]. Recent studies suggest that indeterminate Chagas' disease patients have higher frequency of CD4+CD25high T cells in comparison to cardiac and non-infected individuals in their peripheral blood [20], [21]. Consequently, the measurement of CD4+CD25high T cells suppressive activity in patients with indeterminate and cardiac forms of disease could be an important tool to evaluate a regulatory mechanism that prevents cardiac damage, which was another aim of this work. Treg do not seem to play a major role in regulating the effector responses of CD8 T cells in the myocardium during the acute and chronic experimental T. cruzi infection, since the blockade of CD25 did not change the inflammatory response or parasite burden in mice [13], [22], [23]. However, the treatment with anti-GITR resulted in increased mortality, TNF-α production, and myocarditis with enhanced migration of CD4, CD8, and CCR5 leukocytes to the heart in the T. cruzi infected mice [13]. If Treg could be involved in the control of immune response and cardiac disease progression in Chagas' disease patients is other aim of this work.
An additional lineage of effector CD4+ T helper lymphocytes, with potential regulatory properties, produces IL-17A that acts in several cells types leading the production of GM-CSF, IL-1, IL-6, and TNF-α, activation of NOS2, metalloproteinases and chemokines, resulting in leukocytes recruitment [24]–[27]. Treatment of T. cruzi infected mice with anti-IL-17A mAb lead to increased myocarditis, premature mortality, and decreased parasite load in the heart, suggesting that IL-17 controls the host resistance. Also, IL-17 regulates Th1 cells differentiation, cytokine and chemokine production and the influx of inflammatory cells to the heart tissue [28]. IL-17A−/− mice infected with T. cruzi had a lower survival rate, multiple organ failure, and sustained parasitemia compared with wild-type mice, indicating that IL-17A is crucial to leukocyte activation that are critical for parasite killing [29]. Although it is not very clear, it seems to be a relationship between Tregs and Th17 cells. Differentiation of Th17 in the presence of Treg leads to increased specific cytokine release, what could be due the consumption of IL-2 [30], [31]. Similarly, Treg cell depletion results in a reduced frequency of IL-17 producers through modulation of IL-2 [32]. In addition, Treg can also be converted into a variety of T effector cells, including Th17 cells [33].
The purpose of the present study was to analyze the potential participation of IL-17 and Treg in the development of different clinical manifestations of human chronic Chagas' heart disease. Our hypothesis was that patients with chronic Chagas' disease undergoing cardiomyopathy produce increased levels of IL-17 and have a reduced frequency or suppressive activity of Treg compared with those patients with the indeterminate form of the disease. We provide novel information about immunological mechanisms involved in the human T. cruzi infection that could be used for the development of chemotherapies, as well as for the evaluation of prognostic markers of disease.
The inclusion of the 39 subjects (10 controls) in our investigation had the prior approval of an institutional ethics committee (Hospital das Clínicas de Ribeirão Preto – USP, São Paulo, Protocol number 2285/2007; Brazil). Signed informed consent was obtained from all participants. All patients (n = 29) had at least two positive serology tests for Chagas' disease, as determined by ELISA, immunofluorescence or hemagglutination techniques. All patients underwent a detailed clinical evaluation, 12-lead rest electrocardiogram (EKG), chest X-ray and a 2D-echocardiogram. Twenty one patients had not received etiologic treatment and 8 had received full treatment with benznidazole (5 mg/kg/day) for roughly 60 days. According to their clinical and laboratory characteristics (Table 1), the chagasic patients were divided in 3 groups: Group 1 (n = 10): Patients not treated with benznidazole and not showing signs of or only having mild cardiomyopathy, Group 2 (n = 11): Patients not treated with benznidazole but with moderate/severe cardiomyopathy, Group 3 (n = 8): Patients previously treated with benznidazole (cardiomyopathy-free or mild cardiomyopathy patients). Healthy Individuals from the same endemic areas were included in this study as controls, composing the Group 4 (n = 10). All of them presented negative serologic tests for Chagas' disease and were matched by age and gender with the Chagas' disease patients.
Protein lysate of T. cruzi (Y strain) obtained from LLMCK2 fibroblast cell line was used as the source of antigens. Briefly, the parasites were harvested, washed and submitted to 6 freeze/thaw cycles in liquid nitrogen and 37°C. The lysate was centrifuged at 12,000 g, the supernatant collected and the protein concentration determined.
Peripheral blood was harvested with heparin (50 U/mL) from healthy individuals and Chagas' disease patients. PBMC were isolated using Ficoll-Hypaque (Pharmacia Biotech) density gradient centrifugation, washed, counted, and used for CD4+CD25+ T cell isolation or cultured with specific antigen. PBMC (5×106 cells/mL) were cultured for 48 h with T. cruzi antigen (10 µg/mL) and phytohaemagglutinin (PHA) (1 µg/mL) (Sigma-Aldrich, St. Louis) in 48 wells plates (final volume of 0.5 mL) and labeled with specific antibodies for phenotypic analysis in flow cytometer and determination of cytokine production in the supernatant of PBMC. As the concentration of IL-17 peaked at 48 h culture, we choose this time point for supernatant collection and cytokine assay.
The cultured PBMC were washed in cold phosphate buffered saline (PBS) and samples of 5×105 cells/tube incubated for 30 min at 4°C with PBS-5% rabbit normal serum to block unspecific bidding, followed by the addition of 0.5 µg of phycoerythrin (PE), allophycocyanin (APC) or fluorescein isothiocyanate (FITC)-labeled antibodies anti-CD3, anti-CD4, anti-CD25, anti-GITR, anti-CTLA-4 and anti-CD103 (all from BD-Pharmingen) for additional 30 minutes at 4°C in the dark. To detect the intracellular expression of Foxp3 the cells were fixed with cytofix/cytoperm solution (BD Biosciences) for 15 min at room temperature (RT), washed and stained with anti-Foxp3 or anti-IL17 peridinin chlorophyll protein (PERCP)-labeled, for 30 min at 4°C in the dark. Subsequently, the cells were washed twice and suspended in 100 µL of PBS-1% formaldehyde. In the assays involving intracellular detection of IL-17, the cells were incubated for additional 6 h in the presence of GolgiStop solution, according manufacturer's recommendations (BD Biosciences) and then treated as described above.
Data acquisition was performed using a FACSCanto II (BD) and the multivariate data analysis performed with the FlowJo software (Treestar, USA), after collecting 50,000 events/sample. Distinct gating strategies were used to analyze the regulatory T cell and IL-17-producing CD4 T cell. Characterization of Treg started with gating the lymphocytes on FSC versus SSC dot plot. The T-lymphocyte subpopulations were further selected on FL1 ? anti-CD4 versus FL2 ? anti-CD25 dot plots. The percentage of cells expressing CTLA-4, CD-103, GITR and Foxp3 were analyzed in CD4 T cells, considering three different gates, according to the level of expression (or not) of CD25. The percentage of cells expressing intracellular IL-17 was analyzed within the gate of CD3+CD4+ population.
Cytokine production was assayed in supernatant culture of PBMC stimulated or not with T. cruzi antigen. ELISA sets were IL-10, IL-17, IFN-γ and TNF-α (R&D, Minneapolis, MN), and procedures were undertaken according to manufacturers' instructions. Optical densities were measured at 450 ηm. Results are expressed as picograms per milliliter.
To verify the regulatory function of CD4+CD25+ T cells isolated from PBMC of moderate/severe cardiomyopathy or free/mild cardiomyopathy patients, they were cultured with PBMC (2×105/well) from normal donors, at ratio 1∶5 and 1∶10, in 96-well U-bottom plates, in presence of PHA (1 µg/mL), at 37°C and 5% CO2. CFSE (Molecular Probes) was added at a final concentration of 1.25 µM. The solution was well mixed and incubated at RT for 5 min. An equal volume of serum was used to quench the reaction, after which, the cells were washed with PBS with 5% serum. On day 3 of culture, lymphocytes were collected, washed twice and suspended in 100 µL of PBS-1% formaldehyde. Data acquisition was performed using a FACSCanto II and the multivariate data analysis was performed in the FlowJo software. The data expressed as percentage of inhibition were calculated based on the PHA-induced proliferation of allogeneic T cells cultured without CD4+CD25+ T cells.
Statistical analysis was performed using Mann-Whitney or Kruskal–Wallis tests, performed for the comparison of two or three variables between groups (INSTAT Software; GraphPad). The association between IFN-γ levels, regulatory T cell activity and left ventricular ejection fraction were tested by using the Spearman correlation (INSTAT Software; GraphPad). All values were considered significantly different at P<0.05.
We first aimed to study the ability of cells from patients with different forms of the disease to produce IL-17, IL-10, IFN-γ and TNF-α after T. cruzi antigen stimuli. Similar levels of IL-17 were observed in all groups (Figure 1A). In contrast, cells from free/mild cardiomyopathy patients produced higher amounts of IL-10 than cells from moderate/severe cardiomyopathy patients group (Figure 1B). In addition, the response to T. cruzi antigen regarding the production of TNF-α and IFN-γ was higher in patients with moderate/severe cardiomyopathy (Figure 1C, 1D). The production of IL-17 by CD4+ T cells in PBMC from patients belonging to each experimental group, after being cultured with T. cruzi antigen obtained from trypomastigotes forms was also assessed using flow cytometry analysis. CD3+CD4+IL-17+ T cells from free/mild cardiomyopathy patients (3.73%) displayed increased frequency when compared to healthy individuals (0.99%). Conversely, moderate/severe cardiomyopathy patients (1.23%), Bz-treated patients (1.84%) and healthy individuals had similar frequency of these cells (representative dot plots are shown in Figure 2A). No significant differences were found in the intensities of IL-17 expression (MIF) in CD3+CD4+ T cells among the groups of Chagas' disease patients. When we analyzed the data obtained with the patients of all groups, we found that the percentage of CD4+T cells expressing IL-17 were expressively increased in the cardiomyopathy-free/mild group of patients (1.74±0.92) compared with all the other groups. The mean of the percentage of CD4+T cells expressing IL-17 in moderate/severe cardiomyopathy patients, Bz-treated patients and healthy individuals were 0.99±0.75, 0.90±0.58 and 0.67±0.57, respectively (Figure 2B). These findings were confirmed on confocal examination of PBMC.
To characterize Treg population, CD4 versus CD25 dot plots were done and CD25+ lymphocytes classified in low and high or CD25− T cells (as in Figure 3A). No significant differences in the frequencies of CD4+CD25high, CD4+CD25low and CD4+CD25− T cells were found among patients presenting different clinical forms of the disease as well as in controls (P = 0.118 comparing healthy vs. free/mild cardiomyopathy; P = 0.893, healthy vs. moderate/severe; P = 0.438, healthy vs. treated; P = 0.109, free/mild vs. moderate/severe cardiomyopathy; P = 0.247, free/mild vs. treated; P = 0.494, moderate/severe cardiomyopathy vs. treated) (Figure 3B, C and D). These results suggest that assessing the percentage of CD4+CD25+ could not be a reliable immunological approach to predict the different clinical forms of Chagas' disease.
We next determined the frequency of cell that co-express CD103, GITR, CTLA-4, and Foxp3 on CD4+ T cell expressing high, low or absence of CD25. Free/mild cardiomyopathy patients presented higher frequency of CD4+CD25high T cells expressing Foxp3 (P = 0.033) and CTLA-4 (P = 0.042) than moderate/severe cardiomyopathy patients (Figure 3E). High percentage of CD4+CD25+Low T cells expressing Foxp3 (P = 0.016) and CTLA-4 (P = 0.046) were also observed in free/mild cardiomyopathy patients compared with moderate severe cardiomyopathy patients (Figure 3F). Moreover, severe/moderate cardiomyopathy patients showed lower frequency of CD4+CD25− T cells expressing CTLA-4 (P = 0.035) than free/mild cardiomyopathy patients, and Bz treated Chagas' disease patients (Figure 3G). The mean intensity of fluorescence (MIF) of CTLA-4, CD103, GITR and Foxp3 was similar in all groups studied. Interestingly, the expression of CTLA-4, but not CD103, GITR and Foxp3, in CD4+CD25high T cells was decreased in moderate/severe cardiomyopathy compared with free/mild cardiomyopathy patients and health individuals (Figure 4A). These data show that CTLA-4 expression and frequency of CTL-4+ T cells correlates with less severe cardiac disease. Moreover, it may indicate that treatment with benznidazol, with the consequent parasite elimination, may have important implications in the cardiac disease progression.
We next aimed to study if the reduced frequency of CD4+CD25+ T cell expressing CTLA-4 and Foxp3 that we found in severe cardiomyopathy patients correlated with deficient regulatory activities. CD4+CD25+ T cells from healthy individuals, free/mild cardiomyopathy patients and severe cardiomyopathy patients were sorted, and suppressive activity was evaluated in vitro through co-culture assay with allogeneic T cells stimulated with PHA. The purity of CD4+CD25+ T cells isolated from free/mild cardiomyopathy patients and severe cardiomyopathy patients were about 99%. Interestingly, the inhibitory activity of CD4+CD25+ T cells from healthy individuals (62.95±5.37) (P = 0.0159) and free/mild cardiomyopathy patients (57.40±9.18) (P = 0.0189) were significantly higher than that observed with CD4+CD25+ T cells from moderate/severe cardiomyopathy patients (33.76±4.67), when cultured at a ratio of 1∶5 Treg∶allogeneic T cell (Figure 4B). Of note, no differences were observed among the groups when the ratio of Treg∶effector was 1∶10, possible due to a dilution effect in suppressive activity of these cells (Figure 4C). The impairment in suppressive activity observed in CD4+CD25+ T cells from patients suffering from severe cardiomyopathy correlates with the observation of reduced amounts of CD4+CD25+ T cells expressing CTLA-4 and Foxp3 in this group of patients.
We next correlated LVEF with the levels of IFN-γ in the sera of all Chagas' disease patients and Treg suppressive activity obtained after allogeneic cultures (as described). Our results showed that IFN-γ levels during chronic Chagas' disease are inversely correlated to the LVEF (P = 0.040, r = −0.594) (Figure 5A). Accordingly, the levels of regulatory T cell activity are directly correlated with LVEF (P = 0.022, r = 0.500) (Figure 5B). We thus hypothesized that patients with chronic Chagas' disease undergoing cardiomyopathy produce increased levels of IL-17 and have a reduced frequency or suppressive activity of Treg compared with those patients with the indeterminate form of the disease. To our surprise, however, we found a positive correlation between frequency of CD4+IL-17+ T cell and CD4+CD25+HighFoxp3+ (P = 0.042, r = 0.418) (Figure 5C). In addition, no significant correlation was observed between TNF-α (P = 0.159, r = 0.133), IL-10 (P = 0.265, r = 0.066) production and LVEF.
In this investigation we first evaluated the production of IFN-γ, TNF-α, IL-10 and IL-17 in PBMC obtained from groups of Chagas' disease patients and in a group of benznidazol-treated individuals. The cultures of PBMC from patients with moderate/severe cardiomyopathy produced higher IFN-γ and TNF-α, and lower IL-10 levels than those observed in PBMC culture from free/mild cardiomyopathy patients, which is in accordance with previous reports by other researchers [1], [7], [9]. An imbalance in the production of cytokines IFN-γ and IL-10 was also observed in the present study, assaying these cytokines in the sera from chronic cardiac Chagas' disease patients: This imbalance has been implicated in the pathogenesis of Chagas heart disease [1], [7], [34]. Production of more IFN-γ and less IL-10 in cardiac patients supposedly results in efficient control of parasites replication but with more lesions to myocardium [5].
In addition, the analysis of IFN-γ production by ELISPOT of CD8 T cells from Chagas' disease patients showed that the frequency of IFN-γ producing-CD8 T cells is very low among those patients suffering the most severe form of the disease, and among individuals living in areas of active transmission of the disease, indicating that severe Chagas' cardiomyopathy could be related with the frequency of IFN-γ – producing T cells [31], [35]. On the other hand, one study comparing the levels of mRNA expression for the cytokines IL-5, IL-10, IL-13 and IFN-γ in PBMC from healthy individuals, and patients with cardiomyopathy or indeterminate forms of Chagas disease, found no differences among these groups [11]. Hence, there is not a consensus regarding the exact participation of classic Th1 cytokine profiles in the mechanisms that lead to the cardiac lesions during Chagas' disease.
It is therefore possible that other cytokine and cellular profiles participate in the immunological imbalance observed during Chagas' disease. One candidate is IL-17 which has effectively been involved in the control of parasites and in the induction of myocarditis in T. cruzi experimental infection [24]. In the present study PBMC from free/mild cardiomyopathy patients exhibited a higher expression of IL-17 in CD4+ T cells than that observed in PBMC from patients with severe/moderate cardiomyopathy and in cells from healthy individuals. Likewise, in the experimental model the inhibition of IL-17 resulted in enhanced production of IFN-γ and increased cardiac inflammation [24]. Moreover, impaired activation of immune-related cells that are critical for the killing of T. cruzi is observed in the absence of IL-17A gene [25]. Our data confirmed that PBMC from the group of moderate/severe cardiomyopathy patients produce more IFN-γ and TNF-α and less IL-10 than the cells obtained from the other groups. Cells from the same group of patients expressed more IL-17 when cultured with parasite antigens. In the same way, the infection with the trypanosomatid Leishmania donovani, the etiological agent of Kala Azar (KA), stimulates the differentiation into Th17 cells in PBMC obtained from healthy donors, leading to IL-17 and IFN-γ production [36].
As a result, IL-17 should be important in the control of cardiac inflammation by playing a negative feedback role on the production of IFN-γ and chemokines during T. cruzi infection in humans and mice, modulating the cardiac immune-mediated lesions found in Chagas' disease patients. Here we showed that the production of IL-17 is increased in patients without or with mild cardiac manifestations of the disease, which together with the results showing efficient suppressive activity of Treg in the same group of patients, suggest that IL-17 may be involved in the control of the immune response and, therefore, in the modulation of cardiac disease progression. As pointed before, IL-17 is also crucial for the control of parasite growth and host survival [24], [25]. These data are in agreement with that from a study on a cohort of subjects during a severe outbreak of the infection by the trypanosomatid L. donovani, in which the analysis of Th1, Th2, and Th17 cytokine responses by cultured PBMCs from revealed that IL-17 is associated with protection against severe KA [32].
The frequency of CD4+CD25+ regulatory T cells among patients with different clinical forms of Chagas disease was also examined in the present study. Surprisingly, all groups of patients showed a similar frequency of CD4+CD25+ T cell and CD4+CD25high T cells. This is not in agreement with a previous report showing lower values of CD4+CD25high T cells among school children with the indeterminate form of Chagas disease than that values observed in healthy children [17]. However, the same authors reported later in a study that patients with the indeterminate form of Chagas' disease exhibited a higher frequency of CD4+CD25high T cell expressing Foxp3 and IL-10 as compared to those individuals with cardiomyopathy [37]. The last study was confirmed by a recent report showing that asymptomatic patients had increased amounts of Treg than those with cardiomyopathy [38]. Thus, it is possible to assume that a low frequency of regulatory T cell during early stages of Chagas' heart disease might be associated with the development of more serious chronic manifestations of Chagas' heart disease. As pointed out before, our study does not confirm these data probably due to our very well characterized groups of patients. All patients underwent a detailed clinical evaluation, 12-lead rest electrocardiogram (EKG), chest X-ray and a 2D-echocardiogram. However, in the experimental model of Chagas' disease the inhibition of Treg function with anti-GITR markedly increased the parasitemia, myocarditis and mortality compared with control mice [39].
As we did not detect differences in the percentages of Treg between the groups of patients in the present study, we investigated the suppressive activity of these cells. First, we assayed the expression of surface markers (CD103, CTLA-4 and GITR) as well as the transcriptional factor Foxp3. A higher expression of CTLA-4 and Foxp3 in the CD4+CD25high T cells from free/mild cardiomyopathy patients was observed, when compared to moderate/severe cardiomyopathy patients. Moreover, the analysis of CD4+CD25Low cells population demonstrated that free/mild cardiomyopathy patients and Bz treated patients have a higher occurrence of CTLA-4 than moderate/severe cardiomyopathy patients. The high amount of CD4+CD25+ T cells expressing CTLA-4 and Foxp3 (that activate the regulatory T cell machinery) in CD4+ CD25+ T cells of free/mild cardiomyopathy and Bz treated patients may be related to high suppressor activity of these cells. Furthermore, the greater number of CD4+CD25− T cells of free/mild cardiomyopathy patients and Bz treated patients expressing CTLA-4 than cells from moderate/severe cardiomyopathy patients, probably contributes to the modulation of immune response in the heart. A higher incidence of T cells expressing CTLA-4 among CD4+CD25− T cells from free/mild cardiomyopathy patients when compared to moderate/severe cardiomyopathy patients, also suggest a better negative control of the immune response, since CTLA-4 expression in CD25− T cells is known to suppress the immune response [40].
The suppressive activity of Treg in PBMC from all groups of patients herein described was examined based in their capacity to suppress T cell proliferation. As we suspected, CD4+CD25+ T cells from Chagas' disease patients with severe cardiomyopathy presented reduced capacity to suppress T cell proliferation when compared to free/mild cardiomyopathy patients and healthy individuals. This phenomenon may be correlated by low frequency of CTLA-4 in the CD4+CD25− T cells from cardiac patients. Nevertheless, it was previously reported that cardiomyopathy patients exhibit a higher percentage of CD4+CD25high T cells expressing CTLA-4 [33]. The mechanism leading to reduced expression of Foxp3 and CTLA-4 and consequently, deficient suppressive activity of CD4+CD25+ T cells from patients with cardiomyopathy has not been elucidated. It is possible that a defective control of the immune response by Treg/Th17 may contribute to the pathogenesis of Chagas' heart disease, in a similar way as patients with other inflammatory and autoimmune diseases such as multiple sclerosis, systemic lupus erythematous, type 1 diabetes, psoriasis and rheumatoid arthritis have compromised functional activity of Treg [41], [42].
We also analyzed the levels of cytokines produced by PBMC from patients after in vitro stimulation with T. cruzi antigens, and we showed that IFN-γ production during chronic Chagas' disease is inversely correlated to LVEF, while normal regulatory T cell activity directly correlates with it. In addition, TNF-α production levels were lower in free/mild cardiomyopathy patients than in patients with moderate/severe cardiomyopathy. This finding is in agreement with a previous study reporting that patients with significant left ventricular (LV) dysfunction (LV ejection fraction ≤50%) showed higher levels of TNF-α, compared to Chagas' disease patients without LV dysfunction [43]. Moreover, studies in patients with dilated cardiomyopathy reported a significant increase of TNF-α among these individuals when compared with healthy controls, suggesting that the elevation of TNF-α could be an immune pathogenic mechanism in the progression to cardiomyopathy. Here we showed that the production of TNF-α (and not IFN-γ) tends to be lower among benznidazole-treated individuals. Although further research are required to explore the mechanisms by which benznidazole can induce these differential effects on cytokines production, these findings has been experimentally addressed before, and coincide with our current results. For example, it was shown that IFN-γ mediates the protective effect of benznidazole against T. cruzi infection [44], and slightly inhibits the synthesis of TNF-α in murine cells [45]. The levels of this cytokine may also constitute an important marker of ventricular dysfunction in chronic chagasic cardiomyopathy [46], [47].
One import result found in the present study was a positive correlation between IL-17 and Foxp3 expression in PBMC among Chagas' disease patients. Therefore, more IL-17 and Foxp3 expression is preferentially found in free/mild cardiomyopathy patients. Thus, the expanded Treg are better able to control the inflammatory response in presence of Th17. This data are in agreement with the previous demonstration that Th17 are preferentially differentiated in the presence of Treg [28] due the consumption of IL-2 by Treg [26], [27] Chronic autoimmune inflammation originates when this process is deregulated, and then therapeutic intervention becomes necessary to restore that balance between Th17 and Treg.
It is clear that genetic characteristics of both the host and the parasite are important in determining the outcome of the infection. Our data suggest that genetic aspects of the immune response involved in the functions of Treg, IL-17, and some related genes may deserve further investigation and may shed light on the comprehension of the immune pathogenesis of Chagas' disease.
In summary, our results show that CD4+CD25+ Treg from patients with severe cardiomyopathy display a deficient suppressive activity, leading to uncontrolled production of pro-inflammatory cytokines (TNF-α and IFN-γ) from leukocytes. Moreover, patients with less aggressive forms of the disease (cardiomyopathy free or mild cardiomyopathy individuals) produce higher levels of the cytokines IL-10 and IL-17. Reduced CD4+CD25+ regulatory T cell function and low levels of IL-17 also correlated with more advanced cardiomyopathy. We think that these findings may be helpful in the design of immunotherapeutic approaches for eventual primary, secondary and tertiary prevention of chronic Chagas' cardiomyopathy.
|
10.1371/journal.pcbi.1000624 | Rational Mutational Analysis of a Multidrug MFS Transporter CaMdr1p of Candida albicans by Employing a Membrane Environment Based Computational Approach | CaMdr1p is a multidrug MFS transporter of pathogenic Candida albicans. An over-expression of the gene encoding this protein is linked to clinically encountered azole resistance. In-depth knowledge of the structure and function of CaMdr1p is necessary for an effective design of modulators or inhibitors of this efflux transporter. Towards this goal, in this study, we have employed a membrane environment based computational approach to predict the functionally critical residues of CaMdr1p. For this, information theoretic scores which are variants of Relative Entropy (Modified Relative Entropy REM) were calculated from Multiple Sequence Alignment (MSA) by separately considering distinct physico-chemical properties of transmembrane (TM) and inter-TM regions. The residues of CaMdr1p with high REM which were predicted to be significantly important were subjected to site-directed mutational analysis. Interestingly, heterologous host Saccharomyces cerevisiae, over-expressing these mutant variants of CaMdr1p wherein these high REM residues were replaced by either alanine or leucine, demonstrated increased susceptibility to tested drugs. The hypersensitivity to drugs was supported by abrogated substrate efflux mediated by mutant variant proteins and was not attributed to their poor expression or surface localization. Additionally, by employing a distance plot from a 3D deduced model of CaMdr1p, we could also predict the role of these functionally critical residues in maintaining apparent inter-helical interactions to provide the desired fold for the proper functioning of CaMdr1p. Residues predicted to be critical for function across the family were also found to be vital from other previously published studies, implying its wider application to other membrane protein families.
| Membrane proteins belonging to the Major Facilitator Superfamily (MFS) transport molecules, including drugs, across the membrane and are known to be associated with drug resistance. CaMdr1p is one such MFS major multidrug efflux pump whose over-expression is linked to frequently encountered azole resistance in hospital isolates of C. albicans. Amino acid residues critical for a protein's function are conserved across members of the protein family. However, the traditional measure of conservation is not a useful parameter in mapping a functionally important residue in membrane proteins e.g., hydrophobically conserved stretches form helical transmembrane regions of the protein and are responsible for membrane localization, which individually have limited effect on binding and transport. We have developed a method that uses information theory to score the conservation of a residue relative to its context within the membrane and hypothesize that these residues would be critical for the protein's function. The relevance of predicted residues in the functioning of MFS is validated on CaMdr1p.
| In yeasts, including the pathogenic Candida, an up-regulation of multidrug transporter genes belonging to either ATP Binding Cassette (ABC) or Major Facilitator Superfamily (MFS) is frequently observed in the cells exposed to the drugs leading to the phenomena of multidrug resistance (MDR) [1]. Among the 28 putative ABC and 95 MFS transporter genes identified in the C. albicans genome, only ABC transporters CaCdr1p and CaCdr2p and MFS transporter CaMdr1p, are found to be the major determinants of azole resistance [2],[3]. The reversal of the functionality of these multidrug efflux pump proteins represents an attractive strategy to combat azole resistance.
The major ABC transporters such as CaCdr1p, CaCdr2p bear similar topology and exist as two homologous halves. These, like any other member of the ABC superfamily have four distinct modules: two transmembrane domains (TMDs) each consisting of six transmembrane segments (TMSs) and two nucleotide binding domains (NBDs) located on the cytosolic side of the membrane. Though similar in topology and promiscuity towards substrate specificity, these ABC multidrug transporters of C. albicans also display selectivity to the range of substrates they can export [4].
The transporters belonging to MFS, consists of membrane proteins from bacteria to higher eukaryotes and these are involved in symport, antiport or uniport of various substrates [5],[6]. One of the 17 families of MFS transporters uses the proton motive force to drive drug transport and has been identified in both prokaryotes and eukaryotes [7]. Crystal structures of MFS proteins such as lactose permease (LacY), glycerol-3-phosphate (GlpT), EmrD and oxalate: formate antiporter (OxlT), suggest high structural resemblance among this family of proteins [8]. These consist of 12 TMS, arranged with a similar predicted topology, strongly supporting a common structural architecture or fold across all the MFS transporters [9]–[12]. The fungal MFS members particularly those involved in drug transport are poorly explored in terms of their structure and function [13]. The multidrug MFS transporter CaMdr1p belongs to DHA1 family which is widely distributed and includes both drug-specific and multidrug efflux pumps [14].
Random and site-directed mutational strategies have been extensively used to understand the structure and function of these MDR efflux proteins. For example, random mutational analysis of an ABC transporter, ScPdr5p of budding yeast identified several amino acid residues that alter its substrate specificity and sensitivity to various inhibitors [15],[16]. Tutulan-Cunita et al. observed that several point mutations led to significant changes in drug specificity of ScPdr5p which are distributed throughout the length of the protein [17]. Site-directed mutagenesis followed by an elegant screen done by Golin's group has revealed interactions between TMS 2 and the NBD which may help to define at least part of the translocation pathway for coupling ATP hydrolysis to drug transport mediated by ScPdr5p. Recently, Schmitt et al. have elucidated the role of H1068 in H-loop of ScPdr5p which couples ATP hydrolysis with drug transport [18].
Site-directed mutational analysis of multidrug ABC multidrug transporter CaCdr1p (a close homologue of ScPdr5p) has revealed insight into its drug binding and efflux properties. These studies have implicated some of the amino acid residues of TMS 5, 6, 11 and 12 as the components of the substrate binding pocket(s) of CaCdr1p [19],[20]. Together, these studies suggest that the drug binding sites in CaCdr1p are scattered throughout the protein and probably more than one residue of different helices are involved in binding and extrusion of drugs. However, there is still insufficient information available to predict where and how exactly the most common antifungals such as azoles bind and how are they extruded by CaCdr1p.
Site-directed mutational strategies rely on conservation of residues in a Multiple Sequence Alignment (MSA). The conservation of a residue is calculated from the amino acid frequency distribution in the corresponding column of a MSA. However, the physicochemical conservation is not necessarily responsible for a protein's structure and function but could reflect a more general function such as membrane localization. Thus conservation alone is not sufficient to distinguish between residues responsible for the protein function and membrane localization. Membrane proteins differ from soluble proteins because of their inter-TM hydrophilic and TM hydrophobic propensities, which have allowed the development of efficient membrane protein TM prediction methods [21] and of membrane protein specific substitution matrices [22].
The quantification of residue conservation has evolved over the last few years to the use of information theoretic measures [23]. Relative entropy is a distance measure commonly applied to multiple alignments by comparing the observed frequency distribution with a background distribution. In the present study, we have developed and employed a new method using information theory to rationalize mutation strategies and also applied it to a MFS multidrug transporter CaMdr1p [24]. Relative Entropy (RE) or the Kullback-Liebler divergence is an information theoretic measure of the difference between two probability distributions and has been increasingly applied in bioinformatics to identify functional residues [24],[25]. The use of RE with background frequencies [26] can improve the prediction of a protein's functional residues [27]–[32] as well as detect residues that determine the functional subtype of proteins [28]. Though the basic Kullback-Liebler equation has not changed, its intelligent application in our method calculates Relative Entropy (REM) relative to its context within the membrane. The REM scoring scheme has been improved by treating TM and inter-TM regions of MFS proteins separately which has drastically increased the credibility over the existing methods [23]. In this manuscript, we have compared traditional treatment of conservation, and standard RE, with our improved method. We validated our predictions by replacing the predicted highest REM positions of CaMdr1p with alanine by site-directed mutagenesis. We show that most of these residues when replaced with alanine showed decreased resistance to drugs which was corroborated by abrogated efflux of drugs. Additionally, we could further confirm the functional relevance of each of these high REM residues by predicting their location in deduced 3D model of CaMdr1p and their role in maintaining apparent inter-helical interactions. With this approach, our method enabled us to accurately predict MFS-wide function-specific residues, validated by using CaMdr1p.
A comprehensive non-redundant data set, sourced from all MFS sequences present in the 56.2 release, was generated. This data set was then aligned using a membrane-specific multiple alignment program, which stacked the helices appropriately. A highly conserved residue in a multiple alignment is predicted to have a functional significance. We calculated conservation values using the algorithm from Jalview. Residues shown to be conserved dominate the TM helices, and on closer evaluation are largely hydrophobic residues associated with membrane localization. The traditional relative entropy and our modified treatment of the method (REM) were calculated on the same alignment using scripts written in-house. (Fig. 1 shows a representative section of the alignment along with the REM, RE and conservation scores; see supplementary Table S1 for the REM, RE and conservation scores for the entire MSA). The distribution curve generated on the basis of REM from the MSA is shown in Fig. 2A. Notably the dominant signal using a conventional conservation measure lies in the TM helices and traditional RE also issues high scores to these residues which are less frequent in nature. Our treatment using a REM dampens the membrane localization signals further increasing the signal from atypically occurring conserved residues (Figure S1). Since REM considers conservation as well as the background probability of a residue at a particular alignment position, it is an improved index of the functional significance of a residue. To emphasize this fact, thirty residues with highest values were short-listed each from the conservation list, RE list and REM list. On comparison, it was found that there are thirteen residues shortlisted as both conserved and with high RE while it was found that only nine out of these short-listed columns are both conserved and have high REM (Supplementary Table S2). The thirty alignment positions with high REM were further studied to assess their functional relevance. Expectedly, all residues predicted using REM, are not present in every protein in the family. In CaMdr1p, 16 residues were identical with the most frequently occurring residue in the thirty highest scored alignment columns, and were mutated to alanine to directly validate the prediction (Fig. 2B).
These sixteen out of the top thirty positions, wherein the residue in CaMdr1p matched with the most occurring residue across that alignment position in MSA were analyzed for further studies by site-directed mutagenesis. Interestingly, most of the sixteen residues with high REM turned out to be part of the well-known motifs of the MFS. These motifs are identified as Motif A (GxLaDrxGrkxxl), Motif B (lxxxRxxqGxgaa) which are conserved throughout the MFS, Motif C (gxxxGPxxGGxl) only in 12 and 14-TMS family and Motif D2 exclusive to 12-TMS family [7]. Three out of the sixteen residues short-listed for CaMdr1p; E178, G183 and R184 are a part of Motif A; residues L211, R215 and G219 are a part of Motif B and G256 is a part of motif C. In addition to the known motifs mentioned above, two new motifs have been identified by our study. The residues in these stretches 273Wrxxf277 and 296Pespr300 have high REM scores. However, the known motif D2 does not appear to be highly conserved in our alignment and is thus not predicted to be family-wide function-specific.
All the sixteen residues selected on the basis of high REM were mutated by employing site-directed mutagenesis and were replaced with alanine except G165, G183 and G256 which were replaced by leucine. For functional analysis of the mutant variants, a heterologous hyper-expression system, where GFP-tagged CaMdr1p (CaMDR1-GFP) was stably over-expressed from a genomic PDR5 locus in a S. cerevisiae mutant AD1-8u−, was used [33]. The host AD1-8u− developed by Goffeau's group, was derived from a Pdr1-3 mutant strain with a gain-of-function mutation in the transcription factor Pdr1p, resulting in constitutive hyper-induction of the PDR5 promoter [34]. A single-copy integration of each transformant at the PDR5 locus was confirmed by Southern hybridization (data not shown). Two positive clones of each mutant were selected to rule out clonal variations. These residues with high REM score showed increased drug susceptibility and abrogated efflux of substrates such as [3H] MTX and [3H] FLU (Fig. 3). Of note, there were few exceptions to the list of residues with high REM. For example residues T160, L211, W273 and R274 have high REM values but do not appear to be critical for function of CaMdr1p since the drug susceptibility and efflux are not affected upon replacement of these residues with alanine.
To confirm that the change in susceptibility observed in the mutant variants was not due to their poor expression or mislocalization, we compared the localization of GFP-tagged version of CaMdr1p (CaMDR1-GFP) and its mutant variants by FACS and confocal imaging. A proper localization of all the mutant variants was confirmed by both these methods which showed proper rimmed appearance of GFP-tagged CaMdr1p. The Western Blot analysis further confirmed that the expression levels of CaMdr1p of all the mutant variants were similar thus corroborating FACS and confocal data (Fig. 4).
For evaluating the relevance of high REM residues in CaMdr1p, a 3D homology model was constructed using available crystal structures of lac permease of E. coli (1pv6), glycerol-3-phophate of E. coli (1pw4) and oxalate: formate transporter of O. formigenes (1zc7) as described in Materials and Methods. All the top thirty positions of highest REM are marked in the model which mostly lie in the N-terminal half of the protein (Fig. 5A). Using the homology model, a symmetric contact map of CaMdr1p was generated as discussed in Materials and Methods (Fig. 5B). We exploited this distance plot to ascertain the role of these residues with high REM.
It is apparent that residues L211, R215 and G219 in TMS 4 are within 8 A° distance to many residues of TMS 1, 2 and 3. For example, it can be seen that residue G219 on TMS 4 lie on the same face of the helix and is within 8 A° to the residues G165 and G169 on TMS 2. Indeed, the mutation of predicted G165 and G169 on TMS 2 resulted in abrogated drug susceptibility and transport (data not shown). All the predicted interactions are summarized in Fig. 6A and shown in a pictorial representation of the homology model in Fig. 6B.
The multidrug MFS transporter CaMdr1p harbors a conserved antiporter ‘motif C’ within TMS 5. Our recent study has revealed that the conserved and critical residues of this motif and of TMS 5 are bunched together on the same face of its helical wheel projection and are critical in drug efflux [35]. However, the structure and function aspects of this major multidrug transporter remain poorly understood. To address some of these questions, in this study, we have rationalized conventional mutational strategy and applied computational approach to predict functionally critical residues of CaMdr1p.
The sequence set described in this manuscript represents a comprehensive non-redundant coverage of annotated MFS sequences from SWISSPROT. Many methods have been developed to improve the MSA of membrane protein families for accurate predictions of residues critical for structure and function [36]. Membrane proteins have fold signals which are easily mapped to the primary sequence as TM and inter-TM stretches. Considering the differences in physico-chemical properties of these two regions, membrane protein specific substitution matrices have been developed [22]. However, we argued that a conservation score on the basis of identity or physico-chemical similarity still remains inadequate as the background frequencies of their immediate environmental milieu are radically different with respect to hydrophilic and hydrophobic propensities. This is also apparent from the conservation scores of the MSA wherein a large proportion of the conserved columns correspond to hydrophobic TM regions. Notably, two CaMdr1p residues (F216 and L217) with high conservation but low REM were taken as controls, when replaced with alanine showed no change in the phenotype (data not shown). One of the most basic fold specific signals is the hydrophobic core in globular proteins, and the TM region in membrane proteins. Unlike globular proteins, the hydrophobic TM region is continuous in the membrane protein's primary structure, and indeed this still remains one of the preferred methods to identify membrane proteins, and map their TM regions. While it is intuitive that the synchronous stretch of hydrophobic residues is responsible for membrane localization, the application of a scoring method that can distinguish these residues from family-wide alignment columns associated with other functions has not yet been deployed. In essence, we require a method that can objectively separate the TM signals from other signals. To overcome these limitations, we improved existing method(s) of information theory wherein REM was calculated on the basis of MSA of MFS proteins, keeping in mind the differences in the environmental milieus. We thus treated TM and inter-TM regions by different background probabilities for calculation of REM. These REM scores helped us to predict those sites which have amino acid distributions very different from the respective background distribution thereby statistically predicted to be functionally critical. Not all the residues predicted using REM, are present in every protein in the family. In CaMdr1p, 16 residues were identical with the most frequently occurring residue in the thirty highest scored alignment columns, and were mutated to directly validate the prediction (Fig. 2B). Our results of drug susceptibility assays revealed that almost all of these matched residues with high REM when replaced with alanine displayed sensitivity to the tested drugs and showed abrogated drug transport (Fig. 3). Interestingly, when we mutated residues which had high conservation values but lower REM values (negative control); none showed alterations in drug susceptibilities and thus did not retain the functionally critical stringency as was evident from residues with higher REM. For example, analysis of a few conserved columns of the MSA, such as F216, L217 and L171 having REM values between 0.57-0.44 revealed that their replacement with alanine did not affect the function of CaMdr1p (data not shown). This strengthens the fact that our method takes into account the conservation along with the background frequency and thus lists out residues which affect the function. Also, to check the efficiency of the method, another negative control used was to mutate residues which are having low conservation and low REM values but lie in the vicinity of one of the 16 selected high REM residues. For example, when C225 which is closer to the critical G219 and D235, was mutated to C225A, the functioning of the protein was not affected (data not shown). Similarly, for critical G256, when residues A248, A253 and V254 which are within its vicinity were mutated as A248G, A253G and V254A, the mutant variants continued to behave as WT-CaMDR1-GFP [35].
To further elucidate the role of predicted residues in the functionality of CaMdr1p, a homology model based on the available crystal structures of lac permease, glycerol-3-phophate and oxalate: formate transporter was deduced [9]-[11]. The REM method predicts the relative importance of a residue purely from sequence analysis and is independent of the protein's structure. However, the role a residue plays in the protein's function is not readily apparent from its sequence. We exploited the protein's 3D model as a guide to reason why a residue is functionally critical. The deduced 3D model suggested that similar to other MFS structures, the 12 TM helices of the CaMdr1p span the membrane in such a way that they form the channel pore particularly aligned by residues of TMS 2, 4, 5, 7, 8, 10 and 11. From the deduced homology model of CaMdr1p, a symmetric contact map was generated to highlight the inter-helical interactions of the protein (Fig. 5B). Based on the predictions from the distance map, we could show that many high REM residues are indeed a part of inter-helical interactions (Fig. 6B). It is apparent that more than one residue pair is predicted to be involved in maintaining the interactions between helices (Fig. 6A).
Our aim in developing this method was to identify residues with high specificity which would play a critical role across this entire MFS protein family. Although signals associated with antiporter motifs have been identified using this method, a finer granularity in function such as substrate specificity determining residues is not expected, as these signals would not be family-wide. Since the enlisted residues with high REM values which are functionally critical for CaMdr1p are expected to be family-wide function-specific and thus critical for the entire MFS protein data set, we validated their relevance from the earlier published work. It is known that Motif A of the MFS transporters span an eight residue long loop between TMS 2 and 3 and is suggested to be involved in maintaining a β-turn linking the adjacent TM helices [14]. In the present study, G183 and R184 in the loop between TMS 2 and TMS 3 of CaMdr1p were picked up as family-wide function-specific residues thus corroborating that these residues are a part of Motif A (GxLaDrxGrkxxl) which holds importance throughout the MFS transporters. The hypothesized rocking motion in MFS presumably requires conformational changes in the TMS and the β–turns. In this, the transporter inter-converts between Ci (inward facing) and Co (outward facing) states for translocation of substrates. In glycerol-3-phosphate of E. coli, it was seen that D88 was involved in inter-conversion between these Ci and Co states of the protein [10]. Interestingly, D88 corresponds to E178 of CaMdr1p which also lies in Motif A which upon mutation to alanine turns out to be critical for drug susceptibility and efflux (Table 1).
Motif B (lxxxRxxqGxgaa) of all MFS has a role in energy coupling which spans the N-terminal half of TMS 4 [7]. CaMdr1p contact map reveals that residues in Motif B interface with residues 165GxxxG169 on TMS 2. Motifs rich in glycine and proline residues promote formation of special backbone conformation including kinks in TMS, tight interactions between TMS and very flexible β-turns. In human VAchT, Motif B and the adjacent sequences contain a total of nine notch signatures. A notch allows two helical TMS to approach each other unusually closely because small side chains are located at the interface. R124 of PcaK of Pseudomonas putida which is equivalent to high REM R215 of CaMdr1p of C. albicans is shown to be critical for helix packing [37]. Interestingly, G111 of LacY of E.coli which also occupies a position in the same alignment column is also critical and earlier shown to be a residue at a kink.
Residues from Motif C (gxxxGPxxGGxl) which is exclusive to 12-TMS family are also picked up by our calculations [7]. G150 of LacY of E.coli which is equivalent to high REM G256 of CaMdr1p is function-specific for LacY protein [38]. A stretch of conserved residues 296Pespr300, previously unidentified, at the end of TMS 6 were also predicted with high REM. We have mutated equivalent residues P296A, E297A and T298A of CaMdr1p that overlap with the consensus residues in the stretch and found that cells expressing these mutated variants displayed increased sensitivity to drugs. However, the functional significance of these residues is yet to be established.
There are a few exceptions which emerged from our method. For example, our method did not pick up any residue of Motif D2. This could be an artifact of the method used for alignments in earlier studies. In this study we have employed a membrane protein specific alignment method whereas earlier reports have used standard multiple alignments substitution matrices with smaller data sets. However, when we repeated the alignment using MUSCLE [39] and with the standard substitution matrix (Blosum 62) [40] on the complete data set the motif still did not appear (data not shown). Motif D2 is assumed to have a structural significance as it holds a major kink within TMS 1 but mutations in this motif do not alter the backbone conformation. As an example of the possibly insignificant role of the motif, in human VAchT, the mutation of L49G in this motif completely eliminates propensity for a kink or notch and abolishes activity while normally a glycine itself is expected to be present at this position and is supposed to be involved in maintaining a major kink in this motif [41].
Out of the 16 residues that were mutated, T160A, L211A, W273A and R274A did not lead to any phenotypic changes. It is known that for some of the positions in alignment, the most frequent amino acid does not match with the residue of CaMdr1p at that site. One reason for this could be that some of the functionally important residues co-evolve i.e., these residues may mutate, with compensatory mutation occurring elsewhere in the protein to regain function [42]. T160 where the most frequent residue is a serine at that position may be one such case. Another reason may be that the alignment used in this study involved prediction of TMS with the possibility of errors in demarcating the edges of TM helices. Residues from columns lining the edges of the helices may be wrongly assigned to TM and inter-TM regions. This probably explains the lack of any effect of mutation on residues T160 of TMS 2 and L211 of TMS 4 which lie at the edge of the respective TMS. Other exceptions to our predictions are the mutation of W273 and R274 which though highly conserved and probably a part of the new motif but do not abrogate function upon mutation. Although a few tryptophans in an ABC transporter MRP1, have been shown to be involved in substrate binding and transport [43], generally, in a membrane protein tryptophan residues located on the surface of the molecule are mainly positioned to form hydrogen bonds with the lipid head groups while their hydrophobic rings are immersed in the lipid part of the bilayer [44]. We predict that W273 and R274 may be associated with membrane helix orientation and this function may not be perturbed by mutating them individually through alanine scanning. Alternatively, the tryptophan-arginine residues could be functionally critical in tandem and compensate each other for the loss of either one of them.
Of note, in our predictions, substrate specific residues with high REM are not picked up which predominantly occur in C-terminal of MFS proteins. It should be mentioned that since our alignment considers the entire MFS, residues responsible for substrate specificity would only be selectively conserved within a subfamily and would not have sufficiently strong signals to be visible in this present family-wide study. For this, the same method may be applied to a data set classified on the basis of substrate selectivity to identify residues critical to the functioning of that subfamily.
There are a number of conservation methods known but none has yet achieved both biological and statistical rigor. We have used REM to separate conserved residues from the background function of TM localization. The interpretations support the well-known fact that MFS has a conserved N-terminal half which has residues important for maintenance of a specific fold for this class of proteins while C-terminal half has a more specific role in substrate binding and recognition [7]. Taken together, our study provides an insight into the molecular details of MFS transporters in general and CaMdr1p in particular. Our method of scaled REM calculations improves its performance over other information theoretic methods. Additionally, this study also provides a method for rational mutational analysis not only for MFS proteins but can be applied to any class of membrane proteins and thus makes it possible to predict and locate family-wide functionally relevant residues.
Anti-GFP monoclonal antibody was purchased from BD Biosciences Clontech, Palo Alto, CA, USA. DNA modifying enzymes were purchased from NEB. The drugs cycloheximide (CYH), 4-Nitroquinoline oxide (4-NQO), Methotrexate (MTX) and Protease inhibitors (Phenylmethylsulfonyl fluoride, Leupeptin, Aprotinin, Pepstatin A, TPCK, TLCK) and other molecular grade chemicals were obtained from Sigma Chemicals Co. (St. Louis, MO, USA) Fluconazole (FLU) was generously provided by Ranbaxy Laboratories, India. [3H] Fluconazole was custom prepared and [3H] Methotrexate (MTX) was purchased from Amersham Biosciences, United Kingdom.
Plasmids were maintained in Escherichia coli DH5α. E.coli was cultured in Luria-Bertani medium (Difco, BD Biosciences, NJ, USA) to which ampicillin was added (100 µg/ml). The S. cerevisiae strain used was AD1-8u− (MATa pdr1-3 his1 ura3 Δyor1::hisG Δsnq2::hisG Δpdr5::hisG Δpdr10::hisG Δpdr11::hisG Δycf1::hisG Δpdr3::hisG Δpdr15::hisG), provided by Richard D. Cannon, University of Otago, Dunedin, New Zealand [33], [34]. The yeast strains used in this study are listed in the Supplementary Table S3. The yeast strains were cultured in YEPD broth (Bio101, Vista, CA, USA) or in SD-ura− dropout media (0.67% yeast nitrogen base, 0.2% dropout mix, and 2% glucose; Difco). For agar plates, 2.5% (w/v) Bacto agar (Difco, NJ, USA) was added to the medium.
|
10.1371/journal.pntd.0003963 | The Rapid-Heat LAMPellet Method: A Potential Diagnostic Method for Human Urogenital Schistosomiasis | Urogenital schistosomiasis due to Schistosoma haematobium is a serious underestimated public health problem affecting 112 million people - particularly in sub-Saharan Africa. Microscopic examination of urine samples to detect parasite eggs still remains as definitive diagnosis. This work was focussed on developing a novel loop-mediated isothermal amplification (LAMP) assay for detection of S. haematobium DNA in human urine samples as a high-throughput, simple, accurate and affordable diagnostic tool to use in diagnosis of urogenital schistosomiasis.
A LAMP assay targeting a species specific sequence of S. haematobium ribosomal intergenic spacer was designed. The effectiveness of our LAMP was assessed in a number of patients´ urine samples with microscopy confirmed S. haematobium infection. For potentially large-scale application in field conditions, different DNA extraction methods, including a commercial kit, a modified NaOH extraction method and a rapid heating method were tested using small volumes of urine fractions (whole urine, supernatants and pellets). The heating of pellets from clinical samples was the most efficient method to obtain good-quality DNA detectable by LAMP. The detection limit of our LAMP was 1 fg/µL of S. haematobium DNA in urine samples. When testing all patients´ urine samples included in our study, diagnostic parameters for sensitivity and specificity were calculated for LAMP assay, 100% sensitivity (95% CI: 81.32%-100%) and 86.67% specificity (95% CI: 75.40%-94.05%), and also for microscopy detection of eggs in urine samples, 69.23% sensitivity (95% CI: 48.21% -85.63%) and 100% specificity (95% CI: 93.08%-100%).
We have developed and evaluated, for the first time, a LAMP assay for detection of S. haematobium DNA in heated pellets from patients´ urine samples using no complicated requirement procedure for DNA extraction. The procedure has been named the Rapid-Heat LAMPellet method and has the potential to be developed further as a field diagnostic tool for use in urogenital schistosomiasis-endemic areas.
| Human schistosomiasis is a disease caused by several species of parasitic worms of the genus Schistosoma that is affecting 200 million people, especially in sub-Saharan Africa. Most people are infected with Schistosoma haematobium, the species that causes urogenital schistosomiasis and also bladder cancer in many chronic infections. The definitive diagnostic test is based on microscopic examination of urine samples to detect parasite eggs. This method has low sensitivity, high day-to-day variability and cannot be carried out in the acute phase of the disease since the parasite has not started yet to lay eggs. New high-throughput diagnostic tools would be desirable, permitting early treatment and preventing the pathology associated with chronic infections. An interesting approach is the loop-mediated isothermal amplification (LAMP) technique because of its simplicity in operation and potential use in clinical diagnosis and surveillance of infectious diseases. In this study, we developed and evaluated a LAMP assay for detection of S. haematobium DNA in patients´ urine samples using heated pellets with no complicated requirement procedure for DNA extraction, namely the Rapid-Heat LAMPellet method. This is a new, easy, rapid and cost-effective LAMP method that should prove useful for mass screening in limited-resource settings in urogenital schistosomiasis-endemic areas.
| Human schistosomiasis, a parasitic freshwater snail transmitted disease caused by several species of genus Schistosoma trematode worms, is one of the 17 neglected tropical diseases (NTDs) considered by World Health Organization (WHO) [1]. It is estimated that 732 million persons are at risk of infection worldwide and over 200 million people are infected with this disease in 74 different countries, especially in sub-Saharan Africa [2–4], where both associated morbidity and mortality are a significant barrier to social and economic development [5–7]. It must be also observed that the prevalence of imported schistosomiasis is increasingly a problem in non-endemic areas due to the growing number of international travellers to endemic areas, expatriates and immigrants from endemic countries [8–10]. Although humans are mainly infected by five species of schistosomes, namely Schistosoma mansoni, S. haematobium, S. japonicum, S. mekongi, and S. intercalatum, the main burden of disease in sub-Saharan Africa is usually attributed to two species referred to as the major human schistosomes: S. mansoni, causing hepatic and intestinal schistosomiasis and S. haematobium, the chief cause of urogenital schistosomiasis [3].
More people are infected with S. haematobium than with the other schistosomes; it is estimated that 112 million people suffer from urogenital schistosomiasis [11–14]. The infection typically results in haematuria, anaemia, dysuria and genital and urinary tract lesions, but in severe cases it may also lead to kidney damage. It is well known that the deposition of S. haematobium eggs eventually leds to squamous cell carcinoma of the bladder in many chronically infected individuals [15, 16] the International Agency for Cancer Research (IACR) in association with WHO classified S. haematobium as a Group 1 biological carcinogen [17]. Moreover, most of women infected with S. haematobium suffer from female genital schistosomiasis of the lower genital tract [13]; which impairs fertility [18] and also increases susceptibility of the woman to HIV [19].
For the diagnosis of urogenital schistosomiasis, the gold standard remains microscopic detection of excreted ova in urine samples [20] after using either sedimentation/centrifugation or filtration methods [21]. These conventional methods are inexpensive, easy to perform under field conditions and relatively rapid. However, parasitological diagnosis has classically low sensitivity, especially in low-grade infections and may be affected by day-to-day variability in egg excretion, often missing diagnosis by microscopy [22, 23]. In addition, egg count-based criteria cannot be carried out in the acute phase of the disease since the parasite have not yet started to produce eggs. The collection of a larger number of urine samples per individual on consecutive days instead of a single one may increase the sensitivity of microscopic detection, but is more expensive and also time-consuming [23]. Identifying blood in the urine-micro or macrohaematuria- has been widely and successfully used as a good indicator of S. haematobium infection, mainly in a high prevalence situation. However, haematuria is a nonspecific symptom of urogenital schistosomiasis in areas of low endemicity and can be incorrectly estimated depending on the infection prevalence in an area [24, 25]. Antibody-based assays are useful to confirm S. haematobium infections, but do not distinguish active infection from past exposure, and so low sensitivity and specificity results frequently occur. Moreover, antibody tests are usually negative during acute symptomatic urogenital schistosomiasis. On the other hand, assays that detect circulating antigens seem very promising in the early phase of infection but still lack sensitivity in the diagnosis of light infections [20, 26, 27].
To overcome the drawbacks of both classical parasitological and immunological diagnostic methods, the development of new, more sensitive and specific molecular diagnostic tools for the diagnosis of urogenital schistosomiasis are desirable and still needed. In recent years, several studies have reported the utility of polymerase chain reaction (PCR)-based assays for sensitive and specific detection of S. haematobium DNA in human urine [28–30] and serum [31] samples. However, the PCR-based technologies are not widely used in low-income S. haematobium endemic countries because skilled operators and costly equipment are needed.
In this way, the loop-mediated isothermal amplification (LAMP)
assay [32] offers a field-friendly alternative to PCR-based technologies as it is less time consuming than PCR and can be performed using a simple heating block or water bath, with results read by the naked eye under natural or UV light [33, 34]. Additionally, LAMP reagents can be storage at room temperature for weeks [35], the reaction shows low susceptibility to typical inhibitory compounds occurring in samples [36–38], its robustness against variation of reaction conditions such as pH and temperature has been described [39] and it can operate with minimal handling and processing of DNA samples for amplification [40], [41–43], or even without prior DNA extraction [36]. Thereby, considering these salient advantages over most DNA-based amplification tests, LAMP technology shows a potential use in clinical diagnosis and surveillance of infectious diseases, particularly under field conditions for most NTDs [44, 45].
Several successful approaches for LAMP assay for Schistosoma spp. detection have been recently reported in laboratory settings using experimentally infected animals, such as S. japonicum in rabbits [46, 47] or S. mansoni in mice [48], as well as in field settings for monitoring infected snails with S. mansoni, S. haematobium [49, 50] and S. japonicum [51, 52]. Additionally, a LAMP to detect S. japonicum in human sera has been also reported [53].
Thus, with the aim to develop new, applicable and cost-effective molecular tools for the diagnosis of urogenital schistosomiasis, in our work we have developed a new sensitive and specific LAMP assay for detection of S. haematobium in human urine samples. In this study, the effectiveness of the LAMP assay was evaluated in a number of patients´ urine samples with parasitological proven infection with S. haematobium. Different fractions of urine samples (whole urine, supernatants and pellets) as well as different methods for DNA extraction were used to compare results and cost-effectiveness. To the best of our knowledge, this is the first report using LAMP assay for detection of S. haematobium in human urine samples.
Human urine samples used in this study were obtained as part of public health activities at Hospital Universitario Insular, Las Palmas de Gran Canaria, Spain. Later, samples were sent and stored at CIETUS, University of Salamanca, Spain, for further analyses. Human urine samples were not collected specifically for this study and all were obtained under written informed consent and coded and tested as anonymous samples. Participation of healthy urine donors for obtaining simulated artificial urine samples was voluntary. All participants were given detailed explanations about the aims, procedures and possible benefit of the study. The study protocol was approved by the institutional research commission of the University of Salamanca. Ethical approval was obtained from the Ethics Committee of the University of Salamanca (protocol approval no. 48531).
In the first procedure for DNA extraction we used the i-genomic Urine DNA Extraction Mini Kit (Intron Biotechnology, UK) following the manufacturers´ instructions. DNA samples thus obtained were stored at -20°C until use in LAMP reactions.
In the second procedure, we used the hot NaOH extraction method [54] with minimal modifications in the standard protocol by adding sodium docecyl sulfate (SDS) to ensure disruption of the S. haematobium eggs to release the DNA. Briefly, an equal volume of a 50 mM NaOH solution containing 0.1% of SDS was added to urine aliquots of 100 μl and then heated at 95°C for 30 min. Subsequently, the tubes were centrifugated at 5000 rpm for 5 min and a volume of 50 μL of supernatant was recovered in a new clean tube and mixed with an equal volume of a 1 M Tris-HCl solution at pH 8.0. Each new solution thus obtained was stored at -20°C until further use as template in LAMP assays.
In the third procedure,-named the “Rapid-Heat LAMP method”-, each aliquot of whole urine, supernatant and pellet obtained from each urine sample was heated at 95°C for 15–20 min and then briefly spun to pellet the debris. After this, 2 μL of the supernatant were used immediately as template for each LAMP reaction. The remaining volume of each sample was stored at -20°C. To obtain DNA to be used as template in LAMP reactions to test the remaining 76 clinical urine samples included in the study, we firstly obtained the urinary sediment (pellet) as already indicated and, subsequently, the Rapid-Heat LAMP method was applied.
A set of six oligonucleotide primers were used for the LAMP assay, targeting eight regions in the 2522 base pair (bp) sequence of S. haematobium ribosomal intergenic spacer (IGS) DNA retrieved from GenBank (Accession No. AJ223838) [55]. The outer forward primer (F3), outer backward primer (B3), forward inner primer (FIP), backward inner primer (BIP), and loop forward (LF) and backward (LB) primers were designed using the online Primer Explorer V4 software (Eiken Chemical Co. Ltd, Tokyo, Japan; http://primerexplorer.jp/elamp4.0.0/index.html) according to the general criteria described by Notomi et al. [32] and finally selected based on the criteria described in “A Guide to LAMP primer designing” (http://primerexplorer.jp/e/v4_manual/index.html). The location and nucleotide sequences of the six primers are shown in Fig 1. All the primers were of HPLC grade (Thermo Fisher Scientific Inc., Madrid, Spain). To confirm the specificity for the designed primers in annealing exclusively with the S. haematobium DNA correct target sequence, a BLASTN local search and alignment analysis [56] was carried out in different online databases against currently available nucleotide sequences for other organisms (NCBI; http://blast.ncbi.nlm.nih.gov/Blast.cgi) as well as specifically against human, murine (Ensembl; http://www.ensembl.org/Multi/Tools/Blast) and other related Schistosoma species genomes (Sanger Institute; http://www.sanger.ac.uk/resources/software/blast/).
The outer LAMP primer pair, designated F3 and B3, was initially tested for the amplification of S. haematobium DNA by a touchdown-PCR (TD-PCR) to verify whether the correct target was amplified. The PCR assay was conducted in 25 μL reaction mixture containing 2.5 μL of 10x buffer, 1.5 μL of 25 mmol/L MgCl2, 2.5 μL of 2.5 mmol/L dNTPs, 0.5 μL of 100 pmol/L F3 and B3, 2 U Taq-polymerase and 2 μL (1 ng) of DNA template. Conditions for TD-PCR amplification were as follows: an initial denaturation was conducted at 94°C for 1 min, followed by a touchdown program for 15 cycles with successive annealing temperature decrements of 1.0°C every 2 cycles. For these 2 cycles, the reaction was denatured at 94°C for 20 s followed by annealing at 58°C-55°C for 20 s and polymerization at 72°C for 30 s. The following 15 cycles of amplification were similar, except that the annealing temperature was 54°C. A final extension was performed at 72°C for 10 min.
The specificity of PCR using outer primers F3 and B3 was also tested with 20 heterogeneous DNA samples from other parasites included in the study. The sensitivity of the PCR was also assayed to establish the detection limit of S. haematobium DNA with 10-fold serial dilutions ranging from 0.5 ng/μL to 0.5 atg/μL prepared as mentioned above. The assays were performed with 2 μL of the diluted template in each case, thus resulting a final concentration of DNA ranging from 1 ng/μL to 1 atg/μL. Negative controls (ultrapure water instead of DNA template) were included in each run. The PCR products (5–10 μL) were subjected to 2% agarose gel electrophoresis stained with ethidium bromide and visualized under UV light.
To evaluate the LAMP primer set designed in S. haematobium DNA amplification, we set up the reaction mixture using Bst 2.0 WarmStart DNA polymerase (New England Biolabs, UK) combined with different betaine (Sigma, USA) and MgSO4 (New England Biolabs, UK) concentrations. Thus, LAMP reactions mixtures (25 μL) contained 1.6 μM of each FIP and BIP primers, 0.2 μM of each F3 and B3 primers, 0.4 μM of each LB and LF primers, 1.4 mM of each dNTP (Bioron), 1x Isothermal Amplification Buffer -20 mM Tris-HCl (pH 8.8), 50 mM KCl, 10 mM (NH4)2SO4, 2 mM MgSO4, 0.1% Tween20- (New England Biolabs, UK), betaine (ranging 0.8, 1 or 1.2 M), supplementary MgSO4 (ranging 4, 6 or 8 mM) and 8 U of Bst 2.0 WarmStart DNA polymerase with 2 μL of template DNA. To establish the standard protocol for LAMP reactions mixtures assayed, a range of temperatures (61, 63 and 65°C) was tested in a heating block for 30, 50 or 60 min and then heated at 80°C for 5–10 min to inactivate the enzyme and thus to terminate the reaction. Then, both optimal temperature and incubation time were determined and used in the following tests. Positive (S. haematobium DNA) and negative (no DNA template) controls were always included in each LAMP assay.
To estimate the accuracy of the LAMP assay as a diagnostic test, the percentages of sensitivity, specificity, positive predictive value (PPV) and negative predictive value (NPV) were calculated using the MedCalc statistical program version 15.2.2 (MedCalc Software, Ostende, Belgium) according to the software instruction manual (www.medcalc.org).
To confirm that the expected target was amplified, a PCR reaction was performed using outer primers F3 and B3 to amplify S. haematobium DNA. Thus, a 199 bp amplicon was successful obtained (Fig 2A). In order to determine the lower detection limit of the PCR reaction, a 10-fold serial dilution ranging from 10−1 to 10−9 of S. haematobium DNA was amplified. The minimum amount of DNA detectable by PCR using outer primers was 1 ng (Fig 2B). According to specificity, when DNA samples obtained from other parasites included in the study were subjected to this PCR assay, amplicons were never amplified (Fig 2C).
To establish a standard procedure for the LAMP assay we used the Bst 2.0 WarmStart DNA polymerase applying a range of temperatures (61, 63 and 65°C) for testing different mixtures containing variable concentrations of betaine (ranging 0.8, 1 or 1.2 M) combined with supplementary variable concentrations of MgSO4 (ranging 4, 6 or 8 mM) in a heating block for 30, 50 and 60 min. The best amplification results were obtained when the reaction mixture contained 1 M of betaine combined with supplementary 6 mM of MgSO4 (resulting a final concentration of 8 mM MgSO4 in 1x Isothermal Amplification Buffer) and was incubated for 50 min at 63°C in a heating block (Fig 3A). Once the most favourable conditions and molecular components were established for the LAMP assay, all positive results in subsequent reactions could be clearly visually observed by the naked eye by inspecting the colour change after adding SYBR Green I as well as the typical ladder of multiple bands after electrophoresis on agarose gels.
To determine the specificity of the primers designed, a panel of 20 DNA samples from other parasites were subjected to the LAMP assay. As shown in Fig 3B, only LAMP products were amplified when S. haematobium DNA was used as template and no false positive amplification was observed, thus indicating the high specificity of the established LAMP assay.
Regarding to the sensitivity of the LAMP assay, a 10-fold serial dilution of S. haematobium genomic DNA was amplified by LAMP. The results indicated that the detection limit for the LAMP reaction was 100 fg (Fig 3C). This suggested that the LAMP assay is 104 times more sensitive than the PCR using outer primers F3 and B3 (see Fig 2B). On the other hand, the sensitivity of LAMP assay in simulated fresh human urine samples artificially contaminated with DNA from S. haematobium was also examined. In this case, the detection limit of LAMP assay was 10 fg/μL when performing the DNA extraction with the commercial kit (Fig 4A), whereas the detection limit was established in 1 fg/μL using the Rapid-Heat LAMPellet method for DNA extraction (Fig 4B).
Comparative LAMP results obtained when testing aliquots of whole urine, supernatants and pellets from patients´ urine samples with parasitological confirmed S. haematobium infection after using the three different DNA extraction methods attempted in our study are shown in Figs 5, 6 and 7, respectively.
In LAMP tests using a starting volume of whole patients´ urine samples of 100 μL/each we obtained 15/18 positive results when performing DNA extraction using the i-genomic Urine DNA Extraction Mini Kit (Fig 5A), 11/18 when using the NaOH/SDS extraction method (Fig 5B) and 12/18 when the Rapid-Heat LAMP method was applied (Fig 5C).
In LAMP tests for supernatant fraction of patients´ urine samples we obtained only 3/18 positive results when performing DNA extraction using the i-genomic Urine DNA Extraction Mini Kit (Fig 6A), 4/18 when using the NaOH/SDS extraction method (Fig 6B) and 9/18 when the Rapid-Heat LAMP method was applied (Fig 6C).
Finally, in LAMP tests for the urinary sediment (pellet) obtained from the urine samples we obtained 17/18 positive results when performing DNA extraction using the i-genomic Urine DNA Extraction Mini Kit (Fig 7A), 15/18 when using the NaOH/SDS extraction method (Fig 7B) and a total of 18/18 when the Rapid-Heat LAMP method was applied (Fig 7C). Thus, in general, the higher effectiveness in LAMP amplification of S. haematobium DNA in patients´ urine samples was obtained when the urinary sediment (pellet) was used for DNA extraction; moreover, the simple Rapid-Heat LAMP method provided the best results of the three methods assayed for extracting DNA detectable by LAMP. Thereby, the minimal pellet obtained from urine samples, in addition to the Rapid-Heat LAMP method for DNA detection-hereafter "Rapid-Heat LAMPellet method"-, was set up as the most advantageous procedure to be used in successive LAMP reactions to detect S. haematobium DNA in urine samples and to test all the clinical samples included in our study.
The results of all 94 patients´ urine samples evaluated by duplicated for S. haematobium DNA detection by using the Rapid-Heat LAMPellet method are presented in Table 1. We obtained LAMP positive results in 18/18 confirmed S. haematobium infected urine samples, in 1/9 urine samples with other helminths species confirmed infections (specifically a patient infected with a "hookworm"), in 1/5 urine samples with other agents confirmed infections (specifically a patient infected with Trichomonas vaginalis), in 1/15 urine samples from patients with eosinophilia without a confirmed diagnosis and, finally, in 5/24 urine samples from patients without either eosinophilia and none apparent disease. The seven parasitological S. mansoni-positive urine samples as well as the 16 urine samples from healthy non-endemic donors (used as negative controls samples) were all negative by LAMP. All positive results could be visually observed in tubes by color change after adding SYBR Green I and also after electrophoresis on agarose gels as a ladder of multiple bands of different sizes (S1 Fig). Considering the results obtained, diagnostic parameters for sensitivity and specificity were calculated for our LAMP assay, 100% sensitivity and 86.67% specificity, and also for microscopy detection of eggs in urine samples, 69.23% sensitivity and 100% specificity. The PPV and NPV for both LAMP assay and microscopy were also calculated; all statistic data obtained are showed in Table 1.
Urogenital schistosomiasis due to S. haematobium remains a serious underestimated public health problem, particularly in sub-Saharan Africa. Frequency of urogenital schistosomiasis in travellers, expatriates and migrants is in the same range to that of intestinal schistosomiasis due to S. mansoni [57]. As there is no vaccine to protect against schistosomal infection, mass praziquantel treatment of populations at risk of infection is being conducted routinely in many endemic areas; however, reinfections rapidly occur because of recurrent direct contact with water infected with parasites [58]. Considering the current problems of parasitological, serological and molecular methods in detecting schistosomal infections [59], new, simple, accurate and affordable diagnostic tools are essential for providing specific treatment and for maximizing the success of control of urogenital schistosomiasis in endemic areas; as well as for monitoring drug effectiveness.
Point-of-care tests are being developed as economic evaluation diagnostic technologies for infectious diseases control strategies as they are easy to use and interpret, require minimal laboratory infrastructure, are cost-effective, reduce patient waiting time and potentially therefore reduce loss to follow-up, and may have comparable or higher sensitivity to microscopy [60]. The LAMP technology-as a DNA amplification method- combines rapidity, simplicity and high specificity [32] and has a wide range of possible applications, including point-of-care testing in developing countries [61, 62]. We have developed a LAMP assay for rapid, sensitive, specific and cost-effective detection of S. haematobium in human urine samples, even in the absence of parasites eggs in excreta, as a basis for a potential field diagnostic tool for use in schistosomiasis endemic areas. Besides its excellent performance, the most striking results of this study are the simplicity to perform the whole process without requiring DNA extraction from a small volume of starting urine to get the urinary sediment (pellet) to carry out the molecular analysis. We have named this simple procedure the "Rapid-Heat LAMPellet method".
To accomplish its development, we designed a specific set of six primers targeting eight regions in a species specific sequence of S. haematobium ribosomal IGS [55]. The ribosomal IGS regions within Schistosoma species generally contain unique sequence motifs which are specific to that group of organisms. In addition, the IGS target locus has been already used for successful detection of Schistosoma spp. infection in freshwater snails by real-time PCR and oligochromatographic dipstick rapid technology (PCR-OC) [63]. Several other advantages of these sequences to be use in molecular studies have been already reported elsewhere [55, 64].
Once the primer set was designed, in silico comparisons of the expected 199 bp sequence with the on line available genomes showed the higher homology in alignment length with S. haematobium and no cross-reaction was found, specifically with S. mansoni; this result is especially important as these two species are the main schistosomes producing co-infections in most areas of sub-Saharan Africa [58]. Specificity results obtained in in silico were later verified by PCR using outer primers F3-B3.
After this, we attempted to establish the most suitable reaction mixture for the six specific primers in the LAMP assay. We used the Bst polymerase 2.0 WarmStart as this warm-start version has several advantages compared to wild-type Bst DNA polymerase large fragment, such as faster in obtaining amplification signals [65] and increased stability at room temperature [66]. These features are important when testing a large number of samples under field conditions in endemic areas where limited resources for the maintenance of a cold chain exists. As the LAMP reaction might be facilitated by the addition of loop primers [67] our LAMP assay designed was accelerated by the addition of a pair of loop primers, thus allowing to amplify successfully S. haematobium DNA in only 50 min, whereas a previously described LAMP assay to amplify S. haematobium DNA in freshwater snails takes 120 min to complete the reaction [49, 50].
The specificity of the LAMP assay was determined using a panel of heterogeneous control DNA samples of a number of parasites. The assay specifically produced typical ladder patterns from the target sequence only for S. haematobium DNA. The sensitivity of the LAMP resulted 104 times greater than that of PCR using outer primers (100 fg vs. 106 fg or 1ng, respectively). It is usually considered that LAMP is highly sensitive compared to conventional PCR methods and other studies also found a higher sensitivity when comparing LAMP results in contrast to PCR in amplification of DNA from Schistosoma species, including S. japonicum [47], S. haematobium and S. mansoni [49, 48].
The effectiveness of our LAMP assay was assessed in patients´ urine samples with confirmed S. haematobium infection by microscopic examination. Bearing in mind a potential easy and cost-effective large-scale application in field conditions, we evaluated different DNA extraction methods for their ability to isolate DNA from small volumes of different fractions of human urine samples, including whole urine, urine supernatant and urinary sediment (pellet) to compare results. A simple, quick and economically DNA extraction method for use in combination with small volumes of clinical urine specimens could greatly reduce the infrastructure requirements of collecting, handling, storing and processing the patients´ samples in schistosomiasis endemic areas where limited resources exist.
The three different DNA extraction methods tested in our work were much more efficient in extracting detectable DNA by LAMP when using aliquots of whole urine and pellets than supernatants. This seems to be logical since after centrifugation to remove and retain supernatants, both potential free S. haematobium DNA and parasite eggs-and therefore containing DNA- found in patients´ urine samples should be concentrated at the bottom of the tube, thus improving the sensitivity of the DNA molecular detection methods, as previously described [68]. When using the pellets, the simple rapid-heating method allowed us to obtain a very good-quality detectable DNA that did not compromise LAMP amplification and all the S. haematobium-positive urine samples tested were successfully amplified.
The consistent results in DNA obtained from aliquots of whole urine and pellets when applying a commercial kit may be due to the well-known effectiveness of this procedure to isolate genomic DNA from urine samples suitable for further molecular analyses [69]. Urine specimens contain many inhibitors which may interfere in DNA amplification [70], so removing inhibitors as much as possible by using a kit is convenient to ensure that DNA will be subsequently efficiently amplified. However, since this procedure could be very expensive to use when a large number of samples must be tested, an inexpensive and simple rapid-heating method is much more advantageous. It is also known that DNA purification from samples could be omitted in LAMP reactions, since LAMP assays have shown a significant tolerance to inhibitor substances derived from a number of biological samples [71], [72], [73]. Additionally, other LAMP assays with high sensitivity and no complicate requirement procedure for DNA extraction have been developed for molecular detection and diagnostic of bacterial [74] and parasitic [75] diseases in urine samples. Moreover, a simple heating DNA obtaining method has been also successfully applied with other clinical samples, such us blood [41] and swaps [42] in LAMP amplification of both Plasmodium and Leishmania species nucleic acids, respectively.
To really establish the sensitivity of our LAMP assay in urine samples that most closely resembled the patients´ urine specimens analyzed, we used a panel of simulated human urine samples artificially spiked with S. haematobium genomic DNA. For these samples, to extract DNA as template in LAMP we used both the commercial kit and the rapid-heat methods since these procedures showed the highest efficiency to obtain detectable DNA by LAMP in S. haematobium-positive clinical samples. After extracting DNA with the commercial kit, LAMP detection limit resulted tenfold higher than that obtained using S. haematobium genomic DNA 10-fold serially diluted without DNA extraction (10 fg vs. 100 fg, respectively). Unexpectedly, when heating the simulated samples, we obtained a limit of detection tenfold higher than that obtained when using purified DNA samples by the commercial kit (corresponding to 1 fg vs. 10 fg, respectively). An increased sensitivity has been also reported when using crude DNA extraction methods compared with a commercial method (i.e. DNazol) for template preparation from the pellets or supernatants of nasopharyngeal aspirates for LAMP detection of adenovirus [76]. Thus, the sensitivity value of 1 fg was considered as the lower limit of the detection threshold of the LAMP assay in detecting S. haematobium DNA in human urine samples. By reference, as S. mansoni genome contains approximately 580 fg of DNA [77], theoretically our LAMP assay would detect S. haematobium diluted DNA in urine samples corresponding to less than the equivalent to a single parasite cell. Such sensitivity is a feature of great value to overcome the difficulties of detecting urogenital schistosomiasis in areas of low transmission or in individual cases with a very low worm burden.
Then, taking into account both the high sensitivity and the good-quality detectable S. haematobium DNA by LAMP in easy to obtain and handling heated pellets from clinical urine samples, we tested the remaining 76 specimens included in our study by the Rapid-Heat LAMPellet method. We obtained negative results by LAMP in all parasitologically S. mansoni-positive urine samples tested (corroborating again that no cross-reaction with that schistosome species occurs) and also in urine samples from healthy non-endemic donors used as negative controls. Nevertheless, eight LAMP positive results were obtained when testing patients´ urine samples from other groups which were formerly microscopy-confirmed as S. haematobium-negative. It may be rational to consider that those eight LAMP positive results are truly S. haematobium-infected samples which were undetected in the microscopic analysis since this method is very low sensitive, especially in low-grade infections and high day-to-day variable. Regarding the two LAMP positive results in patients´ urine samples with other microscopy-confirmed infectious diseases (i.e. hookworm and T. vaginalis), it is not uncommon to find co-infections of S. haematobium with other organisms such as bacteria, protozoa and helminths, including the hookworms [78]. It is unlikely that this result is due to a cross-reaction with hookworm since we obtained LAMP negative results in other three patients´ urine samples with microscopy-confirmed infection with this geohelminth. One eosinophilic without confirmed diagnosis patient as well as five non-eosinophilic without apparent pathologic disease individuals had S. haematobium-positive results by LAMP. The presence of absence of eosinophils is usually used as a biomarker for helminthic infections, including schistosomiasis [79]; however, it is not predictive of Schistosoma species infection and may generate inconsistent results [80]. Thus, application of our LAMP method may improve the identification of cases with low-intensity infections as well as in cases which did not pass eggs in urine samples, thus revealing infections in people frequently presumed to be uninfected. Finally, although all patients´ urine samples were tested in duplicate with the same result, it would be very interesting to know how reproducible the technique is when testing in field settings as well.
In conclusion, we have demonstrated that simply rapid-heating urinary pellets for good-quality DNA extraction was effective for use in LAMP assays with regard of detecting S. haematobium in clinical urine samples. This procedure has been named the Rapid-Heat LAMPellet method and it would be well-suited to diagnose urogenital schistosomiasis in resource-limited endemic regions because of its rapidity, easy handling, cost-effectiveness and both high detection specificity and sensitivity. The next step for refining the assay by conducting a field evaluation in an endemic setting should be desirable.
|
10.1371/journal.pntd.0001667 | Genetics of Host Response to Leishmania tropica in Mice – Different Control of Skin Pathology, Chemokine Reaction, and Invasion into Spleen and Liver | Leishmaniasis is a disease caused by protozoan parasites of genus Leishmania. The frequent involvement of Leishmania tropica in human leishmaniasis has been recognized only recently. Similarly as L. major, L. tropica causes cutaneous leishmaniasis in humans, but can also visceralize and cause systemic illness. The relationship between the host genotype and disease manifestations is poorly understood because there were no suitable animal models.
We studied susceptibility to L. tropica, using BALB/c-c-STS/A (CcS/Dem) recombinant congenic (RC) strains, which differ greatly in susceptibility to L. major. Mice were infected with L. tropica and skin lesions, cytokine and chemokine levels in serum, and parasite numbers in organs were measured.
Females of BALB/c and several RC strains developed skin lesions. In some strains parasites visceralized and were detected in spleen and liver. Importantly, the strain distribution pattern of symptoms caused by L. tropica was different from that observed after L. major infection. Moreover, sex differently influenced infection with L. tropica and L. major. L. major-infected males exhibited either higher or similar skin pathology as females, whereas L. tropica-infected females were more susceptible than males. The majority of L. tropica-infected strains exhibited increased levels of chemokines CCL2, CCL3 and CCL5. CcS-16 females, which developed the largest lesions, exhibited a unique systemic chemokine reaction, characterized by additional transient early peaks of CCL3 and CCL5, which were not present in CcS-16 males nor in any other strain.
Comparison of L. tropica and L. major infections indicates that the strain patterns of response are species-specific, with different sex effects and largely different host susceptibility genes.
| Several hundred million people are exposed to the risk of leishmaniasis, a disease caused by intracellular protozoan parasites of several Leishmania species and transmitted by phlebotomine sand flies. In humans, L. tropica causes cutaneous form of leishmaniasis with painful and long-persisting lesions in the site of the insect bite, but the parasites can also penetrate to internal organs. The relationship between the host genes and development of the disease was demonstrated for numerous infectious diseases. However, the search for susceptibility genes in the human population could be a difficult task. In such cases, animal models may help to discover the role of different genes in interactions between the parasite and the host. Unfortunately, the literature contains only a few publications about the use of animals for L. tropica studies. Here, we report an animal model suitable for genetic, pathological and drug studies in L. tropica infection. We show how the host genotype influences different disease symptoms: skin lesions, parasite dissemination to the lymph nodes, spleen and liver, and increase of levels of chemokines CCL2, CCL3 and CCL5 in serum.
| Several hundred million people in 88 countries are living in areas where they can contract leishmaniasis, a disease caused by intracellular protozoan parasites of the genus Leishmania and transmitted to vertebrates by phlebotomine sand flies. Leishmania parasites infect professional phagocytes (neutrophils, monocytes and macrophages), as well as dendritic cells and fibroblasts [1]. The main vertebrate host target cell is the macrophage, where parasites multiply, eventually rupture the cell, and spread to uninfected cells [2]. As macrophages migrate to all mammalian tissues, Leishmania parasites have a great potential for damaging bodily functions. In the dermis, they cause the cutaneous form of the disease (which can be localized or diffuse); in the mucosa, they result in mucocutaneous leishmaniasis; and the metastatic spread of infection to the spleen and liver leads to visceral leishmaniasis. One of the major factors determining the type of pathology is the species of Leishmania [3]. However, the transmitting vector, as well as genotype, nutritional status of the host, and environmental and social factors also have a large impact on the outcome of the disease [3], [4]. That is why even patients, infected by the same species of Leishmania, develop different symptoms [3] and may differ in their response to therapy [5]. The basis of this heterogeneity is not well understood [6], but part of this variation is likely genetic. Numerous potentially relevant genes were reported (reviewed in [7]).
The extent of involvement of Leishmania tropica in human leishmaniasis has been recognized only recently. The western limit of L. tropica appears to be the Greek Island of Zakynthos, the disease has been found in Turkey, Syria, Jordan, Israel, Morocco, Tunisia, Saudi Arabia, Yemen, Iran, Iraq, Afghanistan, Turkmenistan, Pakistan, Kenya, Ethiopia and Namibia, and at its eastern limits in India (reviewed in [8]). While L. major is a zoonosis with mainly rodent (Gerbillidae) reservoir hosts, L. tropica can circulate among humans without the involvement of animal reservoirs, but zoonotic transmission also occurs [9]. Similarly to L. major, L. tropica causes cutaneous leishmaniasis in humans. Moreover, L. tropica was also reported to visceralize and cause an initially not understood systemic illness in veterans returning from endemic areas in the Middle East [10], as well as the classical visceral leishmaniasis (kala-azar) in India [11], and the disseminated cutaneous leishmaniasis accompanied with visceral leishmaniasis in Iran [12].
A suitable animal model for study of this parasite would contribute to functional dissection of the clinical course of infection. Golden hamsters (Mesocricetus auratus) have been considered to be the best model host of the L. tropica infection, but this host is not inbred and hence not suitable for many studies. However, several strains of L. tropica from Afghanistan, India [13], and Turkey [14] have been described to cause cutaneous disease in inbred BALB/c mice, thus providing a better defined host. In comparison with widely studied immune response to L. major infection (reviewed in [15]) and its genetic control (reviewed in [7]), little is known about L. tropica in mouse [13], [14], [16]. Here we aimed to study genetics of susceptibility to L. tropica in the mouse. We analyzed response to L. tropica in CcS/Dem (CcS) recombinant congenic (RC) strains [17] derived from the background strain BALB/cHeA (BALB/c) and the donor strain STS. Each CcS strain contains a different unique random set of about 12.5% genes from the donor strain STS and 87.5% genes from the background strain BALB/c. These strains have been already successfully used for analysis of infection with Leishmania major [18]–[22]. The RC system enabled us to analyze organ pathology and systemic disease after infection with L. tropica and their genetic control.
Males and females of strains BALB/c, STS and selected RC strains [17], [23] were tested. When used for these experiments, RC strains were in more than 38 generation of inbreeding and therefore highly homozygous. During the experiment, male and female mice were placed into separate rooms and males were caged individually. The research had complied with all relevant European Union guidelines for work with animals and was approved by the Institutional Animal Care Committee of the Institute of Molecular Genetics AS CR and by Departmental Expert Committee for the Approval of Projects of Experiments on Animals of the Academy of Sciences of the Czech Republic.
Leishmania tropica from Urfa, Turkey (MHOM/1999/TR/SU23) was used for infecting mice. Amastigotes were transformed to promastigotes using SNB-9 [24], 1×107 stationary phase promastigotes from subculture 2 have been inoculated in 50 µl of sterile saline s.c. into the tail base, with promastigote secretory gel (PSG) collected from the midgut of L. tropica-infected Phlebotomus sergenti females (laboratory colony originating from L. tropica focus in Urfa). PSG was collected as described [25]. The amount corresponding to one sand fly female was used per mouse.
Leishmania major LV 561 (MHOM/IL/67/LRC-L137 JERICHO II) was used for mouse infection. Amastigotes were transformed to promastigotes using SNB-9 [24], 1×107 promastigotes from 6 days old subculture 2 were inoculated in 50 µl of sterile saline s.c. into the tail base.
The size of the skin lesions was measured weekly using a Vernier caliper gauge. The mice infected with L. tropica were killed 21 or 43 weeks after inoculation. Mice infected with L. major were killed 8 weeks after infection. Blood, spleen, liver and inguinal lymph nodes were collected for further analysis.
The current semi-quantitative technique is based on the limiting dilution assay [26]. We modified the procedure by using only a single pre-selected cell concentration, and parasite count was measured with a Coulter Counter CBC5 (Coulter Electronics Inc., Hialeah, FL). In comparison with the original limiting dilution technique, this modified culture method is less laborious and allows rapid estimation of parasite number.
Preparation of cells must be carried out under sterile conditions. During preparation, all samples, which were not immediately worked with, were kept on ice. Inguinal lymph nodes and quarters of spleens were disrupted in a glass homogenizer in complete RPMI (containing 5% of heat-inactivated fetal calf serum (Sigma-Aldrich, USA), 25 mM Hepes (Sigma-Aldrich, USA), 0.0005% β-mercaptoethanol (Serva, Germany), 63.7 µg/ml penicillin (Sigma-Aldrich, USA), and 100 µg/ml streptomycin (Sigma-Aldrich, USA). The homogenate was passed through the nylon filter. The homogenizer was washed 3 times with 3 ml of sterile PBS after processing each lymph node. The samples were then centrifuged 8 min at 300 g, 4°C (centrifuge Eppendorf 5810 R, Eppendorf, Germany). The supernatant was removed and the cells were resuspended in 0.5 ml of complete Schneider's medium supplemented with 20% heat-inactivated fetal calf serum (Sigma-Aldrich, USA), 2% sterile fresh human urine, 50 µg/ml gentamicine (Sigma-Aldrich, USA), 63.7 µg/ml penicillin (Sigma-Aldrich, USA), and 100 µg/ml streptomycin (Sigma-Aldrich, USA). To count cells with a Coulter Counter CBC5 (Coulter Electronics Inc., Hialeah, FL), USA), 50 µl of the cell suspension was diluted in 20 ml of PBS. Ekoglobin (Hemax s.r.o., Czech Republic) was added to the diluent prior to counting to lyse red blood cells.
0.5 ml of the cell suspension (1×105 cells per ml for lymph nodes and 2×105 cells per ml for spleens) was cultivated in complete Schneider's medium in 48-well tissue culture plates (Costar, Corning Inc., USA) at 27°C (Biological thermostat BT 120 M, Labsystem, Finland) for 3 days. Each sample was prepared in triplicate. After incubation, 100 µl of a mixed sample from each well, containing Leishmania parasites released from lymph node or spleen cells were diluted in 20 ml of PBS and the parasite number was counted with the Coulter Counter.
Parasite load was measured in frozen liver samples using PCR-ELISA according to the previously published protocol [27]. Briefly, total DNA was isolated using a standard proteinase procedure. For PCR, two primers (digoxigenin-labeled F 5′-ATT TTA CAC CAA CCC CCA GTT-3′ and biotin-labeled R 5′-GTG GGG AGG GGG CGT TCT-3′ (VBC Genomics Biosciences Research, Austria) were used for amplification of the 120-bp conservative region of the kinetoplast minicircle of Leishmania parasite, and 50 ng of extracted DNA was used per each PCR reaction. For a positive control, 20 ng of L. tropica DNA per reaction was amplified as a highest concentration of standard. A 40-cycle PCR reaction was used for detection. Parasite load was determined by analysis of the PCR product with the modified ELISA protocol (Pharmingen, USA). Concentration of Leishmania DNA was determined using the ELISA Reader Tecan and the curve fitter program KIM-E (Schoeller Pharma, Czech Republic) with least squares-based linear regression analysis.
Levels of GM-CSF (granulocyte-macrophage colony-stimulating factor), CCL2 (chemokine ligand 2)/MCP-1 (monocyte chemotactic protein-1), CCL3/MIP-1α (macrophage inflammatory protein-1α), CCL4/MIP-1β (macrophage inflammatory protein-1β), CCL5/RANTES (regulated upon activation, normal T-cell expressed, and secreted) and CCL7/MCP-3 (monocyte chemotactic protein-3) in serum were determined using Mouse chemokine 6-plex kit (Bender MedSystems, Austria). The kit contains two sets of beads of different size internally dyed with different intensities of fluorescent dye. The set of small beads was used for GM-CSF, CCL5/RANTES and CCL4/MIP-1β and the set of large beads for CCL3/MIP-1α, CCL2/MCP-1 and CCL7/MCP-3. The beads are coated with antibodies specifically reacting with each of the analytes (chemokines) to be detected in the multiplex system. A biotin secondary antibody mixture binds to the analytes captured by the first antibody. Streptavidin-phycoerythrin binds to the biotin conjugate and emits a fluorescent signal. The test procedure was performed in the 96 well filter plates (Millipore, USA) according to the protocol of Bender MedSystem. Beads were analyzed on flow cytometer LSR II (BD Biosciences, USA). As standards were used lyophilized GM-CSF and chemokines (CCL2/MCP-1, CCL3/MIP1α, CCL4/MIP1β, CCL5/RANTES, CCL7/MCP-3) supplied in the kit. Concentration was evaluated by Flow Cytomix Pro 2.4 software (eBioscience, Vienna, Austria). The limit of detection of each analyte was determined to be for GM-CSF 12.2 pg/ml, CCL2/MCP-1 42 pg/ml, CCL7/MCP-3 1.4 pg/ml, CCL3/MIP-1α 1.8 pg/ml, CCL4/MIP-1β 14.9 pg/ml, CCL5/RANTES 6.1 pg/ml respectively.
IFNγ, IL-4, IL-12 and IgE levels in serum were determined using the primary and secondary monoclonal antibodies (IFNγ: R4-6A2, XMG1.2; IL-4: 11B11, BVD6-24G2; IL-12p40/p70: C15.6, C17.8; IgE: R35-72, R35118) and standards from BD Biosciences, USA (recombinant mIFNγ, mIL-4, mIL-12 and purified mIgE: C38-2). ELISA was performed as recommended by BD Biosciences. The ELISA measurement of IFNγ, IL-4, IL-12, and IgE levels was performed by the ELISA Reader Tecan and the curve fitter program KIM-E (Schoeller Pharma, Czech Republic) using least squares-based linear regression analysis. The detection limit of ELISA was determined to be 30 pg/ml for IFNγ, 8 ng/ml for IgE, 16 pg/ml for IL-4 and 15 pg/ml for IL-12.
Inguinal lymph nodes and spleens were fixed in 4% formaldehyde and embedded in paraffin. Immunohistochemical staining of parasites was performed in 5 µm lymph node sections. Slides were deparaffinized with xylene (2 times for 5 min) and rehydrated with 96% ethanol (3 times for 3 min), 80% ethanol (3 min), 70% ethanol (3 min) and PBS (phosphate buffer saline, 3 min). Endogenous peroxidase was quenched with 3% H2O2 in methanol for 10 min. Sections were washed in PBS (10 min) and parasites were stained using anti-Leishmania lipophosphoglycan monoclonal mouse IgM (Code Nr. CLP003A, Cedarlane, Canada) diluted 1∶100 in PBS with 1% BSA (bovine serum albumine, Sigma-Aldrich, USA) and applied for 1 h at 37°C, followed by TRITC-conjugated goat anti-mouse IgM (Code Nr. 115-025-020, Jackson Immunoresearch, USA), also diluted 1∶100 in PBS with 1% BSA and applied for 1 h at 37°C. Cell nuclei were stained with DAPI (4′,6-diamidino-2-phenylindole dihydrochloride) 10 ng/µl (Sigma-Aldrich, USA). For histological analysis, 5 µm spleen and lymph node sections were stained by the routine hematoxylin and eosin method (H&E).
The differences between CcS/Dem strains in parasite numbers in lymph nodes were evaluated by the analysis of variance (ANOVA) and Newman-Keuls multiple comparison using the program Statistica for Windows 8.0 (StatSoft, Inc., Tulsa, Oklahoma, U.S.A.). Strain and age were fixed factors and individual experiments were considered as a random parameter. The differences in parameters between strains were evaluated using the Newman-Keuls multiple comparison test at 95% significance. Difference between sexes in parasite numbers in lymph nodes was analysed by Mann Whitney U test. Analysis of sex influence on lesion development after L. major infection was performed using General Linear Models, Univariate ANOVA, Statistica 8.0 with experiment as a random and age as a fixed parameter.
To study susceptibility to L. tropica we infected both females and males of the strains BALB/c, STS and RC strains CcS-3, CcS-5, CcS-11, CcS-12, CcS-16, CcS-18, and CcS-20.
In females, skin pathology started as a nodule at the site of L. tropica infection appearing between weeks 11 and 20, which transformed in susceptible strains into a skin lesion (Figure 1). Females of the strains BALB/c, CcS-11, CcS-16 and CcS-20 were relatively susceptible to the infection and developed skin lesions after week 18; the largest lesions were observed in CcS-16 (Figure 2A). Females of the strain CcS-16 exhibited skin lesions from week 18 until the end of experiment (week 43). In females of the strains BALB/c, CcS-11 and CcS-20, the lesions partly healed and tended to transform back to nodules after week 30. Interestingly, in females of the strain CcS-11, small skin nodules appeared at week 14, but at 32–42 weeks of infection half of the females died without obvious pathological findings. Only one female died at week 13 in the 21-week experiment. Strains CcS-12 and CcS-18 are intermediate in susceptibility to skin pathology. CcS-12 females developed small skin lesions at the late stages of infection (after week 37), whereas CcS-18 females developed nodules or small lesions that healed. Strains STS, CcS-3 and CcS-5 were resistant to skin lesions development. Females of the strain CcS-3 had small skin nodules at the late stages of infection and did not develop skin lesions during the entire course of the experiment. Females of the strains STS and CcS-5 were resistant to L. tropica, and only a few of them developed small nodules at the site of infection.
Males of the strain CcS-16 developed small lesions from week 22, which later healed, whereas BALB/c males exhibited small lesions from week 30 until the end of the experiment (Figure 2B). Males of the strain CcS-12 developed small skin lesions at the late stages of infection (after week 37). Males of CcS-11 developed no or only small skin nodules, but most of the animals died before week 18 of infection. Similarly as CcS-11 females, they were without obvious pathological findings. Males of the strains STS, CcS-3 and CcS-5 had small skin nodules at the late stages of infection and did not develop skin lesions within the entire course of the experiment. Only a few males of the strains CcS-18 and CcS-20 developed small nodules at the site of infection.
Sex differences observed in susceptibility to L. tropica led us to analyze sex influence in susceptibility to L. major. As our previous research with L. major was focused on analysis of females [28], in this study we have infected with L. major both females and males of strains BALB/c, STS and RC strains CcS-3, CcS-5, CcS-11, CcS-12, CcS-16, CcS-18, and CcS-20. Both males and females of all analyzed strains developed larger skin lesions after infection with L. major than when infected with L. tropica (Table 1, Figure 2A, B). The effect of sex was different in experiments with L. major and L. tropica. After the infection with L. tropica, females of CcS/Dem strains in most cases exhibited more extensive skin pathology than males (Figure 2 A, B), whereas after infection with L. major, skin lesions in males and females of strains BALB/c, STS, CcS-11, CcS-12, CcS-16 and CcS-20 did not differ, whereas males of strains CcS-3 (P = 0.001), CcS-5 (P = 0.001) and probably also CcS-18 (P = 0.043) developed larger skin lesions than females (Table 1).
In vitro culture tests showed that all tested mice, including strains that did not exhibit any skin pathology, contained viable parasites in their inguinal lymph nodes both 21 and 43 weeks after infection (Figure 3). Presence of parasites was also documented by staining of Leishmania in tissue smears with the anti-Leishmania lipophosphoglycan monoclonal antibody (Figure 4) and by histological analysis of hematoxylin-eosin stained lymph nodes smears (Figure S1).
None of the strains contained more parasites in the lymph nodes than the background parental strain BALB/c (Figure 3). At week 21 after infection, females of the strains STS and CcS-5 (P = 0.0002), and males of the strain CcS-5 (P = 0.0093) contained significantly lower parasite numbers than the strain BALB/c. At week 43 after infection, females of the strains STS, CcS-5 and CcS-11 (P = 0.00000001), and STS males (P = 0.0004) had significantly lower parasite load than BALB/c. At week 43, females of strain CcS-18 had higher parasite count than males (P = 0.0209), whereas males of the strain CcS-5 had higher parasite load than females (P = 0.0143). In both experiments counted together, females of CcS-18 (P = 0.0318) strain had higher parasite load than males, whereas STS males had higher parasite numbers than females (P = 0.0097) (Figure 3).
We did not observe any splenomegaly and hepatomegaly induced by infection with L. tropica. However, in vitro cultures have shown that 21 weeks after infection 50% (3 out of 6), 66.7% (2 out of 3) and 16.7% (1 out of 6) spleens of female mice of the strains CcS-3, -18 and -20, respectively, contained viable parasites. We were also able to cultivate parasites from 16.7% (1 out of 6) and 33.3% (2 out of 6) of spleens of males of the strains BALB/c and CcS-20, respectively. Parasite presence in spleens was confirmed by histological examination (Figure 5). Parasite numbers in spleens of other strains were either below the level of detection or absent. Mice of the strain CcS-12 were not tested in the 21-week experiment.
Later we measured parasite load in frozen liver tissues using PCR-ELISA [27]. The presence of parasites after 21 weeks of infection was detected in females of the strains BALB/c 33.33% (2 out of 6), CcS-3 66.67% (4 out of 6), CcS-11 42.86% (3 out of 7), CcS-16 71.43% (5 out of 7), CcS-18 66.67% (2 out of 3) and CcS-20 75% (3 out of 4). Livers of males of the same strains also contained parasites: BALB/c 83.33% (5 out of 6), CcS-3 100% (3 out of 3), CcS-11 80% (4 out of 5), CcS-16 66.67% (4 out of 6), CcS-18 33% (2 out of 6) and CcS-20 83.33% (5 out of 6).
Unfortunately, an additional analysis of spleens and lymph nodes with PCR-ELISA cannot be performed as these organs were completely used for the cultivation assay.
We measured levels of IL-4, IL-12, IFNγ, GM-CSF (granulocyte-macrophage colony-stimulating factor), CCL2 (chemokine ligand 2)/MCP-1 (monocyte chemotactic protein-1), CCL3/MIP-1α (macrophage inflammatory protein-1α), CCL4/MIP-1β (macrophage inflammatory protein-1β), CCL5/RANTES (regulated upon activation, normal T-cell expressed, and secreted), CCL7/MCP-3 (monocyte chemotactic protein-3) and IgE in serum of uninfected and infected mice.
No significant difference was found in IL-4, IL-12, IgE, IFNγ and GM-CSF levels in serum of infected mice in comparison with noninfected controls (data not shown). We have observed an increase of serum levels of CCL2, CCL3 and CCL5 in infected strains; the largest increase was observed in strains CcS-11, CcS-16, CcS-18 and CcS-20. Figure 6 shows chemokine kinetics in females. Peak of increase of levels of these chemokines usually followed the start of lesion development. The increase was greater in females than in males (data not shown). Females of the strain CcS-16 exhibited a unique pattern of kinetics of CCL3 and CCL5 levels, which differed from all other strains (Figure 6) and also from CcS-16 males (Figure 7). We observed two peaks of increase of serum levels of CCL3 and CCL5 in females of CcS-16 (Figure 6); one before the development of skin lesions, the other after the decrease of lesion size, and there were almost no changes in CCL2 level. CcS-16 males had slight increase of CCL3, CCL5 (Figure 7), and CCL2 (data not shown); their kinetics of increase was similar to those in females and males of other strains.
We have analyzed manifestations of the disease and the immunological parameters, which constitute the pathology of leishmaniasis in mice (organ pathology, parasite load in organs, systemic immune response) after infection with L. tropica. Compared with infection with L. major, infection progressed more slowly, some manifestations (skin lesions, parasite load in organs) were less pronounced, some (splenomegaly, hepatomegaly) were absent in the tested strains, and the systemic immune response also differed. These observations reveal new areas where the comparative research of infection by the two species can contribute to a deeper genetic and functional understanding of responses to both and perhaps eventually lead to a common conceptual framework for their interpretation. Such more general scheme could not be readily obtained by analysis of any of them alone. A progress in the heavily under-investigated biology of responses to L. tropica infection may also have a translational potential, which may be significant in view of the recent wider recognition of the etiological role of this parasite in human disease. The areas highlighted by our results include the likely species specificity of at least some susceptibility genes, the dramatically different effects of sex on the response, and a different regulation of some systemic responses.
We have observed clearly different patterns in strains' susceptibility to L. tropica and L. major. Out of the nine strains tested, four (BALB/c, STS, CcS-5, CcS-16) exhibited similar relative susceptibility, whereas susceptibility of five strains (CcS-3, CcS-11, CcS-12, CcS-18 and CcS-20) to these two parasites differed. Strains CcS-3 and CcS-20 are resistant and susceptible, respectively, to development of skin lesions after infection with L. tropica, but intermediate to infection with L. major. The strains CcS-12, CcS-18, and CcS-16 are among the most susceptible to L. major infection [28], (Table 1), whereas with L. tropica CcS-12 and CcS-18 are clearly less susceptible than BALB/c and CcS-16 (Figure 2). The mice of strain CcS-11 are intermediate after infection with L. major, but after infection with L. tropica they died with small or no lesions, low parasite load in lymph nodes and with no detectable parasites in spleens. The cause of mortality of CcS-11 was not revealed by standard histo-pathological investigation. These differences indicate presence of species-specific susceptibility genes. Such genes were indicated also by results of Anderson and coworkers [16] who found that strains BALB/c and C57BL/6 had similar numbers of parasites in ear dermis and exhibited similar ear lesion development after infection with L. tropica. In contrast, these two strains differ greatly in susceptibility to L. major [29], (reviewed in [7] and [15]). Poly-specific response genes that control susceptibility to both L. major and L. tropica probably also exist, as the strains STS and CcS-5 are resistant and BALB/c and CcS-16 are susceptible to cutaneous disease induced by both parasite species (Figure 2, Table 1).
These data complement information about species-specific and poly-specific control of infection to L. donovani, L. infantum, L. mexicana and L. major (reviewed in [7]). Poly-specific and species-specific susceptibility genes are not limited to leishmaniasis, but were also indicated in susceptibility to other pathogens such as Plasmodium (reviewed in [30]). The data on susceptibility to L. tropica and L. major reported here is relevant for investigators of other species-specific responses.
Females of the strain CcS-16 that contains a set of approximately 12.5% genes of the donor strain STS and 87.5% genes of the background strain BALB/c exhibited the largest skin pathology (Figure 1, Figure 2A), exceeding skin manifestations in both parental strains BALB/c and STS. The observations of progeny having a phenotype, which is beyond the range of the phenotype of its parents, are not rare in traits controlled by multiple genes. It was observed in different tests of immune responses of RC strains in vitro [31]–[35] and in vivo [28], [36]. Similarly, analysis of gene expression from livers in chromosome substitution strains revealed that only 438 out of 4209 expression QTLs were inside the parental range [37]. These observations are due to multiple gene-gene interactions of QTLs, which in new combinations of these genes in RC or chromosomal substitution strains can lead to the appearance of new phenotypes that exceed their range in parental strains. Alternatively, with traits controlled by multiple loci, parental strains often contain susceptible alleles at some of them and resistant at others, and some progeny may receive predominantly susceptible alleles from both parents. However, we cannot exclude the possibility that the unique phenotype of CcS-16 may be caused by a spontaneous mutation, which had appeared during the inbreeding, similarly as for example a loss-of-function mutation in pyruvate kinase protecting RC strains AcB55 and AcB61 against malaria, which is absent in both parental strains A/J and C57BL/6 [38].
The strains CcS-3 and CcS-5, which are resistant to L. tropica share common STS-derived segments on chromosome 5 (a small part near the position 131.01 Mb); chromosome 6 (segment 32- 44 Mb); chromosome 8 (0–14.72 Mb) and on chromosome 10 (114.44–125.42 Mb)([23] and unpublished data). Interestingly, the segment on chromosome 10 overlaps with Lmr5, which controls resistance to L. major [19].
Three phenomena related to sex influence on Leishmania infection deserve attention: i) a different sex influence on overall susceptibility to skin pathology after infection with relatively closely related pathogen species L. tropica and L. major (Figure 2, Table 1), ii) different sex influence on strains' susceptibility to development of skin lesions (Figure 2) and on parasite numbers in lymph nodes (Figure 3), and iii) sex influence on chemokine levels in serum (Figure 7).
In contrast to L. major infection in CcS/Dem RC strains where males exhibited either higher or similar pathology as females, in L. tropica experiments females were more susceptible to skin pathology than males. However, lymph nodes of females and males of most RC strains do not differ in L. tropica parasite load. The only exceptions are lymph nodes of the strains CcS-5 (and possibly STS), where males exhibit higher numbers of parasites than females and strain CcS-18, where females exhibit higher number of parasites than males (Figure 3). We have also observed a unique transient early peak of serum level of CCL3 and CCL5 in CcS-16 females, but not in CcS-16 males nor in any other strain (see the following section). These data suggest that some genes controlling susceptibility to L. tropica might be sex dependent or alternatively that this sex influence depends on genotype.
Different sex influence on susceptibility to L. mexicana and L. major was observed in DBA/2 mice where females were highly resistant and males susceptible to lesion development after infection with L. mexicana. On the contrary, although both female and male mice developed ulcerating lesions after infection with L. major, lesions healed in males but not in females [39]. Sex influenced liver parasite burdens after intravenous inoculation of L. major in strains BALB/cAnPt, DBA/2N and DBA/2J, males having higher parasite load than females [40].
Genotype influence on sex differences was described in studies of L. major infection [22], [41]. No sex differences in susceptibility were observed in BALB/cJ mice, whereas male B10.129(10 M)ScSn mice were relatively resistant to cutaneous disease, while females developed non-healing ulcerative lesions followed by parasites' metastases and death [41]
Comparison of L. major susceptibility in two strains, BALB/cHeA and CcS-11, has shown that there is no significant sex influence on skin lesion development, splenomegaly and hepatomegaly in these strains. Parasite numbers in lymph nodes in males of both strains were higher than in females; however in spleens only CcS-11 but not BALB/c males had higher numbers than females. These observations show that sex affects pathology of various organs differently and that this influence is modified by the host genotype [22].
These results indicate that data obtained with L. tropica (different sex influence on susceptibility to two relatively closely related pathogen species, sex and genotype interaction, and different sex influence on pathology in different organs) reflect a more general phenomenon. Other clear sex biases in incidence of disease, parasite burden, pathology, mortality, and immunological response against various parasites, have been observed in humans and in rodents (reviewed in [42]).
No significant difference was found in IL-4, IL-12, IFNγ and GM-CSF levels in serum of infected mice in comparison with noninfected controls (data not shown). This differs from increase of serum levels of IL-4, IFNγ and IL-12 observed in CcS/Dem strains after 8 weeks of infection [28]. Loci controlling serum levels of IL-4, IFNγ, IL-12, TNFα and IL-6 after 8 weeks of L. major infection are described in [19], [21], [22]. Similarly as after L. tropica infection, no increase was observed in serum GM-CSF level after infection with L. major (data not shown).
However, the absence of differences in serum levels of IL-4, IL-12, IFNγ and GM-CSF after infection does not exclude the possibility that they are involved in the local response to L. tropica. To test this alternative future experiments are needed, similar to those performed to establish the role of Fli1 (Friend leukemia integration 1) in L. major infection model [43].
Infection with L. tropica led to increased serum levels of chemokines CCL2, CCL3 and CCL5. The highest increase was observed in the strains CcS-11, CcS-16, CcS-18, and CcS-20 (Figure 6, chemokine kinetics in females). The most prominent was the increase of CCL3/MIP-1α. Unexpectedly and in contrast with the other strains tested, the CcS-16 females but not males exhibited a unique pattern of this systemic reaction, characterized by an additional early peak of chemokine levels before the onset of cutaneous disease. It suggests that these early peaks of CCL3 and CCL5 (Figure 7) might be associated with an increased susceptibility of CcS-16 females to L. tropica. However, they could also reflect a stronger, but ineffective response.
CCL3 is produced by a range of cell types, including monocytes/macrophages, lymphocytes, mast cells, basophils, epithelial cells, and fibroblasts. Similarly, expression of CCL5 can be induced in activated T cells, macrophages, fibroblasts, epithelial and endothelial cells, and mesanglial cells [44], [45]. Although chemokines evolved to benefit the host, inappropriate regulation or utilization of these proteins can contribute to many diseases [45].
Both CCL3 and CCL5 bind to receptors CCR1, CCR5 and to chemokine decoy receptor D6. CCL5 also binds to CCR3 [45]. Genes ccl3 and ccl5 are situated on mouse chromosome 11; genes ccr1, ccr5 and ccbp2 (D6) are located on mouse chromosome 9 (Table S1). In CcS-16 ccl5 is on a STS-derived segment, whereas the strain of origin of ccl3 is not yet known; ccr1, ccr5 and ccbp2 are on BALB/c-derived segment ([23] and unpublished data).
The role of CC-chemokines CCL2, CCL3 and CCL5 in leishmaniasis has been tested in a number of studies (reviewed in [46], [47]). CCL2 and CCL3 stimulate anti-leishmania response via the induction of NO-mediated regulatory mechanisms to control the intracellular growth and multiplication of L. donovani [48]. CCL2 together with CCL3 also significantly enhanced parasite killing in L. infantum infected human macrophages [49]. In analysis of susceptibility to L. major in mouse, CCL5 contributed to host resistance, but CCL2 alone did not correlate with resistance [50]. In humans, CCL2 expression correlated with self healing cutaneous lesions, whereas CCL3 was associated with lesions of chronic progressive diffuse cutaneous leishmaniasis caused by L. mexicana [51]. These studies indicate that a coordinated interaction of several chemokines is important for successful immune response against Leishmania, but also that the role of different chemokines in defense against various Leishmania ssp. might differ. The observed strain differences and the double peak of CCL3 and CCL5 in CcS-16 females provide a novel potential starting point for investigation of the impact of inter-individual differences in chemokine response on pathogenesis of leishmaniasis.
In spite of relatively limited pathological symptoms, we found viable parasites in inguinal lymph nodes of all tested strains (Figure 3, 4). In some strains (CcS-3, -18, -20 females, and CcS-20 and BALB/c males) we also observed visceralization of parasites in the spleen (Figure 5); and in females and males of BALB/c, CcS-3, CcS-11, CcS-16, CcS-18 and CcS-20 we detected parasites in the liver. Similarly as in previous L. major experiments, which mapped genes controlling parasite numbers and demonstrated their distinctness from susceptibility genes [22], our present L. tropica studies confirmed that the extent of the pathological changes in different organs did not directly correlate with parasite load. This was especially obvious in CcS-3 mice, which were resistant to development of skin pathology, but nevertheless contained parasites in lymph nodes, spleen and liver. These data indicate that parasite spread to the different organs and other manifestations of the disease are dependent on the genome of the host. Absence of correlation between parasite load in organs or parasitemia and intensity of disease has been observed also after infection with several other pathogens such as Toxoplasma gondii [52], Trypanosoma brucei brucei [53], Trypanosoma congolense [54], and Plasmodium berghei [55].
In conclusion, the present observation that many of the RC strains tested with the two Leishmania species exhibited different susceptibility to L. major and L. tropica demonstrates existence of species-specific controlling host genes with different functions. Therefore, without combining the two components of variation involved in the outcome of Leishmania infection – genetic variation of the host and species of the parasite - the understanding of the mechanisms of disease will remain incomplete. On the basis of the observed strain differences we will perform linkage analysis of the responsible genes. This information may provide the first step to distinguishing the species-specific from the general genes controlling pathogenesis of leishmaniasis.
|
10.1371/journal.pmed.1002261 | Age-related cognitive decline and associations with sex, education and apolipoprotein E genotype across ethnocultural groups and geographic regions: a collaborative cohort study | The prevalence of dementia varies around the world, potentially contributed to by international differences in rates of age-related cognitive decline. Our primary goal was to investigate how rates of age-related decline in cognitive test performance varied among international cohort studies of cognitive aging. We also determined the extent to which sex, educational attainment, and apolipoprotein E ε4 allele (APOE*4) carrier status were associated with decline.
We harmonized longitudinal data for 14 cohorts from 12 countries (Australia, Brazil, France, Greece, Hong Kong, Italy, Japan, Singapore, Spain, South Korea, United Kingdom, United States), for a total of 42,170 individuals aged 54–105 y (42% male), including 3.3% with dementia at baseline. The studies began between 1989 and 2011, with all but three ongoing, and each had 2–16 assessment waves (median = 3) and a follow-up duration of 2–15 y. We analyzed standardized Mini-Mental State Examination (MMSE) and memory, processing speed, language, and executive functioning test scores using linear mixed models, adjusted for sex and education, and meta-analytic techniques. Performance on all cognitive measures declined with age, with the most rapid rate of change pooled across cohorts a moderate -0.26 standard deviations per decade (SD/decade) (95% confidence interval [CI] [-0.35, -0.16], p < 0.001) for processing speed. Rates of decline accelerated slightly with age, with executive functioning showing the largest additional rate of decline with every further decade of age (-0.07 SD/decade, 95% CI [-0.10, -0.03], p = 0.002). There was a considerable degree of heterogeneity in the associations across cohorts, including a slightly faster decline (p = 0.021) on the MMSE for Asians (-0.20 SD/decade, 95% CI [-0.28, -0.12], p < 0.001) than for whites (-0.09 SD/decade, 95% CI [-0.16, -0.02], p = 0.009). Males declined on the MMSE at a slightly slower rate than females (difference = 0.023 SD/decade, 95% CI [0.011, 0.035], p < 0.001), and every additional year of education was associated with a rate of decline slightly slower for the MMSE (0.004 SD/decade less, 95% CI [0.002, 0.006], p = 0.001), but slightly faster for language (-0.007 SD/decade more, 95% CI [-0.011, -0.003], p = 0.001). APOE*4 carriers declined slightly more rapidly than non-carriers on most cognitive measures, with processing speed showing the greatest difference (-0.08 SD/decade, 95% CI [-0.15, -0.01], p = 0.019). The same overall pattern of results was found when analyses were repeated with baseline dementia cases excluded. We used only one test to represent cognitive domains, and though a prototypical one, we nevertheless urge caution in generalizing the results to domains rather than viewing them as test-specific associations. This study lacked cohorts from Africa, India, and mainland China.
Cognitive performance declined with age, and more rapidly with increasing age, across samples from diverse ethnocultural groups and geographical regions. Associations varied across cohorts, suggesting that different rates of cognitive decline might contribute to the global variation in dementia prevalence. However, the many similarities and consistent associations with education and APOE genotype indicate a need to explore how international differences in associations with other risk factors such as genetics, cardiovascular health, and lifestyle are involved. Future studies should attempt to use multiple tests for each cognitive domain and feature populations from ethnocultural groups and geographical regions for which we lacked data.
| The prevalence of dementia varies around the world, but it is not known whether international differences in rates of cognitive decline contribute to this.
The extent to which risk and protective factors such as sex, education, and apolipoprotein E ε4 allele (APOE*4) carrier status have different associations with dementia in different ethnocultural groups and geographic regions is also not known.
We analyzed cognitive performance data from 42,170 mostly elderly individuals, provided by 14 studies of aging representing 12 countries (Australia, Brazil, France, Greece, Hong Kong, Italy, Japan, Singapore, Spain, South Korea, United Kingdom, United States).
The Mini-Mental State Examination (MMSE) and memory, processing speed, language, and executive functioning test scores all declined with age, and rates of decline accelerated with age.
The 14 studies showed different rates of decline, and decline in MMSE scores was faster for Asians than whites, females than males, and APOE*4 carriers than non-carriers. APOE*4 carriers also declined faster than non-carriers on test of memory, processing speed, and language.
International differences in rates of cognitive decline might contribute to the global variation in dementia prevalence.
Further research is needed to determine whether cardiovascular health, lifestyle, and other risk factors for dementia have different associations with cognitive decline in different ethnocultural groups and geographic regions.
| The age-specific prevalence of dementia varies around the world, reportedly being the highest in Latin America and lowest in sub-Saharan Africa [1]. While age-specific prevalence is a good indicator of the population burden of dementia, the relative risk of dementia in different countries is better reflected in the age-specific incidence data. Unfortunately, such data are frequently lacking, especially in low- and middle-income countries. A reasonable proxy for dementia incidence is the rate of cognitive decline, with the expectation that more rapid cognitive decline will lead to higher rates of dementia in one population than another. Differences in the rates of cognitive decline may also contribute to global variation in late-life cognitive deficits less severe than dementia, such as the prevalence of mild cognitive impairment (MCI), which varies even when applying uniform diagnostic criteria [2], and performance on immediate word list recall tasks [3].
Different rates of cognitive decline have been a focus of research seeking to account for lower cognitive performance scores and more prevalent dementia among blacks than whites in the United States [4]. However, the results have been mixed, with some studies suggesting faster decline in whites [5,6] but others suggesting faster decline in blacks [7,8] or no difference between blacks and whites [7,9,10]. There are only a few studies comparing rates of late-life cognitive decline among international cohorts. One of these studies reported a similar rate of decline in immediate word list recall scores across samples from Europe, the US, China, and Mexico but a slower rate of decline in a sample from India [3]. However, not only were these findings based on one type of cognitive task, they were derived primarily from cross-sectional data and, thus, complicated by cohort effects. Another study found no difference in how Mini-Mental State Examination (MMSE) scores declined with age across six samples from four countries [11], but all of these had predominantly white populations with presumably less cultural and socioeconomic disparities than present among a broader range of international regions.
Given the current state of the research, it is not yet known whether different rates of cognitive decline contribute to the global variation in cognitive functioning and rates of dementia. Also unknown is the extent to which risk and protective factors have different associations with cognitive decline in different ethnocultural groups and geographic regions. One important factor is sex, with decline in cognitive performance found to occur more rapidly in women than men in a Chinese cohort, but not in a Danish cohort [12]. The apolipoprotein E ε4 allele (APOE*4) is an established risk factor for Alzheimer disease (AD) [13] and for the transition from MCI to AD [14]. However, the prevalence of APOE*4 among AD patients varies across geographic regions and is significantly lower in Asia than in Europe and North America [15]. Educational attainment has been considered a likely determinant of cognitive decline rates, but research has been inconclusive, with different (even opposite) effects found in different cohorts [7]. Educational attainment is also a factor that varies substantially among late-life cohorts from around the world (e.g., as shown in Sachdev et al. [2]).
The Cohort Studies of Memory in an International Consortium (COSMIC) is a collaborative effort comprising members from around the world who share data from current or past longitudinal population-based studies of cognitive aging [16]. For the current study, data were available for 14 cohorts, representing 12 countries from North and South America, Europe, Asia, and Australia. Our primary goal was to harmonize these data and compare the rates of age-related decline on various types of cognitive tests across the samples. We also aimed to investigate the extent to which sex, educational attainment, and APOE*4 carrier status were associated with decline. Knowing whether rates of cognitive decline differ across different ethnocultural groups and geographic regions will help to clarify why there is global variation in cognitive functioning and prevalence of dementia. With such a diverse overall sample, our study should also help to clarify how sex, education, and APOE*4 carrier status are associated with cognitive decline.
The total sample size of 42,170 individuals for this project was arrived at by combining the samples of all 14 COSMIC studies contributing longitudinal cohort data (listed in Table 1 with their abbreviations). In most cases the full cohort was not used, as we excluded individuals missing data for age, sex, or years of education. The samples we used varied in size from 785 to 12,630 individuals.
Contributing studies had various assessment schedules and follow-up durations. The number of assessment waves (including baseline) was two for six studies, three for five studies, four for two studies, and 16 for two studies (Bambui and EAS), and the maximum follow-up duration was between 2 and 10 y for all studies except Bambui and EAS (each 15 y). For CFAS, the number and type of follow-up assessments differed among the participants (see http://www.cfas.ac.uk/cfas-i/cfasistudy-design/), and we used an abridged schedule comprising baseline and two follow-up waves that captured the majority of participants (waves S0, C2/S2, CX). For each cohort and assessment wave, the number of participants assessed and the average time since baseline are shown in S1–S3 Tables.
This COSMIC project was approved by the University of New South Wales Human Research Ethics Committee (Ref: # HC12446). Each of the 14 contributing studies had previously obtained ethics approval from their respective institutional review boards, and all participants provided informed consent (see S4 Table for details). Further participant consent was not required, as de-identified health data are not considered to be protected health information under current research principles (e.g., as per the Privacy Rule proposed by the National Institutes of Health: http://privacyruleandresearch.nih.gov/research_repositories.asp).
We obtained information on age, sex, educational attainment, and dementia status at baseline from all studies. Data for educational attainment were provided as years by all studies except ESPRIT, for which categories (e.g., higher primary, long technical or professional) had to be assigned discrete year values based on informed decisions. All but four studies also provided APOE*4 carrier status data (see the references in Table 1 for collection details), which we classified as carriers of one or two ε4 alleles versus non-carriers. Cognitive performance was assessed with scores for the MMSE [31] and for neuropsychological tests representing each of four cognitive domains: memory, language, processing speed, and executive functioning (a visuospatial domain was not included because there were not enough common relevant tests across the cohorts). All studies except EAS and SPAH administered the MMSE. However, EAS administered the Blessed Information Memory Concentration test, and a validated formula was used to convert these scores to MMSE scores [32]. Scores for neuropsychological tests from one or more of the cognitive domains were available for all studies except Bambui, SGS, and ZARADEMP. For each of the domains, we used a single test or type of test as common to all studies as possible. For memory, this was a delayed word list recall test, though the particular test varied between studies. The most commonly used memory test was the Rey Auditory Verbal Learning Test [33]. For studies without a specific memory test, we used the MMSE three-word recall sub-score. Tests were allocated to the remaining domains in a manner reflecting common practice [33–35], though we acknowledge that opinions vary and our approach may differ from how the studies previously allocated tests to domains. We allocated semantic fluency tests, typically the number of animals named in 60s [33], to language (as per Ganguli et al. [36]). Trail Making Tests A and B [33] were allocated to processing speed and executive functioning, respectively (as endorsed by others, e.g., Lim et al. [37]). The tests from each study and variations in type or nonstandard administration are detailed in S5 Table.
The following analyses were performed separately for each cohort and each test. First, scores greater than three standard deviations (SDs) from the mean were considered outliers and excluded (the proportions of outliers and other missing scores are shown in S6 Table). Where required, a logarithmic or other transformation was applied to reduce a distribution’s absolute value of skewness from >1 to <1 before identifying outliers. Next, linear mixed models with random effects terms for intercept and age (but not age2) were applied to the original, untransformed data (with outliers removed) to produce estimates of the mean and SD for common values of age (75 y), education (9 y), and sex (50% female). These estimates were used to transform the raw test scores to standardized Z-scores by subtracting the estimated means from the raw scores and then dividing this difference by the estimated SD. These Z-scores were then used in analyses examining longitudinal associations with age, sex, education, and APOE*4 carrier status. Across the contributing studies, standardized scores for the different tests (or different versions or administration protocols for the same test) used to represent a domain were regarded as equivalent in that they provided a comparable metric to compare effect sizes for relationships across studies and between different types of tests.
The type of analysis employed was dependent upon whether the distribution of test scores was approximately symmetric (|skewness| <1) or more highly skewed (|skewness| >1). If approximately symmetric, linear mixed modelling was used, with fixed effects for age, age2, sex, education, and interactions of both sex and education with age, and with random effects for the intercept and age (but not age2). Age was centered at 75 y (approximately the mean age across all cohorts and waves) to reduce multicollinearity between age and age2. For more highly skewed distributions, we used generalized linear mixed effect modelling with the gamma distribution, featuring the same fixed and random effects as above. Note that because age was centered at 75 y, estimates of the fixed effects of age obtained from the above models (that include an age2 term in the equations) represent model estimates of longitudinal associations with age at 75 y. Associations with APOE*4 carrier status were also investigated by repeating the above analyses, with this variable, as well as its interaction with age, included in the model. The dominant genetic model was used, and there was no race-based stratification in comparisons of ε4 carriers and non-carriers.
Estimates of effect sizes pooled across samples were obtained by meta-analysis (using random effects models) and presented in forest plots. Heterogeneity of effect sizes among samples was evaluated with the I2 statistic, which is the percentage of the total variation that is due to variation between studies, rather than sampling error or chance. We report I2 values derived from fixed effects models that give more appropriate indications of variation across studies. We took values of I2 as corresponding to levels of heterogeneity that were low if less than 40%, moderate if 40%–60%, and substantial or considerable if greater than 60% (as per the Cochrane Collaboration [38]).
We repeated our analyses separately for two racial/ethnic groups, one with all individuals from cohorts predominantly comprising white participants (CFAS, ESPRIT, HELIAD, Invece.Ab, PATH, Sydney MAS, ZARADEMP) and one with all individuals from cohorts predominantly comprising Asian participants (HK-MAPS, KLOSCAD, SGS, SLASI). The statistical significance of differences in pooled corresponding cognitive measures between the two groups was obtained using the means and standard errors (SEs) of the pooled measure derived from the meta-analyses. The SE of the difference between two pooled measures (SEdiff) was calculated as the square root of the sum of the squares of the SEs of the two pooled measures. Differences between the means greater than 1.96 times SEdiff were regarded as statistically significant.
Meta-analyses were also used to obtain pooled estimates of fixed effects of sex, education, and APOE*4 carrier status, as well as the interactions of these risk factors with age, and to examine how consistent these associations were across cohorts. Age was analyzed in years, but for ease of interpretation, Bs and 95% confidence intervals (CIs) are presented using age in decades. CIs were obtained as B±Zα2SE(B), where B is the estimate, SE(B) is the standard error of B, and Zα2 is the upper 97.5% percentile point of the standard normal distribution.
In a number of study/test distributions, ceiling or floor effects had produced data spikes in which a relatively large proportion of scores were of either the minimum or maximum possible value. The most prominent reason for this was the termination of timed tests after a predetermined period and the recording of a score equal to that time. These scores were removed in order to achieve convergence or acceptable model fit (the numbers of scores removed are shown in S6 Table). We subsequently examined whether the removal of these scores affected our results by repeating the meta-analyses used to obtain pooled estimates of the fixed effects for each cognitive measure with studies featuring data spikes excluded.
Our primary analyses used data from all available individuals with sufficient information, including those identified by the contributing studies as having dementia at baseline. The inclusion of individuals with dementia at baseline meant that our evaluations of cognitive change were more likely to truly represent those of aging populations. However, as it is possible that individuals with dementia may decline at rates different from those without dementia, the analyses were repeated with cases of baseline dementia removed. The majority of studies diagnosed or classified dementia using DSM-IV criteria, with the exceptions being Bambui (an MMSE score cutoff point 13/14 appropriate for Brazilian populations with low schooling [39]), CFAS (AGECAT organicity level of O3), ESPRIT (standardized interview by a neurologist incorporating cognitive testing, with diagnoses validated by an independent panel of expert neurologists), HK-MAPS (Clinical Dementia Rating ≥1), and SGS (self-reported medical history). We note that these approaches are not harmonized or necessarily optimal for identifying dementia, including the case in which MMSE scores were used while other criteria for dementia, such as impaired functional ability, were not considered.
The Sydney team created the pooled dataset and performed the analyses. IBM SPSS Statistics 23 was used to create the dataset and identify outliers, the function Ime in the R (version 3.3.1) package mlme (https://www.r-project.org/) was used for linear mixed effects modelling, and the Penalised Quasi-Likelihood method implemented in the program glmPQL of the MASS package [40] was used for generalized linear mixed effects modelling. The meta-analyses were conducted and forest plots made using the metafor package in R [41].
Table 2 shows the demographic characteristics of the 14 cohorts contributing to our longitudinal analyses. All of the cohorts except one (PATH) had a greater proportion of females than males. For nine of the cohorts, the age of the youngest participant at baseline was 60 or more years (it was no less than 54 y for the remaining cohorts). The design of both Invece.Ab and PATH led to their cohorts having a much narrower age range than others. While most cohorts contained some participants with no formal schooling, the mean number of years of education varied considerably. Participants from the Brazilian cohorts (Bambui and SPAH) had the fewest years of formal education. Each cohort was essentially homogenous for race/ethnicity, except for EAS (approximately two-thirds white and one-third black), Bambui (white, black, and Brazilian indigenous native), and SPAH (mostly mixed race and white). Across the cohorts, the total number of individuals with APOE*4 data was 15,199, and 22.9% of these were APOE*4 carriers. However, the proportion of APOE*4 carriers varied across the cohorts, being lowest for the two comprising predominantly Chinese participants (HK-MAPS and SLASI). Across all cohorts, the proportion of individuals with dementia at baseline was between almost zero and 5.8% (not counting two studies that excluded individuals with dementia during recruitment: HK-MAPS and Sydney MAS). The overall proportion of individuals with dementia at baseline was 3.3%.
We used individual participant-level data provided by members of the COSMIC collaboration to investigate rates of cognitive decline in 14 longitudinal population-based studies of cognitive aging, representing 12 countries and 5 continents. We also investigated the extent to which sex, education, and APOE*4 carrier status were associated with cognitive performance and decline across these diverse ethnocultural groups and geographic regions. Our findings were minimally affected when repeating our analyses with cases of baseline dementia removed, probably in large part because the overall proportion of these cases was low (3.3%).
The cognitive measures analyzed in this study were the MMSE and tests representing four cognitive domains: memory, language, processing speed, and executive functioning. For all of these measures and across all cohorts, we found performance to not only decline substantially with age, but to decline more rapidly with increasing age. Processing speed exhibited the strongest decline with age, and the MMSE exhibited the weakest. The rate of age-related change in processing speed was almost -0.25 Z-score units per decade, which was not too dissimilar to the rates of decline for memory and executive functioning but twice the rate of decline we found for the MMSE. This is consistent with previous reports of age-related associations being stronger for processing speed, intermediate for memory, and weaker for language [42]. Processing speed being the cognitive measure most strongly associated with age could be seen as supporting the processing speed theory of cognitive aging [43]. However, it was not our aim to investigate this, and further analyses examining the extent to which change in performance on other domains is driven by changes in processing speed would be required to test this idea. The slowest rate of decline being found for the MMSE could stem from this measure being insensitive to changes at high levels of cognition [44].
Rates of cognitive decline and changes in the rates of decline with age exhibited a degree of heterogeneity across the cohorts. The direction of associations with age was highly consistent across cohorts for all cognitive measures (with no instance of significant improvement rather than decline), but the strength differed. These differences remained when cases of baseline dementia were excluded, suggesting that they could lead to international differences in rates of incident dementia, and thereby contribute to the global variation in prevalence of dementia [1]. Future COSMIC projects will aim to harmonize data on incident dementia across the cohorts and match these to rates of cognitive decline and prevalence of dementia.
Despite the differences seen across all cohorts, our initial results indicated only one significant difference in rates of decline or change in rates of decline between groups of cohorts classified as white or Asian: a slightly faster decline with age on the MMSE in the Asian group. There also seemed to be a group difference in the strengths of the pooled age2 fixed effects between processing speed and language, which was greater for whites than for Asians. Analyses with cases of baseline dementia removed showed some additional age2 effects within each group, amplifying the differences between whites and Asians. Further research is needed to determine the reliability of these differences and their implications.
Across our cohorts, females generally performed better than males on verbal memory tests, although the difference was not large. Previously reported differences in late-life memory performance between men and women have varied depending on where the samples were from. Reports of better memory in women have come from developed nations, including the UK [45], US [46], and Denmark [12]. Conversely, women have shown poorer memory performance than men in samples from developing nations or where women have historically not been afforded the same educational opportunities as men, including India [47] and China [48]. Nevertheless, it should be noted that our findings were relatively consistent across the diverse range of cohorts investigated, including some that may be from developing nations. Better verbal memory performance in women than men could arise via an effect of estrogen [49] or sex-specific cognitive reserve [50]. Our initial finding of faster decline in MMSE scores for females than for males is ostensibly consistent with reports that women exhibit both a steeper decline in general cognition with increasing age [51] and a greater prevalence of AD [52]. However, there was only a trend for this association (p = 0.089) after excluding baseline dementia cases from our analyses. Future COSMIC projects will use harmonized incidence of dementia data to more fully examine sex differences in cognitive decline.
Previous research has consistently found higher levels of educational attainment to be associated with better late-life cognitive functioning [7,53,54], but associations between education and rates of cognitive decline are mixed [7]. Our finding that more years of education was associated with better performance on all cognitive measures is consistent with this. Also consistent are declines with age that were slower for the MMSE but faster for language, though the reasons for the mixed directions of these associations are unclear.
Compared to noncarriers, APOE*4 carriers performed worse on memory, processing speed, and the MMSE (as well as on executive functioning in analyses with studies featuring data spikes excluded). APOE*4 carriers also exhibited greater rates of decline than noncarriers for all measures except executive functioning. Such findings are not unexpected given that APOE*4 is a risk factor for AD [13] and for the transition from MCI to AD [14]. With cases of baseline dementia excluded from our analyses, APOE*4 carriers continued to show significantly poorer performance only for memory, which fits with a recent meta-analysis finding memory to be the cognitive measure most strongly affected in APOE*4 carriers with no diagnosed cognitive impairment [55]. The reasons for extremely low heterogeneity among the cohorts for associations with APOE*4 carrier status and APOE*4 carrier status by age interactions on memory and language are unclear. We note that the differences in APOE*4 carrier prevalence across our cohorts are generally consistent with previously reported racial/ethnic differences [56], particularly the relatively low prevalence for the Chinese (HK-MAPS and SLASI) and Italian (Invece.Ab) cohorts.
Strengths of our study include the large number of independent cohorts from diverse geographical, ethnic, and sociocultural groups and the use of the same or very similar cognitive tests by these studies. Even with analyses based on standardized scores, we expect that the use of common tests helped to minimize heterogeneity across the studies within each of the cognitive domains investigated. Nevertheless, with only one test being used to represent cognitive domains, we caution against generalizing our results to domains rather than viewing them as test-specific associations, although it is noteworthy that the tests used were prototypical of their domains, and that for the memory domain a variety of verbal memory tests were used across the cohorts. We also note that the MMSE has been criticized as psychometrically unsound for assessing cognitive change in healthy older adults [57] and prone to practice effects [58]. Indeed, with the same cognitive tests used repeatedly in all assessment waves, it is possible we underestimated age-related change because of practice effects. Being reportedly stronger in younger adults [59], practice effects could partially explain increasing rates of decline with increasing age. Other limitations include the cohorts differing in size, number of assessment waves, and overall follow-up duration. Despite all being population-based, the use of particular strategies for recruitment and regional specificity may mean that the cohorts are not necessarily representative of the countries or entire populations they were from. Our study did not have data on chronic degenerative diseases or cardiovascular and lifestyle-related factors commonly associated with aging. These factors could have independent associations with cognitive decline, and not controlling for them may lead to overestimating the strength of associations between age and cognitive decline. Limitations also come with having to harmonize some data from among a heterogeneous group of studies. For example, the use of different memory tests by the studies entailed differences in the range of possible scores, which, despite harmonization, potentially influenced the variability within studies and, thus, also potentially influenced our findings of differences between studies.
In conclusion, we found that cognitive performance consistently declined with age, and more rapidly with increasing age, across cohorts from a diverse range of ethnocultural groups and geographical regions. Similar patterns of results were found for analyses that either included or excluded individuals with dementia at baseline. The strengths of the observed associations varied across the cohorts, and there were also some small differences between groups of cohorts classified as white or Asian. This suggests that different rates of cognitive decline might contribute, via different rates of incident dementia, to the global variation in dementia prevalence. Given the diversity of cohorts and our large overall sample size (more than 42,000 individuals), the associations with sex, education, and APOE genotype we found should help to clarify the contributions of these factors to cognitive ageing on a global scale. We intend for future research with COSMIC cohorts to explore how risk factors not investigated in the current study, including other genetic, epigenetic, cardiovascular, and lifestyle-related factors, contribute to cognitive decline and neurocognitive disorders, and to determine the extent to which their associations vary internationally. We also aim to feature populations from ethnocultural groups and geographical regions for which the current study lacked data, including Africa, India, and mainland China. This will provide important information for developing efficacious interventions to prevent or minimize cognitive impairment and dementia in the rapidly aging population worldwide.
|
10.1371/journal.pcbi.1006106 | RIDDLE: Race and ethnicity Imputation from Disease history with Deep LEarning | Anonymized electronic medical records are an increasingly popular source of research data. However, these datasets often lack race and ethnicity information. This creates problems for researchers modeling human disease, as race and ethnicity are powerful confounders for many health exposures and treatment outcomes; race and ethnicity are closely linked to population-specific genetic variation. We showed that deep neural networks generate more accurate estimates for missing racial and ethnic information than competing methods (e.g., logistic regression, random forest, support vector machines, and gradient-boosted decision trees). RIDDLE yielded significantly better classification performance across all metrics that were considered: accuracy, cross-entropy loss (error), precision, recall, and area under the curve for receiver operating characteristic plots (all p < 10−9). We made specific efforts to interpret the trained neural network models to identify, quantify, and visualize medical features which are predictive of race and ethnicity. We used these characterizations of informative features to perform a systematic comparison of differential disease patterns by race and ethnicity. The fact that clinical histories are informative for imputing race and ethnicity could reflect (1) a skewed distribution of blue- and white-collar professions across racial and ethnic groups, (2) uneven accessibility and subjective importance of prophylactic health, (3) possible variation in lifestyle, such as dietary habits, and (4) differences in background genetic variation which predispose to diseases.
| Race and ethnicity are typically unspecified in very large electronic medical claims datasets. Computationally estimating a patient’s missing race and ethnicity from their medical records is important on both an academic and practical basis. Academically, discriminative medical events tell us about racial and ethnic health disparities and divergent genetic predispositions. Practically, imputed race and ethnicity information can substantially improve genetic and epidemiological analyses with these large datasets.
| Electronic medical records (EMRs) are an increasingly popular source of biomedical research data [1]. EMRs are digital records of patient medical histories, describing the occurrence of specific diseases and medical events such as the observation of heart disease or dietary counseling. EMRs can also contain demographic information such as gender or age.
However, these datasets are often anonymized and lack race and ethnicity information (e.g., insurance claims datasets). Race and ethnicity information may also be missing for specific individuals within datasets. This is problematic in research settings as race and ethnicity can be powerful confounders for a variety of effects. Race and ethnicity are strong correlates of socioeconomic status, a predictor of access to and quality of education and healthcare. These factors are differentially associated with disease incidence and trajectories. As a result of this correlation, race and ethnicity may be associated with variation in medical histories. As an example, it has been reported that referrals for cardiac catheterization are rarer among African American patients than in White patients [2]. Furthermore, researchers have reported differences in genetic variation which influence disease across racial and ethnic groups [3]. Due to the association between race, ethnicity and medical histories, we hypothesize that clinical features in EMRs can be used to impute missing race and ethnicity information.
In addition, race and ethnicity information can be useful for producing and investigating hypotheses in epidemiology. For example, variation in disease risk across racial and ethnic groups that cannot be fully explained by allele frequency information may provide insights into the possible environmental modifiers of genes [3].
The task of race and ethnicity imputation can be serialized as a supervised learning problem. Typically, the goal of imputation is to estimate a posterior probability distribution over plausible values for a missing variable. This distribution of plausible values can be used to generate a single imputed dataset (e.g., by choosing plausible values with highest probability), or to generate multiple imputed datasets as in multiple imputation [4]. In our setting, the goal was to impute the distribution of mutually-exclusive race and ethnicity classes given a set of clinical features. Features comprised age, gender, and codes from the International Disease Classification, version 9 (ICD9, [5]); ICD9 codes describe medical conditions, medical procedures, family information, and some treatment outcomes.
Bayesian approaches to race and ethnicity imputation using census data have been proposed [6] and have been used for race and ethnicity imputation in EMR datasets [7]. However, these approaches require sensitive geolocation and surname data from patients. Geolocation and surname data can be missing in anonymized EMR datasets (as in the datasets used here), limiting the utility of approaches which use this information.
Traditionally, logistic regression classifiers have been used to impute categorical variables such as race and ethnicity [8]. However, there has been recent interest in the use of deep learning for solving similar supervised learning tasks. Deep learning is particularly exciting as it offers the ability to automatically learn complex representations of high-dimensional data. These representations can be used to solve learning tasks such as regression or classification [9].
Deep learning involves the approximation of some utility function (e.g., classification of an image) as a neural network. A neural network is a directed graph of functions which are referred to as units, neurons or nodes. This network is organized into several layers; each layer corresponds to a different representation of the input data. As the input data is transformed and propagated through this network, the data at each layer corresponds to a new representation of the sample [9]. For our imputation task, the aim was to learn the representation of an individual as a mixture of race and ethnicity classes where each class is assigned a probability. This representation is encoded in the final output layer of the neural network. The output of a neural network functions as a prediction of the distribution of race and ethnicity classes given a set of input features.
We introduce a framework for using deep learning to estimate missing race and ethnicity information in EMR datasets: RIDDLE or Race and ethnicity Imputation from Disease history with Deep LEarning. RIDDLE uses a relatively simple multilayer perceptron (MLP), a type of neural network architecture that is a directed acyclic graph (see Fig 1).
In addition to investigating the novel utility of deep learning for race and ethnicity imputation, we used recent methods in interpreting neural network models [10] to perform a systematic evaluation of racial and ethnic patterns for approximately 15,000 different medical events. We believe that this type of large-scale evaluation of disease patterns and maladies by race and ethnicity has not been done heretofore.
We aimed to assess RIDDLE’s imputation performance in a multiclass classification setting. We used EMR datasets from Chicago and New York City, collectively describing over 1.5 million unique patients. There were approximately 15,000 unique input features consisting of basic demographic information (gender, age) and observations of clinical events (codified as ICD9 codes). The target class was race and ethnicity; possible values were White, Black, Other or Hispanic (see Table 1). Although race and ethnicity can be described as a mixture, our training datasets labeled race and ethnicity as one of four mutually exclusive classes. For the testing set, we treated the target race and ethnicity class as unknown, and compared the predicted class against the true class. The large dimensionality of features, high number of samples, and heterogeneity of the source populations present a unique and challenging classification problem.
In our experiments, RIDDLE yielded an average accuracy of 0.668, and cross-entropy loss of 0.857 on test data, significantly outperforming logistic regression, random forest classifiers, and gradient-boosted decision tree (GBDT) classifiers across all classification metrics (p < 10−9; see Table 2).
Support vector machines (SVMs) with various kernels were also evaluated. However, SVMs could not be feasibly used with the full dataset as individual trials took longer than 36 hours each (36 hours runtime was the allowed maximum on the system used in our analysis). Additional experiments involving a smaller subset of the full dataset (165K samples) were performed; in such experiments, SVMs could be practically utilized and RIDDLE significantly outperformed the baseline methods across all classification metrics (p < 10−2; see Table E in S1 Supplement).
While the multiclass learning problem appeared relatively hard, RIDDLE achieved class-specific receiver operating characteristic’s (ROC) area under the curve (AUC) values above 0.8 (see Fig 2), and a micro-average (all cases considered as binary) AUC of 0.874—significantly higher than that of logistic regression (mean = 0.854, p = 6.67 × 10−11), random forest (mean = 0.844, p = 2.05 × 10−10) and GBDT (mean = 0.846, p = 1.20 × 10−10) classifiers (see Table 2).
RIDDLE exhibited runtime performance comparable to that of other machine learning methods on a standard computing configuration without the use of a graphics processing unit or GPU (see Table 2).
As explained prior, SVMs were also evaluated but precise runtime measurements could not be obtained as the computational cost was too high. However, on a smaller subset (165K samples) of the full dataset where SVMs could be utilized, RIDDLE exhibited significantly faster runtime performance compared to all SVM methods (p < 10−10; see Table E in S1 Supplement).
In order to replicate real-world applications where data other than race and ethnicity (e.g., features for specific samples) may be missing, we conducted additional experiments to simulate random missing data. A random subset of feature observations (ranging from 10% to 30% of all feature observations) was artificially masked completely at random.
Feature observations at the sample level (e.g., a particular ICD9 code for a specific patient) were randomly deleted to simulate random missing data. The number of whole features was kept fixed—only individual observations were removed. Otherwise, the same classification training and evaluation scheme was used as before. Under simulation of random missing data, RIDDLE significantly outperformed logistic regression, random forest classifiers and GBDTs in classification metrics across all simulation experiments (p < 10−9 for 10% and 20% missing data simulation, p < 10−4 for 30% missing data simulation; see Table 3).
A major criticism of deep learning is the opaqueness of trained neural network models for intuitive interpretation. While intricate functional architectures enable neural networks to learn complex tasks, they also create a barrier to understanding how learning decisions (e.g., classifications) are made. In addition to creating a precise race and ethnicity estimation framework, we sought to identify and describe the factors which contribute to these estimations. We computed DeepLIFT (Deep Learning Important FeaTures) scores to quantitatively describe how specific features contribute to the probability estimates of each class. The DeepLIFT algorithm compares the activation of each node to a reference activation; the difference between the reference and observed activation is used to compute the contribution score of a neuron to a class (see the Methods) [10].
If a feature contributes to selecting for a particular class, this feature-class pair is assigned a positive DeepLIFT score; conversely, if a feature contributes to selecting against a particular class, the pair is assigned a negative score. The magnitude of a DeepLIFT score represents the strength of the contribution.
Using DeepLIFT scores, we were able to construct natural orderings of race and ethnicity classes for each feature, sorting classes by positive to negative scores. The following example ordering shows how the example feature (heart disease) is a strong predictor for the African American class, and a weak (or negative) predictor for the Other class.
We computed the class orderings for all ∼15,000 features (see S1 Data). The orderings of the 10 most predictive features (by highest ranges of DeepLIFT scores) are described in Table 4.
We visualized the orderings of the 25 most common features using both frequencies and DeepLIFT scores (see Fig 3; the full table of features is shown in S1 Data). Frequency-based orderings were obtained by sorting the four classes by the number of samples within a class exhibiting a particular feature. Race and ethnicity class orderings obtained from frequency scores were distinctly different than those obtained from DeepLIFT scores. This suggests that RIDDLE’s MLP network is able to learn non-linear and non-frequentist relationships between ICD9 codes and race and ethnicity categories.
According to orderings constructed using DeepLIFT scores, sex is an important feature for predicting race and ethnicity in our models: men who seek medical attention are least likely to be Other followed by African American men. Men who seek medical attention are most likely to be White or Hispanic.
In addition, specific medical diagnoses convey grains of racial and ethnic information: hypertension and human immunodeficiency virus (HIV) are more predictive for African American and Hispanic individuals than White individuals. This finding is also reflected in medical literature, where it has been reported that African American and Hispanic populations are at significantly higher risk for heart disease [11–13] and HIV [14–16] than their White peers.
The fact that these features are important for imputing race and ethnicity could reflect (1) a skewed distribution of blue- and white-collar professions across racial and ethnic groups, (2) uneven accessibility and subjective importance of prophylactic health care across racial and ethnic groups, and (3) possible variation in lifestyle, such as dietary habits. Further work would involve investigating epidemiological hypotheses on how these environmental factors may affect differential clinical patterns across race and ethnicity.
Some of the genetic diseases are famously discriminative for races and ethnicities. For example, sickle cell disease occurs more frequently in African Americans and Hispanic populations than in the rest of the US population [17]. In our model, sickle cell anemia most strongly predicts for the African American and Hispanic classes over the White or Other classes. It has been reported Lyme disease predominately occurs in Whites, and largely unreported for Hispanics or African Americans [18]. This finding is also reflected in our model, where Lyme disease serves as a strong predictor of the White race. Additional strongly White-predictive diseases and medical procedures include atrial fibrillation, hypothyroidism, prostate neoplasm, dressing and sutures, lump in breast, coronary atherosclerosis. These are primarily diseases of older age, suggesting that lifespan varies across race and ethnicity due to socioeconomic and lifestyle reasons, as reported in literature [19, 20].
These orderings provide a high-level description of community structure, and may reflect socioeconomic, cultural, habitual, and genetic variation linked to race and ethnicity across the population of two large cities, New York City and Chicago.
In our experiments, RIDDLE yielded favorable classification performance with class-specific AUC values of above 0.8. Although, RIDDLE uses a fairly simple deep neural network architecture, RIDDLE displayed significantly better classification performance across all tested metrics compared to the popular classification methods logistic regression, random forest and GBDTs. RIDDLE maintained a robust (and significant) classification performance advantage over competitors in experiments simulating missing data. In other experiments, the use of pre-trained bagged embeddings were not helpful to RIDDLE (see Table H in S1 Supplement).
RIDDLE’s superior accuracy and loss results suggest that RIDDLE produces more accurate probability estimates for race and ethnicity classes compared to currently used techniques. Although results could not be obtained for SVMs due to unacceptably high computational costs, RIDDLE significantly outperformed SVMs in runtime efficiency and classification performance on smaller subsets of the full dataset (see Table E in S1 Supplement).
Furthermore, RIDDLE, without the use of a GPU, displayed runtimes comparable to those of traditional classification techniques. With these findings, we argue that deep-learning-driven imputation offers notable utility for race and ethnicity imputation in anonymized EMR datasets. Our current work simulated conditions where ethnicity was missing completely at random. Future work will involve simulating conditions where race and ethnicity are missing at random or missing not at random, and formalizing a multiple imputation framework involving deep-learning estimators.
However, these results also highlight a growing privacy concern. It has been shown that the application of machine learning poses non-trivial privacy risks, as sensitive information can be recovered from non-sensitive features [21]. Our results underscore the need for further anonymization in clinical datasets where race and ethnicity are private information; simple exclusion is not sufficient.
In addition to assessing the predictive and computational performance of our imputation framework, we made efforts to analyze how specific features contribute to race and ethnicity imputations in our neural network model. Each individual feature may represent only a weak trend, but together numerous indicators can synergize to provide a compelling evidence of how a person’s lifestyle, her social circles, and even genetic background can vary by race and ethnicity.
The aforementioned highlights of race- and ethnicity-influenced patterns of health diversity and disparity (see the Results) can be extended to thousands of codes (please see S1 Data for the complete table of features and corresponding annotations). To the best of our knowledge, this systematic comparison across all classes of maladies with respect to race and ethnicity is done for the first time in our study.
Our study used de-identified, independently collected patient data, and was determined by the Internal Review Board (IRB) of the University of Chicago to be exempt from further IRB review, under the Federal Regulations category 45 CFR 46.101(b).
We used an anonymized EMR datasets jointly comprising 1,650,000 individual medical histories from the New York City (Columbia University) and Chicago metropolitan populations (University of Chicago). Medical histories are encoded as variable length lists of ICD9 codes (approximately 15,000 unique codes) coupled with onset ages in years. Each individual belongs to one of four mutually exclusive classes of race (Other, White, Black) or ethnicity (Hispanic). Features included quinary gender (male, female, trans, other, unknown), and reported age in years. Age was quantized into discrete categories by integer values.
Onset age information of each ICD9 code was removed and continuous age information was coerced into discrete integer categories. Features were vectorized in a binary encoding scheme, where each individual is represented by a binary vector of zeros (feature absent) and ones (feature present). Each element in the binary encoded vector corresponds to an input node in the trained neural network (see Fig 1).
k-fold cross-validation (k = 10) and random shuffling were used to produce ten complementary subsets of training and testing data, corresponding to ten classification experiments; this allowed for test coverage of the entire dataset. From the training set, approximately 10% of samples were used as holdout validation data for parameter tuning and performance monitoring. Testing data was held out separately and was only used during the evaluation process.
We used Keras [22] with a TensorFlow backend [23] to train a deep multilayer perceptron (MLP). Neural network architectures and hyperparameters were selected using randomized grid search on 10,000 samples from the validation data. It has been reported that randomized grid search requires far less computational effort than exhaustive grid search with only slightly worse performance [24]. The final neural network hyperparameters are detailed in Table A in S1 Supplement.
The structural architecture of the neural network was fixed across different k-fold partitions prior to training. The neural network was composed of an input layer of 15,122 nodes, two hidden layers of 512 nodes each, and a softmax output layer of four nodes (see Fig 1).
Dropout regularization was applied to each hidden layer with a dropout rate ranging from 0.2–0.8. Dropout regularizes the neural network by randomly dropping neurons and their connections during training; this limits complex co-adaptations between neurons which may not generalize well outside of the training data [25].
For its nodes, our neural network architecture utilizes either Parametric Rectifier Linear Units (PReLUs) [26] or Rectified Linear Units (ReLUs); the choice of which activation to use was determined during hyperparameter tuning.
PReLUs are variants of rectifier functions:
f ( x ) = { x , x > 0 ; α x , x ≤ 0 , where α is a learned parameter.
where x is the input, and f(x) is the output of the PReLU node. ReLUs are simply PReLUs with the coeficient parameter fixed at α = 0.
The MLP was trained iteratively using the Adam optimizer [27]. The learning rate, which controls the magnitude of updates during gradient descent, was tuned via randomized grid search. Training was performed in a batch-wise fashion; data vectorization (via binary encoding) was also done batch-wise in coordination with training. The large number of samples (1.65M) and attention to scalability necessitated “on the fly” vectorization. The number of training epochs (passes over the data) was determined by early stopping and model caching [24], where the model from the epoch with minimal validation loss was selected. In order to encourage exploration beyond local minima, a number of epochs with poorer validation loss was permitted in accordance to a fixed patience parameter.
Categorical cross-entropy was chosen as the loss function; categorical cross-entropy penalizes the assignment of lower probability on the correct class and the assignment of non-zero probability to incorrect classes.
We evaluated several other machine learning approaches: logistic regression, random forest classifier, gradient-boosted decision trees (GBDTs), and support vector machines (SVMs) with various kernels (linear, polynomial, radial basis function). Traditionally, logistic regression has been used for categorical imputation tasks [8]. We used fast Cython (C compiled from Python) or array implementations of these methods (with the exception of GBDTs) offered in the popular ‘scikit-learn’ library. For the GBDT methods, we used a Python wrapper of the popular XGBoost C library [28].
To handle the multiclass ethnicity imputation problem, we used a one-vs-one implementation of SVMs and a one-vs-all implementation of GBDTs. The implementations of logistic regression and random forest are inherently multiclass. Model hyperparameters were tuned in the same fashion (randomized grid search) as for the deep neural networks. The final hyperparameters are detailed in Tables B-D in S1 Supplement.
In order to replicate real-world scenarios where additional information (other than race and ethnicity) may be absent, we conducted simulation experiments where we randomly removed some proportion of feature data (10%, 20%, or 30%). The number of input features was kept the same as feature observations at the sample level were removed; entire features were not removed.
For example, if 500 patient samples exhibited ICD9 code 401.9 (hypertension NOS) in the training data, we removed, with some fixed probability, the observation of ICD9 code 401.9 for each of the 500 individuals. The entire ICD9 code 401.9 feature was not removed—only sample observations of this feature.
We conducted training and testing pipelines with these new “deficient” datasets in the same fashion as before, using ten train/test partitions of the data given by k-fold cross-validation.
The code used to conduct all experiments is available on GitHub (see S1 Code).
We computed standard accuracy, cross-entropy loss, precision, and recall scores for testing data across all ten experiments. We also computed class-specific ROC AUC scores as well as micro-average and macro-average ROC AUC metrics. Class-specific ROC AUC scores refer to the ROC AUC scores computed by binarizing the classification problem to a specific class. The micro-average ROC AUC score was computed by reducing all multiclass classification problems to binary prediction problems (true class vs. other classes). The macro-average ROC AUC score was calculated by averaging all class-specific ROC scores, weighted by the number of cases in each class.
In addition to evaluating classification performance, we also monitored runtime performance across methods. Models were trained on a standard computing configuration on the Midway compute cluster at the University of Chicago: 16 Intel Sandybridge cores at 2.6 GHz, and 32GB RAM.
Significant differences in performance scores were detected using paired t-tests with Bonferroni adjustment.
We computed DeepLIFT scores to interpret how certain features contribute to probability estimates for each class [10]. The DeepLIFT algorithm takes a trained neural network and produces feature-to-class contribution scores for each passed sample.
DeepLIFT scores describe how differences in values for some input neuron (compared to a reference value) result in differences in output neuron values (compared to a reference value). The DeepLIFT interpretation method relies on a central summation-to-delta property:
Δ t = ∑ i = 1 N C Δ x i , Δ t (3)
where Δt is the difference-from-reference value for an output neuron. CΔxi, Δt is the difference-from-reference value for the output neuron which can be attributed to differences-from-reference value for a neuron xi which is necessary to compute the output neuron; this also serves as the DeepLIFT score. Although DeepLIFT does not use gradient information, DeepLIFT scores are computed using a backpropagation-like algorithm which uses a chaining principle analogous to the chain rule. Unlike gradient-based approaches, DeepLIFT scores can be meaningful and non-zero even when the gradient is zero [10].
To compute DeepLIFT scores for the RIDDLE neural networks, we assumed reference values of zeros for all input neurons because our training features were binary and sparse; furthermore, a value of zero for an input feature naturally indicates the absence of a disease state. Alternatively, population statistics for disease incidences could have been used as reference values. Reference values for the hidden layers were obtained by performing a forward pass using input values of zero (the input reference values).
We computed DeepLIFT scores using separate test samples and models from each of our k-fold cross validation experiments to achieve full coverage of the dataset. Scores were summed across experiments for aggregation purposes. To describe high-level relationships between features and classes, we summed scores across all samples to produce an aggregate score. The aggregate DeepLIFT scores for the ten most predictive features are summarized in Table 4.
As described prior, we computed orderings of race and ethnicity classes with each feature’s DeepLIFT scores. These orderings describe how certain features (e.g., medical conditions) can predict for or against a particular race and ethnicity class. We visualize the orderings defined by DeepLIFT scores for the twenty-five most common features in Fig 3, and compare them to the orderings produced from sorting classes by the total number of feature observations within the class. We visualized the orderings of the 25 most frequently observed features in the dataset in Fig 3. For the visualizations, frequency counts were mean-centered to facilitate comparison to DeepLIFT scores.
|
10.1371/journal.pcbi.1003356 | Beyond GLMs: A Generative Mixture Modeling Approach to Neural System Identification | Generalized linear models (GLMs) represent a popular choice for the probabilistic characterization of neural spike responses. While GLMs are attractive for their computational tractability, they also impose strong assumptions and thus only allow for a limited range of stimulus-response relationships to be discovered. Alternative approaches exist that make only very weak assumptions but scale poorly to high-dimensional stimulus spaces. Here we seek an approach which can gracefully interpolate between the two extremes. We extend two frequently used special cases of the GLM—a linear and a quadratic model—by assuming that the spike-triggered and non-spike-triggered distributions can be adequately represented using Gaussian mixtures. Because we derive the model from a generative perspective, its components are easy to interpret as they correspond to, for example, the spike-triggered distribution and the interspike interval distribution. The model is able to capture complex dependencies on high-dimensional stimuli with far fewer parameters than other approaches such as histogram-based methods. The added flexibility comes at the cost of a non-concave log-likelihood. We show that in practice this does not have to be an issue and the mixture-based model is able to outperform generalized linear and quadratic models.
| An essential goal of sensory systems neuroscience is to characterize the functional relationship between neural responses and external stimuli. Of particular interest are the nonlinear response properties of single cells. Inherently linear approaches such as generalized linear modeling can nevertheless be used to fit nonlinear behavior by choosing an appropriate feature space for the stimulus. This requires, however, that one has already obtained a good understanding of a cells nonlinear properties, whereas more flexible approaches are necessary for the characterization of unexpected nonlinear behavior. In this work, we present a generalization of some frequently used generalized linear models which enables us to automatically extract complex stimulus-response relationships from recorded data. We show that our model can lead to substantial quantitative and qualitative improvements over generalized linear and quadratic models, which we illustrate on the example of primary afferents of the rat whisker system.
| To account for the stochasticity inherent to neural responses, single cells as well as populations of cells are often characterized in terms of a probabilistic model. A popular choice for this task are generalized linear models (GLMs) and related approaches [1]–[6]. These models can often be chosen such that the corresponding maximum likelihood problem is a convex optimization problem where a global optimum can be found. This guarantee comes at a price, as GLMs tightly constrain the computations which can be performed on the input. More complex computations can nevertheless be implemented by choosing a nonlinear feature representation of the input which is then fed into the linear model. In practice, however, it is typically very challenging to select the appropriate feature space because it presupposes a deeper understanding of the cell's nonlinear behavior or unfeasibly large amounts of data.
Several approaches have been suggested to overcome the limitations of the generalized linear model. A natural extension is given by generalized quadratic models [7]–[9]. While a quadratic model represents a true generalization of a linear model, it can also be viewed as a linear model with a quadratric extension of the feature space (and, depending on the parametrization, some additional constraints on the parameters). Consequently, it shares similar limitations. A linear combination of quadratic features might still fail to represent the kind of stimulus properties a neuron responds to, but going to higher-dimensional general-purpose feature spaces quickly leads to overfitting. The number of parameters which need to be estimated grows linearly with the stimulus dimensionality in a linear model, quadratically in a quadratic model, and correspondingly faster if one uses a feature space of higher order.
An alternative approach is offered by nonparametric methods such as maximally informative dimensions (MID) [10]. Here, one first seeks a projection of the stimulus onto a lower-dimensional subspace such that as much information as possible is retained about the presence or absence of a spike. Afterwards, histograms are used to map out the nonlinear dependence of the neuron on the projected stimulus. This approach has the advantage that it can, at least in principle, capture arbitrary dependencies on the stimulus. However, the number of parameters that need to be estimated grows exponentially with the dimensionality of the stimulus subspace. This limits the approach to cells which are selective for only a few stimulus dimensions, although nonlinear extensions of this approach exist [11].
Here, we explore a different tradeoff. We derive a much more flexible neuron model for single cells which can, at least in principle, approximate arbitrary dependencies on the stimulus. The model can be viewed as generalizing generalized linear and quadratic models, but in contrast to quadratic models cannot easily be reduced to a GLM by choosing a different representation of the stimulus. Nonlinear stimulus features are directly learned from the data by maximizing the model's likelihood and do not need to be hand picked. The number of parameters of the model still grows only quadratically with the dimensionality of the stimulus, and the complexity of the model can be tuned to take into account the cell's complexity and the amount of data available. We demonstrate that optimizing this model is feasible in practice and can lead to a better fit than either generalized linear or quadratic models.
In the following—after briefly reviewing generalized linear and quadratic models—we introduce a new model for single cell responses and discuss its properties.
All experimental and surgical procedures were carried out in accordance with the policy on the use of animals in neuroscience research of the Society for Neuroscience and the German law.
In a GLM it is assumed that the output conditioned on some input is distributed according to an exponential family and that the expected output is given bywhere is an invertible nonlinearity. Parameters of the model are the weights and potentially additional parameters of the exponential family. In the following, we will assume that is a representation of the stimulus and indicates the presence or absence of a spike.
A special case of the GLM applicable to our problem is, for instance, the linear-nonlinear-Bernoulli (LNB) model, where the exponential family is given by the Bernoulli distribution. As nonlinearity we might choose the sigmoidal logistic function,(1)In the following, we will derive this linear model from a generative modeling point of view. This will help to motivate and see the connections to the extension presented later.
Let us consider the distribution over the stimulus conditioned on . If equals 1, this distribution corresponds to the spike-triggered distribution. If equals 0, we will call it the non-spike-triggered distribution. At least for the moment, let us assume that both distributions are Gaussian, that is,with means and covariances . Bayes' rule allows us to turn these assumptions into a probabilistic model of the neuron's behavior,Using a few simple calculations, the probability of observing a spike, or firing rate, can be seen to be(2)where(3)Using our assumption of Gaussianity, this reduces to(4)where we have performed the reparametrizationand the bias term is given byIf the spike-triggered and non-spike-triggered covariances are assumed identical, the quadratic term vanishes and we obtain the linear-nonlinear-Bernoulli model from above. Without this assumption, we are left with a quadratic model [7]–[9].
The unconstrained quadratic model is equivalent to a GLM with a quadratic extension of the feature space, since(5)is linear in the parameters . In practice, is often replaced by a low-rank approximation [7]–[9], [12], where controls the rank. The quadratic form in this case is given by(6)When choosing this parametrization, the optimization is no longer a convex problem [9] and the model no longer a GLM. In the following, we will use “quadratic model” only to refer to the unconstrained version—a GLM with a quadratic feature space—and “linear model” to refer to the GLM without quadratic features.
The generative point of view immediately suggests generalizations by loosening the assumptions of Gaussian distributed spike-triggered and non-spike-triggered stimuli. In the following, we consider mixtures of Gaussians as possible candidates,Mixture models provide a good compromise between the assumptions of the tightly constrained generalized linear models and nonparametric approaches such as histograms. By plugging the mixture distributions into Equation 3, we obtain a new neuron model whose complexity can be controlled by adjusting the number of mixture components. We dub this model the spike-triggered mixture model (STM).
In the same manner that we have derived a model for the neuron's dependency on the stimulus, we can incorporate dependence on other features as well. Let be the time past since the neuron fired its last spike. Using Bayes' rule, we obtainwhere here we have made the additional assumption that and are independent given . This assumption is also known as the naive Bayes assumption and is often employed in classification. It has empirically been observed that naive Bayes classifiers often perform well even when the assumption of independence is not met [13], [14].
Taken together, the input to the sigmoid nonlinearity (Equation 2) is given by(7)where represents the prior probability of observing a spike and we have used histograms and to represent the interval distributions, (Figure 1). Note that if we do not constrain the parameters, there are several redundancies in this parametrization. For example, we can multiply both and by a common factor without changing the model's predictions. If we reparametrize the model to get rid of redundancies and in addition assume that one mixture component is enough to represent the non-spike-triggered distribution, the input to the sigmoid takes the much simpler form(8)The assumption of Gaussian distributed non-spike-triggered stimuli is sensible, for instance, if an a priori Gaussian distributed stimulus is used to drive the neuron and the width of each bin of the spike train is small such that the posterior probability of observing a spike is generally also small, since in this caseThe spike history dependent term on the right-hand side of Equation 8 can also be written in terms of a linear filter,where represents the spike history, and is the unit vector with zeros everywhere except at the position of the most recent spike. That is, the only difference to a linear model with history dependent term is that here all but one spike are suppressed by . In our experiments, we found that the two forms of spike history dependency worked equally well for most cells.
It is instructive to compare Equation 8 with Equation 4. While the quadratic model can be cast into the form of a linear model with a quadratic feature space, this is in general not possible for the STM. The function is also known as soft maximum, since it can be viewed as a smooth approximation to the maximum of the . Our model is thus effectively taking the maximum of the responses of a number of quadratic models. Also note that the number of parameters is only a constant times the number of parameters of the quadratic model, which means it still grows only quadratically in the number of stimulus dimensions. But the number of parameters can be reduced further, as discussed in the next section.
Assuming a single non-spike-triggered mixture component as in Equation 8 and ignoring the spike history for the moment, the number of parameters of the STM grows as , where is the stimulus dimensionality and is the number of mixture components. This growth might still be impractical where is large or the amount of available data is small, as is often the case with neural data.
To reduce the number of parameters, we can employ the same trick as for the quadratic model and replace the matrices by low-rank approximations (Equation 6). If we additionally share features between the different components, we obtain(9)The number of parameters now grows as , where is the number of quadratic features contributing parameters, is the number of coefficients , and is the number of parameters added by the linear features . That is, for fixed and , the number of parameters is linear in the number of stimulus dimensions. We will refer to this variant of the model as the factored STM.
We tested our model on spike trains obtained from 18 whisker-sensitive trigeminal ganglion cells of adult Sprague-Dawley rats. Recordings were made with a single electrode (sampling frequency: 20 kHz, bandpass filter: 300–5000 Hz). Manual stimulation was used to identify which whisker the neuron innervated as well as the approximate preferred direction of the whisker, after which the whisker was placed inside a plastic tube driven by a metal stimulator arm. The stimulator arm was programmed to deliver low-pass filtered (100 Hz) Gaussian white noise stimulation along the neuron's preferred movement direction. Stimulation was divided into 50 unfrozen trials in which the stimulation sequence varied between trials, and 50 frozen trials in which a Gaussian white noise sequence was generated for the first trial only and then repeated for each subsequent trial. Spikes were extracted offline on the basis of waveform shape and all cells were categorized as either slowly adapting (SA) or rapidly adapting (RA). Example spike trains of two cells for frozen stimuli are shown in Figure 2.
We extracted 10 ms windows from the stimulus and reduced their dimensionality by keeping the first 10 principal components ( explained variance). We also extracted 25 ms of the spike history preceding each bin of the spike train. The dimensionality of the spike history was reduced to 100 by binning spikes into 100 equally sized bins of width (no bin contained more than 1 spike). We then removed all but the most recent spike from the spike history and used this as input to all models. A linear projection of this vector is equivalent to the form of spike history dependency in Equation 8.
Filters of generalized linear models were first trained assuming a sigmoid nonlinearity. Together with a Bernoulli output distribution, this guarantees a concave log-likelihood such that an optimal solution can be found. Afterwards, we replaced the sigmoid nonlinearity with a more flexible nonlinearity consisting of a sum of Gaussian blobs,where the hyperbolic tangent ensures that the predicted probability of a spike does not exceed 1. We jointly optimized the parameters of this nonlinearity and the linear filter by alternately maximizing the average log-likelihood of the linear-nonlinear model using limited-memory BFGS [15], a standard quasi-Newton method (see Text S1 of the supporting information for gradients of the parameters). In a final step, we used a nonparametric histogram estimate (150 bins) to map out the nonlinearity. Through this multi-step procedure we tried to maximize the chances of finding a linear-nonlinear description of a neuron's behavior where one exists. Note that strictly speaking, this model is no longer a generalized linear model (since the nonlinearities used are not invertible and the nonlinearities' parameters are optimized). Quadratic models were optimized using the same procedure after extending the input by quadratic features.
The parameters of the STM (Equation 7) were initialized by estimating the spike-triggered, non-spike-triggered, and interspike interval distributions. Mixtures of Gaussians were fitted using standard expectation maximization [14], [16] and interval distributions were estimated using histograms. While naive Bayes classifiers often already work well, it can be beneficial to directly optimize the conditional log-likelihood [17]. After initializing the parameters, we thus discriminatively finetuned the parameters using BFGS [18]. We found that this indeed helped to substantially improve the performance where the model depended on both the stimulus and the spike history.
We used between three and five components for the spike-triggered distribution and one and two components for the non-spike-triggered distribution, which was found to work well in preliminary runs on a different but related dataset with similar stimuli. Using two non-spike-triggered components increased the stability of the optimization for some cells. Finally, factored STMs were trained discriminatively using limited-memory BFGS with randomly initialized parameters.
All models were trained on the 50 unfrozen trials and performance was evaluated based on the 50 frozen trials.
We qualitatively and quantitatively compare the performance of the generalized linear, quadratic and spike-triggered mixture model (STM) for different cells and find in both cases that the STM can lead to substantial improvements.
Figure 2 shows spike trains generated by the different models when fitted to one SA cell and one RA cell. The trial-to-trial variability of the responses of most cells in the dataset is quite low. This behavior is well captured by the STM, while the responses of the generalized linear and quadratic models generally seem to be noisier. This difference is more pronounced for SA cells than for RA cells, where all models appear to give a reasonably good fit. Corresponding peristimulus time histograms (PSTHs) can be seen in Figure 3 (details on how the PSTHs were computed are given in the next section).
Similar conclusions can be drawn by looking at spike-triggered distributions (Figure 4). Ensembles of spike-triggered positions and velocities of the time-varying stimulus suggest a complex dependency of the responses on the stimulus for at least some cells. Note, however, that even a linear neuron can produce non-Gaussian spike-triggered distributions when the stimulus is correlated over time and the cell's firing depends on its history of generated spikes. Also note that while here we show 2-dimensional spike-triggered distributions, the input to the models was a 10-dimensional stimulus (and a 100-dimensional spike history), as described above.
To get a better intuition for the degree of nonlinearity of a cell, we can compare the cell's spike-triggered distribution with the spike-triggered distribution of the best matching linear model. In the given examples, the linear model is unable to reproduce the spike-triggered distributions of the cells displayed in Figure 4. For the SA cell, even the quadratic model fails to reproduce many of the features of the neuron's spike-triggered distribution, while the STM's behavior much more closely resembles that of the real cell.
To quantify the performance of the different models, we estimate the cross-entropy or negative log-likelihood,(10)where the expectation is taken over stimuli and spikes generated by the real neuron. We estimate this quantity using 50 frozen trials not used during training of the model. The cross-entropy is a natural measure for comparing different models, as it is the measure which is optimized during maximum likelihood estimation of the parameters, and many other system-identification approaches such as spike-triggered averaging can often be viewed as performing maximum likelihood or penalized maximum likelihood learning [19].
The cross-entropy can be used to lower-bound the mutual information between stimuli and spikes,The better a model distribution approximates a cell's behavior, the smaller the difference will be between the lower bound and the true information transmitted by the cell. Note that this mutual information only quantifies the information carried by one bin of the spike train while we are generally more interested in the information carried by an entire spike train, .
The spike train's mutual information with the stimulus can be decomposed as followswhere denotes the history of spikes preceding time . To correctly quantify the mutual information between the spike train and the stimulus, it is thus imporant to take spike history effects into account. If we also use the fact that a neuron's firing will only be affected by the stimulus preceding a spike, , we getfor the mutual information of the spike train per time bin. Dividing by the bin width yields an information rate (measured in bits per second or similar). Estimating this quantity requires two models: one for approximating the distribution and one for approximating . A model for the former can take the form of Equation 8 but with the stimulus-dependent terms dropped.
Results averaged over all recorded SA cells () and all RA cells () are given in Figure 5. The average improvement of the STM over the quadratic model is 45.40 bit/s for SA cells and 15.48 bit/s for RA cells (for models taking into account spike history). The improvement for the cell with the largest difference to the quadratic model is 95.15 bit/s for SA cells and 43.05 bit/s for RA cells (the cells displayed in Figures 2 to 4). The firing rates of these two neurons were 117 Hz and 52.6 Hz, respectively, so that both numbers roughly correspond to 0.8 bit/spike improvement. These improvements correspond to the amount of information carried by the cells that would have been missed if a quadratic model was used to estimate mutual information instead of an STM. The average differences between the quadratic and the linear model, and the STM and the quadratic model (with and without including spike history) were all significant (one-tailed Wilcoxon signed-rank test, ; Figure 5C and D).
In addition to comparing different models, we can also compute and compare our model's performance to the cross-entropy of a PSTH, which has also been called oracle model [20]. We computed PSTHs by convolving the average response to the frozen stimulus with a Gaussian kernel. We took all but one trial to compute the PSTH and the remaining trial for prediction. That is, the probability of a spike at time in trial was predicted to be(11)where is the number of trials and is a normalized Gaussian kernel of width ,For spike counts larger than 1, the same approach could be taken by using the right-hand side of Equation 11 as the rate parameter of a Poisson distribution. We found it was necessary to add a small offset to the PSTH to achieve good results. Both the offset and the kernel width were automatically chosen from a prespecified set of parameters to minimize the cross-entropy averaged over all trials. That is, for each individual cell, we chose the kernel width with the best prediction performance. The optimal kernel widths were found to be around 0.12 ms and 0.09 ms (full width at half maximum, FWHM) for the SA and the RA cell displayed in Figure 3, respectively.
While the performance of the PSTH does not give us a guaranteed lower bound on the achievable cross-entropy, it gives us a very optimistic estimate of the performance that can be achieved by a model which does not take spike history into account. We found that the PSTH yielded a significantly lower cross-entropy than an STM without history dependency (), but not significantly lower than an STM which takes spike history into account ( and , respectively; Figure 5C and D).
PSTHs for model cells were estimated from 1000 spike trains (sampling spike trains was necessary since the models depend on the spike history) using the same kernel as for the real cell. The variance explained () by the generalized linear model, quadratic model and STM was 0.15, 0.26, and 0.47 for the SA cell, and 0.19, 0.41, and 0.5 for the RA cell (Figure 3), respectively. Note that the explained variance depends heavily on the chosen kernel width and wider kernels would yield larger coefficients.
The high firing rate of the cells and the resulting abundance of data allowed us to neglect regularization and overfitting issues. The training set contained on average about 25,000 spikes for SA cells and 6,700 spikes for RA cells. However, typically much less data is available.
To counter overfitting, different approaches to regularization can be taken. We already suggested reducing the number of parameters of the STM via factorization and parameter sharing (Equation 9). To get an idea of how the factored STM's performance behaves as a function of the available data, we artificially reduced the amount of data by randomly picking a subset of the 50 training trials. Of that subset, we used 50% for validation and 50% for optimization. During optimization, the performance on the validation set was tested every 5 iterations. If it decreased 50 times in a row, training was stopped and the parameters with the lowest validation error until then were kept. Other than early stopping, no other form of regularization was used. The test set was the same as the one used in Figure 5.
Figure 6 shows the performance of the factored STM for different amounts of spikes in the training set. The factored STM used 6 components and 5 quadratic features (246 parameters in total) for the SA cell, and 3 components and 5 quadratic features (198 parameters) for the RA cell. For comparison, we also plot the performance of a generalized linear model (111 parameters) trained with early stopping on a subset of the training data, as well as the performance of non-factored STMs (532 parameters and 400 parameters, respectively) and quadratic (156 parameters) models trained on the entire training set.
For the SA cell, the performance of the factored STM started to decrease more rapidly as soon as less than 5,000 spikes were present in the training set. However, even with 2,500 spikes the average performance was still much better than the performance of a quadratic model trained on the entire dataset. For the RA cell, the performance started to deteriorate at about 2,000 spikes. Note that the performance of the linear model worsened at a similar rate. Reducing the number of parameters further by using half the spike history or six instead of ten principal componets to represent the stimulus did not help to improve performance in the regime of few data points. The performance might however be improved by choosing suitable priors for the parameters, which we did not explore here.
Training with half the dataset of the RA cell (about data points) on average took 9.4 minutes for the factored STM and 2.7 minutes for the linear model with parallelized implementations written in C++ when run on a machine with 12 Intel Xeon E5-2630 cores (2.3 GHz).
We have shown that a spike-triggered mixture model can lead to better performance than either linear or quadratic models, which we illustrated on the example of rat primary afferents. A possible explanation for the improved performance might be that our model can better cope with a cell's adaptation to the stimulus. Because the firing rate of our model is effectively a maximum over a number of quadratic models, the model is able to respond differently in different regions of the stimulus space. Our model may yield even bigger improvements when applied to cells higher up the hierarchy—such as cortical cells—where highly nonlinear dependencies on the stimulus are to be expected [21]. In particular, an interesting empirical question is whether STMs will be able to improve upon quadratic models in modeling complex cells [22]. As a generalization, the STM can capture the same kind of invariances that the quadratic model can capture, but in addition allows us to spend parameters in different ways by adding components instead of quadratic feature dimensions.
Here, we chose to give up on the constraint of convexity to be able to build a more flexible neuron model. In practice, non-convex or even multimodal likelihoods do not have to be an issue. Many local optima of the STM likelihood are created simply by permutations of the parameters of the different mixture components and are therefore unproblematic. We found that initializing mixture models with EM and fine-tuning with an off-the-shelf optimizer worked well for our data and the performance of the resulting model was stable across different intializations. The parameters of the factored variant of the STM (Equation 9) were randomly initialized and gave comparable results (Figure 6).
Alternatively, we could have used support vector machines, kernel logistic regression (KLR) [23] or other kernel based approaches [24] for gaining flexibility while retaining convexity. In KLR, the input to the sigmoid (Equation 2) determining the firing rate takes the form(12)where indexes training points and is a kernel measuring the similarity between stimuli or, more generally, inputs to the neuron. If a Gaussian RBF kernel is used, KLR becomes similar to an STM with all covariance matrices constrained to a multiple of the identity matrix and one mixture component placed on top of each data point (cf. Equation 8).
KLR is equivalent to a linear-nonlinear-Bernoulli model with a cleverly chosen feature space whose dimensionality grows with the number of data points. Hence, one advantage of KLR is that its objective function is convex. Advantages of a parametric model like the one presented in this paper are more readily interpretable parameters and lower computational costs when the number of training points is large. Ultimately, whether kernel based methods or a generative approach should be preferred presumably depends on whether one has a better intuition of what represents a good kernel for the input space, or a better intuition of what represents a good characterization of the spike-triggered distribution.
The idea of using spike-triggered distributions to construct and motivate neuron models is not new. However, most work in this direction has focused on spike-triggered averages and covariances [25]–[29]. Here we used mixtures of Gaussians and histograms to derive a new neuron model, but other distributions might work better in a different context and might be worth exploring.
Yet another related approach is to use feed-forward neural networks [30]–[32]. While standard feed-forward neural networks are in principle also able to represent arbitrarily complex stimulus-response relationships [33], one can hope to get away with fewer parameters, less data, or simpler optimization schemes when using a model tailored to the task at hand. In contrast to general nonlinear regression strategies, a generative approach can lead to much more problem-specific architectures and nonlinearities (Equations 8 and 9). Similar cascades of linear-nonlinear units have been proposed but motivated by physiological rather than statistical considerations [20], [34]–[36].
STMs can easily be extended to model populations of neurons similar to how GLMs are extended to populations by introducing coupling filters [5], [37]. Analogous to how we incorporated dependency on the spike history of a single neuron, a form for the dependency between neurons can also be motivated via a log-likelihood ratio for the distribution of cross-interspike intervals.
Code for fitting STMs is provided at http://bethgelab.org/code/theis2013a/.
|
10.1371/journal.pntd.0002817 | Circulating CD14brightCD16+ ‘Intermediate’ Monocytes Exhibit Enhanced Parasite Pattern Recognition in Human Helminth Infection | Circulating monocyte sub-sets have recently emerged as mediators of divergent immune functions during infectious disease but their role in helminth infection has not been investigated. In this study we evaluated whether ‘classical’ (CD14brightCD16−), ‘intermediate’ (CD14brightCD16+), and ‘non-classical’ (CD14dimCD16+) monocyte sub-sets from peripheral blood mononuclear cells varied in both abundance and ability to bind antigenic material amongst individuals living in a region of Northern Senegal which is co-endemic for Schistosoma mansoni and S. haematobium. Monocyte recognition of excretory/secretory (E/S) products released by skin-invasive cercariae, or eggs, of S. mansoni was assessed by flow cytometry and compared between S. mansoni mono-infected, S. mansoni and S. haematobium co-infected, and uninfected participants. Each of the three monocyte sub-sets in the different infection groups bound schistosome E/S material. However, ‘intermediate’ CD14brightCD16+ monocytes had a significantly enhanced ability to bind cercarial and egg E/S. Moreover, this elevation of ligand binding was particularly evident in co-infected participants. This is the first demonstration of modulated parasite pattern recognition in CD14brightCD16+ intermediate monocytes during helminth infection, which may have functional consequences for the ability of infected individuals to respond immunologically to infection.
| The parasite Schistosoma infects over 200 million people world-wide and can cause serious morbidity. Infection occurs following exposure to larvae (cercariae) which release excretory/secretory (E/S) material to aid their entry into exposed skin. Larvae mature into adult worms that produce hundreds of eggs per day which also release E/S material. Both sources of E/S material have the potential to stimulate the host’s innate immune system. Circulating monocytes are important cells that act as potential sentinels in the recognition of these E/S materials. Different sub-sets of human monocytes can be identified according to their expression of CD14 and CD16 but their role following infection with schistosome helminths has not been investigated. In the current study, three sub-sets (classical, intermediate and non-classical) were enumerated in individuals living in a region co-endemic for S. mansoni and S. haematobium. Although all three monocyte sub-sets bound to fluorescently-labelled schistosome E/S material, the intermediate sub-set had significantly enhanced ability to recognise cercarial and egg E/S in co-infected participants. This is the first demonstration that circulating human monocytes can recognize schistosome E/S antigens and that their ability to do so is modulated by infection which may affect the development of schistosome immunopathology and/or protective immunity.
| Helminth parasites infect over 1 billion of the world’s population causing a range of overt morbid diseases and can exert substantial modulatory effects on the immune system [1], [2]. Schistosoma mansoni and S. haematobium are chronic, blood-dwelling, parasitic helminth infections of humans [3] and are co-endemic in many parts of Africa. Both species can cause life-threatening morbidities including damage to the intestines and liver (S. mansoni), or urogenital tract and kidneys (S. haematobium) [4].
Schistosome infection of the mammalian host is by skin penetration following exposure to waterborne cercariae [5], [6] which release excretory/secretory (E/S) material containing an abundance of glycosylated molecules [7] and proteases [8]. These E/S products aid penetration and migration of larvae, and consequently can act as initial stimuli of the cutaneous innate immune system [9]. Schistosome E/S products released in the first 3 hours after infection (0-3hRP) [10] stimulate both dendritic cells (DC) and macrophages (M) through binding of constituent ligands to pattern recognition receptors (PRR) such as Toll-like receptors (TLRs) [11], and C-type lectins (CLRs) including the mannose receptor (MR) [12]. These E/S products also have immune-modulatory effects on antigen presenting cells (APCs) such as DC in vitro and in vivo [13], particularly after repeated exposures, which can impact on downstream modulation of anti-schistosome responses and immunopathology in the liver [14].
Following migration and maturation, adult schistosome worms pair in the venous blood supplying the intestines (S. mansoni), or the bladder and urogenital tract (S. haematobium), and commence release of hundreds of eggs per day [15]. Mature eggs provide another source of glycosylated E/S products [16] termed egg secreted products (ESP). This E/S material may be critical in mediating the transit of eggs across host tissues to reach the external environment [17] and is implicated as a mediator of egg-related granulomatous immunopathology via induction of pro-fibrotic Th2 responses [18], [19]. Interestingly, an abundantly expressed ESP, Omega 1 [20], mediates Th2 priming via internalization into DC following ligation of the MR [21].
Despite the important role of E/S products in schistosome invasion, tissue migration and transmission of eggs, combined with their observed immunological priming and modulatory capacities in murine infection models, analysis of human immune responses to E/S material is very limited. Recently, we investigated cercarial E/S stimulation of whole blood cultures (WBC) from individuals from a region in Senegal which is co-endemic for S. mansoni and S. haematobium [22]. We identified significantly elevated levels of immune-regulatory IL-10, and increased ratios of IL-10:TNFα in infected individuals indicative of enhanced regulatory immune cell activity [22]. As the WBC culture supernatants were harvested at 24 hours post-stimulation with E/S material, the cytokines produced were most likely derived from the innate immune cell compartment (e.g. monocytes).
In this report, we extend our previous study by examining, for the first time in the context of human helminth infection, the parasite E/S pattern-recognition profiles of circulating monocyte sub-sets. We classified peripheral blood monocytes according to their expression of CD14 and CD16 in order to identify three sub-sets corresponding to ‘classical’ (CD14brightCD16−), ‘intermediate’ (CD14brightCD16+),and ‘non-classical’ (CD14dimCD16+) monocytes which have recently emerged as mediators of divergent immune functions during infectious disease [23]–[30]. Our study shows that intermediate CD14brightCD16+ monocytes have a greater ability to bind both cercarial and egg E/S products than other monocyte sub-sets, and that this capacity is greater in patients co-infected with both schistosome species compared to uninfected controls or those infected with S. mansoni only.
This study was approved by the review board of the Institute of Tropical Medicine, Antwerp, the ethical committee of Antwerp University Hospital and ‘Le Comité National d’Ethique de la Recherche en Santé’ Dakar, Senegal. Written informed consent was obtained from all participants. All community members were offered a single dose of praziquantel (40 mg/kg) and mebendazole (500 mg) after the study to clear helminth infection.
Participants were recruited from the village of Diokhor Tack (N16.19°; W15.88°) in a region co-endemic for S. haematobium and S. mansoni [31]. Each participant provided two stool and two urine samples (with a minimum total volume of 10ml urine) on consecutive days to quantify schistosome eggs microscopically as described previously [22]. Participants were classified as ‘mono-infected’ if they had an S. mansoni egg count ≥1 egg in one or more of their stool samples and ‘co-infected’ if they were also found to have ≥1 S. haematobium egg in one or more of their urine samples. Participants infected with S. haematobium only were not included in this study. Of 54 participants who provided a blood sample, 4 were excluded for providing insufficient samples for parasitological analysis and 9 were excluded for providing insufficient blood volume to conduct all ligand binding assays.
The following ligands were used for binding studies of PBMCs: schistosome cercarial E/S (0-3hRP at 50 μg/ml), egg E/S product (ESP at 25 μg/ml), zymosan-coated AlexaFluor488 conjugated bio-particles (0.5×106/tube; Life Technologies Ltd., Paisley, U.K.) as a positive control, and the fluorescein-labelled polyacrylamide glycoconjugate D-mannose (5 μg/ml; Lectinity Holding Inc., Moscow, Russia). Although zymosan is a yeast-derived ligand, zymosan bio-particles were selected as a positive control for the parasite E/S products because both ligands are heterogeneous in biochemical composition (containing carbohydrates, proteins and glycoproteins) and because like 0-3hRP, zymosan also stimulates in vitro cultured DCs to acquire a pro-Th2 activity [10]. D- mannose acted as a control for mannose receptor-mediated ligand binding, based on the knowledge that 0-3hRP contains an abundance of mannosylated glycans [7] which are important ligands for the macrophage MR [12]. 0-3hRP and ESP were prepared as previously described [10], [17], [32]. After isolation and purification, 0-3hRP and ESP were conjugated to AlexaFluor488 carboxylic acid 2,3,5,6-tetrafluorophenyl ester (Life Technologies Ltd, Paisley, U.K.) using established protocols [33].
Venous blood was collected in heparin coated-tubes [22], separated by density centrifugation (1400 rpm, 25 min, room temperature) on Ficoll (GE Healthcare, Pollards Wood, U.K.) and the resulting PBMC layer re-suspended at 10×106 cells/ml in ice-cold phosphate buffered saline (PBS) containing 0.5% bovine serum albumin and 2 mM EDTA (Sigma Aldrich, St. Louis, U.S.A.). Aliquots of PBMC suspension (5×105 cells) were transferred to 1.5 ml eppendorf tubes containing 50 μl of diluted ligands and monoclonal antibody (mAb) cocktail (see below) on ice. After a brief vortex, the suspension of PBMCs, ligands and mAbs was incubated on ice for 60 mins, with another vortex after 30 mins. For each participant, an aliquot of PBMC was incubated without ligands as a ligand-free control. PBMC were then washed with 900 μl ice cold PBS, pelleted at 800 g for 5 mins, before being re- suspended in 500 μl cell buffer containing 1% formaldehyde. PBMC were stored at 4°C in the dark before analysis by flow cytometry.
PBMC aliquots were surface-stained with fluorescently labeled anti-CD14 (conjugated to allophycocyanin; CD14-APC) and anti-CD16 (conjugated to eFluor450; CD16-ef450) mAb (eBioscience, San Diego, U.S.A.). Data was acquired with a Cyan flow cytometer (Beckman Coulter Ltd., High Wycombe, U.K.) and analysed using FlowJo software version 7.6.5 (TreeStar, Ashland, U.S.A.). Cells were gated according to forward- and side-scatter characteristics (SSC) and 5000–10000 total events acquired. An SSChi gate was used to select for cells with high granularity, which are primarily monocytes, whereas lymphocytes and NK cells were found in the gate for cells with low granularity. Polymorphonuclear granulocytes were mostly excluded by our Ficoll separation. SSChi PBMC were further sub-divided via CD14 and CD16 expression with gates determined relative to an aliquot of cells incubated with isotype control mAb to identify separate monocyte sub-sets. Ligand-free PBMC controls were used to determine threshold fluorescence intensity for AlexaFluor488 or fluorescein above which monocyte sub-sets were considered to be positive for ligand binding (ligand+ gate).
The software package IBM Statistics version 19 (Armonk, U.S.A.) was used for all statistical analyses. Mean proportions of monocyte sub-sets for each participant were calculated from proportions identified by cytometry of multiple aliquots of PBMC used for ligand binding assays (n = 5/participant). Proportions of monocyte sub-sets within the SSChi gate, proportions of each sub-set in the SSChi ligand+ gate and proportions of ligand+ monocytes within each monocyte sub-set met the assumptions for parametric analysis. Thus, comparison between monocyte sub-sets was made using paired t-tests, and comparisons between schistosome infection groups were made using ANOVA. Post-hoc pair-wise comparisons were made for significant ANOVA using Fisher's test. As the total proportions of SSChi cells in the ligand+ gate and median fluorescence intensity (MFI) values did not meet the assumptions of parametric tests, even after transformation, statistical comparisons were made using non-parametric tests. The paired Wilcoxon test was used to compare the proportion of ligand+ SSChi cells relative to the ligand-free control. For comparison between infection groups, a Kruskal Wallis test was first used to determine statistical difference between groups and pair-wise Mann Whitney U tests were used post-hoc to identify which groups differed.
The study population and schistosome infection status of the participants is detailed in Table 1 and comprised a total of 41 individuals aged 6 to 60 years old. When assigned by infection status, the three groups had similar sample sizes, age-range and sex ratios.
Following labelling with mAbs specific to CD14 and CD16, three discrete populations of SSChi monocytes were identified according to their recent characterisation and nomenclature in human peripheral blood; i) CD14brightCD16− ‘classical’ monocytes, ii) CD14brightCD16+ ‘intermediate’ monocytes and iii) CD14dimCD16+ ‘non-classical’ monocytes (Fig. 1A). Of the two CD14+ monocyte sub-sets, the CD14brightCD16− population was more abundant (16.2±1.5% of total SSChi) than the CD14brightCD16+ (6.8±0.7% of total SSChi) sub-set. Although the CD14dimCD16+ population was the most abundant overall (19.5±1.6% of total SSChi), it may include a small number of CD14− granulocytes as previously noted [28] but CD14dim/-CD16+ NK cells were excluded from our analysis as they do not have a high granularity phenotype defined as SSChi.
When the relative abundances of the three monocyte sub-sets were compared by infection status, no statistically significant differences in the proportions of each sub-set between the un-infected, infected with S. mansoni only, or co-infected with S. mansoni and S. haematobium groups were identified (Fig. 1B, p>0.05 for all comparisons).
The ability of SSChi cells to recognise schistosome and non-schistosome pathogen-associated molecular patterns (PAMPs) was investigated by examining their ability to bind each of the fluorescently-conjugated ligands (Fig. 2A). SSChi monocytes bound the two schistosome E/S products, zymosan bio-particles and D-mannose and had significantly greater proportions of cells within the ligand+ gate than their corresponding ligand-free controls (Fig. 2B, p<0.001 for all comparisons).
Within the SSChi population, CD14bright monocytes (both the classical CD16− and intermediate CD16+ sub-sets) were more efficient at binding to cercarial E/S than the CD14dim non-classical population (Fig. 3A; CD14brightCD16− t: 3.57, p<0.01, CD14brightCD16+ t: 6.30, p<0.001). Intermediate CD14brightCD16+ monocytes were also the most efficient at binding egg E/S products compared to the other monocyte sub-sets (Fig. 3B; cf. CD14brightCD16− t: 6.438, p<0.001, and cf. CD14dimCD16+ t: 9.29, p<0.001). However, CD14brightCD16+ intermediate monocytes were less efficient in their binding of the control ligands zymosan and D-mannose than the other monocyte sub-sets (Fig. 3C & D). A greater proportion of classical CD14brightCD16− monocytes bound egg E/S than non-classical CD14dimCD16+monocytes (Fig. 3B; t: 2.87, p = 0.007) and classical monocytes also bound zymosan with the greatest efficiency compared to intermediate (Fig. 3C; t: 6.41, p<0.001) and non-classical monocytes (t: 4.40, p<0.001). The classical and non-classical sub-sets had equivalent binding efficiency to D-mannose (Fig 3D; t: 0.09, p = 0.931) and both bound D-mannose with greater efficiency than the intermediate subset (CD14brightCD16− t: 5.45, p<0.001, CD14dimCD16+ t: 4.48, p<0.001).
Having established functional distinctions between the three SSChi populations in their ligand binding capacity, we investigated whether ligand binding within each monocyte sub-set depended upon participant infection status within the study population. Similar proportions of classical monocytes bound cercarial E/S in all 3 infection groups (Fig. 4A; F2, 38: 2.12, p = 0.135). In contrast, ligand uptake by intermediate monocyte was influenced by infection status (F2, 38: 4.93, p = 0.013) with a significantly greater proportion of intermediate monocytes from co-infected participants binding to cercarial E/S than those from uninfected subjects (Fig. 4A, mean: 55.64±5.31% versus 34.10±4.08%, p = 0.003). A greater proportion of non-classical monocytes also bound cercarial E/S products in mono-infected compared with un-infected (Fig. 4A, 13.64±1.55% versus 7.69±1.19%, p = 0.004) or co-infected participants (9.83±1.12%, p = 0.041).
In terms of the quantity of cercarial E/S bound by each monocyte sub-set (determined by MFI), although classical monocytes bound similar amounts in all infection groups (Fig. 4B, Kruskal Wallis; Χ2: 2.90, p = 0.235), the amount bound by intermediate monocytes differed according to participant infection status (Fig. 4B, Kruskal Wallis; Χ2: 10.01, p = 0.007). Hence, CD14brightCD16+ intermediate monocytes from co-infected patients bound significantly greater quantities of antigen as judged by their higher MFI ( = 231.90) than either mono-infected ( = 34.34, p<0.05) or uninfected participants ( = 10.70, p<0.01). In fact, intermediate monocytes from some co-infected individuals were particularly efficient at binding cercarial E/S (i.e. 7 co-infected participants bound >2-fold greater quantities of cercarial E/S than the group median, Fig. 4B). There was no significant difference between the three infection groups in the amount of cercarial E/S bound by non-classical monocytes (Fig. 4B, Kruskal Wallis; Χ2: 1.89, p = 0.388).
The proportions of classical monocytes that bound schistosome egg E/S products were similar between the three infection groups (Fig. 5A, F2, 38: 0.693, p = 0.506), as were the amounts of egg E/S bound by this sub-set (Fig. 5B, Kruskal Wallis; Χ2: 2.11, p = 0.348). However, significant infection-related differences were evident in the intermediate monocyte population (Fig. 5A; F2, 38: 3.59, p = 0.037). Greater proportions of intermediate monocytes from co-infected subjects bound egg E/S (Fig. 5A; 46.53±5.14%) than those isolated from mono-infected participants (12.72±4.00%; p = 0.015). There was also a non-significant trend for a greater proportion of intermediate monocytes recognising egg E/S in co-infected versus un-infected patients (Fig. 5A; 32.72±6.28%, p = 0.075). In addition, intermediate monocytes varied in the amount of bound egg E/S according to infection status (Fig. 5B; Kruskal Wallis; Χ2: 9.60, p = 0.008) with those from the co-infected group binding significantly greater quantities (MFI: 56.03) than those from un-infected individuals (Fig. 5B; MFI: 5.30, p = 0.015). There was no difference in the proportions of egg E/S+ non-classical monocytes (Fig.5A, F2, 38: 1.64, p = 0.208), nor in the amount of egg E/S uptake by this sub-set between the 3 infection groups (Fig. 5B, Kruskal Wallis; Χ2: 0.972, p = 0.615).
Evaluation of the binding recognition profiles for zymosan (a yeast- derived bioparticle) and D-mannose (a purified glycan) by the different monocyte sub-sets between schistosome infection groups did not reveal any significant differences (Supplementary Figs. S1 & S2). Although not directly comparable with our schistosome-derived E/S products, these observations confirm that the significantly elevated proportions of CD14brightCD16+ intermediate monocytes from co-infected participants that bound cercarial, or egg E/S products were not due to an increase in non-specific binding regardless of the ligand.
Monocytes are recruited to tissue sites of inflammation and infection and are precursors of specific M and DC populations at tissue sites (e.g. the skin and intestines). Thus, monocytes potentially have both innate immune and subsequent APC functions in response to different stages of the schistosome parasite in multiple tissue sites. Although human circulating monocytes have traditionally been identified on the basis of CD14 (a lipopolysaccharide co-receptor) expression, more recently they have been further subdivided according to the expression of the low affinity Fc receptor, CD16, which could define their binding to various ligands and their immune function [28]. In mice, which are an established in vivo experimental model of human schistosomiasis, monocytes have been defined on the basis of expression of the surface markers Ly6C, CD115 and the chemokine receptor CCR2, and are thought to be selectively recruited to inflamed tissues [34]. However, monocytes that are Ly6C- are thought to differentiate into alternatively activated M [34], [35] which dominate after multiple exposures to schistosome cercariae and following chronic long-term infection [14], [36]. However, due to limited characterization of the distinctions between murine and human monocyte markers, it is difficult to directly translate findings with murine models into human disease. On the other hand, a consensus on the definition of monocyte sub-sets in humans based upon their pattern of labeling with anti-CD14 and anti-CD16 mAbs is emerging, despite ongoing debate over the functional distinctions between these sub-sets (see commentary in [29]). Therefore, in our study of PBMC monocytes collected from individuals inhabiting a schistosome-endemic region of northern Senegal, SSChi monocyte sub-sets were classified as being CD14brightCD16- (‘classical’), CD14brightCD16+ (‘intermediate’), and CD14dimCD16+ (‘inflammatory’/‘non-classical’).
We are not aware of a precedent study that has examined the ability of human monocyte sub-sets to differentially bind innate immune cell ligands and thus our study is the first to demonstrate functional distinctions between the three monocyte sub-sets in their capacity for pattern recognition of schistosome-derived E/S. In particular, we show that CD14bright monocytes are more efficient at binding to schistosome E/S antigens than CD14dim non-classical monocytes, highlighting a potential role for classical and intermediate monocytes in innate sensing of human schistosome infections. Moreover, our study is the first to demonstrate that schistosome infection status affects the binding of CD14brightCD16+ intermediate monocytes, but not the other two monocyte sub-sets, to schistosome E/S ligands. We show that a significantly greater proportion of CD14brightCD16+ intermediate monocytes from S. mansoni and S. haematobium co-infected participants recognize and bind schistosome E/S products from cercariae and mature eggs, than the same sub-set of monocytes obtained from mono- and uninfected subjects. Furthermore, this intermediate sub-set also binds greater quantities of E/S antigen. This may be due to greater numbers of S. mansoni eggs in co-infected compared to mono-infected patients (Table 1), to the additional presence of S. haematobium, or to a combination of both factors. Together, this data indicates that schistosome infection affects the surface receptor repertoire of CD14brightCD16+ monocytes, enabling them to become more sensitive to recognition of schistosome-secreted molecules, which may enhance subsequent recruitment of this sub-set to tissue sites of infection. This is potentially of significance in the development of schistosome-specific protective immunity or immunopathology. Indeed, a functional role for intermediate monocytes in the development of severe malaria has previously been proposed [37].
Candidate monocyte surface receptors that may be influenced by infection status and are known to be involved in pattern recognition of schistosome cercarial and egg E/S ligands include surface TLRs (2 and 4) and the phagocytic C-type lectin, MR, previously implicated in glycosylated schistosome molecule recognition [11], [12], [21], [38]-[40]. Ligation of MR by cercarial E/S has an immune modulatory effect [12], possibly acting on TLR signaling [41] as proposed for other schistosome-derived glycans [39]. However, because D-mannose was not differentially recognized in the three infection groups in our study, schistosome E/S recognition in intermediate monocytes may be independent of infection-related changes in MR expression. Furthermore, as binding of zymosan (purified yeast cell wall); a commonly encountered PAMP recognized by both TLR2 and the β-glucan C-type lectin, Dectin-1 [42], to each of the three monocyte sub-sets did not vary significantly between infection groups, it is unlikely that increased recognition of parasite E/S in infected individuals cause alterations in the TLR2 and Dectin1 PRR complex.
The differential expression of surface receptors, other than CD14 and CD16, such as TLRs and C-type lectins, was beyond the scope of this first investigation of monocyte heterogeneity in helminth-infected humans. However, our findings indicate that determining which PRRs are differentially expressed between the various sub-sets in the context of infection status would be pertinent. Moreover, since it is unknown at present whether differential PRR expression is dependent upon the origin (e.g. bacterial, fungal, protozoan or helminth) of the stimulatory ligands to which they are exposed, an investigation of PRR expression by different monocyte sub-sets in response to schistosome antigens alongside appropriate defined control antigens from other pathogen sources (e.g. zymosan), is a valid area for further investigation. In addition, it would be desirable to determine whether binding of the different parasite E/S products to these PRRs proceeds to endocytosis and how this impacts on monocyte-derived APC function. Previous investigations have already begun to further subdivide the three monocyte sub-sets described here according to surface expression of Major histocompatibility molecules and chemokine receptors [37], [43] although their functional relevance to pattern recognition by monocytes has yet to be investigated. It would also be instructive to determine whether activation, or regulatory signals, such as secretion of different cytokines and chemokines, are induced in the respective monocyte sub-sets following ligation of parasite E/S products at the cell surface.
Monocytes expressing CD16 have been regarded as ‘pro-inflammatory’ according to their cytokine secretion profile (i.e. high TNFα and low IL-10 in response to LPS) and their ability to present antigen, suggesting they are more mature than the CD16− classical monocyte sub-set [24]. CD16+ monocytes are also more liable to develop into M or DC [44] and the CD14brightCD16+ intermediate sub-set is usually expanded under inflammatory disorders [28], [45], although our data indicates that this is not the case for schistosomiasis. However, the intermediate monocyte sub-set has also been identified as a major source of the regulatory cytokine IL-10 [46]. The latter contention is supported by data suggesting that CD14brightCD16+ intermediate monocytes may act in an anti-inflammatory manner in response to infectious pathogens such as Plasmodium protozoa [37]. This raises the question as to whether the function of CD14brightCD16+ intermediate monocytes (i.e. having a pro-inflammatory versus a regulatory role) depends on the specific molecular composition of the stimulatory microbial or inflammatory ligand. Interestingly, it has recently been reported that CD16+ monocytes, which are abundant (∼40%) in patients infected with Mycobacterium tuberculosis, fail to differentiate into mature DC [47] and can adversely affect the ability of classical CD14+ monocytes to differentiate into DC [47]. Therefore, CD16+ monocytes under the influence of specific microbial ligands may give rise to immune-regulatory DC, or divert differentiation of tissue macrophages to having anti-inflammatory properties [48]. In this context, we have previously shown that murine bone-marrow derived DC exposed to cercarial E/S fail to mature taking on a ‘modulated’ or ‘regulatory’ phenotype [32] and release abundant regulatory IL-10 [10]. Thus, in light of the report that CD14brightCD16+ intermediate monocytes produce abundant IL-10 [46], it would be pertinent to determine whether the elevated levels of IL-10 released by WBC from schistosome-infected patients in response to stimulation with cercarial E/S [22] are due to elevated numbers of CD14brightCD16+intermediate monocytes that have bound to cercarial and/or egg E/S products.
In conclusion, our study shows that circulating CD14brightCD16+ intermediate monocytes have a hitherto un-appreciated potential to specifically bind schistosome E/S material which may ultimately shape the development and function of monocyte-derived myeloid cells (e.g. M and DC) recruited to parasitized tissue sites during schistosome infection. In addition, the ability of CD14brightCD16+ intermediate monocytes to recognize parasite-derived E/S molecules is enhanced in schistosome-infected patients compared with uninfected individuals, suggesting a mechanism of modulation in surface-expression of parasite pattern recognition receptors on this specific monocyte sub-set. Future lines of study should include: identification of the innate immune cell PRRs involved in the binding of E/S products, investigation of the cytokine secretion and activation profile of ligand+ monocytes, the fate of bound ligands (e.g. internalization and intracellular processing), and analysis of the functional potential of E/S-exposed monocytes (e.g. phagocytosis and/or antigen presentation). In spite of these unknowns, our study indicates that exposure to schistosome-derived E/S products may profoundly influence the function of circulating monocyte sub-sets, which in turn may have substantial modulating effects on human immune reactivity. Importantly, differences in the responsiveness of circulating APC precursors to schistosome E/S material may impact upon permissiveness to invading cercariae at the cutaneous site of infection and the development of immunopathology around eggs sequestered in host tissues.
|
10.1371/journal.pcbi.1002856 | Divide and Conquer: Functional Segregation of Synaptic Inputs by Astrocytic Microdomains Could Alleviate Paroxysmal Activity Following Brain Trauma | Traumatic brain injury often leads to epileptic seizures. Among other factors, homeostatic synaptic plasticity (HSP) mediates posttraumatic epileptogenesis through unbalanced synaptic scaling, partially compensating for the trauma-incurred loss of neural excitability. HSP is mediated in part by tumor necrosis factor alpha (TNFα), which is released locally from reactive astrocytes early after trauma in response to chronic neuronal inactivity. During this early period, TNFα is likely to be constrained to its glial sources; however, the contribution of glia-mediated spatially localized HSP to post-traumatic epileptogenesis remains poorly understood. We used computational model to investigate the reorganization of collective neural activity early after trauma. Trauma and synaptic scaling transformed asynchronous spiking into paroxysmal discharges. The rate of paroxysms could be reduced by functional segregation of synaptic input into astrocytic microdomains. Thus, we propose that trauma-triggered reactive gliosis could exert both beneficial and deleterious effects on neural activity.
| Homeostatic plasticity refers to the ability of neurons and neuronal circuitry to adjust their properties in order to maintain physiologically relevant electrical activity notwithstanding perturbations in synaptic input. Synaptic input is often chronically reduced immediately following brain trauma, and previous studies had suggested that homeostatic synaptic plasticity can aid in the dynamical transition of the traumatized network toward epileptic seizures, a condition known as “post-traumatic epilepsy”. This form of homeostatic plasticity is mediated by glial cells which release regulatory molecules shortly after trauma. In this study we used computational modeling to investigate the mechanisms and the implications of glial mediated plasticity early after trauma. We show that astrocytes (a subtype of glial cells) exert both beneficial and deleterious effects on post-traumatic reorganization of neural activity. This suggests that, in the dysfunctional neuronal network, some aspects of glial-neuronal signaling could alleviate the dynamical transition to pathological activity.
| Post-traumatic epilepsy develops in some but not all head injury cases, depending on the severity of injury and the time elapsed since trauma. Often there is a latent period between the traumatic event and onset of paroxysmal activity [1]. Identification of neurological mechanisms underlying this latency to seizures can offer a possibility for therapeutic intervention. Experimental and modeling studies suggest that this slow transition from normal to paroxysmal activity might depend on homeostatic adjustment of synaptic conductances, connectivity and intrinsic excitability properties [2], [3].
Homeostatic synaptic plasticity (HSP) likely operates on several spatial and temporal scales [4]. Chronic synaptic and neuronal inactivity, such as the one that often occurs following trauma, engages glial cells to release tumor necrosis factor alpha (TNFα) [5], [6], [7]. This relatively slow process (global effects in culture are measurable after ∼48 hours of inactivity [6]) may represent a global “network response” to prolonged inactivity [8]. The strengthening of inputs from the open eye during monocular deprivation is another slow process that is mediated by TNFα [7], [8], [9]. Early after trauma elevated levels of TNFα are likely to be spatially localized to their glial sources, implying spatial localization of homeostatic synaptic plasticity.
Earlier studies showed that TNFα causes a rapid, p55 receptor mediated insertion of neuronal AMPA receptors [10], and endocytosis of GABA receptors [6]. Thus, TNFα could promote epileptogenesis by shifting the excitation-inhibition balance in favor of excitation. Consistent with this, systemic administration of TNFα [11] and constitutive over-expression of TNFα [12] had pro-epileptic effects. Seizure incidence was dramatically reduced in knockout mice lacking p55 TNFα receptors [13], [14]; susceptibility to seizures was reduced following systemic pre-injection of TNFα antibodies [15]. These data suggest that TNFα can promote epileptogenesis [16]. Given the role of TNFα in HSP [7], [8], [9], the implication is that homeostatic synaptic plasticity can drive the traumatized network toward epileptic activity [3], [17], [18], [19].
In our previous studies [3], [18] we showed that trauma-triggered HSP can transform cortical activity from asynchronous spiking (∼5 Hz for pyramidal neurons, ∼10 Hz for inhibitory neurons) to paroxysmal bursting, and we further showed that the pattern of trauma changes the threshold for epileptogenesis [20]. In those studies we implicitly assumed that HSP represents the action of TNFα which is released in response to chronically low levels of neuronal activity incurred by the traumatic injury. We also assumed that HSP adjusted synaptic conductances in a manner that depended on the network-global averaging of neuronal activity. The assumption of global network averaging of neuronal activity is likely to be valid at sufficiently long time after trauma, when levels of TNFα had equilibrated throughout the network. However, at a short time (several hours) after trauma, elevated levels of TNFα are likely to be localized around their glial sources [21], thus implying spatial localization of HSP and spatially heterogeneous disruption of excitation-inhibition balance that may strongly favor the transition to seizures. Given the extensive evidence for the dramatic involvement of TNFα in post-traumatic epilepsy [16], high levels of localized TNFα a short time after brain injury [21] are not consistent with the relatively low incidence of paroxysmal spikes and seizures during that period.
In the present study, we addressed this question by studying the early effects of TNFα mediated HSP hours after trauma. Homeostatic synaptic plasticity restored the average network firing rate to its pre-traumatic level but transformed asynchronous spiking to paroxysmal bursts. Thus, we adopted the rate of paroxysmal burst generation (rather than the network-averaged firing rate) as a measure of network's propensity to exhibit the transition to seizures. Paroxysmal bursts of highly correlated population activity in our model resembled interictal epileptiform discharges (IEDs), which are often considered an important diagnostic feature of epileptic seizures [22], [23], [24]. Thus, a higher rate of population bursting in a post-traumatic model network was considered an indicator of stronger propensity to seizures. With spatially constrained HSP (“local HSP”), representing local synaptic scaling by TNFα, paroxysmal bursts occurred in post-traumatic network at a high rate, with little dependence on the fraction of deafferented neurons (trauma volume). This was in striking contrast to the gradual dependence of burst rate on trauma volume that characterized the later stage of “global HSP”. Properties of paroxysmal discharges could be modulated by functional segregation of synaptic inputs into reactive astrocytic microdomains [25], [26]. Thus, our modeling studies suggest that some aspects of reactive astrogliosis might alleviate paroxysmal activity early after trauma.
The primary goal of the present study was to explore the impact of spatially localized homeostatic synaptic plasticity (HSP) on the emergence of paroxysmal activity in deafferented post-traumatic cortical network and to identify mechanisms that might operate to suppress this activity in the early post-traumatic window. We have shown previously that the spatial pattern of trauma itself can significantly affect the trauma threshold for paroxysmal burst generation [20]. In addition, as we show below, segregated synaptic regulation by astrocytic microdomains can also affect post-traumatic electrical activity.
Homeostatic regulation is likely to operate on a localized spatial scale during the early phase of response to trauma, reflecting the local response of glial cells to nearby synaptic activity. Such a change of scale could in principle affect our earlier conclusions regarding the role of trauma pattern in post-traumatic epileptogenesis [20]. In particular, in our earlier studies [20] we found that the trauma threshold for the emergence of paroxysmal events in post-traumatic network depended on the pattern of trauma itself. In those studies, the extent of trauma was parameterized by the fraction of deafferented model neurons,, which we will refer to as the “volume of trauma”. When burst rate was plotted vs. the trauma volume parameter, focal trauma (spatially contiguous set of deafferented neurons) caused lower burst threshold as compared to diffuse trauma (spatially randomly distributed set of deafferented neurons). Thus, it was important to validate the conclusions of our earlier studies regarding the role of trauma spatial organization in generation of post-traumatic paroxysmal activity.
Here, we assumed that the downregulation of excitatory synaptic conductances in a computational model of a hyperactive pyramidal (PY) neuron was determined by the time-averaged firing rate of the postsynaptic neuron, consistent with postsynaptic synaptic scaling [4]. On the other hand, upregulation of excitatory synaptic input in response to reduced levels of synaptic activity was determined in our model by the time-averaged firing rate of all PY neurons that projected their synapses to a PY neuron under consideration, corresponding to glial scaling of synaptic conductances by TNFα [8]. Thus, the baseline model is described by “local UP” regulation and “local DOWN” regulation of synaptic conductance. This model is referred to below as a local HSP model.
To compare with our previous results [20] we also used a global HSP model, in which both pre- and postsynaptic components of homeostatic synaptic scaling were determined by the global, network-averaged, firing rate of model PY neurons. Thus, the “global HSP” model that was used in our previous studies [20] is described by “global UP” regulation and “global DOWN” regulation of synaptic conductance. A shift in the spatial scale of HSP rule in the present model would correspond to the transition from the early phase of post-traumatic reorganization (during which upregulation of synaptic conductance is constrained to glial sources of TNFα) to the later phase (of equilibrated levels of TNFα).
In some simulations (e.g., Figure 1C) we only changed the spatial scale of presynaptic, upregulating, HSP component from local (averaged only over those PY neurons that project their synapses to a given neuron, as in “local HSP” or baseline model) to global (averaged over all PY neurons in the network, “global UP” model in Figure 1C). Note that the “global UP” model differs from “global HSP” model in that in the former the downregulating postsynaptic HSP component is local (i.e., based on the activity of the specific postsynaptic neuron, similar to the baseline model). In other simulations, the downregulating postsynaptic HSP component was removed altogether from the model network, to assess the impact that homeostatic downregulation of synaptic conductance might have on collective activity; this model is referred to as “global UP no DOWN”. Note again that the “global UP no DOWN” model differs from the “global HSP” model in that in the latter the downregulating component of HSP is present. In yet other simulations, the set of synaptic inputs in the local HSP scheme was further randomly and evenly partitioned into several groups, for each of which we applied equations that described presynaptic component of HSP (Materials and Methods). Such partitioning into sub-groups of synaptic inputs was taken to mimic partition of the cortex into astrocytic microdomains [25], [27]. Finally, we compared two different patterns of trauma: focal trauma (in which a spatially contiguous subpopulation of neurons was deafferented, (e.g., Figure 1A1)), and diffuse trauma (in which deafferentation affected fraction of neurons randomly selected from the entire network (e.g., Figure 1A2)). In what follows, the “baseline” model network configuration is defined as a network with one microdomain per neuron, local HSP rule, and subject to focal trauma.
Within the “early response” scheme of the HSP based on the activity of presynaptic neurons (local HSP, see above and Materials and Methods), the rate of paroxysmal bursts was significantly higher for focal trauma (Figure 1A1) compared to the spatially diffuse trauma (Figure 1A2), and this distinction was observed over a wide range of trauma volume parameter values, . This was generally consistent with earlier studies in which we showed that the spatial pattern of trauma could critically affect the threshold for post-traumatic paroxysmal activity [20]. In Figures 1A1, 1A2, we also plotted the results obtained with the global HSP model (in which both up and down regulation of synaptic conductance was determined based on the global network-wide average over activities of pyramidal neurons), to compare them with the present model, which made use of local HSP. Although the two models produced the same result qualitatively (both showed an increase in the rate of paroxysmal activity above some critical threshold of trauma volume parameter, and in both cases the threshold was higher for the diffuse trauma), the threshold for paroxysmal activity appeared to be much lower in the case of local HSP rule.
The difference between the two HSP models was also reflected in the dynamics of the network-averaged firing rate of the PY neurons, shown in Figure 1B1 for the case of focal trauma. In the local HSP model of the focal trauma, and for relatively small trauma volumes (), the network-averaged firing rate of the pyramidal neurons showed a much steeper transition to its post-traumatic target value compared to the more gradual change observed within the global HSP model (Figure 1B1, compare solid red lines for local HSP model and dashed red lines for global HSP model). In contrast, after more severe trauma () the approach of the network-averaged firing rate of the model PY neurons toward its post-traumatic target value was more similar for both the local and global HSP scenarios (Figure 1B1, black solid and black dashed lines for local and global HSP, respectively).
The quantitative differences between the two HSP models, as reflected in the dynamics of the network-averaged firing rate of model pyramidal neurons, were discernible also within the scenario of diffuse trauma (Figure 1B2). However, in the diffuse trauma scenario, there was no qualitative difference between the firing rate reorganization dynamics in different HSP models; both local and global HSP models caused gradual recovery of firing rate for relatively small trauma volume (Figure 1B2, solid red and dashed red for local and global HSP models, ) and steeper transition in the case of more severe trauma (Figure 1B2, solid black and dashed black for local and global HSP models, ).
Thus, qualitatively, the strongest effect of HSP localization on the post-traumatic reorganization of electrical activity was observed for relatively small volumes of focal trauma. This is consistent with the results in Figures 1A1, 1A2, which show that the primary effect of local HSP is to lower the trauma volume threshold for burst generation and that this effect is more pronounced in the focal trauma scenario.
We next explored the intriguing independence of the paroxysmal burst rate on trauma volume in a model with a local HSP rule. Several mechanisms in the new model of homeostatic plasticity could have been responsible for this observation. It could have been a consequence of normalizing the neuronal firing rates, which was implemented by downregulating postsynaptic component of HSP (based on the firing rate of postsynaptic neuron, as suggested in [4]). Alternatively, the independence of burst rate on the trauma volume could follow from the local scale of the presynaptic component of the HSP, in contrast to the global, network-wide, scale of HSP employed in earlier models of late post-traumatic phase [3], [18], [20]. To test this second possibility, we replaced the local scale of the presynaptic HSP (for which the averaging of firing rates was performed over the set of those model PY neurons that projected synapses to a given neuron) with the global scale of the presynaptic HSP (for which the averaging of firing rates was performed over all model PY neurons). The postsynaptic component of HSP remained local and was determined by the firing rate of the postsynaptic neuron. As shown in Figure 1C1, this manipulation on the spatial scale of the presynaptic HSP component did not result in any significant effect on the burst rate – trauma volume relation.
We then tested the possibility that the apparent independence of burst rate on trauma volume was dominated by the postsynaptic downregulating component of HSP. When the postsynaptic component of HSP was excluded from the model and the global scale scheme was used for the presynaptic component, the linear relation (in the supra-threshold regime) between the burst rate and trauma volume was recovered (Figure 1C2, open green diamonds). Thus, it appeared that the downregulating postsynaptic component of HSP acted as a permissive factor, either allowing or preventing the burst rate to be modulated by the spatial scale of the presynaptic HSP component.
Earlier studies [3] suggested that post-traumatic paroxysmal activity arises because HSP acts to restore the firing rates of pyramidal neurons to their pre-traumatic value. Thus, we reasoned that the above dependence of burst rate on trauma volume during early stages of post-traumatic reorganization might, at least partially, be reflected in the firing rates of PY neurons. Thus, we estimated the dependence of pyramidal firing rate on the neuronal location in the network. For this analysis, neuronal firing rates were sampled from the center-symmetric strip (cross section was 5 cells from the center of the strip) of “neural tissue”, and at each point the firing rates of model PY neurons were averaged over the cross-section of the sampled strip.
With the postsynaptic downregulating component of HSP present, the firing rate of PY neurons (either deafferented or intact) was clamped at ∼5 Hz and did not depend on the spatial scale of presynaptic HSP component (Figures 1D1 and 1E1). Indeed, the postsynaptic HSP component prevented firing rate of any individual PY neuron to exceed its preset target rate. The firing rate of the intact neurons never increased and the firing rate of the deafferented neurons reached the target regardless of the presynaptic HSP model. Therefore the model with global presynaptic scaling was virtually indistinguishable from the baseline model (with focal trauma and local HSP) and for both models the firing rate was independent on the trauma volume (Figures 1E1).
When the global scale presynaptic scheme was combined with exclusion of the postsynaptic downregulating HSP component, the averaged firing rate of intact PY neurons at the traumatized-intact boundary — defined as a set of PY neurons located within one synaptic footprint from the boundary deafferented neuron (see Materials and Methods) — displayed strong dependence on the trauma volume parameter (Figure 1E2, open green diamonds). In this model, increase of the trauma volume led to an increase in the size of the deafferented (less active) PY population and, therefore, required a stronger increase in the firing rate of the intact PY population to keep the overall firing rate constant. This model allowed an increase because the postsynaptic downregulating HSP component was absent. Therefore, for intermediate and high trauma volumes (Figure 1E2), the firing rates of intact PY neurons were higher than those in the baseline model (with focal trauma and local HSP) and the firing rates of deafferented neurons were lower (Figures 1D2, black squares vs. green diamonds). Surprisingly, for intermediate trauma volumes (Figure 1C2, black squares vs. green diamonds) the relation between the corresponding rates of paroxysmal bursts was opposite to the one observed for neuronal firing rates (i.e., for intermediate values of the trauma volume parameter the burst rate in the “global UP no DOWN” model was lower than the burst rate in the baseline model with focal trauma and local HSP). This suggests that the burst rate was limited by the firing rates of deafferented neurons; even when intact neurons fired at higher rate, they only occasionally triggered bursts in the deafferented population. Thus, although the “firing rate–burst rate” relation hypothesis could qualitatively account for the dependence of burst rate on trauma volume, it failed to explain the quantitative differences between the two HSP scenarios (local vs. global models).
In our previous studies we showed that, in the deafferentation model of cortical trauma, paroxysmal activity is generated by the intact pyramidal neurons located at the boundary between intact and deafferented regions [20]. Because our previous studies utilized “global HSP” model, it was important to test whether or not the same conclusions would hold as well for networks with “local HSP” scenario. Figures 2A1, 2B1 show snapshots of spatial activity in deafferented regions of model networks (trauma volume parameter ), for “local HSP” (Figure 2A1) and “global HSP” (Figure 2B1), and in both scenarios it is seen that paroxysmal activity propagates in a wave-like manner, from intact and into the deafferented part of the network (in Figures 2A1, 2B1 image boundaries correspond to the boundary between the intact and deafferented regions of a network, so that only the dynamics in deafferented regions are shown). In Figures 2A2, 2B2 the spatial spread of activity is further quantified by computing and plotting, for two scenarios, the time-averaged firing rates of the model neurons. In these plots, the axis are the spatial dimensions of the network grid, and color codes the firing rate of individual neurons, averaged over a long time window (T = 50 s) after the network had reached its post-traumatic steady state. Destruction of synaptic connectivity between the intact and deafferented parts of the network completely eliminated paroxysmal activity (Figures 2A3, 2B3). This confirms that paroxysmal activity in the deafferented part of the network critically depends on the existence of functional synaptic connectivity with the intact part.
In our previous studies [28] we showed that the rate of post-traumatic paroxysmal bursts may be set not only by the firing rates of intact PY neurons; other determinants of collective activity include the spatial distribution of intact PY neurons and the strengths of their recurrent synapses [28]. This implies that the spatial scale of synaptic connectivity pattern — the spatial extent to which synaptic connections can be formed, as determined by the size of the synaptic footprint (see Materials and Methods for details) — might predispose the traumatized network to become more or less epileptogenic. Indeed, experimental evidence suggests that, following traumatic brain injury, a network is likely to undergo changes in its anatomical connectivity [17], [19], [29], [30]. Further, modeling studies showed that these changes in anatomical connectivity could breach the excitation-inhibition balance and generate epileptic-like seizures [2]. Although reorganization of synaptic connectivity likely occurs on much slower time scales than the ones we study here (days vs. hours), rapid and localized remodeling of synaptic connections was also reported [31].
We addressed the possible interplay of synaptic connectivity and HSP spatial scales by scaling up the size of synaptic footprint in the model. The synaptic footprint is the region from and to which a given model neuron could receive or send synaptic connections, and in the model network it was a 10×10 square centered at the neuron under consideration. As the size of synaptic footprint was scaled (by scaling the dimensions of the square region from and to which synapses could be received/sent), the probability of establishing synaptic contacts inversely depended on the number of potential pre-synaptic partners, as determined by the footprint size, in order to keep the average number of synapses to a given neuron the same, regardless of the footprint size. This allowed us to avoid conflating the effects of footprint size with an increase in synaptic connectivity. As Figure 3A1 shows, the rate of paroxysmal burst generation was smaller for larger synaptic footprint sizes (parameterized as the half-length of square side). This reduction in burst rate was paralleled by an increase in the rate of neuronal firing during the burst, which in turn stemmed from an increase in the number of spikes fired during the burst (Figures 3B1–3 for sample burst profiles, Figures 3C1,2 for quantification of intra-burst spiking activity). By contrast, within the global HSP scheme, manipulations of the synaptic footprint size averaging activity over all model PY neurons had a much weaker effect on the burst-rate trauma-volume relation (Figure 3A2).
Because burst nucleation in our model required the activity of a certain fraction of intact neurons to be sufficiently correlated (in order to be able to “ignite” their deafferented postsynaptic partners) [28], a reduction in the rate of paroxysmal discharge could signal reduced correlation between burst-triggering intact neurons. However, correlation between activities of intact neurons on the boundary did not exhibit any remarkable dependence on the size of synaptic footprint (Figure 3D1). The correlation between activities of deafferented neurons did grow up with the increasing size of synaptic footprint (Figure 3D2). Thus, the reduction in burst rate that was observed for a larger synaptic footprint did not depend on the reduced correlation between burst-igniting neurons.
Another possibility is that a reduced rate of burst generation reflects higher heterogeneity in interconnectedness and synaptic weights for synapses formed among neurons on the boundary between intact and deafferented regions. Indeed, distributions of HSP scaling factors at PY-PY synapses in post-traumatic steady state (Figure 3F) were characterized by larger standard deviations for scenarios with larger footprint sizes (Figure 3E). These results suggest that heterogeneity of interconnectedness and synaptic conductances at the intact-deafferented boundary could help to alleviate the onset of paroxysmal bursting activity, but this comes at the expense of more intense spiking activity during the burst.
Assuming that the heterogeneity of synaptic organization at the boundary between traumatized and intact regions is likely to be important in post-traumatic epileptogenesis, we sought to identify the physiological mechanisms that might mediate this effect. Experimental evidence suggests that homeostatic scaling of synaptic conductances might be at least in part mediated by the soluble tumor necrosis factor alpha (TNFα) that is believed to be released from astrocytes to compensate for low levels of glutamatergic synaptic activity [8]. It is well established that astrocytes can sense glutamatergic synaptic activity and respond to it with diverse spatio-temporal patterns of free cytosolic calcium [32], [33]. Although a typical astrocyte contacts ∼100,000 synapses [34], in a recent study, the calcium-mediated detection and modulation of synaptic release by astrocytes under physiological conditions was local, with regulation occurring independently along astrocytic processes (branches) in groups of 10 s of adjacent synapses [27]. These findings are consistent with the notion of astrocytic microdomains, with each microdomain responsible for the autonomous regulation of a small cluster of spatially proximal synapses [25]. Note that microdomains are morphological feature of astrocytes, and thus astrocytic microdomain should not necessarily contact synapses for regulation; however, any synapse that is regulated by an astrocyte belongs, by definition, to unique astrocytic microdomain.
Spatial localization of astrocytic signaling may translate into autonomous regulation of groups of synapses. Since we consider here the early stage of post-traumatic response (when a relatively high glial TNFα concentration is likely to be spatially constrained to its release sites) such autonomous regulation could increase the heterogeneity of synaptic conductances scaled by glia-mediated HSP. Thus, we hypothesized that functional segregation of synaptic inputs into astrocytic microdomains could help alleviate the rate of paroxysmal discharges in our model networks. To test this hypothesis, for each model PY neuron the set of all collateral synapses to it (from PY and IN neurons) was randomly partitioned into groups (microdomains), such that on average a group of synapses constituted a microdomain). Homeostatic scaling of synaptic conductances (both glutamatergic and GABAergic) for each astrocytic microdomain was determined independently by the time-averaged activity of glutamatergic synapses in it (according to Equations 11,12).
Several computational models were developed to describe interactions between astrocytes and synapses [35], [36], [37]; however, these models linked increased levels of synaptic activity to calcium elevations in astrocytes, and thus cannot explain how low levels of synaptic activity could culminate in astrocytic release of TNFα. Rather than attempting to develop a detailed mathematical model to describe this process, we assumed here that the ultimate effect of astrocytic microdomain activation is to scale synaptic conductances according to Equations 11,12. This approximation allowed us to avoid introducing additional complexity associated with biochemical cascades of activation in astrocytes [37] and to focus on the long-term network effects of interactions between neurons and astrocytes.
Figure 4A shows the dependence of paroxysmal discharge rate on the trauma volume, for several scenarios in which synaptic input to each model PY neuron was partitioned into several microdomains. We considered here scenarios of focal trauma. For values of trauma volume above a critical threshold, the burst rate still did not depend on the volume of trauma but the plateau value of paroxysmal burst rate now depended on the number of microdomains into which synaptic input set was partitioned. When plotted vs. the number of microdomains (for the same trauma volume), the burst rate monotonically decreased for larger numbers of microdomains (Figure 4B), and reached an asymptotic level of ∼0.3 Hz for microdomains, corresponding to a situation in which each synapse to a PY neuron was associated with an unique microdomain (each model PY neuron received, on average, 55 synapses from its fellow PY model neurons, and the maximal number of glutamatergic synapses per PY neuron was 75). For values , some microdomains had zero “synaptic occupancy” (i.e., they had no synapses to regulate) and thus did not take part in homeostatic synaptic plasticity. Notably, for the same trauma volume, the burst rate that emerged as a result of diffuse trauma also depended on the number of microdomains, but this dependence was much weaker compared to that of the corresponding focal trauma (Figure 4B, compare closed squares and open circles). Thus, even though diffuse trauma resulted in a lower rate of bursts in the model with one microdomain, for strongly segregated set of synaptic input it resulted in more frequent bursting than did the focal trauma.
The averaged intra-burst firing rates of PY and IN model neurons were higher for stronger segregation of synaptic input into microdomains (Figure 4C1), as were the average numbers of spikes fired by model neurons during the burst (Figure 4C2). Because the firing rates of model PY neurons in our model could influence the outgoing synaptic conductances through the presynaptic part of HSP rule, we also computed the mean HSP scaling factor separately for the set of PY-PY synapses arriving from deafferented model PY neurons and the set of PY-PY synapses arriving from intact model PY neurons. The value of the HSP scaling factor is directly proportional to the value of synaptic conductance after scaling, and thus could be taken as a measure of how much the outgoing synaptic conductance of the deafferented vs. intact model PY neurons changes as a function of the number of microdomains and the pattern of trauma. Figure 4C3 shows that the mean HSP scaling factor shows increasing trend as a function of for excitatory input from deafferented neurons, but decreases with larger for excitatory input from intact neurons. This effect is qualitatively the same for either diffuse or focal trauma scenarios (Figure 4C3, closed vs. open symbols).
The effect of microdomain partitioning in reducing the rate of paroxysmal bursts was further seen by visual inspection of network activity raster plots (Figures 4D1,2,3). This suggested that the underlying effect of stronger input segregation on paroxysmal burst rate could be similar to that of altered synaptic footprint size (Figure 3A1) – namely, that increased heterogeneity of synaptic input would lead to the decreased burst rate. Indeed, as Figure 4E1 shows, increasing the synaptic footprint size resulted in the downward offset of burst rate, consistent with the results shown in Figure 3A1. Both the mean and the standard deviation of the HSP scaling factor at PY-PY synapses were in general higher for larger synaptic footprint size, for all values of microdomains considered (Figures 4E2,3). The mean value of HSP scaling factor at PY-PY synapses was nearly independent of , while its standard deviation was generally higher for larger (Figures 4E2,3).
Segregation of synaptic input into microdomains allows a more independent scaling of the inputs from deafferented and intact neurons and thus enhances the correspondence between the firing rate of a given presynaptic neuron and the resulting homeostatic scaling of its downstream synaptic conductance. As a result, in segregated inputs scenario, synaptic conductance from intact neurons is weaker than synaptic conductance from deafferented neurons (Figure 4C3). On the other hand, the postsynaptic component of HSP scales all of synaptic conductances (from both deafferented and intact neurons) by the same amount. Thus, the role of intact neurons in burst generation is weaker (by virtue of their weaker synaptic conductance), but intra-burst firing becomes more intense (partially because of the stronger synaptic scaling in deafferented neurons). Thus, random segregation of synaptic input into microdomains acted to reduce the rate of paroxysmal discharges via a HSP-mediated increase in the variance of the synaptic scaling distribution.
Results reported in the previous section suggest that functional segregation of homeostatic synaptic scaling by astrocytes has the potential to alleviate the rate of paroxysmal burst discharges in post-traumatic network albeit the reduction in burst rate is relatively small when only a small number of microdomains is considered. Remarkably however, astrocytes in injured cortex also undergo significant rapid morphological remodeling [26]. Specifically, in the ferrous chloride model of trauma-induced epilepsy, astrocytes that were located relatively close (200 microns) to the boundary between intact and injured cortical regions lost their trademark star shape and elongated in the direction perpendicular to the trauma boundary [26]. Accordingly, the astrocytes at the boundary between intact and injured regions were termed “palisading astrocytes”, to distinguish from “hypertrophic astrocytes” that were located in intact part of the network and still retained their “star shaped” morphology to some extent (Figure 5A). Similar reorganization of astrocytes has also been observed in the kainate-induced epilepsy model [26], suggesting that it may represent a generic response of astrocytes to the trauma-induced alterations in neuronal activity. Because astrocytic morphology critically determines its ability to sense synaptic activity (and inactivity) such trauma-induced reorganization might have important implications for post-traumatic epileptogenesis.
To model the role of astrocytic morphological reorganization in trauma induced epileptic like activity, we used the microdomains scheme (as described in Materials and Methods and above) to model the presynaptic component of HSP and further assumed that synaptic input arriving to a specific model PY neuron was subdivided into two microdomains. One microdomain included the synapses exclusively from intact presynaptic neurons, and the other microdomain included synapses exclusively from deafferented presynaptic neurons (Figure 5B for schematic). This was a critical assumption and it derived from the observation that in the experimental model of trauma the “palisading astrocytes” were aligned perpendicular to the boundary separating the intact and traumatized parts of cortical tissue (Figure 5A). We further assumed that this orientation allows for a better segregation of synaptic input into distinct groups (inputs arriving from traumatized neurons vs. inputs arriving from relatively intact neurons). The postsynaptic component of HSP was modeled according to Equations 11,12.
In the focal trauma scenario, the rate of paroxysmal bursts was significantly lower for the scenario of “segregated inputs” (compared to the model with one microdomain), and this difference was observed for a wide range of trauma volumes considered (Figure 5C1). For comparison, in diffuse trauma model, the segregation of synaptic input did not have any significant effect on the burst rate trauma volume relation (Figure 5C2).
Firing rates of PY neurons varied depending on the neuronal location in the network (Figure 5D1, variation of firing rate along X location). The mean (Figure 5D2) and the standard deviation (Figure 5D3) of the PY-PY HSP scaling factor (computed individually for each model PY neuron) also showed strong dependence on neuronal location, with neurons on the intact-traumatized boundary having zero mean HSP and strong variability in synaptic conductances (Figures 5D2,3). The average over the population of model PY neurons of the mean HSP scaling factor was positively correlated with the trauma volume, and its value in the segregated inputs scenario was somewhat higher than that obtained for the model with one microdomain (Figure 5E1). The average (computed over the population of model PY neurons) standard deviation of the HSP scaling factor was also positively correlated with the trauma volume, and its value in the segregated inputs scenario was significantly higher than that obtained for the model with one microdomain (Figure 5E2).
As expected, the reduction in burst rate was paralleled by an increase in the average number of spikes fired per burst (Figure 5F1), as well as by an increase in intra-burst spiking rate (Figure 5F2) of model PY and IN neurons.
Based on the computational model of deafferentation-induced cortical trauma presented here, we predict that the functionally segregated organization of synaptic input to a neuron could help alleviate the rate of paroxysmal activity in early post-traumatic period (hours to days). The post-traumatic morphological reorganization of astrocytes could further mitigate paroxysmal activity [26]. Thus, potential therapeutic approaches to post-traumatic epilepsy should include the influence of astrocytes.
The increased heterogeneity of synaptic organization, reflected in the higher variability of HSP scaling factors at the boundary between intact and traumatized subnetworks (Figure 5E2), may underlie the reduced rate of paroxysmal bursting, which, however, comes at the expense of more intense spiking activity during the burst. In addition, these modeling results suggest that trauma-induced morphological reorganization of astrocytes on the intact-traumatized boundary could further increase the functional segregation of synaptic input to neurons, thus helping to alleviate paroxysmal activity.
In our previous studies we showed that the pattern of trauma can itself determine the threshold for post-traumatic paroxysmal activity [20]. These observations were also qualitatively reproduced in the present study of early post-traumatic reorganization driven by spatially local HSP rule. In particular, in focal trauma scenarios paroxysmal activity emerged for lower values of trauma volume parameter as compared to the diffuse trauma scenarios (Figure 1). In addition, in focal trauma scenario the rate of paroxysmal discharges did not depend on the trauma volume, but showed strong dependence on it in diffuse trauma cases (Figure 1).
A simple explanation of these simulation results would be as follows: In focal trauma and spatially local HSP rule, neurons at the boundary between intact and deafferented regions get ∼50% of their synaptic input from other intact neurons and another ∼50% from deafferented neurons. Because this breakdown does not depend on the trauma volume, and because paroxysmal bursts are generated at the boundary between intact and deafferented regions [20], the burst rate is not expected to depend on the trauma volume. Conversely, for diffuse trauma and local HSP, the breakdown of synaptic input from intact/deafferented neurons monotonically increases (in favor of deafferented neurons) with trauma volume, thus contributing to the increase in burst rate.
We showed earlier [20] that the propagation of paroxysmal events is undermined in diffuse trauma, where asynchronous activity of intact neurons helps to “dissipate” the correlated firing associated with the network burst. Together with the present results, this suggests that the relative role of trauma volume vs. functional segregation of synaptic input depends on the pattern of trauma, with focal trauma allowing for a more efficient “containment” of paroxysmal activity by functional input segregation. We suggest that functional segregation can be achieved by the morphological reorganization of astrocytes, similar to what was observed in some recent experiments [26].
It is widely recognized that astrocytes assume a critical role in different kinds of epilepsies [38]. Computational modeling suggested that glutamate signaling from astrocytes to neurons may drive spontaneous neuronal oscillations [35] and give rise to paroxysmal depolarization shifts as often seen in epilepsy [39]. Recent experimental study confirmed these earlier modeling results by directly demonstrating that a positive feedback loop between neurons and astrocytes can drive neurons to seizure threshold [40].
In our own models of post-traumatic epileptogenesis [3], [18], [20], paroxysmal activity is a consequence of homeostatic synaptic plasticity, which is mediated in part by TNFα that is released by astrocytes in response to neuronal inactivity [8]. All of the above mechanisms of astrocytic involvement in epileptogenesis are based on abnormal variations in neurotransmitter/cytokine signaling. By contrast, we showed here that structural reorganization of synaptic input regulation by astrocytes could constitute a contra-convulsive mechanism. It is tempting to speculate that such a seizure-suppressing program is switched on a short time after the traumatic event to compensate for pro-seizure influences of neuroinflammation; however, to fully address this issue, more refined clinical data (showing, for example, the relative timing of TNFα release vs. post-traumatic morphological reorganization of astrocytes) is needed.
The main clinically relevant prediction of our model is that a local homeostatic regulation of synaptic activity by astrocytes can lead to distinct network behavior (compared to the global homeostatic regulatory process). This prediction can be tested by targeting glial fibrillary acidic protein (GFAP), which is a common biological marker of trauma-induced morphological transformation of astrocytes and is dramatically increased during reactive gliosis. GFAP helps astrocytes to maintain mechanical strength and thus is instrumental in determining the cell shape and spatial distribution of finer astrocytic processes (microdomains) for synaptic regulation. Thus, our model would predict a higher incidence of seizures in trauma models of GFAP knockout animals. A higher incidence of seizures should occur following failure of astrocytic microdomains to reorganize after trauma. Consistent with this prediction, one experimental study reported that hippocampi of GFAP knockout mice exhibited higher sensitivity to kainate-induced seizures [41]. We predict that this GFAP deficiency affects seizure susceptibility via failed regulation of synaptic conductance by astrocytic microdomains.
The pathological action of homeostatic synaptic plasticity studied here is only one of the several known mechanisms of epileptogenesis. A common, long-recognized, cause of epileptic seizures is impaired clearance of extra-cellular potassium ions [42], [43], [44]. During intense bouts of neuronal activity extracellular potassium may rise to relatively high levels (10–12 mM) thus further depolarizing the neurons and contributing to the onset of seizure [44], [45]. The predominant mechanism of extracellular potassium clearance is through its reuptake and spatial buffering by astrocytic inward rectifying potassium channels [46]. In fluid percussion model of traumatic brain injury, glial contribution to extra-cellular potassium homeostasis is controversial. Some studies (e.g. [47], [48]) reported alteration in glial uptake of potassium; results of another study suggest that glial contribution to potassium uptake is not altered immediately after trauma [49]. Because the relative apposition of glial potassium channels and neurons is likely to depend on the spatial orientation and cell shape of astrocytes, it is plausible that the spatial pattern of extracellular potassium is further affected as a result of astrocytic reorganization. However, whether reactive astrocytes become more efficient in potassium clearance remain to be shown explicitly.
Trauma-induced morphological reorganization of astrocytes, as well as their release of cytokines and gliotransmitters, is part of “reactive gliosis”, a complex set of processes whereby astrocytes undergo various molecular and morphological changes [50]. Because of its role in the formation of glial scar that prevents axon regeneration, reactive gliosis has been associated with the detrimental effects of trauma-induced reorganization of neural circuitry. However, emerging evidence (reviewed in [50]) indicates that reactive astrocytes can have both beneficial and detrimental effects on post-traumatic reorganization.
Management of post-traumatic epilepsy and related convulsive disorders is often done with a variety of anti-epileptic drugs [51]. In particular, phenytoin and/or sodium valproate therapy can prevent early posttraumatic seizures [52], [53]. It was shown that sodium valproate inhibits production of TNFα through inhibition of NF-κB [54]. Extensive experimental data implicating the role of TNFα in seizure generation (reviewed in [16]) and results of our modeling studies thus provide an explanation with regard to the mechanistic action of valproate in early post-traumatic seizure suppression. Interestingly, valproate was found to be less efficient in suppression of late post-traumatic seizures [52]. This may imply that the trauma-induced astrocyte-mediated TNFα signaling possibly represents one out of several pathways of post-traumatic homeostatic regulation that tends to reorganize the network in response to chronic changes in electrical activity. Indeed, other mechanisms, such as trauma-induced changes in anatomical connectivity, have been implicated in post-traumatic epileptogenesis [2], [17], [19], [30], [55]. These mechanisms are likely to operate on much slower time scale than regulation by glial TNFα. The multiple temporal scales of post-traumatic HSP could thus explain the fact that anti-epileptic drugs are more efficient during early post-traumatic period (when they target specific pathway) as opposed to their relative inefficiency during late post-traumatic period (when other HSP pathways are activated).
Synaptic activity can be increased or decreased by molecules released from astrocytes and vice versa. In particular, accumulating evidence indicates that endocannabinoids (eCB) might be involved in compensatory mechanisms to offset the effects of trauma [56]. Endocannabinoids are released from neurons following intense neuronal activity. Recently it was shown that endocannabinoids released from hippocampal neurons can cause phospholipase C dependent calcium elevation in adjacent astrocytes through activation of astrocytic CB1 receptors [57]. The endocannabinoid-mediated activation of astrocytes stimulated the release of glutamate from astrocytes that further promoted the excitability of postsynaptic neurons. On the other hand, application of endogenous cannabinoid was shown to suppress TNFα formation through inhibition of nuclear factor kappa beta (NF-kB) after traumatic brain injury in a CB1-dependent manner [58]; signaling through this pathway is likely to decrease neuronal excitability. Thus, astrocyte-endocannabinoid interaction can contribute to post-traumatic reorganization in a complex manner. How this interaction contributes to the regulation of paroxysmal activity remains to be investigated [59].
Consistent with this emerging evidence, and based on our modeling results, we propose two opposing roles for reactive astrocytes in early post-traumatic epileptogenesis. Specifically, we propose that the release of TNFα promotes the generation of paroxysmal bursts, and that a post-traumatic morphological reorganization of astrocytes acts to suppress bursting activity at the expense of less frequent, but more intense, paroxysmal bursts. In this conceptual model the reorganization of astrocytes might be an adaptive step aimed at reducing the pro-convulsive effects of TNFα. Interestingly, in the ferrous chloride model of epilepsy, seizure suppressing drugs led to a significant reduction in the extent of seizure-related morphological transformation of astrocytes [26]. Thus, seizure itself appears to be a causative factor behind astrocytic reorganization. It is conceivable then that different components of reactive gliosis are coordinated in a way that aims to maintain minimal risk of seizures incurred by the activation of astrocytes. Biochemical dissection of these components and relations between them may help in understanding the causes of post-traumatic epileptogenesis.
The core computational model used in the present study was based on the modeling framework that we earlier developed to study the effect of different trauma patterns on the incidence of network bursts in post-traumatic cortex [20], [28]. Since in the present work we were specifically interested in the transformation of electrical activity during the early phase of post-traumatic response (when synaptic scaling might be constrained by a local glial response), the computational model of homeostatic synaptic plasticity in post-traumatic cortex needed to be adjusted to reflect the spatial localization of HSP. These changes, and the rationale behind them, are described in details below.
A cortical network was modeled as a 2D network on square lattice (80×80 neurons) in which each neuron could establish synapses with its peers with probability within its local footprint (a square 10×10 neuron region from and to which a neuron could receive or send synaptic connections). Model pyramidal neurons received, on average, 55 synapses from other model PY neurons and 12 synapses from model IN neurons. The distribution of synaptic inputs to model PY neurons is shown in Figure 6. The synaptic footprint of each neuron was centered on the neuron, and the size of the footprint was given by dimensions of the corresponding square region. Pyramidal neurons accounted for 80% of network population (5120 out of 6400 neurons), and inhibitory neurons constituted the remaining 20% (1280 out of 6400 neurons). Inhibitory neurons were distributed regularly on the lattice, such that every 5th neuron was inhibitory. Our goal was to understand the general responses of a cortical network to traumatic perturbation and therefore specific features pertaining to specific cortical layers were not included in the model.
The dynamics of neurons were modeled with the one compartmental Morris-Lecar model [60], as described in detail elsewhere [20], [28]. Briefly, the equations that governed the neuronal dynamics were:(1)(2)(3)(4)Phenomenological spike frequency adaptation current was added to model PY neurons to account for the experimentally observed spike frequency adaptation:(5)(6)Synaptic currents at PY-PY synapses had both AMPA and NMDA components, with both AMPA and NMDA conductance attenuated by synaptic depression as described below. Inhibitory synaptic currents did not incorporate synaptic depression.
Synaptic transmission was modeled as a deterministic process, during which synaptic conductance rose instantaneously following the spike and relaxed, with the characteristics time , to zero:(7)The value of maximal synaptic conductance,, depended on the pre- and postsynaptic neurons. Thus, (maximal synaptic conductance from pyramidal neuron to pyramidal neuron) was different from (maximal synaptic conductance from pyramidal neuron to inhibitory interneuron). Values of synaptic conductance are given in Table 1.
The temporal dynamics of the NMDA conductance was modeled as a difference of fast () and slow () exponentially decaying components:(8)(9)The parameter accounted for the efficacy of synaptic transmission. For GABAergic synapses, this parameter was held fixed (D = 1). For excitatory AMPA and NMDA synapses, it evolved according to the following equation, with representing the strength of synaptic short-term depression:(10)Values of parameters are given in Table 1. In the intact network with no deafferentation model pyramidal (PY) and inhibitory (IN) neurons fired with average rates of 5 and 10 Hz, respectively.
In addition to network current, each model neuron received an excitatory current from “the rest” of the cortex (afferent excitation). Synaptic conductance of this current evolved according to: , with . This synaptic conductance was stimulated randomly at times at the baseline Poisson rate of .
Trauma was modeled as cortical deafferentation, following which the frequency of external (afferent) excitation to a fraction of model neurons (both PY and IN) was reduced to 10% of its value in the intact model (from 100 Hz in intact model to 10 Hz in the deafferented state). The parameter represents the fraction of deafferented (injured) neurons and can be thought of as being proportional to the relative volume of deafferented neurons; thus, we refer to this as the “volume of trauma” parameter. As in our previous studies, we considered scenarios of focal and diffuse trauma. In focal trauma, the deafferented neurons were organized in a contiguous block (Figure 1A1, right). In diffuse trauma, the deafferented neurons were picked up at random from the entire model network (Figure 1A2, right).
Recent experimental evidence indicates that TNFα is a permissive, rather than instructive, mechanism that allows the synapses to express homeostatic synaptic plasticity [61]. Thus, our present implementation of HSP aimed to account for both the possible differential nature of up- vs. down- regulation of synaptic conductance and for the permissive (rather than instructive) nature of TNFα signaling. Specifically, we assumed the following mathematical form for the HSP at AMPA synapses of PY-PY pairs:(11) In Equation 11, is the convergence rate of the homeostatic update and is a preset “target rate” of pyramidal neuron firing (our HSP equation was only applied to model pyramidal neurons), typically set at. Upregulation of excitatory synaptic conductance on model PY neuron could occur if the average firing rate in the set of presynaptic partners (collateral pyramidal neurons that project synapses to a given neuron) was below the preset target rate. This regime corresponds to the release of TNFα from glial cells, which is believed to be determined by the levels of synaptic activity in the proximity of these cells (our implicit assumption is that synaptic activity scales with the firing rate of neurons that generate this synaptic activity). By contrast, downregulation of excitatory synaptic conductance on model PY neuron could occur if the firing rate of a neuron under consideration was above the preset target rate.
This modeling assumption stems from the fact that TNFα can only upregulate excitatory synaptic conductance; thus, down regulation of synaptic conductance likely depends on the firing rate of postsynaptic neuron [5]. It is also consistent with earlier findings regarding the role of neuronal activity in downregulation of synaptic conductances onto it [62]. The update scheme for GABA synapses on model pyramidal neurons is the opposite of that of the AMPA synapses on model pyramidal neurons (the average over the presynaptic set was still taken over the presynaptic set of PY model neurons)(12)Schematic presentation of homeostatic plasticity rules for up- and down regulation of synaptic conductances is given in Figure 7A.
The extent of per synapse homeostatic adjustment was computed as percentage of change in synaptic conductance relative to its value in the corresponding intact model. Taking PY-PY synapse as an example, where T is the time after the network had reached its new post-traumatic steady state.
In some simulations, we investigated the effect that functional segregation of synaptic input into microdomains might have on the post-traumatic reorganization of electrical activity in cortical network. For each model PY neuron, the entire set of synapses to it was randomly and equally partitioned into groups, each group on average consisting of synapses. Cortical neurons receive ∼8,000 synapses [63]. In the model network, each neuron received on average ∼80 synapses. Thus, each model synapse can be thought of as representing a group of ∼100 synchronously activated biological synapses, consistent with small (∼1%) amount of synchrony present at any time in cortical input [64]. The number of microdomains, , was varied from (corresponding to the case when all synaptic input is lumped into single astrocytic microdomain) to (corresponding to the case when each microdomain is associated with at most 1 model synapse). It was shown that in the cortex, each astrocytic domain can cover up to 80 microns of dendritic length [65]; taking linear spine density 1.0–1.5 spines/micron [66], it follows that each domain can be associated with up to ∼120 synapses. Thus, the case (1 model synapse per microdomain, 1 model synapse = ∼100 biological synapses) yields values consistent with the experimental findings (up to ∼120 biological synapses per microdomain). A schematic diagram for the case is shown in Figure 7B. The presynaptic component of HSP was computed for each one of these microdomain groups separately, as in Equations 11,12, with being the firing rate averaged over all model PY neurons that belonged to that same microdomain. Downregulating component of HSP was computed as in Equations 11,12, based on the firing rate of postsynaptic neuron.
Several models of astrocyte-synapse interaction [35], [36], [37] have linked increased levels of synaptic activity to astrocyte calcium elevation. Here, we focused on astrocytic responses to synaptic inactivity. Thus, existing models cannot explain post-traumatic activation of astrocytes. Rather than attempting to develop a detailed mathematical model to describe the response of astrocytes to synaptic inactivity, we assumed here that the ultimate effect of astrocytic microdomain activation was to scale synaptic conductances according to Equations 11,12. This assumption allowed us to avoid introducing additional complexity associated with biochemical cascades of astrocyte activation [37] and to focus on the long-term network effects of neuron-astrocyte interaction.
In a separate set of simulations we modeled correlated segregated inputs, whereby the entire set of synapses to each model PY neuron was partitioned into two groups (two microdomains, ) with synapses in one microdomain associated with the inputs from deafferented neurons, and synapses in another microdomain associated with the inputs from intact neurons.
Following deafferentation, homeostatic synaptic plasticity transformed asynchronous spiking to paroxysmal bursts. Thus, we adopted the rate of paroxysmal burst generation as a measure of network's propensity to exhibit the transition to seizures. Paroxysmal bursts of highly correlated population activity in our model resembled interictal epileptiform discharges (IEDs), which are often considered as an important diagnostic feature of epileptic seizures [22], [23], [24]. Thus, a higher rate of population bursting in a post-traumatic network model was considered as an indicator of stronger propensity to seizures.
Paroxysmal bursting events were detected as previously described in [20]. First, the entire simulation time period was partitioned into non-overlapping bins of 100 milliseconds each. Based on this binning, the spike count per bin for each model neuron was obtained. A paroxysmal bursting event was registered in time bin if at least of recorded model neurons (both pyramidal neurons and interneurons) fired action potentials during this time bin, with an average rate of firing greater than the threshold rate. This operational definition was constrained by the minimal fraction of active neurons, , that defined the paroxysmal burst, and by the minimal intra-burst firing rate, , of these active neurons. We set and .
The large size of our model network (6,400 neurons) made analysis of the entire network computationally intractable; thus, we sampled the activity of a subset of neurons (both pyramidal neurons and interneurons, from both deafferented and intact regions), using this sample to estimate the rate of paroxysmal bursts in the network. In the case of a diffusely traumatized network (spatially random deafferentation), the sampling region was a square block (usually 20×20 model neurons) co-centered with the center of the 2D network because spatially random deafferentation resulted in activity roughly symmetric with respect to the center of the lattice. In the case of a focally traumatized network (spatially structured deafferentation), the sampling region was composed of 5 parallel adjacent lines (resulting in a total of 400 model neurons) because the activity generally propagated from intact regions into deafferented regions.
In our earlier studies we showed that, in the deafferentation model of post-traumatic epileptic activity, the paroxysmal activity is generated by model pyramidal neurons with partially intact connectivity that are located at the boundary between intact and deafferented regions [20]. Consequently, activity generated at the boundary by intact neurons takes the shape of paroxysmal bursts (waves) that propagate into the deafferented part of a network. In the present study, we quantified some characteristics of network organization (e.g., mean and standard deviation of HSP scaling factors) and neural activity (e.g., cross-covariance between activities of different neurons) at the network boundary. The boundary was operationally defined as a set of intact neurons located within one synaptic footprint from the boundary deafferented neuron.
|
10.1371/journal.pcbi.1003452 | Global Patterns of Protein Domain Gain and Loss in Superkingdoms | Domains are modules within proteins that can fold and function independently and are evolutionarily conserved. Here we compared the usage and distribution of protein domain families in the free-living proteomes of Archaea, Bacteria and Eukarya and reconstructed species phylogenies while tracing the history of domain emergence and loss in proteomes. We show that both gains and losses of domains occurred frequently during proteome evolution. The rate of domain discovery increased approximately linearly in evolutionary time. Remarkably, gains generally outnumbered losses and the gain-to-loss ratios were much higher in akaryotes compared to eukaryotes. Functional annotations of domain families revealed that both Archaea and Bacteria gained and lost metabolic capabilities during the course of evolution while Eukarya acquired a number of diverse molecular functions including those involved in extracellular processes, immunological mechanisms, and cell regulation. Results also highlighted significant contemporary sharing of informational enzymes between Archaea and Eukarya and metabolic enzymes between Bacteria and Eukarya. Finally, the analysis provided useful insights into the evolution of species. The archaeal superkingdom appeared first in evolution by gradual loss of ancestral domains, bacterial lineages were the first to gain superkingdom-specific domains, and eukaryotes (likely) originated when an expanding proto-eukaryotic stem lineage gained organelles through endosymbiosis of already diversified bacterial lineages. The evolutionary dynamics of domain families in proteomes and the increasing number of domain gains is predicted to redefine the persistence strategies of organisms in superkingdoms, influence the make up of molecular functions, and enhance organismal complexity by the generation of new domain architectures. This dynamics highlights ongoing secondary evolutionary adaptations in akaryotic microbes, especially Archaea.
| Proteins are made up of well-packed structural units referred to as domains. Domain structure in proteins is responsible for protein function and is evolutionarily conserved. Here we report global patterns of protein domain gain and loss in the three superkingdoms of life. We reconstructed phylogenetic trees using domain fold families as phylogenetic characters and retraced the history of character changes along the many branches of the tree of life. Results revealed that both domain gains and losses were frequent events in the evolution of cells. However, domain gains generally overshadowed the number of losses. This trend was consistent in the three superkingdoms. However, the rate of domain discovery was highest in akaryotic microbes. Domain gains occurred throughout the evolutionary timeline albeit at a non-uniform rate. Our study sheds light into the evolutionary history of living organisms and highlights important ongoing mechanisms that are responsible for secondary evolutionary adaptations in the three superkingdoms of life.
| Proteins are biologically active molecules that perform a wide variety of functions in cells. They are involved in catalytic activities (e.g. enzymes), cell-to-cell signaling (hormones), immune response initiation against invading pathogens (antibodies), decoding genetic information (transcription and translation machinery), and many other vital cellular processes (receptors, transporters, transcription factors). Proteins carry out these functions with the help of well-packed structural units referred to as domains. Domains are modules within proteins that can fold and function independently and are evolutionarily conserved [1]–[4]. It is the domain make up of the cell that defines its molecular activities and leads to interesting evolutionary dynamics [5].
Different mechanisms have been described to explain the evolution of domain repertoires in cells [3]. These include the reuse of existing domains [2], [6], interplay between gains and losses [7]–[9], de novo domain generation [1], and horizontal gene transfer (HGT) [10]. Domains that appeared early in evolution are generally more abundant than recently emerged domains and can be reused in different combinations in proteins. This recruitment of ancient domains is an ongoing evolutionary process that leads to the generation of novel domain architectures (i.e. ordering of domains in proteins) by gene fusion, exon recombination and retrotransposition [2]–[4], [11]. For example, aminoacyl-tRNA synthetases are enzymes that charge tRNAs with ‘correct’ amino acids during translation [12], [13]. These crucial enzymes are multidomain proteins that encode a catalytic domain, an anticodon-binding domain, and in some cases, accessory domains involved in RNA binding and editing [13]. Evolutionary analysis suggests that these domains were recruited gradually over time [14]. In fact, recruitment of ancient domains to perform new functions is a recurrent phenomenon in metabolism [15].
In addition to the frequent reuse of domains, the dynamics between gains and losses also impacts the evolution of proteome repertoires [7]–[9]. Previous studies identified high rates of gene gains and losses in 12 closely related strains of Drosophila [7], Prochlorococcus (a genus of cyanobacteria) [16], and 60 isolates of Burkholderia (a genus of proteobacteria) [17]. A recent analysis of Pfam domains [18] revealed that ∼3% of the domain sequences were unique to primates and had emerged quite recently [19]. This implies that emergence of novel domains is an incessant evolutionary process [1]. In contrast, different selective pressures can lead to loss of domains in certain lineages and trigger major evolutionary transitions. For example, the increased rate of domain loss has been linked to reductive evolution of the proteomes of the archaeal superkingdom [20], adaptation to parasitism in cells [21] (e.g. transition from the free-living lifestyle to obligate parasitism in Rickettsia [22]), and ‘de-evolution’ of animals [23], [24] from their common ancestor. In these studies, gain and loss inferences were restricted to only particular groups of phyla or organisms. A global analysis involving proteomes from the three superkingdoms remained a challenge. Finally, changes to domain repertoires are also possible by HGT that is believed to occur with high frequency in microbial species, especially Bacteria [25], [26].
Here, we describe the evolutionary dynamics of protein domains grouped into fold families (FFs) and model the effects of domain gain and loss in the proteomes of 420 free-living organisms that have been fully sequenced and were carefully sampled from Archaea, Bacteria, and Eukarya (Dataset S1). The 420-proteome dataset was previously used by our group to reconstruct the evolutionary history of free-living organisms (see [27]) and was updated here to account for recent changes in protein classification and functional annotation. The dataset is very well annotated, especially regarding organism lifestyles that are otherwise problematic to assign, has already produced patterns of protein and proteome evolution that are very useful (including those described in [27]), and has produced timelines of FF evolution that are being actively mined. We conducted phylogenomic analyses using the abundance (total redundant number of each FF in every proteome) [28], [29] and occurrence (presence or absence) [30], [31] counts of FFs as phylogenetic characters to distinguish the 420 sampled taxa (i.e. proteomes). FF information was retrieved from the Structural Classification of Proteins (SCOP) database, which is considered a ‘gold standard’ for the classification of protein domains into different hierarchical levels [32]. Current SCOP definitions group protein domains with high pair-wise sequence identity (>30%) into a common FF, FFs that are evolutionarily related into fold superfamilies (FSFs), FSFs with similar secondary structure arrangement into folds (Fs), and Fs with common secondary structure elements into a handful of protein classes [33], [34]. A total of 110,800 SCOP domains (ver. 1.75) are classified into a finite set of only 1,195 Fs, 1,962 FSFs and 3,902 FFs. The lower number of distinct FSFs and FFs suggests that domain structure is far more conserved than molecular sequence (e.g. see [35]) and is reliable for phylogenetic studies involving the systematic comparison of proteomes [27]. Another advantage of using SCOP domains is the consideration of known structural and inferred evolutionary relationships in classifying domains into FFs and FSFs [36]. In comparison, evolutionary relationships for the majority of the Pfam domains are unknown. We further restricted the analysis to include only FF domains as they are conserved enough to explore both the very deep and derived branches of the tree of life (ToL) and are functionally orthologous [37]. In contrast, FSF domains represent a higher level in SCOP hierarchy and are more conserved than FFs but may or may not be functionally orthologous. Moreover, high conservation of FSF domains is useful for exploring the deep branches of the ToL but may not be very informative for the more derived relationships.
The analysis of retracing the history of changes in the occurrence and abundance of FF domains on each branch of the reconstructed ToLs revealed that FFs were subject to high rates of gains and losses. Domain gains generally outnumbered losses but both occurred with high frequencies throughout the evolutionary timeline and in all superkingdoms. Remarkably, the gains-to-loss ratios increased with evolutionary time and were relatively higher in the late evolutionary periods. Finally, functional annotations of FFs illustrated significant differences between superkingdoms and described modern tendencies in proteomes.
The 420-proteome dataset used in this study included proteomes from 48 Archaea, 239 Bacteria, and 133 Eukarya. The dataset did not include any parasitic organisms as they harbor reduced proteomes and bias the global phylogenomic analyses (e.g. [38]). FFs were assigned to proteomes using SUPERFAMILY ver. 1.73 [39] hidden Markov models [40], [41] at an E-value cutoff of 10−4 [42]. A total of 2,397 significant FF domains were detected in the sampled proteomes. The definitions of eight FFs in the 420-proteome dataset were updated in SCOP ver. 1.75 and were therefore renamed in our dataset. FFs were referenced using SCOP concise classification strings (css) (e.g. ‘Ferredoxin reductase FAD-binding domain-like’ FF is b.43.4.2, where b represents the class [all-beta proteins], 43 the fold, 4 the FSF and 2 the FF).
We considered the genomic abundance [28], [29] and occurrence [30], [31] of 2,397 FFs as phylogenetic characters to reconstruct phylogenies describing the evolution of 420 free-living organisms (i.e. taxa) using maximum parsimony. The raw abundance values of each FF in every proteome (gab) were log-transformed and divided by the logarithm of maximum value in the matrix (gmax) to account for unequal proteome sizes and variances (see formula below) [29], [43].The transformed abundance values were then rescaled from 0 to 23 (scaling constant) in an alphanumeric format (0–9 and A-N) to allow compatibility with the phylogenetic reconstruction software. The transformed abundance matrix with 24 possible character states was imported into PAUP* 4.0b10 [44] for the reconstruction of abundance trees. For occurrence trees, we simply used 0 and 1 (indicating absence and presence) as the valid character state symbols. We polarized both abundance and occurrence trees using the ANCSTATES command in PAUP* and designated character state 0 as the ancestral state, since the most ancient proteome is closer to a simple progenote organism that harbors only a handful of domains [20], [38]. The stem lineage of this organism gradually increased its domain repertoire, supporting the polarization from 0 to N and Weston's generality criterion, in which the taxic distribution of a set of character states is a subset of the distribution of another [45], [46]. Phylogenetic trees are adequately interpreted when rooted. This provides direction to the flow of evolutionary information and is useful to study species adaptations. In this study, we choose to root trees using the Lundberg method [47]. This scheme first determines the most parsimonious unrooted tree, which is then attached to a hypothetical ancestor. The hypothetical ancestor may be attached to any of the branches in the tree. However, only the branch that gives the minimum increase in overall tree length is selected [48]. This branch, which exhibits the largest numbers of ancestral (plesiomorphic) character states was specified using the ANCSTATES command in PAUP*. Thus, Lundberg rooting automatically roots the trees by preserving the principle of maximum parsimony. This method is simple and free from artificial biases introduced by alternative rooting methods (e.g. the outgroup method). While selection of an appropriate outgroup to root the ToL is virtually impossible, Lundberg rooting provides a parsimonious estimate of the overall phylogeny and should be considered robust as long as the assumptions used to root the trees are not proven false. To evaluate support for the deep branches of ToLs, we ran bootstrap (BS) analysis with 1,000 replicates. Character state changes were recorded by specifying the ‘chglist’ option in PAUP*. Trees were visualized using Dendroscope ver. 3.0.14b [49].
To determine congruence between abundance and occurrence trees, we used the nodal module implemented in the TOPD/FMTS package ver. 3.3 [50]. The module takes as input a set of trees in Newick format and calculates a root mean squared deviation (RMSD) value for each pairwise comparison. The RMSD value is 0 for identical trees and increases with incongruence. To evaluate the significance of calculated RMSD values, we implemented the ‘Guided randomization test’ with 100 replications to determine whether the calculated RMSD value was smaller than the chance expectation. The randomization test randomly changes the positions of taxa in trees, while maintaining original tree topology, and calculates an RMSD value for each random comparison [50]. The result is a random distribution of RMSD values with a mean and standard deviation. The calculated RMSD value was compared with the mean of the random distribution to determine whether the observed differences were better than what would be expected merely by chance.
The spread of each FF was given by its distribution index (f-value), defined by the total number of proteomes encoding a particular FF divided by the total number of proteomes. The f-value ranges from 0 (absence from all proteomes) to 1 (complete presence).
To determine the relative age of FF domains in our dataset, we reconstructed trees of domains (ToDs) from the abundance and occurrence matrices used in the reconstruction of ToLs. The matrices were transposed, treating FFs as taxa and proteomes as characters. The reconstructed ToDs described the evolution of domains grouped into FFs and identified the most ancient and derived FFs (refer to [27] for an elaborate description and discussion on ToDs). To root the trees, we declared character state ‘N’ as the most ancestral state. This axiom of polarization considers that history of change for the most part obeys the ‘principle of spatiotemporal continuity’ (sensu Leibnitz) that supports the existence of Darwinian evolution. Specifically, it considers that abundance and diversity of individual FFs increases progressively in nature by gene duplication (and associated processes of subfunctionalization and neofunctionalization) and de novo gene creation, even in the presence of loss, lateral transfer or evolutionary constraints in individual lineages. Consequently, ancient domains have more time to accumulate and increase their abundance in proteomes. In comparison, domains originating recently are less popular and are specific to fewer lineages. We note that the N to 0 polarization is supported by the observation that FFs that appear at the base of the ToDs are structures that are widespread in metabolism and are considered to be of very ancient origin (e.g. [27]). The age of each FF was drawn directly from the ToDs using a PERL script that calculates the distance of each node from the root. This node distance (nd) is given on a relative scale and portrays the origin of FFs from 0 (most ancient) to 1 (most recent). The geological ages of FFs were derived from a molecular clock of protein folds [51], [52] that was used to calibrate important events in proteome evolution. We have previously shown that nd correlates with geological time, following a molecular clock that can be used as a reliable approximation to date the appearance of protein domains [51], [52].
We used the SUPERFAMILY functional annotation scheme (based on SCOP 1.73) to study the functional roles of FF domains in our dataset [53]–[55]. The SUPERFAMILY annotation assigns a single molecular function to FSF domains (and by extension to its descendant FFs). The annotation scheme gives a simplified view of the functional repertoire of proteomes using seven major functional categories including, i) metabolism, ii) information, iii) intracellular processes, iv) extracellular processes, v) general, vi) regulation and vii) other (includes domains with either unknown or viral functions). We assumed that FFs grouped into an FSF performed the same function that was assigned to their parent FSF. While this simplistic representation does not demonstrate the complete functional capabilities of a cell, it is sufficient to illustrate the major functional preferences in proteomes (refer to [21] for further description and use of the functional annotation scheme in large-scale proteomic studies).
We conducted a GO enrichment analysis [56], [57] on FF domains to identify biological processes [58], [59] that were significantly enriched. For this purpose, the list of FF domains was given as input to domain-centric Gene Ontology (dcGO; http://supfam.org/SUPERFAMILY/dcGO) resource and the most specific and significant associations to GO terms corresponding to different biological processes were retrieved. The statistical significance was evaluated by P-value computed under the hypergeometric distribution [56], while the false discovery rate (FDR) was set to default at <0.01 [60].
We first describe the patterns of FF use and reuse in superkingdoms and then build on this knowledge to infer the meanings of domain gain and loss in proteomes.
A Venn diagram describes the sharing patterns of 2,397 FFs in seven Venn distribution groups (Figure 1A). For simplicity, we name these sets ‘taxonomic groups’ with the understanding that their taxonomic status is endowed by patterns of distribution of FFs in superkingdoms. The number of FFs decreased in the order Eukarya (total FFs = 1,696), Bacteria (1,510) and Archaea (703). Eukarya also had the highest number of superkingdom specific FFs (758), followed by Bacteria (522), and Archaea (89). ABE FFs were universal (i.e. present in all three superkingdoms) and made the third largest group with 484 FFs, while BE was the fourth largest taxonomic group with 414 FFs (Figure 1A). The lowest number of FFs was in AE with only 40 FFs that were unique to both Archaea and Eukarya. The number of Archaea-specific FFs was also low (89) but comparable to the number of akaryotic FFs (i.e. AB = 90). We observed that Archaea was mostly about sharing (or not innovating new FFs). This was evident by the fact that only 13% of the total archaeal FFs were Archaea-specific. This was in striking contrast with Bacteria and Eukarya where superkingdom-specific FFs made large proportions of the FF repertoires with 35% and 45% FFs, respectively (Figure 1A).
We plotted the distribution of domain ages (nd) for FFs in each taxonomic group to determine the order of their evolutionary appearance (Figure 1B) (see Methods). The first FF to appear in evolution was the ‘ABC transporter ATPase domain-like’ (c.37.1.12) FF at nd = 0 in the ABE taxonomic group (Figure 1B). ABC transporters are multifunctional proteins that are primarily involved in the transport of various substrates across membranes [61], [62]. These domains are ubiquitous and highly abundant in extant species and considered to be very ancient. In our timeline, c.37.1.12 appeared first, supporting its widespread presence and significance in cells. ABE was the most ancient taxonomic group spanning the entire time axis with a median nd of 0.24 (Figure 1B). This suggested that the majority of the FFs that were common across all superkingdoms appeared very early in evolution. ABE was followed by the appearances of BE (at nd = 0.15), AB (0.26), B (0.26), E (0.551), A (0.555), and AE (0.57) taxonomic groups, in that order (Figure 1B).
The first complete loss event for any FF in the primordial world likely triggered the appearance of the BE taxonomic group. Our data indicates that this occurred at nd = 0.15 (roughly >3.2 billion [Gyrs] years ago) with the complete loss of the ‘Heat shock protein 90, HSP90, N-terminal domain’ (d.122.1.1) FF in Archaea (Figure 1B). Heat-shock proteins are molecular chaperones that assist in protein folding and clearing of cell debris [63]. These are highly conserved in bacterial and eukaryal species, but relatively less abundant in Archaea. In fact, homologs of Hsp90 or Hsp100 are completely absent in archaeal species [63]. This knowledge is compatible with our finding of loss of d.122.1.1 FF in Archaea that occurred very early in evolution. We propose that this event exemplifies reductive evolutionary processes that were at play early in evolution in nascent archaeal lineages as emergent diversified cells were unfolding different mechanisms of protein folding. In light of our results, Archaea was the first superkingdom to follow reductive trends. The first superkingdom-specific FF appeared in B at nd = 0.26 (∼2.8 Gyrs ago), while both Archaea and Eukarya acquired unique FF domains concurrently at around nd = 0.55 (∼1.6 Gyrs ago) (Figure 1B). Emergence of taxonomic groups in evolution described three important evolutionary epochs: (i) early (0≤nd<0.15), a period before the start of reductive evolution in the archaeal superkingdom, (ii) intermediate (0.15≤nd<0.55), a period marked by early domain discovery in Bacteria, and (iii) late (0.55≤nd≤1), a period during which simultaneous diversification of Archaea and Eukarya occurred (Figure 1B).
To determine the popularity of FFs across organisms, we computed an f-value representing the fraction of proteomes encoding an FF. The median f-value decreased in the order, ABE>AE>E>BE>AB>A>B (Figure 1C). We observed that universal FFs of the ABE taxonomic group were most popular and shared by the majority of the proteomes (median f = 0.58). The FFs in AE and E were also distributed with higher f-values (median f = 0.54 and 0.27). In contrast, most of the bacterial taxonomic groups (e.g. BE, AB and B) had lower median f-values (0.22, 0.10, and 0.02, respectively). The Venn diagram indicated that ∼22% of the total FFs were bacteria-specific (Figure 1A) but the median f-value of those FFs was extremely low (0.02) (Figure 1C). This implied that FFs unique to Bacteria were very unevenly distributed among bacterial species. This also suggested that the rate of FF discovery in Bacteria was very high but their spread was quite limited.
A recent study proposed concepts of economy (i.e. organism budget in terms of number of unique genes and domain structures), flexibility (potential of an organism to adapt to environmental change) and robustness (ability to resist damage and change) to help explain the persistence strategies utilized by organisms in the three superkingdoms [64]. To determine how persistence strategies distributed in our dataset, we redefined economy (i.e. total number of unique FFs in a proteome), flexibility (total number of redundant FFs in a proteome) and robustness (ratio of flexibility to economy). When plotted together on a 3D plot, interesting patterns were revealed (Figure 1D). As expected, the proteomes of the akaryotic microbes in Archaea and Bacteria were most economical but least flexible and robust (Figure 1D). Within these superkingdoms, archaeal proteomes (red circles) exhibited greatest economy but lowest flexibility and robustness. In contrast, Bacteria exhibited intermediate levels of economy, flexibility and robustness. Finally, eukaryal proteomes were least economical but highly flexible and robust (Figure 1D). Table 1 lists the lower and upper bounds for economy, flexibility, and robustness for the three superkingdoms. The median values for the three parameters always increased in the order, Archaea, Bacteria, and Eukarya (Table 1). The analysis revealed that the survival strategy of microbial species lies in encoding smaller domain repertoires while the eukaryal species trade-off economy with more flexibility and robustness and harbor richer proteomes [64]. The number of both unique (economy) and redundant FFs (flexibility and robustness) was considerably higher in eukaryotes.
We compared the distributions of molecular functions in taxonomic groups (Figure 2A) and dated their appearance in evolutionary time (nd) (Figure 2B–H). Metabolism was the most abundant and widely distributed molecular function in organisms, especially in the ABE, BE, and AB taxonomic groups. However, significant deviations were observed in the AE and A taxonomic groups, where informational FFs (e.g. those belonging to the replication machinery) outnumbered FFs in other functional categories (Figure 2A). These results are consistent with previous knowledge regarding high sharing of informational proteins between Archaea and Eukarya and a common metabolic apparatus between Bacteria and Eukarya. This observation has often led to proposals relating the origin of eukaryotes to a confluence between akaryotic cells (reviewed in [65]; see also [66]–[69]). However, our data show that the presence of bacterial metabolic enzymes in Eukarya is better explained by primordial endosymbiotic events leading to mitochondria and plastids in a proto-eukaryote stem cell-line (read below). In comparison, sharing of informational enzymes between Archaea and Eukarya occurred relatively late in evolution and could actually reflect late domain losses in Bacteria. Intracellular processes and general were distributed similarly while regulation and extracellular processes appeared to be preferential only in Eukarya (Figure 2A). The distribution of molecular functions in taxonomic groups was largely in agreement with the distribution previously explained for individual species [21].
We explored the order of evolutionary appearance of molecular functions by generating nd vs. f plots for the seven taxonomic groups (Figure 2B–H). The ABE FFs were present with largest f-values and as expected spanned the entire nd-axis (Figure 2B). In fact, 13 FFs had an f-value of 1.0 indicating universal presence in organisms, while 62 near-universal FFs were present in >95% of the proteomes. ABE FFs were generally enriched in metabolic functions (Figure 2B). This suggested that the last common ancestor of diversified life was structurally and metabolically versatile (e.g. [38]). However, the f-value distribution of ABE FFs followed a bimodal pattern with a significant drop in f during the intermediate evolutionary epoch. Most of the FFs of intermediate age were classified as metabolic (grey circles), informational (red circles), or with intracellular roles (light blue circles) (Figure 2A, 2B). BE followed a distribution similar to ABE but the first FF appeared during the intermediate evolutionary epoch at nd = 0.15 (Figure 2C). This also marked the first loss of an FF in Archaea (boxplot for BE in Figure 1B). This observation implies that Archaea was the first superkingdom to escape from the ancestral community and evolved by streamlining genomes. Perhaps, genome reduction was better suited for harsher environments. Other selective pressures that may have triggered early domain loss in Archaea could include escape from RNA viruses (because RNA is unstable at extreme temperatures) and phagotrophs [70]. The majority of the BE FFs served metabolic, informational and intracellular roles (Figure 2A, 2C), just like ABE. The akaryotic-specific (AB) FFs appeared during the intermediate and late evolutionary epochs and were largely dominated by metabolic and other FFs (Figure 2A, 2D). Most of these FFs had very low f-values (Figure 2D) indicating that this taxonomic group exhibited low popularity levels. In contrast, all of the 40 AE FFs appeared in the late epoch and were dominated by domains involved in informational (red) (Table 2) and regulatory processes (green) (Figure 2A, 2E). This validated the hypothesis that informational enzymes in eukaryotes are very similar to their archaeal counterparts rather than bacterial enzymes [71]–[73]. This argument has been used to propose a sister relationship between Archaea and Eukarya and an ancient origin of Bacteria. However, our analysis revealed that sharing of informational domains between archaeal and eukaryal species was only a recent event (i.e. was evident in the late evolutionary epoch; nd≥0.55) and that the sister relationship between Archaea and Eukarya inferred from the 16S rRNA trees [74] was influenced by the high rates of modern sharing between Archaea and Eukarya (see Discussion) [75]. AE FFs were generally distributed with higher f-values (Figure 2E).
FFs unique to Archaea (A) appeared in the late epoch at nd≥0.55 and were generally distributed with lower f-values (Figure 2F). The discoveries of these FFs were biased towards informational and other domains (Figure 2A, 2F). A large number of bacteria-specific FFs (B) also appeared during the intermediate and late evolutionary epochs (Figure 2G). We note that, in general, bacterial FFs appearing in the intermediate epoch were biased towards informational roles while those that appeared later served metabolic and general roles (Figure 2A, 2G). Lastly, all of the Eukarya-specific (E) FFs appeared in the late epoch (Figure 2H), just like Archaea (Figure 2F). Eukarya discovered a large number of recent FF domains (nd≥0.55) that were involved in regulation (green circles) and extracellular processes (blue circles) and were distributed with relatively high f-values in the eukaryal proteomes (Figure 2A, 2H).
Superkingdom-specific FFs appeared in both Archaea and Eukarya at around the same time, and both showed a tendency to become widespread in species (Figure 2F, 2H). In contrast, the discovery of Bacteria-specific (B) FFs started much earlier but with limited spread (Figure 2G). This suggested that while Archaea was the first superkingdom to follow reductive trends, it was Bacteria that diversified first and was capable of unfolding superkingdom-specific domain structures. The primordial stem-line (that was structurally and functionally complex) later evolved into eukaryotes, possibly after engulfment of already diversified microbes (Discussion). In this regard, we identified a set of mitochondrial FFs, all of which appeared at nd≥0.55, during and after the rise of the E taxonomic group, including the ‘Mitochondrial resolvase ydc2 catalytic domain’ (c.55.3.7; nd = 0.55) and the ‘Mitochondrial cytochrome c oxidase subunit VIIb’ (f.23.5.1; nd = 0.59) FFs (Table 3). Thus, our timelines do not support fusion hypotheses for the origin of eukaryotes linked to a confluence between akaryotes. The fusion scenarios have been discussed elsewhere [65], [70], [76]–[79] and it is beyond the scope of this study to evaluate what model is better. In light of our data that is based on the genomic census of conserved FF domains in hundreds of free-living organisms, we support a phagotrophic and eukaryote-like nature of the host (anticipated in [78], [79]) that acquired the primordial alpha-proteobacterium as an endosymbiont, which later became mitochondria and triggered the diversification of eukaryotes (at nd = 0.55; roughly ∼1.6 billion years ago). A formal test of this hypothesis is warranted and will be explored in a future study. The exercise also revealed that the lower median f-values observed earlier (Figure 1C) were due to the significant drop in f in the intermediate evolutionary epoch. We note that the majority of the bacterial FFs (belonging to the ABE, BE, B and AB taxonomic groups) also appeared during this period and thus affected the overall medians.
We generated rooted ToLs from abundance (Figure 3A) and occurrence (Figure 3B) counts of 2,397 FF domains in the 420 free-living proteomes (see Dataset S1 for taxon names) using maximum parsimony as the optimality criterion in PAUP* 4.0b10 [44]. Both reconstructions recovered a previously established tripartite world of cellular organisms [20], [27], [74], [80]. The archaeal superkingdom always formed a paraphyletic group at the base of the ToLs. The deep branches of the ToLs were occupied by thermophilic and hyperthermophilic archaeal species (Thermofilum pendens and Cand. Korarchaeum) (Figure 3). The archaeal rooting of the ToL is supported by a number of previous studies (e.g. [14], [20], [27], [81]–[83]) and is in conflict with the traditional Archaea-Eukarya sister relationship (Discussion). Bacteria and Eukarya formed strong monophyletic clades that were supported by high BS values (≥99%) and were separated from Archaea with 53% (Figure 3A) and 78% (Figure 3B) BS support. Both ToLs had strong phylogenetic signal (g1 = −0.33 and −0.28). Overall, phylogenomic patterns resembled traditional groupings and supported previous analyses of similar kind [20], [27]. Moreover, the dissimilarity between two reconstructions was 5.37, which was smaller than the mean RMSD calculated from 100 random comparisons (Figure 3) (Methods). Because the ToLs were supported with high confidence and resembled previous analyses [20], [27], they made useful tools for the study of domain gain and loss events on the many branches (read below).
To quantify the relative contributions of domain gains and losses impacting the evolution of superkingdoms, we retraced the history of character state changes (i.e. changes in the abundance or occurrence of FFs) on each branch of the reconstructed ToLs. For each FF domain, we counted the number of times it was gained and lost in different branches of the phylogenetic tree. Gains were recorded when the abundance/occurrence of a particular FF at a node was higher than the corresponding value at the immediate ancestral node. In contrast, losses were incremented when the abundance/occurrence of a particular FF at a node was lower. Because we allowed character changes in both forward and backward directions (Wagner parsimony), each FF character could be both gained and lost a number of times across the many branches of the ToL. This assumption is reasonable as different lineages of organisms utilize domain repertoires differently. Because abundance counts are expected to be higher in the eukaryotic species (especially in metazoa) due to increased gene duplication events and a persistence strategy that favors flexibility and robustness (Figure 1D) [64], we also considered gains and loss statistics from the occurrence trees.
To evaluate the performance of both models, we first compared the number of FFs that were gained (i.e. net sum above zero) and lost (net sum below zero) in both reconstructions. Out of the total 2,397 (2,262 parsimony informative) FF domains in the abundance model, 1,955 (86%) were gained, while only 236 (10%) were lost (Dataset S2). In contrast, occurrence identified 60.1% FFs as gained (1,353/2,249) and 30.5% (686/2,249) as lost (Dataset S3). Nearly 96% (1300/1,353) of the occurrence gains were also gained in abundance while only 26% (178/686) losses were common to both models. This suggested that abundance included nearly all the occurrence gains and likely overestimated the number of gains (due to gene duplications and domain reuse). In contrast, occurrence led to more balanced distributions and likely overestimated losses (read below).
To provide additional support to the gain/loss model, we pruned taxa from the original ToLs leaving only one superkingdom and recalculated character state changes on the pruned trees. This eliminated any biases resulting from the differences in the persistence strategies of the three superkingdoms and yielded four phylogenetic trees, Total (taxa = 420, total FF characters = 2,397), Archaea (48, 703), Bacteria (239, 1,510) and Eukarya (133, 1,696). For each of the four trees, we calculated the sum of gain and loss events for all parsimony informative FF characters and represented the values in boxplots (Figure 4A). In all distributions, medians were above 0 indicating that the sum of net gains and losses was a non-negative number for both abundance (Figure 4A:abundance) and occurrence (Figure 4A:occurrence) models. The exception was the eukaryal tree pruned from the occurrence model, for which the median was exactly zero. The result revealed that while both gains and losses occurred quite frequently, the former was more prevalent in proteome evolution.
The histograms in Figure 4B describe the distributions of gain and loss counts for all parsimony informative FF characters in the Total dataset. When plotted against evolutionary time (nd), results highlighted remarkable patterns in the evolution of domain repertoires. Domain gains outnumbered losses in both abundance (80,904 gains vs. 47,848 losses) and occurrence (17,319 vs. 13,280) tree reconstructions (Figure 4B). The gain-to-loss ratios were 1.69 and 1.30, respectively, indicating an increase of 69% and 30% in gains relative to losses. Relative differences in the numbers of gains (red) versus losses (blue) suggested that gains increased with the progression of evolutionary time in both reconstructions (read below).
We note that different evolutionary processes may be responsible for shaping the proteomes in individual superkingdoms. For example, the origin of Archaea has been linked to genome reduction events [20], [84], while HGT is believed to have played an important role in the evolution of bacterial species [25]. In contrast, eukaryal proteomes harbor an increased number of novel domain architectures that are a result of gene duplication and rearrangement events [6], [43]. Therefore, to eliminate any biases resulting from the effects of superkingdoms in the global analysis (Figure 4B), we recalculated the history of character changes on the pruned superkingdom tress recovered earlier (Figure 4C). For abundance reconstructions, the exercise supported earlier results where the number of gains was significantly higher than the corresponding number of losses for Archaea (4,616 vs. 2,009), Bacteria (36,606 vs. 20,196), and Eukarya (40,515 vs. 25,036) (Figure 4C: abundance). The overall gain to loss ratios decreased from 2.30 in Archaea to 1.81 in Bacteria and 1.62 in Eukarya (Figure 4C: abundance). The increased gain-to-loss ratios in akaryotic microbial species are remarkable; it implies that the rate of gene discovery in akaryotic microbes (by de novo creation, gene duplication, acquisition by HGT and/or recruitment) is higher than the rate in eukaryotes. This tendency in microbial species could be a novel ‘collective’ persistence strategy to compensate for their economical proteomes. For histograms representing occurrence models, global gain-to-loss ratios decreased in the order, Archaea>Bacteria>Eukarya (Figure 4C: occurrence). Remarkably, the ratio in Eukarya dropped below 1 indicating prevalence of domain loss events relative to gains. This result supports recent studies that have proposed the evolution of newly emerging eukaryal phyla via genome reduction [85].
When partitioned into the early, intermediate, and late evolutionary epochs, the gain-to-loss ratios exhibited an approximately linear trend towards increasing gains (Figure 5). For abundance, the ratios increased from 1.32 in the early epoch to 1.45 in the intermediate and 1.96 in the late evolutionary epochs. Similar trends were also observed for occurrence, with calculated ratios of 0.61, 0.97, and 1.68, respectively (Figure 5A). In fact, both gains and losses increased linearly with evolutionary time in all reconstructions. However, accumulation of gains overshadowed the number of losses (Figure 5). Remarkably, the occurrence model suggested predominant losses in the first two phases of evolution (0.61 and 0.97) that were compensated by significantly higher amounts of gains (1.68) in the late epoch. In contrast, abundance failed to illustrate this effect and indicated overwhelming gains in all evolutionary epochs.
When looking at the individual epochs for pruned trees (Figure 5B), we noticed that the rate of domain gain increased with time (as before) (Figure 5A). However, the ratios in the initial two evolutionary epochs were considerably higher in Archaea for both the abundance and occurrence models. For example, Archaea exhibited gain-to-loss ratios of 2.06 and 2.14, in comparison to 1.26 and 1.39 in Bacteria, and 1.55 and 1.67 in Eukarya for early and intermediate evolutionary epochs (Figure 5B:abundance). In contrast, Bacteria exhibited an overwhelming gain-to-loss ratio of 2.88 in comparison to 2.67 in Archaea and 1.61 in Eukarya, in the late evolutionary epoch. Overall, the gain-to-loss ratios increased with evolutionary time in all superkingdoms with the sole exception of Eukarya that had a lower ratio in the late (1.61) compared to the intermediate (1.67) epoch (Figure 5B:abundance).
Results based on occurrence indicated similar trends but with relatively more balanced gain-to-loss ratios and still highlighted the abundance of domain gains in evolution. The individual ratios were 1.42, 1.66, and 2.44 in Archaea, 0.60, 0.91, and 2.61 in Bacteria, and 0.51, 0.95, and 0.95 in Eukarya (Figure 5B:occurrence). Both Bacteria and Eukarya showed increased levels of ancient domain loss. However, Bacteria compensated this decrease by engaging in massive gain events during the late evolutionary epoch (ratio of 2.61). In contrast, Eukarya exhibited an even exchange between FF gain and loss events (ratio = 0.95) in both the intermediate and late epochs. Occurrence results also supported the evolution of Eukarya by gene loss, which is in line with recently published analyses [23], [85]. Abundance also indicated this drop in gene discovery rate for recent domains in Eukarya. However, the drop appears to be compensated by increased duplications of other domains that lead to an increase in the overall number of domains that are gained (Figure 5B: abundance). This apparent discrepancy can be explained by the power of both models in depicting true evolutionary relationships between organisms. Abundance accounts for a number of evolutionary processes such as HGT, gene duplication, and gene rearrangements while occurrence merely describes presence and absence of FFs and because of its more ‘global’ nature fails to illustrate a complete evolutionary picture (Discussion).
To test whether unequal sampling of proteomes per superkingdom was contributing any bias to the calculations of domain gains and losses, we extracted 100 random samples of 34 proteomes each from the three superkingdoms and generated 100 random trees. From each of the random trees, we recalculated the gain-to-loss ratios using both abundance and occurrence models (Figure 6). Random and equal sampling supported the overall conclusion that gains were overwhelming during the evolution of domain repertoires (Figure 6). The median ratios for random trees were 2.47 in Archaea, 2.35 in Eukarya, and 2.34 in Bacteria for abundance reconstructions (Figure 6A). In comparison, the ratios decreased from 2.11 in Archaea to 1.93 in Bacteria and 1.11 in Eukarya for occurrence reconstructions (Figure 6B). Based on the results of random and equal sampling, we safely conclude that the gain of domains in proteomes is a universal process that occurs in all three superkingdoms of life. Moreover, the gain-to-loss ratios increase with time (Figure 5) and their effects are directly responsible for evolutionary adaptations in superkingdoms (Discussion). We also propose that using abundance increases the reliability of the phylogenomic model and accounts for many important evolutionary events, a feat that is not possible when studying occurrence.
We identified FFs that were gained (i.e. net sum of gains and losses was above 0) and lost (net sum below 0) directly from the pruned superkingdom trees. To eliminate any redundancy, we only kept FFs that were gained (or lost) in both abundance and occurrence reconstructions and excluded those where both methods disagreed. Using this stringent criterion, we classified a total of 368 archaeal FFs as being gained and 40 as being lost. In comparison, Bacteria and Eukarya gained 892 and 633 FFs, respectively, while they lost only 148 and 164 FFs. Both gained and lost FFs for each superkingdom were provided as input to the online dcGO resource [56], [57] to retrieve the highly specific and significantly enriched biological process GO terms (Methods). For FFs that were gained, a total of six GO terms were significantly enriched in archaeal proteomes representing biological processes involved in the biosynthesis of nucleotides and metabolism, such as ‘tricarboxylic acid cycle [GO:0006099]’, ‘pyruvate metabolic process [GO:0006090]’, ‘acyl-CoA metabolic process [GO:0006637]’, ‘thioester biosynthetic process [GO:0035384]’, ‘purine nucleobase metabolic process [GO:0006144]’, and ‘pyrimidine nucleoside metabolic process [GO:0006213]’ (Table 4). In comparison, only one biological process in Bacteria (‘polysaccharide catabolic process [GO:0000272]’) and 37 in Eukarya were significantly enriched (Table 4). While, the bacterial GO term corresponded to metabolic roles (similar to Archaea), eukaryal functions encompassed a diverse range of processes including ‘sex determination [GO:0007530]’, regulatory [GO:0044089] and immunological roles [GO:0046634], functions related to the development of mammary glands [GO:0061180], and others (Table 4). Finally, none of the archaeal or eukaryal lost FFs was significantly associated with any of the highly-specific biological process GO terms, indicating that loss of FFs in these two superkingdoms occurred without any functional constraint. In contrast, two biological processes were predicted to be lost from Bacteria including, ‘cellular modified amino acid biosynthetic process [GO:0042398]’, and ‘pyrimidine-containing compound biosynthetic process [GO:0072528]’(Table 5).
We report the evolutionary dynamics of gain and loss events of protein domain FFs in hundreds of free-living organisms belonging to the three cellular superkingdoms. Structural phylogenomic methods were used to reconstruct ToLs from genomic abundance and occurrence of FF domains in proteomes. Standard character reconstruction techniques were then used to trace domain gain and loss events along the branches of the universal trees. Finally, molecular functions and biological processes of FFs were studied using traditional resources. The exercise revealed remarkable patterns:
How reliable is our study? Both abundance and occurrence were congruent with respect to the overall tree topologies and general conclusions drawn from the analyses. Both supported the existence of overwhelming gains in evolution. However, discrepancies also existed especially in the numerical differences for the gain-to-loss ratios among superkingdoms. In general, abundance (apparently) overestimated gains while occurrence underestimated losses. The higher number of gain-to-loss ratios in abundance models is an expected outcome as we are accounting for evolutionary processes such as gene duplications, gene rearrangements, and HGT that are known to increase the representation of genes in genomes. Ancient genes have more time to multiply and increase their genomic abundance compared to newly emergent genes. In contrast, occurrence merely describes the presence or absence of genes and provides a simplified view of the overall landscape of change. Another explanation is the possible existence of methodological artifacts when dealing with genomic occurrence in parsimony analysis that excludes most of the ancient FFs as non-informative characters, when these are present in all proteomes. Moreover, occurrence fails to take into account the weighted contribution of ancient genes to the phylogeny and treats all characters equally. Thus trees built from abundance counts are better resolved at their base while trees built from occurrence behave poorly in this regard [27]. We emphasize that the focus of this study is to highlight the relative contribution of domain gains and losses in the evolution of superkingdoms and not to evaluate which methodology is preferable. The finding that domain gains are overwhelming and increase approximately linearly with evolutionary time in both models is remarkable and suggests that the appearance of novel domains is a continuous process (Figures 4 and 5).
In our phylogenomic model, we rooted ToLs by character absence (i.e. 0) using the Lundberg method. We assumed that proteomes became progressively richer during the course of evolution. However, this implicit assumption did not lead to an increased number of domain gains as character state changes in both forward (e.g. 9 to 22) and reverse (12 to 5) directions were allowed and carried equal weights. Moreover, we evaluated the effects of ToL rooting on the calculations of domain gain and loss statistics by considering outgroup taxa instead of the Lundberg method. Superkingdom trees rooted with outgroup taxa led to similar tree topologies and supported the conclusion of overwhelming gains that we here report (Figure S1). However, we decided to exclude outgroup analysis from this study for two reasons. First, outgroups add an external hypothesis into the model and bias gains and losses by including artificial character changes in the most basal branches leading to outgroup taxa. Second, the selection of the most appropriate outgroups for each superkingdom is a complicated problem and is virtually impossible for the reconstruction of ToLs. However, it would be interesting to study the gain and loss dynamics at different levels of the SCOP hierarchy such as the FSF and F levels of structural abstraction. We expect that patterns reported in this study will remain robust regardless of the SCOP conservation level and will extend the analysis to FSF in a separate publication.
We used maximum parsimony to search for the best possible tree and described the evolution of 420 free-living proteomes using the entire repertoire of 2,397 FFs as phylogenetic characters. We note that parsimony is most appropriate (and gives superior performance) for this kind of analysis as it performs better when the characters are evolving under different evolutionary rates [100]. Moreover, rescaling of raw abundance values into 24 possible character states considerably reduces the likelihood of convergent evolution. Reconstructing evolutionary history of species and studying domain emergence and loss patterns is a difficult problem complicated by a number of considerations (e.g. taxa and character sampling, biases introduced by organism lifestyles, ecological niches of organisms, and non-vertical evolutionary processes). We attempted to eliminate these problems by reconstructing whole-genome phylogenies, sampling conserved FF domains as characters, excluding parasitic and facultative parasitic organisms from study, and by using multistate phylogenetic characters. However, we realize that no method is free from technical and logical artifacts. Our analysis largely depends upon the accuracy of phylogenetic reconstruction methods, current SCOP domain definitions, reliability of function annotation schemes, and literature for organism lifestyle. However, we expect that recovered results will remain robust both with data growth and improvement in available methods and that drastic revisions to existing databases would be unlikely. For that reason we caution the reader to focus on the general trends and main conclusions of the paper (i.e. overwhelming gains and its consequences) rather than the actual numbers and discrepancies between the phylogenomic methods. Quantifying gain and loss events on a global scale is a difficult problem and our work lays foundations for more and improved studies in the future.
We propose that grouping of protein domains into FFs provides a reliable character for a global evolutionary analysis that involves large number of proteomes. FF domains are both sufficiently conserved and informative to explore the many branches on the ToLs. The age and distribution of FFs in organismal groups is biased and carries the power to unfold superkingdom history and explain important structural and functional differences among superkingdoms. Based on our data, we propose the primacy of domain gains over losses over the entire evolutionary period, ongoing evolutionary adaptations in akaryotic microbes, evolution of emerging eukaryotic species by domain loss, an early origin for Archaea, and endosymbiosis leading to mitochondria as a crucial event in eukaryote diversification. Each of these conclusions is important for reconstructing the evolutionary past and predicting evolutionary events in the future.
|
10.1371/journal.pcbi.1003416 | Immune-Mediated Competition in Rodent Malaria Is Most Likely Caused by Induced Changes in Innate Immune Clearance of Merozoites | Malarial infections are often genetically diverse, leading to competitive interactions between parasites. A quantitative understanding of the competition between strains is essential to understand a wide range of issues, including the evolution of virulence and drug resistance. In this study, we use dynamical-model based Bayesian inference to investigate the cause of competitive suppression of an avirulent clone of Plasmodium chabaudi (AS) by a virulent clone (AJ) in immuno-deficient and competent mice. We test whether competitive suppression is caused by clone-specific differences in one or more of the following processes: adaptive immune clearance of merozoites and parasitised red blood cells (RBCs), background loss of merozoites and parasitised RBCs, RBC age preference, RBC infection rate, burst size, and within-RBC interference. These processes were parameterised in dynamical mathematical models and fitted to experimental data. We found that just one parameter , the ratio of background loss rate of merozoites to invasion rate of mature RBCs, needed to be clone-specific to predict the data. Interestingly, was found to be the same for both clones in single-clone infections, but different between the clones in mixed infections. The size of this difference was largest in immuno-competent mice and smallest in immuno-deficient mice. This explains why competitive suppression was alleviated in immuno-deficient mice. We found that competitive suppression acts early in infection, even before the day of peak parasitaemia. These results lead us to argue that the innate immune response clearing merozoites is the most likely, but not necessarily the only, mediator of competitive interactions between virulent and avirulent clones. Moreover, in mixed infections we predict there to be an interaction between the clones and the innate immune response which induces changes in the strength of its clearance of merozoites. What this interaction is unknown, but future refinement of the model, challenged with other datasets, may lead to its discovery.
| Malaria infections often consist of more than one strain of the same parasitic species. Understanding the within-host competition between these various strains is essential to understanding the evolution and epidemiology of drug resistance in malarial infections. The infection process and the competition between strains involve complicated biological processes that are explained by various hypotheses. Mathematical models tested against experimental data provide quantitative measures to compare these hypotheses and enable us to discern the actual biological processes that contribute to the observed dynamics. We use a group of models against experimental data on rodent malaria to test various hypotheses. Such quantitative measures, in understanding rodent malaria, can be considered as a step towards understanding within-host parasite dynamics. Our work presented here demonstrates how confronting mathematical models with data allows the discovery of subtle and novel interactions between hosts and parasites that would be impractical to do in an experiment and allows the rejection of hypotheses that are incorrect. It is our contention that understanding the forces controlling within-host parasite dynamics in well-defined experimental model is a necessary step towards understanding these features in natural infections.
| Malarial infections often consist of more than one strain of the same parasitic species [1]–[3]. Parasite populations of multiple strains interact with one another directly via resource competition and indirectly via the host's immune response to the infection [4], [5]. These interactions affect the population dynamics of the competing strains [2], [4]–[13]. Population dynamics during such mixed infections, when compared to single infections, have been shown to exhibit different mortality rates for the parasites, rates of growth to peak density, maximum parasitaemia and renewed growth within hosts [6]. There is evidence to suggest that higher within-host densities may lead to higher transmission success [7], [14] and competitive interactions which may directly affect the rate of transmission [3], [8]. Such competitive interaction can drive the evolution of virulence in parasites [7], [15]. Consequently, understanding the within-host competition between strains, is essential to understanding the evolution of virulence and drug and vaccine resistance in malarial infections [7], [16]–[18].
Several experimental studies of mixed infections of P. chabaudi clones have demonstrated competitive suppression of less virulent clones by virulent clones [7], [14], [15], [19]. These studies have led to some interesting speculation on the potential mechanisms responsible for the competitive suppression. However, an exact mechanism has yet to be established. An experimental study of mixed infections of two P. chabaudi clones, by Taylor et al., provides evidence for competitive suppression of one of the clones irrespective of initial dose [14]. Mice in three treatment groups were infected with virulent (ER) and avirulent (CR) clones of P. chabaudi with different ratios of initial parasite numbers. The competitive suppression of avirulent clone at the later stages of infection in all three treatment groups were attributed to clone-specific and cross-immunity of the host induced by the parasite strains. However, the exact role of host immune response on the suppression of CR could not be explored.
In another experimental study, 7 genetically closely related strains of P. chabaudi, differing in virulence, were tested against an unrelated, and more virulent strain of P. chabaudi [7]. Densities of individual parasite strains in mixed infections were tracked for 14–50 days. In all infections the virulent strain competitively suppressed the avirulent strains. Among the avirulent strains, the ones that were more virulent in single-strain infections achieved greater parasite densities and also suffered relatively less competitive suppression than the less virulent strains when in competition. The exact mechanism by which the avirulent clone is suppressed could not be established.
Another study showed that a virulent clone obtained a competitive advantage due to larger parasite and gametocyte densities, compared to an avirulent clone, during mixed infections [15]. Compared to respective single infections, both strains experienced reduction in both asexual parasite and gametocytes densities. However, the suppression in the gametocytes density of an avirulent clone was larger compared to the virulent strain during mixed infections. Virulent clones reached larger parasite densities compared to avirulent clones both in single and mixed infections. This study demonstrated the importance of within-host competition in the spread and selection of virulence in parasite evolution.
Recently a series of experiments were designed to study the effects of parasite genotype, residency and time of infection on within-host parasite densities during mixed infections. In these experiments two pairs of distinct clones of P. chabaudi were inoculated into mice either simultaneously or 3 or 11 days apart and their population sizes were tracked using immunofluorescence or quantitative polymerase chain reaction [19]. In all the experiments, at least one of the two clones suffered strong competitive suppression during mixed infections. It was observed that the avirulent clone suffered from competition even when it infected mice before the virulent clone, whereas the virulent clone suffered from competition only when infecting mice after the avirulent clone. It was suggested that host immunity along with competition for resources played an important role in causing the suppression of one of the clones during mixed infections. However, the extent of the contribution of resource limitation and host immune response to competitive suppression could not be disentangled.
In a recent paper examining competition between malaria clones we found direct experimental evidence of immune-mediated competition [20]. This was the first evidence of such competition in any host-parasite system. Two genetically distinct clones of P. chabaudi (AS and AJ) were co-infected into mice. The AS clone is less virulent than the AJ clone, being associated with a lower peak parasitaemia, less RBC loss and less weight loss [21]. In order to determine if the immune response mediated competitive suppression, both immuno-competent and immuno-deficient (T-cell depleted) mice were infected. If competition was mediated by the immune response, then the expectation was that competitive suppression would be weaker in immuno-deficient mice than in immuno-competent mice. Compared to single clone infections, the presence of the AJ clone in mixed infections competitively suppressed the AS clone. Importantly, suppression was alleviated in immuno-deficient mice. The statistical analysis of the data, however, did not allow the determination of the nature, strength and precise timing of the suppression. Moreover, the data suggested that other competitive mechanisms must be important, although what those mechanisms were was impossible to determine.
Our aim in this paper is to re-examine this dataset using a dynamical model-based Bayesian inference approach in order to determine the nature of these competitive interactions, immune mediated or otherwise. Parameterised dynamical (process) mathematical models are fitted to the experimental data. Mechanism can then be inferred from the estimated parameters – i.e., a parameter for a mechanism (such as immune-mediated clearance rates of parasites) that is different across treatments suggests possible causes of competitive interactions [22]–[25]. This approach allows formal and quantitative testing and comparison of hypotheses for the effect of factors that cannot be easily measured empirically.
We briefly describe the experiment here. See [20] for a more detailed description.
Three different phenotypes of 12–14 week old, female BALB/c mice were used: (i) wildtype mice; (ii) female nu/nu mice (“nude mice”; Harlan UK); and (iii) nude mice reconstituted with T cells taken from wildtype mice. The mutation nu is a recessive mutation that blocks the development of the thymus and hence these mice have no mature T-cells which impairs their immune systems [26]. Both nude mice and nude mice reconstituted with T-cells are genetically different from wildtype mice. Only the nude and reconstituted mice were used in the analysis in [20] to allow for the comparison of genetically similar immuno-competent and immuno-deficient hosts; we present data for all three phenotypes here. The wildtype mice provide additional statistical power to discriminate between competing hypotheses about the cause of competitive suppression.
Mice of each phenotype (wildtype, nude, reconstituted) were inoculated intraperitoneally with AS or AJ or AS and AJ parasitised RBCs (pRBCs); resulting in treatment groups. There were seven mice in the treatment groups with nude mice, and six mice in each of the treatment groups with reconstituted and wildtype mice. RBC and parasite densities were measured on days 0, 2, 4, and then daily until day 18 when the experiment was terminated. Measurements were taken at 08:00 hr before asexual merozoites have yet to replicate within pRBCs. RBC density was measured by flow cytometry, parasite density was measured by quantitative PCR. We have previously estimated the error in these measurements [25].
We extend the model of malaria parasite bloodstream asexual replication developed in [25] to mixed infection of two clones and further include RBC age-structure [22] and background loss of pRBCs. We provide a brief description of the model here; the mathematical details with supporting tables of variables, parameters and their priors are given in the Supplementary Materials.
In P. chabaudi, parasitised RBCs (pRBCs) rupture synchronously every 24 hours [27], releasing on average 6–8 parasites (merozoites) into the bloodstream [28]. These newly released merozoites infect further RBCs and the cycle repeats. The rupture of pRBCs (schizogony) occurs at approximately midnight [27], [29].
We use a discrete-time formulation to model the dynamics, where each time step corresponds to a single day. The start of day is defined as the point immediately following rupture of pRBCs, before any infection has occurred (i.e., the point at which merozoites are released into the bloodstream). The script for our model can be accessed at https://code.google.com/p/bayesian-model-based-inference/
We assume that the processes determining RBC and parasite densities occur on two non-overlapping timescales. The first corresponds to the short infection phase during which merozoites infect RBCs, which occurs within a few minutes following schizogony. The second and subsequent timescale (the remainder of the day) corresponds to the RBC turnover phase: the parasites replicate within pRBCs, and new unparasitised RBCs (uRBCs) migrate from the bone marrow and spleen into the bloodstream [24], [30], [31]. At the end of the RBC turnover phase, surviving pRBCs rupture and release new merozoites. In normal, homeostatic, conditions, migration of uRBCs exactly replenishes the natural loss of RBCs [32]. In anaemic conditions RBC production and migration (erythropoiesis) is up-regulated at a rate proportional to the difference between the normal RBC density and the actual density a few days in the past [22], [25], [33], [34]. As discussed below, one possible cause of competition is differential RBC-age preference between the two clones [35], [36]. We therefore extend the model to include age structure of RBCs as in [22], [23]. We distinguish between 1–2 day old immature RBCs (reticulocytes) and the older mature RBCs (normocytes) they develop into.
We model separate, time-dependent, adaptive immune responses against merozoites in the infection phase [37], [38] and pRBCs in the turnover phase as in [25]. We tried three different functional forms for the clearance rates: piecewise linear (as in [25]), exponential and sigmoidal. In addition, we include a constant, low-level background loss rate of free merozoites in the infection phase as in [22], [25], [36], and a constant background loss rate of pRBCs. We also include time-dependent bystander killing of uRBCs in the turnover phase [25], [39], [40]. The mathematical details are given in the Supplementary Material.
Biologically, competition between clones can be mediated by several processes as listed in Table 1. The main difference between the mouse phenotypes is their immuno-competence. Hence, we expect to see an effect of phenotypes in processes that involve host immune response. This allows us to identify all processes including host-immune response that may play a role in competitive suppression. Previous modelling studies of P. chabaudi [22], [36] have shown that clone-specific RBC age preferences can cause competitive suppression of a less virulent clone, when virulence is a function of the age range of RBCs a parasite can invade. Our first hypothesis H1, considers this possibility (Table 1). In our model age-preference is modelled as different merozoite infection rates of reticulocytes and normocytes; and respectively. It turns out, however, that we cannot separately identify (estimate) these two rates; only their ratio , can be estimated (see Supplementary Material for details). Our second hypothesis H2, considers whether the number of merozoites that burst from pRBCs , is different between clones. Evidence that burst sizes are significantly higher for the more virulent clones compared to avirulent ones have been observed previously [22], [23]. Our third hypothesis H3, considers the possibility that competition for resources within multiply parasitised RBCs may cause differential death rates , of the different clones. A previous in vitro study of P. falciparum has shown that diffusible molecules within RBCs can regulate the growth and gametocytogenesis of parasites [41]. Hence, multiple parasites within the same RBC may competitively interfere for these resources. Our fourth hypothesis H4, considers whether RBCs infected by the different clones have different constant background death rates . We do not have a specific process in mind that might cause such a difference, other than it not being caused by clone-specific adaptive immunity (which we consider in hypothesis H7). Our fifth hypothesis H5 considers competition caused by a combination of two processes: differential background loss rates of merozoites , and differential merozoite infection rates of normocytes . Mathematically we cannot separately estimate these two parameters; only their ratio , can be estimated (see Supplementary Material). The parameter can be interpreted as the RBC density at which a single merozoite has a 50% chance of infecting a RBC (assuming no age preference, and in the absence of an adaptive immune response against merozoites). Hence, if one clone has a higher background merozoite loss rate or a lower merozoite infection rate of normocytes, this clone has a lower chance of infecting RBCs at a particular RBC density, and, therefore, is at a competitive disadvantage. In hypotheses H6 and H7 we consider clone-specific adaptive immunity against merozoites and pRBCs respectively.
Competition is incorporated into the model via clone-specific parameters (Table 1). We would expect, after fitting the model to the data, for some of these parameters to exhibit different estimates between clones. We may then infer that competitive suppression is mediated by the processes whose parameters differ between clones. For example, the analysis by Råberg et al. [20] strongly suggested that competitive suppression was mediated by some aspect of immunity (hypotheses H6 and H7), so we might observe weaker adaptive immune clearance of the AJ clone compared to the AS clone.
We test the causes of competition as follows. The full model, described above and in more detail in the Supplementary Material, includes all possible causes of competition. That is, all parameters included in hypotheses H1 to H7 are allowed to be different between the two strains. This so called “all-cause” model is fit to the data. Each single-cause model is obtained by keeping the parameters clone-specific for the cause of interest and making parameters clone-non-specific for all other causes. Each single-cause model is fit to the data. If none of the single-cause models adequately predict the data, we would then examine dual-cause models, and so on. This was not necessary however. The all-cause model acts as a reference because it has the highest maximum likelihood. Any single-cause model that has a maximum likelihood similar to (but necessarily smaller than) the all-cause model fits the data as well as the all-cause model.
There is considerable variability in parasite and RBC dynamics of the mice both between and within the treatment groups. This suggests that there is variability in the underlying processes that govern the dynamics and thus in the parameters. Furthermore, the immune responses are significant sources of variability in vivo and RBC invasion rates may vary between-mice due to the multi-factorial nature of such processes which involve the interaction of many host and parasite proteins. We therefore make no assumption about which parameters are invariant across mice and estimate each parameter separately for each mouse.
In the experiment, measurements were taken at approximately 08:00 hrs, roughly of the time between successive rupture events. We therefore fit the model predictions of RBC and parasite densities at this time. Total RBC density was measured by flow cytometry while the total parasite density was measured using quantitative PCR. We have previously shown that the measurement errors in RBC and log10-parasite densities are normally distributed with standard deviations and respectively [25]. Assuming independence in the errors, the likelihood of the model parameters, given the data for a particular mouse, is simply the product of the likelihoods of the parameters given each data point. We use an adaptive, population based Markov chain Monte Carlo method with power posteriors [24], [42]–[44] to sample the posteriors and compute marginal likelihoods (see below). The Markov chains had a burn-in of samples. Inferences are based on samples thinned to 3,000 samples. Five simulations were run to obtain means and standard errors of the marginal and maximum likelihoods.
We use maximum and marginal likelihoods to compare our competing hypotheses about the causes of competitive suppression. Marginal likelihoods naturally penalise models that over-fit data with too many parameters. Marginal likelihoods are computed for each mouse [44]. Assuming mice are independent, the marginal likelihood over all mice is simply the product of their individual marginal likelihoods.
When comparing two hypotheses the ratio of their marginal likelihoods, their Bayes factor, is a convenient statistic. Bayes factors quantify how much more likely one hypothesis is over another given the observed data [45]. However, when comparing multiple hypotheses it is more convenient to compare the logs of their marginal likelihoods directly. A difference of 1 log would be strong evidence in favour of the more likely hypothesis, and a difference of 2 logs or more would be decisive evidence [45].
The experimental data on nude and reconstituted mice are discussed in [20], [25]. We present the data here in a different format and present the previously unpublished data of infections in wildtype mice.
The average parasite densities for the three mouse phenotypes for single (solid lines) and mixed (dashed lines) infections of the AJ (left panel) and AS (right panel) clones are shown in Figure 1. The results clearly demonstrate the strong competitive suppression of the AS clone in mixed infections (dashed lines) compared to single infections (solid lines) [20]. This is the case for all mouse phenotypes. The AJ clone, in comparison, does not exhibit any significant changes in parasite density during mixed infections when compared to single infection.
Figure 1 also shows that the strength of competitive suppression of the AS clone is stronger in immune-competent mice. This is seen by comparing the diverging densities of the AS clone in nude (dashed black line) and reconstituted mice (dashed red line). This result suggests that the AS clone undergoes immune mediated competition [20].
The all-cause competition model was fit to the single and mixed infection data from [20]. The analysis of the fits to single infections has been reported elsewhere [25] so we only assess the fits to the mixed infections here.
We first tested the fits for the three functional forms of the adaptive immune responses. The sigmoidal response gave the best fits in terms of maximum and marginal likelihoods (see Table 2), the piecewise linear response gave slightly worse fits, and the exponential response gave significantly worse fits. For the rest of the paper we analyse the fits of the sigmoidal model. The results and conclusions from using the piecewise linear model are identical. We do not consider the exponential model any further.
The standardised residuals of the all-cause model for each mouse phenotype are given in the Supplementary Material (Figs. S3, S4 and S5). The Q-Q plots of the all-cause model for each phenotype are given in the Supplementary Material (Figs. S6, S7, S8). The standardised residuals of an adequate model should be approximately normally distributed with mean 0 and standard deviation 1. The overlaid residuals and the Normal Q-Q plot of the fits suggest that the all-cause model is adequately fitting the data with some minor over and under estimation of the dynamics. We can therefore be confident that the all-cause model is adequately explaining the data and so we proceed to the single-cause models.
Figure 2 plots -marginal likelihood against -maximum likelihood of the models tested in Table 1. The all-cause model must have the highest maximum likelihood amongst all our models because it has the most degrees of freedom. We would expect, though, for it to have a low marginal likelihood due to over-fitting. The single-cause models may fall into one of two categories. i) A model may have a substantially poorer fit than the all-cause model causing it to have a substantially lower maximum likelihood. Its marginal likelihood may be lower or higher than the all-cause model. ii) A model may have almost as good a fit as the all-cause model causing it to have a similar maximum likelihood to it and a substantially higher marginal likelihood. Models falling into the latter category are considered minimal adequate models: they predict the data well with as few parameters as possible [46].
It is clear from Figure 2 that only one model falls into the minimal adequate category. The model with clone-specific differences in has a maximum likelihood slightly smaller than the all-cause model, meaning that it predicts the data almost as well. Its marginal likelihood is much higher because it has far fewer parameters. All other models can cause competition (results not shown). However, either their maximum likelihoods are at least an order of magnitude lower or their marginal likelihoods are significantly lower. Figures S9, S10, S11 in the Supplementary Material show marginal against maximum likelihoods for the three mouse phenotypes separately. In all, the model with clone specific differences in consistently has the highest marginal likelihood and similar maximum likelihoods to the all-cause model. We can thus conclude that clone-specific differences in are sufficient to adequately explain the competitive suppression of the AS clone. All other parameters can be assumed to be the same between the two clones.
The fits to individual mice data of the single-cause model with clone-specific are shown in Figs. 3, 4. RBC density in reconstituted and wildtype mice recover after the first peak in parasite density, but then recrudesce around day 14 post infection. By comparison, RBC density does not recover in nude mice and they die.
Figure 5 shows the means (and their standard errors) of the posterior means of among mice within each treatment group. There are six features in Figure 5 that are pertinent for understanding how contributes to competitive suppression of the AS clone.
We discuss the significance of these results next.
In mixed infections of virulent AJ and avirulent AS P. chabaudi clones, the AJ clone competitively suppresses the AS clone [20]. This competition is thought to be mediated partially by the immune response because in immune-deficient mice competitive suppression is alleviated [20]. The aim of this paper was to provide a quantitative assessment of the, possible, multiple factors that cause this competition. Drawing on hypotheses from experimental data and the mathematical modelling literature we built dynamical models and fitted them to the experimental data. The outputs were analysed using a Bayesian inference approach.
We tested seven possible mechanisms that could cause competitive suppression (Table 1). Our results suggest that just one model parameter , the ratio of background loss rate of free merozoites to their infection rate of normocytes, needs to be clone-specific in order to fully explain competition between the AS and AJ clones.
In fact, all of the mechanisms of competition we tested could explain competitive suppression (results not shown). However, these mechanisms did not predict the data as well as a clone-specific (see Figure 2). This does not imply that clone-specific differences in these other mechanisms do not exist. Other modelling work has suggested that clone-specific RBC age preference could cause competitive suppression [22], [6], [47], [48]. In these papers models were fitted to data from single-clone infections and the resulting estimated clone-specific, age-dependent infection rates used to simulate parasite and RBC dynamics in mixed infections. These simulations gave qualitatively similar dynamics to data from mixed infections thus suggesting that RBC age preference can cause competitive suppression. We went a step further in this study by fitting our model to the mixed infection data as well as the single infection data. This allowed us to quantitatively compare this mechanism with many others and demonstrate that, although it can explain competitive suppression, it does less well than clone differences in .
Before we discuss the biological interpretation of we first discuss the differences in its estimates across treatments (refer to Figure 5). In single infections, we found no difference in between clones (, ). In mixed infections, however, is significantly higher for AS than AJ (, ). The difference between and was significantly smaller in immune-compromised nude mice than in immune intact mice, both T-cell reconstituted and wildtype mice (, ). Therefore, we suggest that the reason why AJ competitively suppresses AS is because of clone-specific differences in , and the reason why competitive suppression is stronger in immune-intact mice is because the difference is larger in these mice.
In addition, in nude and reconstituted mice we found that significantly increased between single and mixed infections (nude: , , reconstituted: ). Whereas did not significantly change between single and mixed infections (nude: , , reconstituted: , ). The opposite was the case in wildtype mice: significantly decreased between single and mixed infections (, ) whereas did not significantly change (, ). We can offer no explanation for this qualitative difference between mice phenotypes, other than to note that nude mice and nude mice reconstituted with T-cells are genetically different from wildtype mice.
Our definition of is the ratio of background loss rate of merozoites , to the infection rate of normocytes . Thus it determines how many merozoites successfully invade RBCs; the larger its value the fewer merozoites which are successful. Moreover, because is assumed constant throughout the infection, its effect on parasite and RBC dynamics is felt from the first day of infection. Its effect on parasite dynamics is three fold. 1) It slows growth during the exponential growth phase (compare green (AS) and blue (AJ) lines in Figure 3). 2) This in turn determines the peak parasite density. This is because the timing and strength of the adaptive immune response is the same for both clones and adaptive immunity is the most important driver for halting and reversing parasite growth. If growth is slower (due to a larger ) then peak parasitaemia will be lower. 3) It speeds up the loss of parasites after the peak. All of the differences in the dynamics between the two clones in Figure 3 are due to clone-specific differences in , all other parameters, apart from initial parasite density, are non-specific.
The parameter can be mathematically interpreted as the RBC density at which a single merozoite has a 50% chance of infecting a RBC (assuming no age preference, and in the absence of an adaptive immune response against merozoites). But how do we interpret it biologically? We initially defined it to be the ratio of background loss rate of merozoites , to the infection rate of normocytes . The definition of is straightforward and has been used in one form or another in all published mathematical models of malaria parasite invasion of RBCs; it parameterises the rate at which merozoites infect normocytes in the absence of an immune response. Our definition of is based on the models of Mideo et al. [22] and Antia et al. [36]. These two papers base the value of on in vivo measurements of the loss of invasive ability of free merozoites [49]. These two papers fix the value of and therefore do not estimate its value, which we do here. Thus has always been defined as a property of the parasite and not as a property of the interaction between host and parasite.
Our finding that changes between single and mixed infections does not fit with the above definitions of and . We can think of no valid reason why should change between single and mixed infections. It is unlikely that different parasite clones could interfere with each others ability to find, attach and infect RBCs, especially when they are at very low densities early in the infection. It is possible that antibody against one clone could block the invasion of RBCs by another clone thus changing . However, we observe competitive suppression before an antibody response is activated as well as in T-cell deficient nude mice. Thus it seems unlikely that is changing between single and mixed infections.
This leads us to suggest that our definition of is at fault. It is likely that represents a combination of factors. We argue that one of these factors could be the innate immune response's clearance of free merozoites, and it is this factor that changes during mixed infections. First, is weakest in nude mice and strongest in wildtype mice (Figure 5, nude vs. reconstituted: , , wildtype vs. reconstituted: , ) which suggests that represents the ability of the immune response to clear parasites. Second, the relative difference between and is larger in immune-competent mice than in immune-compromised mice (Figure 5, , ) again suggesting that is determined by the immune response. Finally, in vivo experiments show that parasite growth rate in the exponential phase increases at low parasite dose and saturates at high parasite dose [50]. It was argued that this is because the innate response is limited in its ability to control large numbers of parasites [50]. Thus there is precedent for the argument that the strength of the innate response controls the growth in the exponential phase.
Although clone-specific differences in give the most probable fit to the data (Figure 2), we cannot rule out other clone-specific differences. In particular clone-specific adaptive immune clearance of merozoites and pRBCs. The models of these two hypotheses have an additional three parameters compared to the model of clone-specific . This explains their significantly lower marginal likelihoods. But even with more parameters they still do not fit the data quite as well as clone-specific (Figure 2). This is for the following reason. As the mice have not experienced malaria parasites before, the adaptive immune clearance rate of parasites must be negliglible (we assume 0) on the day of inoculation. The clearance rate must grow over the course of infection leading to the rapid decline of parasite numbers about a week post infection. Therefore the effect of the adaptive response on parasite dynamics is negligible in the first few days post infection. Therefore a model of clone-specific differences in adaptive immune clearance cannot explain the differences in the growth rates of the clones seen in mixed infections. These differences in growth rates are small (Figure 1), hence the similarity in the maximum likelihoods between the models with clone-specific adaptive responses and clone-specific .
Our results leave us with two unanswered questions: Why should the clearance rate of parasites by the innate immune response change between single and mixed infections? And why is the change in clearance rates positive for the AS clone in nude and reconstituted mice and negative for the AJ clone in wildtype mice (Figure 5)? We believe that the most likely answer to these two questions lies in the strength of cross reactive innate responses. The strength of the innate immune response to the parasite is determined by the density of parasites. Naturally the innate response to the AS parasite is higher in mixed than single infections. However, since AJ is the virulent clone, the addition of AS parasite in mixed infections has negligible effect on the total parasite density. Therefore, there is no extra stimulation of the density dependent response as a result of mixed infection. On the other hand, one could imagine that the innate response is dependent not only on density but also on the diversity of the infection, such that, more diverse infections are harder for the immune system to control. This could explain why the innate response against the AJ parasite in wildtype mice generally decreases in mixed infections when compared to single infections. One other possibility could be the interaction between innate responses triggered by schizogony of one clone adversely affecting the other due to the delay in schizogony of the affected clone. We are examining this idea with other data sets [27].
In conclusion, our dynamical model-based inference approach can be used to compare multiple hypotheses about biological processes underlying infection dynamics data. Using this approach we have shown that competitive suppression of an avirulent clone of P. chabaudi is most likely mediated through innate clearance of merozoites acting throughout an acute infection.
|
10.1371/journal.pcbi.1000868 | A Bayesian Approach to the Evolution of Metabolic Networks on a Phylogeny | The availability of genomes of many closely related bacteria with diverse metabolic capabilities offers the possibility of tracing metabolic evolution on a phylogeny relating the genomes to understand the evolutionary processes and constraints that affect the evolution of metabolic networks. Using simple (independent loss/gain of reactions) or complex (incorporating dependencies among reactions) stochastic models of metabolic evolution, it is possible to study how metabolic networks evolve over time. Here, we describe a model that takes the reaction neighborhood into account when modeling metabolic evolution. The model also allows estimation of the strength of the neighborhood effect during the course of evolution. We present Gibbs samplers for sampling networks at the internal node of a phylogeny and for estimating the parameters of evolution over a phylogeny without exploring the whole search space by iteratively sampling from the conditional distributions of the internal networks and parameters. The samplers are used to estimate the parameters of evolution of metabolic networks of bacteria in the genus Pseudomonas and to infer the metabolic networks of the ancestral pseudomonads. The results suggest that pathway maps that are conserved across the Pseudomonas phylogeny have a stronger neighborhood structure than those which have a variable distribution of reactions across the phylogeny, and that some Pseudomonas lineages are going through genome reduction resulting in the loss of a number of reactions from their metabolic networks.
| Metabolic networks correspond to one of the most complex cellular processes. Most organisms have a common set of reactions as a part of their metabolic networks that relate to essential processes such as generation of energy and the synthesis of important biological molecules, which are required for their survival. However, a large proportion of the reactions present in different organisms are specific to the needs of individual organisms. The regions of metabolic networks corresponding to these non-essential reactions are under continuous evolution. Using different models of evolution, we can ask important biological questions about the ways in which the metabolic networks of different organisms enable them to be well-adapted to the environments in which they live, and how these metabolic adaptations have evolved. We use a stochastic approach to study the evolution of metabolic networks and show that evolutionary inferences can be made using the structure of these networks. Our results indicate that plant pathogenic Pseudomonas are going through genome reduction resulting in the loss of metabolic functionalities. We also show the potential of stochastic approaches to infer the networks present at ancestral levels of a given phylogeny compared to deterministic methods such as parsimony.
| Biological networks are under continuous evolution and their evolution is one of the major areas of research today [1]–[6]. The evolution of biological networks can be studied using various approaches such as maximum likelihood and parsimony [7], [8]. The maximum likelihood approach calculates the likelihood of evolution of one network into another by summing over all possible networks that can occur during the course of evolution under the given model. Parsimony, on the other hand, assumes minimum evolution and only considers those networks that correspond to the minimum number of changes between the two networks. However, the problem with these approaches is that enumeration of networks potentially occurring during evolution becomes impractical in the case of biological networks as the number of networks grows exponentially with the network size. Recently, the evolution of biological networks has been studied using stochastic approaches where efficient sampling techniques makes the problem computationally tractable. For example, Wiuf et al. [5] used importance sampling to approximate the likelihood and estimate parameters for the growth of protein networks under a duplicate attachment model. Similarly, Ratmann et al. [6] used approximate Bayesian computation to summarize key features of protein networks. The authors also approximated the posterior distribution of the model parameters for network growth using a Markov Chain Monte Carlo algorithm.
In this work, we focus on metabolic networks. The evolution of metabolic networks is characterized by gain and loss of reactions (or enzymes) connecting two or more metabolites and can be described as a discrete space continuous time Markov process where at each step of the network evolution a reaction is either added or deleted until the desired network is obtained [9]. To give a biologically relevant picture of evolution some reactions may be defined as core (reactions that cannot be deleted during the course of evolution) or prohibited (reactions that cannot be added) in the given networks. The evolution of metabolic networks can then be studied using simple (independent loss/gain of reactions) or complex (incorporating dependencies among reactions) stochastic models of metabolic evolution. We previously presented a neighbor-dependent model for the insertion and deletion of edges from a network where the rates with which reactions are added or removed from a network depend on the fraction of neighboring reactions present in the network [9]. In this model, two reactions were considered to be neighbors if they shared at least one metabolite. The model is summarized in Section ‘Neighbor-dependent model’ below. The neighbor-dependent model depicts a biologically relevant picture of metabolic evolution by taking the network structure into account when calculating the rates of insertion and deletion of reactions from a network. The model is, however, limited in the sense that it does not allow one to measure the strength of the neighborhood structure affecting network evolution.
Here, we present an extended model called the hybrid model that combines an independent edge model, where edges are gained or lost independently, and a neighbor-dependent model of network evolution [9] such that the rate of going from one network to another is a sum of the rates under the two models based on a parameter, which measures the probability of being in the neighbor dependent model. This allows estimation of the neighborhood effect during metabolic evolution. When modeling network evolution, we represent metabolic networks as directed hypergraphs [9]–[11], where an edge called a hyperedge represents a reaction and may connect any number of vertices or metabolites. Representing metabolic networks as hypergraphs not only captures the relationship between multiple metabolites involved in a reaction but also provides an intuitive approach to study evolution since loss or gain of reactions can be regarded as loss or gain of hyperedges.
We use the hybrid model to study the evolution of a set of metabolic networks connected over a phylogeny. Previous attempts to study the evolution of metabolic networks in a phylogenetic context include Dandekar et al. [12] and Peregrin et al. [13]. However, to our knowledge, the stochastic treatment of metabolic evolution over a phylogeny is an unexplored area. Here, the phylogenetic relationship between the networks is established using sequence data since the metabolic annotations available for the majority of genome-sequenced organisms are generated using automated annotation tools based on the similarity of predicted genes to genes of known function and, therefore, contain a huge amount of noise. In addition, we treat the branch lengths obtained using the sequence data as certain. The advantage of fixing branch lengths is that the calculations do not require summing over all branch lengths for the given tree. Calculating the likelihood over a phylogeny then requires a sum, over all possible networks that may have existed at the interior nodes of the tree, of the probabilities of each scenario of events. This is similar to the idea introduced by Felsenstein [14] for observing DNA sequences over a phylogeny. To sample the networks at internal nodes of the tree a Gibbs sampler [15], [16] is presented that samples a network conditioned on its three neighbors, including a parent and two children networks, for given parameter values. A Gibbs sampler for estimating the parameters of evolution that encases the Gibbs sampler for internal networks sampling is also presented. The sampler estimates the evolution parameters without exploring the whole search space by iteratively sampling from the conditional distributions of the trees and parameters. We demonstrate the Gibbs sampler by estimating and comparing the evolution parameters for the metabolic networks of bacteria belonging to the genus Pseudomonas. The Gibbs sampler can also be used to infer the ancestral networks of a given phylogeny. This is shown by inferring the metabolic networks of Pseudomonas spp. ancestors.
In the neighbor-dependent for the evolution of metabolic networks [9] hyperedges are inserted or deleted from a network depending on the fraction of neighboring hyperedges present in the network. Two hyperedges are considered as neighbors if they share a node. The model assumes that the number of nodes in a network remains fixed and there is a set such that of hyperedges connecting these nodes. The model also assumes the existence of a network called Reference Network which contains all these hyperedges. If the hyperedges in the reference network are labeled 1 to then any given network can be represented as a sequence of 0s and 1s such that the -th entry in the sequence is 1 if and only if the hyperedge labeled is present in the network , and 0 otherwise. Let the rate matrix describing the evolution under the neighbor-dependent model be denoted by . An entry in this rate matrix corresponds to the rate of going from a network to a network , which differs from at position . In the neighbor-dependent model, the rate of going from to depends on , and the neighboring hyperedges present in the network , and is given as follows:(1)where the function corresponds to the neighborhood component and is the appropriate entry from the rate matrix for the hyperedge . The rate matrix is given as(2)where is the insertion rate and is the deletion rate.
The neighborhood component weights the insertion and deletion rates by the proportion of neighbors present in the network and is given as follows:(3)The denominator in Equation 3 gives the number of hyperedges present in the current network.
Although the neighbor-dependent model summarized above produces a biologically relevant behavior whereby highly connected reactions are toggled more frequently than the poorly connected counterparts, it does not allow one to determine the strength of the neighborhood structure effecting the evolution of metabolic networks. To overcome this limitation, a parameter can be introduced in the model that corresponds to the neighborhood effect during the course of metabolic network evolution.
Consider two networks and which differ at position . The hybrid model combines the independent edge model where edges are added or deleted independently, and the neighbor-dependent model summarized above such that the rate of going from to is the sum of the rates under the two models based on a parameter , which specifies the probability of being in the neighbor-dependent model. The rate from to is given aswhere the term is the rate under the neighbor-dependent model given by Equation 1 and the term is the rate under the independent edge model corresponding to the appropriate entry from the rate matrix Q given by Equation 2. Substituting the value of from Equation 1, the above equation can be simplified as follows.(4)where the term corresponds to the neighborhood component given by Equation 3.
It can been seen from (4) that the model behaves under the independent edge model when equals 0 and under the neighbor-dependent model described in the previous section when equals 1. For example, consider the toy network shown in Figure 1A. The reference network containing all allowed hyperedges for this example system is also shown in the figure. The system behavior for different values of is illustrated in Figure S1 for the toy network when simulated under the hybrid model along with the number of neighbors for each hyperedge. The rates were calculated at each step using (4). An edge was then selected based on these rates and was inserted if absent from the current network and deleted otherwise. As expected, hyperedges evolve independently when , resulting in similar insertion frequencies for all hyperedges and increasingly reflecting their neighborhood as the value of goes up to unity. The fitness of the model is discussed in the Section ‘Fitness of the hybrid model’ below.
Biological networks are connected over a phylogenetic tree which is known through sequence analysis. Calculating the likelihood over a phylogeny requires a sum, over all possible networks that may have existed at the interior nodes of the tree, of the probabilities of each scenario of events. For example, Figure 1A shows an example system containing three networks , and with a phylogeny connecting the three networks shown in Figure 1B. Let the phylogenetic tree be denoted by . The likelihood of the tree is given as follows.(5)
Here denotes the parameters of the model, which is in the case of the neighbor-dependent model and in the case of the hybrid model. is the marginal probability of observing the root and denotes the pairwise likelihood of evolving from the network to the network conditioned on in time for the given parameters.
In general, the likelihood of a tree with more than three networks can be calculated using the recursion described by Felsenstein [17]. The likelihood at an internal node of the tree is given by the following recurrence relation(6)where and are left and right descendants of the node . The likelihood of the complete tree is then given as(7)where is the marginal probability of observing the root and is given by Equation 6.
Evaluating Equations 5 and 7 requires an algorithm to systematically and efficiently sample networks at the internal nodes of a tree and a method to calculate the pairwise likelihood of network evolution. A Metropolis-Hastings algorithm to calculate the pairwise likelihood based on sampling paths between network pairs was described by Mithani et al. [9], which calculates the likelihood by summing over paths between the given network pairs. To sample networks at the internal node of a tree, a Markov chain can be constructed where states correspond to networks at the internal nodes. The networks can then be sampled using a Gibbs sampler [15], [16] as described in the next section.
Given a set of networks related by a phylogenetic tree, the networks at the internal nodes of the tree can be sampled using a Gibbs sampler. The general idea is to sample each internal network by conditioning on its three neighbors (one parent and two children). This approach for sampling internal networks is similar to the one used by Holmes and Bruno [18] for DNA sequence alignment. However, instead of using linear sequences, the sampler takes into account the network structure when calculating the new state. The procedure is described below.
Consider a network with its three neighbors with branch lengths , . The new network is selected as follows.
Example Consider the network in Figure 2 for which new state is to be calculated. Denote the network by . The three neighboring networks of the network are the networks , and labeled as , and respectively. If denotes the neighborhood component for hyperedge then for the given rate parameters (insertion) and (deletion), and the neighbor-dependence probability the rate matrix is written asFor simplicity, assume that . The system then behaves under the neighbor-dependent model and the rate matrix simplifies toThe transition probability matrix of transforming to is then given asThe transition probability matrices and can be calculated in the similar fashion.
Once the transition probability matrices have been obtained, the sample for the new network can be drawn using Equation 8. For example, if the current configuration of the networks are taken as shown in Figure 2, then the sample for the new state , for hyperedge 1 is drawn from the following distribution:
The samples for hyperedges labeled 2 to 10 can be drawn in a similar fashion to obtain the new network.
The Gibbs sampler described above samples the internal networks on a phylogenetic tree for given parameter values. This can be extended to estimate the parameters of evolution where equals () in case of the neighbor-dependent model and () in case of the hybrid model. One way is to nest it within another Gibbs Sampler which iteratively samples internal networks and parameters from the distributions and respectively. The general outline of the Gibbs sampler is as follows:
The samples for parameters can be drawn using a Metropolis-Hastings algorithm [19], [20] as described next. Since the Metropolis-Hastings algorithm is a well-established method, it suffices here to give details about how a proposal for new parameters can be generated. Readers interested in the general details of the algorithm are referred to Chapter 1 of Gilks et al. [21]. The performance of the Gibbs sampler is discussed in Text S1.
The Metropolis-Hastings procedure described above to sample parameters requires the likelihood of the tree when moving in the parameter space. The likelihood can be calculated using Equation 5 which in turn requires calculation of the pairwise likelihood between network pairs. The pairwise likelihood can be calculated using the Metropolis-Hastings algorithm described in Mithani et al. [9] which calculates the likelihood by summing over all paths between the given network pair. However, for the Gibbs sampler described above in Section ‘Estimation of parameters’ this seems impractical since it will require running the Metropolis-Hastings sampler for all network pairs. An alternate way is to use a pseudo-likelihood value when calculating the acceptance probability for parameters. We calculate the pseudo-likelihood for a given network pair by dividing the network into smaller sub-networks and multiplying the pairwise likelihoods of the individual sub-networks.
Let denote the pseudo-likelihood from the network to the network in time for the given parameter values. This is given aswhere is the pairwise likelihood of evolving sub-network into calculated by solving the exponential . The procedure to obtain sub-networks containing at most hyperedges is outlined below.
An example is given in Figure S2, which shows the sub-networks for the toy network shown in Figure 1 for different values of . The above procedure was used to calculate the pseudo-likelihood of evolution of the toy network to the network (Figure 1A) for different subnetwork sizes, and the results were compared against the likelihood obtained by the MCMC approach described in Mithani et al. [9] and the true likelihood values obtained by evaluating . All likelihood values were conditioned on the starting network. The average CPU time taken by different approaches is shown in Figure 3 and the pseudo-likelihood values are listed in Table S1. The sub-network approach provides a reasonable approximation of the likelihood with a significant time advantage over the MCMC approach.
To see if the hybrid model fitted the metabolic network data better than the neighbor-dependent model, a likelihood ratio test was performed using the metabolic data for the bacteria belonging to the genus Pseudomonas. The results show that the hybrid model fits the metabolic data better than the neighbor-dependent model. For example, consider the metabolic networks in Pseudomonas fluorescens Pf0-1. The maximum likelihood estimates (MLEs) for the evolution of glycolysis/gluconeogenesis map [22] from Pseudomonas fluorescens Pf-5 to P. fluorescens Pf0-1 obtained using the Gibbs sampler described by Mithani et al. [9] were under the neighbor-dependent model and under the hybrid model. Using the MLEs, the likelihood of observing the data under each model was calculated. Assuming that evolution has been taking place for a long time, it is reasonable to use the equilibrium probability of a network to approximate the probability of observing the network. The equilibrium probabilities were calculated using the procedure described by Mithani et al. [9]. The maximum log likelihood obtained under the neighbor-dependent model equaled −76.53 whereas the maximum log likelihood obtained under the hybrid model equaled −63.47. The likelihood ratio test statistic was calculated as under degree of freedom. The -value on 1 degree of freedom suggests that the hybrid model fits the data better than the neighbor-dependent model. The MLEs, maximum log-likelihoods and the -values for different pathway maps in P. fluorescens Pf0-1 used in this analysis are listed in Table 1. The low -values for all the pathway maps suggest a better fit for the hybrid model compared to the neighbor-dependent model. Likelihood ratio tests for other genome-sequenced Pseudomonas strains used in this analysiss showed similar results (data not shown).
The fit of the data was further tested by comparing the degree distributions of the nodes obtained by simulating network evolution under the neighbor-dependent and hybrid models. The MLEs for the evolution of networks obtained under the two models were used as the simulation parameters. For example, when evolving the pathway maps in P. fluorescens Pf0-1, the parameter values listed in Table 1 were used. A total of 60,000 iterations were run with the first 10,000 iteration regarded as burn-in period. Samples were collected every iteration and degree distributions were calculated. The results for the six pathway maps used in this analysis are shown in Figure 4 for P. fluorescens Pf0-1 as an example which suggest a better fit for the hybrid model than the neighbor-dependent model. Similar results (data not shown) were obtained for the other genome sequenced Pseudomonas strains used in this analysis.
To test the Gibbs sampler described in Section ‘Sampling internal nodes’, the three network phylogeny shown in Figure 1 was used. The networks were sampled at the internal nodes for different rate combinations with the neighbor-dependence probability kept constant at 1. The likelihood value was then calculated using Equation 5 by summing over the networks visited by the sampler at each internal node for each rate combination. When calculating the likelihood over the phylogeny, the pairwise likelihood was calculated using matrix exponentiation. A total of 25,000 iterations were run for each rate combination with the first 10,000 iterations regarded as burn-in period. The exact likelihood of the phylogeny was also calculated by matrix exponentiation using all networks at each internal node. The likelihood values estimated using the networks visited by the Gibbs sampler were comparable to those obtained by summing over all 1024 networks. The true and estimated likelihood surfaces for a range of parameter values are shown in Figure S3.
We also ran the Gibbs sampler for parameter estimation for the toy networks. The sampler was run from a random starting value for 60,000 iterations with the first 10,000 iterations regarded as burn-in period. The samples were collected every iteration to reduce computational overhead relating to storage as well as the correlation between samples. A sample MCMC trace for the first 1,000 iterations of the sampler for the rate parameters is shown in Figure S4. The autocorrelation of parameters is plotted in Figure 5 suggesting an exponential decrease in the correlation as the lag between the samples increases. To test the performance of the sampler, the likelihood of evolution for different rate combinations visited by the sampler was also calculated using Equation 5 by summing over networks visited by the sampler with . As before, the pairwise likelihood was evaluated by calculating the exponential of the rate matrix. The maximum likelihood averaged over three runs was found to be for parameters which is very close to the true likelihood obtained by matrix exponentiation (Figure S3).
To study the metabolic evolution in bacteria, we used the Gibbs sampler to estimate the evolution parameters for the metabolic networks of bacteria belonging to the genus Pseudomonas. The diversity of pseudomonads, and the availability of genome-sequence data for multiple plant-associated Pseudomonas fluorescens, Pseudomonas mendocina, Pseudomonas putida, Pseudomonas stutzeri and Pseudomonas syringae strains, along with genome data for clinical isolates of Pseudomonas aeruginosa and for the insect pathogen Pseudomonas entomophila provide an excellent opportunity to use comparative genomic approaches to develop insight into the evolution of metabolic networks. The phylogeny connecting the seventeen genome-sequenced strains of Pseudomonas is shown in Figure 6A. The phylogeny was generated using multilocus sequencing analysis of conserved housekeeping genes ( gltA, gapA, rpoD, gyrB) [23]. The metabolic network data was extracted from the KEGG database [22] on January 2010 for pathway maps across the seventeen Pseudomonas strains shown in Figure 6A using the Rahnuma tool [24]. The evolution parameters were also compared between two Pseudomonas species: P. fluorescens, a saprotroph that colonizes the soil environment, and P. syringae, a plant-pathogen that is found on leaf surfaces and in plant tissues. The phylogenetic relationships between these species is shown in Figures 6B and C. The results are discussed here for the six pathway maps listed in Table 2 as they provide a representative set of different neighborhood characteristics observed across the Pseudomonas strains used in this analysis. The basic information for each network across the seventeen Pseudomonas strains is given in Table S2.
When estimating the parameters, the hyperedges corresponding to the reactions that were common to all seventeen Pseudomonas strains were defined as core edges and the hyperedges corresponding to the reactions not present in any of these seventeen species were defined as prohibited edges. Three independent replicates of the sampler were run from random starting values for 60,000 iterations for P. fluorescens and P. syringae phylogenies, and 110,000 iterations for the phylogeny connecting the seventeen Pseudomonas strains with the first 10,000 iterations regarded as burn-in period in each case. The samples were collected every iteration to calculate the posterior expectations and variances of the parameters. These are listed in Table 2 and the ESS used for parameter estimation are listed in Table S3. The convergence of the algorithm was tested by checking the trace of the MCMC runs initiated from different starting values. An example is shown in Figure S5, which shows the trace for the sampler run on P. fluorescens phylogeny (Figure 6B). The running times and the acceptance percentages of the algorithm are listed in Table S4 for all three phylogenies. We also calculated the number of insertion and deletion events for each reaction as well as at each branch of the Pseudomonas phylogeny for all six pathway maps. These are shown in Figures S6 and S7.
The high insertion to deletion ratio (Table 2) for all three phylogenies for the glycolysis/gluconeogenesis map, pentose phosphate pathway map and pyruvate metabolism map, which are defined as a part of the carbohydrate metabolism of the bacteria in KEGG [22] and for the histidine metabolism map, which is a part of amino acid metabolism, suggests that very few reactions are missing from these networks in one or more Pseudomonas strains used in the analysis, resulting in a highly conserved network. Lysine and phenylalanine pathway maps, on the other hand, have higher deletion rates compared to the insertion rates suggesting a variable reaction distribution across the Pseudomonas phylogeny and instability of these functionalities. The results obtained in this study are consistent with the previous observation that the histidine metabolism map shows conservation of reactions across pseudomonads (Mithani, Hein and Preston, submitted) and that many Pseudomonas strains are able to use histidine as sole carbon and nitrogen source [25] whereas lysine and phenylalanine pathway maps have few conserved reactions across pseudomonads (Mithani, Hein and Preston, submitted) and are poor nutrient sources for these bacteria [25]. The results also indicate that the pathway maps which are highly conserved across the seventeen Pseudomonas strains, i.e. glycolysis/gluconeogenesis map, pentose phosphate pathway map, pyruvate metabolism map and histidine metabolism map, also have higher neighbor dependence probabilities compared to the other two pathway maps, which have variable reaction distribution across the Pseudomonas phylogeny. This might suggest a relationship between the neighborhood structure and the conservation of networks.
The comparison of the evolution parameters between P. fluorescens and P. syringae provides interesting insights into the evolution of the metabolic networks of these bacteria. For example, the insertion and deletion rates are generally higher in P. fluorescens than those in P. syringae suggesting a higher number of insertion and deletion events in P. fluorescens networks compared to P. syringae networks. This was expected since the evolutionary distance between the P. fluorescens strains is greater as compared to P. syringae strains (Figure 6) allowing more time for the networks in P. fluorescens to evolve. A higher deletion rate for lysine and phenylalanine pathway maps in P. syringae compared to P. fluorescens, however, suggests that P. syringae have had a higher number of deletion events than P. fluorescens during the course of evolution. This supports the finding that P. syringae have gone through a high number of deletion events than expected based on the comparison between observed and expected distribution of reactions across the Pseudomonas phylogeny, and the identification of reactions that are uniquely present or absent from a single lineage (Mithani, Hein and Preston, submitted). In addition, a very low insertion to deletion ratio () for lysine metabolism in P. syringae suggests a high number of deletion events in the lineage and consequently the loss of the ability of these bacteria to assimilate lysine. This is in agreement with nutrient utilization assays, which have reported that bacteria belonging to the species P. syringae do not assimilate lysine as a nutrient source [25]. Phenylalanine metabolism also has a higher deletion rate as compared to insertion rate in both P. fluorescens and P. syringae lineages. This in conjunction with experimental data reporting the weak ability of these bacteria to utilize phenylalanine as a nutrient source might lead to a hypothesis that both P. fluorescens and P. syringae are drifting towards losing their ability to assimilate phenylalanine. Overall, the results show that genome reduction is taking place in plant pathogenic bacteria belonging to the species P. syringae at a higher rate than their non-pathogenic counterparts in the species P. fluorescens.
The final aim of this study was to infer reactions present in the common ancestor of Pseudomonas spp. and of individual species of Pseudomonas. One way to address this is to predict that the common ancestor contained all the reactions that are common to existing Pseudomonas. The variable reactions can then be assigned using a parsimonious approach which generates a conservative model of network evolution in which a minimum number of events occur. However, the results above suggest that some lineages, particularly P. syringae, have undergone deletion events relative to the common ancestor and that some reactions absent in one or more modern pseudomonads might be present in the ancestral strain. To take this into account, stochastic approaches such as the Gibbs sampler described in Section ‘Estimation of parameters’ can be used to sample ancestral networks from the posterior distribution of networks and the likelihood of reactions being present at various levels of the phylogeny can be calculated.
To demonstrate this, the Gibbs sampler was run on the pathway maps listed in Table 2. The Gibbs sampler was run with the same settings that were used for parameter estimation and samples for the networks at internal nodes of the Pseudomonas phylogeny (Figure 6A) were collected. The degree distributions of nodes at the ancestral levels of the phylogeny are given in Figures S8, S9, S10, S11, S12, S13. The likelihood of reactions being present at each level was obtained by calculating the proportion of times each hyperedge was present in the sampled networks. The results are shown in Figures 7A–12A. Only alterable reactions, that is the reactions which were neither defined as core nor were defined as prohibited in the networks, are shown.
The ancestral network reconstruction using the Gibbs sampler reported high likelihood values for reactions which are present in all the networks down a lineage and low likelihood values for reactions which show variable distributions across the Pseudomonas phylogeny. For example, in the pentose phosphate pathway map (Figure 8A), the reaction R01066, which is present only in the three P. syringae strains, was assigned a very high likelihood of being present in the common ancestor of P. syringae pv. phaseolicola 1448A and P. syringae pv. syringae B728a as well as in the common ancestor for all the tree P. syringae strains but a very low likelihood of being present for all other internal networks. In contrast, R06836, which is present in sixteen out of the seventeen Pseudomonas strains (absent in P. fluorescens Pf-5), is reported to have high likelihood values of being present in all internal networks of the phylogeny.
Ancestral predictions were also generated under the parsimony model for these networks using the Fitch Algorithm [26]. When assigning the reactions at the ancestral nodes the ties were resolved in favor of presence of reactions. The results are shown in Figures 7B–12B. Reactions for which parsimony failed to resolve ancestral predictions at the root are marked with asterisks (*). Predictions generated for the Pseudomonas common ancestor using parsimony analysis are nearly identical to predictions generated for the P. aeruginosa common ancestor, which would be expected as parsimony assumes minimum evolution. In addition, parsimony generated a conservative model of network evolution in which a minimum number of events occur, but the stochastic approach takes network information into account when predicting ancestral networks. For example, in the case of the lysine degradation map (Figure 9), six out of the ten variable reactions are reported to be absent from all the ancestral networks of the Pseudomonas phylogeny using the parsimony approach whereas the stochastic approach taking the reaction neighborhood data into account assigns non-zero likelihood values to these reactions for being present in the ancestral pseudomonads. Similarly, all four reactions which are predicted to be absent from all ancestral pseudomonads in the histidine metabolism map under the parsimony model have non-zero likelihoods of being present in the ancestral networks using the stochastic approach (Figure 10). The results for ancestral network reconstruction for phenylalanine metabolism (Figure 11), on the other hand, suggested a very low level of conservation of reactions across the Pseudomonas phylogeny using both approaches and most of the variable reactions were predicted to be absent from the common ancestor. The variable distribution of these reactions across the seventeen Pseudomonas strains along with the results of ancestral network reconstruction suggests that these reactions might have been gained independently at organism level.
In this study, we have used a Bayesian approach to study the evolution of metabolic networks. We extended the neighbor-dependent model described by Mithani et al. [9] by introducing a parameter that estimates the probability of being present in the neighbor-dependent model. This not only provides a better fit for the data but also has an advantage over the existing model since it allows one to estimate the strength of neighborhood structure affecting the evolution of given networks. It must, however, be kept in mind that inferring the neighborhood effect solely on the basis of the neighbor-dependence probability might bias the results due to the fact that a high proportion of reactions involved in central metabolism of an organism will always be present due to their functional importance. Using ortholog and synteny data in conjunction with neighbor-dependence probability would lead to better inference of the role of network structure on metabolic evolution. The idea being that if a reaction is present in most of the species that are evolutionarily close to the one being considered then it has a higher chance of being added, and if it is genetically linked to other reactions then they have a greater chance of being consecutively added or deleted.
The neighbor-dependent model [9] and the hybrid model described here define reaction neighborhood as reactions sharing at least one metabolite. Alternate definitions of reaction neighborhood are also possible. For example, one possible alternate is to consider the reaction directions when calculating the neighborhood and to regard two reactions as neighbors only if the metabolite connecting the two reactions is a substrate of one and the product of the other. Similarly, it is also possible to use other measures such as sequence similarity [27]–[29] or network distance measures [30], [31] in conjunction with the network structure to model the evolution of metabolic networks. There are, however, limitations associated with the models of metabolic evolution solely based on network structure and sequence similarity. There are a number of other factors affecting metabolic evolution. These include substrate availability (for example, availability of a new nutrient in the environment may favor the insertion of reactions which bring this new metabolite into the mainstream metabolism), gene expression (for example, a decrease in the gene expression relating to an enzyme catalyzing a reaction may force the metabolic network to find alternate routes) and reaction mechanism (for example, a reaction which is chemically inefficient may be favored for deletion compared to an efficient reaction). Factors such as these must be taken into account when modeling the evolution of metabolic networks to depict a more realistic picture of evolution.
We also presented a Gibbs sampler to sample the networks at internal nodes of a phylogenetic tree where the internal networks were sampled by conditioning on three neighbors (one parent and two children) in an approach similar to the one used by Holmes and Bruno [18] for DNA sequence alignment. The sampler considered the network structure surrounding the hyperedge being sampled in addition to the state of the hyperedge in the three neighboring networks when calculating the new state thus resulting in an informed sampling procedure. When sampling ancestral networks, it was assumed that all sampled networks were valid networks. However, not all networks may be functionally viable. For example, a network might not be able to produce a key metabolite which is required or may result in disconnected components that compromise network functionality. Checking for validity of networks occurring at ancestral nodes is an important area for further research.
A Gibbs sampler to estimate the evolution parameters was also presented. Standard distributions were used to generate proposals for parameters. The standard distributions provide satisfactory mixing of the MCMC sampler with appropriate scaling [32]. The rate parameters were sampled from a gamma distribution where scale and shape parameters were calculated from the current network and the proposals for neighbor dependence probability were generated using a beta distribution with its scale parameters calculated from the networks present at the leaves of the given phylogenetic tree. A uniform prior was used when estimating the parameters, which assigns equal probability to each point in the parameter space. It might be useful to explore the dependence between the number of insertions and deletions on the given phylogeny and to investigate the use of other prior distributions. Besides this, when calculating the likelihood of evolution, it was assumed that the phylogenetic tree was known through sequence analysis. This simplified the problem by not requiring a sum over all possible branch lengths. However, when calculating the tree using sequence data, the branch lengths depend on the set of genes used for generating the tree and their evolutionary distances. Thus, different set of genes used could result in different branch lengths. To be able to make useful inferences using an evolutionary model such as the one described here, this uncertainty in the tree must be taken into account by summing over all possible branch lengths. In addition, the effects of using a phylogenetic tree constructed de novo from metabolic networks [30], [31] on the model need to be further explored.
The evolution parameters were estimated on a phylogeny connecting the metabolic networks of bacteria belonging to the genus Pseudomonas using the Gibbs sampler. The likelihood values for reactions to be present at various levels of the Pseudomonas phylogeny were also calculated using the networks visited by the Gibbs sampler and the results were compared to those obtained using parsimony. The stochastic assignment of reactions in ancestral networks offers an edge over deterministic approaches like parsimony which provides the minimum number of transformations required to explain the evolution of a reaction of the tree and can, therefore, underestimate the total number of changes. In addition, using the MCMC approach based on neighbor dependence takes network structure into account, and may be particularly useful in resolving ancestral predictions at the root of phylogenies, or in situations where parsimony is unable to assign states unambiguously (see Figures 7 and 8).
An important factor affecting the results when estimating the evolution parameters and reconstructing the ancestral networks relates to the use of individual pathway maps. Although computationally tractable, individual pathway maps do not take a complete network perspective and may, therefore, lead to incorrect results by ignoring a part of the reaction neighborhood, the so-called border effect. This is particularly true for reactions which occur at the boundary of a metabolic pathway map, which may have a large number of their neighbors not included in that pathway map. The calculation of reaction neighborhood solely using the pathway map under consideration ignores all neighboring reactions that are not present in the pathway map thus affecting the likelihood values. For example, consider R01424 in phenylalanine metabolism. This reaction is present in thirteen out of seventeen pseudomonads including all three P. syringae strains, all four P. putida strains and two out of the three P. fluorescens strains. It was, therefore, expected that the reaction would have a high likelihood of being present in the common ancestor of P. fluorescens, P. syringae and P. putida but, on contrary, was reported to have a relatively low likelihood at this level (Figure 11). Closer inspection of the reaction revealed that it links the phenylalanine metabolism to the pathway map relating to benzoate degradation via coenzyme A and has neighbors spanning across multiple pathway maps. Phenylalanine pathway map contains only 2 neighboring reactions of the reaction R01424 whereas using the data from all pathway maps relating to metabolism results in 53 neighbors. Thus, evaluating the likelihood of R01424 at ancestral levels of the Pseudomonas phylogeny solely on the basis of reactions involved in phenylalanine metabolism leads to a very poor neighborhood surrounding the reaction and, consequently, weights down the presence of the reaction in the common ancestor resulting in a low likelihood value. A possible solution to overcome this border effect is to use the full network structure when calculating reaction neighborhoods. However, the computational feasibility of using full network structure when calculating reaction neighborhoods requires further investigation.
When performing the analyses the hyperedges present in all seventeen Pseudomonas strains were defined as core and the hyperedges missing from all the strains were defined as prohibited hyperedges. However, the results presented in this analysis suggest that pathogenic bacteria belonging to species P. syringae have gone through a high number of deletion events compared to other species. Assigning core edges solely on the basis of intersection model may, therefore, bias the results towards the loss of reactions which might be essential in non-pathogenic bacteria. Similarly, prohibiting reactions that are not present in any one of the seventeen genome-sequenced strains would prevent the common ancestor from having reactions which might have been lost very early during the course of evolution. To model scenarios like these the provision of having a lineage specific core and prohibited hyperedges must be explored. Alternatively, it might be useful to assign core and prohibited hyperedges using the ortholog data from closely related bacteria, or by incorporating metabolic information from organisms sharing the same environment in the set of permitted reactions. Comparing ancestral network predictions generated using different set of core and prohibited hyperedges might provide clues about the functionality of the common ancestors of the bacteria and the environment the ancestors might have colonized.
Finally, the analysis presented here uses data from the KEGG database. The metabolic annotations available for the majority of genome-sequenced organisms are generated using automated annotation tools based on the similarity of predicted genes to genes of known function and therefore contain a substantial amount of noise. For example, some genes predicted to have a broad enzymatic function are linked to multiple reactions, while others fail to meet the detection threshold for annotation and are therefore recorded as absent. Nevertheless, networks deposited in databases like KEGG are commonly treated as if they are as certain as sequence data, which is a serious error that undermines many present investigations. It would be desirable to take this noise into account while modeling the evolution of metabolic networks. One way would be to use hidden states to model experimentally validate metabolisms which are observed though predicted metabolisms. This will not only enable one to model the noise in the data but also allow correct prediction of a metabolism for an organism using homologous information similar to comparative genome annotation [33].
In summary, evolutionary modeling of metabolic network is an important area. Using statistical models of network evolution such as the one described here not only allow one to investigate how the metabolic networks evolve in closely related organisms but also enable testing of biological hypotheses such as specialization of genomes and identification of regions of metabolic networks that are under high selection.
|
10.1371/journal.ppat.1000069 | Conserved Mosquito/Parasite Interactions Affect Development of Plasmodium falciparum in Africa | In much of sub-Saharan Africa, the mosquito Anopheles gambiae is the main vector of the major human malaria parasite, Plasmodium falciparum. Convenient laboratory studies have identified mosquito genes that affect positively or negatively the developmental cycle of the model rodent parasite, P. berghei. Here, we use transcription profiling and reverse genetics to explore whether five disparate mosquito gene regulators of P. berghei development are also pertinent to A. gambiae/P. falciparum interactions in semi-natural conditions, using field isolates of this parasite and geographically related mosquitoes. We detected broadly similar albeit not identical transcriptional responses of these genes to the two parasite species. Gene silencing established that two genes affect similarly both parasites: infections are hindered by the intracellular local activator of actin cytoskeleton dynamics, WASP, but promoted by the hemolymph lipid transporter, ApoII/I. Since P. berghei is not a natural parasite of A. gambiae, these data suggest that the effects of these genes have not been drastically altered by constant interaction and co-evolution of A. gambiae and P. falciparum; this conclusion allowed us to investigate further the mode of action of these two genes in the laboratory model system using a suite of genetic tools and infection assays. We showed that both genes act at the level of midgut invasion during the parasite's developmental transition from ookinete to oocyst. ApoII/I also affects the early stages of oocyst development. These are the first mosquito genes whose significant effects on P. falciparum field isolates have been established by direct experimentation. Importantly, they validate for semi-field human malaria transmission the concept of parasite antagonists and agonists.
| Malaria is a parasitic infectious disease transmitted by mosquitoes. It impacts half the population of the world and kills 1 to 3 million people every year, the vast majority of whom are children aged below 5 in sub-Saharan Africa. There, the deadliest parasite is Plasmodium falciparum and its most important vector is the mosquito Anopheles gambiae. This study identifies for the first time specific A. gambiae genes that demonstrably regulate the density of mosquito infection by P. falciparum parasites circulating in malaria patients in Africa. These genes function in mosquito lipid transport and intracellular actin cytoskeleton dynamics, and act as an agonist and an antagonist, respectively, of the parasite ookinete-to-oocyst developmental transition. Importantly, our study validates for P. falciparum the concept of mosquito genes that support or hinder parasite development, a concept that we defined previously using a laboratory model system. Thus, the work constitutes a major contribution to understanding meaningful mosquito/parasite interactions in natural transmission conditions.
| Sub-Saharan Africa is the major and persistent focus of malaria, one of the most devastating scourges of humankind. There, P. falciparum is by far the most important human malaria parasite and A. gambiae its most important vector. Laboratory infections of A. gambiae by the convenient rodent model parasite, P. berghei, elicit broad responses encompassing multiple mosquito genes; some belong to classical innate immunity and act systemically, while others participate in local epithelial responses [1]–[6]. Extensive laboratory experiments with this parasite have established that the outcome of infection depends on finely balanced factors that affect, positively or negatively, the developmental cycle in the mosquito, mostly at the bottleneck of invading the midgut epithelium. However, to identify molecular interactions potentially suitable for developing novel interventions to interrupt the human malaria cycle, it is now important to analyze the invasion response and its consequences in more natural settings. Here, we analyze invasion responses and their consequences, using endemic populations of P. falciparum, the human malaria parasite, and a strain of A. gambiae established from mosquitoes collected in the same area.
The importance of field-based analysis of malaria transmission is underlined by recent studies of vector/parasite interactions. Anopheles mosquitoes are largely inhospitable to Plasmodium [7], and most species host few, if any, Plasmodium species in nature. Only a small number of Anopheles/Plasmodium species combinations have evolved to support effectively this parasitism. Even these Anopheles vectors eliminate most of the input parasites. Further, the level of A. gambiae resistance to P. falciparum apparently depends on specific genotype*genotype interactions: certain mosquitoes resist one subset of parasite genotypes while others resist a different subset [8]. Therefore, effective transmission may require specific compatibility between vector and parasite genotypes; conclusions from a particular combination may not apply to other combinations. Consistent with this concept, several different, operationally defined mosquito quantitative trait loci (QTLs) are associated with resistance against specific parasite species or genotypes [9],[10]. For instance, the genetically selected A. gambiae L3-5 strain possesses genetic traits that confer refractoriness to the simian P. cynomolgi parasite and numerous other species of Plasmodium, including non-African P. falciparum isolates, but not (or only very poorly) to its sympatric African P. falciparum isolates [11]. This observation suggests that sympatric mosquito/parasite populations may have co-evolved in permitting significant transmission intensities in the field. Further, the A. gambiae LRIM1 gene promotes and CTL4 inhibits P. berghei lysis and melanization [4]; the same LRIM1/CTL4 module apparently does not affect the outcome of sympatric A. gambiae/P. falciparum infections [12]. Such data suggest that vector immunity may have been co-adapted during co-evolution with the parasite [10],[13]. Nonetheless, P. falciparum clearance and melanization have been observed in infected wild A. gambiae in Africa, and thus constitute low frequency but natural phenotypes, in sympatric combinations [14],[15]. Mosquito loci that regulate the infection load or parasite melanization in the field have been mapped [10],[14] further suggesting that resistance is a default outcome of infection but is compromised in sympatric interactions.
Major cytoskeletal reorganization is a predominant response of parasite-invaded midgut cells [16],[17], and is accompanied by transcriptional regulation of genes implicated in cytoskeletal dynamics [5]. For example the gene encoding WASP, a local activator of actin cytoskeleton reorganization, is upregulated in the midgut epithelium during P. berghei invasion, and its silencing significantly increases P. berghei infection loads [5]. In contrast, P. berghei upregulates the mosquito precursor of Apolipophorin II/I (ApoII/I), a key circulating lipid transport regulator which appears to benefit both vector and parasite: its silencing disrupts mosquito egg development and drastically decreases parasite oocyst numbers [5].
Here, we examined whether local epithelial and systemic but not-classical immune responses of A. gambiae are pertinent to interactions between field isolates of P. falciparum and mosquitoes of a strain derived from a sympatric A. gambiae population. We discovered that in this strain both WASP and ApoII/I respond transcriptionally to P. falciparum as they do to P. berghei, and that their silencing impacts development of both parasites in the same direction. The conservation of these mosquito reactions against two distantly related parasite species allowed us to investigate the mechanism of action of these two genes in the tractable laboratory model setting. Using as tools P. berghei and A. gambiae strains that are either genetically selected (L35) or epigenetically modified (CTL4 kd), we clearly demonstrate that ApoII/I facilitates ookinete invasion of the mosquito midgut as well as development of early oocysts, while WASP acts only at the level of ookinete invasion. Evidently, WASP-mediated actin reorganization in the invaded epithelium is detrimental to both parasites whereas lipid transport by ApoII/I is beneficial, to both mosquito egg and Plasmodium development, in human as well as rodent malaria infections. Thus, despite the existence of genotype*genotype specific interactions [8], some important aspects of mosquito/parasite interactions are evolutionarily conserved. Further, the model laboratory transmission system can provide leads concerning genetic regulators; these must then be validated by translational research in more demanding, field-based systems of human malaria transmission.
Four A. gambiae genes representing diverse systemic or local epithelial responses elicited by P. berghei infections were selected for assessing in well-controlled experiments their involvement in transcriptional responses to infection by sympatric P. falciparum isolates and their RNAi-mediated silencing effect on parasite infectivity. As described in the introduction, ApoII/I and WASP are transcriptionally induced by P. berghei but have opposite effects on parasite development in the vector [5]. Two other genes were also identified as being transcriptionally induced during infection, but, their silencing had no effect on parasite infectivity [5]: CATHB encodes the proapoptotic enzyme Cathepsin B and was tested in light of observed apoptosis of parasite-invaded cells [16],[17], while KIN1 encodes a histidine-rich putative antimicrobial peptide produced under NF-κB control [18]. KIN1 is transcriptionally induced by bacteria and P. falciparum laboratory strain infections [2],[19], suggesting involvement in immunity. Finally, we tested a fifth gene, ApoIII, which encodes a polypeptide known to combine with ApoI and ApoII to form the insect lipophorin [20] and thus may also be involved in a systemic response to Plasmodium.
P. falciparum isolates were sampled during two high malaria transmission seasons, May 2005 and 2006, in a parasitological survey of 3,081 primary school, 5–11 year old, pupils. The survey was conducted in Mfou, a town 30 km outside Yaoundé, Cameroon. It identified an average of 51% P. falciparum infection prevalence and 5.9% gametocyte prevalence (% of blood samples with detectable asexual parasite blood stages (ABS) or gametocytes, respectively; Table S1).
Mosquitoes used in this study were from the Yaoundé strain that was colonized in the Yaoundé area in 1988 [21]. A previous study had demonstrated strong loss of polymorphisms and considerable divergence from natural populations in a 20-year old laboratory colony of a different mosquito, the neotropical species A. albimanus [22]. We examined to what extend the Yaoundé strain mosquitoes are representative of local populations, by determining the respective genetic diversities and calculating their genetic distance (divergence) from field-collected local A. gambiae. The 4ar/r colony of the same species was used as an external reference [23].
We analyzed Single Nucleotide Polymorphisms (SNPs) in 10 immune-related genes spread across the A. gambiae genome (Table S2). The results showed that the mean nucleotide diversity (π) in the Yaoundé strain (π = 0.0081) was only slightly lower than in field mosquitoes (π = 0.0095), and that the difference in diversity across loci was not significant (Mann-Withney U test, P>0.57). In contrast, the 4ar/r strain revealed extremely limited mean diversity (π = 0.0005), significantly lower than the diversity of field collected mosquitoes (P<0.001). The divergence between field and Yaoundé mosquitoes was also significant (P<0.001) but relatively low (Fst = 0.1202, Da = 0.0016); the divergence between 4ar/r and field mosquitoes was almost 4-fold more pronounced (Fst = 0.4697, Da = 0.0055, P<0.001). Therefore, despite some differences, the mosquitoes used in our study are a reasonable approximation of the local A. gambiae population.
Blood samples were donated by 23 naturally infected gametocyte-carrier volunteers and used to infect Yaoundé mosquitoes in a series of experiments as explained below. In the first experiment we assayed whether the gametocyte density in the blood (gametocytaemia) affects the density of mosquito infection, and thus the subsequent gene expression profiles and our gene silencing experiments. Three to five-day old female mosquitoes were allowed to feed via a membrane on infected blood. Non-blood-fed mosquitoes were removed 24h later, and the mean oocyst density (oocysts per midgut) and infection prevalence (% midguts exhibiting at least one oocyst) were determined 8–10 days post infection (Table 1). Gametocytaemia could not be confirmed in two of the experiments (infections 1 and 18; N/C in Table 1), which were thus excluded from the analysis. To investigate the relationship between gametocyte and oocyst densities we fitted a linear model using oocyst density as the response variable and gametocyte density as the explanatory variable. The oocyst density was log-transformed so that its distribution would better resemble a normal distribution. Indeed, a significant slope coefficient (P<0.05) for the gametocyte density was detected, revealing a correlation between input gametocyte and output oocyst numbers. However, residual analysis revealed that the fit of the model was suboptimal (R2 = 0.029); further investigation of the relationship between the two variables would be interesting but is beyond the scope of this study.
To investigate the relationship between the gametocyte and mosquito infection prevalence, a logistic regression model was fitted using data obtained from infections 2 to 17 and 19, excluding the two infections with N/C gametocytaemia and no detected oocysts, as shown in Table 1. In this analysis, the mosquito infection prevalence (absence or presence of infection) was used as the response variable and the gametocyte density as the explanatory variable. No significant association was detected between input gametocyte density and infection prevalence (coefficient P = 0.63586).
We used blood donated by the P. falciparum gametocyte carriers, to infect mosquitoes and profile the expression of the five candidate genes. Two to three independent biological replicates were performed for each gene; each replicate used a pool of 30–50 mosquitoes and blood from a different carrier. As a control for each replicate, we used 30–50 mosquitoes of the same batch as above, fed on blood originating from the same carrier but depleted of gametocytes by exposure to 42°C. The expression levels of each gene were assessed by quantitative real time RT-PCR (qRT-PCR) in the midgut and carcass (tissues remaining after midgut dissection) at two time periods after blood feeding. T1 (1–3 h) represented the pre-ookinete time period, including gamete fertilization and zygote production in the gut lumen of infected mosquitoes, and T2 (22–25 h) corresponded to ookinete invasion of the midgut epithelium. The replicate results were averaged, and the mean expression levels are presented in Figure 1. In parallel experiments, we used the same design to assess the gene expression levels of the five genes in Yaoundé mosquito infections with P. berghei. In this case, the mosquitoes were infected by feeding on mice bearing the ANKA 2.34 strain of P. berghei, whereas their respective controls were fed on mice bearing the non-gametocyte producing ANKA 2.33 strain. In the P. berghei experiments, the number of replicates ranged from 2 to 4 for different genes (Figure 1). The equivalent designs allowed us directly to compare gene expression between human and rodent malaria infections.
Importantly, with either parasite species, and all candidate genes except ApoIII detected higher expression in infected mosquito midguts (especially at T2 when midgut invasion takes place), as compared to their uninfected controls (Figure 1). Despite substantial quantitative variation between biological replicates the trend of induction in infected vs. control samples was highly consistent. This variation between replicates and differences in induction levels between P. falciparum and P. berghei infections may be due to the differences in infection densities: the geometric mean of oocyst densities were 0.7, 1.1 and 10.0 in the three P. falciparum replicate infections, and ranged between 8.6 and 19.8 in the four P. berghei infections. ApoII/I and ApoIII also showed infection-independent temporal induction in the carcass, consistent with the known origin of apolipoproteins in the fat body.
To examine the effect of silencing the five candidate genes on P. falciparum infectivity, we used blood from the gametocyte carriers to infect Yaoundé mosquitoes. For each gene, two groups of over 50 freshly emerged adult female mosquitoes taken from the same rearing culture were randomly apportioned to small cages. One, the experimental group, was injected with double-stranded RNA (dsRNA) corresponding to the examined gene and the other (control) group was injected with dsRNA of the LacZ gene as described previously [5],[24]. After injection, the two groups were housed identically to eliminate all possible confounding factors. Three to four days later, both groups were allowed to feed via a membrane on the same infected blood source, and the oocyst density was determined at day 8–10 post blood feeding (Table S3). Three to five independent biological replicates were performed for each gene. Each replicate used a different mosquito batch and blood from a different carrier. The silencing efficiency was estimated by qRT-PCR in whole mosquitoes in at least three replicates and averaged; it ranged from 81.4% for ApoIII to 51.3% for CATHB (Figure S1A). However, as revealed for ApoI and ApoII, silencing is often higher at the protein than at the RNA level (Figure S1B, C).
The oocyst density data from all replicates were log-transformed to achieve normality and analyzed by the Residual Maximum Likelihood (REML) variance components analysis by fitting a mixed effect model. In this analysis, we treated the kd-control status as a fixed effect and introduced a random effect for the biological replicate (Figure 2A and Table S3). The difference in infection prevalence between gene kd and control mosquitoes was analyzed using the Chi-square goodness-of-fit test. Relative to their matched controls, WASP gene silenced (kd) mosquitoes showed drastic enhancement of P. falciparum infection prevalence (80.9% vs. 44.9%; P<0.001) and a highly significant increase in oocyst density (3.7 fold increase; P<0.001). In sharp contrast, the ApoII/I kd uniquely decreased the P. falciparum oocyst density (−1.6 fold, P<0.001); this silencing also reduced mosquito fitness as it blocked egg development, a phenotype consistent with the proposed function of ApoII/I as the major lipid carrier in the mosquito hemolymph [25]. Silencing the other three genes did not have any significant effect on P. falciparum infection prevalence or oocyst density. ApoIII silencing also had no effect on mosquito egg development, despite its known involvement, together with ApoII/I, in the insect lipophorin.
A parallel analysis of P. berghei infection of Yaoundé mosquitoes showed that silencing WASP and ApoII/I (but not the other three genes) significantly affects infection loads, in the same direction as for P. falciparum (Figure 2B and Table S3). The WASP kd strongly increased P. berghei oocyst density 2.5-fold (P<0.001), whereas ApoII/I kd reduced the density 4.7-fold (P<0.001) and also limited the prevalence, from 85.3% to 61.2% (P<0.001). Earlier, we reported similar results for P. berghei in G3 strain mosquitoes [5].
Meta-analysis of the standardized mean difference of the various biological replicates fully corroborated the above results, confirming the agonist nature of ApoII/I and the antagonist nature of WASP, in both P. falciparum and P. berghei infections. The forest plots for the meta-analysis are shown in Figure S2A.
To infer the parasite stage affected by ApoII/I kd, we investigated the effect of this gene on P. berghei survival in the parasite-refractory L3-5 strain of A. gambiae [11]. L3-5 mosquitoes kill and subsequently melanize or lyse ookinetes as they complete midgut invasion and encounter the hemolymph filtrate in the basal subepithelial space [1],[6]. Therefore, these mosquitoes can be used as a tool to examine the temporal (and spatial) effect of the examined gene against Plasmodium: a decrease in the numbers of melanized ookinetes in ApoII/I-depleted L3-5 mosquitoes would suggest that ApoII/I functions prior to or at the completion of ookinete invasion, whereas no change in the number of melanized parasites would indicate an effect at a later stage. Analysis of the results as described above (REML variance component analysis and Chi-square goodness-of-fit test) revealed a significant 2.7-fold decrease (P<0.001) of the mean melanized parasite density in ApoII/I kd mosquitoes (Figure 3A, 3B and Table S3), as well as a marked decrease (23.9%; P<0.01) of the infection prevalence (Table S3). These results were independently confirmed by meta-analysis of the standardized mean difference of the various experimental replicates (Figure S2B). Living oocysts were not detected and complete disruption of egg development was again observed (data not shown).
The results presented above (Figure 2) clearly disconnect the functions of the A. gambiae ApoII/I and ApoIII in the response to Plasmodium infection and in mosquito egg development. We examined this disconnection further using the L3-5 mosquito infection assay. Indeed, in contrast to ApoII/I, depletion of ApoIII in such mosquitoes increased drastically (2.7-fold, P<0.001) rather than decreased the density of melanized parasites in the midgut (Figure 3A, 3B and Figure S2; and Table S3), further indicating that the two genes have different functions. Since we did not detect a similar significant increase of live oocyst density in ApoIII-depleted Yaoundé mosquitoes (see Figure 2), we hypothesize a role for ApoIII in parasite melanization. This is corroborated by our recent unpublished data showing an inhibitory effect of ApoIII on the prophenoloxidase activation cascade.
The ApoII/I kd data in L3-5 and Yaoundé mosquitoes suggest that this molecule facilitates P. berghei survival either at the pre-ookinete or at the ookinete stage. Another plausible explanation of the L3-5 phenotype could be that ApoII/I is also directly involved with the melanization cascade in these mosquitoes, but as a positive regulator. We tested and excluded this hypothesis on the basis of genetic epistasis experiments that examined the effect of ApoII/I silencing in another genetic background, CTL4 kd Yaoundé mosquitoes (Figure 3C and S2C and Table S3). CTL4 kd leads to direct melanization and subsequent killing of midgut-invading ookinetes [6]. Concurrent silencing of CTL4 and ApoII/I led to a 2-fold drop in the total parasite load (melanized and not), as compared to the load in CTL4 kd alone (P<0.05; Figure 3C). Therefore, the ApoII/I kd apparently affects survival of parasites before they reach the basal subepithelial space where they become melanized. No significant difference was observed in the proportion of melanized ookinetes to live oocysts between CTL4 kd and CTL4/ApoII/I dkd (double knockdown) mosquitoes, further suggesting that ApoII/I is not involved in the melanization reaction per se.
Interestingly, the total parasite numbers in CTL4 kd mosquitoes (Table S3) were significantly higher than in their controls (2.7-fold, P<0.001). Rather than suggesting a novel role for CTL4, we favor the interpretation that a large number of dead ookinetes are melanized in the absence of the melanization inhibitor CTL4, instead of undergoing lysis.
Previous studies have shown that dead ookinetes of the PbGFPCON transgenic parasite line rapidly loose GFP fluorescence, but continue to display the ookinete surface protein P28 until the early oocyst stage [1],[6]. P28 antibody staining and confocal microscopy of infected midguts was used to assess the proportion of live to dead parasites in control (LacZ dsRNA-treated), WASP kd and ApoII/I kd Yaoundé mosquitoes, at various times after the infected bloodmeal (Figure 4). At day 1 only 17% of the ookinetes were alive (GFP and P28 positive) in the ApoII/I kd compared to 28% in the control mosquitoes. Chi-square goodness-of-fit test established the significance of this difference between observed and expected values (P<0.001, x2 = 318.62). The effect was maximal at early day 2, 32–36 h (P<0.001, x2 = 347.25) when only 10% of the parasites were alive in ApoII/I kd compared to 25% in control midguts. A less pronounced difference was observed at late day 2, 44–48 h (P<0.05, x2 = 5.02), when 14% of the parasites (including the newly formed oocysts) were alive in ApoII/I kd compared to 20% in control midguts. No difference was observed at day 3. These results suggest that ApoII/I kd affects the ookinete during midgut invasion and also the early oocyst stages.
The kinetics were different in WASP kd mosquitoes, where dead ookinetes (GFP negative) were already 63% of the total on day 1 vs. 72% in controls (P<0.001; x2 = 143.13); this effect disappeared at 32–36 h, day 2 and day 3. These data suggest that WASP acts at the level of midgut penetration by ookinetes.
The A. gambiae ApoII/I precursor gene is expressed in the mosquito fat body. After cleavage of its 26 aa signal peptide, the protein is secreted into the hemolymph where it is proteolytically processed to release ApoII (688 aa) and ApoI (2618 aa) [25]. We raised monoclonal antibodies against two peptides (residues 59–70 and 3265–3276) targeting ApoII and ApoI, respectively. Western blot analysis showed that these antibodies recognize abundant hemolymph proteins of the respective size (Figure S3). The anti-ApoI antibody detected additional minor bands of lower molecular weight consistent with previous observations in Manduca sexta, that Apolipophorins are susceptible to proteolytic cleavage [20].
We used the antibodies to examine whether ApoI and/or ApoII bind to P. berghei ookinetes or oocysts in infected mosquito midguts. Immunostaining followed by confocal microscopy only detected a weak signal due to non-specific secondary antibody binding (data not shown).
In light of the complete disruption of mosquito egg development in the ApoII/I kd, we stained mosquito eggs with ApoI and ApoII antibodies at days 1, 2 and 3 after a bloodmeal. The staining detected both proteins in the apical area but not in the rest of the cytoplasm of the follicular epithelial cell layer (Figure S3), or within the oocyte and the nurse cells. The apical epithelial staining confirms that our antibodies can indeed recognize ApoII/I native proteins. Taken together, the immunostainings are consistent with a shuttling function of ApoII/I, a complex which is thought to transport lipids from the hemolymph to the follicular epithelium, where the lipid is internalized for storage in the egg while the lipoproteins are released back into the circulation.
In recent years, molecular interactions between mosquitoes and malaria parasites have received considerable attention, as vector mosquitoes represent a major bottleneck in the malaria transmission system. Such interactions have been studied mainly in laboratory models, and have identified several mosquito genes, often immune-related, which affect the infection outcome, positively or negatively. Studies involving the major vector of human malaria, A. gambiae, have been of special interest, but mostly utilized rodent parasites, or laboratory cultures of the human, P. falciparum parasite. In natural, sympatric parasitism systems, immune interactions may be shaped by selection acting on the vector, the parasite, or both and, as such, these effects may be specific to the parasite species. Indeed, the limited studies on interactions between geographically related A. gambiae/P. falciparum have suggested considerable differences between naturally interacting populations and laboratory models [11],[12],[26]. The present study begins to explore the significance of genes that do not belong to the classical immune repertoire, but have been implicated in local epithelial or systemic responses of A. gambiae to P. berghei.
Our study area was a rain forest locale with continuous malaria transmission and exposure peaking during rainy seasons, April to mid-June (when the study was conducted), or September to late November. To by-pass the daunting difficulties of direct experimentation under natural transmission conditions, we have adopted an approach where field isolates of P. falciparum are used to infect a geographically related laboratory colony (Yaoundé) of M molecular form A. gambiae such as 4ar/r. Although this colony was established in 1988 [21], it retains considerable genetic diversity, substantially higher than typically inbred laboratory strains of A. gambiae. Retention of genetic diversity may be due to the population size of this strain which has always been maintained at a high level, thus mitigating unintended selection processes.
While genetic divergence between the Yaoundé strain and local field collected mosquitoes is statistically significant, this divergence is limited compared to what is commonly observed between intraspecific populations and laboratory strains (data herein and in [22]). In conclusion, the Yaoundé strain is slightly diverged from the local population in the area of study, but it is still highly polymorphic and reasonably representative of the natural M form A. gambiae population. Moreover, recent data (A. Cohuet, unpublished data) have revealed that P. falciparum infection levels are highly similar between the Yaoundé strain and a new strain colonized some months ago also from local populations (N'gousso strain, I. Morlais, unpublished). Thus, it appears that the Yaoundé strain has not diverged significantly in respect to susceptibility to P. falciparum infection, despite absence of contact with P. falciparum for a number of generations.
Transcriptional responses of a mosquito gene to a parasite are considered indicative of a role during infection, although constitutively expressed genes may also be implicated in vector/parasite interactions. Previous studies identified drastic differences and limited similarities between mosquito transcriptional responses to P. berghei vs. laboratory [2] or field isolates of P. falciparum [12],[26]. These studies mostly focused on the mosquito immune system and differences were interpreted as due to A. gambiae/P. falciparum co-adaptation. Guided by our detailed microarray study of midgut responses to P. berghei [5], we focused here on five genes, of which only one belongs to the mosquito immune repertoire (KIN1). We report that all but ApoIII are upregulated during midgut invasion by both P. berghei and P. falciparum field isolates. Although some induction is evident in the respective controls at day-1 after a bloodmeal with non-infectious parasites, upregulation is stronger in the presence of ookinetes. Both Apolipophorin genes are also induced in the carcass in a parasite-independent manner, a response consistent with their putative function in lipid transport.
ApoI and II are cleaved from a common ApoII/I precursor and are integral components of the mosquito lipophorin, a versatile and reusable shuttle system for lipid transport [27]. Lipophorin transports dietary lipids from the gut via hemolymph to storage sites such as the fat body, muscles, ovaries and other tissues [28]. It also binds and delivers lipid-linked morphogens and glycophosphatidylinositol (GPI)-linked proteins to target cells of developing embryos [29]. These functions are consistent with the apical localization of ApoII/I protein at the follicular epithelium, and the prevention of ovarian maturation by ApoII/I depletion. ApoIII, the third polypeptide in lipophorin, is thought to counterbalance the increased hydrophobicity due to lipid binding and thus may stabilize lipophorin particles. However, it is evidently dispensable for lipid transport to the ovaries, as silencing this gene has no effect on follicle maturation.
In contrast, ApoII/I is an exemplar protein that benefits both the mosquito vector and the parasite: its depletion compromises egg production, ookinete invasion of the midgut and early oocyst development. The importance of this gene is highlighted by its requirement for development of the avian [30] as well as human and rodent [5] Plasmodium parasites. Indeed, ApoII/I is the first mosquito gene with a confirmed positive role in development of geographically-related P. falciparum field isolates. It validates the concept of mosquito agonists for human as well as rodent and avian parasites, and may prove to be a universal Plasmodium agonist.
Future studies are required to establish the mechanism whereby this important lipoprotein promotes ookinete migration and development. Immunostaining did not detect direct binding to the parasite. An attractive hypothesis is that ApoII/I sequesters lipids from the hemolymph and releases them to the developing oocyst where sporogonic proliferation and massive membrane formation occur. Indeed, in ApoII/I kd mosquitoes some early oocysts appear to be arrested in development and may be targeted for destruction. This documented positive effect of ApoII/I on oocysts does not explain its earlier effect on ookinete invasion. ApoII/I probably has multiple functions. It may help rescue and release ookinetes as a side effect of lipid mobilization from the midgut epithelium, where suggestive massive droplets appear at the basal side where ookinetes emerge [31]. Thus depletion of ApoII/I might trap ookinetes in the toxic environment of the invaded midgut. ApoII/I is also immune-induced in Drosophila hemolymph [32], detected in clotting assays [33], and present in immune-activated haemocytes [34]. Furthermore, in the fat body of Ae. aegypti, ApoII/I is regulated by the Toll/Rel1 immune pathway [30]. Thus, it may also be involved in systemic non-classical immunity.
ApoIII strongly resembles the N-terminal domain of human Apolipoprotein E which is involved in lipid transport, lipopolysaccharide detoxification, phagocytosis and pattern recognition [20]. Although ApoIII is a known partner of ApoII/I in the insect lipophorin, its depletion does not affect Plasmodium development in susceptible mosquitoes, but does increase drastically the density of melanized P. berghei parasites in the L3-5 refractory strain. These data together with our recent unpublished data showing increased prophenoloxidase activity in ApoIII-depleted mosquito hemolymph suggest that ApoIII may interfere with the melanization reaction itself. It is possible that a number of killed but not melanized parasites in the L3-5 mosquitoes are melanized after depletion of the putative melanization inhibitor ApoIII. This hypothesis is consistent with a report that ApoIII in Galleria mellonella dampens activation of the prophenoloxidase cascade by Bacillus subtilis lipoteichoic acid [35], although another study detected an opposite effect [36].
In contrast to ApoII/I, WASP is a parasite antagonist, as its local induction in the midgut reduces mosquito infection by P. berghei and more so by P. falciparum. Similarly, ApoII/I is induced more strongly in P. falciparum infected midguts, but protects P. berghei better. The levels of induction of WASP and ApoII/I do not directly correlate with their functional potency. Whether such discrepancies reflect co-adaptation of naturally interacting species or other evolutionary processes remains to be determined.
WASP is thought to play a key role in actin cytoskeleton rearrangements in epithelial cells. Our previous in vivo imaging analysis of midgut invasion by P. berghei revealed extensive actin-based motility of the damaged epithelium and an actin-rich structure surrounding ookinetes as they exit the epithelial cell layer [37]. A similar fibrillar organelle-free structure was observed previously around P. gallinaceum ookinetes in a refractory, lytic strain of A. gambiae [38]. Depletion of positive regulators of actin polymerization, such as WASP, increases P. berghei density, whereas depletion of negative regulators decreases the density; therefore we proposed that this actin-rich structure which we named “parasite hood” might be a defense reaction against invading parasites [5],[17]. Recently another study renamed this structure “organelle-free actin zone” and reported that it is required for the clearance of dead parasites [39]. However, this conclusion cannot explain our observed differences in live parasite density that follow silencing of actin cytoskeleton regulators. The formation, regulation and role(s) of this actin-based structure require further investigation. Although WASP silencing has a similar effect on the P. falciparum infection load, no conclusive evidence has been reported to date as to whether P. falciparum is also associated with a hood or causes damage to the invaded epithelium similar to that reported for P. berghei [16].
P. falciparum is the deadliest human parasite. Its association with A. gambiae exacts a devastating toll in Africa. Through several thousand years of A. gambiae/P. falciparum co-evolution, the parasite apparently adapted to reduce the burden on the vector by limiting infection intensity [13],[40], while securing successful transmission to humans. Presumably, the vector also adapted to reduce infection loads thus avoiding the fitness cost of immune system activation [13],[41]. Indeed, recent studies showed that several A. gambiae genes act as positive or negative regulators of immune reactions against P. berghei (a parasite that this mosquito has never encountered in nature), but do not affect sympatric P. falciparum infections [12],[42]. During co-evolution, parasites may also have developed specific mechanisms to modulate activation levels of the vector immune system. These diverse possibilities merit further analysis.
The present study, encompassing both high and low infection densities, demonstrates that genes implicated in local epithelial or systemic (but non-classical immune) responses can have similar effects but different activity levels against two different parasite species. These reactions may have some margin for adjustment in sympatric vector/parasite combination, but adjustment is probably limited for responses that are essential for the vector, e.g. to reconstitute the damaged midgut epithelium after parasite invasion (WASP) or to support reproduction (ApoII/I). Therefore, unlike widely adjustable immune responses that require investigation in natural interacting species, essential vector responses may be studied conveniently in model vector/parasite systems. Moreover, such conserved, robust and not widely adjustable interactions may be ideal for development of novel malaria control strategies.
The Yaoundé colony was originally established at OCEAC, Cameroon, from a population of A. gambiae s.s. caught in a quarter of Yaoundé and adapted to feeding on parafilm membrane feeders [21]. This colony belongs to the M molecular and Forest chromosomal forms (standard chromosomal arrangement). Yaoundé, 4ar/r [23] and refractory L3-5 [11] mosquitoes were cultured in the insectary using standard methods. For the SNP analysis, A. gambiae larvae were collected in Simbock (03°51′N, 11°30′E), a South Cameroon village near Yaoundé and reared in the insectary until adult emergence.
DNA was isolated from legs of 8 adult females from each of the laboratory colonies (Yaoundé and 4ar/r) and from the field collected M form A. gambiae, as described by Morlais et al. [43]. The M molecular form mosquitoes were distinguished from S form mosquitoes by a PCR assay [44]. PCR primer pairs for 10 immune related genes [45] were designed using Primer3 (http://frodo.wi.mit.edu/cgi-bin/primer3/primer3_www.cgi). The PCR reactions and PCR primers are presented in the Protocol S1 and in Table S4. Both DNA strands of PCR products were sequenced using an the Applied Biosystems 3730 sequencer, assembled and verified using SeqScape (Applied Biosystems).
Sequence alignments were performed using ClustalW in MEGA 3.1 [46]. Polymorphism analyses and molecular population genetic test statistics were calculated using DnaSP 4.10 [47]. The nucleotide diversity within each population was estimated as the average pairwise nucleotide difference per site (π). Divergence between the wild populations and laboratory strains was assessed by sequence-based F statistics (Fst), analogous to Wright F statistics [48], which was calculated according to [49]; the net genetic distance was measured using Da [50].
Parasites were detected by microscopic examination of Giemsa-stained thick films of blood taken from volunteers by finger pricking. Gametocyte density was estimated assuming a standard number of 8000 WBC/µl of blood and by counting visible gametocytes against 1000 WBC. Children with asexual parasitaemia exceeding 1000 parasites/µl were treated with amodiaquine and artesunate combination according to national guidelines. Asymptomatic gametocyte-positive children were enrolled as volunteers, following procedures approved by the Cameroonian and WHO ethical review committees.
For the expression profiling experiments, two groups of mosquitoes from the same rearing culture were used per replicate: the test group fed on blood donated by a gametocyte carrier, whereas the control group fed on the same blood which was previously incubated at 42–43°C for 12 min under constant shaking at 500 rpm for gametocyte inactivation. For the gene silencing experiments, we also used two groups of freshly emerged female mosquitoes per gene and per replicate, also separated randomly from the same rearing culture in small containers. The first group was subjected to silencing of the examined gene whereas the second control group was injected with dsRNA of the LacZ gene. In both cases, the two groups were housed under the same microclimate and treated identically, both before and after the blood feeding.
Mosquitoes were allowed to feed via a membrane on blood donated by P. falciparum gametocyte carries. To eliminate transmission blocking immunity factors, the carrier serum was replaced by non-immune AB serum [51]. Blood samples (700 µl each) were transferred into pre-warmed (37°C) artificial membrane feeders and exposed to mosquitoes that were previously starved for 12 hours, according to standard procedures [21]. To determine the levels of infection, mosquito midguts were dissected 8–10 days post blood feeding and stained with 2% mercurochrome before microscopic examination.
Three P. berghei clones, the gametocyte-producer ANKA 15cy1A (2.34), the non-gametocyte-producer ANKA 15cy1A (2.33) and the GFP-expressing PbGFPCON strain [52], were used for the various mosquito infections. Similar to what is described in the previous paragraph, two mosquito groups were used in each of the expression profiling and gene silencing experiments. For expression profiling, the test group fed on mice infected with the ANKA 2.34 strain and the control group fed on mice infected with the non-gametocyte producing ANKA 2.33 strain. In the gene silencing experiments, the two groups were treated as described for the P. falciparum infections and in [5], and fed on mice infected with the PbGFPCON strain. The infections were performed as described [53].
Production of dsRNA and mosquito gene silencing was performed as described previously [5],[24].
QRT-PCR expression profile analysis of mosquito genes were performed as described previously [5]. Briefly, total RNA was extracted using the TRIZOL reagent (Invitrogen) from 30–50 mosquitoes for expression profiling and from 10 adult female mosquitoes to determine the gene silencing efficiency. For the expression profile analysis, two to four independent biological replicates were performed, which used different batches of mosquitoes fed on different blood sources (different gametocyte carriers for the P. falciparum infections and different infected mice for the P. berghei infections). The results of each biological replicate were the average of two technical replicates, in which the same RNA samples were processed in duplicate in the same qRT-PCR plate. The A. gambiae S7 ribosomal gene was used as an internal control to normalize the amount of RNA between the various samples (i.e. between kd and control mosquitoes). For assessment of the gene silencing efficiency, dsRNA-injected mosquitoes (control and kd) were collected before blood feeding. Gene-specific primers for ApoII/I, WASP, CATHB and KIN1 were used for qRT-PCR and dsRNA as previously described by [5]. The qRT-PCR and dsRNA primers for ApoIII were: forward, 5′-GCCGTGCAGGGAAGCTT-3′; reverse, 5′-ATCTTGTCCTTGATGCTCATGA-3′ for qRT-PCR and forward, 5′-TAATACGACTCACTATAGGGTCCAGTCGATCATGAGCATCA-3′; reverse, 5′-TAATACGACTCACTATAGGGAGCTTCTTGAGCGCGTCCT-3′ for dsRNA production.
Peptides corresponding to the A. gambiae ApoI N-terminal sequence (+HN-CGNYAQQKTPKDKKQ-COO−) and ApoII C-terminal sequence (+HN- CGLQQSDKENKQT-COO−) were synthesized (ams biotechnology) and used by the EMBL Monoclonal Antibody Facility to immunize mice and thereby generate hybridoma cell-lines.
Hemolymph was obtained by clipping the proboscis of female A. gambiae mosquitoes and collecting a hemolymph droplet in a pipette tip filled with reducing SDS loading buffer (0.25M Tris pH6.8, 40% Glycerol, 8% SDS, 8% β-mercaptoethanol). Proteins were resolved by discontinuous SDS-PAGE, with 5% stacking and 8% resolving gel, and subsequently transferred to Hybond-P membranes (Amersham Biosciences) in a Trans-Blot SD Semi-Dry Transfer Cell (BioRad) blotting chamber. Unspecific antibody binding was reduced by blocking the membranes overnight (o/n) at 4°C in blocking buffer (1% Tween and 3% Milk powder in PBS). Supernatants of hybridoma cell lines containing primary antibodies were diluted 1∶100 in blocking buffer, and membranes were incubated o/n at 4°C. Anti-mouse IgG conjugated to horseradish peroxidase (Promega) was used as a secondary antibody at 1∶15000 dilution in blocking buffer; incubation was performed for 1–3 h at RT. Blots were developed using Western Lightning Chemiluminescence Reagent Plus Kit (PerkinElmer Life Sciences).
Mosquito midgut epithelia and ovaries were dissected in ice-cold PBS and fixed in 1×PBS pH 7.2, 4% Formaldehyde, 1 µmol EGTA, 2 µmol MgSO4. After washing (PBS at RT for 3×15 min) and blocking (midguts: 1× PBS pH7, 0.1% Triton X-100, 1% BSA; ovaries: 1× PBS pH7, 0.2% Saponin, 1% BSA), the tissues were incubated with monoclonal antibodies against ApoI (1∶100) and ApoII (1∶100) followed by incubation with Alexa-647-conjugated goat anti-mouse antibodies (Molecular Probes) at 1∶1,500 for 90 min at RT. Tissues were washed and mounted on microscope slides using VECTASHIELD Mounting Medium with DAPI (Vector Labs). For negative controls, tissues were incubated in blocking buffer without antibodies. For parasite killing assays, infected midguts were incubated with a Cy-3-tagged monoclonal antibody against P. berghei P28 at 1∶750 (a kind gift of R. E. Sinden). Mounted tissues were observed on a Leica SP2 or Leica SP5 confocal microscope or Leica DMR microscope, respectively. Image processing was performed with the ImageJ v1.36b software.
The oocyst density data were log-transformed [log10 (n+1)] so that their distribution resembles a normal distribution. A linear model was used to examine the relationship between the oocyst densities and gametocyte densities, in which the oocyst density was the response variable and the gametocyte density was the explanatory variable. The correlation between the gametocyte and mosquito infection prevalence was investigated by fitting a logistic regression model, where the absence or presence of infection was used as the response variable and the gametocyte density as the explanatory variable. In the RNAi experiments, the oocyst density data were analyzed by the REML variance components analysis by fitting a mixed effect model. The kd-control status was treated as a fixed effect and we introduced a random effect for the biological replicate. For each dataset a combined P-value is reported for the fixed effects. The difference in the infection prevalence between kd and their control mosquitoes were analyzed using the Chi-square goodness-of-fit test where the observed values were fitted to expected values. Similarly, the same test was used to examine the difference in the proportion of live/dead parasites between gene kd and their control mosquitoes in the killing assays. All the above statistical tests were performed using the GenStat software.
Meta-analysis of the standardized mean difference of the various biological replicates in the gene silencing experiments, and calculation of the total standardized mean difference with 95% Confidence Interval were performed by using the Comprehensive Meta-analysis software (Biostat, version 2). This analysis uses two models, fixed and random, and the total standardized mean difference is given both for the fixed and the random effect model.
Ensembl accession numbers for reported genes and proteins are as follows: ApoII/I (AGAP001826-PA, AGAP001826); WASP (AGAP001081-PA, AGAP001081); KIN1 (AGAP005888-PA, AGAP005888); CATHB (AGAP007684-PA and AGAP007684-PB, AGAP007684); ApoIII (ENSANGESTG00000005174); CTL4 (AGAP005335-PA, AGAP005335); CEC2 (AGAP000692-PA, AGAP000692); SCRB10 (AGAP000016-PA, AGAP000016); STAT2 (AGAP000099-PA, AGAP000099); SRPN11 (AGAP001377-PA, AGAP001377); GNBPB2 (AGAP002729-PA, AGAP002729); PPO9 (AGAP004978-PA, AGAP004978); LRIM1 (AGAP060348-PA, AGAP060348); TEP15 (AGAP008364-PA, AGAP008364); TEP4 (AGAP010812-PA, AGAP010812); TOLL10 (AGAP011187-PA, AGAP011187). |
10.1371/journal.pgen.1002539 | Rapid Analysis of Saccharomyces cerevisiae Genome Rearrangements by Multiplex Ligation–Dependent Probe Amplification | Aneuploidy and gross chromosomal rearrangements (GCRs) can lead to genetic diseases and the development of cancer. We previously demonstrated that introduction of the repetitive retrotransposon Ty912 onto a nonessential chromosome arm of Saccharomyces cerevisiae led to increased genome instability predominantly due to increased rates of formation of monocentric nonreciprocal translocations. In this study, we adapted Multiplex Ligation–dependent Probe Amplification (MLPA) to analyze a large numbers of these GCRs. Using MLPA, we found that the distribution of translocations induced by the presence of Ty912 in a wild-type strain was nonrandom and that the majority of these translocations were mediated by only six translocation targets on four different chromosomes, even though there were 254 potential Ty-related translocation targets in the S. cerevisiae genome. While the majority of Ty912-mediated translocations resulted from RAD52-dependent recombination, we observed a number of nonreciprocal translocations mediated by RAD52-independent recombination between Ty1 elements. The formation of these RAD52-independent translocations did not require the Rad51 or Rad59 homologous pairing proteins or the Rad1–Rad10 endonuclease complex that processes branched DNAs during recombination. Finally, we found that defects in ASF1-RTT109–dependent acetylation of histone H3 lysine residue 56 (H3K56) resulted in increased accumulation of both GCRs and whole-chromosome duplications, and resulted in aneuploidy that tended to occur simultaneously with GCRs. Overall, we found that MLPA is a versatile technique for the rapid analysis of GCRs and can facilitate the genetic analysis of the pathways that prevent and promote GCRs and aneuploidy.
| In this study we describe an adaptation of Multiplex Ligation–dependent Probe Amplification (MLPA) for use in the study of gross chromosomal rearrangements (GCRs) that occur in S. cerevisiae mutants with increased genome instability. Our previous study found that the presence of a Ty912 element on a nonessential arm of chromosome V resulted in increased rates of non-reciprocal monocentric translocations arising from recombination between the Ty912 on chromosome V and ectopic Ty elements on other chromosomes. Using MLPA, we observed that the majority of the translocations targeted six different translocation hotspots even though there were at least 254 potential targets for Ty-mediated translocations in the S. cerevisiae genome. Most of the observed translocations were formed by RAD52-dependent recombination, although we also identified a RAD52-independent recombination pathway that promoted the formation of the same types of translocations at lower rates. Finally, we found that defects in the ASF1-RTT109–dependent histone H3 lysine 56 (H3K56) acetylation pathway caused increased rates of both Ty-mediated translocations and whole-chromosome duplications (aneuploidy). This aneuploidy often occurred simultaneously with Ty-mediated translocations. Overall, our results demonstrate that MLPA is a rapid, inexpensive method that allows the analysis of the large number of GCRs needed to understand the pathways that suppress or promote genome instability.
| Genome stability is important for normal cellular survival and growth. In contrast, genome instability is associated with abnormal cellular growth. For example, tumor cells often contain multiple genome rearrangements and/or exhibit aneuploidy, and such events are thought to contribute to the development and progression of cancer [1]–[4]. Genome rearrangements are also associated with inborn genetic diseases. For instance, copy number changes mediated by segmental duplications are associated with a diversity of genetic diseases [5] and whole chromosome aneuploidy can cause diseases like Down Syndrome [6]. While the association of genome rearrangements and aneuploidy with human genetic disease is well established, the genetic factors that suppress or enhance genome rearrangements and aneuploidy are less well understood.
We previously developed quantitative genetic assays for measuring the rates at which GCRs occur in Saccharomyces cerevisiae. These assays, and modified versions of them, select for progeny that lose a nonessential chromosome arm due to a GCR mediated by either non-repetitive [7]–[10], low-copy repeat [9], [11], or high-copy repeat DNA [12] and allow quantitative genetic analysis of the pathways that suppress or promote the formation of GCRs [13]. Genetic studies using these assays have revealed that numerous genes and pathways contribute to genome stability by suppressing the formation of gross chromosomal rearrangements (GCRs) and/or the loss or gain of whole chromosomes [7], [11], [14]–[22]. To fully understand the mechanisms by which GCRs are formed, it is often necessary to determine their structures and sequence their rearrangement breakpoints. However, such analysis of rearranged genomes remains a challenge, particularly due to the large number that must be analyzed to determine the mechanisms by which the GCRs were formed. A number of different techniques have been used to analyze the structure of GCRs including: 1) Pulse Field Gel Electrophoresis (PFGE) to determine the size of rearranged chromosomes [12], [15], [23]–[27]; 2) different methods for PCR amplification and sequencing of rearrangement breakpoints [7], [11], [12], [20]–[23], [26]–[28]; 3) cloning and/or restriction mapping of rearrangement breakpoints [12], [14], [15]; 4) array Comparative Genomic Hybridization (aCGH) that allows for the identification of regions of copy number change but does not provide information about connectivity of the rearranged regions [12], [14], [15], [23], [26], [27]; and 5) next-generation DNA sequencing, which has the potential to provide considerable detail about the structure of genome rearrangements [29]. However, all of these methods fail to scale when analyzing a large number of rearranged genomes, due to either high costs or the tedious natures of the methods. In this study, we adapted a PCR-based method, Multiplex Ligation-dependent Probe Amplification (MLPA), to supplement the analysis of Ty1-mediated GCRs in a manner that scales well in terms of both cost and time.
MLPA is a multiplex ligation-dependent amplification technique that has been used to identify duplications, deletions, and aneuploidy in human cells [30]–[32]. Briefly, multiple pairs of oligonucleotide probes are designed such that each probe in a pair hybridizes next to the other member of the pair at regions of interest in the genome. The total length of each pair of probes is distinct and is used to identify specific regions in the genome on the basis of the length of the final MLPA product. The probes are hybridized to genomic DNA, and then adjacent probes are ligated and amplified using a common pair of fluorescently labeled oligonucleotide primers. Products are separated and their length and fluorescent intensities measured using capillary electrophoresis. Analysis of the fluorescent intensities allows the determination of copy number differences between control and experimental samples. The main advantage of MLPA is its ability to provide copy number variation data for targeted regions in a rapid, inexpensive, and highly parallel manner. While MLPA does not provide the dense genome-wide coverage of aCGH or next-generation sequencing, it can cover multiple regions of interest simultaneously at a density sufficient for many types of genetic studies. In the present study we demonstrate the utility of MLPA for analyzing GCRs by investigating the target site bias of Ty1-mediated GCRs, RAD52-independent formation of Ty1-mediated translocations, and aneuploidy induced by deletion of RTT109, a gene encoding a histone acetyltransferase.
In previous work we demonstrated that insertion of Ty912 between CIN8 and NPR2 in a nonessential terminal region of chromosome V (the +Ty912 GCR assay) (Figure 1a), led to an increase in the rate of accumulating GCRs [12]. When we screened 88 independent GCR-containing strains derived from either wild-type or one of 11 different mutant strains, we found that all (88 of 88) of these GCR-containing strains contained a deletion of chromosome V from Ty912 to TEL05L (the left telomere of chromosome V) and almost all (82 of 88) of these GCR-containing strains also contained a duplicated region from another chromosome arm bounded by an ectopic Ty1, Ty2, or solo delta sequence at one end and a telomere at the other end. Structural studies demonstrated that these duplication-containing strains each contained a translocation consisting of the centromere-containing fragment of chromosome V joined to the duplicated region of the target chromosome, with a junction involving Ty912 and the bounding ectopic Ty element of the duplicated region. 94% of the translocations were simple translocations with a single junction involving Ty912 and a single target ectopic Ty element and 6% appeared to involve a dicentric translocation intermediate that underwent secondary rearrangements. In addition, each GCR-containing strain also contained a wild-type copy of the chromosome from which the duplicated sequence was derived. Overall, this analysis indicated that all of the Ty912-mediated translocations observed were formed by a non-reciprocal recombination-mediated translocation mechanism.
To better characterize the distribution of the observed chromosome arm duplications, we developed a MLPA probe set capable of identifying duplicated and deleted chromosome arms. A “telomeric” probe set was designed to identify copy number changes at genomic loci located at the ends of each chromosome arm in S. cerevisiae. The probes were designed to hybridize between the telomeres of each chromosome and their closest respective Ty1, Ty2, or solo delta element (Figure 1b, 1c); however, four probes (chrII-L, chrIV-R, chrIX-R, and chrXV-L) could not be designed to meet this criterion due to the lack of suitable non-repetitive DNA regions near the telomeres of these chromosome arms, and were instead designed to hybridize immediately centromeric to the terminal Ty elements present at the 4 chromosome ends (Table S1). As a result, the telomeric probe set theoretically detects 98.4% (250/254) of the possible translocation-associated duplications resulting from Ty912-mediated nonreciprocal translocations targeting ectopic Ty-related elements in the S288C reference genome.
Using this telomeric MLPA probe set, we first reanalyzed 5 isolates that were derived from the wild-type +Ty912 GCR assay strain and that were each previously identified by aCGH to contain a translocation chromosome associated with a chromosome arm duplication [12]. The MLPA results concurred with the previous aCGH results and, in each case, identified both the chromosome V-L deletion and the associated chromosome arm duplication identified previously (Figure 1d). We next screened 112 newly isolated independent GCR-containing strains derived from the wild-type +Ty912 GCR assay strain. The MLPA data revealed that all (112 of 112) GCR-containing strains lost the left arm of chromosome V, and almost all (106 of 112) contained a duplicated region of another chromosome arm (Table S2). The remaining isolates either had no detectable duplication (5 of 112) or could not be unambiguously assigned to a rearrangement class (1 of 112).
We calculated a pair of expected distributions for the chromosome arm duplications based on the assumption that Ty912 could recombine with either all annotated ectopic Ty1 elements at equal frequency or all annotated ectopic Ty1 and delta elements (in this analysis, each of the 13 Ty2 elements were included as 2 separate delta elements; see Methods) at equal frequency (Figure 2a, 2b). Analysis of the 106 observed duplicated chromosome arms isolated in the wild-type +Ty912 assay strain (Figure 2c; Table S2) revealed numerous chromosome arm duplications that did not contain any annotated full length Ty1 sequences in the S288c reference sequence (chrIII-L, chrIII-R, chrIX-L, chrXIII-L, chrXIV-R, chrXV-L, and chrXVI-L), indicating that, consistent with our previous results [12], both ectopic Ty1 and delta elements are likely to have mediated the observed chromosome arm duplications. We found the distribution of observed chromosome arm duplications to be significantly different from a theoretical distribution that assumed that all Ty1 and delta elements acted as translocation targets (Monte Carlo Sampling of a Multinomial Distribution; 2000 replicates; Empirical p = 5.00×10−4). Several chromosome arm duplications were significantly overrepresented in the observed distribution compared to the theoretical distribution, including duplications of chrIII-R (28 times; exact binomial; p = 5×10−18), chrV-R (27 times; exact binomial; p = 1×10−9), chrXIV-L (8 times; exact binomial; p = 4×10−5), and chrX-R (10 times; exact binomial; p = 2×10−3). Together these 4 classes of chromosome arm duplications represented 69% (73 of 106) of the observed duplications.
We next sought to identify if there was any bias in this analysis due to the orientation of Ty912 on chromosome V. We and others have previously shown that translocations targeting centromere-oriented Ty elements yield dicentric chromosomes that must undergo secondary rearrangements to eliminate a centromere [12], [16], [23]. Due to this requirement of secondary rearrangements, these events are likely to be under-represented relative to the proportion of recombinations involving telomere-oriented Ty elements. Thus, it is possible that inclusion of centromere-oriented Ty elements in the list of possible recombination targets biased our analysis.. To check for such bias, we compared the observed distribution of chromosome arm duplications to the distribution of potential telomere-oriented Ty1 and delta element translocation targets (Figure S1). The chrIII-R, chrV-R, chrXIV-L, and chrX-R duplications were still significantly over-represented when compared to this theoretical distribution (exact binomials; p = 2×10−13, 1×10−8, 4×10−4, and 1×10−2, respectively). Thus, our MLPA analysis of a large number of GCRs isolated from a single genetic background confirmed the chromosome arm duplication bias we previously noted when we used aCGH to analyze a smaller number of GCR-containing strains isolated from 12 different wild-type and mutant strains [12].
In order to investigate the possibility that specific Ty-related elements were responsible for the observed chromosome arm duplication bias in the wild-type +Ty912 GCR assay, we created a series of MLPA probe sets specific for chrIII-R, chrV-R, chrX-R, and chrXIV-L. These probe sets contained one or two pairs of primers designed to hybridize between every pair of Ty1, Ty2, solo delta, centromeric, and telomeric element along each chromosome arm, except for a few cases of closely collocated Ty loci (Figure 3a–3d; Tables S3, S5, S6, S7).
We first used the chrIII-R MLPA probe set (Figure 3a; Table S3) to analyze the 28 GCRs involving chromosome III right arm duplications. Two patterns of chromosome arm duplications were observed: those in which the chrIII-R probes telomeric, but not centromeric, to the YCRWdelta8-YCRWdelta9-YCRWdelta10 locus (61%; n = 17) were duplicated and those in which the chrIII-R probes telomeric, but not centromeric, to the YCRWdelta11 locus (39%; n = 11) were duplicated (Figure 3a; Table 1). These loci have been previously termed fragile site 1 (FS1) and fragile site 2 (FS2) and were originally identified as hotspots of Ty recombination under conditions of DNA replication stress caused by reduction of the levels of different replicative DNA polymerases [15]. FS1 was found to replace the SRD1 and YCRWdelta8-YCRWdelta9-YCRWdelta10 loci in the S288c annotated sequence with a pair of tandem full length Ty1s oriented in a direct repeat orientation. FS2 was found to contain two Ty1s oriented in an inverted head-head configuration with a 283 bp spacer sequence in between the two Ty1s instead of the single annotated YCRWdelta11 locus. Other studies have also confirmed that these loci show differences from the SGD S288c reference sequence [27], [33]. We confirmed the presence of both the tandem direct repeat pair of full length Ty1s at FS1 and the pair of inverted repeat Ty1s at FS2 in our strains by PCR and sequencing. These results demonstrate that MLPA can be used to identify specific translocation fusion junctions and that FS1 and FS2, even in the presence of normal rates of DNA replication, likely have fragile site activity that results in increased frequencies of Ty-mediated translocations.
We next analyzed the GCR-containing strains associated with other overrepresented chromosome arm duplications (chrV-R, XIV-L, and X-R) using the remaining chromosome arm-specific MLPA probe sets (Tables S5, S6, S7). The MLPA data revealed the existence of Ty-mediated translocation hotspots on each of the chromosome arms (Figure 3b–3d; Table 1). The majority of duplications (18 of 27) of the right arm of chromosome V appeared to target the linked YERWdelta17, YERWdelta21, and YERCTy1-1 loci (note that this region also contains an unannotated partial Ty sequence [12]), with a smaller proportion (5 of 27) of duplications targeting the region containing the linked YERCdelta14, YERCdelta15, and YERCdelta16 loci. This is in agreement with several studies that observed the involvement of the linked YERWdelta17, YERWdelta21, and YERCTy1-1 loci in Ty-mediated repair of DNA double strand breaks (DSBs) [9], [26], including our own prior observation of a translocation in which Ty912 of our assay targeted the centromere-oriented YERCTy1-1 on chrV-R and resulted in a dicentric translocation chromosome. This dicentric chromosome then underwent additional rounds of rearrangements to yield a monocentric translocation that had duplicated the telomeric end of chrV-R [12]. Likewise, 75% (6 of 8) of the chromosome arm duplications involving the left arm of chromosome XIV appeared to target YNLCTy2-1. All of the duplications on the right arm of chromosome X (10 of 10) appeared to involve the YJRWTy1-1/YJRWTy1-2 tandem Ty1 locus. Overall, these results indicate that 63.2% (67 of 106) of the total Ty-mediated chromosome arm duplications identified in the wild-type +Ty912 assay strain could be accounted for by only 6 Ty target regions on 4 chromosome arms, which contrasts with the presence of 254 potential targets in the S. cerevisiae genome.
One possible explanation for the observed Ty target bias is that Ty912 preferentially recombines with regions of high sequence homology. Although this explanation was at odds with the results for chromosome III (Ty912 has 93% average sequence identity with Ty-related elements on chromosome III-L involved in 2 chromosome arm duplications and 77% average sequence identity with Ty-related elements on chromosome III-R involved in 28 chromosome arm duplications), overall sequence identity may not be the appropriate measure of homology that mediates recombination with Ty912, as a previous study examining DSB-induced HR between Ty elements has noted [27]. We therefore investigated the possibility that translocations mediated by recombination between divergent homologous regions are suppressed by mismatch repair [11], [34], [35]. Thus, if sequence homologies drive the target distribution bias seen for Ty-mediated translocations, an msh2Δ mutation that eliminates suppression of homeologous recombination should alter the distribution of target Ty elements and chromosome arm duplications.
An msh2Δ mutation resulted in an approximately 3-fold increase in the rate of Ty912-mediated GCRs (Table 2). All 44 GCRs isolated in the msh2Δ mutant were associated with a chromosome V-L deletion and 43 of these GCRs were also associated with a chromosome arm duplication. In order to look for significant changes between the msh2Δ strain relative to the wild-type strain, we calculated the fold change between the observed duplication rate of each chromosome arm in the msh2Δ mutant relative to an expected duplication rate for each chromosome arm (see Methods). The expected duplication rate assumed that the msh2Δ mutation evenly increased the duplication rates of all chromosome arms by the bulk-fold change between the msh2Δ mutant and wild-type GCR rates and hence the msh2Δ mutation was expected to increase the rate of each duplication 3-fold (Table 2); in other words, we assumed the msh2Δ mutation did not preferentially affect the rate of any specific translocation (see Methods). The duplication rates for each chromosome arm in the msh2Δ mutant were generally within the 95% confidence interval of the predicted msh2Δ rates and were approximately three-fold higher than the wild-type rates, consistent with the idea that the msh2Δ mutation increased both the average GCR rate and the translocation rate of each chromosome arm by 3-fold. The only exceptions were the duplication rates of chromosome arms II-R, XVI-R, and V-R. Chromosome arm II-R and XVI-R duplications were not previously seen among the 112 wild-type GCRs analyzed and occurred at about a five-fold higher rate in the msh2Δ mutant than predicted from the maximum wild-type rate calculated for these 2 sites. The duplication rate of chromosome V-R was approximately four-fold lower than predicted from the wild-type rate (Figure 2c). The duplication rates of all three of these chromosome arms were outside the 95% confidence interval of their predicted duplication rates calculated from the measured msh2Δ bulk rate (p<0.05). In addition, the number of chromosome V duplications observed in the msh2Δ mutant was significantly lower than that seen in the wild-type strain (Fisher's Exact Test; p = 0.0122). However, GCRs associated with duplicated chromosome arms containing the translocation hotspots (chrIII-R, V-R, X-R, and XIV-L) were seen in both the msh2Δ +Ty912 assay strain and the wild-type +Ty912 assay strain in roughly equal proportions (70% vs. 69%). These results indicate that while an MSH2-dependent function may both modestly affect the rate of specific translocations and be potentially important for the formation of translocations involving the chromosome V-R hotspot, mismatch repair is largely not responsible for the biased distribution of translocations in the wild-type +Ty912 assay strain. This supports the idea that the sequence homology relationships between Ty912 and the other Ty elements in the S. cerevisiae genome are unlikely to be the sole, or major, determinant of translocation target site selection.
One possible reason for the fact that six hotspots comprise nearly 70% of the observed chromosome duplications is that the six hotspots lie in close physical proximity to Ty912, as such physical proximity may determine the ease with which homologous sequences can be targeted by HR. Using previous data that mapped the 3-dimensional spatial relationship of all chromosomes in wild-type cells [36], we analyzed whether any of the six observed hotspots as well as the 20+ kb regions containing each hotspot were in close proximity to the region of chromosome V where Ty912 was inserted. Analysis of the Chromosome Conformation Capture-on-Chip (4C) data generated with HindIII indicated that 2 of the 6 hotspots showed limited association with Ty912 on the left arm of chromosome V, whereas analysis of the data generated with EcoRI indicated that none of the hotspots were located adjacent to chrV-L (Table S4). Thus, we found minimal to no interactions between the region surrounding Ty912 and the observed hotspot locations. Consequently, it is unlikely that physical proximity plays a large role in the selection of the most commonly used translocation targets.
Most of the GCRs identified using the +Ty912 assay appear to result from RAD52-dependent Ty-mediated HR resulting in non-reciprocal translocations [12]. In S. cerevisiae, RAD52-dependent HR is primarily mediated by two pathways, which are dependent upon RAD51 and RAD59, respectively [37]. Surprisingly, a rad51Δ single mutant had a significantly increased rate of Ty-mediated GCRs compared to wild type (Wilcoxon rank sum test; p = 4.89×10−7), whereas a rad59Δ mutation decreased the GCR rate of the rad51Δ mutant strain to one equivalent to the GCR rate of the wild-type +Ty912 assay strain (Wilcoxon rank sum test; p = 0.89) (Table 2; Table S2). The rad59 smutation had no significant effect on the Ty-mediated GCR rate of a wild-type strain (Wilcoxon rank sum test; p = 0.34). Interestingly, our previous results indicate that the GCR rate of the rad51Δ rad59Δ +Ty912 assay strain was 53-fold higher that that seen for the rad51Δ rad59Δ −Ty912 assay strain, suggesting that Ty912-mediated GCRs occur at low rates in the rad51Δ rad59Δ mutant (Wilcoxon rank sum test; p = 2.09×10−6) [12].
To investigate how a rad51Δ mutation affected the formation of GCRs, we analyzed 46 GCR-containing isolates derived from a rad51Δ +Ty912 assay strain (Table S2). All 46 isolates contained a chromosome V-L deletion, and 42 of these isolates each contained a chromosome arm duplication (Figure 4a). Four isolates did not have any accompanying chromosome arm duplications, which was a small, but statistically significant, increase compared to the bulk fold increase in the total GCR rate (p<0.05). This suggests that a rad51Δ mutation results in a small increase in the rate of formation of broken chromosomes that are subsequently healed by de novo telomere addition [19]. There was a statistically significant decrease in the proportion of GCRs associated with chromosome III-R duplications arising from the rad51Δ mutant compared to a wild-type strain (Fisher's exact test; p = 4.16×10−4) and, using the same method that was used to analyze the effect of an msh2Δ mutation on the rate of individual chromosome arm duplications, we found that the rate of chromosome III-R duplications was significantly decreased as well (p<0.05; Figure 4b). In contrast, the rates of the chromosome arm duplications involving the chromosome V-R, X-R, and XIV-L hotspots were not preferentially affected by the rad51Δ mutation. Additionally, chromosome IX-L and XII-R duplications were statistically significantly increased in the rad51Δ mutant compared to the relative bulk effect of a rad51Δ mutation. There were also statistically significant increases in the rates of duplication of several chromosome arms not originally observed among the 112 wild-type derived Ty912-mediated GCRs (chromosomes II-R, VII-L, XI-L, XII-L, XIII-R, and XV-R) (Figure 4b; p<0.05). Overall, these results suggest that RAD51 is important for the formation of GCRs involving the hotspots on chromosome III-R and is important for the suppression of GCRs involving many other Ty-related translocation targets.
To better understand the function of RAD59 in promoting GCRs in a rad51Δ mutant, we analyzed 30 Ty912-mediated GCRs obtained from a rad51Δ rad59Δ double mutant (Table S2). We did not analyze a large number of GCRs isolated from a rad59Δ single mutant as a rad59Δ mutation did not significantly change the GCR rate in the +Ty912 assay (Wilcoxon rank sum test; p = 0.343) (Table 2). All 30 GCRs derived from the rad51Δ rad59Δ double mutant contained the chromosome V-L deletion and 29 of these GCRs were also associated with single chromosome arm duplications. Compared to the bulk fold increase in the GCR rate relative to the wild-type GCR rate, the rad51Δ rad59Δ double mutant had statistically significant decreases in the duplication rates of chromosomes III-R, IV-R, and XII-R, and an increase in the duplication rate of chromosome IX-L (p<0.05). We also noted chromosome X-L (n = 1), chromosome XI-L (n = 2) and chromosome XI-R (n = 5) duplications, which were not seen in the wild-type chromosome arm distribution. The rad51Δ rad59Δ double mutant had a lower overall GCR rate than the rad51Δ single mutant (Wilcoxon rank sum test; p = 1.13×10−5) (Table 2). Compared to the bulk affect of a rad59Δ mutation in a rad51Δ single mutant (see Methods), the rad51Δ rad59Δ double mutant had decreased rates of duplication of chromosome arms IV-R and XII-R, as well as increased rates of duplication of chromosome arms XI-L and XI-R (Figure 4c; p<0.05). In addition, the duplication rates of the chromosome arms associated with the chromosome III-R, V-R, X-R and XIV-L hotspots were reduced in the rad51Δ rad59Δ double mutant relative to the rad51Δ single mutant to the same extent as the bulk GCR rate (Figure 4c, p>0.05). Overall, these results support the view that a RAD51-dependent HR pathway is important for suppressing Ty912-mediated translocations and that most of the translocations seen in the rad51Δ single mutant are promoted by RAD59-dependent HR.
To verify that the duplicated chromosome arms associated with the GCRs formed in the rad51Δ rad59Δ double mutant were indeed fused to chromosome V, we analyzed 7 rad51Δ rad59Δ-derived GCRs using PFGE followed by Southern blotting with a probe specific to MCM3, an essential gene on chromosome V. The results showed that all 7 GCRs (I1–I7) were associated with abnormally large chromosome Vs, consistent with the duplicated chromosome arms being fused to deleted chromosome Vs (Figure 4d), and reminiscent of previously investigated nonreciprocal translocations [12]. We did not verify the nature of the fusion junctions in these translocation chromosomes, although, as discussed below, we did verify the nature of several such translocation junctions identified in rad52Δ single and rad52Δ rad51Δ rad59Δ triple mutants.
There was a 5-fold decrease in the GCR rate of the rad52Δ +Ty912 assay strain compared to that of the wild-type +Ty912 assay strain (Wilcoxon rank sum test; p = 1.69×10−05), consistent with the bulk of the Ty912-mediated GCRs being formed by HR (Table 2). The GCR rate of the rad52Δ +Ty912 assay strain, however, was previously determined to be slightly, but not strictly significantly, higher than the GCR rate of the rad52Δ −Ty assay strain (Wilcoxon rank sum test; p = 0.0625) [12], suggesting that Ty912-mediated GCRs may still occur at low rates in the rad52Δ +Ty912 assay strain. To further investigate the role of RAD52 in the formation of Ty1-mediated GCRs, we used MLPA to analyze 50 independent GCRs derived from a rad52Δ +Ty912 assay strain (Table S2). All 50 GCRs were associated with the chromosome V-L deletion. We did not detect any additional genome alterations in 34 (68%) of these isolates, a finding consistent with a high rate of GCRs mediated by de novo telomere additions in a rad52Δ mutant [11], [19], [38]. In 16 (32%) of the isolates, we observed additional chromosome arm duplications, with each isolate containing one chromosome arm duplication (Figure 5a). This represented a significant decrease in the proportion of GCRs associated with chromosome arm duplications when compared those obtained from a wild-type strain (Fisher's Exact test; p = <2.2×10−16). Interestingly, the rad52Δ mutant shared none of the significant chromosome arm duplication changes seen in the rad51Δ rad59Δ double mutant when these two mutant strains were compared with a rad51Δ single mutant, although this may be due to the small number of chromosome arm duplications analyzed in the case of the rad52Δ mutant (Figure 4c).
To confirm that the observed chromosome arm duplications seen in the rad52Δ mutant were associated with the formation of nonreciprocal translocations involving chromosome V-L and the duplicated chromosome arms, we further analyzed 7 rad52Δ-derived GCR-containing isolates, two of which (I15 and I18) did not have chromosome arm duplications as determined by MLPA and five of which had single arm duplications (I16 [chrIII-R], I17 [chrIV-R], I19 [chrX-R], I20 [chrX-R], and I21 [chrV-R]). Analysis of these isolates by PFGE followed by Southern blotting utilizing a probe to the chromosome V gene MCM3 revealed that each of the isolates without chromosome arm duplications contained a smaller than wild-type chromosome V (consistent with the known chromosome V-L deletion associated with a de novo telomere addition) and that the isolates with chromosome arm duplications each had a larger than wild-type chromosome V (Figure 5b). No other chromosomes appeared to have altered lengths. Analysis of these same 7 GCR-containing strains by aCGH revealed that all 7 had a deletion of the left arm of chromosome V between TEL05L and Ty912 (Figure S2). Additionally, each of the 5 strains identified by MLPA to contain a chromosome arm duplication had an additional duplicated region of DNA bounded by a full length Ty1 or solo delta element at one end and a telomere at the other end (Figure S2; Table S8); the amplified chromosome arms identified by MLPA were the same as those identified by aCGH. PCR amplification of the breakpoints of 3 of the GCRs (I17, I19 and I20) associated with chromosome arm duplications confirmed that these GCRs were the result of a fusion between chromosome V-L at Ty912 and the Ty or delta element bounding the duplicated region associated with each GCR (Figure 5c; Table S8); the other 2 breakpoints could not be amplified. In addition, PCR amplification and sequencing of the breakpoints of four independent translocations (I22–25) occurring in the rad52Δ +Ty912 assay strain and identified by MLPA to be associated with a chromosome arm duplication similarly demonstrated the presence of translocation chromosomes that resulted from fusions between chromosome V-L at Ty912 and the Ty or delta elements bounding the duplicated region associated with each GCR (Figure 5c; Table S8). Thus, deletion of RAD52 decreases, but does not completely eliminate, HR that promotes Ty-mediated nonreciprocal translocations [19], [38].
We next examined whether RAD51 and RAD59, which encode homologous pairing proteins [37], and the Rad1–Rad10 complex, which processes non-homologous single-stranded DNA tails that form during single-stranded annealing [39], were required for the formation of Ty912-mediated translocations formed by RAD52–independent HR. To this end, we investigated to differing degrees the rate and structure of Ty912-mediated GCRs in rad52Δ and rad52Δ rad59Δ double mutants, the rad52Δ rad51Δ rad59Δ triple mutant, and the rad52Δ double mutant. All four mutant strains had GCR rates that were statistically similar to the GCR rate of a rad52Δ single mutant (Table 2) (Wilcoxon rank sum test; p = 0.21, 0.10, 0.95 and 0.79, respectively) and had statistically similar frequencies of GCRs containing and lacking chromosome arm duplications (Fisher's Exact Test with Bonferroni correction for multiple hypothesis testing; all p values>0.0167) (Figure 6a; Figure S3; Table S2) [40]. The chromosome arm duplication patterns of the rad52Δ double and triple mutants were also highly similar to that of the rad52Δ single mutant (Figure 6a; Figure S3), unlike that of the rad51Δ rad59Δ double mutant (see above).
To confirm that the chromosome arm duplications seen in the rad52Δ rad51Δ rad59Δ mutant were associated with the formation of nonreciprocal translocations involving chromosome V-L and the duplicated chromosome arms, we analyzed 7 GCRs isolated from this mutant by PFGE followed by a Southern blotting utilizing a probe to MCM3. Four GCRs were shown by MLPA to be associated with single chromosome arm duplications (I10 [chrIV-R], I11 [chrX-R], I12 [chr-XIIR], I13 [chrXII-R]) and three were not associated with a chromosome arm duplication (I8, I9, I14) (Figure 4d). The results revealed that the four GCRs associated with chromosome arm duplications (I10, I11, I12, I13) were each associated with an abnormally large chromosome V and the three GCRs that did not involve a chromosome arm duplication (I8, I9, I14) were each associated with a smaller than wild-type chromosome V. The smaller than wild-type chromosome Vs were consistent with the formation of GCRs mediated by de novo telomere addition. No other chromosomes appeared to be altered in these 7 GCR-containing strains. PCR amplification of the breakpoints of 2 of the GCRs derived in the rad52Δ rad51Δ rad59Δ +Ty912 assay strain (I11 & I12) confirmed the presence of translocation chromosomes that were the result of a fusion between chromosome V-L at Ty912 and the Ty1 element bounding the duplicated region associated with each GCR (Figure 6b). Thus, deletion of RAD51 and RAD59 in a rad52Δ mutant does not completely eliminate HR that promotes Ty-mediated nonreciprocal translocations.
We next investigated the potential role of two different pathways involving Rtt109 in suppressing GCRs and aneuploidy. Both Vps75, a histone chaperone, and Asf1, a nucleosome assembly factor, form separate complexes with Rtt109 that acetylate H3K56 in vitro, although only the Asf1-Rtt109 complex appears to be required for acetylation of H3K56 in vivo [47]. We found that a vps75Δ single mutant had essentially the same rate of Ty912-mediated GCRs as the wild-type strain, whereas an asf1Δ single mutant had an increased rate of Ty912-mediated GCRs similar to that seen for a rtt109Δ mutant (Table 2). Analysis of 45 vps75Δ and 55 asf1Δ derived GCR-containing isolates for whole chromosome duplications revealed no whole chromosome duplications among the vps75Δ-derived GCR-containing isolates and 7 instances of asf1Δ-derived GCR-containing isolates containing whole chromosome duplications (Figure 7a; Table S2). The frequency of whole chromosome duplications in the asf1Δ mutant was not significantly different from that of an rtt109Δ mutant (Fisher's Exact test; p = 0.207). Additionally, the distribution of chromosome arm duplications seen for the GCRs derived from the vps75Δ mutant (40/45 GCRs) was strikingly similar to that seen with the wild-type strain, while the distribution of chromosome arm duplications seen for the GCRs derived from the asf1Δ mutant (50/55 GCRs) more closely reflected that of the rtt109Δ mutant (Figure 7b, 7c; Table S2).
RLF2 (also known as CAC1) encodes a subunit of CAF-1, a chromatin assembly complex that mediates nucleosome assembly in cooperation with ASF1 [48]. Consistent with this, an asf1Δ double mutant was previously demonstrated to have a synergistic increase in the rate of single copy sequence-mediated GCRs compared to both rlf2Δ and asf1Δ single mutants [49]. The rlf2Δ single mutant strain was similar to the wild-type strain and had only a small 2-fold increase in the Ty912-mediated GCR rate (Table 2). When we analyzed 46 rlf2Δ derived GCR-containing isolates by MLPA, we detected no whole chromosome duplications and a chromosome arm duplication pattern similar to that seen in the wild-type strain, except for increased rates of chromosome III-L, IX-L, XI-L, and XVI-R arm duplications (Figure S4; Table S2). Similarly, there was no significant increase in the frequency of whole chromosome duplications in an asf1Δ rlf2Δ double mutant compared to an asf1Δ single mutant (Fisher's Exact test; p = 0.483) (Figure 7a), even though the double mutant had a synergistic 120-fold increase in the GCR rate compared to the asf1Δ and rlf2Δ single mutants (Table 2). Moreover, 75% (9/12) of the chromosome arm duplications seen in the asf1Δ rlf2Δ mutant (29/30 GCRs) were seen in an asf1Δ mutant, while the remaining 25% (3/12) chromosome arm duplications were observed in the profile of an rlf2Δ mutant. Thus, despite the synergistic increase in GCR rate observed in an asf1Δ rlf2Δ mutant, the distribution of chromosome arm duplications appeared to be additive, with the increase in aneuploidy likely a result of only the asf1Δ mutation. Taken together, these results suggest that the rtt109Δ-dependent increase in whole chromosome duplications is linked to defects in the ASF1-RTT109-dependent H3K56 acetylation pathway.
To investigate whether aneuploidy results from the inability of the ASF1-RTT109 complex to acetylate histone H3K56, we tested mutations that altered histone H3K56 for their effect on the accumulation of whole chromosome duplications. For this analysis, we constructed strains with chromosomal hht1-hhf1Δ hht2-hhf2Δ double deletions carrying plasmids encoding various unacetylatable mutant alleles of HHT2 and a wild-type copy of HHF2. In this manner, we were able to construct hht1-hhf1Δ hht2-hhf2Δ HHF2 hht2::H3K56G and hht1-hhf1Δ hht2-hhf2Δ HHF2 hht2::H3K56R mutant strains. The strain containing the hht2::H3K56G mutation and the strain containing the hht2::H3K56R mutation both exhibited increased GCR rates that were slightly lower than that of an rtt109Δ mutant (Table 2). All (46/46) and 97.8% (44/45) of the GCR-containing isolates derived from the hht2::H3K56G mutant and hht2::H3K56R mutant, respectively, were associated with single chromosome arm duplications; there were 6 and 2 isolates of the hht2::H3K56G and hht2::H3K56R mutant strains, respectively, that had both a whole chromosome duplication as well as a GCR (Figure 7a). The frequency of whole chromosome duplications in the hht2 mutants was either not significantly different (hht2::H3K56G, Fisher's Exact test; p = 0.289) or slightly reduced (hht2::H3K56R, Fisher's Exact test; p = 0.0155) compared to the rtt109Δ mutant. We further investigated the distribution of chromosome arm duplications for the GCRs derived from the hht2 mutant strains to the distribution seen for the wild-type strain. We focused on seven chromosome arm duplications (chrIII-R, IV-R, V-R, VII-R, VIII-L, XII-R, and XVI-R) that shared duplication differences in the rtt109Δ and asf1Δ single mutants when compared to the distribution observed for the wild-type strain (Figure 7b; Table S2). The profile seen for the hht2::H3K56G mutant shared 4 of the 7 changes (Hypergeometric test; p = 0.0187), while the profile seen for the hht2::H3K56R mutant shared 5 of the 7 changes (Hypergeometric test; p = 7.72×10−6), supporting the view that these two hht2 mutations are similar to the rtt109Δ and asf1Δ in regard to their effects on the suppression of GCRs. Overall, these results suggest that the loss of acetylation at histone H3K56 results in high rates of GCRs and whole chromosome duplications.
In this study, we adapted MLPA to identify chromosome arm duplications and deletions, as well as whole chromosome duplications, in order to provide insights into the processes that suppress and promote GCRs in S. cerevisiae. Compared to previous methods used to analyze GCRs [7], [11], [12], [14], [15], [20]–[29], this method is rapid, affordable, and of sufficiently high resolution to provide useful structural insights to guide subsequent analysis. We validated the utility of this method by investigating bias in target site selection of Ty1-mediated translocations, RAD52-independent Ty1-mediated translocations that appear to occur by HR, and whole chromosome duplications that occur at increased rates in an rtt109Δ mutant. This ability to generate structural information for a large numbers of GCRs allowed us to demonstrate the existence of a number of translocation target hotspots and demonstrate that these hotspots likely mediate translocations by different mechanisms. In addition, we obtained evidence for a RAD52-independent HR pathway that can promote Ty1-mediated translocations, as well as evidence for an association between whole chromosome duplications and GCRs in an rtt109Δ mutant. These results demonstrate that the methods developed here will facilitate future analysis of GCR structures in other mutant backgrounds, which will likely reveal yet other unanticipated aspects of the mechanisms that prevent GCRs.
We previously observed that GCRs isolated in the presence of Ty912 located on a nonessential arm of chromosome V were almost exclusively associated with a loss of the region of chromosome V-L from the Ty912 to the left telomere and a duplication of a terminal region of another chromosome arm [12]. In nearly all cases, these GCRs were the products of nonreciprocal translocations mediated by HR between Ty912 on chromosome V-L and an ectopic Ty element on a target chromosome. Using MLPA probes hybridizing to the telomeric ends of the chromosomes, we identified four chromosome arms (the right arm of chromosome III, the right arm of chromosome V, the left arm of chromosome XIV, and the right arm of chromosome X) that were the target of approximately 70% of the translocations in a wild-type strain and thus appeared to contain hotspots for HR events resulting in Ty912-mediated translocations. Detailed analysis using MLPA probes specific to these chromosome arms demonstrated that there were only 6 hotspots that mediated these translocations versus at least 254 potential targets for Ty-mediated translocations in the S. cerevisiae genome, and that furthermore, these hotspots were composed of Ty elements.
We considered a number of potential explanations to account for the existence of the observed translocation hotspots; however, no single factor could explain the observed translocation target distributions. First, sequence homology with Ty912 did not explain the hotspots, as the target Ty elements at the hotspots did not share the highest degree of homology with Ty912. Additionally, the target site distribution was relatively unaffected by an msh2Δ mutation that increased the efficiency of homeologous recombination, suggesting that the hotspot distribution was not determined by sequence homology. Second, the location of essential genes likely played little or no role in generating the target site distribution. The region of the left arm of chromosome V that is deleted in the GCRs detected by the assay contains only non-essential genes, and all of the translocations detected were non-reciprocal translocations that only duplicated other regions of the genome. Hence, no essential gene was ever observed to be deleted or would be expected to be deleted in this assay. Third, the distribution of centromere- (52.4% of Ty elements) vs. telomere- (47.6% of Ty elements) oriented Ty elements does not explain the distribution of target hotspots as almost all of the chromosome arms contain centromere- and telomere-oriented Ty elements, and the hotspot usage observed was significantly different compared to the distribution of all Ty elements as well as the distribution of only telomere-oriented Ty elements. Furthermore, translocations mediated by centromere-oriented Ty elements, which result in intermediate dicentric translocation chromosomes, were recovered using our assay, and at least 1 of the 6 observed translocation hotspots has been previously demonstrated to mediate the formation of dicentric translocations that undergo additional rounds of rearrangements [12]. Finally, analysis of the data from 4C experiments [36] showed that the spatial proximity of the hotspot targets relative to Ty912 on the left arm of chromosome V was also unable to explain the hotspots. Thus, target selection seems likely to reflect structural features of either individual Ty elements that are the targets of Ty-mediated translocations, or structural features of the region of the genome in which they reside.
The translocation hotspots on chromosome III-R were located at two previously identified fragile sites (FS1 and FS2) that were induced by down-regulation of the replicative DNA polymerases [14], [15]. Consistent with previous studies [15], [27], [33], each of these fragile sites contains a pair of Tys that were targeted by Ty912-mediated translocations, with FS1 containing 2 tandemly repeated Ty1 sequences and FS2 containing a pair of inverted Ty1 sequences separated by a short spacer sequence. The fact that we identified these same two fragile sites using our assay that detects spontaneous rearrangements implies that these sites are also fragile under conditions of normal DNA replication. We previously suggested that Ty912-mediated translocations might occur by a mechanism in which a broken chromosome V was repaired by break-induced replication (BIR) in which the Ty912 on a broken chromosome V promotes strand invasion at the site of ectopic Ty elements and primes DNA synthesis, resulting in copying the terminal region of the target chromosome from the target Ty element to the telomere onto the end of the broken chromosome V [12]. However, the observation that FS1 and FS2 are hotspots for Ty912-mediated translocations suggests at least two other possible mechanisms: HR between two DSBs, one on chromosome III-R and one on chromosome V-L, or BIR initiated by a DSB at either FS1 or FS2 that then invades chromosome V at Ty912 and is followed by loss of the intact chromosome V left by BIR during the selection for the GCR. The fact that another Ty hotspot on chromosome X-R also appears to contain tandem Ty1 elements (YJRWTy1-1 and YJRWTy1-2) (Table 1) suggests that this site might also represent a fragile site, although other there are Ty1 loci in the genome that also contain tandem Ty1 sequences and these did not appear to be translocation target hotspots sites. The hotspot on chromosome XIV-L and the two hotspots on chromosome V-R, one of which has previously been observed as a target of Ty-mediated translocations [9], [12], [26], are not annotated to contain tandem or inverted Ty elements and it is not clear what causes these sites to act as translocation hotspots.
We previously observed that a large proportion of Ty912-mediated translocations were mediated by RAD52-dependent HR and we obtained genetic evidence that a RAD51–dependent HR pathway primarily suppressed Ty-mediated translocations in wild-type strains, whereas a RAD59–dependent HR pathway promoted Ty-mediated translocations in the absence of RAD51. In agreement with this, we found that a rad51Δ mutation increased both the rate of accumulation and the diversity of chromosome arm duplications, and that a rad59Δ mutation decreased the rate of most of the individual types of GCRs that occurred in a rad51Δ mutant. Interestingly, in the rad51Δ mutant, several of the chromosome arms that were duplicated at increased rates did not have full-length Ty1 elements but only contained solo delta elements (chrVII-L, chrIX-L, chrXI-L, and chrXII-L). As RAD59-dependent HR promotes SSA between short repeat sequences [50] this raises the possibility that some of the translocation targets that show increased targeting in the rad51Δ mutant may reflect increased rates of HR between short repeated sequences mediated by RAD59-mediated HR. The majority of the GCRs isolated from rad51Δ and rad59Δ single mutants and rad51Δ rad59Δ double mutant strains were the result of nonreciprocal translocations between Ty912 and ectopic Tys like those seen in wild-type strains. A striking finding was the almost complete elimination of translocations mediated by the chromosome III-R hotspots in the rad51Δ mutant in contrast to the other GCRs, including those mediated by the other translocation hotspots whose rates were increased in the rad51Δ mutant and decreased in the rad51Δ rad59Δ double mutant. This suggests that structural features of the translocation targets affect the mechanism of translocation. For example, the chromosome III-R hotspot may be a particularly good substrate for BIR given that the translocations mediated by this hotspot were highly dependent on RAD51, consistent with prior results [16], [24], [51], whereas the other translocation targets might be more amenable to aberrant repair by RAD59-dependent single-stranded annealing (SSA) of broken chromosomes at the site of Ty elements [50], [52], [53].
The results described here also support the idea that Ty-mediated translocations can occur via a RAD52-independent HR pathway. Deletion of RAD52 reduced the rate of Ty-mediated GCRs significantly below the wild-type rate, although not completely to the level observed in a rad52Δ mutant in the absence of the Ty912 element. Consistent with this, a significant proportion of GCRs in a rad52Δ mutant with the Ty912 present appeared to result from de novo telomere addition after chromosome breakage. However, there was also a significant fraction of translocations mediated by HR between Ty912 and ectopic Ty elements in a rad52Δ mutant. These Ty-mediated translocations were the same types of translocations seen in wild-type strains. At first glance, our data suggest that the Ty912-mediated GCRs in a rad52Δ mutant might occur by the same type of RAD52-independent recombination seen for other repetitive elements, which is believed to involve SSA followed by half-crossover mechanism [54], [55]. However, simultaneous deletion of both RAD51 and RAD59, or deletion of RAD1 (which can promote SSA by removal of nonhomologous DNA flaps [56]), in the rad52Δ mutant did not reduce the GCR rate further and resulted in distributions of GCRs that were similar to that seen in the rad52Δ single mutant. Furthermore, we demonstrated that several of the translocations seen in the rad52Δ rad51Δ rad59Δ triple mutant were due to HR between Ty912 and ectopic Ty elements. This suggests that there might be other mechanisms for promoting recombination or that other nucleases besides Rad1–Rad10 can cleave off non-homologous tails that form during SSA [57], [58].
Previous studies have observed occasional examples of strains containing GCRs and independent whole chromosome duplications [12], [18], [23]. In the present study, we used MLPA to detect whole chromosome duplications in large numbers of independent GCR-containing rtt109Δ-derived isolates. We found a statistically significant association of GCRs and whole chromosome duplications in these isolates. Because an rtt109Δ mutation caused an increase in the accumulation of whole chromosome duplications, but the presence of a whole chromosome duplication did not increase the GCR rate of an rtt109Δ mutant, and because the presence of GCRs did not increase the accumulation of whole chromosome duplications in an rtt109Δ mutant, it seems likely that in an rtt109Δ mutant, whole chromosome duplications tend to arise with GCRs. RTT109 encodes a histone H3 lysine 56 acetylase [41], [42] and our analysis demonstrated that the suppression of GCRs and the suppression of whole chromosome duplications associated with GCRs was due to the role of the Rtt109-Asf1 complex in the acetylation of H3K56. The mechanism by which acetylation of H3K56 suppresses GCRs and aneuploidy is currently unclear; however, the observation that rtt109Δ mutants have defects in recovery from DNA damage induced checkpoint activation [43], [44], [59], but are not DNA repair defective per se, suggests aberrant recovery from arrest due to the DNA damage that results in GCRs might also result in mis-segregation of chromosomes during mitosis.
General methods have been previously described [7], [11], [12], [60]. Genomic DNA preparations were quantified using a Qubit Fluorometer (Invitrogen). All strains used in this study were derivatives of S288c and are described in Table S12. Plasmids pFX04 and pFX06 [61] were used to create hht2::H3K56G HHF2 and hht2:H3K56R HHF2 strains. Ty1 fusion junctions were amplified with Velocity DNA Polymerase (Bioline) using the following protocol: 2 min 98°C initial denaturation step; 25 cycles of 12 sec 98°C denaturation, 30 sec 63.8°C annealing, 3 min 30 sec 72°C extension; final 4 min 72°C extension.
Primer design followed the recommendations for synthetic-probes available on the MRC-Holland website (http://www.mrc-holland.com), with the exception that the minimum length differences between amplification products was designed to be 2 base pairs. All primers were purchased and synthesized by ValueGene (http://www.valuegene.com). We designed a total of six sets of primers. Two primer sets were designed to detect copy number variation of different chromosome arms. The set of 32 telomeric probe pairs (Table S1; Figure 1b) was designed to hybridize to unique sequence located between chromosome telomeres and the most distal Ty1 or solo delta elements; four probes (corresponding to the left arm of chromosome II, the right arm of chromosome IV, the right arm of chromosome IX, and the left arm of chromosome XV) were designed to hybridize centromeric to the most distal Ty1 or solo delta elements due to the lack of suitable unique sequence in the preferred region. The set of 32 centromeric probe pairs (Table S9; Figure 1b) was designed to hybridize to unique sequence located centromeric to the Ty1 or solo delta elements closest to the centromere of each chromosome arm. Ty2 elements were represented in this analysis as two independent solo delta elements per Ty2 element as these delta elements contain the majority of the homology between Ty2 and Ty1 elements. We also designed four primer sets capable of pinpointing duplications along whole chromosome arms for chromosomes III-R, V-R, X-R, and XIV-L (Tables S3, S5, S6, S7, respectively). These primer sets generally contained pairs of probes that hybridized immediately telomeric and centromeric to each Ty1 and solo delta element on the pertinent chromosome arm (Figure 3a–3d).
All amplification reagents were purchased from MRC-Holland (http://www.mrc-holland.com). The MLPA reaction has been previously described for human genomic DNA [32]. Briefly, probes are hybridized to chromosomal DNA, ligated, and amplified by PCR using universal priming sequences (Figure 1c). Modifications for analysis of S. cerevisiae genomic DNA were as follows: 5 ng of template genomic DNA was used instead of 50–100 ng; a 10 min initial 98°C denaturation step was used instead of 5 min; 23 (for chromosome arm specific probe sets) or 25 (for telomeric or centromeric probe sets) PCR cycles were used instead of 35 cycles; and, importantly, the use of thin walled PCR tube strips and plates.
Fragment separation was carried out as suggested by MRC-Holland on an ABI 3730XL sequencer using POP7 polymer and GS500-LIZ sizing standard (ABI). 8.95 µl of Hi-Di Formamide (ABI) and 0.05 µl of GS500-LIZ sizing standard were mixed with 1 µl of the MLPA PCR product per isolate for a total volume of 10 µl. Then an 82°C heat denaturation step was performed for 2 minutes followed by an incubation step at 4°C for 5 minutes before analysis on the ABI3730XL sequencer instead of the suggested 80°C/4°C steps. Raw peak data for each run was obtained via the GeneMapper software from ABI.
Data analysis was performed using Python (version> = 2.5) and as described by MRC-Holland (http://www.mrc-holland.com).
The observed rate of the duplication of each chromosomal arm for a strain was taken to be the total CanR 5FOAR rate of the strain multiplied by the fraction of GCRs associated with the duplication of that chromosomal arm. The expected rate of a test strain relative to a control strain was calculated by multiplying the observed duplication rate for each chromosome arm in the control strain by a scaling constant. The scaling constant was the sum of the rates of the duplications of all chromosome arms in the test strain divided by the sum of the rates of the duplications of all chromosome arms in the control strain. This scaling constant was very similar to the ratio of CanR 5FOAR rates of the test and control strains for those strains dominated by GCRs associated with chromosome arm duplications, but properly handles mutations like rad52Δ which results in a substantial fraction of GCRs lacking chromosome arm duplications. In cases where no duplications for an individual chromosome arm were found in the control strain, then the expected rate was estimated as the upper limit of the duplication rate of that chromosomal arm (the total CanR 5FOAR rate divided by the number of duplications). In cases where no chromosome arm duplication was found in the test strain, then the upper limit of the rate was used if it was less than the expected rate, otherwise the ratio of observed vs. expected rates was set to 1. In cases where no duplications for an individual chromosome arm were found in both the test and control strain, then the ratio of observed vs. expected rates was set to 1.
R (version≥2.11.1) was used to calculate p-values for Wilcoxon rank-sum tests, Fisher exact tests, the Monte Carlo sampling of multinomial distributions, the Hypergeometric test and the Binomial calculation. Ninety-five percent confidence intervals for the median GCR rates were calculated using a two-sided nonparametric test (http://www.math.unb.ca/~knight/utility/MedInt95.htm). Individual chromosome arm duplication rates were considered to be statistically significantly different from the bulk rate of a mutant if the calculated individual chromosome arm duplication rates fell outside the transformed 95% confidence interval of the bulk fold rate of the mutant strain.
|
10.1371/journal.pgen.1003769 | The Uve1 Endonuclease Is Regulated by the White Collar Complex to Protect Cryptococcus neoformans from UV Damage | The pathogenic fungus Cryptococcus neoformans uses the Bwc1-Bwc2 photoreceptor complex to regulate mating in response to light, virulence and ultraviolet radiation tolerance. How the complex controls these functions is unclear. Here, we identify and characterize a gene in Cryptococcus, UVE1, whose mutation leads to a UV hypersensitive phenotype. The homologous gene in fission yeast Schizosaccharomyces pombe encodes an apurinic/apyrimidinic endonuclease acting in the UVDE-dependent excision repair (UVER) pathway. C. neoformans UVE1 complements a S. pombe uvde knockout strain. UVE1 is photoregulated in a Bwc1-dependent manner in Cryptococcus, and in Neurospora crassa and Phycomyces blakesleeanus that are species that represent two other major lineages in the fungi. Overexpression of UVE1 in bwc1 mutants rescues their UV sensitivity phenotype and gel mobility shift experiments show binding of Bwc2 to the UVE1 promoter, indicating that UVE1 is a direct downstream target for the Bwc1-Bwc2 complex. Uve1-GFP fusions localize to the mitochondria. Repair of UV-induced damage to the mitochondria is delayed in the uve1 mutant strain. Thus, in C. neoformans UVE1 is a key gene regulated in response to light that is responsible for tolerance to UV stress for protection of the mitochondrial genome.
| The majority of fungi sense light using the White Collar complex (WCC), a two-protein combination of a photoreceptor and a transcription factor. The WCC regulates circadian rhythms, sexual development, sporulation, metabolism, and virulence. As such, the exposure to light controls properties of fungi that are beneficial and detrimental to people, depending on the species and its interaction with humans. Despite the importance of light on fungal biology, the underlying evolutionary benefit of light-sensing in fungi has remained a mystery. Here we identify a DNA damage repair endonuclease, Uve1, required for UV stress tolerance in the human pathogen Cryptococcus neoformans. UVE1 is a direct target of the WCC in C. neoformans, and UVE1 homologs are also regulated by WCC in two other major lineages of fungi, the Ascomycota and Mucoromycotina. The divergence of the three groups indicates that for about a billion years the same transcription factor complex has regulated a common gene to protect fungal genomes from deleterious effects of light. Curiously, in C. neoformans Uve1 localizes to mitochondria and contributes to mitochondrial DNA repair, implicating its importance in genome repair of this organelle. Thus, light-sensing in fungi exists to protect them against harmful light, and likely all other responses to light relate to or are a secondary consequence of this selective pressure.
| The ability to sense light provides well-known advantages to organisms, such as adapting to photosynthetic light sources in plants and for vision in animals, yet the benefits of light-sensing in non-photosynthetic and non-motile organisms are less established. The fungi contain a suite of potential photoreceptor proteins, with the White Collar complex (WCC) being found throughout the fungal kingdom, except for species in which the two genes encoding the complex were lost. Light influences different responses in different fungi, including phototropism, induction of pigmentation, asexual and sexual sporulation, changes is primary and secondary metabolism, and regulating the circadian clock: most of which are controlled, where established, by the WCC [1]–[3]. A major question is what advantage is provided in using light as an environmental signal to regulate these processes. One compelling hypothesis is that protecting DNA from damage provides a selective pressure, and that wavelengths in the visible spectrum are sensed to indicate the presence of deleterious ultraviolet radiation. However, a DNA repair system common to light-sensing fungi and that acts directly downstream of the WCC is unknown to date.
The White collar-1 and White collar-2 proteins interact to form a complex (WCC) capable of sensing blue and near UV light. Both proteins were originally characterized in the ascomycete fungus Neurospora crassa [4]–[7] where WC-1 acts as a photoreceptor. In N. crassa, WCC has roles to play both in light and dark environments. In the dark the WCC has a major function as a circadian clock component, via regulation of FRQ gene expression [8]–[10]. Light-dependent functions of WCC include conidiation, carotenoid production and mating [11], [12]. The photons and the signal are transduced via the chromophore flavin adenine dinucleotide (FAD) bound within an N-terminal specialized type of PAS domain (from the Per, Arnt, Sim proteins), named the LOV (light, oxygen and voltage) domain [13]. The two proteins interact using other PAS domains, to form the transcription factor complex that regulates transcription of target genes via their GATA-type zinc finger DNA-binding domains [4], [5], [7], [14].
The human pathogen Cryptococcus neoformans is a member of the phylum Basidiomycota, a distant relative to N. crassa. The fungus grows vegetatively as a budding yeast and during mating in a dikaryotic filamentous form. C. neoformans is divided into two varieties: var. grubii is the most prevalent in the clinic and var. neoformans is less common but was the most amenable to experimental methods until the discovery of an opposite mating partner for var. grubii about a decade ago [15]. C. neoformans primarily causes disease in immunocompromised people. The closest relative to C. neoformans is C. gattii, which is also a human pathogen but with a greater tendency to infect immunocompetent individuals [16]. Homologs of N. crassa wc-1 and wc-2 are present in Cryptococcus species, designated BWC1/CWC1 and BWC2/CWC2, and have been characterized in strains of both var. grubii and var. neoformans [17], [18]. The Bwc1-Bwc2 complex has three known functions in C. neoformans: it represses mating in the light, promotes virulence, and provides protection against UV light. The downstream targets of Bwc1-Bwc2 that control these functions remain to be elucidated.
Several genes have been identified from C. neoformans that are regulated by light. These include SXI1α and MFα1, found within the mating type locus and encoding a transcription factor and pre-pheromone protein required for sexual reproduction [17]. Their expression could explain the repression of mating by light. However, the regulation of these transcripts was compared after a long time exposure of 24 h in the light or constant darkness, such that this time point likely reflects indirect regulation by Bwc1-Bwc2. More recently a microarray experiment was carried out in C. neoformans to detect light-regulated transcripts with an hour of exposure to light. The HEM15 gene, encoding ferrochetalase that is the last step in the heme biosynthetic pathway, was identified as a gene under control of Bwc1-Bwc2 [19]. However, the phenotype of the knockout hem15 strain differs significantly from mutation of BWC1 or BWC2. For example, HEM15 is essential for viability, and yet the bwc1 and bwc2 mutants do not show any growth defects in the light or dark. Another light-regulated gene is CFT1, required for iron uptake and virulence [20], yet no iron-dependent phenotype of bwc1 and bwc2 mutants is known. Thus, while CFT1 and HEM15 may be targets of the Bwc1-Bwc2 complex, they likely play minimal roles in the physiological response of C. neoformans to light. Moreover these two genes are also under the control of other transcription factors in addition to WCC. As a transcription factor complex, it was puzzling that more light-regulated genes were not identified by microarray analysis that could potentially explain the phenotypes of deleting the WCC from C. neoformans.
The microarray results of C. neoformans contrast to ascomycete species. For instance, in N. crassa a microarray study in wild type, wc-1Δ, and wc-2Δ strains at different time periods identified 314 light-regulated genes, constituting 5.6% of the total detectable transcripts in the genome [11]. These genes were grouped into two broad categories, the early or late light responsive genes (ELRGs and LLRGs). Some of these ELRGs are involved in the synthesis of vitamins, photo-protective pigments, prosthetic groups and cofactors, cellular signaling, DNA processing, circadian rhythm and secondary metabolism. Many of the LLRGs are implicated in carbohydrate metabolism, fatty acid oxidation and free radical detoxification. In N. crassa there is also a link between the WCC and DNA repair, for example through the clock gene prd-4, which is a cell cycle checkpoint kinase 2 [21]–[23].
As the White Collar complex acts as an UV/blue light photoreceptor and mutation of the complex causes an increase in sensitivity to UV light, potential targets are hypothesized to be genes involved in repairing DNA damage for survival under UV stress. The UVE1 gene of C. neoformans was previously identified in a UV sensitive strain in a collection of insertional mutants [24]. The product of UVE1 is a homolog of an apurinic/apyrimidinic endonuclease that is best characterized in fission yeast Schizosaccharomyces pombe. In S. pombe the gene was called UVDE for UV damage endonuclease, and renamed uve1 for consistency with nomenclature [25], [26]. The mus-18/UVE-1 gene is the homolog characterized from N. crassa [27]. The protein removes UV-induced cyclobutane pyrimidine dimers and 6-4 photoproducts, acting in its own pathway termed the UVDE-dependent excision repair (UVER) pathway. UVDE recognizes single-stranded DNA nicks, apurinic/apyrimidinic sites, and nucleotide mismatches [26], [28]–[30], a suite of DNA lesions that also extends a possible role for UVDE in repairing the equivalent types of DNA damage caused by reactive oxygen species [31].
In S. pombe, UVDE is localized and functional in both the nucleus and mitochondria, and was suggested to act as a reserve mechanism for repairing UV-induced DNA damage in the mitochondria [32]. Homologs of UVE1 are present in a subset of species in the Archaea, Bacteria and Eukaryotes [25], [33]–[37], with one exception being humans where there is no homolog. A preliminary northern blot experiment suggested that UVE1 in C. neoformans var. neoformans is a light-regulated gene with two isoforms, triggering this investigation. We hypothesized that UVE1 is a downstream target of the WCC that functions in repairing UV-induced damage, and tested this hypothesis in the experiments described below.
A T-DNA insertion mutant within the promoter of the UVE1 gene was identified previously [14]. Under UV stress conditions the mutant showed negligible survival as compared to the wild type (KN99α) and the unexposed strains (Figure 1A). Transformations derived from Agrobacterium T-DNA delivery can have phenotypes that are not due to the insertion of the T-DNA into the host genome. The UV sensitivity phenotype of the original mutant was verified by constructing UVE1 gene replacement strains for both C. n. var. neoformans and C. n. var. grubii. The knockout strains in both varieties had reduced survival after exposure to UV (Figure 1A,B). To confirm that the UV sensitivity phenotype was because of the absence of UVE1, an uve1Δ strain was complemented with a wild type copy of UVE1. The UVE1 complemented strain completely rescued the UV sensitivity phenotype (Figure 1). These results show that in C. neoformans the UVE1 gene is required for survival under UV stress.
The responses of the uve1 mutant to stresses other than UV light were tested, showing no other phenotypes (Figure S1). The gene also plays no major role in the formation of mating filaments (Figure S2A, B). A prior analysis, using comparative growth in a pool of 48 strains in the mouse lung, suggested the UVE1 gene has no role in virulence [38]. To examine the uve1Δ strain in isolation, its virulence was tested in an insect model. While the bwc1Δ strain is less pathogenic than wild type in this model, the uve1Δ strain is equally virulent as wild type (Figure S2C). Thus, the only established function of Uve1 in C. neoformans is in response to UV stress.
Since UVE1 is required for surviving exposure to UV light, regulation of UVE1 was tested in response to light. Wild type strains of C. n. var. neoformans, C. n. var. grubii and C. gattii were grown in the dark, and one replicate provided a one hour exposure to white light. Expression of the UVE1 gene was examined by northern blot analysis on polyA RNA purified from these cultures. The UVE1 transcript levels were higher in light grown conditions for all wild type strains (Figure 2). Two transcripts were observed for var. neoformans strain JEC21, one longer (L, for light) induced specifically under light conditions and another shorter (D, dark) expressed in the dark. For var. grubii and C. gattii one isoform of UVE1 was expressed in response to light with negligible expression in dark grown conditions.
The role for the Bwc1 photoreceptor in UVE1 induction by light was examined in the bwc1Δ deletion strains of C. neoformans. In the var. neoformans deletion there was loss of induction of the longer isoform by light, and residual expression of this isoform in both light and darkness. There was no observed effect of bwc1Δ on the shorter isoform in dark grown conditions. Dark isoforms expressed equally well in both light and dark conditions in some replicates, possibly reflecting the age of the culture or shading by other cells. In the var. grubii bwc1Δ mutant, the UVE1 transcript was barely detectable under either illumination regime (Figure 2). These analyses in the bwc1 mutant backgrounds indicate that UVE1 expression is controlled by the Bwc1-Bwc2 complex.
Rapid amplification of cDNA ends (RACE) was used in the var. neoformans strain to define the 5′ and 3′ ends of the two transcripts that were produced in the light and the dark. The UVE1 dark isoform is part of the UVE1 light isoform, with the dark isoform starting in the middle of the light isoform and both sharing a common 3′ end (GenBank accessions KF234405 and KF234406; Figure 3). Alignment of the Uve1 homologs from various fungi indicated that the dark isoform of C. neoformans is truncated and missing key residues in the active site of this protein (Figure 3; Figure S3).
The light-dependent regulation of UVE1 in wild type strains suggested that induction of the gene prior to UV exposure would correlate with increased UV resistance. The wild type, bwc1Δ and the complemented strains were grown in complete darkness overnight. One set was kept in the dark and the other set exposed to light for 2 hours, prior to treatment of both sets with UV light (Figure S4). All three strains grown in darkness showed similar levels of sensitivity to UV light. Exposure of the strains with the wild type copy of BWC1 to light before UV stress increased their resistance to UV, a property not seen for the bwc1Δ strain. Hence, light signaled by Bwc1 promotes UV resistance.
Two isoforms of UVE1 were observed in the var. neoformans strain, raising the possibility that differential transcript sizes may be a common feature for DNA repair genes in C. neoformans. To identify other genes involved in protecting the fungus against UV damage, a collection of 1200 defined knock out mutants in the var. grubii background [38] was screened for those sensitive to UV light. 13 strains were identified, including the uve1Δ strain in the collection (Figure S5A). For the corresponding genes, northern analysis were performed for var. neoformans and var. grubii cultured under light and dark conditions (Figure S5B). No altered size or induction in response to light was observed, as had been for the UVE1 gene.
To examine if UVE1 photoregulation is common among fungi, northern blot analysis of UVE1 was performed in two fungi, Neurospora crassa and Phycomyces blakesleeanus (Figure 4). The species are light-sensing models in the phylum Ascomycota and subphylum Mucoromycotina, respectively. In N. crassa the expression of UVE1 has already been reported in a microarray study of the light-induced genes [11]. For the N. crassa wild type strain we observed by northern blot analysis the induction of one transcript in light grown conditions and minimal expression of the UVE1 transcript in dark grown mycelia, confirming the previous microarray data. For the P. blakesleeanus wild type strain two transcripts were present in light grown conditions that were both absent in samples grown in the dark. Based on the expressed sequence tag information in the genome database, the 5′ end of the UVE1 homolog is shared between the two transcripts and the 3′ end differs, the reverse of the situation in C. neoformans var. neoformans. The longer isoform in P. blakesleeanus has a 3′ extension due to transcriptional read-through into the 3′ neighboring gene.
We also examined the transcript profile of UVE1 in a white collar-1 mutant (wc-1) of N. crassa and a madA-madB mutant of P. blakesleeanus (madA and madB are functional wc-1 and wc-2 homologs in this species [39]). We observed complete loss of light-dependent induction of UVE1 in these mutant strains of N. crassa and P. blakesleeanus. As an additional control to show that UVE1 induction was not an indirect effect of light on the media, UVE1 expression was examined in S. pombe, a “blind” species because it encodes no homologs of the WCC. We observed equal transcript levels of the S. pombe UVE1 homolog in cultures grown under light and dark conditions (Figure 4). These studies demonstrate that UVE1 is photoregulated among highly-diverged fungal species, and substantiates the WC-1 dependent light-induction of UVE1 in the fungal kingdom.
The protein sequences predicted for the two isoforms in var. neoformans from RACE were examined by bioinformatic approaches for their subcellular localization. PSORT II and MitoProt analysis of UVE1 light and dark isoforms predicts the longer form to be most likely mitochondrial and no specific localization pattern for the dark isoform. To confirm these predictions, light and dark isoforms of UVE1 were fused to the N-terminal end of GFP and expressed in the uve1Δ strain AI191. To assess mitochondrial localization we used MitoTracker red, which specifically stains respiring mitochondria. Confocal fluorescence microscopy for the GFP-fused light form of Uve1 showed co-localization of Uve1-GFP with MitoTracker red, giving a yellow fluorescence in merged images (Figure 5A). No expression of Uve1-GFP light form was observed in the nucleus, confirmed by co-staining with Hoechst (Figure S6). For the dark isoform of Uve1-GFP, GFP localization was throughout the cell, but clearly excluded from the mitochondria (Figure 5B; Figure S6). Transformation of the UVE1-GFP constructs into a uve1Δ genetic background enabled a test of their functionality in complementing the UV sensitive phenotype. Expression of the light form of UVE1 rescued in part the UV sensitive phenotype of strain AI191; however, the dark isoform fused to GFP did not. These experiments suggest that the light isoform of UVE1 is localized solely to the mitochondria, and as such it protects the mitochondrial genome from lethal effects of UV-induced DNA damage in C. neoformans.
The function of Uve1 in the fungi at a biochemical level is best characterized in S. pombe [25], [29], [40], [41]. Alignment of the two homologs suggests that they are similar, sharing residues within the active site of the enzyme (Figure S3). To infer functional similarity, a cross-species complementation test was performed. An uve1Δ knockout was generated in S. pombe by replacing the gene via homologous recombination with the KanMX cassette that confers resistance to G-418. We then expressed in this S. pombe knockout strain the cDNA clones of light or dark isoforms of UVE1 from C. neoformans var. neoformans. The UV sensitivity of the S. pombe strains was tested (Figure 6A, B). The uve1 deletion strain was highly sensitive to UV irradiation, as was the control strain transformed with the empty vector. In the strain expressing the light isoform of UVE1, UV sensitivity was rescued as the strain survived UV doses equivalent to the wild type strain. For the uve1::KanMX+C. neoformans dark isoform, no rescue in the UV sensitive phenotype was observed. These observations suggest that the C. neoformans light isoform of UVE1 is functionally active and has the equivalent biochemical functions of Uvde from S. pombe required for repair of DNA damaged by UV exposure. It also suggests that the dark isoform of C. neoformans Uve1 may not have any function in conferring protection against UV.
C. n. var neoformans Uve1 was localized in S. pombe as a GFP fusion to see if the rescue of UV sensitivity phenotype in S. pombe is due to complementing mitochondrial or nuclear genome repair by Uve1 (Figure 6C, D). The Uve1-GFP construct was functional, complementing the UV sensitive phenotype of the S. pombe uve1 mutation (Figure 6A, B). The localization of Uve1 (L)-GFP is in part nuclear, as confirmed by co-localization with the nuclear Hoechst stain (Figure 6C). No localization in mitochondria was observed (Figure 6D). These results suggest that the Cryptococcus Uve1 protein, which seems to be important for mitochondrial DNA repair in C. neoformans, also plays a role in nuclear DNA repair in S. pombe. This observed localization pattern conforms to previous reports where Uve1 in S. pombe repairs nuclear DNA after UV stress, rather than mitochondrial DNA [32].
If Uve1 localizes to mitochondria in C. neoformans, it is expected to play a role in mitochondrial DNA repair in consequence of DNA damage due to UV stress. As no nuclear localization was observed for Uve1, a negligible role of this endonuclease is expected in nuclear DNA damage repair. We performed a PCR-based DNA damage assay to assess the role of Uve1 in mitochondrial and nuclear DNA repair post-UV stress. The assay is based on the principle that damaged DNA impedes the progression of Taq polymerase on the template DNA in a PCR reaction [42]. Hence, there is an inverse relationship between the amount of DNA damage and PCR amplification products. Long template size increases the sensitivity of assay, as the longer the DNA the more chances of encountering damage (dimers). Small template amplification serves as a control, minimizing chances of encountering a damaged DNA strand, and is used for normalization of the amount of starting DNA or mitochondrial DNA copy number.
We compared DNA damage between the nuclear genome and mitochondrial genome of UV treated samples for both wild type and uve1Δ (Figure 7). After one hour there is delayed repair for both nuclear and mitochondrial genomes in the uve1Δ strain compared to wild type, probably reflecting retrograde signaling between the mitochondria and nucleus or that repair pathways of the nuclear genome can be ATP dependent [43]–[48]. The key observation is that in the uve1Δ strain there is lag in the mitochondrial repair. For later time points of 4 hour and 6 hours, the uve1Δ strain repair of mitochondrial genome is delayed as compared to wild type strain, which returned to normal (Figure 7). At 6 hours recovery, amplification of the mitochondrial genome in the wild type is as it was prior to DNA damage, but for uve1Δ the lesion damage still persists. However, the uve1Δ strain shows more efficient repair of lesions in the nuclear genome. For the 4 hour and the 6 hour time points, as the repair process of the mitochondrial genome initiates in uve1Δ by some unknown mitochondrial DNA repair enzymes, the repair of the nuclear genome is faster and comparable to wild type levels. These data implicate Uve1 in the efficient repair of UV-damaged mitochondrial DNA, with evidence for a complex interplay between mitochondrial and nuclear repair and the contributions of other repair pathways.
No induction of the UVE1 transcript encoding the functional form of the protein was observed under light conditions in bwc1Δ mutants (Figure 2). We examined if UVE1 is a direct target of the Bwc1-Bwc2 complex through two approaches. First, we tested if overexpression of Uve1 could rescue the UV hypersensitive phenotype of the bwc1Δ mutant in C. neoformans. Uve1 from var. neoformans was expressed under a galactose-inducible promoter (PGAL7) in the bwc1 deletion mutants of var. neoformans and var. grubii. The strains were grown overnight in media with glucose or galactose as the primary carbon source, serial diluted and plated, and UV sensitivity tests performed. Results from the var. neoformans strains are illustrated in Figure 8, and var. grubii in Figure S7. Uve1 overexpression rescued the UV sensitive phenotype of bwc1Δ as comparable to wild type when induced by galactose. In contrast, there was only slight rescue in strains grown in non-inducing glucose (Figure 8). For the bwc1Δ+PGAL7-UVE1 overexpression strains and wild type strains we performed northern analysis to check the levels of UVE1 induction (Figure S8). The levels of UVE1 transcripts were comparable between the galactose-induced PGAL7-UVE1 and the light-induced wild type strains. These results provide one piece of evidence for the Bwc1-Bwc2 complex directly controlling UV resistance in C. neoformans through regulation of the effector protein Uve1.
The second piece of evidence that UVE1 is a direct target of Bwc1-Bwc2 comes from gel mobility shift assays. Bwc2 has a C-terminal GATA-type zinc finger whose binding target sites are not known in C. neoformans. We searched the promoter of UVE1 for putative Bwc2 binding sites based on those used by the WCC of N. crassa [11], [14], [49]. For instance, one light regulated element (LRE) found in the frq promoter is TCGATCCGCTCGATCCCCT, with the underlined nucleotides similar to a TCGATCTTCATCTCGATCTCCA sequence found in the promoter of C. neoformans UVE1. We amplified and radiolabeled the UVE1 promoter region with this site and performed gel mobility shift assays with recombinant Bwc2 (amino acids 26 to 383) expressed and purified from Escherichia coli (Figure 9A). A retardation in gel migration of the UVE1 promoter DNA was observed, that increased with higher Bwc2 concentration or by adding zinc which is expected for a zinc finger protein (Figure 9B). The nature of the higher mobility forms is unknown, but may represent aggregation of Bwc2 monomers. Control interactions using a non-specific DNA fragment confirmed the specificity of Bwc2 for the UVE1 promoter. These observations indicate that the UVE1 promoter is a direct target for Bwc2 binding.
Light influences diverse aspects of fungal biology, presumably by acting on unique pathways in specific species. However, the potential for conserved regulation also exists, and this is predicted to reflect the original selective pressure(s) and current maintenance of light-sensing in extant fungi. Some responses to light in ascomycete fungi relate to protection against damage caused by light. For instance, expression of the DNA repair enzyme photolyase and genes for biosynthesis of carotenoid pigments are often induced by light. There are previous reports of links between light via its input in circadian rhythms with DNA repair in fungi, as well as in mice and humans. For instance PRD-4 is a checkpoint kinase 2 homolog in N. crassa that is regulated by the WCC and contributes to DNA repair [21]. Another DNA repair protein, XPA, has been shown to have circadian rhythm dependent oscillations in mouse brain [50], [51]. However, between 2.8–6% of genes are regulated at the transcript level in response to a light exposure in ascomycete species [11], [52], [53], yielding a long list of candidate genes for further analysis. In contrast, the basidiomycete C. neoformans may serve as a simpler model for understanding the evolution of light-sensing in fungi, because (a) it does not encode a photolyase gene and pigmentation is not induced by light, (b) few genes are induced in response to light at the transcript level [19], (c) there is no evidence of photoadaptation [17], a trait that influences the intensity of the response to light, and (d) the White collar complex contains only one protein with a zinc finger DNA binding domain [17], [18].
Here, we identify the UVE1 gene as a downstream target of the WCC in C. neoformans, and show that homologs are also light regulated in species that represent two other major branches in the fungal kingdom. We suggest that UVE1 acts as the key factor controlling the UV sensitive phenotype caused by mutating BWC1 or BWC2 in C. neoformans (Figure 1, Figure 10, Figure S4), and likely plays similar roles in other fungi to survive the deleterious effects of sunlight, which has UV wavelengths as an inevitable DNA damaging component.
Northern blot analysis and characterization of 5′ and 3′ ends of C. neoformans var. neoformans UVE1 showed two transcripts of differing size for the UVE1 gene. The functional complementation experiments (Figure 6) in S. pombe uve1Δ by the homologous UVE1 long isoform implies that the C. neoformans protein has similar DNA repair activities; such as against cyclobutane pyrimidine dimers, 6-4 photoproducts, apurinic/apyrimidinic sites, and stretches of single-stranded DNA nicks or gaps. Moreover the localization of C. neoformans Uve1-GFP in S. pombe is in part nuclear rather than mitochondrial (Figure 6). This explains the functional complementation of the UV stress tolerance phenotype in S. pombe uve1Δ strains, as in S. pombe the nuclear UVER pathway is attributed for survival under UV stress [32]. The UVE1 short isoform did not show any functional complementation. Bioinformatic analysis to identify the active domain of UVE1 (pfam03851) provides a possible explanation for the inactivity of the short isoform, because it does not encode the complete conserved region (Figure 3, Figure S3). Another possibility may be the absence of subcellular localization signals on the dark form, such that the protein is rendered inactive due to improper compartmentalization. Correct subcellular localization of proteins involved in DNA repair has been implicated in countering genotoxic stress [54], as improper localization can result in loss-of-function and may even lead to disease development in humans [55], [56].
The light isoform of Uve1 in Cryptococcus localizes to the mitochondria, as shown by Uve1-GFP fusion studies. C. neoformans is an obligate aerobe: it cannot survive loss of mitochondrial function from something like unrepaired DNA damage. Based on the following evidence: (a) no observed localization of Uve1 in nucleus (Figure S6); (b) localization of Uve1 to mitochondria (Figure 5); (c) strains with the UVE1 gene deleted exhibited reduced survival under UV stress and Uve1-GFP partially complements the UV sensitive phenotype of the uve1Δ strain; and (d) reduced mitochondrial DNA damage repair in uve1Δ strain comparatively to the wild type strain (Figure 7) suggest that Uve1 in Cryptococcus is required for protection of mitochondrial DNA for survival under UV stress. Overexpression of UVE1 in bwc1 mutants of either var. grubii or var. neoformans restores UV sensitivity to the wild type level (Figure 8), and recombinant Bwc2 physically binds to the promoter of UVE1 to cause a gel mobility shift (Figure 9B). These data further corroborate the hypothesis that UVE1 is a direct downstream target of Bwc1.
The light-induced genes in C. neoformans were previously sought by a whole genome microarray expression analysis in var. neoformans [19]. The UVE1 gene was not identified in that study although UVE1 transcript data are present for five of the six biological replicates, with an average 1.15 fold difference between dark and light treatment. By contrast, quantification by ImageJ of our northern blot data normalized to actin levels indicates that the longer isoform is upregulated about 16 fold in the light (Dataset S1). UVE1 remained undetected in microarray experiments because the 70-mer probe on the array is common to both light and dark transcript isoforms (Figure 3). The dark isoform must be under control of another transcription factor, using elements within the light isoform to drive transcription. These findings demonstrate one limitation of the microarray technique in comparison to more comprehensive transcript analysis techniques like tiling arrays or RNA-seq, or to conventional hybridization techniques like northern blotting that can detect alternative transcripts.
Two other phenotypes associated with mutation of BWC1 or BWC2 in C. neoformans are loss of the inhibition of mating by light and reduced virulence. To further verify the role of Uve1 in other BWC related phenotypes, we examined the role of Uve1 in C. neoformans mating by crossing uve1Δ strains of both mating types under light and dark conditions. We did not find any contribution of this gene in mating, as the uve1Δ strains behaved like wild type for repression of mating by light (Figure S2). Similarly, the phenotype of the uve1Δ mutant in animal studies does not phenocopy that of bwc1Δ or bwc2Δ. A large-scale analysis of virulence has been undertaken in C. neoformans, measuring competitive survival of strains in mouse lungs [38]. Both bwc1Δ and bwc2Δ strains showed reduced proliferation, consistent with their reported role in virulence [17]. In contrast, the uve1Δ mutant had no defect in this virulence assay. We corroborated these results in a wax moth larvae model of virulence (Figure S2). We also examined a possible role of UVE1 under oxidative stress based on the mild phenotype observed in S. pombe [31], but did not find any phenotypic difference in the uve1Δ strains compared to wild type. Thus, we postulate that Bwc1-Bwc2 modulates its function via more than one downstream target (Figure 10A); of which those for virulence and mating remain to be discovered in future studies. We propose a model for the relationship between light-sensing via the WCC and Uve1 function in the mitochondria in C. neoformans (Figure 10B). It is possible that this model applies also to other fungal species for the protection of mitochondrial, nuclear or both genomes under UV stress.
The White Collar complex is conserved across the fungal kingdom, with homologs present in the chytrids, Mucoromycotina, Glomeromycotina, Ascomycota and Basidiomycota. However the complex has been lost in some fungal lineages, like its absence from the Saccharomycotina [1], [2], [57]. One important question is whether any WCC downstream targets are conserved. Transcript comparisons by northern blot analysis in N. crassa and P. blakesleeanus, members of the Ascomycota and Mucoromycotina, demonstrate that UVE1 homologs are photo-regulated in at least one species in each of these fungal groups. The UVE1 homolog is also induced by light, as measured by microarray studies, in Aspergillus nidulans and N. crassa [11], [53]. Absence of regulation in bwc1 and madA-madB mutants in N. crassa and P. blakesleeanus further implicate White Collar-dependent regulation of UVE1. This suggests that in fungi that have White Collar, UVE1 is regulated in a light-dependent manner, and this regulation is lost or alternative regulation evolved in fungal species missing White Collar proteins. The presence of LOV domain containing flavin-binding photoreceptor proteins and UVE1 homologs in bacteria, like Bacillus subtilis [25], [58], warrant further examination if through convergent evolution the LOV domain type photoreceptors might be involved in regulation of UVE1 expression even more widely.
The repair of photo-damage by Uve1 is conserved in many fungi and important under UV stress, irrespective of the presence of base excision (BER) or nucleotide excision repair (NER) pathways. Repair of DNA damage from UV by Uve1 is faster in comparison to NER [40]; under ancient environments in which ultraviolet levels were higher than today Uve1 could have provided a selective advantage. We estimate from genome sequencing projects that 95% of the fungal genomes encoding WC-1 also encode a copy of UVE1. Many of the exceptions have homologs of photolyase present in their genome, which may play an equivalent role as Uve1, and can also repair mitochondrial DNA [59], [60].
In summary, light triggers a number of physiological and morphological changes in fungi. The advantages of using light as a signal that are conserved have remained unclear although there is increasing evidence for a role in protecting cells from damage. Here, we demonstrate that protection of DNA, including the mitochondrial genome, through photo-regulation of Uve1 provides a benefit that is present in fungi that are able to sense light through the White Collar Complex.
Gene knockout cassettes were constructed by fusion of around 1000 bp flanks 5′ and 3′ of the UVE1 gene with nourseothricin acetyltransferase (NAT) coding sequence for strain JEC21 (var. neoformans, serotype D) and neomycin phosphotransferase (NEO) coding sequence for strain KN99α (var. grubii, serotype A). Oligonucleotide primer sequences are listed in Table S1. To make JEC21 uve1Δ, 5′ and 3′ gene flanks were amplified by primer set AISV030/AISV034 and AISV032/AISV035, respectively, using JEC21 genomic DNA. To make KN99α uve1Δ, 5′ and 3′ gene flanks were amplified by primer set ai830/ai831 and ai832/ai833, respectively, using KN99α genomic DNA. The NAT and NEO ORFs were amplified by primer set ai290/ai006. Overlap PCR was performed to obtain 5′-UVE1-NAT-UVE1-3′ and 5′-UVE1-NEO-UVE1-3′ cassettes by mixing equimolar ratio of 5′-UVE1, NAT, UVE1-3′ for JEC21 and 5′-UVE1, NEO, UVE1-3′ for KN99α. Primers used to perform overlap PCR were AISV030/AISV032 and ai830/ai833 for JEC21 and KN99α, respectively. About 2 µg of 5′-UVE1-NAT-UVE1-3′ or 5′-UVE1-NEO-UVE1-3′ cassette was transformed into strains JEC21 and KN99α using biolistic delivery with a PDS 1000/He particle delivery system (Bio-Rad, Hercules, CA) [61]. Gene replacement was confirmed by PCR and Southern blots for the correct integration of the gene cassette. Strains and genotypes are provided in Table S2.
For complementation of UVE1 in serotype A, a wild type copy of UVE1 was amplified with primers ALID0001 and ALID0002 and cloned into the pCR2.1 TOPO plasmid. The insert was excised with BamHI-XhoI and subcloned into the BamHI-SalI site of pPZP-NATcc. The plasmid was transformed into strain AI191 (uve1::NEO) by biolistics, with positive transformants selected for growth on yeast extract-peptone-dextrose (YPD)+nourseothricin (100 µg/ml) plates. The plasmids used or constructed in this study are listed in Table S3.
The UV sensitivities of strains were tested by applying UV stress in an XL-1500 UV cross linker (Spectronics Corporation, Lincoln, NE). Unless otherwise stated, strains were exposed to the laboratory ambient light (400–800 LUX) during experiments.
Cultures were prepared for C. neoformans strains KN99α, AI81 (bwc1Δ), JEC21, AI5 (bwc1Δ), C. gattii (R265), S. pombe L972, N. crassa wild-type FGSC 4200, N. crassa (wc-1) FGSC 4398, P. blakesleeanus wild-type NRRL1555, and P. blakesleeanus (madA madB) mutant L51. All strains of each species were plated with equal optical density or numbers of spores in duplicates on 15 cm diameter petri dishes containing YPD, and kept in darkness. For N. crassa only, 50 ml liquid cultures were grown in 50 ml YPD medium. Cultures were grown for 23 h or 47 h depending on the growth kinetics of the species. On completion of 23 h or 47 h, one of the sets was exposed to cool white light (dual Sylvania 4100 K 32W bulbs) of 1600–2600 Lux for 1 h and the other set left in the dark. On completion of the 24 h or 48 h time periods both the light and dark cultures were scraped in the light or under safe red light (GBX LED safelight, Kodak, Rochester, NY). N. crassa cultures were harvested directly from the liquid medium. All cultures were pelleted, frozen using dry ice+ethanol, lyophilized and stored at −80°C.
Total RNA was isolated using Trizol reagent (Invitrogen, Grand Island, NY). To address the low transcript abundance of UVE1, total RNA was further purified for polyA mRNA isolation starting with 1 mg total RNA, using the PolyATract Kit (Promega, Madison, WI), except for P. blakesleeanus and C. gattii where 40 µg of total RNA were used. RNA samples were resolved on 1.4% agarose denaturing formaldehyde gels and blotted on to Zeta Probe membrane (Bio-Rad). Probes for northern analysis were amplified using specific primer sets (Table S4) and radiolabeled with [α-32P] dCTP (PerkinElmer, Waltham, MA) using the RediPrime II labeling kit (Amersham, Pittsburg, PA). The blots were stripped and re-probed with fragments of actin homologs as loading controls. Autoradiograms were scanned, and transcript levels were compared by ImageJ analysis (Dataset S1).
The localization of the two isoforms of Uve1 within the cell was assessed by fusions to green fluorescent protein (GFP). C. n. var. neoformans strain JEC21 genomic DNA was used as the template to amplify the UVE1 light (L) isoform using primers AISV001/AISV003 and UVE1 dark (D) isoforms using primers AISV002/AISV003. The histone 3 promoter for Cryptococcus (PH3) and GFP-NAT were amplified from the pPZP-GFP-NATcc plasmid using ai255/AISV005 (overlap primer for L) or ai255/AISV006 (overlap primer for D) for PH3, and AISV004/ai256 for GFP-NAT. The overlap construct was amplified by mixing equimolar ratios of the three amplicons for the light and dark isoforms using primers M13F and M13R. About 2 µg of gel purified UVE1 L and D overlap constructs were transformed into strain AI191 by biolistics. Positives clones were selected by their growth on YPD+100 µg/ml nourseothricin plates and confirmed by PCR, DNA sequencing, western blotting for GFP, and fluorescence signal.
Strains AISVCN28 and AISVCN02 were used for localization of the Uve1 light and dark isoforms fused to GFP. Strains were stained with MitoTracker Red CMXRos (Invitrogen) at 3 nM, kept in the dark for 20 min, washed and suspended in phosphate buffered saline (PBS) and used for microscopy. Cells were imaged using Olympus confocal microscopes FLUOVIEW FV10i or FV300.
Cultures for C. neoformans strains KN99α (wild type) and AI191 (uve1Δ) were grown overnight in YPD and washed with distilled water. Cells were suspended in phosphate buffered saline to 4×104 cells/ml. For each strain totals of 150 ml cells were distributed in 30 ml aliquots for time points 0 min (after stress), 1 h, 4 h, 6 h and control (no stress). For each aliquot the cells were placed in 15 cm petri dishes and exposed to UV light (50 J/m2) using a UV cross linker. Immediately after the UV stress, cells were transferred to 50 ml tubes and kept on ice. Control and 0 min cells were pelleted and frozen in liquid nitrogen. For 1 h, 4 h and 6 h time points, cells were re-suspended in YPD and incubated at 30°C for these respective times, then centrifuged to pellet and snap frozen. All samples were lyophilized, and DNA was extracted by the CTAB buffer method [62].
The relative DNA damage to the mitochondrial and nuclear genomes were assessed using a PCR assay based on established methods [42]. Concentrations of DNA samples from each treatment were standardized by measuring them by spectrophotometry and making appropriate dilutions. Primers used for amplification of fragments of the mitochondrial genome were AISV87/AISV91 (11 Kb). PCR conditions for long mitochondrial PCR were 94°C 4 min, 23 cycles for 98°C 10 s, 68°C 15 min, and a final extension of 72°C 10 min using Ex Taq (Takara, Kyoto, Japan). For nuclear long amplification (8 kb) primers were AISV85/AISV95. Conditions for long nuclear PCR were 94°C 4 min, 23 cycles for 94°C 20 s, 58°C 20 s, 72°C 6 min and a final extension of 72°C 7 min. Short amplification primers for mitochondrial genome were AISV89/AISV99 and for the nuclear genome AISV85/AISV97 amplifying about 250 bp. PCR conditions for small mitochondrial and nuclear amplicons were 94°C 4 min, 23 cycles for 94°C 20 s, 55°C 20 s, 72°C 1 min, and a 72°C 7 min final extension. All PCR amplicons were resolved on agarose gels, and intensities were quantified using ImageJ software. DNA damage was compared by calculating relative amplification of large PCR fragments of the UV treated samples to that of the respective untreated controls using the method reported in reference [42], and adjusting for differences between nuclear and mitochondrial amplicon sizes.
The Galleria mellonella virulence assay followed methods that were previously described [63]. Overnight cultures in YPD medium for strains KN99α, AI191 and AI181 were washed three times with PBS. Cells were suspended in PBS to 2×107 cells/ml. For each strain 11–12 larvae were injected with 5 µl of the cells, as well as the control PBS. Wax moth were incubated at 37°C and survival monitored daily.
An S. pombe uve1 knockout strain was constructed to serve for the functional analysis of UVE1 isoforms from C. neoformans. For the construction of the gene knockout cassette, genomic DNA from S. pombe strain L972 was used to amplify around 320 bp 5′-uve1 and 300 bp 3′-uve1 fragments using primer pairs AISV007a/AISV009 and AISV010/AISV011, respectively. The KanMX fragment was amplified using primer set AISV007/AISV008 from plasmid pFA6a-GFP(S65T)-kanMX6. Overlap PCR was performed to generate the gene knockout cassette using primer set AISV007a/AISV011. Around 2 µg of the PCR construct were transformed into S. pombe (strain MM72-4A ura4-D18 h−) by lithium acetate transformation and cells plated on to YPD+100 µg/ml G-418. Gene knockouts were confirmed by PCR and Southern blotting. S. pombe uve1 knockout strain AISVSP1 was selected for C. neoformans UVE1 complementation studies.
UVE1 cDNA was reverse transcribed from RNA of C. neoformans strain JEC20 using Superscript III First strand Synthesis System (Invitrogen), as per company instructions (JEC20 is isogenic to JEC21, with a different MAT allele; [64]). The synthesized cDNA was amplified by site directed mutagenesis to abolish an NdeI site inconvenient for subcloning while conserving the encoded amino acid residue, and to introduce NdeI and BamHI restriction sites at the start and end of the UVE1 gene. Primers used for amplification of fragment 1 of the L and D form were AISV014/AISV013 and AISV015/AISV013, respectively. Primers used for amplification of fragment 2 were AISV012/AISV016. Overlap PCR was performed to amplify full L and D genes from fragment 1 and 2, using primers AISV014/AISV016 for UVE1 L form and AISV015/AISV016 for UVE1 D form. The NdeI and BamHI digested cassettes were ligated into the NdeI-BamHI site in the pREP42 vector enabling expression from an nmt promoter [39]. Positive clones were confirmed by sequencing. Plasmids containing UVE1 L and D isoforms were transformed into S. pombe strain AISVSP1 (ura4-D18 uve1::kanMX) by the lithium acetate method. Empty vector pREP42 was transformed into strain AISVP1 as a control. Positive S. pombe transformants were selected on minimal medium without uracil and were confirmed by PCR.
C. neoformans var. neoformans Uve1 (L) C- terminal GFP localization in S. pombe was done by fusion of UVE1 (L) to GFP by overlap PCR. For fragment 1, UVE1 (L) was amplified from pREP42-UVE1 (L) using primers AISV014/AISV003 and the fragment 2, GFP, was derived from pPZP-GFP-NATcc using primers AISV004/AISV066. Overlap PCR joining fragments 1 and 2 was performed by primer set AISV014/AISV066. The overlap PCR product was cloned into pCR 2.1 TOPO, and transformed by heat shock into E. coli DH5α. Positive clones were selected and sequenced. A plasmid containing the desired Nde1-UVE1-GFP-BamHI overlap was digested with NdeI and BamHI. The Nde1-UVE1-GFP-BamHI digest was ligated with NdeI and BamHI digested pTN157. Plasmid pTN157-UVE1-GFP was transformed into S. pombe strain AISVSP1 (genotype ura4-D18 uve1::kanMX) by the lithium acetate method. Positive S. pombe transformants were selected on minimal medium without uracil and were confirmed by PCR. Transformants were examined for fluorescence signal and their UV resistant phenotype. Confocal microscopy was performed for strain AISVSP15. For both mitochondrial and nuclear staining, cells were grown in Edinburgh Minimal Medium (EMM). Mitochondrial staining was performed with MitoTracker Red CMXRos (Invitrogen) at a final concentration of 3 nM in water, kept in the dark for 20 min, washed and suspended in PBS and used for microscopy. Hoechst 33342 was used to stain the nucleus. Cells grown in EMM were washed and suspended in Hoechst (1 µg/ml in water) for 10 min. Cells were washed and suspended in PBS, and microscopy was performed.
Overnight cultures for strains L972, AISVSP1, AISVSP2, AISVSP3, AISVSP4 and AISVSP15 in YES media were subcultured the following day. Strains in exponential phase were dotted on YES medium in ten-fold serial dilutions. One set of plates was exposed to UV light (120 J/m2) using the UV cross linker, and another plate kept unexposed. Both UV treated and unexposed sets were incubated at 30°C for 2 days. For UV dose response experiments exponentially growing cells for strains at the same optical density were ten-fold serially diluted and equal volumes for all dilutions of the cells were plated on YES media. The control was kept unexposed to UV and others were exposed at UV doses of 60, 120 and 180 J/m2. All plates were incubated at 30°C for 3 days, and colony forming unit data analyzed for percentage survival.
The UVE1 gene was over-expressed in bwc1 knockout backgrounds to assess if UVE1 can rescue the UV sensitive phenotype of bwc1 mutation. A fusion cassette of the JEC21 GAL7 promoter (PGAL7) and UVE1 gene was made by overlap PCR. The GAL7 promoter was amplified by primer set AISV025/AISV028 and UVE1 was amplified using primer set AISV027/AISV026 on JEC21 genomic DNA. The PGAL7-UVE1 fusion cassette was amplified by mixing GAL7 promoter and UVE1 PCRs in equimolar ratios by primer set AISV025/AISV026. The PGAL7-UVE1 fusion cassette was cloned in a TA vector (pCR 2.1 TOPO; Invitrogen), and transformed by heat shock in E. coli Top10 cells. Positive clones were selected and sequenced. A plasmid containing PGAL7 -UVE1 was digested with BamHI and XhoI. The fragment was ligated with pPZP-NEO Agrobacterium vector digested with the same enzymes, and transformed into Top10 cells. Positive clones were selected by their growth on kanamycin and verified by PCR and restriction digestion. The pPZP-NEO-PGAL7-UVE1 vector was transformed by electroporation into A. tumefaciens strain EHA105. Selected positive Agrobacterium transformants were co-cultured with C. neoformans strains AI5 and AI81. Positive transformants were selected for growth on YPD+G-418+cefotaxime plates. Transformants were cultured overnight in yeast nitrogen base (YNB) medium +2% galactose or 2% glucose, and 10-fold serial dilutions placed on YPD medium plates. Dotting was performed in duplicate and one set was exposed to UV radiation at 120 J/m2. Both sets were incubated at 30°C for 2 days.
JEC21 RNA was reverse transcribed to make cDNA using the Superscript III First strand Synthesis System (Invitrogen). A fragment of BWC2 cDNA was amplified using primers AISV040/AISV041, containing BamHI and EcoRI restriction sites. The amplicon was digested with BamHI and EcoRI, and ligated into the pRSETA vector (Invitrogen) digested with the same enzymes. Top10 cells were transformed with the ligation product. An error free clone was identified by sequencing. The plasmid containing BWC2 was transformed into BL21(DE3)pLysS cells and selected on ampicillin+chloramphenicol SOB medium plates. Protein induction and expression using 1 mM IPTG was performed as described for the pRSET expression system by Invitrogen. The (Histidine)6-tagged Bwc2 protein was semi-purified using Pure Proteome Nickel Magnetic beads (Millipore Corporation, Billerica, MA) as per the manufacturer's instructions.
For electrophoretic mobility shift assays (EMSA), a 244 bp (P1) region 5′ of the start codon of UVE1 containing putative Bwc2 binding sites was amplified using primer set AISV019/AISV020. A control nonspecific DNA fragment (NS) of 278 bp was amplified from pRS426 vector using primers ALID1229/ALID1230. About 600 ng of the amplified fragment from UVE1 promoter region and nonspecific probe was radiolabeled with [γ-32P] dATP (PerkinElmer) using T4 polynucleotide kinase (New England Biolabs, Ipswich, MA). Labeling conditions were 1X T4 polynucleotide buffer, 5 µl 6000 Ci/mmol γ-32P ATP, 10 units T4 polynucleotide kinase in a 50 µl reaction mixture at 37°C for 30 min. The radiolabeled probes were purified using PCR purification columns (Qiagen, Germantown, MD). The composition of 5X EMSA binding buffer used was 20% w/v glycerol, 5 mM MgCl2, 250 mM NaCl, 2.5 mM EDTA pH 8, 10 mM DTT with or without 12 µM ZnSO4 (reference [65] with slight modifications). 20 to 50 µg of purified Bwc2 protein were pre-incubated with 1X EMSA binding buffer for 20 min at room temperature for all reactions. Total reaction mixture of 20 µl consisted of 1X EMSA binding buffer, 20 to 50 µg of purified protein, cold probe (if added), 1 µl of radiolabeled probe and 1X phosphate buffer, followed by 50 min incubation at room temperature. For competition reactions, 600 ng and 900 ng of the cold probes were included in the reaction mixture prior to addition of radiolabeled probes. At the end of 50 min incubation, samples were transferred to ice followed by addition of EMSA loading dye. Samples were loaded on 6% polyacrylamide Tris-borate-EDTA (TBE) gels (Invitrogen), and run at 130 V for 2 h in 1X TBE at 4°C. Autoradiography was performed by exposure to Gene Mate Blue ultra autoradiography films.
|
10.1371/journal.pcbi.1005881 | Firing rate equations require a spike synchrony mechanism to correctly describe fast oscillations in inhibitory networks | Recurrently coupled networks of inhibitory neurons robustly generate oscillations in the gamma band. Nonetheless, the corresponding Wilson-Cowan type firing rate equation for such an inhibitory population does not generate such oscillations without an explicit time delay. We show that this discrepancy is due to a voltage-dependent spike-synchronization mechanism inherent in networks of spiking neurons which is not captured by standard firing rate equations. Here we investigate an exact low-dimensional description for a network of heterogeneous canonical Class 1 inhibitory neurons which includes the sub-threshold dynamics crucial for generating synchronous states. In the limit of slow synaptic kinetics the spike-synchrony mechanism is suppressed and the standard Wilson-Cowan equations are formally recovered as long as external inputs are also slow. However, even in this limit synchronous spiking can be elicited by inputs which fluctuate on a time-scale of the membrane time-constant of the neurons. Our meanfield equations therefore represent an extension of the standard Wilson-Cowan equations in which spike synchrony is also correctly described.
| Population models describing the average activity of large neuronal ensembles are a powerful mathematical tool to investigate the principles underlying cooperative function of large neuronal systems. However, these models do not properly describe the phenomenon of spike synchrony in networks of neurons. In particular, they fail to capture the onset of synchronous oscillations in networks of inhibitory neurons. We show that this limitation is due to a voltage-dependent synchronization mechanism which is naturally present in spiking neuron models but not captured by traditional firing rate equations. Here we investigate a novel set of macroscopic equations which incorporate both firing rate and membrane potential dynamics, and that correctly generate fast inhibition-based synchronous oscillations. In the limit of slow-synaptic processing oscillations are suppressed, and the model reduces to an equation formally equivalent to the Wilson-Cowan model.
| Since the seminal work of Wilson and Cowan [1], population models of neuronal activity have become a standard tool of analysis in computational neuroscience. Rather than focus on the microscopic dynamics of neurons, these models describe the collective properties of large numbers of neurons, typically in terms of the mean firing rate of a neuronal ensemble. In general, such population models, often called firing rate equations, cannot be exactly derived from the equations of a network of spiking neurons, but are obtained using heuristic mean-field arguments, see e.g. [2–6]. Despite their heuristic nature, heuristic firing rate equations (which we call H-FRE) often show remarkable qualitative agreement with the dynamics in equivalent networks of spiking neurons [7–10], and constitute an extremely useful modeling tool, see e.g. [11–28]. Nonetheless, this agreement can break down once a significant fraction of the neurons in the population fires spikes synchronously, see e.g. [29]. Such synchronous firing may come about due to external drive, but also occurs to some degree during spontaneously generated network states.
As a case in point, here we focus on partially synchronized states in networks of heterogeneous inhibitory neurons. Inhibitory networks are able to generate robust macroscopic oscillations due to the interplay of external excitatory inputs with the inhibitory mean field produced by the population itself. Fast synaptic processing coupled with subthreshold integration of inputs introduces an effective delay in the negative feedback facilitating the emergence of what is often called Inter-Neuronal Gamma (ING) oscillations [30–38]. Modeling studies with networks of spiking neurons demonstrate that, in heterogeneous inhibitory networks, large fractions of neurons become frequency-entrained during these oscillatory episodes, and that the oscillations persist for weak levels of heterogeneity [30, 32, 34]. Traditional H-FRE (also referred to as Wilson-Cowan equations) fail to describe such fast oscillations. To overcome this limitation, explicit fixed time delays have been considered in H-FRE as a heuristic proxy for the combined effects of synaptic and subthreshold integration [9, 10, 36, 39].
Here we show that fast oscillations in inhibitory networks are correctly described by a recently derived set of exact macroscopic equations for quadratic integrate-and-fire neurons (that we call QIF-FRE) which explicitly take into account subthreshold integration [40]. Specifically, the QIF-FRE reveal how oscillations arise via a voltage-dependent spike synchronization mechanism, missing in H-FRE, as long as the recurrent synaptic kinetics are sufficiently fast. In the limit of slow recurrent synaptic kinetics intrinsically generated oscillations are suppressed, and the QIF-FRE reduce to an equation formally identical to the Wilson-Cowan equation for an inhibitory population. However, even in this limit, fast fluctuations in external inputs can drive transient spike synchrony in the network, and the slow synaptic approximation of the QIF-FRE breaks down. This suggests that, in general, a correct macroscopic description of spiking networks requires keeping track of the mean subthreshold voltage along with the mean firing rate.
Additionally, the QIF-FRE describe the disappearance of oscillations for sufficiently strong heterogeneity which is robustly observed in simulations of spiking networks. Finally, we also show that the phase diagrams of oscillatory states found in the QIF-FRE qualitatively match those observed in simulations of populations of more biophysically inspired Wang-Buzsáki neurons [30]. This shows that not only are the QIF-FRE an exact mean-field description of networks of heterogeneous QIF neurons, but that they also provide a qualitatively accurate description of dynamical states in networks of spiking neurons more generally, including states with significant spike synchrony.
Recurrent networks of spiking neurons with inhibitory interactions readily generate fast oscillations. Fig 1 shows an illustration of such oscillations in a network of globally coupled Wang-Buzsáki (WB) neurons [30]. Panels (a,c) show the results of a numerical simulation of the network for fast synapses (time constant τd = 5 ms), compared to the membrane time constant of the neuron model (τm = 10 ms). Although the neurons have different intrinsic frequencies due to a distribution in external input currents, the raster plot reveals that fast inhibitory coupling produces the frequency entrainment of a large fraction of the neurons in the ensemble. This collective synchronization is reflected at the macroscopic scale as an oscillation with the frequency of the synchronous cluster of neurons [41, 42]. Indeed, panel (a) shows the time series of both the mean synaptic activation variable S, and the mean firing rate R, which display ING oscillations. Panels (b,d) of Fig 1 show the disappearance of the synchronous state when the synaptic kinetics is slow (τd = 50 ms).
A heuristic firing rate description of the spiking network simulated in Fig 1 takes the form [1, 5]
τ m R ˙ = - R + Φ ( - J τ m S + Θ ) , (1a) τ d S ˙ = - S + R . (1b)
where R represents the mean firing rate in the population, S is the synaptic activation, and the time constants τm and τd are the neuronal and synaptic time constants respectively [39, 43]. The input-output function Φ, also known as the f-I curve, is a nonlinear function, the form of which depends on the details of the neuronal model and on network parameters. Finally, J ≥ 0 is the synaptic strength and Θ is the mean external input current compared to threshold. In contrast with the network model, the H-FRE Eq (1) cannot generate sustained oscillations. In fact, a linear stability analysis of steady state solutions in Eq (1) shows that the resulting eigenvalues are
λ = - α ( 1 ± 1 - β ) , (2)
where the parameter α = (τm + τd)/(2τm τd) is always positive. Additionally, the parameter β = [4τm τd(1 + Jτm Φ′)]/(τm + τd)2 is also positive, since the f-I curve Φ(x) is an increasing function, and its derivative evaluated at the steady state is then Φ′ > 0. Therefore the real part of the eigenvalue λ is always negative and hence steady states are always stable, although damped oscillations are possible, e.g. for strong enough coupling J. Introducing an explicit fixed time delay in Eq (1) can lead to the generation of oscillations with a period on the order of about twice the delay [9, 10, 36]. Nonetheless, inhibitory networks of spiking neurons robustly show oscillations even in the absence of explicit delays, as seen in Fig 1. This suggests that there is an additional mechanism in the network dynamics, key for driving oscillatory behavior, which H-FRE do not capture.
Here we show that the mechanism responsible for the generation of the oscillations shown in Fig 1 is the interplay between the mean firing rate and membrane potential, the dynamics of which reflect the degree of spike synchrony in the network. To do this, we use a set of exact macroscopic equations which have been recently derived from a population of heterogeneous quadratic integrate-and-fire (QIF) neurons [40]. We refer to these equations as the QIF-FRE. The QIF-FRE with exponential synapses have the form
τ m R ˙ = Δ π τ m + 2 R V , (3a) τ m V ˙ = V 2 - ( π τ m R ) 2 - J τ m S + Θ , (3b) τ d S ˙ = - S + R . (3c)
where Δ is a parameter measuring the degree of heterogeneity in the network and the other parameters are as in the H-FRE Eq (1). Eq (3) are an exact macroscopic description of the dynamics in a large network of heterogeneous QIF neurons with inhibitory coupling. In contrast with the traditional firing rate equations Eq (1), they contain an explicit dependence on the subthreshold state of the network, and hence capture not only macroscopic variations in firing rate, but also in spike synchrony. Specifically, a transient depolarizing input which drives the voltage to positive values (the voltage has been normalized such that positive values are suprathrehsold, see Materials and methods) will lead to a sharp growth in the firing rate through the bilinear term in Eq (3a). Simulations in the corresponding network model reveal that this increase is due to the synchronous spiking of a subset of neurons [40]. This increase in firing rate leads to a hyperpolarization of the mean voltage through the quadratic term in R in Eq (3b). This term describes the effect of the neuronal reset. This decrease in voltage in turn drives down the mean firing rate, and the process can repeat. Therefore the interplay between mean firing rate and mean voltage in Eq (3) can generate oscillatory behavior; this behavior corresponds to transient bouts of spike synchrony in the spiking network model.
It is precisely the latency inherent in this interplay which provides the effective delay, which when coupled with synaptic kinetics, generates self-sustained fast oscillations. In fact, in the limit of instantaneous synapses (τd → 0), Eq (3) robustly display damped oscillations due to the spike generation and reset mechanism described in the preceding paragraph [40]. Contrast this with the dynamics in Eq (1) in the same limit: the resulting H-FRE is one dimensional and hence cannot oscillate.
On the face of things the Eq (3) appear to have an utterly distinct functional form from the traditional Wilson-Cowan Eq (1). In particular, the f-I curve in H-FRE traditionally exhibits an expansive nonlinearity at low rates and a linearization or saturation at high rates, e.g. a sigmoid. There is no such function visible in the QIF-FRE which have only quadratic nonlinearities. However, this seeming inconsistency is simply due to the explicit dependence of the steady-state f-I curve on the subthreshold voltage in Eq (3). In fact, the steady-state f-I curve in the QIF-FRE is “typical” in the qualitative sense described above. Specifically, solving for the steady state value of the firing rate in Eq (3) yields
R * = Φ ( - J τ m R * + Θ ) , (4)
where
Φ ( I ) = 1 2 π τ m I + I 2 + Δ 2 . (5)
The f-I curve Eq (5) is shown in Fig 2 for several values of the parameter Δ, which measures the degree of heterogeneity in the network. Therefore, the steady-state firing rate in Eq (3), which corresponds exactly to that in a network of heterogeneous QIF neurons, could easily be captured in a heuristic model such as Eq (1) by taking the function Φ to have the form as in Eq (5). On the other hand, the non-steady behavior in Eq (3), and hence in spiking networks as well, can be quite different from that in the heuristic Eq (1).
We can investigate the emergence of sustained oscillations in Eq (3) by considering small amplitude perturbations of the steady-state solution. If we take R = R* + δReλt, V = V* + δVeλt and S = S* + δSeλt, where δR, δV, δS ≪ 1, then the sign of the real part of the eigenvalue λ determines whether the perturbation grows or not. Plugging this ansatz into Eq (3) yields three coupled linear equations which are solvable if the following characteristic equation also has a solution
− 2 J τ m R * = ( 1 + τ d λ ) [ ( 2 π τ m R * ) 2 + ( τ m λ + Δ π τ m R * ) 2 ] . (6)
The left hand side of Eq (6) is always negative and, for τd = 0, this implies that the solutions λ are necessarily complex. Hence, for instantaneous synapses, the fixed point of the QIF-FRE is always of focus type, reflecting transient episodes of spike synchrony in the neuronal ensemble [40]. In contrast, setting τd = 0 in the H-FRE, the system becomes first order and oscillations are not possible. This is the critical difference between the two firing rate models. In fact, and in contrast with the eigenvalues Eq (2) corresponding to the growth rate of small perturbations in the H-FRE, here oscillatory instabilities may occur for nonvanishing values of τd. Fig 4 shows the Hopf boundaries obtained from Eq (6), as a function of the normalized synaptic strength j = J / Θ and the ratio of the synaptic and neuronal time constants τ = Θ τ d / τ m, and for different values of the ratio δ = Δ/Θ —see Materials and methods, Eqs (19)–(21). The shaded regions correspond to parameter values where the QIF-FRE display oscillatory solutions.
We have seen that the oscillations which emerge in inhibitory networks for sufficiently fast synaptic kinetics are correctly described by the firing rate equations Eq (3), but not by the heuristic Eq (1). The reason for this is that the oscillations crucially depend on the interaction between the population firing rate and the subthreshold membrane potential during spike initiation and reset; this interaction manifests itself at the network level through spike synchrony. Therefore, if one could suppress the spike synchrony mechanism, and hence the dependence on the subthreshold membrane potential, in Eq (3), the resulting equations ought to bear resemblance to Eq (1). In fact, as we saw in Fig 3, the two firing rate models become more similar when the synaptic kinetics become slower.
Next we show that the two models become identical, formally, in the limit of slow synaptic kinetics. To show this, we assume the synaptic time constant is slow, namely τ d = τ ¯ d / ϵ where 0 < ϵ ≪ 1, and rescale time as t ¯ = ϵ t. In this limit we are tracking the slow synaptic dynamics in while the neuronal dynamics are stationary to leading order, i.e.
R * = Φ ( - J τ m S + Θ ) . (9)
Therefore, the dynamics reduce to the first order equation
τ d S ˙ = - S + Φ ( - J τ m S + Θ ) . (10)
Notably, this shows that the QIF-FRE Eq (3), and the H-FRE (1), do actually have the same dynamics in the limit of slow synapses. In other words, Eq (10) is formally equivalent to the Wilson-Cowan equations for a single inhibitory population, and this establishes a mathematical link between the QIF-FRE and Heuristic firing rate descriptions. Additionally, considering slow second order synaptic kinetics (not shown), allows one to reduce the QIF-FRE with either alpha or double exponential synapses to a second-order system that formally corresponds to the so-called neural mass models largely used for modeling EEG data, see e.g. [6, 55–58].
Firing rate models, describing the average activity of large neuronal ensembles are broadly used in computational neuroscience. However, these models fail to describe inhibition-based rhythms, typically observed in networks of inhibitory neurons with synaptic kinetics [30–38]. To overcome this limitation, some authors heuristically included explicit delays in traditional FRE, and found qualitative agreement with the oscillatory dynamics observed in simulations of spiking neurons with both synaptic kinetics and fixed time delays [9, 10, 36, 39]. Nonetheless it remains unclear why traditional H-FRE with first order synaptic kinetics do not generate inhibition-based oscillations.
Here we have investigated a novel class of FRE which can be rigorously derived from populations of spiking (QIF) neurons [40]. Networks of globally coupled QIF neurons with fast inhibitory synapses readily generate fast self-sustained oscillations. The corresponding exact FRE, which we call the QIF-FRE, therefore also generates oscillations. The benefit of having a simple macroscopic description for the network dynamics is its amenability to analysis. In particular, the nonlinearities in Eq (3), which arise due to the spike initiation and reset mechanism in the QIF model, conspire to generate damped oscillations which reflect transient spike synchrony in the network. This oscillatory mode can be driven by sufficiently fast recurrent inhibitory synaptic activation, leading to sustained oscillations. This suggests that any mean-field description of network activity which neglects subthreshold integration will not properly capture spike-synchrony-dependent dynamical behaviors, including fast inhibitory oscillations.
The QIF-FRE have also allowed us to generate a phase diagram for oscillatory behavior in a network of QIF neurons with ease via a standard linear stability analysis, see Fig 4. This phase diagram agrees qualitatively with that of an equivalent network of Wang-Buzsáki neurons, suggesting that the QIF-FRE not only provide an exact description of QIF networks, but also a qualitatively accurate description of macroscopic behaviors in networks of Class I neurons in general. In particular, the QIF-FRE capture the fragility of oscillations to quenched variability in the network, a feature that seems to be particularly pronounced for Class 1 neuronal models compared to neural models with so-called Class 2 excitability [59].
Finally we have shown that the QIF-FRE and the heuristic H-FRE are formally equivalent in the limit of slow synapses. In this limit the neuronal dynamics is slaved to the synaptic activation and well-described by Eq (10), as long as external inputs are stationary. In fact, in the absence of quenched heterogeneity (Δ = 0), the approximation for slow synapses Eq (10) is also fully equivalent to the reduction for slow synapses in networks of Class 1 neurons derived by Ermentrout in [60]. This further indicates that the QIF-FRE are likely valid for networks of Class 1 neurons in general. However, we also show that in the more biologically plausible scenario of time-varying external drive some degree of neuronal synchronization is generically observed, as in (Fig 6), and the slow-synapse reduction Eq (10) is not valid.
The results presented here are obtained under two important assumptions that need to be taken into account when comparing our work to the existing literature on fast oscillations in inhibitory networks. (i) A derivation of an exact firing rate model for a spiking neuron network is only possible for ensembles of QIF neurons, which are the canonical model for Class 1 excitability [45, 61]. Although many relevant computational studies on fast inhibitory oscillations also consider Class 1 neurons [30, 32, 34, 39, 62–64], experimental evidence indicates that fast spiking interneurons are highly heterogeneous in their minimal firing rates in response to steady currents, and that a significant fraction of them are Class 2 [65–68] —but see also [69]. (ii) Our derivation of the QIF-FRE is valid for networks of globally coupled QIF neurons, with Lorentzian-distributed currents. In this system inhibition-based oscillations are only possible when the majority of the neurons are self-sustained oscillators, i.e. for Θ > 0 in Eq (14), and are due to the frequency locking of a fraction of the oscillators in the population [41, 42] —as it can be seen in the raster plot of Fig 1(c). In this state, the frequency of the cluster of synchronized oscillators coincides with the frequency of the mean field. The value of the frequency itself is determined through an interplay between single-cell resonance and network effects. Specifically, the synchronized neurons have intrinsic spiking frequencies near that of the mean-field oscillation and hence are driven resonantly. This collective synchronization differs from the so-called sparse synchronization observed in inhibitory networks of identical Class 1 neurons under the influence of noise [34, 36, 62, 63]. In the sparsely synchronized state neurons fire stochastically at very low rates, while the population firing rate displays the fast oscillations as the ones reported here.
Oscillatory phenomena arising from single-cell resonances, and which reflect spike synchrony at the population level, are ubiquitous in networks of spiking neurons. Mean-field theory for noise-driven networks leading to a Fokker-Planck formalism, allows for a linear analysis of the response of the network to weak stimuli when the network is in an asynchronous state [43, 70]. Resonances can appear in the linear response when firing rates are sufficiently high or noise strength sufficiently low. Recent work has sought to extend H-FRE in this regime by extracting the complex eigenvalue corresponding to the resonance and using it to construct the linear operator of a complex-valued differential equation, the real part of which is the firing rate [29]. Other work has developed an expression for the response of spiking networks to external drive, which often generates resonance-related damped oscillations, through an eigenfunction expansion of the Fokker-Planck equation [71]. Our approach is similar in spirit to such studies in that we also work with a low dimensional description of the network response. In contrast to previous work our equations are an exact description of the macroscopic behavior, although they are only valid for networks of heterogeneous QIF neurons. Nonetheless, the QIF-FRE are simple enough to allow for an intuitive understanding of the origin of fast oscillations in inhibitory networks, and in particular, of why these oscillations are not properly captured by H-FRE.
We model fast-spiking interneurons, the dynamics of which are well-described by the Hodgkin-Huxley equations with only standard spiking currents. Many models of inhibitory neurons are Class 1 excitable [72], including for example the Wang-Buszáki (WB) [30], and the Morris-Lecar models [73]. Class 1 models are characterized by the presence of a saddle-node bifurcation on an invariant circle at the transition from quiescence to spiking. One consequence of this bifurcation structure is the fact the spiking frequency can be arbitrarily low near threshold. Additionally, near threshold the spiking dynamics are dominated by the time spent in the vicinity of the saddle-node itself, allowing for a formal reduction in dimensionality from the full neuron model to a reduced normal form equation for a saddle-node bifurcation [2, 45, 61]. This normal form, which is valid for any Class 1 model near threshold, is known as the quadratic integrate-and-fire model (QIF). Using a change of variables, the QIF model can be transformed to a phase model, called Theta-Neuron model [74], which has an strictly positive Phase Resetting Curve (PRC). Neuron models with strictly positive PRC are called Type 1 neurons, indicating that perturbations always produce an advance (and not a delay) of their phase. In general, Class 1 neurons have a Type 1 PRC [45], but see [75, 76].
In a network of QIF neurons, the neuronal membrane potentials are { V ˜ i } i = 1 , … , N, which obey the following ordinary differential equations [7, 64, 74]:
C d V ˜ i d t = g L ( V ˜ i - V t ) ( V ˜ i - V r ) ( V t - V r ) + I 0 , i (11)
where C is the cell capacitance, gL is the leak conductance and I0,i are external currents. Additionally, Vr and Vt represent the resting potential and threshold of the neuron, respectively. Using the change of variables V ˜ i ′ = V ˜ i - ( V t + V r ) / 2, and then rescaling the shifted voltages as V i = V ˜ i ′ / ( V t - V r ), the QIF model (11) reduces to
τ m V ˙ i = V i 2 + I i (12)
where τm = C/gL is the membrane time constant, Ii = I0,i/(gL(Vt−Vr))−1/4 and the overdot represents derivation with respect to time t. Note that in the model (12) the voltage variables Vi and the inputs Ii do not have dimensions. Thereafter we work with QIF model its simplest form Eq (12). We assume that the inputs are
I i = η i - J τ m S , (13)
where J is the inhibitory synaptic strength, and S is the synaptic gating variable. Finally, the currents ηi are constants taken from some prescribed distribution that here we consider it is a Lorentzian of half-width Δ, centered at Θ g ( η ) = 1 π Δ ( η - Θ ) 2 + Δ 2 . (14)
In numerical simulations the currents were selected deterministically to represent the Lorentzian distribution as: ηi = Θ + Δtan(π/2(2i − N − 1)/(N + 1)), for i = 1, …, N. In the absence of synaptic input, the QIF model Eqs (12) and (13) exhibits two possible dynamical regimes, depending on the sign of ηi. If ηi < 0, the neuron is excitable, and an initial condition V i ( 0 ) < - η i, asymptotically approaches the resting potential - - η i. For initial conditions above the excitability threshold, V i ( 0 ) > - η i, the membrane potential grows without bound. In this case, once the neuron reaches a certain threshold value Vθ ≫ 1, it is reset to a new value −Vθ after a refractory period 2τm/Vθ (in numerical simulations, we choose Vθ = 100). On the other hand, if ηj > 0, the neuron behaves as an oscillator and, if Vθ → ∞, it fires regularly with a period T = π τ m / η i. The instantaneous population mean firing rate is
R = lim τ s → 0 1 N 1 τ s ∑ j = 1 N ∑ k ∫ t - τ s t d t ′ δ ( t ′ - t j k ) , (15)
where t j k is the time of the kth spike of jth neuron, and δ(t) is the Dirac delta function. Finally, the dynamics of the synaptic variable obeys the first order ordinary differential equation
τ d S ˙ = - S + R . (16)
For the numerical implementation of Eqs (15) and (16), we set τs = 10−2 τm. To obtain a smoother time series, the firing rate plotted in Fig 3 was computed according to Eq (15) with τs = 3 ⋅ 10−2 τm.
Recently Luke et al. derived the exact macroscopic equations for a pulse-coupled ensemble of Theta-Neurons [77], and this has motivated a considerable number of recent papers [78–86, 88]. This work applies the so-called Ott-Antonsen theory [89–91] to obtain a low-dimensional description of the network in terms of the complex Kuramoto order parameter. Nevertheless, it is is not obvious how these macroscopic descriptions relate to traditional H-FRE.
As we already mentioned, the Theta-neuron model exactly transforms to the Quadratic Integrate and Fire (QIF) model by a nonlinear change of variables [45, 61, 74]. This transformation establishes a map between the phase variable θi ∈ (−π, π] of a Theta neuron i, and the membrane potential variable Vi ∈ (−∞, +∞) of the QIF model Eq (12). Recently it was shown that, under some circumstances, a change of variables also exists at the population level [40]. In this case, the complex Kuramoto order parameter transforms into a novel order parameter, composed of two macroscopic variables: The population-mean membrane potential V, and the population-mean firing rate R. As a consequence of that, the Ott-Antonsen theory becomes a unique method for deriving exact firing rate equations for ensembles of heterogeneous spiking neurons —see also [92–94] for recent alternative approaches.
Thus far, the FRE for QIF neurons (QIF-FRE) have been successfully applied to investigate the collective dynamics of populations of QIF neurons with instantaneous [40, 86, 87], time delayed [95] and excitatory synapses with fast synaptic kinetics [96]. However, to date the QIF-FRE have not been used to explore the dynamics of populations of inhibitory neurons with synaptic kinetics —but see [83] for a numerical investigation using the low-dimensional Kuramoto order parameter description. The method for deriving the QIF-FRE corresponding to a population of QIF neurons Eq (12) is exact in the thermodynamic limit N → ∞, and, under the assumption that neurons are all-to-all coupled. Additionally, if the parameters ηi in Eq (13) (which in the thermodynamic limit become a continuous variable) are assumed to be distributed according to the Lorentzian distribution Eq (14), the resulting QIF-FRE become two dimensional. For instantaneous synapses, the macroscopic dynamics of the population of QIF neurons (12) is exactly described by the system of QIF-FRE [40]
τ m R ˙ = Δ π τ m + 2 R V , (17a) τ m V ˙ = V 2 - ( π τ m R ) 2 - J τ m R + Θ , (17b)
where R is the mean firing rate and V the mean membrane potential in the network. With exponentially decaying synaptic kinetics the QIF-FRE Eq (17) become Eq (3). In our study, we consider Θ > 0, so that the majority of the neurons are oscillatory —see Eq (14).
To investigate the existence of oscillatory instabilities we use Eq (6) written in terms of the non-dimensional variables and parameters defined previously, which is
− 2 j r * = ( 1 + λ ˜ τ ) [ ( 2 π r * ) 2 + ( λ ˜ + δ π r * ) 2 ] . (22)
Imposing the condition of marginal stability λ ˜ = i ω ˜ in Eq (22) gives the system of equations
0 = 2 j r * + 4 π 2 r * 2 + 4 v * 2 - ( 1 - 4 v * τ ) ω ˜ 2 (23a) 0 = ω ˜ ( 4 v * - 4 π 2 r * 2 τ - 4 v * 2 τ + τ ω ˜ 2 ) (23b)
where the fixed points are obtained from Eq (4) solving
0 = v * 2 - π 2 r * 2 - j r * + 1 , (24)
with
v * = - δ 2 π r * Eq (23b) gives the critical frequency
ω ˜ = 2 τ ( π τ r * ) 2 + τ v * ( τ v * - 1 ) .
The Hopf boundaries can be plotted in parametric form solving Eq (24) for j, and substituting j and ω ˜ into Eq (23a). Then solving Eq (23a) for τ gives the Hopf bifurcation boundaries
τ ± ( r * ) = π 2 r * 2 - 1 + 7 v * 2 ± ( π 2 r * 2 - 1 ) 2 - ( 14 + 50 π 2 r * 2 ) v * 2 - 15 v * 4 16 v * ( π 2 r * 2 + v * 2 ) . (25)
Using the parametric formula
( j ( r * ) , τ ± ( r * ) ) ± = ( v * 2 / r * + 1 / r * − π 2 r * , τ ± ( r * ) ) .
we can be plot the Hopf boundaries for particular values of the parameter δ, as r* is changed. Fig 4 shows these curves in red, for δ = 0.05 and δ = 0.075. They define a closed region in parameter space (shaded region) where oscillations are observed.
We perform numerical simulations using the the Wang-Buzsáki (WB) neuron [30], and compare them with our results using networks of QIF neurons. The onset of oscillatory behavior in the WB model is via a saddle node on the invariant circle (SNIC) bifurcation. Therefore, the populations of WB neurons near this bifurcation are expected to be well described by the theta-neuron/QIF model, the canonical model for Class 1 neural excitability [45, 74].
We numerically simulated a network of N all-to-all coupled WB neurons, where the dynamics of each neuron is described by the time evolution of its membrane potential [30]
C m V i ˙ = - I Na , i - I K , i - I L , i - I syn + I app , i + I 0 .
The cell capacitance is Cm = 1 μF/cm2. The inputs Iapp (in μA/cm2) are distributed according to a Lorentzian distribution with half width σ and center I ¯. In numerical simulations these currents were selected deterministically to represent the Lorentzian distribution as I app , i = I ¯ + σ tan ( π / 2 ( 2 i - N - 1 ) / ( N + 1 ) ), for i = 1, …, N. The constant input I0 = 0.1601 μA/cm2 sets the neuron at the SNIC bifurcation when Iapp = 0. The leak current is
I L , i = g L ( V i - E L ) ,
with gL = 0.1 mS/cm2, so that the passive time constant τm = Cm/gL = 10 ms. The sodium current is
I Na , i = g Na m ∞ 3 h ( V i - E Na ) ,
where gNa = 35 mS/cm2, ENa = 55 mV, m∞ = αm/(αm + βm) with αm(Vi) = −0.1(Vi + 35)/(exp(−0.1(Vi + 35) − 1)), βm(Vi) = 4exp(−(Vi + 60)/18). The inactivation variable h obeys the differential equation
h ˙ = ϕ ( α h ( 1 - h ) - β h h ) ,
with ϕ = 5, αh(Vi) = 0.07exp(−(Vi + 58)/20) and βh(Vi) = 1/(exp(−0.1(Vi + 28)) + 1). The potassium current follows
I K , i = g K n 4 ( V i - E K ) ,
with gK = 9 mS/cm2, EK = −90 mV. The activation variable n obeys
n ˙ = ϕ ( α n ( 1 - n ) - β n n ) ,
where αn(Vi) = −0.01(Vi + 34)/(exp(−0.1(Vi + 34)) − 1) and βn(Vi) = 0.125exp(−(Vi + 44)/80).
The synaptic current is Isyn = kCm S, where the synaptic activation variable S obeys the first order kinetics Eq (16) and k is the coupling strength (expressed in mV). The factor Cm ensures that the effect of an incoming spike to the neuron is independent from its passive time constant. The neuron is defined to emit a spike when its membrane potential crosses 0 mV. The population firing rate is then computed according to Eq (15), with τs = 10−2 ms. In numerical simulations we considered N = 1000 all-to-all coupled WB neurons, using the Euler method with time step dt = 0.001 ms. In Fig 1, the membrane potentials were initially randomly distributed according to a Lorentzian function with half width 5 mV and center −62 mV. Close to the bifurcation point, this is equivalent to uniformly distribute the phases of the corresponding Theta-Neurons in [−π, π] [2, 7, 61, 74]. The parameters were chosen as I ¯ = 0 . 5 μ A / cm 2, σ = 0.01 μA/cm2 and k = 6 mV. The population firing rate was smoothed setting τs = 2 ms in Eq (15).
In Fig 5, we systematically varied the coupling strength and the synaptic time decay constant to determine the range of parameters displaying oscillatory behavior. For each fixed value of τd we varied the coupling strength k; we performed two series of simulations, for increasing and decreasing coupling strength. In Fig 5 we only show results for increasing k.
All quantities were measured after a transient of 1000 ms. To obtain the amplitude of the oscillations of the mean membrane potential, we computed the maximal amplitude V ¯ max - V ¯ min over time windows of 200 ms for 1000 ms, and then averaged over the five windows.
|
10.1371/journal.ppat.1000905 | Environmental Factors Determining the Epidemiology and Population Genetic Structure of the Bacillus cereus Group in the Field | Bacillus thuringiensis (Bt) and its insecticidal toxins are widely exploited in microbial biopesticides and genetically modified crops. Its population biology is, however, poorly understood. Important issues for the safe, sustainable exploitation of Bt include understanding how selection maintains expression of insecticidal toxins in nature, whether entomopathogenic Bt is ecologically distinct from related human pathogens in the Bacillus cereus group, and how the use of microbial pesticides alters natural bacterial populations. We addressed these questions with a MLST scheme applied to a field experiment in which we excluded/added insect hosts and microbial pesticides in a factorial design. The presence of insects increased the density of Bt/B. cereus in the soil and the proportion of strains expressing insecticidal toxins. We found a near-epidemic population structure dominated by a single entomopathogenic genotype (ST8) in sprayed and unsprayed enclosures. Biopesticidal ST8 proliferated in hosts after spraying but was also found naturally associated with leaves more than any other genotype. In an independent experiment several ST8 isolates proved better than a range of non-pathogenic STs at endophytic and epiphytic colonization of seedlings from soil. This is the first experimental demonstration of Bt behaving as a specialized insect pathogen in the field. These data provide a basis for understanding both Bt ecology and the influence of anthropogenic factors on Bt populations. This natural population of Bt showed habitat associations and a population structure that differed markedly from previous MLST studies of less ecologically coherent B. cereus sample collections. The host-specific adaptations of ST8, its close association with its toxin plasmid and its high prevalence within its clade are analogous to the biology of Bacillus anthracis. This prevalence also suggests that selection for resistance to the insecticidal toxins of ST8 will have been stronger than for other toxin classes.
| Bacillus thuringiensis is one of the most useful bacteria in insect pest management: it is used as an environmentally friendly biopesticide and its insecticidal toxins are incorporated into genetically modified crops. Concerns for its ongoing economic exploitation include the rapidity with which insects can evolve resistance to its toxins and whether B. thuringiensis is ecologically and genetically distinct from closely related strains that cause infections in humans and domestic animals. We found that natural bacterial populations in soil and on cabbage leaves were dominated by an insect-pathogen specialist genotype, and this genotype was better than its close relatives at establishing populations on leaves where its hosts were likely to be feeding. Spraying microbial pesticides and the addition of insect hosts increased the proportion of insect-pathogen specialists in the bacterial population, confirming that application of these biopesticides is a safe means of insect control. Populations of B. thuringiensis were transient on plant material, suggesting that selective pressure for resistance can be similarly transient. However, the genotype that dominates the natural community has been economically exploited more than any other, and selection for resistance to this strain may have occurred in natural populations of insects prior to the use of B. thuringiensis in pest control.
| The Bacillus cereus group contains a number of clinically and economically important bacteria including the entomopathogen Bacillus thuringiensis (Bt); Bacillus anthracis, the causative agent of anthrax; and B. cereus sensu stricto, a species involved in potentially lethal food-poisoning as well as a wide range of opportunistic infections [1]. Bt is widely exploited in insect pest management. Its utility derives from the large quantities of proteinaceous toxins that form crystalline parasporal inclusions, termed the Cry (crystal) and Cyt (cytolytic) proteins [2]. The vast majority of commercially viable genetically modified (GM) insect resistant crops express one or more of these toxins and these crops covered approximately 46 million ha (worldwide) in 2008 [3]. Bt formulations are also the most successful organic microbial pesticide, with target hosts ranging from mosquitoes to Lepidopteran pests of agriculture, horticulture and forestry [4]. Although the structure and mode of action of Bt Cry toxins has been intensively studied, the biology and ecology of the bacterium is not fully characterized [1], [5]. Despite high levels of pathogenicity, the ability of B. thuringiensis strains to grow and sporulate effectively within insect cadavers is variable [6], [7]. Reports of natural outbreaks or epizootics of Bt are very rare in the field [8] and effective transmission of Bt between larvae has been difficult to demonstrate experimentally [9] and can require a high density of hosts and/or cannibalism [10].
Although it has been challenging to demonstrate how Bt is transmitted between hosts in the field, this bacterium is readily isolated from soil and plant material [11], [12], [13], [14], [15]. The variable success of Bt as a pathogen, the abundance of this bacterium in soil and plant material, and the reported lack of correlation between host abundance and the abundance of entomopathogenic Bt [14] remains puzzling and has also led to wide speculation on the ecological niche of Bt. It has been suggested that Bt is a soil micro-organism with incidental insecticidal activity [14]; that Bt is part of the phylloplane microbiota and has evolved to provide symbiotic protection against insect attack [15], [16]; or that Bt may be part of the commensal gut microbiota of many insects without causing overt disease [1]. With the exception of a quantitative examination of the claim that Bt can reproduce as a commensal [5] each of these hypotheses remain largely untested.
Improved understanding of the ecology of Bt and the dominant forces shaping the population structure of the B. cereus group will help address key issues for the ongoing exploitation of Bt such as: (1) managing the potential for widespread resistance to microbial sprays and GM crops [17]; (2) understanding the selective forces maintaining insecticidal toxin expression in natural populations; and (3) establishing whether insecticidal Bt is ecologically distinct from B. cereus strains that are capable of infecting humans [18]. Our aims, therefore, were to understand if Bt behaves as a true entomopathogen in the field by exploring the effect of the presence of an insect host on the abundance and population structure of the B. cereus group; to investigate how the application of live microbial pesticides might alter the indigenous bacterial population structure and thereby affect selection for the evolution of resistance to Bt; and finally to understand the ecology and phylogeny of Bt in comparison to that of its close relatives in the B. cereus group. In other words, are environmental and entomopathogenic strains from a single community ecologically distinct? And do environmental conditions that promote the proliferation of Bt also promote the proliferation of other members of this group?
In order to examine how environmental conditions are involved in generating and maintaining population structure within the B. cereus group we used a novel combination of experimental ecology and multi-locus sequence typing (MLST). MLST is an increasingly popular genotyping technique based upon the nucleotide sequences of loci within several housekeeping genes. For each locus, unique sequences are assigned unique allele numbers, and each specific allelic combination is assigned a sequence type (ST) [19], [20]. An MLST scheme has been developed for the B. cereus group and has been used to study the phylogenetic relationships and population structure of these bacteria [21], [22], [23], [24]. Although MLST has been applied to hundreds of Bacillus isolates (http://pubmlst.org/bcereus/) to our knowledge, no study has applied this technique in combination with the experimental manipulation of the Bacillus population or its hosts.
In this study we have explored the ecology of this group of organisms in relation to an insect host, the larvae of Plutella xylostella (the diamondback moth). A manipulative field trial was used to assess how bacterial abundance and population structure would be affected by the presence or exclusion of host insects and by the application of a commercial Bt biopesticide. MLST of strains collected during the trial was used to examine treatment effects upon the population genetic structure. All strains were phenotypically scored for the presence of entomocidal Cry toxins. Alleles and STs were assigned to habitat specific clades, based on genealogical reconstructions using CLONALFRAME [23], and the proportions of these alleles and STs amongst strains of each treatment group were compared. We have shown how insect hosts and biopesticide application can increase the B. cereus sensu lato population and increased the representation of entomocidal strains within this population. We found strong evidence for ecological differentiation between B. cereus clades both in terms of habitat association and the manner in which they responded to insect hosts. In contrast to previous MLST studies of the B. cereus group, we found a near epidemic population structure with a single genotype dominating the community, this genotype proved better able to colonize leaf surfaces (even in the absence of hosts) than non-pathogenic relatives in clade 3 and a higher proportion of isolates with this genotype expressed Cry toxin genes in comparison with other B. thuringiensis genotypes.
The density of B. cereus group bacteria on leaf material was increased by the application of a Bt biopesticide, ten days after spraying (Figure 1, mixed model ANOVA, time point*spray interaction, df = 1, Likelihood ratio = 25.5, p<0.0001). However, bacterial density in leaf samples was not affected by the addition of P. xylostella insect hosts to experimental enclosures (df = 1, Likelihood ratio = 0.315, p = 0.575), nor did the presence of these hosts interact with microbial pesticide application (df = 1, Likelihood ratio = 0.661, p = 0.416). In contrast, both the presence of hosts and application of Bt biopesticide affected bacterial density in soil (Figure 1). Host addition increased bacterial densities in soil over the whole course of the experiment (df = 1, Likelihood ratio = 7.77, p = 0.0053). Bt biopesticide application had only a transient effect on bacterial density in soil (spray*time point interaction, df = 2, Likelihood ratio = 9.63, p = 0.008); post-hoc treatment comparisons confirmed that sprays had their strongest influence on bacterial density on the 10 day post-spraying time point (effect size 1.04, t = 3.83, p = 0.0002). A marginally significant three-way interaction also suggested that this temporary effect of spraying was strongest when insect hosts were present (three-way interaction, df = 2, Likelihood ratio = 6.02 p = 0.049).
A CLONALFRAME analysis of the pubMLST isolates database and our own isolates resolved three major clades, as found in previous analyses [22], [24], [25]. The inclusion of isolates not incorporated in previous analyses revealed a deep branching genetic structure indicating the presence of additional, less populous, clades (Clades 4 & 5, Figure 2). The names of clades 1–3 follow previous studies: clade 1 contains B. anthracis, B. cereus emetic strains and other clinical isolates as well as approximately one third of the B. thuringiensis strains (Figure 3). Clade 2 contains the majority of B. thuringiensis isolates as well as a large number of isolates involved in enterotoxin associated food poisoning and septicaemia, and two isolates of Bacillus mycoides (Figures 2 & 3). Clade 3 contains all the strains designated as Bacillus weihenstephanensis in the database as well as three STs identified as Cry producing thuringiensis (STs 190 196, 200; Figure 3) and two isolates of B. mycoides; the remaining B. mycoides and two isolates of Bacillus pseuodomycoides were placed in a distinct subgroup of clade 3 (Figure 2). Clades 4 and 5 contain predominantly isolates of environmental origin collected from the Silwood Park campus of Imperial College, Ascot UK [26] as well as a small number of clinical isolates, some of which were previously associated with clade 2 (STs 101 and 111; Figure 3) [25]. All except one of the isolates recovered in this study were placed within clade 2 or clade 3, the remaining isolate mapped to clade 5.
The sequence type of the experimentally applied microbial pesticide is ST8 (Figure 2) and it is therefore unsurprising that the clade 2 isolates were dominated by ST8. However, ST8 was the commonest genotype in unsprayed plots and in the soil prior to the application of any biopesticide to the site and there was no prior use of Bt biopesticides on the farm on which the study was located. Thus, while we recovered 112 isolates with the ST8 genotype, 61 of these were recovered either before spraying with biopesticidal Bt (in T0 soil samples) or from unsprayed enclosures, indicating that this genotype dominated the natural community. Of the other STs recovered from clade 2, two have been identified previously as thuringiensis isolates (ST16 and ST18). Both ST16 and 18, and additional STs within clade 2 have been isolated as B. cereus in human infections, either from cases of diarrheal food poisoning (ST24) or from infections in immune-suppressed patients (ST18, 24, 166). We found both Cry toxin positive and Cry null (phenotypic B. cereus) isolates of ST8, ST18, ST24 and ST166 (Figure 2). The characteristics of all isolates with novel STs have been detailed in the online supporting information (Text S1).
We hypothesized that if Bt reproduces in agricultural environments as an entomopathogen we should be able to detect the signature of this proliferation in the altered population genetic structure of the B. cereus group in response to the experimental addition of hosts and or biopesticidal sprays. With a single exception all leaf-isolated bacteria in this study mapped to clade 2, as did all Cry toxin production (Figure 2). These patterns suggested that entomopathogens are predominantly restricted to clade 2, although not all strains in clade 2 are necessarily entomopathogens. The remaining STs were primarily located within the weihenstephanensis clade (clade 3), which are typically soil bacteria [27] (Figure 3).
We tested whether our experimental treatments had altered the representation of entomopathogens in the natural population by using both an allele-based analysis and an isolate based analysis to investigate complementary hypotheses. The allele-based analysis used a genotypic definition of entomopathogenicity based on MLST variation, and allows us to test whether the entomopathogenic niche is associated with particular chromosomal genotypes. In contrast, the isolate-based analysis used a morphological definition of pathogenicity (Cry toxin production) that is associated with possession of virulence plasmids. Thus these two complementary analyses allow us to explore whether chromosomally linked traits and/or potentially mobile elements are important for the exploitation of insect hosts in the field.
An allele-based analysis of our MLST-characterized strains was conducted that incorporated the possibility of allele exchange between strains. We defined strains based upon sample origin (leaf or soil): the alleles of isolates that had been recovered from leaf material were defined as having a pathogenic origin, and the alleles of all isolates that were only recovered from soil were defined as non-pathogenic. A habitat associated genotypic definition allows us to test the hypothesis that chromosomal traits associated with proliferation and/or colonization on leaf surfaces are important for the success of B. cereus group entomopathogens as well as allowing us to test the hypothesis that growth within hosts is more significant on a population level than epiphytic proliferation in the absence of insect hosts. A structure analysis tested how the presence of host insects and biopesticidal sprays affected the proportional representation of putative entomopathogens in the population using 104 isolates from time point T10 and 84 isolates from time point T28. The structure analysis indicated that a greater proportion of alleles with a pathogenic origin (leaf habitat associated alleles) were recovered from the enclosures with insect hosts (74% of alleles) than from enclosures without insect hosts (65% of alleles; Figure 4). Spraying with Bt biopesticide increased the proportion of alleles originating from putatively pathogenic strains, but only when insects were present in experimental enclosures, 87% of alleles originated from pathogenic STs compared to 58% in unsprayed plots. Where insects were absent from experimental plots the effect of spraying was negated: the probabilities of alleles having a pathogenic origin were 65% and 66% in sprayed and unsprayed plots respectively (Figure 4).
In this analysis we defined entomopathogens as those strains that expressed the plasmid-borne Cry toxin parasporal inclusions. This allows us to test whether Cry toxin expression in the B. cereus group has a detectable fitness benefit in these bacteria in the field and in the presence of hosts. This analysis incorporated every isolate from time points T10 and T28 (a total of 301 isolates), all isolates were scored phenotypically by microscopic examination.
The habitat from which isolates originated had a strong effect on the proportion of crystal producers in the population, those with leaf origin having a much greater probability of being Cry positive than soil derived isolates (df = 1, χ2 = 103, p<0.0001; Figure 5). Biopesticide application and the addition of hosts both increased the proportion of crystal producing strains (df = 1, χ2 = 12.2, p<0.0001; df = 1, χ2 = 12.6, p<0.0001 respectively). Moreover, in this analysis, the effect of hosts did not depend upon spray application (spray*host interaction, df = 1, χ2 = 0.85, p = 0.36; Figure 5).
Mapping the sample origin of the field study isolates onto the genealogical tree revealed a strong correlation between clade identity and sample origin (Figure 2). The association of host/habitat with ST-based genotype was also investigated within the global B. cereus dataset (Figure 3). Whilst the soil was a universal reservoir for all isolates, only strains from clades 2 and 5 were isolated from leaf material (Figure 2). In contrast, within the global dataset there was a weaker correlation between clade and habitat/host association with STs. In the global dataset clades 1 and 2 in particular contain STs associated with both insect and vertebrate pathology. The restriction of clade 3 bacteria to the soil also does not hold in the wider data set, which includes isolates from a range of studies with different sampling regimes. An isolate based analysis of the pubMLST database and other sources does suggest a higher degree of clade-level specialization in the B. cereus group than this ST-based exploration of the data. However, the pubMLST database is not a natural population and an isolate-based analysis is potentially more subject to reporting and research biases than an exploration of the ecology of the recorded diversity at an ST level (Text S1). Both an ST based and isolate based analyses of genotype/ecology correlations have been included in the online supporting information for comparison (Text S1: Supporting Figure 1, Supporting Figure 2).
The association of isolates from clade 2 with plant material in this study was not, however, dependent upon the presence of insect hosts or the proliferation of bacteria within the hosts. This is best illustrated by examining the enclosures that did not receive biopesticidal sprays. A simple comparison of the proportion of isolates identified as ST8 (the commonest member of clade 2 in this study) shows that the putatively entomopathogenic ST8 was much more highly represented in the leaf samples than in the soil samples (χ2 = 24.7, df = 2, p<0.0001; Figure 6) while the presence of hosts had no effect on the proportion of isolates identified as ST8 in this subset (χ2 = 0.014, df = 1, p = 0.91; Figure 6).
An additional greenhouse experiment showed that B. thuringiensis ST8 is much better than clade 3 B. cereus at colonizing the leaves of growing plants from experimentally inoculated soil. Plants were grown either in autoclaved compost (negative control); inoculated with a Bt ST8 isolate. Counts of bacteria in the B. cereus group in leaf samples from Chinese cabbage seedlings were measured using selective media containing polymyxin and a lecithinase indicator (egg yolk emulsion). There was at least an order of magnitude more B. cereus group bacteria in the leaf samples from the Bt ST8 treatment than in the B. cereus clade 3 treatment (epiphytic bacteria t = −6.11, p<0.0001; endophytic bacteria t = −4.96, p = 0.001; Figure 7). No B. cereus group bacteria were recovered from control plant leaf washes and there was a single colony recovered from the leaf homogenate (endophytic) control samples. The counts of bacteria in the B. cereus clade 3 treatment were not significantly different from the controls (epiphytic bacteria t = 1.25, p = 0.222; endophytic bacteria t = 0.31, p = 0.76). We also scored microscopically a subsample of B. cereus group isolates from the leaves of plants in Bt ST8 treatment - all 92 isolates were positive for the presence of parasporal crystalline inclusions, a phenotypic trait not shown by any of the B. cereus clade 3 strains.
Despite almost a century since its discovery we still know very little about the reproduction and field ecology of Bt, in part due to a lack of manipulative field experiments. In contrast to previous claims [14] we have shown that insect hosts can alter B. cereus group populations by both increasing population density and the proportion of strains expressing Cry toxins. Although we found no obvious Bt-killed cadavers on plant material and a transient population of B. cereus/Bt on the leaf surface, a moderate density of hosts (typically no more than 4–5 larvae per plant) nearly doubled the population of B. cereus group spores in soil. The rarity of cadavers on leaves, the transient low level population of Bt on plant surfaces [28], and the substantial accumulation of spores in soil suggests that opportunities for host-host transmission on leaves may be rare. Instead these results suggest a life cycle in which spores from cadavers, or cadavers themselves, quickly end up in soil with host-host transmission following on from a resting phase in a soil reservoir [29].
We performed MLST allele based and mobile element based (Cry toxin) analyses to explore how insect hosts in the field and biopesticide application altered population structure. We hypothesized that alleles linked to entomopathogenically important loci should increase in frequency in the population in response to the experimental addition of insect hosts. The MLST analysis showed that pathogenic alleles/STs only increase in frequency when both biopesticides and hosts are present. An analysis of Cry toxin carriage (with a larger sample size) found that hosts increase the frequency of toxin production independent of biopesticide spraying alone. Our overall interpretation is therefore that Cry toxins are beneficial to the B. cereus group when hosts are present and that biopesticides have a stronger impact on the bacterial community when they can proliferate within hosts. Nevertheless, MLST genotype and Cry toxin expression were tightly linked in this natural population and chromosomal genotype was strongly implicated in isolate ecology and in the ability of bacteria to colonize plant leaves. This association predominantly arose because one genotype ST8, was associated with most of the Cry toxin carriage and also dominated the leaf community, this high prevalence on leaves persisted in unsprayed experimental enclosures. We experimentally verified that several ST8 isolates have an improved ability to colonize growing seedlings from soil relative to non-pathogenic isolates in the B. weihenstephanensis dominated clade 3. This plant colonizing ability is independent of the possession of Cry toxin plasmids [30]. Other Bacillus spp. can colonize and adhere to leaf surfaces [31] and vegetative cells of Bt, B. cereus and B. anthracis have been found associated with plant roots [32], [33], [34]. The available evidence suggests that Bt and B. anthracis do not reproduce at a substantial level in soil or epiphytically, but that Bt can readily establish populations on leaves [4], [35], [36]. This study suggests that the movement of Bt from the soil reservoir to the aerial parts of plants, where susceptible hosts are present [4], is a key feature of its ecology. A similar transmission problem must also exist for B. anthracis which commonly infects ungulate hosts orally [37]. Therefore the association of Bt and B. anthracis with plants is probably not driven by the potential for epiphytic growth, but in order to increase the likelihood of infecting hosts.
A key aim of this study was to explore the ecology of Bt in relation to the selection pressure on Cry toxin resistance in natural populations. The poor persistence of high densities of Bt on plant foliage implies that relatively infrequent biopesticide applications will have little long-term effects on the Bacillus community. Elevated densities and changes in population structure effects were practically undetectable 28 days after spray application. More frequent Bt applications could have some impact on selection for resistance by altering the abundance or diversity of pathogens in the soil reservoir. The impact of changes in soil on selection pressure for resistance will be strongly mitigated by the very patchy natural distribution of Bt on plants. The real danger for the evolution of resistance arises if sprays are applied at a rate that results in a near continuous population of Bt on plants. We found that sprays in a temperate environment lend to high densities 10 days after application. Unsurprisingly, spray applications as frequent as every two or three days in low latitudes have led to multiple independent instances of the evolution of resistance to microbial pesticides in Lepidoptera on vegetable crops [38], [39].
The population structure of the natural B. cereus group community on plants also has implications for resistance evolution. This bacterial community was dominated by a single genotype (ST8) in agreement with work in the UK (R. Ellis unpubl. dat. available at http://www.pubmlst.org). This genotype is also used in the most successful biopesticidal sprays based on B. t. kurstaki HD-1 (DiPel). It produces predominantly two very similar Cry toxins (Cry1Ac, Cry1Ab) that are exploited in the majority of GM Bt corn and cotton varieties [40]. Several studies have found that a very high proportion of natural isolates carry Cry1Ac and Cry1Ab genes [41], [42]. Selection for resistance to naturally occurring toxins produced by ST8 may therefore have already elevated the frequency of resistance in insects prior to the use of Bt in pest management [43], [44]. The initial frequency of resistance to rarer toxin genes carried by other strains should be lower because of weaker past selection, and exploiting less common toxins may therefore have some benefits for resistance management.
This community-level study revealed a very tight relationship between bacterial clade and habitat/host association although this pattern is not repeated in the global B. cereus pubMLST dataset. On a global level, clades 1–3 have some differences in the proportion of isolates linked to particular hosts or disease symptoms (Figure 3) [22] but these generalizations do not hold well at a finer taxonomic resolution. Strains associated with insecticidal Cry toxins, pneumonia, septicaemia, and diarrheal food poisoning are widely distributed in both clades 1 and 2 [22] while clades 4 and 5 are almost as diverse. This poor correlation between pathogenicity and taxonomy in the global dataset results, at least in part, from the intrinsic mobility of many B. cereus virulence factors. Many major virulence factors are plasmid-borne in this group, e.g. B. anthracis pX01 and pX02 plasmids, the cereulide emetic toxin plasmid in B. cereus and the Cry toxin plasmids in B. thuringiensis [45], [46], [47]. Natural conjugation rates between B. cereus and B. thuringiensis strains within hosts can be extremely high (10−1 transconjugants per recipient) [48]. Chromosomal virulence factors are widely and variously distributed throughout the group [21] but may also have some horizontal mobility [49]. Nevertheless, occasional horizontal mobility should not preclude selection and recent clonal expansion producing a strong association between plasmid-associated phenotype and genotype at a fine geographical scale. Other studies using geographically coherent populations have also found associations between clade and Cry toxin gene expression [18], [27]. Sampling differences in different studies will also blur genotype habitat correlations in a global dataset. For example, many of the plant associated bacteria in clades 3–5 were recovered from a single study based on the plant Rumex obtusifolius [26]. This plant has a much lower growth habit than the B. oleracea used in this study. Mature Rumex leaves are close to the ground, are readily splashed with soil, and therefore carry more bacteria that are soil specialists. Thus, at a global level recombination, geographical divergence and diverse sampling regimes may be able to mask any local biological association between clade and ecology.
Evidence for dynamic Cry plasmid loss and gain was found in this study. Several STs were associated with isolates that either expressed or lacked Cry toxins and which are therefore formally Bt or B. cereus respectively. This is typical of the group: several of these STs have been previously identified as Bt (ST18 - B. t. pakistani; STs 56 and 57 - B. t. darmstadiensis); and described as B. cereus in wound infections (ST18, ST57) [22], [50] or diarrheal food poisoning (ST56); [25] and at least two additional genotypes (STs 15 and 109) show this dual identity [21], [22], [24], [51]. Clinical infections with B. cereus group bacteria have very rarely been linked to Bt [52], [53], [54]. This absence of recorded Bt infections in humans may have arisen because B. cereus clinical isolates are not routinely screened for Cry toxins or because real biological differences between Cry expressing and Cry null strains determine distinct ecological niches. Ecological differences could arise because certain genotypes have a much lower probability of carrying Cry plasmids or because regulatory cross-talk between plasmid and chromosome can affect virulence gene expression. Loss of the cereulide plasmid has been linked to a shift in site of infection type [25] and absence of Cry toxins has been associated with improved saprophytic growth in soil [55]. Ecological differentiation in the B. cereus group has been ascribed to changes in gene expression [49], and one unexplored possibility is that plasmids may be partly responsible for this variation.
This natural population of Bt and B. cereus was dominated by a single successful genotype [56]. Previous analyses of the global B. cereus database [22], food borne isolates [21], or clinical isolates [25], [57] have interpreted the B. cereus group as having a more reticulate population structure with more numerous or less dominant clonal complexes. Exceptions include the highly clonal nature of B. anthracis and the near clonal population structure of B. cereus emetic strains associated with cereulide poisoning [58]. Another study of an ecologically distinct population found that B. cereus clades 1 and 2 were both dominated by clonal complexes [27]. This discrepancy between data from unstructured sample collections and ecologically coherent studies may arise because the relative frequency of genotypes in a strain collection does not reliably represent their frequency in the field. Nevertheless, in our natural population, ST8 and its close relatives represent a highly dominant clonal complex of clade 2 strains in a manner strikingly similar to the way in which B. anthracis clones are prominent within clade 1. The analogy is not perfect: Bt ST8 may recombine more freely with relatives than B. anthracis and have more mobile plasmids, although B. anthracis virulence plasmids have now been found in other B. cereus group lineages [59]. Nevertheless, evidence is accumulating which suggests that the ST8 genotype is a successful specialist pathogen of Lepidoptera. Despite having the largest number of isolates in the pubMLST database (50 at this time) ST8 has never been associated with a clinical infection and is therefore strictly an invertebrate pathogen. In contrast to many sequence types, we found that a very high proportion of isolates were associated with Cry toxin expression; B. anthracis has a similar attachment to its virulence plasmids. One possible explanation for this pattern is that Cry toxin plasmids, despite their mobility, will not confer any long-term advantage to bacteria in the absence of unlinked genes that enable them to colonize leaves. Therefore, as we collect more data on the B. cereus group, the more the entomopathogenic specialists, such as ST8, resemble vertebrate specialists, such as B. anthracis, in terms of ecology and population genetics.
Twenty four fine wire mesh enclosures were each planted with six 4-week old Chinese cabbages (Brassica pekinensis var. “One Kilo SB”) in May 2006, in a field margin at Wytham Farm, Oxfordshire, UK. This farm has been managed primarily as pasture for at least three decades, a crop that is not typically sprayed with Bt based pesticides, and there has no been no recorded commercial use of these products at this site. We imposed two treatments in a balanced factorial experiment, using cage as a replicate. These treatments were addition/exclusion of a Lepidopoteran host and presence/absence of a Bt-based biopesticidal spray. The host addition enclosures were seeded with Plutella xylostella eggs and larvae in early June (50 eggs and 15 second instar larvae per plant) and with an additional 10 adults per cage in early July. Sprayed cages were inoculated with B. t. kurstaki HD-1 (DiPel DF, Valent Biosciences) on 20 July 2006 (T0) using 400 ml per cage from a stock of 40 g l−1 of formulated product.
Full sampling of the experiment took place 10 days and 28 days after sprays were applied. From each cage two independent soil samples and six independent leaf samples were collected; soil samples were approximately 1 g and were taken from the top 1–2 cm of the soil surface, leaf samples were approximately 2×2 cm and were taken from both emerging and fully mature cabbage leaves. Sampling methods and isolation of B. cereus group bacteria followed described protocols [26] except that all soil samples and leaf washes were pasteurized (65°C for 30 minutes) prior to plating out. All leaf and soil samples were weighed prior to processing. Two soil samples per cage were also taken immediately before spraying. A subset of randomly selected colonies (2 per soil sample or 6 per leaf) were streaked and stored as glycerol stocks. However, only independently sampled isolates (i.e. a single isolate from each sample tube) were included in the MLST analyses. In all, 384 samples were taken from the fully balanced experiment at time points 10 and 28 days post spraying (2 time points ×24 cages ×8 samples per cage) plus an additional 48 samples from the soil at time point 0. Only 288 independent isolates were available for the CLONALFRAME analysis as many leaf samples did not include any members of the B. cereus group. The MLST allele based analysis of experimental effects on population structure used only independent isolates from the fully balanced experiment in time points 10 and 28.
The crystalline parasporal inclusions of entomocidal Cry toxins produced by Bt were detected via oil-immersion microscopy. In brief, colonies were grown on Luria-Bertani (LB) agar plates until sporulation. After streaking on glass slides, bacteria were fixed in 100% methanol (10 minutes) and then stained with 0.05% w/v Coomassie Blue in a solution of 45% methanol (v/v) and 10% glacial acetic acid (v/v) for 20 minutes. Slides were quickly destained in a solution of 45% methanol (v/v) and 10% glacial acetic acid and washed in de-ionized water before inspection. Isolates were scored as potentially pathogenic if square or bipyrimidal blue stained inclusions were produced in sporulated cultures.
Bacterial DNA was extracted using the CTAB method [60] and diluted to ∼150 ng/µl in nuclease-free H2O. PCR reactions containing 0.25 µM of each of the appropriate forward and reverse primers [24], 0.2 mM dNTPs (Bioline), 0.625 U Taq DNA Polymerase (Qiagen), 2.5 µl 10× buffer and 2 µl template DNA were prepared in a final volume of 25 µl. Reactions were carried out at the following conditions: initial denaturation at 94°C for 240 s followed by 10 cycles of 94°C for 30 s, 55°C for 30 s and 72°C for 80 s, followed by 15 cycles of 93°C for 30 s, 53°C for 30 s and 72°C for 80 s, followed by 25 cycles of 92°C for 30 s, 53°C for 30 s and 72°C for 80 s. A final extension period of 72°C for 480 s completed the reaction. Resulting DNA products were precipitated with 20% polyethylene glycolate solution and re-hydrated with nuclease-free H2O.
Sequencing reactions containing 0.2 µM of the appropriate forward or reverse primer, 0.33 µl Big Dye Terminator v3.1 mix (Applied Biosystems), 2 µl 5× buffer and 2 µl hydrated PCR product were prepared in a final volume of 10 µl. Cycling parameters were carried out as follows: initial denaturation of 96°C for 120 s followed by 38 cycles of 96°C for 20 s, 50°C for 10 s, 60°C for 240 s and a final extension of 72°C for 240 s. Sequencing products were precipitated with 5 M sodium acetate solution and subsequently separated and detected with an ABI Prism 3730 automated DNA sequencer (Applied Biosystems). The resulting sequence traces were aligned and edited using the Phineus software (http://www.phineus.org), and assigned allele numbers. Finally, each 7-locus allelic profile was assigned a sequence type.
In the autumn of 2009 we conducted a greenhouse leaf colonization experiment to compare the ability of B. cereus clade 3 bacteria and Bt ST 8 to colonize the leaf surfaces of Chinese cabbage seedlings (B. pekinensis) from soil. Seeds were surface sterilized in sodium hypochlorite [30] and planted in autoclaved soil based compost, with two seeds per 40 mm diameter pot. We set four negative controls (no inoculation); a treatment using three B. cereus clade 3 isolates (recovered from the 2006 field experiment) and a treatment with four B. thuringiensis ST8 isolates, each isolate being replicated four times. One ST8 isolate (D0 4ai) was recovered from the 2006 field experiment prior to spraying with DiPel, another isolate was derived from a stock of DiPel WP; two additional isolates were recovered from wild Brassica oleracea growing on sea cliffs in Dorset, UK [61]. Spores for soil inoculation were grown on B. cereus specific agar (Oxoid, UK), as per described protocols and enumerated using calibrated spectrophotometer readings. Soil was inoculated with 2×106 spores g−1. Seedlings were cultivated in an insect-proof glasshouse with supplementary heating. After five weeks two leaf samples (200 mm2) were cut from each experimental pot. Epiphytic bacteria were recovered from plant surfaces using the same methods as in the field experiment [26]. Endophytic bacteria were recovered from homogenized surface sterilized samples [62]. B. cereus group bacteria were cultured from unpasteurized samples with Bacillus cereus specific agar.
Analysis of bacterial densities used maximum likelihood mixed model ANOVA with enclosure as a random factor and spray, insect and time point as fixed factors. Significance testing was carried out via sequential deletion of terms from a full model in all analyses. For maximum likelihood models we have reported the degrees of freedom (df) of the model being tested as well the Likelihood ratio (the ratio of the likelihoods of statistical models in the test). Analysis of proportional data (STs, Cry toxin producers) used generalized linear modelling with binomial errors. The analysis of the leaf colonization experiment used a split-plot design and linear mixed effect modelling with bacterial isolate nested within block (fixed factors) and bacterial clade as a fixed factor. Model assumptions (normality, homoscedasticity, error distribution) were confirmed with graphical analyses. The above analyses were carried out in R (http://www.r-project.org).
A genealogy of the STs was produced using clonalframe, a model-based approach for determining bacterial microevolution [23]. This model determined the clonal relationships within the population with improved accuracy compared with standard phylogenetic inference techniques for recombining bacteria because it distinguished between point mutation and recombination - the two major sources of allelic polymorphisms. This model has been used successfully to distinguish species and subspecies clades within the B. cereus group [22]. Analysis was carried out on 228 isolates from the current study augmented with data from 286 unique STs from the B. cereus MLST database at http://www.pubmlst.org [63]; we specifically used data only from isolates associated with published papers in addition to isolates submitted by authors of this paper. A total of 315 unique STs were used to produce a genealogy of the entire Bacillus group. In each case, sequences of all 7 loci were concatenated and analyzed in 3 independent runs of clonalframe, each consisting of 100,000 iterations with the first 10,000 burn-in iterations discarded. Using the clonalframe tree comparison tool, described in the user guide, satisfactory convergence and mixing were confirmed using the Gelman–Rubin test [64], [65]. The consensus tree represents combined data from three independent runs with 75% consensus required for inference of relatedness.
The assignment of probability of potential origin of alleles to populations (pathogen/soil bacterium) was calculated for alleles from isolates within each sample group individually and the percentage of all isolates attributed to each origin population was determined as the sum of these probabilities. The hypothesis that different groups/treatments harbour different allele types was tested using the no admixture model in structure [66]. The clonalframe genealogy was used to define background population structure and individual alleles were independently assigned to source with a training set of alleles from pathogenic and non pathogenic strains distinguished from the test data with use of the usepopinfo flag. Probabilistic analysis in structure used 10,000 burn-in iterations and 10,000 subsequent iterations.
|
10.1371/journal.pntd.0003255 | Caseating Granulomas in Cutaneous Leishmaniasis | Caseating granulomas are often associated with a mycobacterial infection (TB) and are thought to be exceedingly rare in cutaneous leishmaniasis (CL). However, no large series has accurately documented the incidence of caseating granulomas in CL.
A multiregional cohort consisting of 317 patients with CL [Syria (157), Pakistan (66), Lebanon (47), Saudi Arabia (43), Ethiopia (2) and Iran (2)] was reviewed. Clinical [age, sex, disease duration, lesion type and geographic and anatomic location] and microscopic data [presence of and type of granuloma, Ridley's parasitic index (PI) and pattern (RP)] were documented. Presence of microorganisms was evaluated using special stains (GMS, PAS, AFB and Gram) and polymerase chain reaction (PCR) for TB and CL. All cases included in this study were confirmed as CL by PCR followed by restriction fragment length polymorphism analysis for molecular speciation and were negative for other organisms by all other studies performed. Categorical and continuous factors were compared for granuloma types using Chi-square, t-test or Mann-Whitney test as appropriate.
Granulomas were identified in 195 (61.5%) cases of CL and these were divided to 49 caseating (25.2%), 9 suppurative (4.6%) and 137 tuberculoid without necrosis (70.2%). Caseating and tuberculoid granuloma groups were significantly different in terms of the geographical source, with more cases harboring caseating granulomas in Saudi Arabia (p<0.0001). Histologically, both groups were also different in the distribution of their RP (p<0.0001) with a doubling RP3 in caseating granulomas (31% vs. 15%) as opposed to doubling of RP5 in tuberculoid granuloma group (38% vs. 19%). Time needed to achieve healing (RP5) was notably shorter in tuberculoid vs. caseating group (4.0 vs. 6.2 months). Parasitic Index, CL species and other considered variables did not differ for the granuloma type groups.
In our multiregional large cohort, a notable 18.2% of all CL cases harbored caseating granulomas therefore; CL should be considered part of the differential diagnosis for cases with caseating granulomas in endemic regions, especially considering that the regions included in our cohort are also endemic for TB. Of note, cases of CL with caseating granulomas also showed a slower healing process, with no association with specific species, which may be due to worse host immune response in such cases or to a more aggressive leishmania strains.
| Cutaneous leishmaniasis displays a wide spectrum of clinical and microscopic findings. The microscopic manifestations have been categorized into five groups. The type of granulomatous response defined in group V is usually tuberculoid in nature with exceedingly rare cases described with necrotizing granulomas in contrast to cutaneous infections with tuberculosis and other mycobacteria that are typically associated with necrotic granulomas. The old world countries endemic for cutaneous leishmania also happen to be endemic for other granulomatous diseases such as leprosy, tuberculosis and cutaneous mycoses. The most common diagnostic approach used in these countries is still microscopic examination despite the advances in molecular diagnostic techniques and culture methods. We document an 18.2% incidence of caseating granulomas in cutaneous leishmania. Hence, cutaneous leishmania should be part of the differential diagnosis for cases with caseating granulomas in endemic regions in addition to other causative infections.
| Granulomas are characterized by the predominance of histiocytes that evolved into epithelioid cells, forming more or less defined aggregates that may contain various other inflammatory cells. The formation of granulomas is the result of a complex inflammatory interplay between a persistent non-degradable antigen and the host's chronic immune system of macrophage activity, Th1 cell response, B-cell overactivity, circulating immune complexes, and a vast array of biological mediators [1], [2]. Granulomas can be divided according to the inflammatory cells involved, their arrangement and the presence of associated features into necrotizing (caseous and suppurative) and non-necrotizing (tuberculoid, sarcoidal, xanthogranulomas, palisaded, foreign body etc.) [3], [4]. (Figure 1)
Cutaneous leishmaniasis (CL) displays a wide spectrum of clinical and microscopic findings that has been extensively described. Ridley et al. proposed grouping the microscopic manifestations into five groups. Group I represents an almost normal appearing skin biopsy with patches of collagen degeneration. Group II shows a predominant severe necrotizing process in the dermis. In Group III, the dermis is involved by a diffuse and heavy mixed inflammatory infiltrate. Group IV shows scattered Langhans giant cells and primitive epithelioid histiocytes. Well-formed granulomas and well-developed epithelioid histiocytes are prominent in Group V [5], [6], [7]. In contrast to cutaneous infections with Tuberculosis (TB) and other mycobacteria that are typically associated with necrotic granulomas [8], [9], [10], the type of granulomatous response defined in CL, group V, is tuberculoid in nature with exceedingly rare cases described with necrotizing granulomas [11], [12], [13]. Furthermore, Ridley modified parasitic index quantifies the parasitic load of amastigotes in cutaneous lesions and has a numerical score from 1 to 6 as displayed in table 1.
The old world countries endemic for CL also happen to be endemic for other granulomatous diseases such as leprosy, tuberculosis and cutaneous mycoses. The most common diagnostic approach used in these countries is still microscopic examination despite the advances in molecular diagnostic techniques and culture methods [14], [15].
In our large cohort of 317 patients with molecularly proven diagnosis of CL, we encountered a significant number of caseating granulomas overlapping clinically and microscopically with the diagnostic picture of TB. This unusual finding can hamper the expedient diagnosis of CL in endemic and nonendemic regions especially with international travel and influx of immigrants from areas of the world where this parasite is nonendemic [16], [17]. Considering the implications of this new finding, we report the accurate incidence of caseating granulomas in CL and its related variables.
A search of the pathology and dermatology archives at the American University of Beirut Medical Center, Beirut, Lebanon; Tishreen University, Lattakia, Syria; Saad Specialist Hospital, Al Khobar, Kingdom of Saudi Arabia; and Shaukat Khanum Memorial Cancer Hospital and Research Centre, Lahore, Pakistan was performed. The search included patients with skin lesions diagnosed with CL between 1992 and 2010. Cases with sufficient clinical data and material for diagnosis confirmation by PCR were included. Cases with mucocutaneous leishmaniasis, visceral leishmaniasis and those who had received prior treatment were excluded. This study was approved by the American University of Beirut Institutional Review Board and the patient data used in this study was anonymized.
Formalin-fixed paraffin-embedded tissue blocks of 4.0-mm punch biopsies were obtained for each patient. Three hundred and seventeen skin biopsies from 317 patients (1 biopsy per patient) with CL in Lebanon (n = 47), Syria (n = 157), Saudi Arabia (n = 43), Pakistan (n = 66), Ethiopia (n = 2) and Iran (n = 2) were evaluated. Part of this material has been used in a previous publication by the same group [18], [19]. Clinical data collected included patient's age, gender, geographic location, eruption type (papule, nodule, verrucous or scar), duration and anatomic site. Microscopically, the hematoxylin and eosin stained sections were retrieved on all cases and reviewed by two pathologists (IK and JA). Multiple parameters were recorded including the modified Ridley's parasitic index (PI, Table 1),) [20] Ridley's pattern (RP) and the presence and the type of granuloma (Figure 1). Patients with immunosuppresion were excluded from the study. Despite the wide period of time (1992–2010), all cases were diagnosed and triaged in the same method.
PCR confirmation was performed on all included cases with subsequent molecular subspeciation performed on all PCR-positive cases, following previously published protocol [21]. In brief, DNA was extracted from ribbons originating from the formalin-fixed paraffin-embedded tissue blocks. PCR assay was performed to amplify the Leishmania ribosomal internal transcribed spacer 1 (ITS1) using the primers LITSR (5′ -CTGGATCATTTTCCGATG-3′) and L5.8S (5′-TGATACCACTTATCGCACTT-3′). This was carried out using the VersoTM 1-Step ReddyMix Kit (Thermo Fisher Scientific Inc, Surrey, UK) in an amplification reaction of 50 µL. The Px2 thermal cycler (Thermo Electron Corporation, Waltham, Massachusetts, USA) was used for amplification under the following steps and conditions: 95°C for 2 min, 35 cycles of (95°C for 20 seconds, 53°C for 30 seconds, 72°C for 1 minute) and 72°C for 6 minutes. Following PCR amplification, digestion of the ITS1- PCR amplicons with restriction enzyme HaeIII was performed for restriction fragment length polymorphism (RFLP) analysis and consequent subspeciation. The ITS1 RFLP technique used allowed the identification of all clinically significant strains, including Leishmania tropica, Leishmania major, Leishmania braziliensis, Leishmania donovani, Leishmania aethiopica and Leishmania infantum [22].
The TB PCR test used for the detection of Mycobacterium DNA is the MTB/RIF assay (CEPHEID, Sunnyvale, USA), and performed according to manufacturers' instructions. The 3 specific primers in the Xpert MTB/RIF assay amplify a portion of the rpoB gene (accession number: AF057488.1) containing the 81 base pair “core” region. The 5 probes A, B, C, D, and E are able to differentiate between the conserved wild-type sequence and mutations in the core region that are associated with RIF resistance. This assay is performed on the GeneXpert platform by CEPHEID.
Special stains were conducted after deparaffinization and hydration with distilled water. GMS, PAS, AFB and GRAM stains were performed in the traditional way with standardized techniques [23], [24].
Continuous variables were analyzed by t-test or Mann-Whitney rank sum test as appropriate.
Categorical variables were analyzed using chi-square test. A two-tailed p<0.05 was required for statistical significance.
The age of the patients recruited for this study ranges from 1 year old to 92 years old (median 24 years, SD = 21.23). The male gender was predominant (n = 175, 55.2%). The most common site involved was the head and neck (n = 145, 45.7%). Patients presented mainly with a plaque/nodular lesion (n = 156, 49.2%) followed by ulcer/verrucous lesions (n = 144, 45.4%). The duration of the lesion at biopsy time varied from 2 weeks to 132 months (median 5 months, SD = 10.6).
Microscopic evaluation of the 317 cases of CL is illustrated in table 2. Sequencing showed a predominance of the leishmania tropica species in the majority of the studied 317 cases (n = 279, 88.0%) and in both groups of granuloma. Leishmania major was identified in the rest (n = 38, 11.9%). All cases were negative by MTB/RIF assay.
Out of the 317 cases there was a tuberculoid granuloma in 137 (43.2%) cases vs. a necrotizing granuloma in 58 (18.2%) cases including the 9 suppurative granuloma cases of CL. The majority of the cases with necrotizing granulomas were of caseating type (84.5%) the rest were suppurative in nature (Figure 2). For statistical purposes, cases with caseating and suppurative granulomas were lumped together as cases with necrotizing granulomas.
The only statistically significant clinical difference between the tuberculoid granuloma and the necrotizing granuloma cases was the geographic location. As such patients from Saudi Arabia exhibited a higher predominance of necrotizing granuloma (20.7%) than tuberculoid granuloma (8.0%). In contrast, patients from Syria showed a higher prevalence of tuberculoid granuloma cases (61.3%) than necrotizing granuloma cases (43.1%) with a p<0.0001.
For the microscopic features, the only statistically significant difference (p<0.0001) was the distribution of the RP where RP 2 and RP3 were more frequently encountered in cases with necrotizing granulomas (29.3% and 31%) vs. cases with tuberculoid granulomas(1.5% and 15.3%) respectively. RP 5 was mostly noted in association with tuberculoid granulomas (38.0%) than necrotizing granulomas (19.0%). The parasitic index variation between the two groups was not statistically significant (p = 0.09, see table 3).
Leishmaniasis is a group of protozoan disease transmitted to human beings by the bite of female sandflies of the genera Phlebotomus in the old world and Lutzomyia in the new world [25]. CL is the most common form of Leishmaniasis and is endemic in more than 70 countries worldwide [15]. The diagnosis of CL is based on microscopic findings that seem to depend on the stage of evolution of the lesion. Early stage typical findings include a dense diffuse infiltrate of lymphocytes, plasma cells and histiocytes containing amastigotes throughout the dermis (RP 3–4). As the lesion progresses, tuberculoid granulomas with very few organisms if any impinge on the epidermis (RP 5) rendering the diagnosis more difficult [26].
The frequency of caseating granulomas in CL cases has been debatable in the literature.
Two studies identified caseating granulomas in association with CL. The first study performed by Boer et al. on a northern European population of 19 patients with CL identified granulomatous dermatitis in all cases. Nine of these granulomas were tuberculoid, five were sarcoidal and 2 were caseating. Sequencing of the following cases revealed all Leishmania to be L. infantum. In this study, 18 of 19 patients were misdiagnosed clinically and nine were also misdiagnosed histopathologically and included the atypical presentation of sarcoidal and necrotizing granulomas [17].
The second study included a larger cohort from Mexico and included 73 biopsies with localized CL. This study identified 21/73 (28.7%) cases of necrotizing granulomas vs. 32/73 (43%) cases of unorganized granulomas without necrosis. There was no relationship regarding the age and sex of the patient with the microscopic findings and the response to treatment. The responsible leishmania strain isolated was L. Mexicana [11].
On the other hand, the following 2 relatively large series showed no evidence of caseating granulomas in association with CL. Venkataram et al. described a series of 40 patients with CL from the Sultanate of Oman. Non-necrotizing granulomas were present in eight patients. The rest had no granulomas. As other parts of the Near East, leishmania tropica was the causing strain in this study [27]. In a larger series of 149 patients from Madrid during an outbreak of CL, the most encountered microscopic manifestation was non-necrotizing granulomas in 100 cases (67%) followed by a rich lymphohistiocytic infiltrate in 46 cases (31%). L. infantum was identified as the causative agent in 98% of the cases [25].
Our study represents the largest cohort of patients (n = 317) coming from different counties in the Near East region (Syria, Lebanon, Saudi Arabia, Ethiopia, Pakistan and Iran). The longest time to biopsy in our cohort, 132 months, compared to a maximum time of 42 months in the other discussed studies probably out of the socioeconomic status of the population and the provided medical care [25]. The high occurrence of necrotizing granulomas (18.2%) noted in our study has been rarely described in the literature except for a cohort of 71 patients from Mexico (28.7%) as previously mentioned. Moreover, our study of the Near East region relates some interesting variables to the occurrence of necrotizing granulomas that can raise the possibility of the diagnosing of CL. Especially that at the granulomatous stage of CL, identification of the amastigotes in the tissue can be very challenging which may lead the investigator toward other infectious diseases such as tuberculosis, leprosy and fungal infections.
Leishmania tropica has been reported to be the strain responsible for CL in the study coming from Oman which is the same strain recognized in 88% of our cases [27]. In addition, the dominant RP identified in our series are RP3 and RP4. Conversely, the most identified RP in the above discussed studies is granulomatous dermatitis (RP5) especially the tuberculoid type except for the study coming from Oman. The later study showed concordant RP with ours which happened to be the only study with the same causative strain “L. Tropica” as our study in contrast to L. Infantum and L. Mexicana identified in the other discussed studies. This can be explained by the adjunct geographical and close environmental factors between the Oman cohort and our cohort. These findings support the importance of the interaction of several factors related to the strain, vector and the host on the final clinicopathologic manifestation of the disease.
In our series, patients with caseating granulomas showed a slower healing process than patients with tuberculoid granulomas (6.2 months vs. 4.0 months) with no significant association with the specific strain or age group. This may be explained by inadequacy of the host immune response. In fact, studies have shown that necrotizing granulomas are due to the formation of immune complexes between antibodies and excess antigen in the center of the lesion. This setting develops when cell-mediated immunity, initially strong, begins to decline for unknown reasons, allowing the parasite to proliferate out of macrophage control. If cell-mediated immunity then improves again the number of microbes diminishes and immune complexes will form in antibody excess causing epithelioid granuloma formation instead of necrosis [2], [10]. This also explains the difficultly in identification of leishmania amastigotes during the granulomatous stage of the disease (late stage, RP5).
In conclusion, we document an 18.2% incidence of caseating granulomas in CL. Hence, CL should be part of the differential for cases with caseating granulomas in endemic regions in addition to TB and other causative infections. In addition, cases of CL with caseating granulomas also showed a slower healing process, with no association with specific species, which may be due to worse host immune response in such cases or to a more aggressive leishmania strains.
|
10.1371/journal.pgen.1007123 | Unlinking the methylome pattern from nucleotide sequence, revealed by large-scale in vivo genome engineering and methylome editing in medaka fish | The heavily methylated vertebrate genomes are punctuated by stretches of poorly methylated DNA sequences that usually mark gene regulatory regions. It is known that the methylation state of these regions confers transcriptional control over their associated genes. Given its governance on the transcriptome, cellular functions and identity, genome-wide DNA methylation pattern is tightly regulated and evidently predefined. However, how is the methylation pattern determined in vivo remains enigmatic. Based on in silico and in vitro evidence, recent studies proposed that the regional hypomethylated state is primarily determined by local DNA sequence, e.g., high CpG density and presence of specific transcription factor binding sites. Nonetheless, the dependency of DNA methylation on nucleotide sequence has not been carefully validated in vertebrates in vivo. Herein, with the use of medaka (Oryzias latipes) as a model, the sequence dependency of DNA methylation was intensively tested in vivo. Our statistical modeling confirmed the strong statistical association between nucleotide sequence pattern and methylation state in the medaka genome. However, by manipulating the methylation state of a number of genomic sequences and reintegrating them into medaka embryos, we demonstrated that artificially conferred DNA methylation states were predominantly and robustly maintained in vivo, regardless of their sequences and endogenous states. This feature was also observed in the medaka transgene that had passed across generations. Thus, despite the observed statistical association, nucleotide sequence was unable to autonomously determine its own methylation state in medaka in vivo. Our results apparently argue against the notion of the governance on the DNA methylation by nucleotide sequence, but instead suggest the involvement of other epigenetic factors in defining and maintaining the DNA methylation landscape. Further investigation in other vertebrate models in vivo will be needed for the generalization of our observations made in medaka.
| The genomes of vertebrate animals are naturally and extensively modified by methylation. The DNA methylation is essential to normal functions of cells, hence the whole animal, since it governs gene expression. Defects in the establishment and maintenance of proper methylation pattern are commonly associated with various developmental abnormalities and diseases. How exactly is the normal pattern defined in vertebrate animals is not fully understood, but recent researches with computational analyses and cultured cells suggested that DNA sequence is a primary determinant of the methylation pattern. This study encompasses the first experiments that rigorously test this notion in whole animal (medaka fish). In statistical sense, we observed the very strong correlation between DNA sequence and methylation state. However, by introducing unmethylated and artificially methylated native genomic DNA sequences into the genome, we demonstrated that the artificially conferred methylation states were robustly maintained in the animal, independent of the sequence and native state. Our results thus demonstrate that genome-wide DNA methylation pattern is not autonomously determined by the DNA sequence, which underpins the vital role of DNA methylation pattern as a core epigenetic element.
| DNA methylation is central to the epigenetic control of transcription in vertebrates and is essential for cell differentiation and embryonic development [1–3]. While the cytosines in cytosine-guanine (CpG) dinucleotides are extensively methylated throughout vertebrate genomes, unmethylated CpGs are commonly found clustered at high density inside gene regulatory elements, such as promoters and enhancers. Previous studies have revealed that the methylation state of regulatory regions governs the expression of their associated genes [4,5]. Furthermore, aberrant changes in the methylation state can lead to deregulated transcription, resulting in cellular dysfunction, diseases and developmental abnormality [6,7].
Given its direct governance on transcription, the methylation landscape needs to be precisely specified and modulated. The DNA methylation pattern is established and maintained through highly dynamic biological processes, in which the methylome undergoes substantial, yet precise, changes. For instance, differentiating cells faithfully acquire specific methylation landscapes that are unique to their committed cell types [8–10]. Remarkably, in human and mice, the DNA methylome is extensively erased [11,12] and fully reconstituted during gametogenesis and early embryonic development [13–15]. These facts suggest that the methylation landscape is pre-defined by genetic information. Thus, deciphering how the methylation pattern is encoded is a prerequisite for understanding of differentiation processes and the pathogenesis of various diseases [6,16–18]. However, by what means the methylation pattern is defined in vivo remains enigmatic.
Researches for the past decade proposed that DNA methylation pattern depends on local sequence context. In particular, in silico analyses asserted that there is the strong statistical association between sequence variants and differential DNA methylation states in vertebrates, from fish [19] to human [20]. A number of recent in vitro studies using cultured cells further demonstrated that high CpG density or the presence of specific DNA sequences that contain transcription-factor binding sites is capable of autonomously determining local hypomethylation in the globally methylated genome [21–24]. These recent in silico and in vitro reports support the notion that DNA methylation pattern is primarily determined by local sequence context [21]. However, the anticipated sequence-dependency of DNA methylation is in contradiction to the pioneer in vitro experiments in early 80’S [25–27], in which the methylation status of exogenous DNAs (either artificially CpG-methylated or completely unmethylated) was found maintained with certain fidelity for many cell generations upon stable genome integration. Given these opposing results, the sequence-dependency of DNA methylome seems less concrete than recently anticipated.
Importantly, the above ideas have never been well demonstrated nor rigorously tested in vivo. In this respect, the report by Long et al. [28] provided valuable insights by studying the DNA methylation state of the 42-Mbp fragment of human chromosome 21 in the Tc1 trans-chromosomic mice, as well as the mouse genome loci-containing transgene constructs that were artificially transposed into the zebrafish genome. Their results suggested the existence of sequence-dependent DNA methylation in vivo, but their analyses only focused on non-native sequences (i.e. examining human genomic sequence in mouse, or mouse genomic sequence in zebrafish). Likewise, Li et al. [29] examined the methylation status of a transgene across three generations in rat and found the stable acquisition and inheritance of DNA methylation pattern, but the transgene examined was composed of a mouse promoter and human gene. Thus, it is difficult to draw a general conclusion with these studies on the causal relationship between DNA sequence and methylation in native context in vivo.
Herein, we report the first experiments that rigorously tested the governance of DNA methylation state by nucleotide sequence in vivo. The small laboratory fish, medaka (Oryzias latipes), was chosen as an experimental model for their relatively small genome size (approx. 700 Mbp), short generation time (2.5 to 3 months), ease of in vivo genetic manipulation, oviparity, in addition to their capability of producing 10–20 fertilized embryos per pair on daily basis [30,31]. Importantly, the medaka has polymorphic inbred lines from two geographically separated subpopulations living in the northern and southern part of Japan (2.5–3% SNP rate, for review, see [32]), and their genomes and methylomes were already decoded [19,33,34]. Although vertebrates could have variable DNA methylation dynamics, particularly during early embryonic development (e.g., the genome-wide methylation erasure immediately after fertilization is highly extensive in human and mice [11,12], but very subtle or virtually absent from sheep [35], medaka [36] and zebrafish [37]), the ultimate zygotic DNA methylation landscape is highly conserved from fish to mammals [37–39]. In addition, since an extensively methylated genome is believed to be prerequisite for the onset of vertebrate evolution [40–42], the molecular mechanisms and logic underlying the patterning of DNA methylome are likely conserved among vertebrates. Hence, observations made on medaka can potentially shed light on the postulated, yet unproven, link between genomic sequences and DNA methylation in vertebrates. Contrary to expectation, our results suggest that nucleotide sequence, by itself, cannot dictate its own methylation state in vivo, which argues against the prevailing view of DNA methylation in vertebrates.
Statistical association between medaka genomic sequences and local methylation states was modelled using support vector machine (kmer-SVM [43]). Hypomethylated and hypermethylated genomic loci (a.k.a. hypomethylated domains, “HypoMDs”, and hypermethylated domains, “HyperMDs”, respectively) at the blastula stage (Stage 11 according to Iwamatsu [44]) were identified using the same criteria as described by Nakamura et al. [45] (see also Fig 1A). While HypoMDs and HyperMDs are not readily discernible in terms of length and GC composition (S1 Fig: panel A & B), they bear conspicuous difference in their sequence pattern, allowing robust in silico classification and accurate prediction of the methylation states by the SVMs based solely on nucleotide sequence (Fig 1B: area under precision-recall curve ≥ 0.83, versus 0.08 from the random classifier). Consistent with the fact that the median CpG density in HypoMDs is higher than that in HyperMDs (S1 Fig: panel C), sequence pattern enriched in HypoMDs display higher frequency of CpG (S1 Table: left columns). Furthermore, CpG-masking prior to the training of SVM could still result in models with modest classification performance (Fig 1C: area under precision-recall curve ≥ 0.53), suggesting that specific, CpG-free DNA motifs are also differentially enriched in HypoMDs and HyperMDs (S1 Table: columns on the right). All these reinforce the notion that, similar to other vertebrates, there is the strong statistical association between genomic DNA sequences and their methylation states in medaka.
To test the dependency between genomic sequences and their methylation state in vivo, we generated transgenic fish that ectopically carry full-length HypoMD or HyperMD, along with their 1.5 to 2-kb up- and down-stream sequences. To distinguish the endogenous and the ectopic copies of the assayed sequences, we specifically selected HypoMD and HyperMD that are differentially methylated in two closely related, inbred strains of medaka: HdrR and HNI [32], i.e. being a HyperMD in HdrR but exists as HypoMD in HNI, or vice versa (Fig 2). The differential states of these homologous sequences in the two strains were presumably due to minor variation in their nucleotide sequences [19]. These transgenic fish helped reveal not only if the differential methylation state is genuinely due to sequence polymorphisms, but also if genomic sequence at ectopic loci could stably recapitulate its endogenous state over a substantial timeframe and across generations (i.e. > 6 months, for the collection of F2 transgenic embryos).
Three transgenic lines were examined, in which DNA sequences from HNI (either endogenously HyperMD or HypoMD) were inserted into the host drR strain (outbred, parental strain of HdrR) (see Fig 2 for schematic illustration). Host drR and inserted HNI sequences were easily discriminated by SNPs. In concordance with the notion that nucleotide sequence can autonomously determine its own methylation state, the integrated full-length HypoMDs were completely unmethylated in the F2 transgenic blastula embryos (Fig 2A & 2B: “core”). However, on the other hand, the integrated full-length HyperMD (Fig 2C: “core”) were also found poorly methylated in the transgenic embryos, which is in stark contrast to its native hypermethylated state. Moreover, while all flanking sequences tested are endogenously hypermethylated in both strains, they were poorly methylated ectopically (Fig 2A–2C: “flank (L)” and “flank (R)”). In fact, substantial de novo methylation was not evident throughout all three integrated sequences, regardless of inside HypoMD, HyperMD, or their flanking regions. Since the transgene constructs were initially propagated in E. coli as bacterial plasmids and were thus completely devoid of CpG methylation prior to transgenesis, these observations suggested that the initial absence of CpG methylation on the transgenes was faithfully maintained regardless of their sequence and respective endogenous methylation states for at least 6 months and across 3 animal generations. This indicates that these assayed genomic sequences (1) do not carry methylation determination information and/or (2) randomly integrated into loci (e.g., inside or in close proximity to expression cassettes) that were under strong influence of preexisting epigenetic factors.
Given the above unexpected observations, a substantial number of genomic fragments of medaka was interrogated to comprehensively test the general presumption that genomic sequences can genuinely determine their own DNA methylation state in vivo. Medaka genomic DNA was digested and enriched for CpG-containing fragments (approx. 40–220 bp; extended to approx. 184–364 bp with adapters) using a library preparation method akin to that was designed for reduced representation bisulfite sequencing (RRBS) [46]. The PCR-amplified (hence, unmethylated) fragments were labeled (methylation at the N6 position of adenine in the Dam sites, 5’-GATC-3’, of the adapters), followed by (or without) artificial CpG methylation in vitro, then introduced into medaka zygotes at the one-cell stage, and allowed for highly efficient I-SceI-mediated random genome integration (see Fig 3A for graphical procedures). According to Thermes et al. [47], the integration event was expected to occur at the one-cell stage, i.e. immediately after injection. At the blastula stage (2000 to 4000 cells per embryo), after the removal of unintegrated fragments by size-selection and DpnI-digestion (S2 Fig: panel A), the methylation state of the integrated fragments was determined via bisulfite PCR and high-throughput sequencing. The assayed integrated fragments encompassed nearly the entire range of GC content and CpG density of HypoMDs and HyperMDs (S3 Fig vs S1 Fig). Approximately equal number of CpGs from HypoMDs and HyperMDs were assayed (S4 Fig: top vs bottom panels).
In spite of the strong statistical association between nucleotide sequence and methylation states, the integrated genomic fragments failed to recapitulate their endogenous methylation state at ectopic locations. The methylation rate at endogenous loci and that at ectopically integrated locations showed essentially zero statistical correlation: Spearman’s ρ ≤ 0.08, Kendall’s τ ≤ 0.07 (see also S5 Fig for the biplots). Without prior artificial methylation, CpGs on the integrated fragments were almost entirely unmethylated regardless of their endogenous states (Fig 3B: upper-left vs lower-left panel). The lack of sequence dependency was further illustrated by a drastically different ectopic methylation pattern when the genomic fragments were artificially methylated prior to injection and genome integration (Fig 3B: left panels vs right panels). The sharp contrast in the ectopic methylation patterns suggested that nucleotide sequence does not carry adequate information for its own methylation state, or the integrated fragments could escape de novo DNA methylation (which occurs at some point between 64-cell stage and blastula stage [36]) and any expected sequence-dependent demethylation in early medaka embryos.
The artificially methylated, integrated fragments contained a substantial number of unmethylated CpGs when examined at the blastula stage (Fig 3B: upper- and lower-right panels). These unmethylated CpGs were unlikely due to incomplete artificial methylation prior to injection for the following reasons. The methylase (CpG DNA methyltransferases M.SssI) used is known to completely methylate CpGs in all sequence context [48]. This was routinely achievable by our optimized reaction regimen (see S6 Fig for examples using bacterial genomic DNA and vector library that have higher CpG frequencies per unit weight of DNA than the medaka genome). The observed unmethylated CpGs could be caused by demethylation in the injected embryos. However, such demethylation could not be directly inferred as recapitulation of the endogenous methylation state, since there was essentially zero correlation between the endogenous and ectopic states (Fig 3B: upper-right vs lower-right panel; see also panel B in S5 Fig). In addition, the observed loss of premethylated state was unrelated to the endogenous chromatin accessibility (hence, potential binding of- or recognition by- transcription factors), as CpGs originated from heterochromatin and euchromatin were equally susceptible to the loss of methylation (right panels of S7 Fig; note the peaks at 0% methylation rate in the histograms along Y-axes). We also compared the nucleotide sequences (10 bp from both up- and down-stream) encompassing CpGs that were demethylated to those that were maintained as hypermethylated using kmer-SVM with the same parameters as above. However, the resultant SVMs were highly imprecise and insensitive (S8 Fig: area under precision-recall curve ≤ 0.47, versus 0.43 from random classifier). Moreover, the overall ectopic methylation states, as well as the demethylation, of the integrated fragments do not correlate with their size or CpG density (S9 Fig). Together, we concluded that the observed demethylated state was not related to intrinsic sequence features of the genomic fragments.
Given that the injected genomic fragments were (1) only partial fragments of HypoMDs or HyperMDs and may lack the presumed sequence features that are required for autonomous determination of methylation state, and (2) integrated into random genomic positions where they might be influenced by local chromatin state, we speculated that the observed demethylation might be due, at least in part, to the local epigenetic state of the integrated loci (i.e. position effect; E.g., integrated into preexisting HypoMDs or somewhere under the influence of trans-acting hypomethylation determining elements, hence rendered hypomethylated). Subsequent experiments were thus conducted at pre-specified genomic loci to control for the possible position effect.
In order to examine whether full-length HyperMDs and HypoMDs can autonomously determine their own methylation state at an inert genomic location, six unmethylated HyperMDs and eleven pre-methylated HypoMDs were injected into one-cell stage medaka embryos and integrated into the gene desert region presumably devoid of any possible influence of active regulatory elements (see also Fig 4A). The integration was achieved by the highly efficient, PhiC31 integrase-mediated site-specific integration in medaka and was expected to occur at the one-cell stage [49]. Methylation states of the integrated sequences were examined at the blastula stage. Autonomy in methylation state determination by the full-length, integrated sequences would manifest as remethylation of the unmethylated HyperMDs, as well as active or passive loss of the methyl groups on the premethylated HypoMDs, after genome integration (see also Fig 4B for illustration of the logic of the experiment).
In concordance with the above experiments, all of the unmethylated, integrated HyperMDs failed to acquire methylation (Fig 4C). Likewise, the pre-methylated, integrated HypoMDs remained hypermethylated (Fig 4D), with very limited number of CpG dinucleotides (i.e. only 4 out of the 202 CpGs inspected) having no methylation (Fig 4D: blue dots on the integrated, ectopic copies of HypoMDs/Loci 1, 4, 6, and 9). Since we were unable to determine the methylation state of these distinct CpGs in the premethylated plasmid library (as plasmid DNA converts very poorly in bisulfite reaction), it is possible that these CpGs were not fully methylated prior to injection. However, as aforementioned, the M.SssI methyltransferase used in the pretreatment has no known sequence specificity. The observed absence of methylation probably reflects highly localized loss of methyl groups on these specific CpGs. Collectively, the above results indicate that the overall, ectopically introduced nucleotide sequences were not perused and the artificially conferred methylation states (i.e. hypomethylation in the HyperMDs, and hypermethylation in the HypoMDs) were robustly maintained in vivo.
Finally, we edited the methylation state in situ to exclude the risk of artifacts possibly incurred by ectopic genome locations. The methylation state of two HypoMDs were edited in situ via CRISPR-Cas9-triggered homology directed repair (HDR) and artificially methylated repair templates (see Fig 5A for illustration of concept behind the experiment). Consistent with the aforementioned observations, in spite of the original hypomethylated state, the loci were rendered largely hypermethylated in the edited blastula embryos (Fig 5B & 5C). Since the observed lack of restoration of native methylation state could be due to the seemingly limited time allowed for the recapitulation (from injection to sampling, i.e. from 1-cell stage to blastula: approx. 8 hrs, encompassing 11–12 rounds of cell divisions), we repeated the editing experiment and extended the endpoints to later developmental stages at 3- (Stage 31) and 7- (50% hatched and free-swimming; i.e. Stage 39) day-post-fertilization (i.e. day-post-injection). Yet, the edited alleles remained hypermethylated in the mid-/late-stage embryos (S10 Fig). Significant loss of methyl groups could only be observed on two distinct, adjacent CpGs in one of the two edited loci (S10 Fig, panel B: the 1st and 2nd CpG). Taken together, these observations indicate that genomic sequence and its methylation state were not coupled even at the endogenous position.
Although DNA methylation is the best characterized epigenetic signature [50], the molecular basis and logic of its establishment still remain elusive. Given that CpG dyads are predominantly methylated unless they are clustered at high density [51], it is generally presumed that hypermethylation is the default state of vertebrate genomes and specific regions (i.e. gene regulatory elements) are protected from de novo methylation, rendering them hypomethylated [21,24,52–55]. Intensive researches for the past decade have demonstrated that the protection on the genomic loci is possibly mediated by nucleosome positioning [56–58] and/or the recruitment of a myriad of proteins [12,59–63] which eventually block off local access of DNA methyltransferases or remove methylation on cytosines in vicinity through oxidation and thymine DNA glycosylase (TDG)-mediated base excision repair. However, little is known about how are these factors specifically predisposed on the preselected loci.
As aforementioned, recent in vitro studies demonstrated that nucleotide sequence features (especially high CpG density and the presence of certain transcription factor binding sites) autonomously determined the local hypomethylated state [21–24,64,65]. However, this preposition has never been rigorously verified in vivo, presumably due to the fact that interrogation of genomic sequences at genome-wide scale requires large number of subject animals, which is prohibitive with classical mammalian models (e.g., rodents). With the use of medaka as an alternative vertebrate model, our results definitively showed that there is no immediate connection between DNA methylation state and underlying nucleotide sequence in vivo, in spite of their strong statistical association. By manipulating and controlling the methylation state of genomic sequences prior to reintegration into the genome, we demonstrated that the artificially established methylation states were predominantly maintained in medaka in vivo, independent of their nucleotide sequences and native methylation states. This also appears to be true for the transgene that passed across generations. Our results thus argue against not only the recently inferred determining role of DNA sequence on the methylation landscape, but also the longstanding belief that there is a default state (i.e. hypermethylated) for the vertebrate genomes.
In fact, the postulated strict sequence-dependency seems paradoxical to the concept of epigenetics itself. There are accumulating reports for the last two decades that DNA methylation could be perturbed by transient physiological stress or chemical exposure. More importantly, the perturbed states could be highly persistent and inheritable, while the underlying genomic sequence remains unchanged [66–68]. These observations highlighted that DNA methylation pattern is not directly coupled with the underlying nucleotide sequence in vivo, in spite of what has been recently shown in silico and in vitro.
However, our results do not rule out the existence of highly confined, local sequence-dependent DNA methylation. As proposed by Richards [69], the sequence-dependency of epigenetic signatures may vary with actual sequence-context, i.e. some nucleotide sequences may favor or even fully mandate certain methylation state, while others may be completely independent of DNA methylation. Although the artificially established hypermethylated state of HypoMD sequences examined in this study was mostly maintained after genome integration, we observed spontaneous, complete loss of methyl groups on some CpGs in the eleven pre-methylated HypoMDs, as well as within one of the in situ edited loci. This suggests the presence of local sequence elements that facilitate demethylation on specific CpGs, although their effect was spatially confined. As previously demonstrated in vitro, some DNA motifs, in particular several transcription factor binding sites (reviewed by Blattler and Farnham [70]), are indeed instructive to DNA methylation and may account for the change of methylation state in specific loci upon differentiation [18,71]. Importantly, their effect was also demonstrated to be limited to no more than a few tens of base pairs up- and down-stream [22,23]. It is thus likely that the restricted governing range (< 100 bp) of these DNA sequences is insufficient to account for the span of HypoMDs (median length > 1 kb).
The apparent lack of sequence dependency can be explained by the involvement of epigenetic factors in DNA methylation in vivo, as suggested by Kaminsky et al. [20]. Genomic fragments tested in the present study and previous works were all purified prior to reintegration into the genome, hence lacked any associated factor(s) that can modulate DNA methylation. Future experiments will need to address the presumed methylation determining factor(s), their deposition onto specific locations of the genome, and their inheritance across cell division and animal generations. The strong link between DNA methylation, nucleosome position and histone modifications [72–74] could provide a hint for further investigation.
In summary, with the use of medaka as a vertebrate model, our data presented herein oppose the recent proposition that the genome-wide DNA methylation pattern in vertebrates is primarily and autonomously designated by the underlying genomic sequence in vivo, but instead provide insights into potential involvement of other epigenetic factor(s) in defining the DNA methylation landscape. Our results demonstrate that the DNA methylation landscape and genomic sequence are not directly coupled, which underpin the widely-observed plasticity of DNA methylation along differentiation, as well as the transgenerational inheritance of perturbed DNA methylation in vivo. However, it is worth noting that vertebrate species could have variable methylation dynamics of DNA methylation during development and growth, especially during early embryonic stages, although underlying molecular mechanisms are probably conserved. This is true even within the same clade of vertebrate species, such as mammals [35]. Further investigation in other vertebrate models will definitely be needed before generalization of our observations made on medaka.
The culture and handling of medaka and their embryos followed the protocols and guidelines published in "Medaka: Biology, Management, and Experimental Protocols" (ISBN: 9780813808710). Experiments were conducted with the permission of Life Science Research Ethics and Safety committee of the University of Tokyo (Permission number: 14–05).
Published whole genome bisulfite sequencing reads of medaka blastula embryos [75] were fetched from the Data Bank of Japan (accession number: SRX149583). Individual reads were trimmed to remove primers, adapters, and low quality basecalls (Phred score ≤ 3) using BBDuk from the BBTools ver. 35.85 [76]. Trimmed reads were mapped to the latest (as of the time of this writing) medaka genome assembly ver. 2.2.4 [34,77] (all genome coordinates reported herein refer to this assembly version) using bwa-meth ver. 0.2.0 [78]. Methylation rates of the mapped CpG dyads were then extracted using MethylDackel ver. 0.2.1 [79] with the default quality filters of MAPQ score ≥ 10 and Phred score ≥ 5. Only those CpG dinucleotides with coverage of ≥ 5× were considered as valid calls [75] and the final mean coverage after filtering was 8×. The same filtering criteria were also applied to all experiments throughout this study, wherever they are applicable. And, unless otherwise specified, the endogenous methylation states of sequences assayed in this study were directly extracted from this mapped, filtered dataset.
HypoMDs calling followed the same definition as previously published [19,45]. Specifically, any stretch of ten or more hypomethylated (methylation rate < 40%) CpGs with no more than four interleaving non-hypomethylated (methylation rate ≥ 40%) or undetermined (unsampled, unmappable or low coverage) dyads were called as HypoMD. HyperMDs were analogously defined as any stretch of at least ten hypermethylated (methylation rate > 60%) CpG dyads containing no more than four interleaving non-hypermethylated (methylation rate ≤ 60%) or undetermined CpGs.
To elucidate whether HypoMDs and HyperMDs contains distinct sequence features, genomic sequences of all called HypoMDs (N = 18435) and HyperMDs (N = 231516) were subjected to supervised classification using kmer-SVM (support vector machine with string-, i.e. nucleotide sequences-, based spectrum kernel) [43]. The default, recommended parameters and k = 6 (i.e. 6-mer) were used. Proportionally higher weights were assigned to HypoMDs (weight = 231516 / 18435 = 12.56) than HyperMDs (weight = 1) to offset the imbalanced sample sizes. Classification performance was gauged by 10-fold cross-validation and the area under precision-recall curves. Since HypoMDs have a higher average CpG density than HyperMDs (S1 Fig: panel C), CpG density might act as a confounding factor that outweighs and conceals non-CpG-containing sequence features. The impact of CpG density was hence controlled for by masking all CpG dinucleotides (i.e. from ‘CG’ to ‘NN’) and the SVM model was retraining using the same parameters as listed above.
Genomic regions that show contrasting methylation state between the HdrR and HNI strains were identified as described by Uno et al. [19]. HNI-specific HyperMD and HypoMDs, along with their 1.5 to 2-kb upstream and downstream sequences, were randomly selected and cloned using primers pairs F2-01 through F2-03 (see S2 Table for the oligo sequences). Medaka’s beta-actin promoter and EGFP coding sequences was amplified using primer sets F2-04 and F2-05, respectively. Amplified fragments were stitched together and cloned into pBlueScript-SK using In-Fushion assembly mix (Clonetech, Japan). The vectors were pre-treated with 5 units of I-SceI meganuclease in 20 μL of 1× I-SceI digestion buffer (New England Biolabs, USA) at room temperature for 1 hour and injected into medaka (drR strain) embryos at 1-cell stage following standard procedures [30]. Embryos that displayed stable, ubiquitously strong GFP fluorescence were raised and crossed with wild-type drR fish. GFP-positive F1 were inter-crossed to produce F2 generation. GFP-positive F2 embryos at blastula stage, i.e. Stage 11 by Iwamatsu [44], were sampled for genomic DNA extraction (see Method 1 in S1 Text). The purified genomic DNA was then bisulfite-converted using the MethylEasy Xceed Rapid DNA Bisulphite Modification Kit (Genetic Signatures, Australia) following manufacturer’s recommended procedures, except that DNA denaturation was carried out at 42°C for 20 mins. The stably integrated Hypo/HyperMDs and their flanking regions were PCR-amplified using BSP primers designed in MethPrimer [80] (primer set F2-06 through F2-11) and ExTaq polymerase (Takara Bio, Japan) under reaction conditions listed in Method 4 in S1 Text. BSP products were TA-cloned using TOPO TA Cloning Kit, Dual Promoter (Thermo Fisher Scientific, USA) and Sanger-sequenced (outsourced to FASMAC Co, Japan). Quality check and methylation rate quantification were carried out in QUMA [81] ver. 1.1.3 with default parameters.
To test whether nucleotide sequences can autonomously determine their own methylation state in vivo at genome-wide scale, CpG-rich genomic fragments were captured and injected into medaka zygotes for random reintegration into the genome, then fished out to check for their methylation state. The capturing method was akin to those described for reduced representation bisulfite sequencing (RRBS). In fact, procedures up to the size selection of adaptor-ligated genomic fragments closely followed those optimized for RRBS [46]. The adaptor-ligated fragments were then enriched and amplified by extension PCR, which also introduced (from 5’ to 3’, in this order) I-SceI target sites, bisulfite PCR (BSP) primer binding sites (i.e. for primer F3-01F and F3-01R), and the Dam methylation site (5’-GATC-3’) to the products’ termini. The pool of amplified fragments was then Dam-methylated by incubating with Dam methylase (New England Biolabs) to facilitate downstream counter-selection of unintegrated fragments. Dam-methylated fragments were split into two equal halves with one half used directly for injection after purification and the other half subjected to artificial methylation using CpG methytransferase M.SssI (New England Biolabs) prior to injection. Detailed procedures are available as supplementary information (Method 2 & 3 in S1 Text).
Immediately prior to injection, the fragments (final concentration: 10 ng/μL) were pre-treated I-SceI meganuclease as above. Medaka zygotes were injected with Dam-methylated or Dam+CpG-methylated fragments at 1-cell stage. Around 500 embryos were injected with each pools of fragments and were allowed to develop to the blastula stage at 28°C. The embryos were visually inspected under dissecting microscope with dead or malformed embryos discarded. Ultimately, 496 (86%) and 433 (92%) embryos injected with Dam-methylated and Dam+CpG-methylated fragments, respectively, developed normally to the blastula stage, and from which genomic DNA with fragments integrated was extracted (Method 1 in S1 Text). While most of the unintegrated fragments were presumably removed using our optimized DNA extraction method that includes size selection by PEG precipitation, carryover was further minimized by incubating the extracted DNA with 2 μL of FastDigest DpnI (Thermo Fisher Scientific, USA) in a 20 μL of 1X NEB Buffer 2 (New England Biolabs) for a total of 72 hours at 37°C in an incubator. This was followed by routine phenol-chloroform extraction and isopropanol precipitation. The precipitated DNA was finally re-dissolved in 20 μL of freshly dispensed Milli-Q water (Merck Millipore, USA).
Efficient removal of unintegrated fragments was indicated by the parallel use of uninjected, spike-in control. Approximately twice the amount of the injection cocktail was spiked into the lysate of uninjected blastula embryos, which was then processed as described above. Relative quantity of library with or without integration was gauged by real-time PCR (THUNDERBIRD SYBR qPCR Master Mix, TOYOBO, Japan; in Agilent Stratagene Mx3000P, USA) using the library-specific primers F3-01F and F3-01R. In parallel, input DNA was also quantitated using primers F3-04F and F3-04R. Amplification plots were imported into qpcR v1.4.0 [82], where the relative quantities were determined after sigmoidal modeling (all adjusted R2 = 1.00).
The purified genomic DNA was then bisulfite-converted as above. Integrated fragments were enriched via PCR using primers F3-01F and F3-01R. The BSP products were dA-tailed and ligated to Illumina TruSeq adapters, pooled, and sequenced using Illumina MiSeq system. Detailed library preparation procedures are described in Method 4 in S1 Text. Sequencing outputs were minimally trimmed, mapped to genome, and called for methylation rate as aforementioned, except bwa-mem’s “-U” switch was set to its default.
In order to relate the methylation state of the integrated fragments to possible binding or recognition by DNA-binding proteins (e.g., transcription factors), we identified DNase I hypersensitive sites (DHS) by remapping the publicly available DNase-seq dataset of drR medaka blastula embryos (accession number: SRX1032807 [83]) to the medaka genome assembly v2.2.4. Adaptor trimming and alignment was accomplished using BBmap v37.36 [76] with default parameters. Aligned reads were filtered for a minimum MAPQ of 20. MACS v2.1.1.20160309 [84] was subsequently used to called 112987 peaks (DHS) with the following switches: “-g 6.3e+8 --nomodel --shift -50 --extsize 100 -q 0.01”. Vast majority (> 96%) of the assayed fragments were originated completely from either inside or outside-, but not spanning across the boundaries-, of DHS (S3 Table).
An engineered transgenic line that carries an attP site inside a gene desert on chromosome 18 for PhiC31 integrase-mediated recombination was used for site-specific integration of the full-length, unmethylated HyperMDs (i.e. PCR-amplified, cloned, and without pretreating with M.SssI) and pre-methylated HypoMDs (i.e. PCR-amplified, cloned, and pretreated with M.SssI) with lengths of 300–400 bp.
PhiC31 integrase coding sequence was amplified from pPGK-PhiC31o-bpA (a gift from Philippe Soriano; Addgene plasmid #13795) and attached to SV40 nuclear localization sequence (NLS) using primer pair F4-01 and Phusion polymerase (Thermo Fisher Scientific), then blunt-end-cloned using Zero Blunt PCR Cloning Kit (Thermo Fisher Scientific). Cloning direction and proper coding sequence were checked via Sanger sequencing (by FASMAC Co). PhiC31 integrase mRNA was generated from the constructed template via in vitro transcription (Method 5 in S1 Text).
Six HyperMDs (see S1 Dataset) with flanking BSP primer binding sites (for F3-01F and F3-01R) and Dam-sites (downstream of the BSP primer sites) were directly synthesized by Thermo Fisher Scientific and Integrated DNA Technologies (USA) as double-stranded DNA and cloned into the targeting vector pEx_MCS-attBtagRFPt (a gift from Joachim Wittbrodt; Addgene plasmid #48876). Eleven HypoMDs were amplified from drR genomic DNA and extended to include BSP primer binding sites and Dam-sites on both ends using primer sets F4-02 through F4-12, then cloned into the targeting vector pEx_MCS-attBtagRFPt. HyperMD-containing targeting vectors were propagated in dam+ E. coli (DH5α) (Thermo Fisher Scientific) and pooled in approximately equimolar amount. HypoMD-containing targeting vectors were similarly processed, except that the pooled library was further artificially methylated with CpG methyltransferase M.SssI and purified as aforementioned (Method 3 in S1 Text). Individual plasmid libraries (final concentration: 10 ng/μL) was injected with PhiC31 integrase mRNA (100 ng/μL) into >200 embryos of PhiC31 transgenic strain [49] at 1-cell stage. Injected embryos were reared at 28°C to blastula stage, screened for normal development (> 85%), homogenized, and extracted for genomic DNA (Method 1 in S1 Text). The extracted DNA was digested with DpnI to degrade unintegrated vectors, re-purified, bisulfite-converted, subjected to PCR via ExTaq polymerase, TA-cloned, Sanger-sequenced, and quantified for methylation rate as aforementioned.
To ensure the injected but unintegrated vectors were efficiently removed, the above injection was also carried out without PhiC31 integrase mRNA. These injected embryos were processed in parallel with those injected with integrase mRNA up to DpnI digestion. The relative abundance of undigested libraries (both unintegrated and integrated) was quantified and normalized to amount of input genomic DNA using real-time PCR as described above (see also S2 Fig: panel B).
Homology directed repair was triggered by CRISPR-Cas9-induced double-strand breaks. spCas9 mRNA was produced from pMLM3613 (a gift from Keith Joung; Addgene plasmid #42251) via in vitro transcription (Method 5 in S1 Text). The HypoMDs, chr17:6415960–6416269 (Locus 1) and chr21:25260707–25262742 (Locus 2), were randomly chosen as targets for editing. sgRNAs targeting these regions were designed using CCTop [85]. The six top-ranked guide sequence designs (sets F5-01 and F5-02, for the two loci, respectively) were synthesized (Thermo Fisher Scientific) and in vitro transcribed (Method 5 in S1 Text). To construct the repair template, these genomic regions (with 6 mutations to the targeted spCas9 PAMs, i.e. from ‘NGG’ to ‘NGC’, in order to protect the template from being cleaved by spCas9) along with their up- and down-stream sequences (800 bp on both sides) as homology arms were synthesized (Integrated DNA Technologies), assembled, cloned into pCR-BluntII vector (Thermo Fisher Scientific) using NEBuilder HiFi assembly mix (New England Biolabs), and propagated by dam+ E. coli. The repair templates were artificially methylated in vitro using CpG methyltransferase M.SssI and purified as described above. For each of the target regions, sgRNA cocktail, spCas9 mRNA, and artificially Dam+CpG-methylated repair template were co-injected into medaka (drR strain) embryos at 1-cell stage at ultra-high concentrations (25 ng/μL each, 600 ng/μL and 10 ng/μL, respectively, i.e. 750 ng/μL of RNA and 10 ng/μL DNA in total) to maximize editing rate. Injected embryos were reared at 28°C for approx. 8 hours to blastula stage, screened for normal development (> 75%) and extracted for genomic DNA, which was DpnI-treated to degrade the repair template, re-purified, and bisulfite-converted as aforementioned. The BSP primer pairs (F5-03 for Locus 1; F5-04 for Locus 2) were designed using MethPrimer 2.0 and screened for the presence of native Dam-site(s) (5’-GATC-3’) within the target region. The amplification products were gel-purified and directly Sanger-sequenced from both ends. The methylation rate of each CpG was estimated from the sequencing chromatograms as: C ÷ (C + T) × 100%, where C and T are the called peak height in the ‘cytosine’ (i.e. methylated cytosines, after bisulfite PCR) and ‘thymine’ (i.e. unmethylated cytosines, which were converted to uracil by bisulfite treatment, then to thymine by PCR) channels, respectively. The signal intensities were extracted in R 3.3.3 [86] using the sangerseqR package (version 1.12.0) [87]. To estimate the editing rate, regions containing the sgRNA target sites was PCR-amplified from unconverted DNA using primer sets F5-05 through F5-09. Editing rate was gauged by the relative frequency of mutated sgRNA PAMs (5’-NGC-3’; on the edited alleles) versus the native PAMs (5’-NGG-3’; i.e. unedited alleles) from the Sanger sequencing trace using the same approach as described above. Editing efficiency was estimated to be 92.04% and 85.10% for Locus 1 and 2, respectively.
To collect edited embryos at later developmental stages (3 and 7 day-post-fertilization; dpf), the above cocktail was diluted 10-fold (in Milli-Q water; Merck Millipore) immediately prior to injection to reduce the toxicity (manifested after gastrulation) of ultra-high nuclei acid concentration at the expense of efficient editing. DNA extraction and subsequent processing were carried as above. Estimated editing efficiency for Locus 1 = 9.56% (at 3 dpf) and 7.81% (at 7 dpf); Locus 2 = 27.16% (at 3 dpf) and 18.69% (at 7 dpf). In order to enable comparison across sampling time-points with variable editing rates, the estimated methylation rates were normalized to the editing efficiency (i.e. “normalized methylation rate” = “methylation rate” ÷ “editing rate”). Raw values prior to normalization are available in S1 Dataset.
|
10.1371/journal.pbio.1001051 | Neurosteroid Dehydroepiandrosterone Interacts with Nerve Growth Factor (NGF) Receptors, Preventing Neuronal Apoptosis | The neurosteroid dehydroepiandrosterone (DHEA), produced by neurons and glia, affects multiple processes in the brain, including neuronal survival and neurogenesis during development and in aging. We provide evidence that DHEA interacts with pro-survival TrkA and pro-death p75NTR membrane receptors of neurotrophin nerve growth factor (NGF), acting as a neurotrophic factor: (1) the anti-apoptotic effects of DHEA were reversed by siRNA against TrkA or by a specific TrkA inhibitor; (2) [3H]-DHEA binding assays showed that it bound to membranes isolated from HEK293 cells transfected with the cDNAs of TrkA and p75NTR receptors (KD: 7.4±1.75 nM and 5.6±0.55 nM, respectively); (3) immobilized DHEA pulled down recombinant and naturally expressed TrkA and p75NTR receptors; (4) DHEA induced TrkA phosphorylation and NGF receptor-mediated signaling; Shc, Akt, and ERK1/2 kinases down-stream to TrkA receptors and TRAF6, RIP2, and RhoGDI interactors of p75NTR receptors; and (5) DHEA rescued from apoptosis TrkA receptor positive sensory neurons of dorsal root ganglia in NGF null embryos and compensated NGF in rescuing from apoptosis NGF receptor positive sympathetic neurons of embryonic superior cervical ganglia. Phylogenetic findings on the evolution of neurotrophins, their receptors, and CYP17, the enzyme responsible for DHEA biosynthesis, combined with our data support the hypothesis that DHEA served as a phylogenetically ancient neurotrophic factor.
| Dehydroepiandrosterone (DHEA) and its sulphate ester are the most abundant steroid hormones in humans, and DHEA was described as the first neurosteroid produced in the brain. DHEA is known to participate in multiple events in the brain, including neuronal survival and neurogenesis. However, to date no specific cellular receptor has been described for this important neurosteroid. In this study, we provide evidence that DHEA exerts its neurotrophic effects by directly interacting with the TrkA and p75NTR membrane receptors of nerve growth factor (NGF), and efficiently activates their downstream signaling pathways. This activation prevents the apoptotic loss of NGF receptor positive sensory and sympathetic neurons. The interaction of DHEA with NGF receptors may also offer a mechanistic explanation for the multiple actions of DHEA in other peripheral biological systems expressing NGF receptors, such as the immune, reproductive, and cardiovascular systems.
| Dehydroepiandrosterone (DHEA) is a steroid, produced in adrenals, in neurons and in glia [1]. The physiological role of brain DHEA appears to be local, i.e. paracrine, while that produced from adrenals, which represents the almost exclusive source of circulating DHEA, is systemic. The precipitous decline of both brain and circulating DHEA with advancing age has been associated with aging-related neurodegenerative diseases [1],[2]. It is experimentally supported that DHEA protects neurons against noxious conditions [3]–[6]. DHEA exerts its multiple pro-survival effects either directly modulating at micromolar concentrations γ-aminobutiric acid type A (GABAA), N-methyl-D-aspartate (NMDA), or sigma1 receptors, or following its conversion to estrogens and androgens. We have recently shown that nanomolar concentrations of DHEA protect sympathoadrenal PC12 cells from apoptosis [7]. PC12 cells do not express functional GABAA or NMDA receptors and cannot metabolize DHEA to estrogens and androgens [8]. The anti-apoptotic effect of DHEA in PC12 cells is mediated by high affinity (KD at nanomolar levels) specific membrane binding sites [9]. Activation of DHEA membrane binding sites results in an acute, transient, and sequential phosphorylation of the pro-survival MEK/ERK kinases, which, in turn, activate transcription factors CREB and NFκB, which afford the transcriptional control of anti-apoptotic Bcl-2 proteins. In parallel, activation of DHEA membrane binding sites induces the phosphorylation of PI3K/Akt kinases, leading to phosphorylation/deactivation of the pro-apoptotic Bad protein and protection of PC12 cells from apoptosis [10].
In fact, the anti-apoptotic pathways in sympathoadrenal cells initiated by DHEA at the membrane level strikingly resemble those sensitive to neurotrophin nerve growth factor (NGF). NGF promotes survival and rescues from apoptosis neural crest–derived sympathetic neurons (including their related sympathoadrenal cells) and sensory neurons involved in noniception. NGF binds with high affinity (KD: 0.01 nM) to transmembrane tyrosine kinase TrkA receptor and with lower affinity (KD: 1.0 nM) to p75NTR receptor, a membrane protein belonging to the TNF receptor superfamily [11]. In the presence of TrkA receptors, p75NTR participates in the formation of high affinity binding sites and enhances NGF responsiveness, leading to cell survival signals. In the absence of TrkA, p75NTR generates cell death signals. Indeed, docking of TrkA by NGF initiates receptor dimerization and phosphorylation of cytoplasmic tyrosine residues 490 and 785 on the receptor. Phosphotyrosine-490 interacts with Shc and other adaptor proteins resulting in activation of PI3K/Akt and MEK/ERK signaling kinase pathways [11]. These signals lead to the activation of prosurvival transcription factors CREB and NFκB, the subsequent production of anti-apoptotic Bcl-2 proteins, and prevention of apoptotic cell death of sympathetic neurons and sympathoadrenal cells, including PC12 cells [12].
Intrigued by the similarities in the prosurvival membrane signaling of DHEA and NGF, we set out to examine in the present study whether the anti-apoptotic effects of DHEA are mediated by NGF receptors. To address this issue we employed a multifaceted approach, designing an array of specific experiments: we used RNA interference (RNAi) to define the involvement of TrkA and p75NTR receptors in the anti-apoptotic action of DHEA; we assessed membrane binding of DHEA in HEK293 cells transfected with the TrkA and p75NTR plasmid cDNAs, using binding assays, confocal laser microscopy, and flow cytometry; to investigate the potential direct physical interaction of DHEA with NGF receptors, we tested the ability of immobilized DHEA to pull-down recombinant or naturally expressed TrkA and p75NTR receptors; finally, we examined the ability of DHEA to rescue from apoptosis NGF receptor sensitive dorsal root ganglia sensory neurons of NGF null mice and NGF deprived rat superior cervical ganglia sympathetic neurons in culture [13]. We provide evidence that DHEA directly binds to NGF receptors to protect neuronal cells against apoptosis, acting as a neurotrophic factor.
To test the involvement of NGF receptors in the anti-apoptotic effect of DHEA in serum deprived PC12 cells we have used a combination of three different sequences of siRNAs for TrkA and two different shRNAs for p75NTR transcripts [14]. The effectiveness of si/shRNAs was shown by the remarkable decrease of TrkA and p75NTR protein levels in PC12 cells, observed by immunoblotting analysis, using GAPDH as reference standard (Figure 1B). Scrambled siRNAs were ineffective in decreasing TrkA and p75NTR protein levels and did not significantly alter the effect of DHEA (unpublished data). FACS analysis of apoptotic cells (stained with Annexin V) has shown that DHEA and membrane impermeable DHEA-BSA conjugate at 100 nM diminished the number of apoptotic cells in serum deprived PC12 cell cultures from 53.5%±17.6% increase of apoptosis in serum free condition (control) to 6%±1.4% and 13%±5.2%, respectively (n = 8, p<0.01 versus control) (Figure 1A). Decreased TrkA expression in serum deprived PC12 cells with siRNAs resulted in the almost complete reversal of the anti-apoptotic effects of NGF and DHEA or DHEA-BSA membrane-impermeable conjugate (Figure 1A). Co-transfection of serum deprived PC12 cells with the si/shRNAs for TrkA and p75NTR receptors did not modify the effect of the TrkA deletion alone. Furthermore, transfection of serum deprived PC12 cells with shRNAs against p75NTR receptor alone did not significantly alter the anti-apoptotic effects of NGF and DHEA, suggesting that their anti-apoptotic effects are primarily afforded by TrkA receptors.
Transfection of serum deprived PC12 cells with the siRNAs against the TrkA transcript fully annulled the ability of DHEA to maintain elevated levels of anti-apoptotic Bcl-2 protein (Figure 1B). Again, transfection with the shRNA against p75NTR receptor alone did not significantly affect Bcl-2 induction by DHEA, further supporting the hypothesis that TrkA is the main mediator of the anti-apoptotic effect of DHEA in this system.
It appears that the ratio of TrkA and p75NTR receptors determines the effect of DHEA or NGF on cell apoptosis and survival. Indeed, both NGF and DHEA induced apoptosis of nnr5 cells, a clone of PC12 cell line, known to express only pro-death p75NTR receptors (Figure 1C), confirming the pro-apoptotic function of this receptor. Blockade of p75NTR expression by shRNA almost completely reversed the pro-apoptotic effect of both agents. The anti-apoptotic effect of NGF and DHEA was remarkably restored after transfection of nnr5 cells with the TrkA cDNA, the efficacy of reversal being proportionally dependent on the amount of transfected TrkA cDNA (Figure 1C).
DHEA was also controlling the response of NGF receptor-positive cells, by regulating TrkA and p75NTR receptor levels, mimicking NGF. Serum deprived PC12 cells were exposed to 100 nM of DHEA or 100 ng/ml of NGF for 12, 24, and 48 h; TrkA and p75NTR protein levels were measured in cell lysates with immunoblotting, using specific antibodies against TrkA and p75NTR proteins, and were normalized against GAPDH. Both NGF and DHEA significantly increased pro-survival TrkA receptor levels in the time frame studied, i.e. from 12 to 48 h (n = 5, p<0.01) (Figure S1). Furthermore, DHEA and NGF significantly decreased p75NTR receptor levels between 24 and 48 h of exposure (n = 5, p<0.01).
We have also tested the anti-apoptotic effects of DHEA in neural crest deriving superior cervical ganglia (SCG), a classical NGF/TrkA sensitive mammalian neuronal tissue, containing primarily one class of neurons, principal sympathetic neurons. Indeed, NGF and TrkA receptors are absolutely required for SCG sympathetic neuron survival during late embryogenesis and early postnatal development [13],[15]. TrkC receptors are barely detectable after E15.5, and no significant TrkB receptors are present in the SCG at any developmental stage [16]. Dispersed rat sympathetic SCG neurons at P1 were isolated and cultured for at least 7 d in the presence of 100 ng/ml NGF before the experiments are performed, in order to obtain an enriched, quasi-homogenous (95%) neuronal cell culture. Enriched SCGs were then incubated in the presence of 100 ng/ml NGF or in the same medium as above but lacking NGF and containing a polyclonal rabbit anti-NGF-neutralizing antiserum in the absence or the presence of 100 nM DHEA. Withdrawal of NGF strongly increased the number of apoptotic sympathetic neurons stained with Annexin V, while DHEA effectively compensated for NGF by decreasing the levels of apoptotic neurons. This effect was blocked by a specific TrkA inhibitor, thus suggesting the involvement of TrkA receptors as the main mediator of the anti-apoptotic action of DHEA (Figure 2).
We have previously shown the presence of specific DHEA binding sites to membranes isolated from PC12, primary human sympathoadrenal, and primary rat hippocampal cells, with KD at the nanomolar level [9]. The presence of DHEA-specific membrane binding sites on PC12 cells has been confirmed by flow cytometry and confocal laser microscopy of cells stained with the membrane impermeable DHEA-BSA-FITC conjugate. In contrast to estrogens, glucocorticoids and androgens displaced [3H]DHEA from its membrane binding sites, acting as pure antagonists by blocking the anti-apoptotic effect of DHEA in serum deprived PC12 cells [9]. In the present study, we repeated this series of experiments using membranes isolated from HEK293 cells transfected with the plasmid cDNAs of TrkA or p75NTR receptors.
HEK293 cells (not expressing TrkA or p75NTR) were transfected with an empty vector (control) or a specific TrkA or p75NTR vector; transfection efficiency was assessed by Western blot (Figure 3A and C,F inserts), confocal laser microscopy, and flow cytometry (Figure 3B,D). Saturation binding experiments have shown that [3H]-DHEA bound to membranes isolated from HEK293 cells, transfected with the cDNAs of TrkA or p75NTR receptors. Membranes isolated from HEK293 cells transfected with the empty vector showed no specific binding. The KD values calculated after Scatchard analysis of saturation curves were, for incubation of membranes at 25°C for 30 min, 7.4±1.75 nM and 5.6±0.55 nM for TrkA or p75NTR, respectively (n = 3) (Figure 3A,C), and for overnight incubation of membranes at 4°C, 7.8±3.1 nM and 5.9±1.7 nM for TrkA or p75NTR, respectively (n = 3) (Figure S2). DHEA was previously shown to bind with low affinity (KD: 2 µM) to androgen receptors (AR) [17]. We have thus tested the hypothesis that specific binding of DHEA to membranes of HEK293 cells transfected with the TrkA and p75NTR cDNAs might be due to the presence of AR receptors, induced by the transfection with NGF receptors. However, RT-PCR analysis showed no detectable levels of androgen receptors mRNA in RNA preparations isolated from naïve and TrkA or p75NTR transfected HEK23 cells (Figure S3).
Transfection of PC12 cells, endogenously expressing NGF receptors, with shRNAs against both TrkA and p75NTR receptors resulted in a complete loss of [3H]-DHEA specific membrane binding (Figure 3E,F). To rule out the possibility that the loss of specific binding might be due to the transfection process, we tested binding of [3H]-DHEA to membranes isolated from PC12 cells transfected with siRNA against GAPDH. Saturation binding and Scatchard analysis have shown that [3H]-DHEA bound to membranes from PC12-siRNA GAPDH cells with a KD = 1.068±0.43 nM (Figure 3E).
The selectivity of DHEA binding to HEK293TrkA and HEK293p75NTR cell membranes was examined by performing heterologous [3H]-DHEA displacement experiments using a number of non-labeled steroids or NGF. Binding of [3H]-DHEA to membranes isolated from both HEK293TrkA and HEK293p75NTR cells was effectively displaced by NGF (IC50: 0.8±0.2 and 1.19±0.45 nM, respectively) (Figure S4). NGF was also effective in displacing [3H]-DHEA binding on membranes isolated from PC12 cells (IC50: 0.92±0.32 nM, unpublished data). Estradiol failed to displace [3H]-DHEA from its binding to membranes from HEK293TrkA and HEK293p75NTR cells at concentrations ranging from 0.1 to 1000 nM. In contrast, displacement of [3H]-DHEA binding to membranes from both HEK293TrkA and HEK293p75NTR cells was shown by sulfated ester of DHEA, DHEAS (IC50: 6.1±1.1 and 8.1±1.2 nM, respectively, n = 3), and testosterone (Testo) (IC50: 5.3±2.1 and 7.4±3.2 nM, respectively). Glucocorticoid dexamethasone (DEX) effectively competed [3H]-DHEA binding to membranes from HEK293TrkA (IC50: 9.5±4.6 nM) but was ineffective in displacing DHEA binding to membranes from HEK293p75NTR cells. Homologous [125I]-NGF displacement experiments with unlabeled NGF confirmed the presence of specific NGF binding on membranes from both HEK293TrkA and HEK293p75NTR cells with IC50 0.3±0.09 and 1.7±0.38 nM, respectively. It is of note that in contrast to unlabeled NGF, DHEA was unable to displace binding of [125I]-NGF to membranes isolated from HEK293TrkA and HEK293p75NTR transfectants (unpublished data).
Incubation of PC12 cells with the membrane impermeable, fluorescent DHEA-BSA-fluorescein conjugate results in a specific spot-like membrane fluorescent staining [9]. In the present study, we have tested the ability of DHEA-BSA-FITC conjugate to stain HEK293TrkA and HEK293p75NTR transfectants. Fluorescence microscopy analysis revealed that DHEA-BSA-FITC clearly stained the membranes of HEK293TrkA and HEK293p75NTR cells (Figure 3B,D). No such staining was found in non-transfected HEK293 cells (unpublished data) or in HEK293 cells transfected with the vectors empty of TrkA and p75NTR cDNAs (Figure 3B,D). Furthermore, BSA-FITC conjugate was ineffective in staining both transfectants (unpublished data). We have further confirmed the presence of membrane DHEA-BSA-FITC staining of HEK293TrkA and HEK293p75NTR cells with flow cytometry (FACS) analysis (Figure 3B,D). Specific staining was noted in both transfectants. No such staining was seen in non-transfected HEK293 cells (unpublished data) or in HEK293 cells transfected with the empty vectors (Figure 3B,D). In both fluorescence microscopy and FACS experiments membrane staining of TrkA or p75NTR proteins in HEK293TrkA and HEK293p75NTR cells was also shown using specific antibodies for each protein (Figure 3B,D).
Our binding assays with radiolabeled DHEA suggest that DHEA physically interacts with NGF receptors. To test this hypothesis we covalently linked DHEA-7-O-(carboxymethyl) oxime (DHEA-7-CMO) to polyethylene glycol amino resin (NovaPEG amino resin) and tested the ability of immobilized DHEA to pull down TrkA and p75NTR proteins. Precipitation experiments and Western blot analysis of precipitates with specific antibodies against TrkA and p75NTR proteins (Figure 4A) showed that immobilized DHEA effectively precipitated recombinant TrkA and p75NTR proteins, while pre-incubation of the recombinant proteins with DHEA or NGF in excess abolished the ability of DHEA-PEG to pull down both receptors. Similar results were obtained when cell extracts isolated from HEK293 cells transfected with TrkA and p75NTR cDNAs, PC12 cells, and whole rat brain were treated with immobilized DHEA (Figure 4B, panels marked with A). No precipitation of TrkA and p75NTR proteins was shown with polymer-supported DHEA-7-CMO incubated with cell extracts from untransfected HEK293 cells or HEK293 cells transfected with the empty vectors. A control experiment was performed with NovaPeg amino resin (no DHEA-7-CMO present), which was found ineffective in precipitating TrkA and p75NTR proteins (Figure 4). The presence of TrkA and p75NTR receptors in HEK293TrkA and HEK293p75NTR transfectants and in PC12 and fresh rat brain was confirmed with Western blot analysis using specific antibodies against TrkA and p75NTR proteins and GAPDH as reference standard (Figure 4, panels marked with B).
Previous findings have shown that NGF controls the responsiveness of sensitive cells through induction of TrkA phosphorylation and regulation of the levels of each one's receptors [18]. We compared the ability of NGF and DHEA to induce phosphorylation of TrkA in HEK293 cells transfected with the cDNAs of TrkA receptors. HEK293TrkA transfectants were exposed for 10 and 20 min to 100 nM of DHEA or 100 ng/ml of NGF, and cell lysates were immunoprecipitated with anti-tyrosine antibodies and analyzed by Western blotting, using specific antibodies against TrkA receptors. Both NGF and DHEA strongly increased phosphorylation of TrkA as early as 10 min, an effect which was also maintained at 20 min (Figure 5A). We also tested the effects of DHEA and NGF in PC12 cells, endogenously expressing TrkA receptors. Naive or siRNATrkA transfected PC12 cells were incubated for 10 min with DHEA or NGF, and cell lysates were analyzed with Western blotting, using specific antibodies against Tyr490-phosphorylated TrkA and total TrkA. Both NGF and DHEA strongly induced the phosphorylation of TrkA in naive PC12 cells, effects which were diminished in siRNATrkA transfected PC12 cells (Figure 5A). The stimulatory effect of DHEA on TrkA phosphorylation might be due to an increase of NGF production. To test this hypothesis, we measured with ELISA the levels of NGF in culture media of HEK293 and PC12 cells exposed for 5 to 30 min to 100 nM of DHEA. NGF levels in culture media of control and DHEA-treated HEK293 and PC12 cells were undetectable, indicating that DHEA-induced TrkA phosphorylation was independent of NGF production. DHEAS mimicked the effect of DHEA and rapidly induced (within 10 min) the phosphorylation of TrkA receptors in HEK293 transfected with the TrkA cDNA expression vector (Figure S5). On the other hand, testosterone, while capable of displacing DHEA binding to TrkA receptors, was unable to increase phosphorylation of TrkA in the same system (Figure S5).
We compared the ability of NGF and DHEA to induce phosphorylation of TrkA-sensitive Shc, ERK1/2, and Akt kinases. Serum deprived naive or siRNATrkA transfected PC12 cells were incubated for 10 min with 100 nM DHEA or 100 ng/ml NGF and cell lysates were analyzed with Western blotting, using specific antibodies against the phosphorylated and total forms of kinases mentioned above. Both DHEA and NGF strongly increased phosphorylation of Shc, ERK1/2, and Akt kinases in naive PC12 cells, effects which were almost absent in siRNATrkA transfected PC12 cells, suggesting that both DHEA and NGF induce Shc, ERK1/2, and Akt phosphorylation via TrkA receptors (Figure 5A).
The effectiveness of DHEA to promote the interaction of p75NTR receptors with its effector proteins TRAF6, RIP2, and RhoGDI was also assessed. It is well established that NGF induces the association of p75NTR receptors with TNF receptor-associated factor 6 (TRAF6), thus facilitating nuclear translocation of transcription factor NFκB [19]. Furthermore, p75NTR receptors associate with receptor-interacting protein 2 (RIP2) in a NGF-dependent manner [20]. RIP2 binds to the death domain of p75NTR via its caspase recruitment domain (CARD), conferring nuclear translocation of NFκB. Finally, naive p75NTR interacts with RhoGDP dissociation inhibitor (RhoGDI), activating small GTPase RhoA [21]. In that case, NGF binding abolishes the interaction of p75NTR receptors with RhoGDI, thus inactivating RhoA. We co-transfected HEK293 cells with the plasmid cDNAs of p75NTR and of each one of the effectors TRAF6, RIP2, or RhoGDI, tagged with the flag (TRAF6) or myc (RIP2, RhoGDI) epitopes. Transfectants were exposed to 100 nM DHEA or 100 ng/ml NGF, and lysates were immunoprecipitated with antibodies against flag or myc, followed by immunoblotting with p75NTR specific antibodies. Both DHEA and NGF efficiently induced the association of p75NTR with effectors TRAF6 and RIP2, while facilitating the dissociation of RhoGDI from p75NTR receptors (Figure 5B).
NGF null mice have fewer sensory neurons in dorsal root ganglia (DRG) due to their apoptotic loss [13]. Heterozygous mice for the NGF deletion were interbred to obtain mice homozygous for the NGF gene disruption. The mothers were treated daily with an intraperitoneal injection of DHEA (2 mg) or vehicle (4.5% ethanol in 0.9% saline). Embryos were collected at E14 day of pregnancy and sections were stained for Caspase 3 and Fluoro jade C, markers of apoptotic and degenerative neurons, respectively. ngf−/− embryos at E14 showed a dramatic increase in the number of Fluoro Jade C and Caspase 3 positive neurons in the DRG compared to the ngf+/− embryos (Figure 6A,B). DHEA treatment significantly reduced Fluoro Jade C and Caspase 3 positive neurons in the DRG to levels of ngf+/− embryos. Furthermore, TrkA and TUNEL double staining of DRGs has shown that in ngf+/− embryos, numbers of TUNEL-positive apoptotic neurons were minimal, while TrkA positive staining was present in a large number of neuronal cell bodies of the DRG and their collaterals were extended within the marginal zone to the most dorsomedial region of the spinal cord. On the contrary, in DRG of ngf−/− embryos levels of TUNEL-positive apoptotic neurons were dramatically increased, while TrkA neuronal staining was considerably decreased and DRG collaterals of the dorsal funiculus were restricted in the dorsal root entry zone (Figure 6C). DHEA treatment resulted in a significant increase of TrkA positive staining and the extension of TrkA staining within the marginal zone to the most dorsomedial region of the spinal cord similarly to the ngf+/− embryos (Figure 6D), while staining of TUNEL-positive apoptotic neurons was decreased to levels shown in ngf+/− embryos.
DHEA exerts multiple actions in the central and peripheral nervous system; however, no specific receptor has been reported to date for this neurosteroid. Most of its actions in the nervous tissue were shown to be mediated via modulation, at micromolar concentrations, of membrane neurotransmitter receptors, such as NMDA, GABAA, and sigma1 receptors. DHEA may also influence brain function by direct binding, also at micromolar concentrations, to dendritic brain microtubule-associated protein MAP2C [22]. In the present study we provide evidence that DHEA binds to NGF receptors. This is the first report showing a direct binding of a steroid to neurotrophin receptors. Saturation experiments and Scatchard analysis of [3H]-DHEA binding to membranes isolated from HEK293 cells transfected with the cDNAs of TrkA and p75NTR receptors showed that DHEA binds to both membranes (7.4±1.75 nM and 5.6±0.55 nM for TrkA or p75NTR, respectively). Non-radioactive NGF effectively displaced [3H]-DHEA binding to both membrane preparations, with IC50: 0.8±0.2 and 1.19±0.45 nM, respectively. Furthermore, pull-down experiments using DHEA covalently immobilized on NovaPEG amino resin suggest that DHEA binds directly to TrkA and p75NTR proteins. Indeed, polymer-supported DHEA-7-CMO effectively pulled down recombinant TrkA and p75NTR proteins and precipitated both proteins from extracts prepared from cells expressing both receptors (HEK293TrkA, HEK293p75NTR, and PC12 cells and freshly isolated rat brain). Interestingly, DHEA was unable to effectively displace binding of [125I]-NGF on membranes isolated from HEK293TrkA and HEK293p75NTR transfectants. It is possible that dissociation of binding of peptidic NGF from its receptors lasts longer due to the multiple sites of interaction within the binding cleft of this large peptidic molecule compared to smaller in volume steroid. Another explanation might be that NGF and DHEA bind to different domains of NGF receptors, the NGF domain being non-recognizable by DHEA. It is of note that antidepressant amitryptiline cannot chase NGF from TrkA receptors because it binds to a different domain on TrkA protein compared to NGF. Indeed, other small molecules, like antidepressant amitriptyline and gamboge's natural extract gambogic amide, bind in the extracellular and the cytoplasmic juxtamembrane domains of TrkA receptor, although with much lower affinity compared to DHEA (Kd 3 µM and 75 nM, respectively) [23],[24]. The domains of TrkA and p75NTR proteins involved in DHEA binding were not defined in the present study. Mutagenesis assays combined with NMR spectroscopy are planned to map the domains of both receptors related to DHEA binding.
Our findings suggest that binding of DHEA to NGF receptors is functional, mediating its anti-apoptotic effects. Indeed, blocking of TrkA expression by RNAi almost completely reversed the ability of DHEA to protect PC12 cells from serum deprivation-induced apoptosis and to maintain elevated levels of the anti-apoptotic Bcl-2 protein. Additionally, in dispersed primary sympathetic neurons in culture, DHEA effectively compensated NGF deprivation by decreasing the levels of apoptotic neurons, an effect which was reversed by a specific TrkA inhibitor, further supporting the involvement of TrkA receptors in the anti-apoptotic action of DHEA. Finally, DHEA effectively rescued from apoptosis TrkA-positive dorsal root ganglia sensory neurons of NGF null mouse embryos.
It appears that the decision between survival and death among DHEA-responsive cells is determined by the ratio of TrkA and p75NTR receptors. In fact, DHEA and NGF induced apoptosis of nnr5 cells, a clone of PC12 cells expressing only pro-death p75NTR receptors. The pro-death effects of both agents were completely blocked by p75NTR shRNA and were remarkably restored after transfection of nnr5 cells with the TrkA cDNA. It is of note that during brain development the ratio of TrkA to p75NTR varies tempospatially [25]. Thus, the ability of DHEA to act in a positive or negative manner on neuronal cell survival may depend upon the levels of the two receptors during different stages of neuronal development.
Binding of DHEA on both TrkA and p75NTR receptors was effectively competed by sulfated DHEA, DHEAS (IC50: 6.1±1.1 and 8.1±1.2 nM, respectively), suggesting that DHEAS may also bind to NGF receptors. Testosterone displaced DHEA binding to TrkA and p75NTR (IC50: 5.3±2.1 and 7.4±3.2 nM, respectively), while synthetic glucocorticoid dexamethasone displaced DHEA binding only to pro-survival TrkA receptors (IC50: 9.5±4.6 nM). In a previous study we had shown that both steroids effectively displaced DHEA from its specific membrane binding sites of sympathoadrenal cells, acting as DHEA antagonists by blocking its anti-apoptotic effect and the induction of anti-apoptotic Bcl-2 proteins [9]. Our findings suggest that testosterone and glucocorticoids may act as neurotoxic factors by antagonizing endogenous DHEA and NGF for their binding to NGF receptors, explaining previously published data. Indeed, testosterone was shown to increase NMDA and GABAA-mediated neurotoxicity [26],[27]. Our findings suggest that testosterone may act as a neurotoxic factor by also antagonizing the neuroprotective effects of endogenous DHEA. Furthermore, glucocorticoids show a bimodal effect on hippocampal neurons causing acutely an increase in performance of spatial memory tasks, while chronic exposure has been associated with decreased cognitive performance and neuronal atrophy [28]. Acute administration of glucocorticoids results in a glucocorticoid receptor-mediated phosphorylation and activation of hippocampal TrkB receptors, exerting trophic effects on dentate gyrus hippocampal neurons [29], via an increase in the sensitivity of hippocampal cells to neurotrophin BDNF, the endogenous TrkB ligand known to promote memory and learning [30]. However, overexposure to glucocorticoids during prolonged periods of stress is detrimental to central nervous system neurons, especially in aged animals, affecting mainly the hippocampus. It is possible that part of neurotoxic effects of glucocorticoids may be due to their antagonistic effect on the neuroprotective effect of endogenous DHEA and NGF, via TrkA receptor antagonism. The decline of brain DHEA and NGF levels during aging and in Alzheimer's disease [28] might exacerbate this phenomenon, rendering neurons more vulnerable to glucocorticoid toxicity. Indeed, glucocorticoid neurotoxicity becomes more pronounced in aged subjects since cortisol levels in the cerebrospinal fluid increase in the course of normal aging, as well as in relatively early stages of Alzheimer's disease [28].
A number of neurodegenerative conditions are associated with lower production or action of both DHEA and NGF [31],[32]. Animal studies suggest that NGF may reverse, or slow down the progression of Alzheimer's related cholinergic basal forebrain atrophy [32]. Furthermore, the neurotrophic effects of NGF in experimental animal models of neurodegenerative conditions, like MPTP (Parkinson's disease), experimental allergic encephalomyelitis (multiple sclerosis), or ischemic retina degeneration mice [33]–[35] support its potential as a promising neuroprotective agent. However, the use of NGF in the treatment of these conditions is limited, because of its poor brain blood barrier permeability. It is of interest that DHEA also exerts neuroprotective properties in some of these animal models [7],[36]. These findings suggest that synthetic DHEA analogs, deprived of endocrine effects, may represent a new class of brain blood barrier permeable NGF receptor agonists with neuroprotective properties. We have recently reported the synthesis of 17-spiro-analogs of DHEA, with strong anti-apoptotic and neuroprotective properties, deprived of endocrine effects [37], which are now being tested for their ability to bind and activate NGF receptors.
We have previously defined the pro-survival signaling pathways that are initiated by DHEA at the membrane level [3]. These pathways include MEK1/2/ERK1/2 and PI3K/Akt pro-survival kinases. We now provide experimental evidence that DHEA activates these kinases via TrkA receptors. Down-regulation of TrkA receptors using siRNAs resulted in an almost complete reversal of the ability of DHEA to increase the phosphorylation of kinases Shc, Akt, and ERK1/2. In addition to TrkA receptors, binding of DHEA to the low affinity NGF receptor was also functional, affording the activation of p75NTR receptors. Unlike TrkA receptors, p75NTR lacks any enzymatic activity. Signal transduction by p75NTR proceeds via ligand-dependent recruitment and release of cytoplasmic effectors to and from the receptor. Indeed, DHEA like NGF facilitated the recruitment of two major cytoplasmic interactors of p75NTR, TRAF6 and RIP2 proteins. Additionally, DHEA-mediated activation of p75NTR led to the dissociation of bound RhoGDI, a protein belonging to small GTPases and interacting with RhoA [21]. A schematic representation of our findings is shown in Figure 7.
Previous findings suggest that DHEA protects PC12 cells against apoptosis via pertussis toxin (PTX) sensitive, G protein-associated specific plasma membrane-binding sites [9]. Indeed, PTX was shown to partially reverse the anti-apoptotic effects of DHEA and its membrane impermeable DHEA-BSA conjugate, as well as their effects on prosurvival kinases PI3K/Akt, the activation of transcription factor NFkappaB, and the phosphorylation and inactivation of apoptotic protein Bad [10]. Interestingly, the prosurvival effects of NGF in sympathetic neurons and PC12 cells are also partially reversed by PTX [38]. Furthermore, the NGF-dependent activation of Akt is partially attenuated by PTX, indicating the participation of G(i/o) proteins. In the same study, NGF-induced phosphorylation of Bad and transcriptional activity of NFkappaB were also shown to be sensitive to PTX [38]. It appears that other NGF-driven pathways are sensitive to PTX too. For instance, in PC12 cells and primary cortical neurons the NGF-induced phosphorylation of tuberin (a critical translation regulator holding a central role in NGF-promoted neuronal survival) is partially blocked by PTX, suggesting the participation of G(i/o) proteins [39]. Finally, NGF-dependent activation of the p42/p44 mitogen-activated protein kinase (p42/p44 MAPK) pathway in PC12 cells was effectively blocked by PTX [40]. However in HEK293 cells transfected with TrkA receptors, PTX was unable to affect the induction of TrkA phosphorylation by NGF or DHEA (Figure S5). These findings considered together suggest that TrkA receptors may use down-stream G protein-coupled receptor pathways, after binding and activation by NGF or DHEA, to control neuronal cell survival.
It is worth noticing that the interaction of DHEA with the NGF system was first suggested 15 years ago by Compagnone et al., showing co-localized staining of CYP17, the rate limiting enzyme of DHEA biosynthesis, and NGF receptors in mouse embryonic DRGs [41]. About one-fifth of CYP17-immunopositive DRG neurons in the mouse were found to be also TrkA-immunopositive. Among the TrkA-expressing cells, about one-third also express CYP17, while p75NTR-expressing neurons represent only 13% of the cells in the DRG. Thus, about one-fifth of CYP17-immunopositive neurons may be able to respond to both DHEA and NGF stimulation, an observation compatible with our data, presented in Figure 6C. A recent report further supports the interaction of DHEA with NGF receptors. Indeed, DHEA was shown to act as a keratinocyte-deriving neurotrophic signal, mimicking NGF in promoting axonal outgrowth of NGF non-producing but TrkA positive sensory neurons, an effect blocked by TrkA inhibitor K252a [42].
CYP17 is expressed in invertebrate cephalochordata Amphioxus [43]. Amphioxus is also expressing TrkA receptor homologous AmphiTrk, which effectively transduces signals mediated by NGF [44]. Phylogenetic analysis of neurotrophins revealed that they emerged with the appearance of vertebrates (530–550 million years ago), when complexity of neural tissue increased [45]. Invertebrate cephalochordata like Amphioxus are positioned on the phylogenetic boundary with vertebrates (600 million years ago). It is thus tempting to hypothesize that DHEA contributed as one of the “prehistoric” neurotrophic factors in an ancestral, simpler structurally invertebrate nervous system [46]; then, when a strict tempospatial regulation of evolving nervous system of vertebrates was needed, peptidic neurotrophins emerged to afford rigorous and cell specific neurodevelopmental processes.
In conclusion, our findings suggest that DHEA and NGF cross-talk via their binding to NGF receptors to afford brain shaping and maintenance during development. During aging, the decline of both factors may leave the brain unprotected against neurotoxic challenges. This may also be the case in neurodegenerative conditions associated with lower production or action of both factors. DHEA analogs may represent lead molecules for designing non-endocrine, neuroprotective, and neurogenic micromolecular NGF receptor agonists.
PC12 cells were transfected with specific si/shRNAs for blocking the expression of TrkA and/or p75NTR receptors. More specifically, three siRNAs and two shRNAs for TrkA and p75NTR, respectively, were obtained. The sequences for TrkA siRNAs (Ambion) were: GCCUAACCAUCGUGAAGAG (siRNA ID 191894), GCAUCCAUCAUAAUAGCAA (siRNA ID 191895), and CCUGACGGAGCUCUAUGUG (siRNA ID 191893). Sequences for p75NTR (Qiagen) were: GACCUAUCUGAGCUGAAA (Cat. No. SI00251090) and GCGUGACUUUCAGGGAAA (CatNo SI00251083).
Rat TrkA was expressed from the pHA vector backbone and rat p75NTR was expressed from the pCDNA3 vector backbone (InVitrogen) using a full length coding sequence flanked by an N-terminal hemagglutinin (HA) epitope tag. Plasmids to express RIP2 [19] and RhoGDI [36] were myc-tagged, while TRAF6 [19] was FLAG-tagged, as previously described.
The origin of antibodies was as follows: Bcl-2 (Cat. No. C-2, sc-7382, Santa Cruz Biotechnology Inc.), phospho TrkA (Cat. No. 9141, Cell Signaling), TrkA (Cat. No. 2505, Cell Signaling, was used for Western Blotting and Cat. No. 06-574, Upstate, was used for immunostainings), p75NTR (Cat. No. MAB365R, Millipore), c-myc (Cat. No. 9E10, sc-40, Santa Cruz Biotechnology Inc.), phospho ERK1/2 (Cat. No. 9106, Cell Signaling), Erk1/2 (Cat. No. 9102, Cell Signaling), phospho-Shc (Tyr239/240) Antibody (Cat. No. 2434, Cell Signaling), Shc (Cat. No. 2432, Cell Signaling), phospho-Akt (Ser473) (Cat. No. 9271, Cell Signaling), Akt (Cat. No. 9272, Cell Signaling), anti-FLAG (M2) mouse monoclonal (Cat. No. F1804, Sigma), pTyr (Cat. No. sc-508, Santa Cruz Biotechnology Inc.), active Caspase-3 (Cat. No. ab13847, Abcam), Tyrosine Hydroxylase (Cat. No. ab6211, Abcam), anti-rabbit-R-phycoerythrin conjugated (Cat. No. P9537, Sigma), anti-mouse-fluorescein conjugated (Cat. No. AP124F, Millipore), anti-rabbit Alexa Fluor 488 (Cat. No. A21206), anti-rabbit Alexa Fluor 546 (Cat. No. A10040), and GAPDH (Cat. No. 2118, Cell Signaling).
PC12 cells were obtained from LGC Promochem (LGC Standards GmbH, Germany) and nnr5 cells from Dr. C.F. Ibáñez (Karolinska Institute). Both cell types were grown in RPMI 1640 containing 2 mM L-glutamine, 15 mM HEPES, 100 units/ml penicillin, 0.1 mg/ml streptomycin, and 10% horse serum, 5% fetal calf serum (both charcoal-stripped for removing endogenous steroids) at 5% CO2 and 37°C. HEK-293 cells were obtained from LGC Promochem. Cells were grown in DMEM medium containing 10% fetal bovine serum (charcoal-stripped for removing endogenous steroids), 100 units/ml penicillin, and 0.1 mg/ml streptomycin, at 5% CO2 and 3°C. HEK-293 and PC12 cells were transfected with Lipofectamine 2000 (InVitrogen) according to manufacturer's instructions. Transfected cells were typically used on the 2nd day after transfection.
PC12 cells were cultured in 12-well plates, and 24 h later they were transfected with the si/shRNAs for TrkA and/or p75NTR. Twenty-four hours later the medium was aspirated and replaced either with complete medium (serum supplemented) or serum free medium in the absence or the presence of DHEA or DHEA-BSA conjugate at 100 nM. Apoptosis was quantified 24 h later with annexin V-FITC and PI (BD Pharmingen) according to our protocol [8].
HEK293 cells were allowed to grow on gelatin-coated glass coverslips for 24 h in culture medium, and 24 h later they were transfected with the cDNAs for TrkA, and p75NTR receptors or the vector alone. Staining was performed 48 h after transfection. Culture medium was aspirated and transfectants were washed twice with PBS buffer. Primary antibodies against TrkA (rabbit, Upstate, No. 06-574, diluted 1∶100) or p75NTR (mouse monoclonal ab, MAB365R, Millipore, dilution 1∶500) were added for 30 min at 37°C. Secondary antibodies, anti-rabbit-R-phycoerythrin conjugated (Sigma, No. P9537), and anti-mouse-fluorescein conjugated (No. AP124F, Millipore) were added at 1∶100 dilution and transfectants were incubated for 30 min at 37°C; then they were washed three times with PBS and counterstained with Hoechst nuclear stain (Molecular Probes) for 5 min. Transfectants were also incubated with the DHEA-BSA-FITC or the BSA-FITC conjugates (10−6M) for 15 min at room temperature in the dark; then they were washed with serum free culture medium and incubated for another 15 min in serum free culture medium containing 4% BSA. Coverslips were mounted to slides with 90% glycerin and were observed with a confocal laser scanning microscope (Leica TCS-NT, Leica Microsystems GmbH, Heidelberg, Germany), mounted with a digital camera.
HEK293 cells were cultured in 12-well plates, and 24 h later they were transfected with the cDNAs for TrkA and/or p75NTR receptors, or the vector alone. Staining was performed 48 h later. Transfectants (5×105 cells) were pelleted and incubated with 20 µl of the primary antibodies against TrkA or p75NTR receptors for 30 min over ice. Afterwards, transfectants were washed three times with PBS and 20 µl of the secondary antibodies, and anti-rabbit-R-phycoerythrin conjugated and anti-mouse-fluorescein conjugated were added, as described above. For DHEA-BSA-FITC binding on cells, 20 µl (100 nM) were added on the pelleted cells for 10 min at RT, and then they were washed with serum free culture medium and incubated for another 15 min in serum free culture medium containing 4% BSA. Transfectants were washed twice with PBS, resuspended in 500 µl of PBS, and were analyzed in a Beckton-Dickinson FACSArray apparatus and the CELLQuest software (Beckton-Dickinson, Franklin Lakes, NJ).
NovaPEG amino resin (loading value 0.78 mmol/g) was purchased from Novabiochem. NMR spectra were recorded on a Varian 300 spectrometer operating at 300 MHz for 1H and 75.43 MHz for 13C or on a Varian 600 operating at 600 MHz for 1H. 1H NMR spectra are reported in units of δ relative to the internal standard of signals of the remaining protons of deuterated chloroform, at 7.24 ppm. 13C NMR shifts are expressed in units of δ relative to CDCl3 at 77.0 ppm. 13C NMR spectra were proton noise decoupled. IR spectra was recorded at Bruker Tensor 27. Absorption maxima are reported in wavenumbers (cm−1).
3β-Acetoxy-17,17-ethylenedioxyandrost-5-ene (0.74 g, 1.98 mmol) and N-hydroxy phthalimide (0.71 g, 2.2 mmol) were dissolved in acetone (39 mL) containing 1 mL of pyridine. The mixture was stirred vigorously at room temperature and sodium dichromate dihydrate (0.89 g, 3 mmol) was added. Additional portions of solid sodium dichromate dihydrate (0.89 g, 3 mmol) were added after 10 and 20 h stirring at room temperature. After reaction completion (48 h), the mixture was diluted with dichloromethane, filtered through a bed of celite, and the filtrate was washed with water, saturated sodium bicarbonate solution, and brine. The organic layer was dried over anhydrous sodium sulfate, the solvent evaporated in vacuo, and the residue purified by flash column chromatography using hexane/acetone/25% NH4OH (85∶15∶0.1 mL) as eluent to afford 3β-acetoxy-17,17-ethylenedioxyandrost-5-ene-7-one (0.6 g, yield: 78%). 1H NMR (CDCl3, 300 MHz) δ: 0.87 (s, 3H), 1.21 (s, 3H), 1.26–2.00 (m, 14H), 2.05 (s, 3H), 2.20–2.51 (m, 3H), 3.84–3.92 (m, 4H), 4.68–4.76 (m, 1H), 5.70 (d, J = 1.58 Hz, 1H).
To a solution of 3β-acetoxy-17,17-ethylenedioxyandrost-5-en-7-one (0.1 g, 0.26 mmol) in pyridine (1.9 mL) was added O-(carboxymethyl)hydroxylamine hemihydrochloride (0.11 g, 0.52 mmol) and the reaction mixture was stirred overnight under argon. After completion of the reaction, the solvent was evaporated and the residue was diluted with ethyl acetate. The organic layer was washed with water and brine, dried over anhydrous sodium sulfate, and the solvent was evaporated in vacuo to afford 3β-acetoxy-17,17-ethylenedioxyandrost-5-en-7-one7-(O-carboxymethy1) oxime as a white foam (0.12 g, yield: 100%). 1H NMR (CDCl3, 300 MHz) δ: 0.88 (s, 3H), 1.13 (s, 3H), 1.16–1.95 (m, 12H), 2.04 (s, 3H), 2.25–2.59 (m, 5H), 3.84–3.95 (m, 4H), 4.59 (d, J = 2.29 Hz, 2H), 4.62–4.73 (m, 1H), 6.51 (d, J = 1.47 Hz, 1H).
To a solution of 3β-acetoxy-17,17-ethylenedioxyandrost-5-en-7-one-7-(O-carboxymethy1) oxime (0.12 g, 0.26 mmol) in a mixture of acetone/water (5∶1, 6.3 mL) was added p-toluenesulfonic acid monohydrate (0.019 g, 0.10 mmol), and the reaction mixture was stirred until the starting material was consumed (48 h). The solvent was evaporated in vacuo and the residue was diluted with ethyl acetate. The organic layer was washed with water and brine, dried over anhydrous sodium sulfate, and the solvent was evaporated in vacuo to afford 3β-acetoxy-androst-5-en-7,17-dione 7-(O-carboxymethy1) oxime as a white foam (0.11 g, yield: 100%). 1H NMR (CDCl3, 600 MHz) δ: 0.90 (s, 3H), 1.15 (s, 3H), 1.20–1.95 (m, 12H), 2.05 (s, 3H), 2.09–2.68 (m, 5H), 4.63 (d, J = 4.18 Hz, 2H), 4.65–4.71 (m, 1H), 6.56 (d, J = 1.39 Hz, 1H).
To a solution of 3β-acetoxy-androst-5-en-7,17-dione 7-(O-carboxymethy1) oxime (0.11 g, 0.26 mmol) in methanol (3.9 mL) was added LiOH (1.5 mL, 1.5 mmol, 1N solution), and the reaction mixture was stirred until the starting material was consumed (4 h). The solvent was evaporated in vacuo and the residue was diluted with water. The solution was acidified with 10% hydrochloric acid and DHEA-7-CMO precipitated as a white solid, which was isolated by filtration (0.097 g, yield: 100%). 1H NMR (CDCl3/CD3OD, 600 MHz) δ: 0.90 (s, 3H), 1.14 (s, 3H), 1.20–2.75 (m, 17H), 3.49–3.54 (m, 1H), 4.54 (s, 2H), 6.54 (s, 1H).
3β-Hydroxy-17-oxoandrost-5-en-7-O-(carboxymethyl)oxime (DHEA-7-CMO) (192 mg, 0.511 mmol) in DMF (5 mL) was treated with HOBt (69 mg, 0.511 mmol) and DIC (0.08 mL, 0.511 mmol), and the resulting mixture was stirred at room temperature for 30 min. This solution was added to NovaPEG amino resin (130 mg, 0.102 mmol, 0.78 mmol/gr) (pre-swollen with DMF for 1 h) and the slurry was shaken at room temperature overnight. The mixture was filtered, the resin was sequentially washed with dichloromethane (3×), methanol (3×), and diethyl ether (3×), and was dried in vacuo overnight. Yield 175 mg (100%), loading value 0.61 mmol/gr. 13C NMR (gel phase, CDCl3) δ: 220.66, 170.15, 157.10, 154.15, 113.11, 72.57, 66.59, 49.92, 47.86, 42.15, 38.46, 37.08, 36.53, 35.49, 31.20, 30.71, 24.96, 20.15, 18.05, 13.95; IR: νmax/cm−1 2865 (s), 1735 (m), 1669 (w), 1653 (w), 1637 (w), 1456 (m), 1348 (w), 1289 (w), 1247 (w), 1093 (s), 946 (w).
HEK293 cells were transfected with the appropriate plasmids (TrkA, p75NTR, RIP2, TRAF-6, and RhoGDI) by using Lipofectamine 2000 (Invitrogen). Cells were harvested 48 h after transfection and suspended in lysis buffer (50 mM Tris-HCl, 0.15 M NaCl, 1% Triton-X100, pH 7.4) supplemented with protease inhibitors. Lysates were precleared for 1 h with Protein A-Sepharose beads (Amersham) and immunoprecipitated with the appropriate antibody (pTyr, Flag, or c-myc) overnight at 4°C. Protein A Sepharose beads were incubated with the lysates for 4 h at 4°C with gentle shaking. In the case of immobilized DHEA-7-CMO, HEK293 or PC12 cells lysates or purified receptors (both from R&D Systems, Recombinant Mouse NGF R/TNFRSF16/Fc Chimera, Cat. No.: 1157-NR and Recombinant Rat Trk A/Fc Chimera, Cat. No.: 1056-TK) were incubated overnight at 4°C with the NovaPEG amino resin alone or conjugated with DHEA. Beads were collected by centrifugation, washed four times with lysis buffer, and resuspended in SDS loading buffer. Proteins were separated by SDS/PAGE, followed by immunoblotting with specific antibodies.
PC12 or HEK293 cells lysates were electrophoresed through a 12% SDS-polyacrylamide gel, and then proteins were transferred to nitrocellulose membranes, which were processed according to standard Western blotting procedures, as previously described [8]. To detect protein levels, membranes were incubated with the appropriate antibodies: Bcl-2 (dilution 1∶500), phospho TrkA (dilution 1∶500), total TrkA (dilution1∶500), p75NTR (dilution 1∶500), phospho Shc (dilution 1∶1000), total Shc (dilution 1∶1000), phospho Akt (dilution 1∶500), total Akt (dilution 1∶500), phospho ERK1/2 (dilution 1∶500), and total ERK1/2 (dilution 1∶500). Proteins were visualized using the ECL Western blotting kit (ECL Amersham Biosciences, UK), and blots were exposed to Kodak X-Omat AR films. A PC-based Image Analysis program was used to quantify the intensity of each band (Image Analysis, Inc., Ontario, Canada).
To normalize for protein content the blots were stripped and stained with GAPDH antibody (dilution 1∶1000); the concentration of each target protein was normalized versus GAPDH. Where phosphorylation of TrkA or kinases was measured, membranes were first probed for the phosphorylated form of the protein, then stripped, and probed for the total protein.
Superior cervical ganglia (SCG) were removed from newborn (P0–P1) rat pups and dissociated in 0.25% trypsin (Gibco, 15090) for 30 min at 37°C. After dissociation SCG neurons were re-suspended in culture medium (Gibco, Neurobasal Cat. No. 21103) containing 1% fetal bovine serum (FBS), 100 units/ml penicillin, 0.1 mg/ml streptomycin, 3 µg/ml araC antimitotic, and 100 ng/ml NGF (Millipore, 01-125). Cells were plated on collagen coated 24-well plates and cultured for 5 d prior to use. For NGF withdrawal experiments, cells were washed twice with Neurobasal containing 1% FBS and fresh culture medium lacking NGF and containing anti-NGF antibody at 1∶50 dilution (Millipore, AB1526). DHEA, TrkA-inhibitor (Calbiochem, 648450) and anti-p75NTR (mouse, MAB365R Millipore) were used at 100 nM, 100 nM, and 1∶50, respectively.
ngf+/− mice [13] were obtained from the Jackson Laboratory and maintained on C57BL/6 background. All procedures described below were approved by the Animal Care Committee of the University of Crete, School of Medicine. Animals were housed in cages maintained under a constant 12 h light–dark cycle at 21–23°C, with free access to food and tap water. Genotyping was performed on tail DNA using the following primers: NGFKOU2 (5′CCG TGA TAT TGC TGA AGA GC3′), NGFU6 (5′CAG AAC CGT ACA CAG ATA GC3′), and NGFD1 (5′TGT GTC TAT CCG GAT GAA CC3′). Genomic PCR reactions containing the 3 primers were incubated for 32 cycles at 95°C (30 s)/59°C (30 s)/72°C (1 min).
Mice heterozygous for the NGF null mutation were interbred to obtain mice homozygous for the NGF gene disruption and the first day of gestation determined by the discovery of a copulation plug. The mothers were treated daily with a subcutaneous injection of DHEA (2 mg/day) or vehicle (4.5% ethanol in 0.9% saline) starting from the third day after gestation. Animals were collected at E14. At the day of collection the mothers were deeply anesthetized with sodium pentobarbital (Dolethal 0.7 ml/kg i.p.) followed by transcardial perfusion with saline solution containing heparin for about 7 min, and with 4% PFA, 15% Picric Acid, 0.05% GA in phosphate buffer 0.1 M, for another 7 min. After the perfusion the embryos were collected and maintained in the same fixative overnight at 4°C. Embryos were then washed in 0.1 M phosphate buffer and cryoprotected by using 10% sucrose followed by 20% sucrose overnight at 4°C. Finally, embryos were frozen in OCT in iso-pentane over liquid nitrogen for 5 min and the frozen tissues were stored for later use at −80°C. The samples were sectioned (20 µm) and mounted onto Superfrost plus slides (Menzel-Glaser J1800AMNZ). Slides were left to air-dry overnight at room temperature (RT) and were then either used immediately or were fixed in cold acetone for 1 min and stored at −80°C for later use.
Stored or fresh slides were fixed for 15 min in cold acetone at 4°C and left to dry for 10 min at room temperature. They were then washed in PB 0.1 M, then in TBS, and incubated for 45 min with 10% horse serum in TBS-T 0.1%. The normal serum was drained off and the primary antibodies (anti-TrkA diluted 1∶400 and active Caspase-3 diluted 1∶50), diluted in TBS-T 0.1% with 1% horse serum, were added. Sections were incubated for 4 h at RT and overnight at 4°C; they were then washed in TBS-T 0.1% and the anti-rabbit secondary antibodies (Alexa Fluor 488 and Alexa Fluor 546, 1∶1000 in TBS-T 0.1%) were added for 6 h at RT. Sections were washed in TBS-T, TBS, and in PB 0.1 M and were coverslipped with Vectashield (Vector, H-1400) and visualized in a confocal microscope. TUNEL (Roche, Cat. No. 12156792910) and Fluoro-Jade C (Millipore, Cat. No. AG325) staining of apoptotic and degenerating neurons, respectively, was performed according to the manufacturer's instructions.
For the statistical evaluation of our data we have used analysis of variance, post hoc comparison of means, followed by the Fisher's least significance difference test. For data expressed as percent changes we have used the nonparametric Kruskal-Wallis test for several independent samples.
|
10.1371/journal.pgen.1004767 | Tetraspanin (TSP-17) Protects Dopaminergic Neurons against 6-OHDA-Induced Neurodegeneration in C. elegans | Parkinson's disease (PD), the second most prevalent neurodegenerative disease after Alzheimer's disease, is linked to the gradual loss of dopaminergic neurons in the substantia nigra. Disease loci causing hereditary forms of PD are known, but most cases are attributable to a combination of genetic and environmental risk factors. Increased incidence of PD is associated with rural living and pesticide exposure, and dopaminergic neurodegeneration can be triggered by neurotoxins such as 6-hydroxydopamine (6-OHDA). In C. elegans, this drug is taken up by the presynaptic dopamine reuptake transporter (DAT-1) and causes selective death of the eight dopaminergic neurons of the adult hermaphrodite. Using a forward genetic approach to find genes that protect against 6-OHDA-mediated neurodegeneration, we identified tsp-17, which encodes a member of the tetraspanin family of membrane proteins. We show that TSP-17 is expressed in dopaminergic neurons and provide genetic, pharmacological and biochemical evidence that it inhibits DAT-1, thus leading to increased 6-OHDA uptake in tsp-17 loss-of-function mutants. TSP-17 also protects against toxicity conferred by excessive intracellular dopamine. We provide genetic and biochemical evidence that TSP-17 acts partly via the DOP-2 dopamine receptor to negatively regulate DAT-1. tsp-17 mutants also have subtle behavioral phenotypes, some of which are conferred by aberrant dopamine signaling. Incubating mutant worms in liquid medium leads to swimming-induced paralysis. In the L1 larval stage, this phenotype is linked to lethality and cannot be rescued by a dop-3 null mutant. In contrast, mild paralysis occurring in the L4 larval stage is suppressed by dop-3, suggesting defects in dopaminergic signaling. In summary, we show that TSP-17 protects against neurodegeneration and has a role in modulating behaviors linked to dopamine signaling.
| Parkinson's disease (PD) is characterized by the progressive loss of dopaminergic neurons. While hereditary forms are known, most cases are attributable to a combination of genetic and environmental risk factors. In PD models, dopaminergic neurodegeneration can be triggered by neurotoxins such as 6-hydroxydopamine (6-OHDA). This drug, which is taken up by the presynaptic dopamine reuptake transporter (DAT-1), also causes the selective death of C. elegans dopaminergic neurons. We found that TSP-17, a member of the tetraspanin family of membrane proteins, protects dopaminergic neurons from 6-OHDA-induced degeneration. We provide evidence that TSP-17 inhibits the C. elegans dopamine transporter DAT-1, leading to increased neuronal 6-OHDA uptake in tsp-17 mutants. TSP-17 also protects against toxicity conferred by excessive intracellular dopamine. TSP-17 interacts with the DOP-2 dopamine receptor, possibly as part of a pathway that negatively regulates DAT-1. tsp-17 mutants have subtle behavioral phenotypes that are partly conferred by aberrant dopamine signaling. In summary, we have used C. elegans genetics to model key aspects of PD.
| Parkinson's Disease (PD) is the second most common neurodegenerative disease, after Alzheimer's disease, and affects ∼2% of the population aged over 65 years. Loss of dopaminergic neurons is a pathological hallmark of PD [1], [2] and aspects of this neurodegeneration have been modeled in C. elegans [3], [4]. The etiology of PD is largely unknown and its heritability is generally rather low; however ∼5–10% of cases are associated with monogenetically inherited mutations [5]. Approximately 15 disease loci are known, most of which are conserved in C. elegans [6], [7]. The vast majority of PD cases are ‘sporadic’ with no clear family history. Besides aging, epidemiological studies have shown risk factors for ‘sporadic’ PD to include a long-term history of rural living, farming, well-water drinking and pesticide exposure. The most extreme examples of toxin-induced PD-like symptoms were linked to the accidental exposure to MPTP (N-methyl-4-phenyl-1,2,3,6-tetrahydropyridine). Similar to sporadic PD cases, PD-like symptoms resulting from MPTP exposure could be alleviated by administration of the dopamine precursor L-3,4-dihydrooxyphenylalanine (L-DOPA) [8]. Exposure to pesticides such as paraquat and rotenone has also been implicated in PD development [9]. The disease is therefore thought to be triggered by a combination of environmental factors and genetic susceptibility [5].
MPTP, paraquat and rotenone all block the mitochondrial electron transport chain, leading to oxidative damage [10], and have been extensively used to model PD neurodegeneration. 6-Hydroxydopamine (6-OHDA), an oxidation product of dopamine, is another neurotoxin widely used in mammalian PD models to induce the specific degeneration of dopaminergic neurons [11]. 6-OHDA was initially identified as a metabolite of dopamine [12], and there is some evidence that 6-OHDA exposure might be linked to PD. 6-OHDA was also identified as a naturally occurring amine in human urine, and has been detected at higher concentrations in PD patients [13]. Furthermore, high 6-OHDA levels were found in postmortem brain samples from PD patients [14]. It has been reported that 6-OHDA interaction with oxygen results in the production of reactive oxygen species (ROS), which in turn trigger free radical-mediated neuronal degeneration [2], [12]. Other dopamine metabolites may also cause oxidative damage [15]. Nevertheless, the mechanism by which 6-OHDA induces neuronal degeneration remains largely unknown [16].
Although there is no treatment to prevent or halt neuronal loss, L-DOPA administration is still one of the most effective treatments for alleviating PD symptoms [17], [18]. However, the effectiveness of L-DOPA declines over time. Prolonged L-DOPA treatment is also potentially neurotoxic [11], [15]. Although not confirmed in a large longitudinal study of L-DOPA use in PD patients (ELLDOPA trial), this nevertheless remains a major concern [19].
C. elegans has been used as a model to study the structure and function of the nervous system, which in hermaphrodite worms consists of 302 neurons [20], [21]. C. elegans dopaminergic neurons are functionally related to those of humans. The genes driving the biochemical processes involved in dopamine metabolism (as well as most PD-associated loci) [6] are also highly conserved in worms [22]. Dopaminergic neurons can be readily visualized in vivo using appropriate GFP markers. Analogous to vertebrate systems, dopaminergic neurons undergo neurodegeneration upon treatment with 6-OHDA. It has been shown that 6-OHDA can enter dopaminergic neurons through the DAT-1 dopamine transporter and thus trigger their degeneration [3]. The exact type of cellular death that occurs following 6-OHDA intoxication is unknown. Electron microscopy has shown apoptotic-like condensed chromatin structures in dying neurons, suggesting that 6-OHDA induces apoptosis. However, 6-OHDA-induced neurodegeneration in C. elegans is independent of CED-4/Apaf1 and CED-3/caspase, two components of the core apoptotic machinery [3]. In an independent study, inactivation of C. elegans autophagy genes partially suppressed 6-OHDA-induced dopaminergic death, suggesting that autophagy might also be involved in this process [23].
During synaptic transmission most of the released dopamine is transported back into the presynaptic terminal by the dopamine reuptake transporter (DAT1) (for a review, see [24]. Therefore, activity of this transporter affects the duration and extent of dopamine signaling. Mammalian cell experiments led to the identification of several proteins that interact with DAT1 to modulate its activity, cell surface expression and trafficking. These include protein kinase C, dopamine D2 receptors (discussed below), SNCA and parkin [25]–[28]. The physiological actions of dopamine are mediated by conserved seven-transmembrane dopamine receptors, designated D1–5. Dopamine receptors are coupled to guanosine triphosphate-binding proteins (G proteins) and are classified into D1 or D2 type dopamine receptors based on their antagonistic effect on adenylyl cyclase activity [29], [30]. D1 dopamine receptors, DOP-1 in worms, are solely found in postsynaptic dopamine-receptive cells, whereas in C. elegans the D2 type receptors DOP-2 and DOP-3 are expressed pre and postsynaptically, respectively [31]–[33].
In vertebrates, the dopamine system plays a crucial role in regulating movement, reward and cognition. Dopamine-deficient newborn mice die as a result of severe motor impairments [34], [35]. In contrast, C. elegans mutants defective in dopamine synthesis are viable, thus facilitating investigations into dopamine-mediated behavior in these animals. Dopaminergic neurons in C. elegans are required for specific, well-described and quantifiable behaviors, often associated with locomotion and feeding. For instance, the basal slowing response allows worms to reduce their speed when encountering a bacterial lawn, which is their food source [36]. Another behavior mediated by dopamine signaling is referred to as “swimming-induced paralysis” (SWIP): dat-1-deficient worms exhibit rapid paralysis in liquid, unlike wild-type controls [37].
Using an unbiased forward genetic approach we identified tsp-17 as a gene that protects dopaminergic neurons from 6–OHDA-mediated neurodegeneration. We provide evidence that TSP-17 regulates DAT-1 transporter activity. Furthermore, our results suggest that DAT-1 regulation by TSP-17 is partly mediated by D2 dopamine receptors.
In order to find genes that protect dopaminergic neurons, we performed a genetic screen for mutants conferring hypersensitivity to 6-OHDA. By adapting procedures initially established by Nass et al. [3] and using the same pdat-1::GFP reporter that highlights dopaminergic neurons, we screened ∼2500 F2 ethyl methanesulfonate (EMS)-mutagenized worms at the L1 developmental stage by incubating with 10 mM 6-OHDA for 1 h. This procedure, which is based on reduced, altered, or absent pdat-1::GFP expression, does not lead to neurodegeneration in >95% of wild-type worms, thus allowing the identification of mutants conferring hypersensitivity to 6-OHDA. Of the initial five mutant candidates, only gt1681 maintained a strong hypersensitive phenotype upon backcrossing (Figure 1A, Figure S1). 6-OHDA-induced degeneration of both wild-type and gt1681 neurons exhibits the same morphological features and pattern of degeneration initially described by Nass et al. [3]. Axonal blebbing becomes apparent (Figure 1B, inset, arrows) a feature also consistent with morphological changes previously observed by electron microscopy. Worms were scored 24, 48 and 72 h after intoxication. Neurons were lost in less than 10% of wild-type worms after 72 h. In contrast, all dopaminergic neurons were lost in ∼40% of gt1681 worms and partial dopaminergic loss was observed in an additional ∼30% of mutant worms after only 24 h (Figure 1A). The extent of neurodegeneration was further increased 72 h after intoxication, with ∼90% of worms displaying total dopaminergic loss at the adult stage (Figure 1A). Enhanced neurodegeneration in the gt1681 background, albeit to a lesser extent, also occurred in L2, L3 and L4 larvae treated with 6-OHDA; no such enhancement was seen in adults (Figure 1C). To exclude the possibility that neurodegeneration might be caused by increased net 6-OHDA uptake at the organismal level, we took advantage of the partial growth retardation conferred by 6-OHDA treatment. By scoring for progression to ensuing developmental stages, we found the growth of wild-type and gt1681 worms to be similarly retarded upon toxin treatment, suggesting that gt1681 specifically affects dopaminergic neurons (Figure 1D).
The gt1681 mutant is recessive in hermaphrodites (Figure 2A). Genetic linkage was established by single nucleotide polymorphism (SNP) mapping, which placed gt1681 on the left arm of the X chromosome. Using unc-20 and lon-2 genetic markers to perform three-factor mapping, the locus was further refined to ∼10 map units. A cross between an unc-20 gt1681 lon-2 triple mutant and the CB4856 “Hawaii” mapping strain enabled us to assess the position of single recombination events relative to gt1681. This analysis localized gt1681 to an interval between nucleotides 3,659,480 and 3,737,466 on the physical map. In parallel, next generation sequencing revealed a single exonic mutation within this interval, leading to a guanine to adenine substitution in the C02F12.1 open reading frame and resulting in a glycine to glutamic acid change at position 109 of the encoded protein (Figure 2B). C02F12.1 encodes a tetraspanin family, integral membrane protein called TSP-17 (see below). Rescue of the phenotype by a fosmid (WRM0626aC02) encompassing tsp-17 and by a tsp-17-encoding transgene (Figure 2C) provides further evidence that gt1681 confers 6-OHDA hypersensitivity. Hypersensitivity is also conferred by the vc2026 allele, a substitution obtained via the Million Mutation Project [38] that results in a glycine to arginine change at position 109 (Figure 2B, 2D). Finally, two deletion alleles, generously provided by the Japanese Knockout Consortium, affecting the first exons of tsp-17 also confer hypersensitivity to 6-OHDA-mediated neurotoxicity (Figure 2B, 2D) as does the trans-heterozygous gt1681/tm4995 mutant combination (Figure 2A).
Tetraspanins constitute a large protein family, with 30 and 21 members encoded in the human and C. elegans genomes, respectively [39]–[41]. Most tetraspanins have not been functionally characterized. In vertebrates, tetraspanins are suggested to be involved in cell–cell fusion, cell adhesion, cell motility and tumor metastasis [42]. In C. elegans, TSP-12 is involved in modulating Notch signaling, and specific hypodermal TSP-15 expression is required to mediate covalent tyrosine–tyrosine cross-linking during cuticle formation [43], [44]. C. elegans tsp-17 is predicted to encode two isoforms. The large isoform, C02F12.1b, encodes a 312 amino acid protein containing four TM domains. The short isoform, C02F12.1a, encodes a 243 amino acid protein that, unlike typical tetraspanins, contains only three transmembrane domains and does not have an intracellular N-terminus. The amino acid change at position 109 in gt1681 affects a highly conserved residue in the third transmembrane domain of the long isoform (Figure 2B, 2E). We confirmed expression of mRNAs encoding for both isoforms, and verified the predicted intron–exon structure (Figure 2B). Using BLAST protein analysis of C. elegans TSP-17, we found the most likely human orthologs of TSP-17 to be CD63, Tspan5 and CD82 (Figure 2E). A previous phylogenetic analysis placed TSP-17 within the human CD82 subfamily [45]. However, our attempts to firmly establish an orthologous relationship between TSP-17 and a single human tetraspanin or a distinct subfamily of human tetraspanins were unsuccessful. Our phylogenetic analysis included all tetraspanins from several nematodes, arthropods, cnidarians and chordates (Figure S2). We speculate that the rapid evolution of this protein family, as often occurs with membrane proteins, compromised our ability to firmly identify a human ortholog of C. elegans TSP-17.
To assess the TSP-17 expression pattern, we used biolistic bombardment to generate transgenic worms (TG2439) expressing a tsp-17::GFP gene fusion (NM001) under the control of its own promoter and 3′UTR. A dat-1 (promoter)::mCherry fusion (PBI001) was co-bombarded to mark dopaminergic neurons. The tsp-17::GFP gene fusion largely suppressed the hypersensitivity phenotype conferred by tsp-17, thus confirming its functionality (Figure 2C, bar 3). Importantly, fusion protein expression was observed in all dopaminergic neurons: it was uniform along axons and dendrites of both dorsal and ventral pairs of CEP neurons, as well as in ADE neurons (Figure 3A–I, arrows indicate axons and dendrites) and in the posterior PDE neurons. Within the cell body, the TSP-17::GFP fusion seems to be excluded from the nucleus, a pattern that is more evident in a “close-up” image of a PDE neuron, where the signal appears to form a ring-like structure around the nucleus (Figure 3J–L arrowheads). mCherry aggregates (which are not linked to neurodegeneration) form dot-like structures in dendrites and axons (arrows), and the surrounding TSP-17 fluorescent signal suggests plasma membrane expression (arrow, Figure 3K). TSP-17 enrichment at the plasma membrane can be observed most prominently in the large cells of the vulva and the sheath cells enclosing the spermatheca (Figure 3M, N). In the spermatheca, TSP-17::GFP expression is also clearly enriched around the nucleus (Figure 3N, arrowheads), possibly localizing to the nuclear membrane or endoplasmic reticulum (Figure 3N, arrowhead). Analysis of subcellular localization in the vulva and spermatheca revealed that the TSP-17::GFP (gt1681) mutant protein is uniformly expressed in the cytoplasm, with a loss of enrichment at the plasma membrane and around the nucleus (Figure S3A). Thus, the gt1681 mutation, which leads to an amino acid change in the fourth transmembrane domain, might compromise the membrane localization of TSP-17 and therefore block its function. TSP-17::GFP is also expressed in multiple neurons throughout worm development. For instance, the NSM serotonergic neuron, which is characterized by extensive axon sprouting, shows TSP-17::GFP expression along its entire length (Figure 3O). Prominent expression was also observed in the muscles of early stage larvae (Figure 3P). Finally expression also appears to be apparent in muscles of the adult head (Figure 3B, C, H, I). In summary, the TSP-17::GFP expression indicates that TSP-17 is expressed in dopaminergic neurons. Transgene expression in dopaminergic neurons was also confirmed by analyzing a TSP-17::GFP expressing transgenic strain crossed to a DAT-1 reporter strain (Figure S3B). We cannot rule out expression of TSP-17 not uncovered by the transgene, due to missing regulatory sequences. We next wanted to investigate whether TSP-17 expression in dopaminergic neurons protects them from 6-OHDA-mediated neurodegeneration. By direct injection of transgenes into the gonad, we generated transgenic worms overexpressing TSP-17 under the control of the dat-1 promoter. Consistent with TSP-17 expression in dopaminergic neurons, we found partial rescue of the hypersensitivity conferred by gt1681 (Figure 2C, compare bars 1, 2 and 4). Interestingly, overexpression of TSP-17 and TSP-17 (gt1681) under the dat-1 promoter led to spontaneous neurodegeneration (Figure S4A, B, respectively). This phenotype tended to be more severe following TSP-17 (gt1681) overexpression. Taken together, these data indicate that TSP-17 indeed functions in dopaminergic neurons, and that excessive TSP-17, especially the mutant form, leads to spontaneous neurodegeneration.
We next wished to address how TSP-17 protects dopaminergic neurons. We hypothesized that TSP-17 might affect dopamine synthesis, or dopamine and 6-OHDA uptake or degradation. Dopamine metabolism is itself a source of oxidative stress and may initiate ROS-mediated injury to dopaminergic neurons. The link between excessive dopamine exposure and toxicity is controversial, but overexpression of CAT-2, the rate-limiting enzyme in dopamine synthesis in C. elegans, is reported to lead to age-dependent degeneration of dopaminergic neurons [46]. We repeated these experiments, and indeed found that neurodegeneration conferred by CAT-2 overexpression in dopaminergic neurons is enhanced in the gt1681 mutant background (Figure 4A). In contrast, we found CAT-2 overexpression to confer a strong resistance toward 6-OHDA-dependent neurodegeneration in both wild-type and gt1618 backgrounds (Figure 4B). We consider it likely that 6-OHDA resistance conferred by CAT-2 overexpression can be explained by reduced 6-OHDA uptake into dopaminergic neurons in the presence of excessive levels of intracellular dopamine. Our results indicate that tsp-17 protects against 6-OHDA toxicity and toxicity caused by excessive dopamine.
Since these genetic interactions suggest that dopamine levels could be altered in tsp-17 mutants, we next investigated behavioral phenotypes associated with dopamine. Dopamine synthesis and release are required for the basal slowing response, in which worms reduce their speed when encountering a bacterial lawn [36]. We did not observe a defect in this response, indicating that both dopamine synthesis and extracellular dopamine sensing by receptors are intact in tsp-17 mutants (Figure S5A). One of the most accessible phenotypes thought to be associated with excessive extracellular dopamine is the SWIP (Swimming Induced Paralysis) phenotype [37]. While wild-type worms placed into a drop of water maintain their thrashing frequency dat-1 mutants become progressively paralyzed. The SWIP phenotype is ascribed to excessive extracellular dopamine as a consequence of the reuptake defect in the dat-1 mutant. Excessive extracellular dopamine triggers paralysis by hyperactivating the DOP-3 receptor expressed on cholinergic neurons and hence blocking acetylcholine release [33]. To perform this experiment, we placed L4 worms into drops of water and scored their ability to swim over a period of 30 minutes. As expected, we found that wild-type but not dat-1 mutant worms can swim for 30 minutes with no change in the speed or pattern of swimming. All four tsp-17 mutants showed a partial SWIP phenotype (Figure 5A). This phenotype is probably caused by dopaminergic signaling because it can be rescued by deletion of the dop-3 dopamine receptor and by deletion of the cat-2 tyrosine hydroxylase (Figure 5A and Figure 5B). It was surprising to find a SWIP phenotype in tsp-17 mutants as we argue that tsp-17 inhibits dat-1 function (see below). While elucidating the exact mechanism of how TSP-17 affects behavioral phenotypes will require further investigation we speculate that hyper-activation of DAT-1 in tsp-17 strains could trigger a feedback loop that transiently enhance extracellular dopamine levels inducing the weak SWIP phenotype we observe.
We also tested for a SWIP phenotype in L1 stage worms, and found that all tsp-17 mutants tested, except the gt1681 allele, behaved similarly to dat-1 mutants (Figure 5C, D). This phenotype, however, is not suppressed by a dop-3 mutation or blocked by a cat-2 mutation (Figure 5D and Figure S5B). We discovered that the “L1 SWIP phenotype” is linked to lethality because worms placed onto agar plates after SWIP assay show reduced viability (Figure S5C, D). Thus, the L1 “swimming-induced lethality” phenotype is unlikely to be related to dopamine levels. Given that TSP-17 is expressed in body wall muscles in L1 larvae, we speculate that swimming-induced lethality might be caused by a muscle defect.
To systematically test whether TSP-17 protects dopaminergic neurons by modulating dopamine metabolism, catabolism, reuptake or signaling, we performed a genetic epistasis analysis. As expected, tsp-17 dat-1 double mutants were completely resistant to 6-OHDA-induced neurodegeneration, consistent with the notion that TSP-17 does not bypass 6-OHDA uptake by the DAT-1 dopamine transporter (Figure 6A). We observed no alterations in 6-OHDA sensitivity in cat-2 (tyrosine hydroxylase), bas-1 (aromatic amino acid decarboxylase/AAADC) and cat-1 (VMAT ortholog required for dopamine packaging) tsp-17 double mutants, indicating that TSP-17 is unlikely to affect levels of dopamine synthesis or packaging (Figure S6).
As 6-OHDA can enter dopaminergic neurons through the DAT-1 transporter owing to its structural similarity to dopamine [3], [47], we wondered whether DAT-1 localization or activity is modified in a tsp-17 mutant background. Having established that 6-OHDA hypersensitivity in tsp-17 worms depends on the DAT-1 transporter (Figure 6A), we tested the hypothesis that enhanced DAT-1 transporter activity may contribute to enhanced 6-OHDA-mediated neurotoxicity. Using a functional pdat-1::dat-1::YFP translational fusion, we found that overexpression of this transgene generated by bombardment does not confer overt 6-OHDA hypersensitivity (Figure 6A, Figure S7). Furthermore, the localization of DAT-1::YFP was similar between wild-type and tsp-17 mutants worms (Figure S8A), a notion further confirmed by Structural Illumination ‘super resolution’ images of CEP dendrites (Figure S8B). Additionally, photobleaching experiments indicated that ∼half of DAT-1::YFP is in the mobile fraction and that the t1/2 is around 30 seconds in both wild-type and tsp-17(gt1681) worms (Figure S8C–E). We thus aimed to test whether TSP-17 negatively regulates DAT-1 activity using a pharmacological approach. We confirmed previous reports that imipramine specifically inhibits the DAT-1 transporter in the worm [3] (Figure 6B, left panels, wild-type 0.25 mM and 1 mM). We reasoned that if DAT-1 is hyperactive in tsp-17 (gt1681), relatively more imipramine should be needed to inhibit DAT-1 activity and prevent neurodegeneration. We thus treated wild-type, tsp-17 (gt1681) worms and wild-type worms overexpressing DAT-1::YFP with 10 mM 6-OHDA and increasing doses of imipramine (Figure 6B, middle and right panels). We indeed found that higher levels of imipramine are needed to reduce neurodegeneration in DAT-1::YFP overexpressing worms and in tsp-17 (gt1681) worms, and that the effect being stronger in the tsp-17 (gt1681) mutant. Reduced levels of neurodegeneration levels were most clearly observed when concentrations of 0.125 mM and 0.25 mM imipramine were used (Figure 6B). This result provides evidence that DAT-1 activity may be higher in the tsp-17 mutant background. We aimed to provide further support for this hypothesis by directly measuring dopamine uptake, following previously described procedures. We macerated C. elegans embryos to establish primary embryonic cell cultures, and used these for dopamine uptake assays [48], [49]. Using two concentrations of tritiated dopamine, we indeed found increased dopamine uptake in tsp-17 mutants (Figure 6C, D). We note that we found this in 7/8 repeat experiments. However, we also note that only a very small proportion of tissue culture cells are dopaminergic neurons and that the absolute amount of dopamine uptake is low especially in the wild-type background.
Our combined genetic, pharmacological and biochemical analysis suggests that TSP-17 modulates DAT-1 activity. Previous studies using tissue culture-based assays demonstrated that dopamine receptor activation might promote DAT-1 activity [25], [50], [51]. Consistent with these results, we found dop-2 and dop-3 mutant worms to be partially resistant to high doses of 6-OHDA compared to wild-type (Figure 7A). We therefore investigated whether tsp-17 genetically interacts with dopamine receptors to modify DAT-1 activity and confer differential 6-OHDA sensitivity. This was done by assessing the sensitivity of tsp-17 mutants in the absence of the C. elegans DOP-1 D1-like receptor and/or in the absence of the DOP-2 and/or DOP-3 D2-like receptors. C. elegans DOP-1 is expressed in a variety of cells, including cholinergic neurons, mechanosensory neurons, head muscles and neuronal support cells. DOP-3 is expressed postsynaptically and its antagonism of DOP-1 in cholinergic neurons is required for the regulation of locomotion [33]. The DOP-2 receptor is expressed both postsynaptically and presynaptically. When expressed presynaptically, it acts as an autoreceptor on the plasma membrane of dopaminergic neurons. We found that dop-1; tsp-17 (gt1681) was as sensitive to 6-OHDA as the respective tsp-17 single mutant. In contrast, 6-OHDA hypersensitivity was reduced in dop-2; tsp-17 (gt1681) and dop-2; tsp-17 (tm4994) and in dop-3; tsp-17 (gt1681) and dop-3; tsp-17 (tm4994) double mutant worms (Figure 7B, C and Figure S9) Our genetic data thus argue that TSP-17 might inhibit DOP-2 and DOP-3 function, which in turn might be required for full DAT-1 transporter activity (Figure 7A, E). Given that deletion of dop-2 and dop-3 only partially rescues 6-OHDA hypersensitivity in tsp-17 mutants, we speculate that TSP-17 also inhibits DAT-1 activity independently of DOP-2 and DOP-3.
We next aimed to investigate how TSP-17 might regulate DAT-1 or D2-like receptors to modulate DAT-1 activity. Given that these are integral membrane proteins, we employed the split-ubiquitin membrane-based yeast two-hybrid system [52]. In this system, a C-terminal ubiquitin moiety fused to a transmembrane protein and a transcription factor is used a bait. An N-terminal ubiquitin moiety is used as the “prey.” Upon “reconstruction” of the split ubiquitin, this molecule is recognized by a protease, which cleaves the transcription factor, thus promoting reporter gene activation. By employing various bait and prey fusions with TSP-17, DAT-1 and DOP-2, we could not find a direct interaction between TSP-17 and DAT-1 using the split-ubiquitin system (Figure 7D). In contrast, we found that DOP-2 and TSP-17 may indeed interact. The specificity of this interaction was clearly revealed when the beta-galactosidase reporter assay was used as an output. In addition, yeast colony formation on his-3 or his-3 ade-2 plates was enhanced when the corresponding reporters where used (Figure 7D). Thus, TSP-17 might modulate DOP-2 activity by a direct physical interaction, consistent with TSP-17 affecting ligand binding, downstream signaling or membrane trafficking of DOP-2-like receptors. Our genetic data also suggest that TSP-17 might also act via other factors to dampen DAT-1 activity (Figure 7B).
Using C. elegans as a model and employing unbiased genetic approaches, we aimed to find neuroprotective genes that alleviate the 6-OHDA-induced degeneration of dopaminergic neurons. Based on our genetic data, which is supported by the characterization of several alleles and transgenic rescue experiments, we provide compelling evidence that TSP-17 protects dopaminergic neurons from 6-OHDA-mediated toxicity. TSP-17 appears to function in dopaminergic neurons, and our combined genetic, pharmacological and biochemical evidence suggests that it might act by antagonizing DAT-1 dopamine transporter activity. We do not know how TSP-17 regulates DAT-1 at a mechanistic level.
TSP-17 is a member of the evolutionarily conserved family of tetraspanins, comprising 20–50 kDa membrane proteins that contain four transmembrane domains. A characteristic feature of tetraspanins is their ability to form lateral associations with each other and with other proteins. Such interactions are thought to lead to a dynamic assembly, resulting in the formation of a network of molecular interactions referred to as the tetraspanin web [41], [53]. Tetraspanins are thought to have regulatory functions in the ligand binding, downstream signaling, protein trafficking and proteolytic activities of associated proteins [42], [54]. In C. elegans, only two tetraspanins have known functions. TSP-15 appears to be required to activate the BLI-3 dual oxidase to regulate H202 production at the plasma membrane and thus alter dityrosine cross-linkage of extracellular matrix proteins [44], [55]. Genetic evidence suggests that TSP-12, most closely related to human TSPAN33, appears to facilitate Notch signaling redundantly with TSP-14. Thus conserved tetraspanins likely function by facilitating γ-secretase cleavage of the membrane-bound form of Notch, thus promoting nuclear localization of this transcription factor [43].
DAT-1 hyperactivity in the tsp-17 mutants could result from altered DAT-1 localization or abundance at the cell membrane; alternatively, TSP-17 might indirectly regulate DAT-1 activity. Using a functional DAT-1::YFP construct, we did not see any obvious change in DAT-1 expression, localization, or change in half life in tsp-17 mutants and we thus favor the idea that TSP-17 regulates DAT-1 activity. Our finding that TSP-17 genetically and biochemically interacts with the DOP-2 D2-like dopamine receptor, suggests an indirect mode of DAT-1 regulation by TSP-17 (Figure 7E). Our genetic analysis provides evidence that TSP-17 might in part regulate DAT-1 via DOP-2 and DOP-3 dopamine receptors (Figure 7E). We found that depletion of the D2-like dopamine receptors, DOP-2 and/or DOP-3, in tsp-17 mutants leads to a moderate reduction in the 6-OHDA hypersensitivity conferred by tsp-17, while D2-like dopamine receptor single knockout strains show the same 6-OHDA sensitivity as wild-type worms. Thus, our analysis suggests that tsp-17 genetically interacts with D2-like dopamine receptors, in line with our observation that TSP-17 directly binds to DOP-2. In mammalian systems, dopamine autoreceptors are reported to have a major role in providing inhibitory feedback to adjust the rate of neuronal firing, dopamine synthesis and dopamine release in response to the dopamine level in the synaptic cleft [30], [32]. Several studies suggest that vertebrate D2 dopamine receptors also modulate DAT-1 activity to regulate the dopamine level in the synaptic cleft. Cass and Gerhardt used pharmacological approaches to demonstrate that inhibition of D2 class dopamine receptors significantly inhibits DAT function [50]. Two independent studies provided evidence that D2 receptors regulate both the activity and cell surface expression of DAT-1 [25], [51]. Nevertheless, further investigations are required to establish functional links between C. elegans DOP-2 receptors and DAT-1 activity. The ability of TSP-17 to inhibit DAT-1 both via DOP-2 and independent of D2-like receptors (Figure 7E) suggests that TSP-17 modulates the activity of multiple signaling proteins. Indeed, our observation of excessive neurodegeneration following wild-type, and especially mutant, TSP-17 overexpression in dopaminergic neurons hints that malfunctioning and/or excessive TSP-17 blocks pathways needed to maintain the integrity of dopaminergic neurons. The enhanced defect associated with overexpression of mutant TSP-17 that fails to show the correct cytoplasmic localization hints the neurotoxicity might be conferred by the sequestration of TSP-17 interacting proteins essential for neuronal survival.
Dopamine neuronal dysfunction has been associated with several common neurobehavioral disorders, including drug addiction, schizophrenia and attention-deficit hyperactivity disorder [32], [56]–[58]. The DAT-1 dopamine transporter plays a central role in dopamine signaling, and it is likely to be subjected to complex modes of regulation. DAT-1 is the target of psychoactive addictive drugs such as cocaine and amphetamine, and DAT1 overexpression leads to increased amphetamine sensitivity [59]–[63]. Mechanisms related to dopamine signaling tend to be evolutionarily conserved. Thus, studies aimed to genetically define modulators of dopamine signaling and 6-OHDA-mediated toxicity will provide important insights into the mechanisms of dopamine signaling in health and disease.
Idiopathic PD is thought to be triggered by a combination of environmental factors and genetic susceptibility, and a case has been made that exposure to environmental toxins such as the pesticides paraquat and rotenone leads to increased PD [9]. Indeed, chemical and tissue culture studies have provided evidence that increased dopamine levels may lead to enhanced neurodegeneration, probably through the generation of toxic intermediates such as the neurotoxic product of dopamine oxidation, 6-OHDA [13], [15], [64]–[68]. The specificity of 6-OHDA entry into dopamine neurons depends on DAT, and DAT antagonists can block uptake [3], [4], [11], [47]. Interestingly, DAT-1 hyperactivity in tsp-17 mutants further enhances the neurodegeneration conferred by elevated dopamine synthesis in CAT2 tyrosine hydroxylase-overexpressing worm strains. Thus, DAT-1 hyperactivity might enhance neurodegeneration by further increasing the intracellular concentration of dopamine and/or toxic metabolites. DAT1 expression or activity has not been linked to PD, but it is intriguing that among dopamine neurons those residing in the substantia nigra express the highest DAT levels in vivo and are most strongly affected in PD [4], [60].
Strains were grown at 20°C under standard conditions, unless indicated otherwise. N2 Bristol was used as the wild-type strain. The tsp-17(tm4994) and tsp-17(tm5169) mutants were generated and kindly provided by Shohei Mitani of the National Bioresource Project for the Nematode (http://www.shigen.nig.ac.jp/c.elegans/). Details of the respective alleles are described by the National Bioresource Project for the Nematode and by WormBase (www.wormbase.org). All mutants were outcrossed a minimum of four times to the TG2435 vtIs1[pdat-1::gfp] strain originally generated by the Blakely laboratory (BY200) and repeatedly crossed into the N2 background.
TG2435 vtIs1[pdat-1::gfp; rol-6] V,
TG1681 vtIs1 V; tsp-17(gt1681) X,
TG2436 vtIs1 V; tsp-17(tm4994) X,
TG2437 vtIs1 V; tsp-17(tm5169) X,
TG2438 vtIs1 V; tsp-17(gk276386) X,
TG2462 vtIs1 V; CB4856,
TG2463 vtIs1 V; lon-2(e678) unc-20(e112) X,
TG2464 vtIs1 V; tsp-17(gt1681) unc-20(e112) X,
TG2465 vtIs1 V; tsp-17(gt1681) lon-2(e678) X,
TG2395 cat-2(e1112) II; vtIs1 V,
TG2394 cat-2(e1112) II; vtIs1 V; tsp-17(gt1681) X,
TG2396 bas-1(tm351) III; vtIs1 V,
TG2397 bas-1(tm351) III; vtIs1 V; tsp-17(gt1681) X,
TG2399 vtIs1 V; cat-1(e1111) X,
TG2398 vtIs1 V; cat-1(e1111) tsp-17(gt1681) X,
TG2400 dat-1(ok157) III; vtIs1 V,
TG2401 dat-1(ok157) III; vtIs1 V; tsp-17(gt1681) X,
TG2404 amx-1(ok659) III; vtIs1 V,
TG2403 amx-1(ok659) III; vtIs1 V; tsp-17(gt1681) X,
TG2406 amx-2(ok1235) I; vtIs1 V,
TG2405 amx-2(ok1235) I; vtIs1 V; tsp-17(gt1681) X,
TG2408 amx-2(ok1235) I; amx-1(ok659) III; vtIs1 V,
TG2407amx-2(ok1235) I; amx-1(ok659) III; vtIs1 V; tsp-17(gt1681) X,
TG2410 vtIs1 V; dop-1(vs100) X,
TG2409 vtIs1 V; dop-1(vs100) tsp-17(gt1681) X,
TG2412 vtIs1 dop-2(vs105) V,
TG2411 vtIs1 dop-2(vs105) V; tsp-17(gt1681) X,
TG2414 vtIs1 V; dop-3(vs106) X,
TG2413 vtIs1 V; dop-3(vs106) tsp-17(gt1681) X,
TG2466 vtIs1 dop-2(vs105) V; dop-3(vs106) X,
TG2467 vtIs1 dop-2(vs105) V; dop-3(vs106) tsp-17(gt1681) X,
TG2415 vtIs1 dop-2(vs105) V; dop-1(vs100) dop-3(vs106) X,
TG2416 vtIs1 dop-2(vs105) V; dop-1(vs100) dop-3(vs106) tsp-17(gt1681) X,
UA57 baIn4[pdat-1::gfp pdat-1::cat-2],
TG2402 baIn4[pdat-1::gfp pdat-1::cat-2],; tsp-17 (gt1681) X,
TG2470 gtIn2469[pdat-1::dat-1::yfp::let-858 3′UTR, unc-119(+)]; gtIn2468[pdat-1::mcherry::let858 3′UTR, unc-119(+)]; unc-119(ed3) III,
TG2471 gtIn2469[pdat-1::dat-1::yfp::let-858 3′UTR, unc-119(+)]; gtIn2468[pdat-1::mcherry::let858 3′UTR, unc-119(+)]; unc-119(ed3) III; tsp-17(gt1681) X,
TG2439 gtIn2439[ptsp-17::tsp-17::gfp::tsp-17 3′UTR, pdat-1::mcherry::let858 3′UTR, unc-119(+)]; unc-119(ed3) III,
TG2472 tsp-17(gt1681) X; gtIn2439[ptsp-17::tsp-17::gfp::tsp-17 3′UTR, pdat-1::mcherry::let858 3′UTR, unc-119(+)]; unc-119(ed3) III,
TG2440 gtEx2440[pdat-1::tsp-17::cfp:: let-858 3′UTR, unc-119(+)]; unc-119(ed3) III; vtIs1 [pdat-1::gfp; rol-6] V,
TG2473 vtIs1 [pdat-1::gfp; rol-6] V; tsp-17(gt1681) X; gtEx2440 [pdat-1::tsp-17::cfp:: let-858 3′ UTR, unc-119(+)],
TG2474 vtIs1 [pdat-1::gfp; rol-6] V; unc-119(ed3) III; gtEx2474[pdat-1::tsp-17(G74E)::cfp:: let-858 3′UTR, unc-119(+)],
TG2478 cat-2(e1112) II; vtIs1V; tsp-17(tm4994) X,
TG2475 dat-1(ok157) III; vtIs1V; tsp-17(tm4995) X,
TG2477 vtIs1; dop-3(vs106) tsp-17(tm4995) X,
TG2476 dat-1(ok157) III; vtIs1V; dop-3(vs106) X,
EMS was added to 4 ml synchronized young adult worms in M9 buffer to a final concentration of 25 mM and incubated for 4 h at 20°C. Mutagenized worms were washed in M9 buffer and incubated at 15°C. Synchronous F1-generation L1 larvae were used for screening. F2-generation L1 larvae from mutagenized TG2435 dat-1::gfp (BY200) worms were used for the mutagenesis screen. L1 larvae were intoxicated with 10 mM 6-OHDA. After 72 h, worms with the highest incidence of neurodegeneration were isolated and scored as hypersensitive. SNP mapping of mutants was done as previously described [69].
To obtain synchronized L1 larvae, 1–10 adult worms (24 h post-L4 stage) were incubated in 70 µl M9 without food on at 20°C, with shaking at 500 rpm for 27–40 h to lay eggs. After hatching, all L1 larvae were collected. Approximately 50 L1 larvae were added to an assay mix (50 µl) containing 10 mM 6-OHDA and 40 mM ascorbic acid, and incubated for 1 h at 20°C, with shaking at 500 rpm. For co-treatment with imipramine or haloperidol, the respective compounds were added to the assay mix at the same time as 6-OHDA. After a 1-h incubation, M9 buffer (100 µl) was added to the assay mix, and the solution containing L1 worms was then transferred to an unseeded NGM plate. After 30 min, L1 worms were individually picked and transferred onto a fresh NGM plate seeded with a line of OP50 bacteria to ease subsequent scoring. Intoxicated worms were incubated at 20°C and scored for dopaminergic neurodegeneration every 24 h for 3 days. All 6-OHDA treatments were done in triplicate and at least 80–100 worms were tested for each strain and condition.
All worms used for SWIP analysis were grown on NGM plates seeded with E. coli OP50 bacteria. For each test, 5–10 L4 hermaphrodites or 10 L1 worms were placed into 40 µl water in a single well of a Pyrex Spot Plate. Paralyzed worms were counted at 1-min intervals using a Leica dissecting microscope [70]. L1 worms were hand picked from seeded plates, 12 hours after the addition of embryos, obtained by bleaching.
For semi-quantitative analyses of 6-OHDA-induced degeneration, worms were examined using a Leica fluorescent dissecting microscope. The absence of all eight dopaminergic neurons in worms was scored as “complete loss.” The presence of a complete, intact set of eight dopaminergic neurons was scored as “no loss.” Any intermediate situation, for example a damaged or absent subset of dopaminergic neurons or missing dendrite portions, was scored as a “partial loss.” Neurodegeneration resulting from cat-2 overexpression was scored using developmentally synchronized worms, as indicated. A DeltaVision microscope (Applied Precision) was used to acquire images. All images were analyzed using softWoRx Suite and softWoRx Explorer software (Applied Precision).
Total RNA was isolated and reverse transcribed from wild-type C. elegans (N2) using an RNeasy mini kit (QIAGEN). Coding regions of dop-2c (K09G1.4c) and dat-1 (T23G5.5) were amplified and cloned into pBT3-STE vectors (Dual Systems Schlieren) for expression of a fusion protein containing the C-terminal half of ubiquitin (Cub) and the artificial transcription factor LexA-VP16. tsp-17b (C02F12.1b) cDNA was amplified and cloned into prey vector pPR3-STE for expression of a fusion protein containing a mutated version of the N-terminal half of ubiquitin (NubG). Constructs were verified by DNA sequencing, and sequences of the respective constructs can be provided upon request. Yeast transformations and pairwise interaction assays were done according to the protocol of Dualsystems Schlieren.
Embryonic cells were prepared as described previously (Christensen, M, et al 2002, Neuron). The uptake assay was done according to Carvelli et al. (2004). Briefly, C. elegans cells cultured for 2 days were washed twice with KRH buffer (120 mM NaCl, 4.7 mM KCl, 1.2 mM KH2P04, 10 mM Hepes, 2.2 CaC12, 10 mM glucose, 0.1 mM ascorbic acid and 0.1 mM tropolone and 0.1 mM pargyline mono amine oxidase inhibitors) and incubated with 50 or 250 nM [3H]-dopamine for 20 min at room temperature. Uptake was terminated by three washes of ice-cold KRH buffer, and cells were lysed by incubation with 1% SDS for 20 min. [3H]-dopamine uptake was measured in each genetic background, based on radioactive counts, using a scintillation counter (PerkinElmer Liquid Scintillation Analyzer Tri-Carb 1800TR). Total cell numbers were determined with a hemocytometer and were used to normalize radioactive counts. Cell numbers varied between experiments but were not biased towards mutant or control strain: There were 400,000/400,000, 75,000/150,000 and 1,000,000/400,000 cells for control/mutant strain, respectively. Cell extraction and uptake assays were always done simultaneously for both strains. The error bars depict the standard error of the means (SEM).
Neurodegeneration and SWIP assay data are presented as the average of three biological replicates, and error bars represent the standard error of the mean, unless otherwise indicated. When assaying neurodegeneration statistical significance was calculated using the Chi-Sqare test using Yates p-values. http://www.quantpsy.org/chisq/chisq.htm. The statistical significance of differences in the SWIP assays (Figure 5) was calculated using the two-tailed t-test.
|
10.1371/journal.pgen.1004024 | Huntington's Disease Induced Cardiac Amyloidosis Is Reversed by Modulating Protein Folding and Oxidative Stress Pathways in the Drosophila Heart | Amyloid-like inclusions have been associated with Huntington's disease (HD), which is caused by expanded polyglutamine repeats in the Huntingtin protein. HD patients exhibit a high incidence of cardiovascular events, presumably as a result of accumulation of toxic amyloid-like inclusions. We have generated a Drosophila model of cardiac amyloidosis that exhibits accumulation of PolyQ aggregates and oxidative stress in myocardial cells, upon heart-specific expression of Huntingtin protein fragments (Htt-PolyQ) with disease-causing poly-glutamine repeats (PolyQ-46, PolyQ-72, and PolyQ-102). Cardiac expression of GFP-tagged Htt-PolyQs resulted in PolyQ length-dependent functional defects that included increased incidence of arrhythmias and extreme cardiac dilation, accompanied by a significant decrease in contractility. Structural and ultrastructural analysis of the myocardial cells revealed reduced myofibrillar content, myofibrillar disorganization, mitochondrial defects and the presence of PolyQ-GFP positive aggregates. Cardiac-specific expression of disease causing Poly-Q also shortens lifespan of flies dramatically. To further confirm the involvement of oxidative stress or protein unfolding and to understand the mechanism of PolyQ induced cardiomyopathy, we co-expressed expanded PolyQ-72 with the antioxidant superoxide dismutase (SOD) or the myosin chaperone UNC-45. Co-expression of SOD suppressed PolyQ-72 induced mitochondrial defects and partially suppressed aggregation as well as myofibrillar disorganization. However, co-expression of UNC-45 dramatically suppressed PolyQ-72 induced aggregation and partially suppressed myofibrillar disorganization. Moreover, co-expression of both UNC-45 and SOD more efficiently suppressed GFP-positive aggregates, myofibrillar disorganization and physiological cardiac defects induced by PolyQ-72 than did either treatment alone. Our results demonstrate that mutant-PolyQ induces aggregates, disrupts the sarcomeric organization of contractile proteins, leads to mitochondrial dysfunction and increases oxidative stress in cardiomyocytes leading to abnormal cardiac function. We conclude that modulation of both protein unfolding and oxidative stress pathways in the Drosophila heart model can ameliorate the detrimental PolyQ effects, thus providing unique insights into the genetic mechanisms underlying amyloid-induced cardiac failure in HD patients.
| Huntington's disease (HD) is associated with amyloid-like inclusions in the brain and heart, and accumulation of amyloid protein is associated with neurodegeneration and cardiomyopathy. Recent studies suggest that HD patients show increased susceptibility to cardiac failure. However, the mechanisms by which disease-causing poly-glutamine repeats (PolyQ) cause heart dysfunction in these patients are unclear. We have developed a novel Drosophila heart model that exhibits significant GFP-positive aggregates upon HD-causing PolyQ expression in myocardial cells resulting in PolyQ length-dependent physiological defects. Modulation of protein folding and oxidative stress pathways in this system reduced the number of aggregates and reversed the cardiac dysfunction in response to expression of disease-causing PolyQ. The ability to explore PolyQ-associated mechanisms of cardiomyopathy in a genetically tractable whole organism, Drosophila melanogaster, promises to provide novel insights into the relationship between amyloid accumulation and heart dysfunction. Our findings not only impact the understanding of PolyQ-induced cardiomyopathy but also other human cardiac diseases associated with oxidative stress, mitochondrial defects and protein homeostasis.
| Amyloidosis constitutes a large group of diseases characterized by the misfolding of proteins and the accumulation of protein aggregates in different tissues [1]–[3]. Huntington's disease (HD) is an inherited neurodegenerative disorder caused by mutations in the Huntingtin (HTT) protein which result in expanded Poly-glutamine (PolyQ, CAGn) repeats that cause aggregation-prone amyloidosis [4]–[7]. The molecular mechanism that leads to HD is not fully understood and presently no effective treatment exists [4], [8]–[10]. It has been well established that the length of the PolyQ repeat is important in the progression of disease [4], [5]. HTT with 6–35 PolyQ repeats does not cause HD. However, HTT with more than 40 PolyQ (CAG40) repeats results in HD [4], [5], [11]. In general HD is primarily considered as an aggregation-based disease; however, some studies have shown that disease-causing PolyQ repeats in HTT make it prone to misfolding and aggregation [4]–[7], [12]–[15].
HTT is expressed in several tissues in addition to the brain, including heart and skeletal muscles [11], [16]–[18] and is known to be involved in protein trafficking, vesicle transport and transcriptional events [4], [11]. HD is also associated with skeletal muscle atrophy [11], [19] and multiple epidemiological studies have shown that cardiovascular diseases and cardiac failure are the second leading cause of mortality in HD patients [8], [18], [20]. Cardiac failure is implicated as the cause of death in over 30% of HD patients, compared to 2% of the age-matched non–HD patients [8]–[11], [18], [20]. Although, the mechanism whereby mutant HTT causes muscle atrophy and cardiac defects is not known, it is possible that an increase in protein misfolding and the consequent high energy burden in cardiac cells play roles [8], [10], [11], [16], [18]. In support of this, recent evidence demonstrates nuclear and cytoplasmic PolyQ aggregates in non-CNS tissue [11], [16], [18]. Furthermore, neuronal expression of mutant HTT protein with expanded PolyQ or cardiac-specific expression of only the PolyQ pre-amyloid oligomers in mice leads to cardiac defects [10], [21], [22]. Furthermore, expression of mutant PolyQ-81 in mice and in rat neonatal cardiomyocytes results in amyloid as well as PolyQ-positive aggregates in the cytoplasm and over-expression of a chaperone αB-crystallin reduces PolyQ-induced aggresomes [21], [23]. Moreover, reduction of aggresomes upon over-expression αB-crystallin results in higher levels of amyloid oligomer and enhances toxicity [23]. Despite the availability of cell and mouse models to examine/investigate PolyQ expression in the heart, little is known about the mechanism that leads to cardiac dysfunction.
We have previously established the Drosophila heart as a useful model system to generate insights into the genetic basis of heart development and to elucidate the genetic interactions underlying heart physiology and age-dependent deterioration [24]–[27]. Recently, we showed in this genetic model that knock-down of the chaperone UNC-45 significantly reduced myosin expression and led to severe cardiac dilation [28]. Protein folding and oxidative stress pathways have previously been implicated in the development/pathology of HD [5], [6], [10], [18], [19], however, their involvement with cardiac phenotypes has not been explored. In the current study, we manipulate these two pathways to attempt to suppress the cardiac defects induced by mutant HD-associated PolyQ repeat lengths. These defects include the accumulation of GFP-positive aggregates, mitochondrial defects, oxidative stress and both functional and morphological cardiac abnormalities. We also found that cardiac over-expression (OE) of the chaperone UNC-45 suppressed amyloid deposition and ameliorated the heart function defects to some extent. In addition, OE of SOD as well as feeding with the dietary antioxidant resveratrol also partially suppressed the amyloid-induced cardiac dysfunction, whereas hydrogen peroxide feeding aggravated the heart defects. Our data suggest that the protein folding and ROS pathways interact in mediating the effects of mutant HTT as a near complete reversal of the cardiac defects was achieved when both pathways were modulated simultaneously in flies expressing disease-causing PolyQ repeats. Thus, our data show a deleterious effect of mutant PolyQ aggregates on cardiac function and indicate that these effects are the result of protein misfolding and/or concomitant oxidative stress.
To evaluate cardiac function following PolyQ-induced cardiomyopathy, we obtained Drosophila transgenic lines [5] expressing enhanced-GFP-tagged control or mutant Htt fragments (UAS-Httex1-QneGFP) with different PolyQ repeat lengths (Q25, Q46, Q72, and Q103). For simplicity, Httex1-Q25-eGFP, Httex1-Q46-eGFP, Httex1-Q72-eGFP and Httex1-Q102-eGFP are referred to as PolyQ-25, PolyQ-46, PolyQ-72 and PolyQ-102, respectively. Using the heart-specific driver Hand, we observed severe cardiac defects and/or extreme dilation upon expression of disease-causing PolyQ-46, PolyQ-72 and PolyQ-102, whereas the shorter PolyQ-25 had no measurable effect. Figure 1A shows contracted hearts from 1-week old flies with cardiac-specific expression of PolyQ-72 and an age-matched control expressing PolyQ-25. The heart tube expressing PolyQ-72 is clearly much less contracted during systole than is the PolyQ-25 control. Heart wall diameters during both systole and diastole are indicated by double headed arrows in the M-mode records produced from high speed movies; the line of pixels used to produce these records is indicated by the blue line in the still image. The ability of the PolyQ-72 heart to contract during systole is much reduced compared to the control heart. A comparison of representative M-mode records for 3-week old flies from each of the transgenic lines and wild type controls is shown in Figure 1B and demonstrates a progressive increase in cardiac arrhythmia with increased length of PolyQ. A comparison of PolyQ-25, PolyQ-46 and PolyQ-72 heart contractility, dilation and arrhythmia is also shown in supplementary movie S1. Hearts expressing the longer repeats, PolyQ-72 and PolyQ-102, also showed significant cardiac dilation in addition to arrhythmia (Figure 1B). Hearts from flies with cardiac-specific expression of mutant PolyQ also exhibited additional functional and morphological defects including floppy, non-contractile ostia and one or more non-contractile myocardial cells, primarily in the conical chamber (CC) and the adjacent chamber (Figure 1C). In PolyQ-72 and PolyQ-102 expressing hearts there were frequent asystolic periods as well as hearts that were completely unable to beat. The incidence of these qualitative defects is quantified in Figure 1C. In addition to these heart function defects, cardiac specific expression of the two longer PolyQ proteins (PolyQ-72 and PolyQ-102) significantly shortens the lifespan of flies (Fig. 1 D and Table S1).
Quantification of functional parameters from the high-speed movies demonstrated a significant increase in diameters during both systole (Figure 2A) and diastole (Figure 2B) compared to hearts expressing PolyQ-25 and wild-type controls; this dilation was more severe for the longer PolyQ repeats (PolyQ-72 and PolyQ-102). The observed cardiac dilation was accompanied by a significant reduction in heart contractility, measured as a decreased fractional shortening (% FS) in the PolyQ-46, PolyQ-72 and PolyQ-102 expressing hearts (Figure 2C). Long PolyQ repeats induced significant increases in both the systolic and diastolic intervals and this effect again appeared to be dependent on the “dose” of PolyQ (Figure 2D, 2E). The incidence in cardiac arrhythmias was also quantified (arrhythmia index) [27]–[29], and showed a PolyQ dose-dependent increase (Figure 2F, see also Figure 1B). Cardiac specific expression of PolyQ-25 did not significantly alter any of the measured cardiac function parameters when compared to hearts from control flies that lacked any PolyQ expression (Hand-Gal4/+). Similar alterations in cardiac physiological parameters were observed in response to disease-causing PolyQ expression (PolyQ-46, PolyQ-72 and PolyQ-102) in younger, 1 week old flies (Figure S1A to S1F). Taken together our data indicate that all the cardiac defects we observe are PolyQ-length dependent and suggest that the PolyQ-72 repeat length is sufficient to exert the maximal deleterious effect on these hearts.
To explore whether the cardiac physiological dysfunction in response to cardiac-specific expression of PolyQ is the result of amyloid accumulation we used a Green Fluorescent Protein (GFP) tag to visualize PolyQ proteins and amyloid deposits and phalloidin to detect F-actin, revealing the myofibrillar organization within myocardial cells. Hearts expressing non-disease causing PolyQ-25 (control) show densely packed actin-containing myofibrils arranged in a circumferential pattern within the cardiomyocytes (Figure 3A) and GFP-tagged PolyQ was found to be distributed homogeneously throughout the cytoplasm (Figure 3B). In contrast, expression of disease-causing PolyQ-72 resulted in noticeably reduced myofibrillar content and in severe myofibrillar disorganization (Figure 3C, dashed box). Interestingly, we also observed the presence of many GFP-positive aggregates of various sizes throughout the cardiomyocytes (Figure 3D). We used a well-documented filter trap assay [30]–[32] to confirm this increase in aggregate formation upon expression of mutant PolyQ-72. Although expression of both control PolyQ-25 and mutant PolyQ-72 protein was virtually the same relative to histone H2B (Fig. 3G, top and middle panels), the heart-specific expression of mutant PolyQ-72 resulted in a significant increase in GFP-positive aggregates compared to control hearts (Figure 3G, bottom panel).
We also used antibody against muscle myosin to explore the effect of mutant PolyQ on the myosin organization within the myofibrils. Myosin organization in myofibrils from control hearts exhibits a similar circumferential arrangement as for F-actin (thick arrow in Figure 3E); however, the myosin pattern appears significantly aberrant upon expression of PolyQ-72 (Figure 3F). In fact the majority of the staining visible in Fig. 3F is due to myosin in the non-cardiac longitudinal muscle fibers that run ventrally along the cardiac tube (thin arrows in Figure 3E and 3F). Some disorganized myosin-containing myofibrils are still seen upon expression of mutant PolyQ (Figure 3F, dashed boxes). This phenotype is reminiscent of knock-down of the myosin-specific chaperone UNC-45 [28], suggesting that long PolyQ aggregates might interfere with chaperone function. These data indicate that the presence of toxic aggregates leads to a reduction in cardiac myosin-actin content with disorganized myofibrils.
The ultrastructure of Drosophila cardiac muscle has been described previously in detail by Lehmacher et al. [33]. Transmission electron micrographs of transverse sections of 4 week-old hearts from PolyQ-25 control flies reveal a layer of contractile cardiomyocytes and a supporting layer of non-cardiac ventral-longitudinal fibers (VL, Figure 4A). Myocardial cells from PQ-25 controls contain mitochondria with densely packed cristae (4A, MT) that can be seen adjacent to the myofibrils (Figure 4A, MF). In contrast, micrographs from PolyQ-46 hearts show evidence of myofibrillar degeneration (4B, arrow) and severe mitochondrial fragmentation (4B, B', asterisks). Such mitochondrial fragmentation and alterations in cristae structure have previously been linked to increased apoptotic activity in primary striatal cultures from YAC128 HD transgenic mice as well as in neuronal expression of mutant HTT protein with expanded PolyQ in a mouse heart model [10], [34]. These defects are even more severe in hearts expressing longer forms of PolyQ-72 with near complete loss of myofibrillar architecture.
We also looked for autophagosome/lysosome structures and observed a significant amount of LysoTracker positive punctae upon expression of mutant PolyQ-72 (Figure S2 E). Significantly, most of the staining is co-localized with PolyQ-GFP punctae (Figure S2 F). In contrast, almost no GFP- or LysoTracker-positive punctae are seen upon expression of non-disease causing PolyQ-25 (Figure S2 A–C) suggesting a direct link between expression of mutant PolyQ and activation of the autophagy pathway.
The mitochondrial defects, myofibrillar disorganization and cardiac function abnormalities observed upon expression of mutant PolyQ could arise from aggregate-induced oxidative stress. We used dihydroethidium (DHE) to evaluate the role of reactive oxygen species (ROS) production in mediating the effects of disease-causing PolyQ. Cardiac expression of PolyQ-46 and PolyQ-72 resulted in 2- and 5-fold increases in DHE staining respectively (Figure 5E and 5H) compared to age-matched PolyQ-25 control hearts (Figure 5B). Furthermore, a number of the mutant-PolyQ induced GFP-aggregates colocalize with areas of strong DHE staining (arrows in Figure 5D and E; and Figure 5G and H) while expression of PolyQ-25 shows almost no GFP-positive punctae or DHE staining. These results confirm an association between PolyQ-induced aggregates and oxidative stress.
To explore whether induction of oxidative-stress could aggravate the PolyQ phenotype, flies expressing PolyQ-25 and PolyQ-46 in cardiac tissue were fed H2O2 for 3-weeks during adulthood. PolyQ-46 expressing hearts in the presence of oxidant showed significantly increased cardiac dilation. However, no such enlargements of cardiac diameters were seen in PolyQ-25 expressing hearts in the presence of oxidant (Figure 6A–6B). Although fractional shortening was decreased and cardiac arrhythmias were increased in PolyQ-46 expressing hearts by H2O2 feeding, both were affected to a similar extent as were control PolyQ-25 expressing hearts (Fig. 6C–F). Feeding oxidant to non-PolyQ expressing wild-type flies (Hand-Gal4/+) had similar minimal effects on cardiac parameters as for the PolyQ-25 controls (Figure S3A–S3F). The increased incidence of arrhythmia and reduced contractility of hearts expressing PolyQ-72 was also aggravated in the presence of oxidant (Figure S4A, S4B).
While PolyQ-46 hearts in the absence of oxidant do exhibit sparsely distributed amyloid-aggregates, the presence of H2O2 results in an increase (30%) in the density of aggregates (green GFP-punctae, Figure 6G, 6I). Furthermore, treatment with oxidant also resulted in more myofibrillar disorganization and loss compared to age-matched PolyQ-46 without oxidant (compare Figures 6G and 6H). Interestingly, muscle fiber organization remained virtually unchanged in hearts from wild-type controls (Hand/+) and PolyQ-25 expressing controls when fed H2O2 (compare Figure S3G with S3H and S3I with S3J). These data support the idea that oxidative stress enhances the accumulation of mutant PolyQ aggregates (GFP-punctae), myofibrillar disorganization and loss of actin-containing myofibrils and that these aggregates contribute to the cardiac dilation and heart function defects we observe. Additionally, our results show that oxidative stress exerts differential effects on heart function and structure depending upon the presence or absence of PolyQ aggregates. It is also possible that treatments with H2O2 may lead to secondary stress (such as initiation of the heat shock program). Assuming secondary stresses in response to H2O2 were similar between these two groups of flies, the only explanation for these more severe defects in the PolyQ-46 expressing group is an interaction between ROS signaling and mutant PolyQ expression. The H2O2 treatments may also affect other tissues such as neuronal tissue; therefore, it will be interesting to determine if control and mutant PolyQ affect cardiac and neuronal tissue under oxidative stress in a similar manner.
Because mutant PolyQ expressing hearts exhibited oxidative stress, mitochondrial defects as well as aggravated cardiac defects in response to hydrogen peroxide feeding, we tested whether over-expression of superoxide dismutase (SOD) could rescue the PolyQ-induced cardiomyopathy. We over-expressed SOD-1 or SOD-2 along with PolyQ-72 in fly hearts and examined the effect on cardiac function (Figure 7A–7F). In hearts from 3-week old flies over-expressing SOD-1 along with PolyQ-72 the cardiac dilation was significantly reduced compared to hearts expressing PolyQ-72 alone and was nearly the same as for wild-type Hand/+ hearts (Figure 7A, 7B). Cardiac contractility was also improved in SOD overexpressing hearts (Figure 7C). Both diastolic and systolic intervals were significantly lower (Figure 7D, 7E) and the incidence of arrhythmias was significantly reduced to nearly wild-type levels (Figure 7F) in PolyQ-72 hearts overexpressing SOD compared to hearts expressing PolyQ-72 alone. Analysis of the myofibrillar organization also showed a partial rescue of PolyQ-72–induced myofibril disarray upon SOD-1 expression and both the size and the density of mutant PolyQ aggregates were markedly reduced (Figure 7G, 7H, 7K, 7L). Co-expression of SOD-2 and PolyQ-72 produced a similar suppression of the PolyQ-72-induced cardiomyopathy (Figure S5A to S5F). No cardiac defects were seen when SOD-1 or SOD-2 was co-overexpressed with the non-disease causing PolyQ-25 (not shown).
Over-expression of SOD has been shown to suppress the cardiac defects associated with knock-down of mitochondrial assembly regulatory factor (MARF) [35]. We attempted to rescue Poly-Q associated cardiac abnormalities with transgenic expression of MARF. However, over-expression of MARF does not result in any significant suppression of cardiac defects associated with mutant PolyQ-72 (Figure S5A to S5F). In fact, it has been shown that over-expression of mitofusion 2 promotes cardiomyocyte apoptosis via a mitochondrial death pathway in cultured mammalian cardiomyocytes [36].
We also tested the effects of feeding flies the antioxidant resveratrol. As for SOD-over-expression in PolyQ-72 hearts, resveratrol treatment reduced the dilated systolic and diastolic diameters, the diastolic and systolic intervals, and the arrhythmias (Figure 7A, 7B, 7F). It increased contractility and significantly reduced the PolyQ-72-induced increase in aggregate size and density (Figure 7C, 7I, 7K, 7L). These data indicate that the cardiac defects seen in response to expression of disease-causing PolyQ can be partially suppressed upon over-expression of antioxidant agents, such as SOD or resveratrol.
Since over-expression of SOD or feeding with the antioxidant resveratrol rescued mutant PolyQ-induced cardiac defects, we examined whether over-expression of SOD could rescue the mitochondrial defects associated with expression of mutant PolyQ-46. In contrast to 4 week-old PolyQ-46 hearts, which contained areas of myofibrillar degeneration (Figure 8A, arrow), the majority of myofibrils in PolyQ-46 hearts overexpressing SOD were intact (Figure 8B, arrow). Hearts expressing PolyQ-46 contained fragmented mitochondria (Figure 8A, A' asterisks), while PolyQ-46 hearts overexpressing SOD contained normally shaped mitochondria with densely packed cristae (Figure 8 B, B'), similar to PolyQ-25 controls (Figure 4).
It is known that accumulation of mutant PolyQ interferes with the protein folding machinery in neurons and it has been predicted that PolyQ has the same effect in the heart [37], [38]. Therefore, we reasoned that over-expression of the chaperone UNC-45, which may enhance proper protein folding, might improve cardiac function in hearts compromised by disease-causing PolyQ expression. To address this we over-expressed UNC-45 along with PolyQ-72 in the fly heart. Indeed, transgenic over-expression of UNC-45 completely suppressed mutant PolyQ-72 induced cardiac dilation (compare to Hand/+ wild-type controls, Figure 7A, 7B). UNC-45 over-expression improved contractility (Figure 7C) and the regularity of the heart rhythm (Figure 7F). Most interestingly, over-expression of UNC-45 in the presence of PolyQ-72 significantly reduced the density of mutant PolyQ aggregates compared to hearts expressing PolyQ-72 alone (Figure 7G, 7J, 7L, 7M, compare Figure 7G with 7J and compare lane 1 and 4 in Figure 7M). However, the mean aggregate size was not altered (Figure 7K). Additionally, hearts overexpressing UNC-45 showed a slightly more wild-type organization of actin-containing myofibrils (Figure 7J). Over-expression of UNC-45 in the presence of PolyQ-72 also resulted in a restoration some of the normal structure and content of the myosin-containing myofibrillar network (Figure S6B), which were nearly absent in the cardiomyocytes upon expression of PolyQ-72 alone (Figure S6A). These results suggest that disease-causing PolyQ may act by interfering with chaperone function, which is required for proper myosin folding/accumulation [28], [39]. In contrast, over-expression of UNC-45 alone or with PolyQ-25 resulted in only minor changes in functional cardiac parameters (Figure S7A to S7F). These results suggest that protein unfolding may play a role in mediating PolyQ-induced cardiomyopathy.
We examined whether improving protein folding and oxidative stress pathways might interact to suppress PolyQ-induced cardiac defects. To test this, we co-expressed UNC-45 and SOD-1 in conjunction with PolyQ-72. Co-expression of UNC-45 and SOD-1 restored cardiac contractility (Figure 9C) and suppressed cardiac dilation (Figure 9A, 9B), as well as cardiac arrhythmias (Figure 9F, and Movie S2). Over-expression of UNC-45 and SOD-1 also nearly completely suppressed the formation of GFP-positive aggregates that were dramatically induced by expression of PolyQ-72 (Figure 9H, 9K, 9P, 9Q). Furthermore, PolyQ-72 hearts expressing both UNC-45 and SOD-1 exhibited more organized actin-containing myofibrillar structures (Figure 9G vs. 9J). Together this suggests that mutant PolyQ aggregates induced by abnormal protein folding and increased oxidative stress are linked to cardiac physiological and structural defects.
To further confirm an association of both protein-unfolding and oxidative stress pathways with PolyQ-induced cardiomyopathy, we co-expressed PolyQ-72 and UNC-45 in the presence of the antioxidant resveratrol (Figure 9). Over-expression of UNC-45 in the presence of resveratrol almost completely suppressed PolyQ-induced cardiac dilation, cardiac arrhythmia and amyloid aggregation (Figure 9A, 9B, 9F, 9N, 9P, 9Q). Furthermore, the cardiac contractility was improved as was myofibrillar organization compared to PolyQ-72 expression alone (Figure 9C, 9M). Over-expression of UNC-45 and SOD or over-expression of UNC-45 in the presence of resveratrol also reduced the diastolic and systolic intervals to wild-type levels compared to age-matched PolyQ-72 (Figure 9D, 9E). A summary of cardiac parameters for wild-type controls as well as hearts expressing PolyQ-25 and PolyQ-72 (with or without antioxidant treatment) is shown in Figure S8A. Finally, mutant PolyQ-induced lifespan reduction was rescued by transgenic over-expression of SOD or co-expression of SOD and UNC-45 but not with UNC-45 over-expression (Figure S8B and Table S1). Overall, the genetic interactions that we have identified in this study demonstrate that the protein-misfolding and oxidative stress pathways induced by accumulation of HD-causing PolyQ aggregates are linked and associated with cardiac dysfunction.
Huntingtin protein is expressed in many tissues including the heart and epidemiological studies suggest that HD patients have a higher susceptibility to cardiac failure compared to age-matched controls without HD [8], [9], [11], [16], [17]. However, the cellular mechanisms underlying the cardiac dysfunction in HD have yet to be studied in the heart. Using the genetically tractable model system Drosophila, we now show a direct correlation between the levels of amyloid accumulation, overall ROS production and the severity of cardiac dysfunction. Cardiac-specific expression of disease-causing Htt-PolyQ (PolyQ-46, PolyQ-72 and PolyQ-102) all elicited cardiac dysfunction compared to hearts expressing the non-disease-causing PolyQ-25. In addition the qualitative as well as quantitative defects that we observed in response to PolyQ expression were dose-dependent. Since mutant Htt-PolyQ protein was expressed specifically in the heart, it is unlikely that our observations reflect a neuronal contribution to these cardiac defects. Our data suggest that the increased risk of cardiac disease in HD patients is possibly due to cardiac amyloid accumulation, mitochondrial defects as well as oxidative stress and that the severity of disease depends upon the length of the PolyQ repeat (Figures 3–5).
Our data also demonstrate that the likely cause of the observed functional defects is the severe myofibrillar disorganization and reduced myosin and actin content in myocardial cells resulting from cardiac-specific expression of disease causing PolyQ (Figure 3C and 3F). Recently we showed that the chaperone UNC-45 is required for preserving myosin accumulation/folding in Drosophila cardiomyocytes, as its reduction leads to severe disorganization of myosin-actin containing myofibrils and thus sarcomeres [28]. The current results extend this observation and are the first demonstration of a role for UNC-45 in amyloidosis-induced cardiac defects. In support of this hypothesis, it has previously been shown that nuclear or cytoplasmic aggregates (inclusion bodies) of polyglutamine proteins contain chaperones involved in protein folding [11], [37]. Furthermore, and consistent with our results, over-expression of the chaperone αB-crystallin reduces PolyQ-induced aggregation in rat neonatal cardiomyocytes; however, over-expression of αB-crystallin enhances amyloid oligomer formation and toxicity [23]. In the present study co-over-expression of UNC-45 with disease-causing PolyQ-72 dramatically reduced amyloid aggregate density (Figure 7L and 7M) and ameliorated cardiac dysfunction by decreasing the incidence of cardiac arrhythmia, suppressing the mutant-Htt-induced cardiac dilation (Figure 7A, 7B) and improving cardiac contractility to a dramatic extent (Figure 7C). Importantly, over-expression of UNC-45 in the presence of PolyQ-72 restored myosin-containing myofibrils (Figure S6), suggesting that one effect of amyloid aggregation is to interfere with proper folding of muscle myosin in cardiomyocytes.
The fact that UNC-45 over-expression did not completely suppress the mutant Htt-PolyQ-induced cardiac physiological defects and lifespan reduction is consistent with the idea that amyloid accumulation affects additional cellular pathways that result in cardiac abnormalities. As reported for αB-crystallin, suppression of aggregates is not sufficient to reduce toxicity [23] and this possibility may also exist in the case of UNC-45. It is also possible that the overall high level of oxidative stress produced by mutant PolyQ is the main determinant for lethality. Expression of mutant PolyQ leads to mitochondrial defects due to increased oxidative stress [4]–[10], [40]–[42]. Several neuronal studies have shown that expression of mutant polyQ affects SOD expression [43]–[46]. Manipulation of SOD seems to be directly correlated with levels of oxidative stress in several neurodegenerative diseases [4]–[10], [40]–[42]. Additionally, SOD over-expression reduces diabetic cardiomyopathy and some forms of neurodegeneration by reducing oxidative stress [47], [48]. However, neither UNC-45 nor SOD has been shown previously to suppress the PolyQ-induced phenotypes in either neuronal or cardiac animal disease models. Our data also support a role for oxidative stress pathways in amyloid-induced cardiac dysfunction and lethality. Treatment with oxidants aggravated the moderate effects of PolyQ-46 on heart function, causing an increase in amyloid aggregate density and more severe cardiac defects (Figure 6). This suggests a possibly causal relationship between oxidative stress, the formation of aggregates and cardiac dysfunction. Furthermore, our ultrastructural analysis clearly shows mutant PolyQ-induced mitochondrial defects, while DHE staining indicates that excess ROS production occurs upon expression of mutant PolyQ (Figures 4 and 5). Interestingly, some of the PolyQ aggregates co-localize with concentrated DHE staining (Figure 5). Significantly, we were able to reduce the size and density of mutant PolyQ-aggregates as well as the severity of the PolyQ-72-induced cardiac defects by over-expression of SOD or by feeding the anti-oxidant resveratrol (Figure 7). This is consistent with findings that resveratrol provides protection in neuronal models of Huntington's disease [49]–[54]. Interestingly, the anti-oxidant resveratrol has been shown to affect expression of anti-oxidative enzymes, including enhanced expression of SOD-1 [49]–[54].
Expression of mutant PolyQ may both induce oxidative stress and interfere with protein folding pathways [4], [21], [40]–[42]. A study using cultured mouse neurons showed that oxidative stress increases PolyQ aggregation and that over-expression of SOD1 in conjunction with the chaperone HSP-70/HSP-40 could suppress Htt-polyQ-induced aggregation and toxicity [42]. However, simultaneous manipulation of both of these genetic pathways has not previously been attempted in vivo. In addition to neurons, expression of the mutated Htt protein or expression of pre-amyloid oligomers cause cardiac defects by affecting several pathways including oxidative stress, mitochondrial abnormalities, presence of protein aggregates and increased autophagosomal content [10], [21], [30], [31], [34]. However, no attempt had thus far been made to suppress PolyQ-induced cardiac defects, a crucial step for understanding the mechanistic basis of disease progression and amelioration. Indeed, in our in vivo cardiac model, co-expression of UNC-45 and SOD-1 or expression of UNC-45 in the presence of resveratrol had a tendency to suppress the PolyQ-72-induced amyloid aggregation and concomitant cardiac dilation more efficiently than either treatment alone (Figures 7 and 9). Thus, our results suggest that suppression of both protein aggregates and ROS may be required for the amelioration of PolyQ-induced cardiomyopathy. As HD is primarily a neurological disease, the effect of such suppression is worth exploring in neural tissues.
In addition to interfering with protein folding pathways, expression of mutant PolyQ may lead to myofibril loss by directly interacting with muscle proteins. Previous studies have suggested that mutant PolyQ may bind directly to contractile proteins and disturb their function [55], [56]. Integrity of contractile proteins is also required for maintaining mitochondrial organization and cardiomyocyte function [35], [57]–[60]. Additionally, expression of aggregation-prone mutant PolyQ may induce oxidative stress due to mitochondrial damage in the cardiomyocytes, which are heavily dependent on mitochondrial function and are vulnerable to oxidative stress [60]–[62]. For example, knockdown of SOD results in mitochondrial defects and severe dilated cardiomyopathy phenotype in a mouse model [63]. Our results do show a dramatic increase in overall ROS levels in mutant PolyQ expressing hearts. Moreover, the GFP-positive PolyQ aggregates co-localize with areas of strong DHE staining and the observation that antioxidant treatments partially rescue the cardiac defects further support this hypothesis.
Overall, accumulation of amyloid in the cardiomyocytes can induce mechanical deficits by affecting the integrity of contractile proteins as well as mitochondria and lead to cardiomyocyte death, possibly through activation of autophagy. Consistent with our finding, a similar mechanism has been proposed for cardiomyopathy associated with amyloid producing mutant αB-crystallin [58]–[60], [64]. Both mutant αB-crystallin and mutant PolyQ caused aggregate formation in cardiomyocytes suggesting a common mechanism for underlying cardiomyocyte degeneration [21], [23], [58]–[60], [64]. It is unclear at this point whether the presence of toxic aggregates in cardiomyocytes is directly interfering with mitochondrial organization leading to cardiac defects or whether oxidative stress produced by mutant PolyQ leads to mitochondrial dysfunction that triggers cardiomyocyte dysfunction.
A full understanding of all the molecular details involved in mutant PolyQ induced cardiomyopathy will require additional study but we have now identified some of the key players and interactions in vivo. Although dense granular deposits, immunoreactive to an anti-Huntingtin antibody, have been found in muscle tissue from an HD patient, no such study has been performed on HD heart biopsy samples [65]. Thus, our study suggests that it would be useful to look for accumulation of amyloid protein in the hearts of HD patients, especially those with heart disease. Delineating how these aggregates might be toxic to cells will be critical not only for an understanding of PolyQ-induced cardiomyopathy but also for gaining insights into aggregation-based neural degeneration. The Drosophila heart model provides a genetically tractable system whereby these interactions can be examined in the context of a functioning organ. Indeed, elucidating the genetics underlying PolyQ-induced cardiomyopathy should also have an impact on our understanding of other cardiac diseases associated with oxidative stress, mitochondrial dysfunction, the unfolded protein response and proteostasis in general.
To evaluate cardiac function following PolyQ-induced cardiomyopathy, we obtained Drosophila transgenic lines expressing enhanced-GFP-tagged mutant Htt fragments (UAS-Httex1-QneGFP) with different PolyQ lengths (Q25, Q46, Q72, and Q102) [5]. Heart-specific expression was achieved with a UAS-Gal4 system using the Hand driver [66], [67], crossed to the different UAS-Httex-GFP lines (Httex1-Q25-eGFP, Httex1-Q46-eGFP, Httex1-Q72-eGFP and Httex1-Q102-eGFP). For simplicity Httex1-Q25-eGFP, Httex1-Q46-eGFP, Httex1-Q72-eGFP and Httex1-Q102-eGFP are referred to as PolyQ-25, PolyQ-46, PolyQ-72 and PolyQ-102, respectively in this study. F-1 progeny (i.e. Hand-Gal4/+, Hand-Gal4>UAS-PolyQ-25, Hand-Gal4>UAS-PolyQ-46, Hand-Gal4>UAS-PolyQ-72 or Hand-Gal4>UAS-PolyQ-102) of each transgenic cross were collected, separated by sex and cultured at 25°C. Lifespan of female progeny was determined with survivorship being monitored every third day with a food change as previously described [28]. Adult flies were analyzed at 1 and 3 weeks of age. The cardiac tissue-specific Hand-Gal4 driver was gift from Eric Olsen [68]. Transgenic unc-45, SOD and MARF lines were generated as previously described [69]–[71].
Semi-intact hearts were prepared as described previously [29], [72]. Direct immersion optics were used in conjunction with a digital high-speed camera (up to 200 frame/sec, Hamamatsu EM-CCD) to record 30 s movies of beating hearts; images were captured using HC Image (Hamamatsu Corp.). Cardiac function was analyzed from the high speed movies using semi-automatic optical heartbeat analysis software (a MatLab-based image analysis software) which quantifies heart period, diastolic and systolic diameters, diastolic and systolic intervals, cardiac rhythmicity, fractional shortening and produced the M-mode records [29], [72].
Dissected hearts (from 1 and 3 week old flies) were briefly exposed to 10 mM EGTA and then fixed with 4% paraformaldehyde in PBS as previously described [73]. Fixed hearts were probed with myosin antibody followed by goat-anti-rabbit IgG-Cy5 (Chemicon, Temecula, CA) and Alexa555-phalloidiin (Invitrogen, Carlsbad, CA) to stain F-actin. Fluorescence imaging of Drosophila heart tubes was carried out using an Apotome Imager Z1 (Zeiss) and an AxioCam MRm (Zeiss) as previously described [73], [74]. To detect extended PolyQ-induced aggregates in the fly heart, we used Htt-PolyQ-GFP [5] in conjunction with anti-myosin or phalloidin.
Dihydroethidium (DHE) and LysoTracker were employed for the detection of oxidative stress and autophagosomes/lysosomes respectively using a modified protocol previously described for use in other tissue [71], [75]. Briefly, semi-intact hearts were prepared as described above and stained with DHE (Molecular Probes, Carlsbad, CA) at 2 µM final concentrations in artificial hemolymph for 30 min, followed by three washes with artificial hemolymph. Hearts were relaxed with 10 mM EGTA and mounted in Vectashield. A similar staining procedure was applied for the staining with LysoTracker red (Molecular Probes, Carlsbad, CA): 30 min, 1 µM final concentration in artificial hemolymph. Fluorescence imaging was carried out using an Apotome Imager Z1 (Zeiss) and an AxioCam MRm (Zeiss) as previously described [73], [74]. DHE intensity was quantified using ImageJ software.
Semi-intact heart preparations were prepared for transmission electron microscopy using a modified protocol described previously [76]. Briefly, hearts were relaxed with 10 mM EGTA followed by a primary fixation protocol (3% formaldehyde, 3% glutaraldehyde in 0.1 M cacodylate buffer, pH 7.4) and secondary fixation (1% OsO4, 100 mM phosphate buffer, and 10 mM MgCl2, pH 7.4). The samples were block stained in 2% uranyl acetate and dehydrated with an acetone series, followed by orientation and embedding in Epon-filled BEEM capsules. Polymerization was performed at 60°C under vacuum. Thin sections (50 nm) were cut using a Diatome diamond knife on a Leica ultramicrotome and picked up on formvar-coated grids. Slices were stained with 2% uranyl acetate for 10 min and Sato's lead stain [77] for 2 min. Images were obtained at 120 kV on a FEI Tecnai 12 transmission electron microscope.
We employed standard genetic and transgenic techniques [69] to co-express chaperones or SOD in flies expressing UAS-PolyQ in the heart using the Gal4 driver. Genetic crosses using multiple balancers were carried out with transgenic flies expressing unc-45, SOD-1 or SOD-2. Briefly, the wild-type genomic dunc-45 was used as previously described [69]. To ameliorate PolyQ-72 induced cardiac defects, adult males homozygous for the PolyQ-72 (w1118/Y; +/+; PolyQ-72/PolyQ-72, flies were crossed with female flies homozygous for the wild-type dunc-45 transgene on the 1st chromosome P[w+, dunc-45/P[w+, dunc-45]; Hand-Gal4/Cyo; +/+ [69]. The following progeny (P[w+, dunc-45]/w1118; Hand-Gal4/+; PolyQ-72/+) were analyzed to determine ability of UNC-45 over-expression to suppress Poly-Q induced cardiac defects. A similar genetic suppression approach was used with the wild-type dunc-45 transgene on the 2nd chromosome after crossing w1118/w1118; P[w+, dunc-45]/P[w+, dunc-45]; +/+ to w1118/Y; Cyo/Hand-Gal4; PolyQ-72/PolyQ-72 to obtain w1118/w1118; P[w+, dunc-45]/Hand; +/PolyQ-72. For expression of SOD-1 (2nd chromosome), w1118/w1118; P[w+, SOD-1]/P[w+, SOD-1]; +/+ flies were crossed to w1118/Y; Cyo/Hand-Gal4; PolyQ-72/PolyQ-72 to obtain w1118/w1118; P[w+, SOD-1]/Hand; +/PolyQ-72. Finally, for co-expression of unc-45, SOD-1 in PolyQ-72 expressing flies, females homozygous for the wild-type dunc-45 transgene on the 1st chromosome P[w+, dunc-45]/P[w+, dunc-45]; Hand-Gal4/Cyo; +/+ were crossed to w1118/Y; P[w+, SOD-1]/Cyo; PolyQ-72/PolyQ-72 to obtain P[w+, dunc-45]/w1118; Hand-Gal4/SOD-1; PolyQ-72/+. Similar genetic approaches were used to co-express UNC-45 or SOD in flies expressing UAS-Poly-25 in the heart using the Gal4 driver. To study the effects of MARF over-expression on mutant PolyQ-induce cardiac defects, flies with MARF overexpressing transgenes on their second or third chromosomes were crossed with extant stocks (w1118/Y; Cyo/Hand-Gal4; PolyQ-72/PolyQ-72) to obtain w1118/w1118; P[w+, MARF]/Hand; +/PolyQ-72 or w1118/w1118; +/Hand; P[w+, MARF]/PolyQ-72. We did not see any difference in the time to eclosion or the number of progeny between the mutant PolyQ and cardiac-specific SOD1/UNC-45/MARF overexpression lines.
For treatment with resveratrol, flies expressing Hand-Gal4>UAS-PolyQ (control or disease causing), Hand-Gal4; w1118 or (P[w+, dunc-45]/w1118; Hand-Gal4/+; PolyQ-72/+) were raised separately in standard media in the presence or absence of resveratrol (final concentration, 1 mg/ml, a dose previously used in the Drosophila model [48]). Flies were collected on eclosion, separated by sex and cultured at 25°C; food was changed every 3 days. Adult flies were analyzed at 3 weeks of age. Additionally, we did not see any difference in the time to eclosion or the number of progeny between the mutant PolyQ and resveratrol fed organisms (concentration 1 mg/ml). For treatment with H2O2, flies were collected on eclosion and separated by sex. One group was cultured on food containing 1% H2O2 and a second group received standard food. Food was changed every 3 days and adult flies were analyzed at 3 weeks of age.
Semi-intact hearts from flies expressing various length Htt-PolyQ-GFP [5] were prepared and fixation was carried out at described above. The numbers of aggregates in micrographs taken using the GFP wavelength (488 nm) were quantified using ImageJ software (Particle Count and Analysis function). Briefly, each immunofluorescence micrograph was divided into 50×50 µM square boxes (2500 µm2 unit area) and the total number of aggregates within each box was quantified. Between three and four boxes per heart were analyzed and the number of aggregates per box was averaged for each heart. We also quantified the 2D surface area of the aggregates within the box using ImageJ software. Thus, we quantified mutant Htt-polyQ deposits in terms of the total number of aggregates per unit area as well as in terms of the size of aggregates from at least four hearts per genotype. For each heart aggregate density/size was determined for each of three 50×50 micron regions and averaged and four to six hearts were examined for each genotype/treatment.
We used a filter trap assay for the quantification of Htt-polyQ GFP aggregates in the Drosophila heart as previously described [30]–[32]. Briefly, hearts were dissected (30–50/genotype) and harvested in SDS lysis buffer (20 mM Tris-HCl, pH 7.5, 200 mM NaCl, 2% SDS); samples were diluted to 1 µg/100 µl with TBS (20 mM Tris-HCl, pH 7.5, 500 mM NaCl) and 500 ng were loaded onto a cellulose acetate membrane (0.2 µm pore size, Whatman, Piscataway, NJ), using the 96-well BioDot Apparatus (Bio-Rad, Hercules, CA). Protein content of homogenized heart samples was determined using the DC Protein Assay Kit II (Bio-Rad, Hercules, CA). Non-specific binding sites were blocked using 5% nonfat milk in TBA buffer for 2 hours. Immunodetection was performed after incubation with mouse anti-GFP (Covance Research Products, Dedham, MA) and secondary antibodies (KPL, Inc., Gaithersburg, MD) using Thermo Scientific SuperSignal West Dura Substrate on a Bio-Rad ChemiDoc XRS System. A parallel series of 500 ng samples was blotted onto nitrocellulose (Bio-Rad, Hercules, CA) to confirm GFP expression in non-aggregating PolyQ-GFP, using H2B antibody (Cell Signaling, Danvers, MA) as a loading control. Quantification of the immunoblot's density was carried out using ImageJ software.
For all quantitation except lifespan analysis, statistical significance was determined using one-way analysis of variance (ANOVA) followed by Dunnett's post-hoc test to determine significance between groups with Prism 6.0 (Graph Pad) software. Significant differences were assumed for p<0.05. For lifespan studies, data were analyzed using the Gehan-Breslow-Wicoxon test followed by multiple comparisons between control and experimental groups. Significance was taken at p values less than the Bonferroni-corrected threshold of p<0.0125.
|
10.1371/journal.ppat.1007953 | Novel lactate dehydrogenase inhibitors with in vivo efficacy against Cryptosporidium parvum | Cryptosporidium parvum is a highly prevalent zoonotic and anthroponotic protozoan parasite that causes a diarrheal syndrome in children and neonatal livestock, culminating in growth retardation and mortalities. Despite the high prevalence of C. parvum, there are no fully effective and safe drugs for treating infections, and there is no vaccine. We have previously reported that the bacterial-like C. parvum lactate dehydrogenase (CpLDH) enzyme is essential for survival, virulence and growth of C. parvum in vitro and in vivo. In the present study, we screened compound libraries and identified inhibitors against the enzymatic activity of recombinant CpLDH protein in vitro. We tested the inhibitors for anti-Cryptosporidium effect using in vitro infection assays of HCT-8 cells monolayers and identified compounds NSC158011 and NSC10447 that inhibited the proliferation of intracellular C. parvum in vitro, with IC50 values of 14.88 and 72.65 μM, respectively. At doses tolerable in mice, we found that both NSC158011 and NSC10447 consistently significantly reduced the shedding of C. parvum oocysts in infected immunocompromised mice’s feces, and prevented intestinal villous atrophy as well as mucosal erosion due to C. parvum. Together, our findings have unveiled promising anti-Cryptosporidium drug candidates that can be explored further for the development of the much needed novel therapeutic agents against C. parvum infections.
| Cryptosporidium parvum is a protozoan parasite that can cause a life-threatening gastrointestinal disease in children and in immunocompromised adults. The only approved drug for treatment of Cryptosporidium infections in humans is nitazoxanide, but it is not effective in immunocompromised individuals or in children with malnutrition. C. parvum possesses a unique lactate dehydrogenase (CpLDH) enzyme that it uses for generating metabolic energy (ATP) via the glycolytic pathway to fuel its growth and proliferation in the host. We have identified novel inhibitors for the enzymatic activity of CpLDH. Further, we have demonstrated that two of the CpLDH inhibitors effectively block the growth, proliferation and pathogenicity of C. parvum at tolerable doses in immunocompromised mice. Together, our findings have unveiled novel CpLDH inhibitors that can be explored for the development of efficacious therapeutic drugs against C. parvum infections.
| The zoonotic and anthroponotic protozoan parasite, Cryptosporidium parvum, is a major cause of diarrheal diseases in children under the age of two, resulting in significant morbidity and mortality in poor-resource areas of developing countries [1]. In livestock, particularly in calves, lambs and goat kids, it causes a serious diarrheal syndrome, culminating in growth retardation and high neonatal mortalities [2–4]. C. parvum is highly prevalent because of its enormous capacity to reproduce in infected livestock, resulting in large amounts of infective parasite oocysts being shed in animal feces, and contaminating water sources as well as the general environment. The parasite oocysts in the environment are difficult to eliminate because of their resistance to virtually all kinds of chemical disinfectants, as well as to commonly used water treatments such as chlorination [5]. The efficacy of the only FDA-approved anti-Cryptosporidium drug in humans, nitazoxanide, is modest. Of particular concern, nitazoxanide is ineffective in those individuals most at risk for morbidity and mortality due to Cryptosporidium infections, including malnourished children and immunocompromised individuals [6]. There is currently no vaccine against Cryptosporidium infections.
Efforts to develop fully effective drugs against Cryptosporidium have largely been hampered by the lack of genetic tools for functional interrogation and validation of potential molecular drug targets in the parasite. Recently, however, a CRISPR/Cas9 gene manipulation approach [7], and a morpholino-based targeted gene knockdown approach [8, 9] in C. parvum have been developed. The completed and annotated genome sequence of Cryptosporidium indicates that, while the parasite lacks genes for conventional molecular drug targets found in other important protozoan parasites, it has several genes encoding plant-like and bacterial-like enzymes that catalyze potentially essential biosynthetic and metabolic pathways in Cryptosporidium [10]. Using a morpholino-based approach for targeted gene knockdown in C. parvum, we have previously validated that the C. parvum lactate dehydrogenase gene (CpLDH) that encodes a bacterial-like enzyme, is essential for survival, virulence and reproduction of C. parvum both in vitro and in vivo [8, 9].
In the present study, we screened compound libraries and identified compounds with inhibitory effect against the enzymatic activity of recombinant CpLDH protein in vitro. Among the identified CpLDH inhibitors, we have demonstrated that two of the inhibitors can effectively block the growth, proliferation and pathogenicity of C. parvum in vivo at tolerable doses, suggesting that they are potential candidates for development of drugs against C. parvum infections.
By sequencing, the cloned open reading frame of CpLDH gene was verified to be 966 bp long, and 99.79% identical to that reported in the genome database (GenBank accession number AF274310.1). It coded for a 321 amino acids long protein with amino acid residue substitutions of F-198-L, R-251-K and K-295-E when compared to that in GenBank. The expressed and purified His-tagged CpLDH protein was of the expected molecular size of about 34 kDa (Fig 1A). By analyzing the in vitro catalytic activities of recombinant CpLDH, we found that it depicted more activity in catalyzing the reduction of pyruvate to lactate than the oxidation of lactate to pyruvate. We found that recombinant CpLDH enzymatic catalytic activity was consistent with the Michaelis-Menten kinetics on pyruvate, NADH, lactate and NAD+ (Fig 1B–1E). The Lineweaver–Burk representation of the saturation curves (insets in Fig 1B–1E) showed that the Km of recombinant CpLDH for pyruvate was at least 54-fold lower than that for lactate, while its Vmax for pyruvate was 123-fold higher than that for lactate (Table 1). Our obtained enzymatic kinetic parameters for recombinant CpLDH in comparison to those reported previously for C. parvum CpLDH [11] are summarized in Table 1.
We found that recombinant CpLDH had more catalytic activity for the reduction of pyruvate to lactate than for the oxidation of lactate to pyruvate. Therefore, we used the assay for reduction of pyruvate to lactate to screen chemical compounds for inhibitors of the enzymatic activity of recombinant CpLDH in vitro. Within the group of the 27 diverse chemical compounds (S1 Table) [12], we identified three compounds (NSC22225, NSC37031 and NSC158011) that significantly (P < 0.05) inhibited the catalytic activity of recombinant CpLDH for the reduction of pyruvate to lactate (Fig 2). On the other hand, among the 800 compounds in the Mechanistic Set IV (S2 Table), we found 20 that had significant (P < 0.05) inhibitory effect on the catalytic activity of recombinant CpLDH (Fig 3). Those 20 compounds included: NSC51148, NSC1771, NSC626433, NSC349438, NSC10447, NSC85561, NSC73413, NSC657799, NSC686349, NSC638352, NSC34931, NSC253995, NSC70929, NSC70925, NSC56817, NSC79688, NSC18298, NSC71948, NSC22842, NSC33006, and NSC82116. The rest of the compounds either had no effect or augmented the catalytic activity of recombinant CpLDH and were thus not pursued further.
To decipher the differences in the interactions of NSC150811 and NSC10447 (S1 Fig) with CpLDH and human LDH proteins, an in silico molecular docking using Autodock Vina [13] was performed to determine the most energetically favorable poses of the compounds complexed with the rigid structures of both CpLDH and human LDH. Both compounds were found to bind most favorably into the co-factor-binding pocket of the CpLDH and human LDH, where NADH binds to reduce pyruvate to lactate.
NSC158011 complexed with CpLDH with an affinity of -6.4 kcal/mol. (Table 2). The ligand was surrounded in hydrophobic and hydrophilic interactions. Hydrophobic interactions occurred with the Ile-100, Ala-80, and Ile-15 residues and the nonpolar aromatic rings of the molecule, while polar interactions occurred with the Asn-97 and Gln-14 residues and the highly polar thio-amide group (S2 Fig). Although secondary amines do not possess a strong dipole moment, it is possible that the positively-charged Arg-85 residue interacts with the resonance-stabilized deprotonated thio-amide moiety of NSC158011. The docked NSC158011 possessed little solvent exposure, likely due to its folded nature and position, tight within the protein pocket (S2 Fig). NSC158011 complexed with human LDH with an affinity of -7.2 kcal/mol. (Table 2). The hydrophobic Phe-119, Ile-120, Val-116, Val-98, Ala-96, Val-26, and Val-28 residues interacted with the non-polar aromatic rings of the NSC158011 (S3 Fig). Remarkably, these same non-polar aromatic rings are also heavily solvent-exposed (S3 Fig).
NSC10447 complexed with CpLDH with an affinity of -7.6 kcal/mol. (Table 2). The ligand was involved primarily in polar interactions with the surrounding Ser81, Thr-79, Thr-229, Thr-231, and positive-charged Arg-85 residues, mediated through interactions with semi-polar carbonyl carbons abundant on one side of the molecule (S4 Fig). The non-polar side of the molecule interacted favorably with the hydrophobic Tyr-233 residue internal to the protein (S4 Fig). There was a weak hydrogen-bonding interaction between the backbone of Asn-97 and an alcohol group on the ligand. Additionally, the hydrophilic carbons had weak but notable solvent exposure (S4 Fig). NSC10447 complexed with human LDH with an affinity of -7.1 kcal/mol. (Table 2). The ligand was involved primarily in hydrophobic interactions with the surrounding Val-98, Ala-96, Val-94, Phe119, Val-26, Tyr-83, and Val-116 residues that stabilized the non-polar moiety of the molecule (S5 Fig). Interestingly, there was a weak, polar, anti-bonding interaction with the Thr-95 residue. The polar alcohol groups and the attached carbons were heavily solvent exposed in the final docking conformation (S5 Fig).
The results of the molecular docking simulation showed that due to the high level of interaction between the two lead compounds and the residues within the LDH cofactor-binding pocket, NSC158011 and NSC10447 each bound favorably to both CpLDH and human LDH proteins. It can be proposed that the compounds act as competitive inhibitors for the LDH enzyme, binding favorably to the hydrophobic residues internal to the co-factor-binding pocket and blocking the enzyme from binding NADH, thus preventing the hydride transfer that powers the conversion of pyruvate to lactate.
All the compounds that we found to have inhibitory effect against recombinant CpLDH activity were first analyzed for in vitro cytotoxicity in a mammalian cell line, HCT-8 (American Type Culture Collection Item number: CCL244) before testing their anti-Cryptosporidium efficacy. For cytotoxicity screening, varying concentrations of each compound (from 0 to 700 μM) were tested in triplicate using the WST-1 cell proliferation assay and the half maximal inhibitory concentration (cytotoxicity IC50 values) of the compounds in HCT-8 cells (S3 Table) were derived from dose–response curves using GraphPad PRISM software. To test the compounds’ efficacy against C. parvum in vitro, an initial screen was performed using concentrations that were at least 50% lower than the compounds’ respective cytotoxicity IC50 values (S3 Table). NSC10447 and NSC158011 from the diverse group and Mechanistic Set IV group, respectively, were found to significantly (P < 0.05) inhibit C. parvum proliferation in vitro at 48 h post-infection. Therefore, these two compounds were selected for secondary analysis of anti-Cryptosporidium efficacy using varying concentrations and durations of culture to derive the IC50 values for the inhibition of parasite proliferation. For each compound, the assays were done in two formats: (1) by adding the compound to the HCT-8 cells culture shortly before infecting them with C. parvum sporozoites, with the goal to assess whether the compounds would block host cell invasion by sporozoites, and (2) by adding the compounds to the cells 2 h post-infection to determine the effect of the compounds on intracellular parasites. When the cultures were analyzed at 48 h post-infection, compound NSC158011 was found to have a significant (P < 0.05) concentration-dependent effect of inhibiting proliferation of intracellular C. parvum merozoites in HCT-8 cells starting at 10 μM (with 40 μM blocking parasite growth almost completely) relative to the control infected cultures without compound treatment (Fig 4A). Treating the cultures with NSC158011 compound 2 h post-infection also resulted in a concentration-dependent reduction in parasite proliferation, but with a slight decrease in compound efficacy relative to treating at the time of infection (Fig 4A). By using GraphPad PRISM software, the half maximal inhibitory concentration (IC50) values of NSC158011 for C. parvum in vitro were derived from the dose–response curves. The NSC158011 IC50 values at 48 h post-infection for inhibition of C. parvum growth when compound was added immediately, or 2 h after infecting the HCT-8 cells were 14.88 and 15.81 μM, respectively. Analysis of the inhibitory effect of NSC158011 on parasite proliferation at 72 h post-infection, depicted similar concentration-dependent effects (Fig 4B), with IC50 values of 15.63 and 16.50 μM when compound was added immediately or 2 h post-infection, respectively. After infecting HCT-8 cells, C. parvum is able to proliferate for 3–4 days before becoming growth-arrested. Therefore, during both observation time-points of 48 h and 72 h post-infection, C. parvum if untreated was expected to be in proliferative phase. Consistently, in the untreated infected cells, the relative parasite load at 72 h post-infection was about 2-fold that observed at 48 h post-infection (Fig 4A and 4B).
Compound NSC10447 also depicted a concentration-dependent inhibitory effect on parasite growth, both at 48 h (Fig 5A) and 72 h (Fig 5B) time points of observation. The efficacy of NSC10447 when added immediately or 2 h after infecting the cells was similar (Fig 5A). The NSC10447 IC50 values at 48 h post-infection for inhibition of C. parvum growth when compound was added immediately, and 2 h after infecting the HCT-8 cells were 72.65 and 79.52 μM, respectively. Consistently, NSC10447 had a concentration-dependent inhibitory effect on parasite growth at 72 h time point of observation (Fig 5B), with IC50 values of 83.63 and 95.17 μM, when the compound was added immediately, and 2 h after infecting the HCT-8 cells, respectively. Noteworthy, in all instances, NSC158011 depicted significantly higher in vitro efficacy against C. parvum than NSC10447 (Table 3). The IC50 values of NSC158011 and NSC10447 for the inhibition of the catalytic activity of recombinant CpLDH in vitro were 76.59 μM and 46.33 μM, respectively (Table 3).
We used paromomycin as the positive control treatment. Using in vitro assays the cytotoxicity of paromomycin in HCT-8 cells has been reported to be negligible even when used at concentrations above 1000 μM [14, 15]. Therefore, we tested the in vitro efficacy of paromomycin against C. parvum at varying concentrations up to a maximum of 1000 μM, and found it to have a concentration-dependent effect of inhibiting C. parvum growth in vitro, both at 48 h (Fig 6A) and 72 h (Fig 6B) post-infection, with IC50 values of 450 and 400 μM, respectively. Others have previously reported paromomycin to have an IC50 of 711 μM for inhibition of C. parvum growth in HCT-8 cells [15]. There was no notable significant difference in paromomycin inhibitory effect against C. parvum between starting the treatment immediately or 2 h after infection of the HCT-8 cells.
Compound NSC158011 and NSC10447 that were found to inhibit C. parvum growth in vitro were selected for in vivo testing using a mouse infection model. Prior to use in mice, the highest tolerable doses in mice for the two compounds were found to be 400 mg/kg and 1000 mg/kg for NSC158011 and NSC10447, respectively. These doses consistently did not induce any toxicity signs (changes from normal physical activity, respiration, body temperature, feeding pattern, body posture, fur condition or occurrence of death) over 7 days of daily oral gavage in mice. In the case of NSC158011, the dose of 400 mg/kg intraperitoneal administration in mice has also been previously shown not to be toxic to mice [16]. Thus, the doses of 400 mg/kg and 1000 mg/kg were selected as the highest doses for testing the efficacy of NSC158011 and NSC10447, respectively, against C. parvum growth and proliferation in mice. Paromomycin at 100 mg/kg daily by oral gavage was used as a positive control. The daily load of C. parvum oocysts shed in mice feces was determined using real time PCR quantification of the C. parvum 18s rRNA gene. As expected, in the feces of untreated infected mice, C. parvum genomic DNA was almost undetectable during the first 2 days post-infection, but was detectable from 3 days post-infection, and increased progressively with increase in the number of days post-infection (Tables 4–6). The notable lower C. parvum DNA in the feces of the untreated infected mice in Table 6 when compared to those for Tables 4 and 5 is because the infection assays for Tables 3 and 4 were done using freshly purified oocysts, while those for Table 6 were done using oocysts that were purified 3 months earlier, and thus their infectivity could have been lower. We found that NSC158011 at 400 mg/kg significantly (P < 0.05) reduced shedding of C. parvum oocysts in mice feces, comparable to the efficacy of paromomycin (Table 4).
As expected, C. parvum DNA was consistently undetectable at all time points sampled in the uninfected mice (Table 4). By day 7 post-infection, both NSC158011 and paromomycin treatment had reduced the shedding of C. parvum in mice feces by about 3000-fold when compared to the untreated infected mice samples (Table 4). This suggested that NSC158011, at 400 mg/kg, had sustained anti-Cryptosporidium efficacy in vivo comparable to that of 100 mg/kg of paromomycin. We titrated the dose of NSC158011 to determine the effect of lower dosages. We observed a dose-dependent reduction in efficacy of NSC158011, with 200 mg/kg having about 10-fold lower efficacy than paromomycin during the last three days of treatment. Consistently, 100 mg/kg of NSC158011 showed lower efficacy than 200 mg/kg NSC158011 (Table 5).
For testing the in vivo anti-Cryptosporidium efficacy of NSC10447, in addition to testing the highest tolerable dose of 1000 mg/kg, we also tested lower doses of 250 mg/kg and 500 mg/kg. While the C. parvum DNA was undetectable in the uninfected mice’s feces, the untreated infected mice had readily detectable C. parvum DNA by day 3 post-infection, that then increased progressively with increase in number of days post-infection (Table 6).
From day 3 until day 8 of treatment, compared to the infected untreated, mice treated with NSC10447 at 250, 500 and 1000 mg/kg showed sustained significantly lower (by at least 50%) C. parvum DNA load in their feces (Table 6). There was a notable dose-dependent effect, with 1000 mg/kg having the highest efficacy (Table 6). Notably, on day 7 and 8 post-infection, the 1000 mg/kg dose of NSC10447 maintained anti-Cryptosporidium efficacy that was comparable to that of paromomycin, while both 250 and 500 mg/kg doses depicted lower efficacies than paromomycin (Table 6).
During C. parvum infection, usually the distal small intestines are severely affected, characterized by villous atrophy, erosion and ulceration of the intestinal mucosa. Thus, we performed histopathological examination of the distal small intestines of the experimental mice at 9 days post-infection. As expected, while uninfected mice maintained the integrity of the intestinal mucosa (Fig 7A and Fig 8A), infected untreated mice had microscopic lesions characterized by villous atrophy and mucosal erosion (Figs 7B and 8B). Infected mice treated with NSC158011 maintained intact intestinal mucosa and villi (Fig 7D) whose integrity was similar to that of mice treated with paromomycin (Fig 7C). Likewise, mice treated with NSC10447 also prevented villous atrophy and maintained the integrity of the intestinal mucosa (8D-F) similar to treatment with paromomycin (Fig 8C). We enumerated the mean percentage of denuded intestinal villi in 4 randomly chosen microscopic fields per sample from representative histopathology images. We observed that in infected mice, just like treatment with paromomycin, treatment with NSC158011 and NSC10447 reduced the percentage of denuded intestinal villi by 7-fold or more compared to infected untreated mice (Fig 7E and Fig 8G). These findings corroborated the observations that treatment with NSC158011 and NSC10447, just like paromomycin, inhibited C. parvum oocysts shedding in mice’s feces to almost undetectable levels.
Because of the lack of genetic tools for identifying essential molecular components in Cryptosporidium, screening for potential drug lead-compounds against Cryptosporidium has been based on molecular targets identified in other protozoan parasites such as Toxoplasma and Plasmodium. However, the completed and annotated Cryptosporidium genome sequence shows the absence of conventional drug targets being pursued in other protozoan parasites [10]. Nevertheless, the completed genome sequence of Cryptosporidium has unveiled a number of bacterial-like and plant-like classic and novel drug molecular targets that now require functional characterization and validation using genetic tools [10]. Among the identified potential drug molecular targets, is the C. parvum lactate dehydrogenase (CpLDH), which is a bacterial-type lactate dehydrogenase enzyme that the parasite uses to generate metabolic energy (ATP) in the glycolytic pathway [11, 17, 18]. Importantly, C. parvum lacks both the Krebs cycle and the cytochrome-based respiration chain [10], suggesting that the glycolysis pathway is the sole energy source in C. parvum [19–21]. Consistently, using morpholino-based targeted knockdown of CpLDH, we recently showed that CpLDH is essential for growth, propagation and viability of C. parvum in vitro and in vivo [8, 9]. Corroboratively, previous studies have shown that known inhibitors of lactate dehydrogenase enzymes, gossypol and FX11, are able to inhibit the enzymatic activity of CpLDH [11]. However, both gossypol and FX11 are not specific to CpLDH, but also inhibit mammalian lactate dehydrogenases, implying that they would be toxic to mammalian cells. Regardless, it is noteworthy that CpLDH is unique to C. parvum, and is very significantly different from the lactate dehydrogenase enzymes found in mammals [17].
In the present study, we first established the in vitro enzymatic kinetic parameters of the natively purified recombinant CpLDH protein. Consistent with previous reports by others [11], we found that recombinant CpLDH preferentially catalyzed the reduction of pyruvate to lactate, and displayed Michaelis-Menten enzymatic kinetics. Using the in vitro enzymatic assay, we identified 29 chemical compounds that inhibited the catalytic activity of recombinant CpLDH for the reduction of pyruvate to lactate. Lactate dehydrogenase is a key enzyme for the anaerobic respiration in which pyruvate is reduced to lactate, with the concomitant oxidation of NADH to NAD+ [22]. Thus, we tested the candidate compounds for toxicity in a mammalian cell line (HCT-8) and selected only those that were tolerable at high micromolar concentrations (IC50 > 140 μM) as candidate compounds for further testing. The cytotoxicity IC50 values of the candidate compounds were at least 2-fold higher than the cytotoxicity IC50 values of known mammalian lactate dehydrogenase inhibitors (gossypol and FX11) in HCT-8 cells [11]. We subsequently tested the candidate compounds for anti-Cryptosporidium effect using in vitro infection assays of HCT-8 cells monolayers and identified compounds NSC158011 and NSC10447 that sustainably inhibited the proliferation of intracellular C. parvum. The HCT-8 cells were infected with excysted C. parvum sporozoites that infect host cells and transform into proliferative merozoites. In C. parvum sporozoites and merozoites, CpLDH is expressed and localized in the cytosol [18], suggesting that it is utilized for energy generation during these parasite stages that are important for host cell invasion and intracellular parasite growth. Interestingly, NSC158011 has been previously shown to inhibit the catalytic activity of the Plasmodium faclciparum phosphoethanolamine methyltransferase enzyme, and to inhibit in vitro intracellular growth of the parasite [23]. However, based on the completed genome sequence of C. parvum, there are no homologs of genes encoding a phosphoethanolamine methyltransferase in C. parvum. Therefore, our findings suggest that the anti-Cryptopsoridium activity of NSC158011 is associated with its ability to inhibit the catalytic activity of CpLDH which is an essential enzyme for survival and growth of C. parvum, both in vitro and in vivo [8, 9].
At amino acid sequence level, CpLDH is only 25% identical to human LDH, with the active site conformation of CpLDH being significantly different from that of the human LDH [24]. Further, in the 3-dimensional structure model of the two enzymes, the helix-loop portion of CpLDH is more proximal to the active site loop than it is in the human LDH [24]. Additionally, the co-factor binding site of human LDH possesses a network of hydrogen-bonding formed by a serine residue with NAD+, while the co-factor binding site of CpLDH only forms two hydrogen bonds with NAD+ [24]. This is thought to lower the CpLDH affinity for NAD+/NADH than human LDH [24]. When we modeled NSC158011 and NSC10447 onto the 3-D crystal structure of CpLDH and human LDH, we found that both NSC158011 and NSC10447 bind to the NAD+ co-factor binding site. Interestingly, in the docking simulation, NSC10447 displayed better affinity for CpLDH than human LDH, while NSC158011 displayed better affinity for the human LDH. We had selected NSC10447 and NSC158011 based on their low toxicity in a human cell line, but high inhibitory activity against C. parvum, though these molecules still bind to the human LDH crystal structure. A docking simulation calculates the free energy of the interaction between a protein and a ligand but does not consider the interaction between the ligand and its surrounding solvent (solvation energy). Due to the unfavorable high solvent exposure of the non-polar, aromatic rings in NSC158011 docked to human LDH, it can be inferred that the binding stability is greatly reduced. The docking pose of NSC158011 to CpLDH exposes the non-polar regions of the molecule to less solvent, leading to a much more stable interaction. These ligand binding properties suggest that NSC158011 and NSC10447 would more effectively compete out the binding of NAD+ to CpLDH than to human LDH. This is consistent with our observations that both NSC158011 and NSC10447 effectively inhibit C. parvum growth and replication (both in vitro and in vivo) at concentrations that are not toxic to mammalian (including human) cells.
Typically, solvent-exposed protein pockets like the one in LDH are often not targeted in lead-compound optimization due to their poor binding characteristics, but the results of our in silico docking reveal that solvent-exposed protein pockets may be useful for enhancing lead-compound selectivity. Importantly, these differences in ligand-binding stability between CpLDH and human LDH offer prospects for identifying inhibitors that would specifically target CpLDH, without being toxic to mammalian host cells, and would thus be potential lead-compounds for development of effective anti-Cryptosporidium drugs.
We observed that NSC158011’s IC50 for the inhibition of recombinant CpLDH in an in vitro enzymatic assay was higher than its IC50 for inhibition of C. parvum growth. Based on our observation that NSC158011 binds to the co-factor binding site in CpLDH, the likely reason for this discordance is that in the recombinant CpLDH enzymatic assay in vitro, excessive amounts of NADH co-factor (1 mM) were used that in turn required high concentration of NSC158011 to effectively compete out the co-factor and reduce the generation of the product. In comparison, intracellular (intra-parasite) levels of co-factor are likely much lower (μM range). For instance, in human cells the absolute concentration of NADH has been reported to be in the range of 97 to 168 μM [25, 26]. The lower intracellular concentrations of NADH when compared to the higher in vitro concentrations, would translate into lower concentrations of NSC158011 to effectively compete out the co-factor and register a decrease in CpLDH activity, and subsequent reduction in parasite growth. Importantly, the chemical structure of NSC158011 suggests that it possesses promising drug-like properties that render it amenable to drug development [23].
Using doses that were tolerable in mice, we tested the in vivo efficacies of NSC158011 and NSC10447 in Gamma interferon knockout mice (B6.129S7-Ifng) that when infected with C. parvum, develop debilitating clinical disease, with completion of the parasite life cycle and shedding of oocysts in feces [27]. We found that both NSC158011 and NSC10447 consistently significantly reduced the shedding of C. parvum oocysts during the experimental period of 9 days, and prevented the occurrence of villous atrophy and intestinal mucosal erosion that is associated with C. parvum infection. NSC158011 displayed better efficacy than NSC10447, both in vitro and in vivo, with lower anti-Cryptosporidium IC50 values. Importantly, compared to the only FDA-approved nitazoxanide that lacks efficacy in immunocompromised individuals, both NCS158011 and NSC10447 were efficacious against C. parvum in the immunocompromised mice we used in the study.
In conclusion, we have demonstrated NSC158011 and NSC10477 as specific inhibitors for CpLDH that have efficacy against C. parvum both in vitro and in vivo. Thus, our findings provide promising anti-Cryptosporidium drug candidates that can be explored further for the development of much needed novel cryptosporidiosis therapeutic interventions.
All experiments involving the use of mice and Holstein calves were carried out in accordance with guidelines and protocols number 17024 and 18108, respectively, approved by the University of Illinois Institutional Animal Care and Use Committee, in compliance with the United States Department of Agriculture Animal Welfare Act and the National Institute of Health Public Health Service Policy on the Humane Care and Use of Animals guidelines.
For all experiments, the AUCP-1 isolate of C. parvum was used. The parasites were maintained and propagated in male Holstein calves in accordance with the guidelines of protocol number 18108 approved by the University of Illinois at Urbana-Champaign, USA. Freshly shed C. parvum oocysts in calf feces were extracted and purified by sequential sieve filtration, Sheather's sugar flotation, and discontinuous sucrose density gradient centrifugation, essentially as previously described [28, 29]. The purified oocysts were rinsed and stored at 4°C in 50 mM Tris–10 mM EDTA, pH 7.2, and used within 3 months. Sporozoites were excysted from C. parvum oocysts following the method described previously [30]. Briefly, to about 1 × 108 purified C. parvum oocysts suspended in 500 μl of PBS, an equal volume of 40% commercial laundry bleach was added and incubated for 10 minutes at 4°C. The oocysts were washed four times in PBS containing 1% (w/v) bovine serum albumin and resuspended in Hanks balanced salt solution, incubated for 60 minutes at 37°C, and mixed with an equal volume of warm 1.5% sodium taurocholate in Hanks balanced salt solution. The excysted sporozoites were collected by centrifugation and resuspended in supplemented PBS. The sporozoites were purified by passing the suspension through a sterile 5 μM syringe filter (Millex) and enumerated with a hemocytometer.
The coding sequence of CpLDH was cloned from cDNA prepared from the AUCP-1 isolate of C. parvum, and the His-tagged CpLDH recombinant protein expressed in Escherichia coli, and purified in native form essentially as previously described [8]. Briefly, cDNA was prepared from total RNA extracted from the AUCP-1 isolate of C. parvum, and the coding sequence of CpLDH (Genebank accession number AF274310.1) was PCR-amplified from the cDNA using the primer pair 5’-CTCGAGATGATTGAAAGACGCAAGA-3’ (Forward, with the XhoI restriction site italicized and start codon in bold) and 5’-GGATCCTTATGCTCCAGCTGGT-3’ (Reverse, with the BamHI site italicized and stop codon in bold). The PCR amplicon was cloned at the XhoI/BamHI site of the pET15b expression vector in-frame with the hexahistidine (His-tag) at the N-terminal and sequenced to confirm identity. The recombinant expression vector was transformed into protein expression E. coli BL21-CodonPlus-DE3-RIL (Stratagene). Transformed E. coli was cultured at 37°C in Luria broth medium (supplemented with 100 μg/ml ampicillin and 34 μg/ml chloramphenicol) to an A600 of 0.8 followed by addition of 1 mM isopropyl-β-d-thiogalactopyranoside to induce protein expression. The expression E. coli was harvested and lysed under native conditions by sonicating in lysis buffer (50 mM NaH2PO4, 300 mM NaCl, 10 mM Imidazole, pH 8.0) containing a 1x EDTA-free protease inhibitor cocktail, 600 units benzonase and 30 kU lysozyme (EMD Millipore). The lysate was clarified by centrifugation and the His-tagged recombinant protein purified under native conditions by nickel-affinity chromatography according to the manufacturer's instructions (Novagen). The wash buffer used contained 50 mM NaH2PO4, 300 mM NaCl and 20 mM Imidazole, pH 8.0, while the elution buffer was composed of 50 mM NaH2PO4, 300 mM NaCl, 250 mM Imidazole, pH 8.0. The eluate was dialyzed using a buffer containing 5 mM Hepes–KOH (pH 7.8) and 0.5 mM DTT. The purity of the recombinant protein was determined by SDS/PAGE, and the concentration measured using a Qubit 3.0 fluorometer (Life technologies). The in vitro enzymatic activity of the recombinant CpLDH protein for catalyzing the reduction of pyruvate to lactate was determined by measuring the change in optical density of a 100 μl reaction mixture containing 10 mM pyruvate, 1 mM NADH, 100 mM Tris, pH 7.5 and varying concentrations of CpLDH recombinant protein at 25°C. On the other hand, the catalytic activity of CpLDH recombinant protein for the oxidation of lactate to pyruvate was determined by measuring the change in optical density of a 100 μl reaction mixture containing 100 mM lactate, 1 mM NAD+, 100 mM Tris, pH 9.2, with varying concentrations of CpLDH recombinant protein at 25°C. For determining the kinetic parameters, a fixed concentration of CpLDH recombinant protein was used in reactions with varying substrate and co-factor concentrations (pyruvate from 0.5–15 mM; NADH from 0.25–1.5 mM for the reduction reaction, while for the oxidation reaction lactate varied from 25–125 mM; NAD+ from 0.05–1.5 mM). In all assays, reaction mixtures without recombinant CpLDH protein were included as negative controls. All assays were performed in triplicate and repeated at least thrice. The change in optical density was measured every 15 seconds for a total of 2 minutes using a Spectra Max 384 Plus plate reader (Molecular Devices) at a wave length of 340 nm.
The chemical compounds were obtained from the National Cancer Institute/Developmental Therapeutics Program Open Chemical Repository. They consisted of a diverse set of compounds (n = 27) (S1 Table) reported previously [12], and a Mechanistic Set IV compounds (n = 800) (S2 Table). The compounds were reconstituted in dimethyl sulfoxide (DMSO) as stock solutions. Just before use, aliquots of the stock solutions were diluted in sterile distilled water to generate working solutions, such that the final amount of DMSO added to the reaction mixtures was less than 1% (V/V). The compounds were tested for their inhibitory effect against the enzymatic activity of recombinant His-tagged CpLDH for catalyzing the reduction of pyruvate to lactate. The reactions were performed in 100 μl reaction volume containing 10 mM pyruvate, 1 mM NADH, 100 mM Tris, pH 7.5, 15 ng/μl of recombinant CpLDH protein with or without 20 μM of compound. Control reactions without recombinant CpLDH protein were included. Reactions were performed in triplicate and repeated at least thrice. The change in optical density was measured every 15 seconds for a total of 2 minutes using a Spectra Max 384 Plus plate reader (Molecular Devices) at a wave length of 340 nm. The mean percent inhibition effect of each compound on recombinant CpLDH activity was derived using the following formula:
Mean Percent Inhibition (MPI) = (ΔOD340 of reaction with compound / ΔOD340 of reaction without compound) X 100
Where:
Compounds with inhibitory effect against the enzymatic activity of recombinant CpLDH were tested for cytotoxicity in a human cell line, HCT-8 (American Type Culture Collection Item number: CCL244), that was used for in vitro culture of C. parvum. A colorimetric assay using the cell proliferation reagent WST-1 (Roche, USA) for the quantification of cell viability was performed. HCT-8 cells were cultured in 96-well plates in 200 μl of RPMI 1640 medium without phenol red (Life Technologies), but supplemented with 2 g/L of sodium bicarbonate, 2.5 g/L of glucose, 10% FBS (Gibco, USA), 1× antibiotic–antimycotic (Gibco), and 1× sodium pyruvate (Gibco). When the cells were confluent, the old medium was replaced with fresh medium with or without varying concentrations of chemical compound. After 24 h of culture, 10 μl of the cell proliferation reagent WST-1, (for quantification of cell viability) was added to each well, mixed and the plates incubated for 1 h at 37 C with 5% CO2 in the dark. Following incubation, 150 μL of the medium from each well was transferred to a new 96-well plate and quantification of the formazan dye produced by metabolically active cells was read as absorbance at a wavelength of 420 nm using a scanning multi-well spectrophotometer (Spectra Max 384 Plus; Molecular Devices, USA). Three independent assays were performed and the dose–response curves of the means of triplicate assays were generated using GraphPad PRISM software to derive the half maximal inhibitory concentration (IC50) of compounds in HCT-8 cells.
HCT-8 cells were cultured in supplemented RPMI-1640 medium in 96-well plates. When the cells were confluent, old medium was replaced with fresh medium. To one set of wells, varying concentrations of recombinant CpLDH inhibitors (reconstituted in DMSO and diluted in RPMI medium) were added, while another set was left without inhibitors. Paromomycin was used as a positive control drug reconstituted in distilled sterile water. Then, 4 x 104 freshly excysted sporozoites were added to each well and incubated at 37°C with 5% CO2. After 2 h incubation, varying concentrations of CpLDH inhibitors were added to the set of infected cells that were not treated initially. Control infected cells without inhibitors, but with added DMSO volumes equivalent to those used in the wells with inhibitors, were included. The cells were maintained in culture for a total of either 48 h or 72 h and processed for immunofluorescence assay as previously described [8, 16]. Briefly, medium was decanted and the cells fixed with methanol-acetic acid (9:1) for 2 minute at room temperature. The cells were rehydrated and permeabilized by two successive washes with buffer (0.1% Triton X-100, 0.35 M NaCl, 0.13 M Tris-base, pH 7.6) and blocked with 5% normal goat serum, followed by staining with antibody to C. parvum (SporoGlo; Waterborn, Inc.) overnight at 4°C. The stained cells were washed twice with PBS, followed by water, and then imaged with an inverted fluorescence microscope. Fluorescence quantification was done using ImageJ version 1.37v software (NIH). Assays were performed in triplicate and repeated at least thrice.
To propose a model for the specific binding of NSC15801 and NSC10447 to lactate dehydrogenase, the crystal structure of CpLDH complexed with substrate pyruvate and cofactor analogue 3-acetylpyridine adenine dinucleotide (APAD) was obtained from the RCSB protein database (4ND2). The chemical structures of NSC158011 and NSC10447 were obtained from the PubChem library. The protein crystal structure was loaded into the AutoDockTools software suite and a search location box was drawn encompassing the co-factor analogue APAD in one subunit of the homotetrameric protein. APAD was removed from the active site of the crystal structure using the Swiss-Model DeepView software. Using Autodock Vina (Scripps Institute, USA) as previously done [13], polar hydrogen atoms were added to the APAD-deficient LDH structure and its non-polar hydrogen atoms were merged. NSC15801 and NSC10447 were each docked into the empty co-factor binding site of one subunit of the protein with exhaustiveness = 10. Both compounds were docked using a 40 × 40 × 40 Å grid box, and all single bonds within the ligands were set to allow free rotation. The procedure was subsequently repeated with the crystal structure of human LDH (1I0Z) and a 24 × 14 × 20 Å grid box around the co-factor binding pocket of the new structure. Docking results were visualized using VMD (University of Illinois at Urbana-Champaign, USA) as previously done [31]. The most energetically favorable result of each docking was then loaded as a protein-ligand complex into the Schrödinger Maestro software (Schrodinger LLC, USA) to investigate the nature of the protein-ligand interactions and propose a mechanism for the lead compounds’ inhibition of LDH. Because the presence or absence of natural LDH substrate, pyruvate, complexed within the active site did not significantly affect binding of the compounds in CpLDH or human LDH, those models were excluded in the final molecular docking simulations.
Gamma interferon knockout mice (B6.129S7-Ifng), 8 weeks of age, were purchased from Charles River, USA. The care and use of the mice was done following the guidelines of protocol number 17024 approved by the University of Illinois at Urbana-Champaign, USA. The animals were allowed to acclimatize for 1 week before experiments commenced. Stock solutions of recombinant CpLDH inhibitors reconstituted in DMSO were diluted in sterile distilled water to reduce the final amount of DMSO in the solution to less than 1% (v/v) before administering them to mice. Prior to testing the anti-Cryptosporidium effect of the inhibitors in mice, the tolerability of each inhibitor was tested by oral gavage using varying dosages (100–2000 mg/kg body weight) of each inhibitor in groups of mice (three mice for each dose) daily for 7 days. Mice were observed daily for signs of toxicity including changes from normal physical activity, respiration, body temperature, feeding, posture, fur condition or occurrence of death. The highest dose (1000 mg/kg for NSC10447, and 400 mg/kg for NSC158011) of each inhibitor that did not induce any toxicity signs over the 7 days of administration was used as the maximum dose limit for subsequent in vivo experiments. The subsequent dosages of NSC10447 used in the mice infection assays were 250, 500 and 1000 mg/kg mouse body weight, while the NSC158011 dosages used were 100, 200 and 400 mg/kg mouse body weight. Mice were allocated to groups as follows: “Infected plus inhibitor treatment”; “Infected minus inhibitor treatment”; “Uninfected minus inhibitor treatment” and “Infected plus paromomycin treatment”. Each group contained at least three mice. Each mouse in the infection groups received 5,000 C. parvum oocysts (resuspended in 50 μl of PBS) by oral gavage. Mice were housed individually in cages lined with sterile gauze as bedding. One day post-infection (PI), daily oral gavage administration of recombinant CpLDH inhibitor or paromomycin commenced and continued for a total of 7 days. Untreated mice received an equivalent volume of sterile distilled water (containing DMSO equivalent to the amount administered in the inhibitor-treated group) by oral gavage. Fecal pellets were collected daily from each cage and placed in individual sterile 15 ml conical tubes. An equivalent volume of PBS containing a cocktail of penicillin (100 units/ml), streptomycin (100 μg/ml), chloramphenicol (34 μg/ml) and amphotericin (0.25 μg/ml) was added to the fecal samples and stored at 4°C until use for quantification of C. parvum genomic DNA load. Three independent replicate infection assays were performed. At 9 days PI, mice were euthanized and 5 cm of the distal small intestine resected 2 cm anterior to the cecum and immediately submerged in 4% buffered formalin. The intestinal tissues were submitted for histopathology to the Veterinary Diagnostic Laboratory at the University of Illinois at Urbana-Champaign. Briefly, intestinal tissues preserved in 4% buffered formalin were washed in 70% ethanol and embedded in 1% agar and then processed for paraffin embedding. For hematoxylin and eosin staining, five μm transverse and cross sections were cut and processed and stained following standard procedures of the Veterinary Diagnostic Laboratory. Sections were imaged using a Zeiss microscope and images captured with a color camera.
Genomic DNA was extracted from individual fecal samples collected from mice at different days as described above. For each sample, 220 mg of homogenized feces were used to extract genomic DNA using the QIAamp PowerFecal DNA Kit (Qiagen, USA) following the manufacturer’s instructions. Quantification of the amount of C. parvum 18s rRNA gene (GenBank accession number AF164102) was performed essentially as described previously [9]. Briefly, the primer pair 5′-CTGCGAATGGCTCATTATAACA-3′ (Forward), and 5′-AGGCCAATACCCTACCGTCT-3′ (Reverse) was used to generate a 240 bp amplicon from C. parvum genomic DNA by conventional PCR. The PCR product was fractionated on agarose gel, extracted using the QIAquick Gel extraction kit (Qiagen, USA), and the concentration measured by Nanodrop Spectrophotometer (Fisher, USA). Ten-fold serial dilutions of the extracted DNA fragment were made and used as quantification standards for real-time PCR. Each real time PCR mixture contained 2 μl of DNA template, 1 μl of primer mix (500 nM each), and 10 μl of SsoAdvanced Universal SYBR Green Supermix (Bio-Rad, USA), with the final volume made up to 20 μl with nuclease-free water. The cycling conditions included an initial denaturation for 10 min at 98°C, 40 cycles at 98°C for 15 s and 60°C for 1 min, and a final melting curve step. Cycling was performed using a 7500 Real Time PCR System (Applied Biosystems, USA). DNA quantities were derived by the system software using the generated quantification standard curves.
Statistical analyses were performed using two-tailed Student’s t test. P values of 0.05 or less were considered significant.
|
10.1371/journal.ppat.1005361 | A Shift from Cellular to Humoral Responses Contributes to Innate Immune Memory in the Vector Snail Biomphalaria glabrata | Discoveries made over the past ten years have provided evidence that invertebrate antiparasitic responses may be primed in a sustainable manner, leading to the failure of a secondary encounter with the same pathogen. This phenomenon called “immune priming” or "innate immune memory" was mainly phenomenological. The demonstration of this process remains to be obtained and the underlying mechanisms remain to be discovered and exhaustively tested with rigorous functional and molecular methods, to eliminate all alternative explanations. In order to achieve this ambitious aim, the present study focuses on the Lophotrochozoan snail, Biomphalaria glabrata, in which innate immune memory was recently reported. We provide herein the first evidence that a shift from a cellular immune response (encapsulation) to a humoral immune response (biomphalysin) occurs during the development of innate memory. The molecular characterisation of this process in Biomphalaria/Schistosoma system was undertaken to reconcile mechanisms with phenomena, opening the way to a better comprehension of innate immune memory in invertebrates. This prompted us to revisit the artificial dichotomy between innate and memory immunity in invertebrate systems.
| Schistosomiasis is the second most widespread tropical parasitic disease after malaria. It is caused by flatworms of the genus Schistosoma. Its life cycle is complex and requires certain freshwater snail species as the intermediate host. Given the limited options for treating S. mansoni infections, much research has focused on a better understanding of the immunobiological interactions between the invertebrate host Biomphalaria glabrata and its parasite S. mansoni. Recently, we demonstrated the existence of a time-dependent and genotype-dependent acquired innate immune memory in B. glabrata snails. A primo-infection of the Lophotrochozoan vector snail, Biomphalaria glabrata, with Schistosoma mansoni totally protected the snail against a secondary challenge. Learning more about the immunobiological interactions between B. glabrata and S. mansoni could have important socioeconomic and public health impacts by changing the way we attempt to eradicate parasitic diseases and prevent or control Schistosomiasis in the field.
| The environment of an invertebrate is filled with complex and changing populations of microorganisms, including potential pathogens. This engenders selective pressures comparable to those experienced by Gnathostomes [1], and means that invertebrates should possess sophisticated immune systems capable of dealing with such pathogens. Indeed, recent studies have shown that the immune defenses of invertebrates are more complex and specific than previously thought, and the existence of innate immune memory or priming has been suggested [2–6].
To date, the observations of invertebrate innate immune memory have been mainly phenomenological and based on ecological or phenotypic studies, and little work has addressed the potential molecular and cellular mechanisms underlying these processes. The innate immune system of invertebrates includes the barrier functions of the epithelium, the cellular immune response and the humoral defense response [7,8]: the former refers to mucosal immunity and hemocyte responses (phagocytosis, encapsulation, melanization), while the latter includes pathogen recognition receptors (PRRs), antimicrobial peptides, coagulation, and the production of cytolytic molecules or reactive intermediates of oxygen and nitrogen [7,8]. During the primary infection of an invertebrate, the pathogen is recognized at the first encounter and the cellular and humoral defense responses are coordinated to neutralize the intruder. To the best of our knowledge, the existing studies investigating the molecular mechanisms of innate immune memory in invertebrates have all suggested that the cellular immune response and/or hemocyte phagocytosis is/are improved upon a subsequent encounter with the same pathogen [9–11]. For example, in Porcellio scaber, enhanced phagocytosis was demonstrated after a first encounter with different bacteria species (Bacillus thuringiensis/Escherichia coli) [12]. For Anopheles gambiae, hemocytic differentiation evidenced by increased mRNA levels of hemocyte-specific genes, such as thioester-containing protein 1 (TEP1) and leucine rich repeat immune protein 1 (LRIM1) was reported following initial exposures to Plasmodium and bacteria [5]. In shrimp, bacterial challenge was followed by an enhanced cellular immunity characterized by a significant increase in the percentage of phagocytic cells [13]. The innate immune memory response in invertebrates has been previously described as being involved in two mechanisms, namely:
However, the international community working on invertebrate innate immunity believes that the existing innate immune memory observations cannot be used in isolation. Instead, they should be used to construct hypotheses that need now to be exhaustively tested with rigorous functional, cellular, biochemical and molecular methods, until all alternative explanations are eliminated [10,14]. To fully understand the capabilities of invertebrate immune systems, we must use global molecular approaches at the whole-organism level to investigate the mechanisms that form the basis for innate immune memory, while being guided by the existing observations and phenotypic descriptions. Without this demonstration, the existence of immune memory process in invertebrates will remain controversial and doubted by many immunologists [for the polemic, see [14,15]].
Recently, we demonstrated that a primo-infection of the Lophotrochozoan vector snail, Biomphalaria glabrata, with Schistosoma mansoni protected completely the snail against a homologous secondary challenge [4]. Total protection was reached 10 days after primo-infection, and was maintained for the rest of the snail’s lifespan. Our findings provided evidence for the existence of a time-dependent acquired innate immune memory in B. glabrata snails [4,16]. Moreover, we used homologous and heterologous challenges to demonstrate that this innate immune memory was genotype-dependent, in that the protection decreased with increasing neutral genetic distance between the parasites used for the primo-infection and the secondary challenge [4].
Here, we sought to investigate the molecular mechanisms of innate immune memory in B. glabrata in response to S. mansoni infections. We precisely describe the snail’s immune response phenotypes, then characterize the molecular determinants involved in B. glabrata innate immune memory via a global transcriptomic approach and a targeted proteomic analysis of plasmatic factors performed in combination with RNA interference. This work provides the first global investigation of the molecular processes supporting innate immune memory in an invertebrate model. Moreover, Biomphalaria snails have an important role in the transmission of Schistosomiasis the second most widespread human parasitic disease after malaria causing 200,000 deaths annually. More emphasis on snail-related research, on the role of snails and parasite intramolluscan larval stages in transmission is essential for changing the way we attempt to eradicate parasitic diseases [17,18]. Learning more about the immunobiological interactions between B. glabrata and S. mansoni could have important socioeconomic and public health impacts by contributing to the discovery of new ways to prevent and/or control Schistosomiasis diseases by limiting the parasite in the field.
Further details are provided in S1 Appendix and Fig 1.
The present work utilized a strain of Biomphalaria glabrata originating from Brazil (BgBRE), along with its 100% compatible sympatric strain of Schistosoma mansoni (SmBRE). Both were as previously described [19].
One hundred and forty BgBRE snails were primo-infected with 10 miracidia of SmBRE each; 60 were secondarily challenged 25 days later with 10 miracidia of SmBRE each. For the RNAseq approach, pools of 20 BgBRE snails were recovered at 1, 4, 15, and 25 days post primo infection (DPPI); these were designated 1DPPI, 4DPPI, 15DPPI and 25DPPI, respectively. Twenty more snails were recovered at each of 1, 4 and 15 days after the secondary challenge; equimolar amounts of each of these experimental groups were joined together into a single sample designated as days post-secondary challenge (DPC) sample. Two pools of 20 uninfected snails (designated naive1 and naive2) were sampled and used as control replicates. For the proteomic approach, 200 BgBRE snails were primo-infected and secondarily challenged with 10 miracidia (per round) of SmBRE. Plasma sampling was performed on the same schedule described for the RNAseq experiments. Five biological replicates (5 independent pooled plasmas from 10 snails) were recovered from naïve snails (controls), and 15DPPI, 25DPPI and 15DPC snails.
Forty-eight hours after primo-infection or secondary challenge with 10 miracidia of SmBRE, snails were fixed in Halmi’s fixative (mercuric chloride 4.5%, sodium chloride 0.5%, trichloroacetic acid 2%, formol 20%, acetic acid 4% and 10% of picric acid-saturated aqueous solution). Fixed mollusks were dehydrated and embedded in paraffin as previously described [20,21]. Transverse histological sections (10-μm thick) were cut and stained using azocarmine G and Heidenhain’s azan. The following serial steps were used: (i) re-hydration (toluene, 95, 70, 30% ethanol and distilled water); (ii) coloration (azocarmine G, 70% ethanol / 1% aniline, 1% acetic alcohol, distilled water, 5% phosphotungstic acid, distilled water, Heidenhain’s azan) and (iii) dehydration (95% ethanol, absolute ethanol, toluene). The preparations were then mounted with Entellan (Sigma Life Science, St. Louis Missouri, USA) and subjected to microscopic examination.
Total RNA was extracted from naive1, naive2, 1DPPI, 4DPPI, 15DPPI, 25DPPI and DPC samples using nitrogen and the TRIzol reagent (Sigma Life Science, St. Louis Missouri, USA). mRNAs were sequenced in paired-end, 72-bp read length, with three samples multiplexed per lane, using an Illumina Genome Analyzer 2 (Montpellier Genomix (MGX), Montpellier, France).
De novo transcriptomes were assembled using high-quality reads (phred > 29) from all seven sequenced samples and an in-house pipeline created using the Velvet (v1.2.01), Oases (v0.2.04) and CDhit EST (v4.5.4) programs [22]. A consensus reference transcriptome was optimized using various parameters, including k-mer length, insert length and expected coverage, as previously described [22]. To suppress non-molluscan transcripts, a BLAST-based transcriptome subtraction strategy was used. Transcripts were compared against the B. glabrata draft genome, and transcripts with identities and coverages of less than 80% were deleted. This led to a final subtraction of 45.1% of the transcripts, most of which corresponded to S. mansoni. The final size of the transcriptome was 159,711 transcripts.
Quality reads (Phred score >29) were aligned to the assembled transcriptome using the C++ script, Bowtie2 (v2.0.2) (mapping quality score 255), which was run locally using the Galaxy Project server [23]. The DESeq2 software [24] (v2.12; http://www.bioconductor.org/packages/release/bioc/html/DESeq2.html) was run under the default settings to compare duplicate samples from uninfected snails (naive1 and naive2) versus infected samples to quantify differential gene expression (P value < 0.1). A heatmap was constructed to analyze transcript expression patterns (log2 fold change) using Hierarchical Ascending Clustering (HAC) with Pearson correlation, as applied by the Cluster 3.0 [25] and JavaTreeView software packages. We used quantitative real time PCR (Q-RT-PCR) to ascertain RNAseq results. For that purpose the correlation between the rpkm (reads per kilobase per million mapped reads) of RNAseq and the ct (cycle threshold) of Q-RT-PCR was tested and confirmed (see S1 Fig). Primer sequences used in Q-RT-PCR were available in S3 Table.
Small interfering RNAs (siRNAs) were used to knock down FREP2 (GenBank accession number gi|16303186), FREP3 (gi|18389116) and FREP4 (gi|16303188). Three siRNA duplexes (Eurogentec) were used in conjunction with Invivofectamine transfecting agent (Invitrogen, CA). Two hundred ng of pooled FREP duplex siRNA or GFP siRNA (used as control) were injected into the cardiac sinus of B. glabrata snails as previously described [26]. One, Two and four days following siRNA injection, the knock-down efficiency was confirmed by Q-RT-PCR analysis of FREP2, -3 and -4 mRNA expression. For phenotypic analysis, 90 BgBRE snails were infected with 10 miracidia of SmBRE. Twenty-one days after primo-infection, these snails were individually injected with or without siFREPs or siGFP. Four days later, each snail was subjected to a secondary challenge with 10 miracidia of SmBRE. Fifteen days post-challenge, snails were fixed and the parasite prevalence was determined by identifying schistosome primary sporocysts (SpI), in snails tissue as previously described [27,28].
Hemolymph was collected from the head-foot region of 50 BgBRE snails (5 pools of 10 snails, 5 biological replicates) at each time point of the infection process (Naïve, 15DPPI, 25DPPI, and 15DPC). Hemocytes were pelleted by centrifugation at 2500 rpm for 15 min at 4°C, and discarded. Hemoglobin was separated from the plasma by ultracentrifugation at 55,000 rpm for 2.5 h at 4°C and the supernatant plasma samples were recovered. Protein concentration of the samples were estimated using a 2D Quant kit (GE Healthcare life sciences). The plasma (5 biological replicates for each of the 4 time points) was subjected to 2D gel electrophoresis on 12% SDS-PAGE gels using 100 μg of plasma, denaturing buffer, and 17 cm IPG Strips, pH 3–10 non-linear gradient (BioRad). Gels were stained with mass spectrometry (MS)-compatible silver staining, and comparative analysis of digitized proteome maps was performed using the PDQuest 7.4.0 image analysis software (Bio-Rad). Spots showing obvious qualitative and quantitative (at least 2-fold) differences were excised from the gel and characterized by nanoscale capillary liquid chromatography on an Ultimate 3000 coupled to a LTQ-Orbitrap tandem mass spectrometer (nanoLC–MS/MS) mapped to Swiss prot-trembl and the Biomphalaria glabrata Brazil transcriptome (http://ihpe.univ-perp.fr/enseignement/). A protein was considered to be correctly identified if at least two peptides were confidently matched to database sequences with an overall Mascot score greater than 50 [29].
Plasma samples were recovered from 15DPPI or naïve snails as described above. PBS-snail solution (8.41 mN Na2HPO4, 1.65 mN NaH2PO4, 45.34 mM NaCl) was used as the control injection. Twenty μl of each sample (naïve plasma, 15DPPI plasma, or TBS Tween solution) were injected into 25 BgBRE snails per group. One group of 48 BgBRE snails was left untreated and used as a control for infection. Fifteen days after injections, the snails in the four experimental groups were infected with 10 miracidia/snail of SmBRE. This period of 15 days was used to ensure that an observed phenotype was not a direct effect of the injected molecules, but rather reflected their ability to activate the snail’s immune response pathways. Fifteen days after infections, snails were dissected and the parasite prevalence was quantified as described above.
To test for significant differences in prevalence, Fisher’s exact test was used, with P ≤ 0.05. For the proteomic approach, statistically significant quantitative differences between spots were tested using a Mann-Whitney U test, as applied with the PDQuest 7.4.0 software (BioRad).
The laboratory and experimenters possessed an official certificate of the French Ministry of National Education, Research, and Technology, CNRS and DRAAF Languedoc Roussillon for experiments on animals, animal housing, and animal breeding (# A66040; decree # 87–848, October 19, 1987; and authorization # 007083).
Three types of immuno-biological interactions were observed following infection of B. glabrata by S. mansoni. After primo-infection, a compatible interaction was characterized by the ability of miracidia to penetrate, transform into sporocysts and develop normally in snail tissues (30 to 40% of the entering miracidia produce SPI) (Fig 2A). In an incompatible interaction, miracidia were immediately recognized, encapsulated and killed by hemocytes; in this case, a multicellular capsule could be observed surrounding the parasite (60 to 70% of the entering miracidia were killed) (Fig 2B). Finally, in primed snails subjected to a secondary challenge, the sporocysts degenerated in the snail tissues. Encapsulation was never observed nor was the accumulation of hemocytes observed near the parasite (100% of the entering miracidia were killed by humoral factors) (Fig 2C). These findings indicate that a cellular immune response was the main outcome following a primo-infection, whereas an exclusively humoral immune response was found in primed snails subjected to a secondary challenge (Fig 2).
The RNAseq values were validated by a correlation analysis performed with our quantitative Q-RT-PCR data (S1 Fig, S3 Table). We obtained a good correlation (R2 = 0.768), indicating that the representations of transcripts in our RNAseq results appeared to be proportional to those obtained through Q-RT-PCR amplifications.
This RNAseq experiment allowed us to identify 1887 differentially expressed transcripts, most of which were found in the 1DPPI, 25DPPI and DPC samples (Fig 3, S1 Table). At 1DPPI, most of the differentially expressed transcripts were under-represented (Fig 3). This could reflect the immuno-suppression/modulation induced by S. mansoni parasites within the first hours of an infection, when the parasites develop their molecular mimicry strategy [30,31]. At 25DPPI and DPC, the proportions of over- and under-represented transcripts were more comparable (Fig 3). Hierarchical clustering was used to sort the differentially represented transcripts into six clusters, three each corresponding to over-represented transcripts (clusters 1–3) and under-represented transcripts (clusters 4–6) (Fig 3, S1 Table). Cluster 2 included transcripts that were specific to the primo-infection and not activated following the secondary challenge. Clusters 1 and 3 appeared to be the most promising possible sources of innate immune memory candidates. Cluster 1 comprised transcripts that were over-represented following primo-infection and remained highly expressed throughout infection (sustained response); it included immune molecules known to be involved in the cellular immune response (macrophage mannose receptors, thrombospondin), as well as certain pathogen recognition receptors (PRRs) (selectins and C-type lectins). In contrast, cluster 3 comprised transcripts that were over-represented exclusively following the secondary challenge (Fig 3), including certain PRRs (FREPs, C-type lectins) and circulating immune effectors (anti-microbial peptides such as mytimacin and LBP/BPI, and biomphalysin). Interestingly, clusters 4, 5, and 6, which contained the under-represented transcripts, included immune molecules belonging to the same immune functional groups represented in clusters 1 to 3 (Fig 3). This may indicate that there is a trade-off among the immune variants involved in the specific response to S. mansoni parasites, with some members of the same family (i.e., variants or isoforms) being down-regulated to permit the over-expression of other variants. In cluster 4, notably, most of the under-represented transcripts were involved in the cellular and epithelial immune response (e.g., macrophage mannose receptors, extracellular matrix compounds, mucins). This indicates a down regulation of epithelial (mucosal immunity) and cellular immunity two major components of innate immune response following secondary challenge. The complete list of differentially represented transcripts is presented in S1 Table.
To estimate total FREP expression (Fig 4A) over the entire innate immune memory process based on our RNAseq data, we summed the fold changes for all the differentially represented FREP transcripts (identified in Fig 3). A huge induction (5.096-fold) of FREPs was observed following the secondary challenge (DPC, Fig 4A). SiRNA-mediated knockdown of FREPs were used in combination to ascertain the role played by these FREP candidates in innate immune memory. Knockdown of FREPs was confirmed by Q-RT-PCR; the results were normalized with respect to the mRNA expression levels in naïve snails, and compared to the corresponding levels in siGFP-control-injected snails (Fig 4B). The best knockdown level was reached at 96 h post-siRNA injection, when we observed a decrease of 7.98-, 1.56- and 3.60-fold change for FREP 2, 3 and 4, respectively (Fig 4B). The effect of FREP knockdown on the innate immune memory phenotype was assessed in a typical homologous innate immune memory assay. In untreated snails, the parasite prevalence after secondary challenge was 0% (Fig 4C), indicating that innate immune memory had effectively protected them against a subsequent exposure to S. mansoni. Injection of siGFP did not significantly change the parasite prevalence after secondary challenge (4%; Fig 4C). Following injection of the FREP siRNA, however, 15% of the snails became infected with S. mansoni following secondary challenge (Fisher’s exact test P < 0.05) (Fig 4C). Thus, FREP knockdown partially altered the innate immune memory phenotype, rendering primed snails more susceptible to S. mansoni infection.
To validate our RNAseq results, we focused our attention on the plasmatic compartment of B. glabrata, which should contain the actors of the humoral innate immune memory response. We performed global comparative 2D gel electrophoresis on plasma samples obtained from naive, 15DPPI, 25DPPI, and 15DPC snails. Our quantitative and qualitative bioinformatic analyses identified 29 differentially expressed spots corresponding to 62 different proteins (Fig 5A, S2 Table).
A hit map of the quantitated expression ratios was generated for the entire infection process (Fig 5B, S2 Table). Four clusters were identified (Fig 5B): the proteins in cluster 1 were up-regulated following the secondary challenge; those in cluster 2 were up-regulated after the primo-infection and remained highly expressed throughout infection (sustained response); those in cluster 3 were up-regulated after the primo-infection and down-regulated following secondary challenge those of cluster 4 were down-regulated. The differentially expressed proteins were also subjected to functional classification based on their putative functions and/or structural features. Five functional groups were identified: innate immune response proteins (receptors, effectors, and regulators), scavengers of reactive oxygen species, cell signaling proteins, gametogenesis-related proteins, and glycolysis-related proteins. Thirty seven percent of the identified molecules belonged to the innate immune response group (23 of 62 proteins) (Fig 5B). Consistent with our RNAseq results, we observed an important enrichment of potential humoral immune candidates in the plasma of primed snails. The molecules involved in immune recognition and opsonization included numerous isoforms belonging to different families of circulating immune receptors that were also identified in our transcriptomic analysis (e.g., C-type lectins, β-1,3-glucan binding protein, and hemagglutinin/amebocyte aggregation factors). Most of these isoforms were found in clusters 1 and 2, suggesting that they may actively contribute to the innate immune memory response (Fig 5B). Among the immune effector molecules, chitinase (spot 3002, Fig 5B), fatty acid binding protein (spots 4004 and 1001, Fig 5B), and biomphalysin (spot 7315, Fig 5B) were found in clusters 1, 2 and 4, respectively. Biomphalysin was over-represented following primo-infection and secondary challenge in the transcriptomic data, but down regulated at 15DPPI and 15DPC in our proteomic data (cluster 4, spot 7315; Fig 5B). Thus, it seems that biomphalysin is highly transcribed during the innate immune memory response against S. mansoni parasites, but is thereafter consumed at the protein level. Some known regulators of the immune response (e.g., thymosin, spot 3002; serpin, spot 1314; and β chain acetylcholine binding protein, spots 2206 and 3008) were associated with protein clusters 2 and 4 (Fig 5B). In addition, other proteins known to be involved in host/parasite interactions were identified as being differentially expressed, including some anti-oxidant molecules, which were identified in protein clusters 1 and 2 (e.g., glutathione peroxidase 3 precursor, spot 1803; glutathione-S-transferase mu 3-like, spots 7009 and 8006; and thioredoxin peroxidase, spot 4004; Fig 5B).
Proteins involved in gametogenesis were also identified by our global proteomic analysis. They included schistosomin (spot 1004), which is involved in the regulation of gametogenesis [32]; pathogen-related protein (spots 2002 and 1009), which is a seminal fluid protein [33,34]; and ovipostatin (spot 1106), which is involved in oviposition and hatching [35]. These proteins were found in clusters 3 and 4, and were thus down-regulated following primo-infection and/or secondary challenge (Fig 5B) (see S2 Appendix for details).
A plasma experiment was carried out to confirm the role played by humoral factors in innate immune memory. Plasmas were recovered from naïve or 15DPPI snails and injected into naïve snails. The snails were subjected to infection 15 days post-plasma-injection, and analyzed for phenotypes at 15 days post-infection (Fig 5C). Injection of saline solution has no significant effect on the prevalence of parasitic infection compared to the untreated control (88% and 90%, respectively) (Fig 5C). Transfer of naive snail plasma also failed to have any significant effect (prevalence of 77%) (Fig 5C). However, a statistically significant decrease in the prevalence of infection was observed for snails injected with primed snail plasma (from 90% to 68%; P = 0.04). This protection remained efficient at 15 days post-transfer, confirming that plasmatic humoral factors function to activate the immune system and prepare the snail to answer a subsequent encounter with the parasite.
Innate immune memory, which is a process through which an organism acquires a (more or less) specific and long-lasting protection against later challenges that persists even after the pathogen is neutralized, has been described in diverse invertebrate phyla [2,3,5,6,14,36–39]. This suggests that innate immune memory exists in a broad range of invertebrates.
A form of innate immune memory was previously demonstrated in B. glabrata in response to S. mansoni infection, and shown to be dependent on the genotype of the pathogen [4]. Here, we provide evidence that innate immune memory in B. glabrata is not associated with an enhancement of the cellular immune response, as has been suggested in other invertebrate species [5,12,13]. Instead, our histological analysis (Fig 2) revealed that in primed snails, the parasites of a secondary challenge fail to develop into sporocysts and are killed by the host, without any observable cellular immune response/hemocytic reaction. Together, these findings show that, after a primo-infection, each successive encounter with a similar parasite initiates an exclusive humoral immune defense response in B. glabrata.
Accordingly, we set out to identify the humoral factors through which primed snails are able to neutralize the sporocysts of a secondary challenge. A whole-snail RNA sequencing approach was conducted to identify molecular candidates that might be involved in the innate immune memory response. For RNAseq approach the cluster 1 (sustained-response transcripts) and cluster 3 (secondary challenge-specific transcripts) contained the most promising candidates (Fig 3). Numerous circulating or cell-surface PRRs were found. These immune receptors, which recognize terminal sugar residues on the glycans that are attached to the surface proteins of some microorganisms, act in the pathogen recognition and clearance processes of innate immunity. The identified PRRs belong to various families, including the macrophage mannose receptors, selectins, C-type lectins and fibrinogen-related proteins (FREPs). Cytotoxic and cytolytic molecules were also identified, as were anti-oxidant molecules. The latter finding suggests that the reactive oxygen species pathway may be activated as part of the humoral innate immune memory response of B. glabrata. Moreover, RNAseq cluster 4 (molecules down-regulated upon secondary challenge) included transcripts whose products are involved in the epithelial and cellular immune response, such as macrophage mannose receptors, extracellular matrix compounds and mucins. This strengthens our contention that there is a shift from cellular or mucosal immunity to humoral immune response during the innate immune memory response in B. glabrata snails.
An unexpected result of this study was the identification of RNAseq cluster 3 and proteomic cluster 1. Until now, innate immune memory was believed to be the result of two potential regulatory processes: (i) a sustained immune response consisting of the long-lasting up-regulation of immune molecules after a primo-infection (as observed in RNAseq cluster 1 and proteomic cluster 2); or (ii) a recalled response that consists of the ability to store information and recall it later for a faster and more powerful response against subsequent pathogenic exposure (never been observed in any of the RNAseq or proteomic clusters) [36,39]. Here, we provide the first report of a novel unexpected regulatory process for invertebrate innate immune memory, in which transcripts are specifically up regulated following the secondary challenge (RNAseq cluster 3, proteomic cluster 1) without having been previously activated by the primo-infection. This new observation warrants further study, such as through the identification of its potential molecular hallmarks, including the genetic (activation of transcription factors) and/or epigenetic (DNA methylation, chromatin markers, lncRNA, miRNA, etc.) factors that form the basis for this exclusive secondary challenge innate immune memory response.
To validate the potential roles played by the identified molecular candidates in the observed innate immune memory phenotype, we performed three sets of experiments: we used RNAi to knockdown certain FREPs and examined the priming response; we characterized the snail plasma proteome; and we examined the effect of injecting naïve snails with plasma from primed snails.
As FREP lectins showed extensive differential representation following the secondary challenge (see Fig 4A), and they are known to play key roles in the immunobiological interactions between B. glabrata and S. mansoni, they were chosen as candidates for RNAi invalidation. The FREPs comprise one or two amino-terminal immunoglobulin domains (IgSF) and a carboxyl-terminal fibrinogen domain (FBG). They belong to a multigene family of at least 14 members, and undergo somatic rearrangements leading to remarkable diversification [40,41]. FREPs can precipitate soluble antigens derived from trematodes. Their expression levels increase in response to infection with S. mansoni [42,43], and they form immune complexes with similarly highly polymorphic and individually variable mucins (the SmPoMucs) that act as antigens of S. mansoni [44]. The high level of diversification within FREP family members results in a huge proportion of partial FREP transcript sequences generated by the RNAseq de novo assembly [22]. We were unable to precisely identify the FREP variants/isoforms present in the various RNAseq clusters. Thus, to confirm the role of FREPs in innate immune memory, we invalidated FREP 2, 3, and 4, as they had previously been demonstrated to be involved in the response of B. glabrata to trematodes. More specifically, FREP 2 is involved in immune complexes [44]; FREP 3 knock-down reverts the snail resistance status to trematodes [45]; and microarray analysis showed that FREP 4 is over-expressed following infection by S. mansoni [43]. SiRNA-mediated knock-down of FREP 2, 3 and 4 (Fig 4B) was found to reduce the innate immune memory phenotype by 15% (Fig 4C). These results indicate that these FREPs are involved in B. glabrata innate immune memory, but demonstrate also that additional molecular partners (and/or other FREP variants) also play roles in the pathogen recognition and innate immune memory of this snail.
Our characterization of the plasmatic proteome confirmed that most of the molecules identified in the RNAseq data were present and differentially expressed at the protein level. The presence of these molecules in snail plasma also confirms that they act as circulating humoral factors. The most promising candidates for participation in innate immune memory were found in protein cluster 1 (secondary challenge-specific proteins) and protein cluster 2 (sustained-response proteins) (Fig 5). Numerous PRRs were identified (e.g., macrophage mannose receptors, C-type lectin, hemagglutinin, and β -1, 3-glucan binding proteins). Interestingly, we did not recover any FREP in this proteomic analysis of primed snail plasma. We speculate that the FREPs were lost from our analysis due to their precipitation with parasitic antigens [44]. Alternatively, and as suggested previously for other families of highly variable molecules [46], the different variants may be expressed at such low levels that they are not visible on 2D gel electrophoresis. Indeed, we recently demonstrated that FREPs are expressed at a low level compared to other PRRs in naive B. glabrata snails [22]. Plasmatic proteome was also composed of cytotoxic and cytolytic molecules, including antimicrobial peptides [AMPs; e.g., hydramicin, mytimacin and lipopolysaccharide binding protein/bactericidal permeability-increasing protein (LBP/BPI)], and the biomphalysin. We recently reported the molecular cloning and functional characterization of the B. glabrata biomphalysin and its involvement in the killing of S. mansoni [47]. Biomphalysin, which shares a common architecture with proteins belonging to the aerolysin superfamily, is strictly expressed in immune-competent cells. Recombinant biomphalysin was shown to bind to parasitic membranes and exhibit cytotoxic activity toward S. mansoni sporocysts. Our RNAseq and proteomic analyses primarily recovered immune recognition factors, many of which had multiple isoforms. This suggests the involvement of both translational and post-translational regulation. Additional work is now needed to clarify the functions of these molecules in the humoral innate immune memory response of B. glabrata. For example, C-type lectins are known to be involved in the innate immune memory responses of mollusks; their expression levels increased in the scallop, Chlamys farreri, following vaccination with heat-killed Vibrio anguillarum, and successive challenges with V. anguillarum or Micrococcus luteus enhanced this protection [11]. In addition to their roles in pathogen recognition and opsonization, lectins also possess direct cytotoxic activities. Examples of cytotoxic lectins include plant-derived ricin from Ricinus communis beans [48], the fungus-derived N-acetyl-D-galactosamine-specific lectin from Schizophyllum commune [49], and the invertebrate-derived hemolytic lectin, CELIII, from Cucumaria echinata [50]. Thus, it seems reasonable to hypothesize that some of the lectins identified in the present study may participate in the humoral innate immune memory response both as recognition receptors and as cytotoxic/cytolytic molecules involved in killing the S. mansoni SpIs of the secondary challenge.
Lastly, we explored whether the transfer of plasma from primo-infected snails could provide recipient naïve snails with enhanced immunity. In a previous study, plasma transfer from S. mansoni-resistant Biomphalaria tenagophila to susceptible snails was associated with a transfer of resistance [51]. Here, we found that transfer of primed snail plasma to naive snails significantly reduced the prevalence of S. mansoni infection by more than 20% compared to controls (Fig 5). A comparable experiment was previously performed using the mosquito, Anopheles gambiae, and its bacterial pathogens [5]. The transfer of cell-free hemolymph from infected mosquitoes into healthy mosquitoes triggered increases in the hemocyte populations of transferred mosquitoes, indicating that humoral factors could promote a cellular immune response that protected mosquitoes against subsequent bacterial challenges [5]. In Biomphalaria, plasma transfer experiments demonstrated that soluble humoral factors are released into the hemolymph of primed snails, and that the transfer of such factors activates/regulates the humoral immune response in recipients and confers enhanced antischistosomal immunity against subsequent encounters with the pathogen.
To conclude, we evidence for the first time the molecular basis of innate immune memory in an invertebrate model, and we demonstrate the role of humoral factors in such phenomenon. The previous studies addressing the molecular mechanisms of innate immune memory in invertebrates have mainly suggested the involvement of elevated hemocyte phagocytosis [9,10]. In B. glabrata, innate immune memory seems to be supported by humoral factors that trigger the degeneration/death of the parasite. Moreover the innate immune memory protection was previously shown to decrease with increasing neutral genetic distance between the parasite used for the primo-infection and the one used for the secondary challenge [4]. We hypothesized that better protection against a homologous (vs. heterologous) secondary infection involves the mobilization of specific repertoires of B. glabrata immune receptors to target certain subsets of S. mansoni genotypes. Our present findings prompt us to speculate that the genotype-dependent innate immune memory of B. glabrata may be supported by a diverse repertoire of FREPs and other PRRs (Figs 4 and 5). As previously suggested [22], PRRs might serve as collaborative recognition factors that can be processed as homologous or heterologous multimers, which then act as immune recognition complexes to increase the host’s PRR repertoire and mediating anti-pathogen responses. Once recognized, the pathogen is neutralized by the release of cytotoxic/cytolytic circulating factors. Here, we demonstrate that biomphalysin is highly transcribed following secondary challenge and consumed at the protein level during innate immune memory response, suggesting that biomphalysin may play a major role in neutralizing the pathogen following immune recognition. Clearly, future studies are needed to fully understand how these molecules mediate and regulate the specificity of the innate immune memory defense seen in B. glabrata.
To reconcile mechanisms with phenomena a molecular characterisation of innate immune memory in Biomphalaria/Schistosoma model was undertook. Transcriptomic, proteomic, and functional validation lead us to a new molecular comprehension of innate immune memory processes and prompted us to revisit the artificial dichotomy between innate and memory immunity in invertebrate systems.
|
10.1371/journal.pgen.1005630 | Convergent Evolution During Local Adaptation to Patchy Landscapes | Species often encounter, and adapt to, many patches of similar environmental conditions across their range. Such adaptation can occur through convergent evolution if different alleles arise in different patches, or through the spread of shared alleles by migration acting to synchronize adaptation across the species. The tension between the two reflects the constraint imposed on evolution by the underlying genetic architecture versus how effectively selection and geographic isolation act to inhibit the geographic spread of locally adapted alleles. This paper studies the balance between these two routes to adaptation in a model of continuous environments with patchy selection pressures. We address the following questions: How long does it take for a novel allele to appear in a patch where it is locally adapted through mutation? Or, through migration from another, already adapted patch? Which is more likely to occur, as a function of distance between the patches? What population genetic signal is left by the spread of migrant alleles? To answer these questions we examine the family structure underlying migration–selection equilibrium surrounding an already adapted patch, treating those rare families that reach new patches as spatial branching processes. A main result is that patches further apart than a critical distance will likely evolve independent locally adapted alleles; this distance is proportional to the spatial scale of selection (σ / s m, where σ is the dispersal distance and sm is the selective disadvantage of these alleles between patches), and depends linearly on log(sm/μ), where μ is the mutation rate. This provides a way to understand the role of geographic separation between patches in promoting convergent adaptation and the genomic signals it leaves behind. We illustrate these ideas using the convergent evolution of cryptic coloration in the rock pocket mouse, Chaetodipus intermedius, as an empirical example.
| Often, a large species range will include patches where the species differs because it has adapted to locally differing conditions. For instance, rock pocket mice are often found with a coat color that matches the rocks they live in, these color differences are controlled genetically, and mice that don’t match the local rock color are more likely to be eaten by predators. Sometimes, similar genetic changes have occurred independently in different patches, suggesting that there were few accessible ways to evolve the locally adaptive form. However, the genetic basis could also be shared if migrants carry the locally beneficial genotypes between nearby patches, despite being at a disadvantage between the patches. We use a mathematical model of random migration to determine how quickly adaptation is expected to occur through new mutation and through migration from other patches, and study in more detail what we would expect successful migrations between patches to look like. The results are useful for determining whether similar adaptations in different locations are likely to have the same genetic basis or not, and more generally in understanding how species adapt to patchy, heterogeneous landscapes.
| The convergent evolution of similar phenotypes in response to similar selection pressures is a testament to the power that selection has to sculpt phenotypic variation. In some cases this convergence extends to the molecular level, with disparate taxa converging to the same phenotype through parallel genetic changes in the same pathway, genes, or even by precisely the same genetic changes [1–3]. Convergent adaptation also occurs within species, if different individuals adapt to the same environment through different genetic changes. There are a growing number of examples of this in a range of well studied organisms and phenotypes [4]. Such evolution of convergent phenotypes is favored by higher mutational input, i.e., higher total mutational rate and/or population size [5]. The geographic distribution of populations can also affect the probability of parallel mutation within a species: a widespread species is more likely to adapt by multiple, parallel mutations if dispersal is geographically limited, since subpopulations will adapt via new mutations before the adaptive allele arrives via migration [6]. Standing variation for a trait can also increase the probability of convergence, as this increases the probability that the selective sweep will be soft (beginning from a base of multiple copies), which leads to genetic patterns similar to convergent adaptation [7, 8] whether or not the copies derive from the same mutation,
Intuitively, convergence is also more likely when geographically separated populations adapt to ecologically similar conditions. The probability that convergent adaptations arise independently before adaptations spread between the populations by migration will be larger if these adaptive alleles are maladapted in intervening environments, since such adverse conditions can strongly restrict the spread of locally adapted alleles [9].
One elegant set of such examples is provided by the assortment of plant species that have repeatedly adapted to patches of soil with high concentrations of heavy metals (e.g., serpentine outcrops and mine tailings) [10–13]; the alleles conferring heavy metal tolerance are thought to be locally restricted because they incur a cost off of these patches. Similarly, across the American Southwest, a variety of species of animals have developed locally adaptive cryptic coloration to particular substrates, e.g., dark rock outcrops or white sand dunes [14]. One of the best-known examples is the rock pocket mouse (Chaetodipus intermedius), which on a number of black lava flows has much darker pelage than on intervening areas of lighter rock [15]. Strong predator-mediated selection appears to favour such crypsis [16], and, perhaps as a result of this strong selection against migrants, at least two distinct genetic changes are responsible for the dark pigmentation found on different outcrops [17]. Similar situations have been demonstrated in other small rodent systems [18–20] and in lizards [21].
In this paper, we study this situation, namely, when a set of alleles provide an adaptive benefit in geographically localized patches, separated by inhabited regions where the alleles are deleterious. The main questions are: Under what conditions is it likely that each patch adapts in parallel, i.e., convergently through new mutation, and when is it likely that migration carries these alleles between patches? How easy will it be to detect adaptive alleles that are shared by migration, i.e., over what genomic scale will a population genetic signal be visible?
We work in a model of continuous geography, using a variety of known results and new methods. In the section Establishment of a locally adaptive allele due to mutational influx we develop a simple approximation, Eq (2), for the rate at which patches become established by new mutations. The most important conceptual work of the paper occurs in the section Establishment of a locally adaptive allele due to migrational influx, where we develop an understanding of the process by which alleles move from an existing patch to colonize a novel patch, culminating in Eq (11) for the rate at which this happens. We combine these two results in the section The probability of parallel adaptation between patches to discuss the probability of parallel adaptation, Eq (12). To understand the genomic signal left by adaptations shared by migration, in the section Length of the hitchhiking haplotype, we study the time it will take for an allele to transit between patches, (Eq (18)), and thus the length of haplotype that hitchhikes with it (Eq (19)). Finally, in the section Applications we apply this work to understand the geographic scale of convergent evolution in Chaetodipus intermedius.
Consider a population spread across a landscape to which it is generally well adapted, but within which are patches of habitat to which individuals are (initially) poorly adapted. (When we refer to “patches” it is to these pieces of poor habitat.) Suppose it takes only a single mutational change to create an allele (B) that adapts an individual to the poor habitat type. The required change could be as specific as a single base change, or as general as a knockout of a gene or one of a set of genes. This change occurs at a (low) rate of μ per chromosome per generation, and has fitness advantage sp relative to the unmutated type (b) in these “poor” habitat patches. Outside of these patches the new allele has a fitness disadvantage of sm in the intervening areas, with sp and sm both positive. (Here we define “fitness” to be the intrinsic growth rate of the type when rare.) We assume that the disadvantage sm is sufficiently large that, on the relevant timescale, the allele is very unlikely to fix locally in these intervening regions. (The case where sm = 0 requires a different approach, which we do not treat here.)
We are interested in the establishment of mutations in the “poor” patches by either migration or mutation, which depends on whether the allele can escape initial loss by drift when rare. Therefore, we need not specify the fitness of the homozygote; only that the dynamics of the allele when rare are determined by the fitness of the heterozygote. More general dominance will only make small corrections to the dynamics until initial fixation, with the exception of the recessive case, which we omit. In other words, we follow the literature in treating the diploid model as essentially haploid.
We also assume population density ρ is constant across the range (even in the “poor” patches). The variance in offspring number is ξ2, and that the mean squared distance between parent and child is σ2 (i.e., σ is the dispersal distance). We will deal with migration by immediately appealing to the central limit theorem, treating the total distance traveled across t dispersal events as Gaussian with variance tσ2, and do not discuss the (interesting) cases where rare, long-distance dispersal events are more important (see e.g., [6, 22, 23] for discussion).
We first compute the time scale on which a new B mutations will appear and fix in a single, isolated poor habitat patch in which no B allele has yet become established. As we are interested in patches where local adaptation can occur, we will assume that the patch is larger than the cutoff for local establishment mentioned above.
Let p(x) be the probability that a new mutant B allele that arises at location x relative to the center of the patch fixes within the poor habitat patch. Under various assumptions, precise expressions can be found for p(x) [29], but results will be more interpretable if we proceed with a simple approximation. The total influx per generation of mutations that successfully establish is the product of population density ρ, mutation rate μ, and the integral of p(x) over the entire species range:
λ mut = ρ μ ∫ p ( x ) d x . (1)
This depends in a complicated way on the patch geometry and selection coefficients, but still scales linearly with the mutational influx density ρμ. If we consider a patch of area A, whose width is large relative to σ / 2 s m, then a reasonable approximation is to ignore migration, so that p(x) = pe ≈ 2sp/ξ2 within the patch, and p(x) = 0 otherwise. This approximates the integrand p(x) in Eq (1) by a step function, which will be a good approximation if the patch is large relative the scale over which p(x) goes from 0 to pe, or if p(x) is close to pe at some point and is symmetric about the edge of the patch. We examine this more generally via exact calculation of p(x) in the section Numerical calculation of the probability of establishment.
The rate at which mutations arise and colonize a patch of area A is then
λ mut ≈ 2 s p ρ A μ / ξ 2 , (2)
i.e., the product of mutational influx in the patch (ρAμ) and the probability of establishment (pe ≈ 2sp/ξ2). If this rate is low, then the time (in generations) until a mutation arises that will become locally established within the patch is exponentially distributed with mean 1/λmut. Assuming that once a mutation becomes established it quickly reaches its equilibrium frequency across the patch, the time scale on which new patches become colonized by the B allele from new mutation is therefore approximately 1/λmut.
Now suppose that there are two patches separated by distance R (i.e., the shortest distance between the two is R). If the B allele has arisen and become established in the first patch, but has not yet appeared in the second, we would like to know the time scale on which copies of B move between patches by migration (supposing that no B allele arises independently by mutation in the meantime). To determine this, we study the fine-scale genealogy of alleles that transit between patches, obtaining along the way other useful information about the genealogy of B alleles. Doing this we arrive at Eq (11) for the rate at which an allele established in one patch spreads to a neighboring one by migration.
Migration–selection balance ensures that there will be some B alleles present in the regions where they are disadvantageous, but only rarely, far away from the patch where B is established. Denote the expected frequency of allele B at displacement x relative to the patch by q(x), and assume that the shape of the patch is at least roughly circular. Following [24] or [9], one can write down a differential equation to which q(x) is the solution, and show that q(x) decays exponentially for large ∣x∣, with a square-root correction in two dimensions:
q ( x ) ≈ C ( | x | 2 s m / σ ) − ( d − 1 ) / 2 exp ( − | x | 2 s m / σ ) for large | x | , (3)
where d is the dimension (d = 1 or 2), and C is a constant depending on the geographic shape of the populations and the selection coefficients. In applications we fit C to the data; to get concrete numbers from theory we take C = 1 if necessary.
As J. Hermisson pointed out in comments on an earlier draft, this functional form has a simple intuition: local migration leads to the exponential decay, since if each migrant at distance x produces an average of c descendants that make it to distance x + 1, then in one dimension, the number of migrants must decay as exp(−cx); and the square-root term in two dimensions enters because the available area grows with x. The “renewal” aspect of this same argument suggests that for Eq (3) to hold, ∣x∣ must be larger than a few multiples of σ.
These asymptotics fit simulations quite well, as shown in Fig 1. To be clear about the assumptions implicit here, below we provide a derivation of Eq (3) in section The equilibrium frequency, as well as justification for the asymptotics below in section Asymptotic solution for the equilibrium frequency. In one dimension, the equation can be solved to give q(x) in terms of Jacobi elliptic functions [37]; see the Supporting Information.
This expected frequency q(x) is the time-averaged occupation frequency, i.e., the total number of B alleles found across T generations near location x, per unit area, divided by T. If, for instance, q(x) = .01 and the population density is ρ = 10 individuals per unit area, then in a survey tract of unit area around x we expect to find one individual every 10 generations—or, perhaps more likely, 10 individuals every 100 generations. This points to an important fact: close to the original patch, the “equilibrium frequency” q(x) describes well the density of the B allele at most times, but far away from the patch, the equilibrium is composed of rare excursions of families of B alleles, interspersed by long periods of absence. An example of one of these rare excursions is shown in Fig 2. The relevant time scale on which B alleles migrate between patches is given by the rate of appearance of such families.
This suggests decomposing the genealogy of B alleles into families in the following way. First, consider the genealogy of all B alleles that were alive at any time outside of the original patch. This is a collection of trees, each rooted at an allele living outside the patch whose parent was born inside the patch. Next, fix some intermediate distance r0 from the established patch, and erase from the genealogy every allele that has no ancestors living further away than r0 to the patch. This fragments the original set of trees into a collection of smaller trees that relate to each other all B alleles living outside of r0 at any time, and some living inside of r0; we refer to these trees as “families”. If r0 is large enough that there are not too many families and interactions between family members are weak, then these families can be approximated by a collection of independent spatial branching processes whose members are ignored if they enter the original patch, illustrated in Fig 3. (This can be made formal in the limit of large population density, also taking r0 large enough that the chance of reaching the original patch is small.) The opportunity for adaptation depends on how often these families of B alleles encounter the new patch. Suppose that the area occupied by the new patch is S. We can learn about the rate of arrival of families at S from the time-averaged number of B alleles expected in S were it not a patch (i.e., if B was still maladaptive in S), which from Eq (3) is
ρ q ( S ) = ( outflux of families ) × ( mean occupation in new patch per family ) , (4)
where q(S) = ∫S q(y)dy. The quantity we wish to compute, the effective rate at which families of migrant B alleles establish in the new patch, is
λ mig ( S ) = ( outflux of families ) × ( probability a family establishes in new patch ) . (5)
The quantities on the right-hand side depend implicitly on r0, but their product does not.
Now that we have expressions for the mean rates of adaptation by new mutation, Eq (2), and by migration from an already colonized patch, Eq (11), it seems helpful to step back and review the assumptions underlying the asymptotic results we have used, or will use below. Our results should hold exactly in the limit of large, circular patches sufficiently far apart, large population density, and small selection coefficients of equal magnitude.
To be more precise, the mutation rate results most obviously apply if sm ≈ sp, if the patch diameter is large relative to σ / min ( s m , s p ), and the patch is not too far from circular. If sm ≪ sp then a strip of width σ / s m should be added to the outside of the patch in computing A for Eq (2), and if sm ≫ sp, a strip of width σ / s p should be subtracted from the patch. As for the migration rate, we assume that each patch is large enough to support a stable population of B alleles (A 1 / d > ( σ / 2 s p ) tan - 1 ( s m / s p )). The geometry and size of the patch will also affect the approximation of Eq (48). Next, in using Eq (3), we assume that the inter-patch distance is large relative to the characteristic length (R > σ / s m), and that local drift is not so strong that the equilibrium frequency is at least approximately attained (using Wright’s effective neighborhood size, sm ≫ 1/(ρσd)). The last requirement is necessary because if the B allele fixes in large neighborhoods where it is deleterious, we cannot approximate its dynamics via a branching process. We also neglect the time for migration-selection equilibrium to be reached. As discussed above, we also assume that migration to a new patch takes place over a number of generations; if there are sufficient rare, long-distance migration events that would move between patches in a single hop, this would require a different analysis.
To test the robustness of our results to the various approximations we used, we used individual-based simulations on a one-dimensional lattice of 501 demes, with ρ haploid individuals per deme, dispersal to nearby demes with σ = 0.95, and run for 25,000 generations. More details are given below in section Simulation methods, and the number of simulations used and parameter values are given in supplementary S1 and S2 Tables. To estimate the mean time until adaptation by mutation, we used one centrally located patch of 99 demes and a mutation rate of μ = 10−5, and to estimate the mean time until adaptation by migration, we used two centrally located patches of 99 demes separated by a varying number of demes, with one patch initialized to have the B allele at frequency 0.8. In each case, we measured the “time to adaptation” as the amount of time until at least 100 B alleles were present in the previously unadapted patch. Fig 4 summarizes how the results compare to theory, excluding parameter combinations that violate the assumptions discussed above, or where a majority of simulations did not adapt by 25,000 generations. S1 and S2 Figs show all times, and depictions of typical simulation runs are shown in S3, S4, S5, S6, S7, S8, S9 and S10 Figs.
The agreement is reasonable for both. The theoretical value 1/λmut underestimates the mean time to mutation by a factor of around 2 that increases slightly with sm. This is to be expected for two reasons: First, we compute the time to reach 100 B alleles, while theory predicts the time until the progenitor of those 100 B alleles arose. Second, the expression for λmut neglects effects near the boundary of the patch, and the larger sm is, the harder it is for mutations that arise near the edge of the patch to establish. The theoretical values 1/λmig again has a margin of error of a factor of about 2. S10 Fig shows simulations at more parameter values.
We now turn to the question of whether the separated patches adapt by parallel genetic changes or by the spread of a migrant allele between the patches. As only a single mutation is required for individuals to adapt to the poor habitat patch, subsequent mutations that arise after an allele becomes established in the patch gain no selective benefit. Similarly, an allele introduced into a patch by migration will not be favored by selection to spread, if the patch has already been colonized. Therefore, mutations selectively exclude each other from patches, over short time scales, and will only slowly mix together over longer time scales by drift and migration, an assumption we also made in [6]. In reality, different mutations will likely not be truly selectively equivalent, in which case this mixing occurs over a time-scale dictated by the difference in selection coefficients.
We assume that once a B allele is introduced (by migration or mutation) it becomes established in the poor habitat patch rapidly if it escapes loss by demographic stochasticity. Under this approximation, and the “selective exclusion” just discussed, after one of the patches becomes colonized by a single B allele, the other patch will become colonized by migration or mutation, but not by both. As such, the question of how the second patch adapts simply comes down to whether a new mutation or a migrant allele is the first to become established in the second patch. To work out how adaptation is likely to proceed, we can compare Expressions (2) and (11) above for the effective rates of adaptation by new mutation and by migration. We work in one dimension, as the square root term appearing for two dimensions is relatively unimportant.
We first consider the order of magnitude that our parameters need to be on in order for adaptation via mutation or migration to dominate. Let γ = min(1, sm/pe) and w = A/A′—since A is the area of the not-yet-adapted patch and A′ is the area of its closest σ / 2 s m strip, w is approximately the width of the patch in units of σ / 2 s m. Effective migration and mutation rates will be on the same order if A μ ≈ 2 A ′ γ s m exp ( - R 2 s m / σ ), where R is the distance between the patches. In other words, the migrational analogue of “mutational influx” (μρA) is 2 ρ A ′ γ s m exp ( - R 2 s m / σ ), which depends fairly strongly on the selective disadvantage sm between patches. Equivalently, the rates are roughly equal if R / σ = log ( 2 γ s m / ( w μ ) ) / 2 s m, which gives the critical gap size past which adaptation will be mostly parallel in terms of selection coefficients, patch width, and mutation rate.
If we take μ = 10−5, the patch width to be ten times the length scale σ / 2 s m so w = 10 (and −log(wμ) ≈ 9.2), and sm > pe so that γ = 1, then adaptation is mostly parallel for patches separated by gaps larger than R / σ > ( 9 . 2 + log ( 2 s m ) ) / 2 s m. If the selective pressure is fairly strong (say, sm = .05), then for convergence to be likely, the distance between patches must be at least 21.8 times the dispersal distance (i.e., R/σ > 21.8). If sm is much smaller, say sm = .001, the required distance increases to R/σ > 67.
If the mutation rate is higher—say, μ = 10−3—the required separation between patches is only reduced to R/σ > 7.3 with sm = .05. If sm = .001, then with this higher mutational influx this calculation predicts that mutation will always be faster than migration—however, this should be taken with caution, since as discussed above, this model does not hold if R is of the same order as σ or if sm is small enough that local drift is more important.
We can go beyond these rough calculations to find the probability of parallel adaptation if we are willing to take our approximations at face value. Doing so, the time until the first of the two patches is colonized by the B allele will be approximately exponentially distributed with mean 1/(2λmut). Following this, the time until the second patch is subsequently colonized (via either migration or new mutation) will be exponentially distributed with mean 1/(λmut + λmig). Both these rates scale linearly with population density (ρ) and the probability of establishment of a rare allele (pe ≈ 2sb/ξ2, for pe > sm), so that the overall time to adaptation is robustly predicted to increase with offspring variance (ξ2) and decrease with population density and beneficial selection coefficient. Furthermore, the probability that the second adaptation is a new mutation, i.e., the probability of parallel adaptation, with γ = min(1, sm/pe), is
λ mut λ mut + λ mig ( R ) = w μ / ( 2 s m ) w μ / ( 2 s m ) + C γ ( R 2 s m / σ ) − ( d − 1 ) / 2 exp ( − 2 s m R / σ ) , (12)
so that the probability of parallel mutation should increase approximately logistically with the distance between the patches, on the spatial scale σ / 2 s m.
We tested this using the same simulations as Fig 4, by using the empirical distributions of the respective times to adaptation to estimate the probability, for each parameter combination, that a new, successful mutation appears before a successful migrant arrives from another, already adapted patch. The results are compared to Eq (12) in Fig 5.
If a patch adapts through new mutation or a rare migrant lineage, the genomic region surrounding the selected site will hitchhike along with it [42], so that at least initially, all adapted individuals within a patch share a fairly long stretch of haplotype. Pairs of adapted individuals within a patch will initially share a haplotype of mean genetic length of about log(ρAsp)/sp around the selected site, as long as the patch is reasonably well mixed by dispersal (otherwise see [43]). This association gets slowly whittled down by recombination during interbreeding at the edge of the patch, but there will always be longer LD nearby to the selected site [44].
When an already adapted patch colonizes another through migration, the newly colonized patch will inherit a long piece of haplotype around the selected site from the originating patch. The genetic length of this haplotype is roughly inversely proportional to the number of generations the allele spends “in transit”, because while very rare, the allele will be mostly in heterozygotes, and so each generation provides another opportunity for a different haplotype to recombine closer to the transiting B allele. A large, linked haplotype may still arrive and fix in the new patch, in which case the haplotype has literally hitchhiked across geography! Fig 6 shows a simulation of such an event, including the lineage that founds an adaptive population on a second patch, and the length of the haplotype shared between this lineage and one in the original patch. The time that the lineage is outside the region where the B allele is common (dark red in Fig 6), the haplotype that accompanies it is broken down rapidly. After the lineage establishes on the patch, the rate of decay of the haplotype is slowed significantly, since most others with which it recombines have similar haplotypic backgrounds.
Above we argued that a good model for this transit is a single Brownian trunk lineage surrounded by a cloud of close relatives of random size K whose chance of surviving until t generations is 1 − ke(t). Consider a single such family, and let τ be the (random) time at which it first hits the new patch, with τ = ∞ if it never does. We assume that the length of the hitchhiking segment depends only on the time spent by the trunk lineage of the first successful migrant “in transit” between the patches, i.e., τ conditioned on τ < ∞. In so doing, we neglect recombination between B alleles (since they are at low density in transit), and the possibility that more than one successful migrant family is in transit at once (so that faster migrants would be more likely to have arrived first).
Since each generation spent in transit provides an opportunity for recombination, if recombination is Poisson, the length of the haplotype (in Morgans) initially shared between populations on each side of the selected locus is exponentially distributed with mean τ. Therefore, if L is the length of hitchhiking segment on, say, the right of the selected locus, then
P { L > ℓ } = E [ e - ℓ τ | τ < ∞ ] . (15)
Computing this depends on 1 − ke(t), the probability a family survives until t. Since 1 - k e ( t ) ≃ e - s m t / E [ K ], we approximate the lifetime distribution of a family by an exponential with mean 1/sm. We can then use standard results on hitting times of d-dimensional Brownian motion that is killed at rate sm (see [45] 2.2.0.1 and 4.2.0.1). In particular, if the patch is circular with radius w and lies at distance R from the already adapted patch, then
E [ e - ℓ τ ] = e - R 2 ( s m + ℓ ) / σ d = 1 K 0 ( ( R + w ) 2 ( s m + ℓ ) / σ ) K 0 ( w 2 ( s m + ℓ ) / σ ) d = 2 , (16)
where K0 is a modified Bessel function of the second kind. We are interested in lineages that manage to reach the patch before being killed, i.e., having τ < ∞, which occurs with probability P { τ < ∞ } = lim ℓ → 0 E [ exp ( - ℓ τ ) ].
To keep the expressions simple, in the remainder of this section we only compute quantities for one dimension. By Bayes’ rule,
P { L > ℓ } = exp - R σ 2 ( ℓ + s m ) - 2 s m . (17)
This can be differentiated to find that the expected transit time is
E [ τ ∣ τ < ∞ ] = ( R 2 s m / σ ) × 1 / ( 2 s m ) (18)
and that Var [ τ ∣ τ < ∞ ] = ( R 2 s m / σ ) × 1 / ( 2 s m ) 2. (A saddle point approximation provides an alternative route to these expressions.)
The form of Eq (17) implies that if Y is an exponential random variable with rate R 2 / σ, then L has the same distribution as ( Y + s m ) 2 - s m. Furthermore, the expected length of shared hitchhiking haplotype is
E [ L ] = σ 2 s m / R + σ 2 / R 2 (19)
and Var [ L ] = 2 s m σ 2 / R 2 + 4 σ 3 2 s m / R 3 + 5 σ 4 / R 4. For two dimensions, asymptotics of Bessel functions show that L has the same tail behavior, but how other properties might change is unclear.
As a rule of thumb, Eq (18) means that families who successfully establish move at speed 2 s m σ towards the new patch—if sm, the strength of the selection against them, is smaller, the need to move quickly is less imperative. Then, Eq (19) means that the haplotype length is roughly the length one would expect given the mean transit time, so more weakly deleterious transiting alleles arrive with them shorter haplotypes. Note that we have become used to seeing σ divided by 2 s m, rather than multiplied; it appears here because σ / 2 s m is a length scale, and to convert it to a speed this is divided by 1/2sm.
Coat color in the rock pocket mouse Chaetodipus intermedius is a classic example of local adaptation [14, 15]. These mice live on rocky outcrops throughout the southwestern United States and northern Mexico, and generally have a light pelage similar to the predominant rock color. However, in some regions these mice live on darker substrates (e.g., old lava flows), and in these patches have much darker pigmentation, presumably to avoid visual predators. Some of the largest patches of dark rock range from 10km to 100km wide and lie 50–400km from each other, and dark-colored populations of C. intermedius have been found on many of these. (However, patches of all sizes occur across all scales in a heterogeneous manner.) [46] demonstrated that on one of these flows (Pinacate), much of the change in coloration is due to an allele at the MC1R locus. This dark allele differs from the light allele by four amino acid changes, and has a dominant or partially dominant effect depending on the measure of coat color. The Pinacate allele is not present in a number of other populations with dark pelage, suggesting these populations have adapted in parallel [17, 46]. However, [47] reasoned that, elsewhere in the range, multiple close dark outcrops may share a dark phenotype whose genetic basis has been spread by migration despite intervening light habitat.
A key parameter above was the dispersal distance divided by the square root of strength of selection against the focal allele between patches, σ / s m. [48] studied the frequency of the dark MC1R allele and coat color phenotypes, at sites across the (dark) Pinacate lava flow and at two nearby sites in light-colored rock. On the lava flow the dark MC1R allele is at 86% frequency, while at a site with light substrate 12km to the west (Tule), the frequency is 3%. The dark allele was entirely absent from Christmas pass, a site with light substrate 7km north of Tule, and 3km further from the lava flow. In the other direction, the dark MC1R allele was at 34% at a site with light substrate 10km to the east of the flow (O’Neill). Note that such apparent asymmetry is expected, since as noted above the migration–selection equilibrium can be highly stochastic. These numbers give us a sense of a plausible range of the parameters. Assuming the dark allele is at 50% frequency at the edge of the lava flow, we can fit Formula (3) to these frequencies (similar to [48]). Doing this for Tule we obtain σ / s m ≈ 3 km, and for O’Neill, σ / s m ≈ 30 km, giving us a range of cline widths. We also need estimate of sm. Using σ ≈ 1km [49, 50], these cline widths imply that sm = 1/9 and 1/900 are reasonable values.
The mutational target size μ for the trait is unclear. While the Pinacate dark haplotype differs from the light haplotype at four amino acid residues, it is likely that not all of these changes are needed for a population to begin to adapt. Also, there a number of genes besides MC1R at which adaptive changes affecting pigmentation have been identified in closely related species and more broadly across vertebrates [51]. To span a range of plausible values, we use a low mutation rate of μ = 10−8 (a single base pair), and a high mutation rate μ = 10−5 (a kilobase). Finally, we set A = 100km2 (roughly the size of the Pinacate patch). In Fig 7 we show the dependence of the probability of parallel mutation on the distance between lava flow patches using these parameters, showing that parallel mutation should become likely over a scale of tens to a few hundred kilometers between patches.
Given the large selection coefficient associated with the dark allele on the dark substrate, we expect the initial haplotype associated with either a new mutation or migrant allele to be large. Fig 7 also shows how long the founding haplotype shared between populations is expected to be, from Eq (19). The initial length can be quite long between geographically close patches (tens of kilometers). However, for the wider cline width (σ / s m = 30 km), adaptation by migration can still be likely for patches 100km apart, but the shared basis may be hard to detect, as the length of shared haplotype can be quite short.
This paper is an investigation into the basic question: What is the spatial resolution of convergent local adaptation? In other words, over what spatial scale of environmental patchiness will the process of adaptation develop independent solutions to evolutionary problems? The answer to this depends most strongly on σ / s m, the dispersal distance divided by the square root of the strength of selection against the allele between patches. It depends much more weakly on the selective benefit within the patches or (perhaps surprisingly) the population density, although these two factors will determine the time-scale over which adaptation will occur (and note that population density could affect the selection coefficients). This is in contrast with models of panmictic populations [8, 52, 53] and geographically spread populations adapting to homogeneous selection pressures [6], where the probability of multiple, independently arising adaptive alleles increases with the population size. However, in all of these models the dependence on the beneficial selection coefficient is absent or weak, due to the fact that selection both aids establishment of new alleles and the spread of existing alleles (but see [53] for the complications of varying population sizes).
We have also shown that while weaker selection against alleles will make sharing of adaptations between patches easier, it also makes it harder to spot such sharing, since the lucky alleles that manage to colonize new patches move slower, and thus carry a shorter shared haplotype. This issue is amplified by the fact that the length of haplotype shared within patches decays over time, potentially making the identification of shared adaptations to old selection pressures difficult.
Perhaps the most useful rule-of-thumb quantities we found were the following. The effective rate of migration into an as-yet-unadapted patch from an already-adapted patch distance R away—the analogue of the mutational influx μρA—is 2 ρ A ′ s m exp ( - R 2 s m / σ ). Equivalently, the critical gap size between patches past which adaptation is likely independent is R = ( σ / 2 s m ) log ( 2 s m / ( w μ ) ). Finally, successfully transiting migrant lineages move at rate σ 2 s m, and so shared haplotype lengths between patches will be of order R / ( σ 2 s m ).
In developing the set of approximations in the paper we have ignored a number of complicating factors. We now briefly discuss these.
We have focused on relative rates of adaptation, since in applications where adaptation has occurred, the question is whether adaptations in distinct patches have appeared independently or not. However, any adaptation that does occur may have to make use of standing variation, if mutation rates are low. The case of a panmictic population was studied by [8], and we study the case of a continuous, spatial population in [54]. If parallelism in local adaptation of the sort we study here is due to standing variation rather than new mutation, then the dynamics of adaptation should not depend strongly on migration patterns (but the initial spatial distribution of standing variation may).
We have mostly ignored the issue of dominance by dealing with essentially haploid models, and appealing to the fact that the dynamics we study occur where the mutation is rare, and hence mostly present only in heterozygotes. Our results should hold as a good approximation to dominant and partially dominant alleles (with sm the selection against heterozygotes). If, however, the mutation is recessive, then it is essentially neutral where rare, and so would encounter much less resistance to spreading between patches. The shape of the cline obtained is given by [24]. This is counteracted, however, by the increased difficulty with which the mutation would establish once arriving in a new patch, if the beneficial effect is also recessive. As such it is not clear what our intuition should be about the contribution of recessive alleles to adaptation via migration. Further work is needed to put empirical observations of local adaptation by recessive alleles in a theoretical context. It is similarly unclear how the model should extend to a polygenic trait.
To provide context for the results on shared haplotype length in section Length of the hitchhiking haplotype it is important to also understand the process by which haplotypes are whittled down within patches. The initial haplotype that sweeps within a patch will be dispersed over time by recombination. Likewise, the haplotype that is shared between patches coadapted by migration will also break down (Eq 19). However, a long time after the initial sweep, we may still expect to find individuals within the patch sharing longer haplotypes around the selected locus than with individuals elsewhere, since selection against migrants decreases mean coalescence times within the patch near the selected locus. The literature on clines (e.g., [44]) has important information, but more work is needed to provide robust estimates for these processes. Questions about the genomic length-scale of signals of sweeps shared by migration have also been addressed in discrete population settings [55, 56], reviewed in [57]. This work has shown that the length of the shared swept haplotype is often significantly shorter than the sweep within each patch, resulting in a pattern of shoulders of elevated FST between adapted populations some distance away from the shared selected allele. It would be of interest to see how similar patterns can arise in a continuous population setting, as a way of uniting these results.
We have also ignored the possibility of very long distance migration, instead focusing on local dispersal (hence Gaussian by the central limit theorem). However, dispersal distributions can be very heavy tailed, with a small fraction of individuals moving very long distances indeed [22, 58]. In addition, over long time-scales, very rare chance events (mice carried off by hurricanes and the like; [59, 60]) could play a role in spreading migrant alleles if adaptation by other means is sufficiently unlikely. Such tail events could greatly increase the probability of shared adaptation above that predicted by our model. Furthermore, if adaptive alleles do move between distant patches via rare, long distance migration then they will be associated with a much longer shared haplotype than predicted by Eq (19). As such, we view our results as a null model by which the contribution of long distribution migrants to adaptation could be empirically judged.
We have studied circular patches of habitat at long distances from each other. Real habitat geometry can be much more complex, e.g., with archipelagos of patches of varying sizes, or patches connected by long, skinny corridors, for instance. The work of [61] comes closest to a general theory of balanced polymorphisms in such habitats. It is possible that our techniques could be applied in their much more general setting, as both are based, fundamentally, on branching process approximations. It is also interesting to think about the probability of convergent adaptation to continuously varying environments, e.g. replicated environmental clines.
The falling cost of population genomic sequencing means that we will soon have the opportunity to study the interplay of adaptation with geography and ecology across many populations within a species. Our work suggests that even quite geographically close populations may be forced to locally adapt by repeated, convergent, de novo mutation when migration is geographically limited and selective pressures are divergent. Thus, systems where populations have been repeatedly subject to strong local selection pressures may offer the opportunity to study highly replicated convergent adaptation within a similar genetic background [1]. Such empirical work will also strongly inform our understanding of the ability of gene flow to keep ecologically similar populations evolving in concert [62]. Our results suggest that adaptation to shared environments is certainly no guarantee of a shared genetic basis to adaptation, suggesting that rapid adaptation to a shared environment could potentially drive speciation if the alleles that spread in each population fail to work well together [63].
First we briefly describe the simulations we used for illustration and validation (the R code used is provided in S1 Scripts and at http://github.com/petrelharp/spatial_selection). We simulated forward-time dynamics of the number of alleles of each type in a rectangular grid (either one- or two-dimensional) of demes with fixed size N. Each generation, each individual independently chose to reproduce or not with a probability r depending on her type and location in the grid; locally beneficial alleles were more likely to reproduce. Each extant individual then either remained in the same location with probability 1 − m or else migrated a random number of steps in a uniformly chosen cardinal direction; for most simulations m = 0.2 and the probability of migrating k steps was proportional to 2−k for 1 ≤ k ≤ 5. In 2D, diagonal steps were also used. This gave us values of σ = 0.95 deme spacings in 1D and σ = 0.74 in 2D. Once migrants were distributed, each deme was uniformly resampled back down to N individuals. (Although we described the simulation in terms of individuals, we kept track only of total numbers in an equivalent way.)
The base probability of reproduction in each generation in simulations for type b alleles was r = 0.3; this was then multiplied by 1 + s to get the probability of reproduction for type B, where the value of s is either sm or sp depending on the individual’s location. This determines the values of sm and sp reported in the figures, and do not depend on the basic rate of reproduction. However, to obtain values for sp and sm when comparing theory to simulation, we computed the rate of intrinsic growth, i.e., the s so that the numbers of B alleles when rare would change by est after t generations in the absence of migration. (The resulting values are close to the first notion of s, but give better agreement with theory, which uses the second definition.)
To sample lineages, we first simulated the population dynamics forwards in time, then sampled lineages back through time by, in each generation, moving each lineage to a new deme with probability proportional to the reverse migration probability weighted by the number of B alleles in that deme in the previous time step. If more than one lineage was found in a deme with n alleles of type B, then each lineage picked a label uniformly from 1 … n, and those picking the same label coalesced. Since reproduction is Poisson, this correctly samples from the distribution of lineages given the population dynamics.
When rare, copies of a new mutant allele are approximately independent and experience a uniform selective benefit; and can therefore be treated as a branching process. Furthermore, whether or not a new, beneficial mutation establishes or is lost to demographic stochasticity is determined by this initial phase where it is rare. Fortunately, the probability that a branching process dies out can be found as a fixed point of the generating function of the process [38]. Therefore, we calculated explicitly the generating function for a spatial branching process with nearest-neighbor migration on a one-dimensional lattice and a Poisson number of offspring with mean 1 + s, where s could vary by location, and iterated this forward to convergence to obtain 1 − p(x), the probability a single mutation appearing at x would fail to establish. We considered two situations: where s is a step function, and where it has a linear transition. These solutions are shown, and parameters described, in Fig 8.
Here (and at other parameter choices) we see that the probability of establishment p(x) goes to the equilibrium value (approximately pe = 2s/ξ2) within the patch; the transition is fairly symmetrical about the edge of the patch, even if the edge of the patch is not sharp. Additional experimentation indicated that the fit remains equally good for other parameter values, even if migration can move further than one deme and offspring numbers are not Poisson. This lends credence to our approximation that the integral of p(x) over the entire range is close to pe multiplied by the area of the patch.
For completeness, and clarity as to the scalings on the relevant parameters, here we provide a derivation of the differential equations referred to above Eq (3), and establish the asymptotics given in that equation. One route to the “equilibrium frequency” of the allele outside the range where it is advantageous is as follows; see [9] or [29] (or [64] or [65] or [24]) for other arguments in equivalent models, and see [66] and/or [67] for a general framework for the stochastic processes below.
Suppose that the population is composed of a finite number of small demes of equal size N arranged in a regular grid, and that selection (for or against) the allele is given by the function s(x), with x denoting the spatial location. Each individual at location x reproduces at random, exponentially distributed intervals, producing a random number of offspring with distribution given by X who then all migrate to a new location chosen randomly from the distribution given by x + R, where they replace randomly chosen individuals. If x + R is outside of the range, then they perish. Each individual’s time until reproduction is exponentially distributed: the reproduction rate is 1 if it carries the original allele, or is 1 + s(x) if it carries the mutant allele. Suppose that the number of offspring X has mean μ; the variance of X will not enter into the formula (but assume X is well-behaved). Also suppose that the migration displacement R has mean zero and variance σ2; in more than one dimension, we mean that the components of the dispersal distance are uncorrelated and each have variance σ2.
Let Φ t N ( x ) be the proportion of mutant alleles present at location x at time t, and Φt(x) the process obtained by taking N → ∞ (which we assume exists). Denote by δx a single unit at location x, so that e.g. Φ t N + δ x / N is the configuration after a mutant allele has been added to location x. For 0 ≤ ϕ ≤ 1, we also denote by X ¯ ϕ the random number of mutant alleles added if X new offspring carrying mutant alleles replace randomly chosen individuals in a deme where the mutant allele is at frequency ϕ (i.e. hypergeometric with parameters (X, Nϕ, N(1 − ϕ))); similarly, X ˜ ϕ is the number lost if the new offspring do not carry the allele (i.e. hypergeometric with parameters (X, N(1 − ϕ), Nϕ)). (We like to think of Φ t N as a measure, but it does not hurt to think of ΦN as a vector; we aren’t providing the rigorous justification here.) Then the above description implies that for any sufficiently nice function f(Φ) that
∂ ∂ t E f ( Φ t N ) = N ∑ x E ( 1 + s ( x + R ) ) Φ t N ( x + R ) f Φ t N + X ¯ Φ t ( x + R ) N δ x - f ( Φ t N ) + E 1 - Φ t N ( x + R ) f Φ t N - X ˜ Φ t ( x + R ) N δ x - f ( Φ t N ) (20)
= μ ∑ x E ∂ ϕ ( x ) f ( Φ t ) Φ t ( x + R ) - Φ t ( x ) + s ( x + R ) Φ t ( x + R ) ( 1 - Φ t ( x ) ) + O 1 N . (21)
In the final expectation, R and Φ are independent. This follows by taking first-order terms in 1/N in the Taylor series for f, and the fact that E [ X ¯ ϕ ] = ϕ μ and E [ X ˜ ϕ ] = ( 1 - ϕ ) μ. We can see two things from this: First, since this is a first-order differential operator, the limiting stochastic process Φ obtained as N → ∞ is in fact deterministic (check by applying to f(Φ) = Φ(x)2 to find the variance). Second, if we want to rescale space as well to get the usual differential equation, we need to choose Var[R] = σ2 and s(x) to be of the same, small, order; this is another way of seeing that σ / s is the relevant length scale (as noted by [9]). More concretely, suppose that the grid size is ϵ → 0, that Var[R] = (σϵ)2, and that the strength of selection is s(x)ϵ, suppose that Φt(x) is deterministic and twice differentiable, and let ξ(t, x) = Φt/ϵ(x); then the previous equation with f(Φ) = Φ(x) converges to the familiar form:
∂ t ξ ( t , x ) = μ σ 2 2 ∑ k = 1 d ∂ x k 2 ξ ( t , x ) + s ( x ) ξ ( t , x ) ( 1 - ξ ( t , x ) ) . (22)
Here we have taken first the population size N → ∞ and then the grid size ϵ → 0; we could alternatively take both limits together, but not if ϵ goes to zero too much faster than N grows. One reason for this is that at finite N, the process Φt is an irreducible finite-state Markov chain with absorbing states at 0 and 1; therefore, the inevitable outcome is extinction of one type or another, which is not the regime we want to study.
In one dimension, we are done (and discuss exact solutions in S1 Text); in higher dimensions, we are more interested in the mean frequency at a given distance r from a patch. If we take a radially symmetric patch centered at the origin (so s only depends on r), and let ξ(t, r) denote the mean occupation frequency at distance r, then the polar form of the Laplacian in d dimensions gives us that Eq (22) is
∂ t ξ ( t , r ) = μ σ 2 2 ∂ r 2 ξ ( t , r ) + σ 2 d - 1 2 r ∂ r ξ ( t , r ) + s ( r ) ξ ( t , r ) ( 1 - ξ ( t , r ) ) . (23)
A radially symmetric equilibrium frequency ξ(t, x) = q(∣x∣), with s(r) = −s < 0 for all r > r0, solves for r0 < r < ∞,
∂ r 2 q ( r ) + d - 1 r ∂ r q ( r ) - 2 s σ 2 q ( r ) ( 1 - q ( r ) ) = 0 (24)
lim r → ∞ q ( r ) = lim r → ∞ ∂ r q ( r ) = 0 0 < q ( r ) < 1 (25)
Since q(r) → 0 as r → ∞, so q(r) (1 − q(r)) ≈ q(r), it can be shown that the true equilibrium frequency q is close, for large r, to the solution to
∂ r 2 u ( r ) + d - 1 r ∂ r u ( r ) - 2 s σ 2 u ( r ) = 0 (26)
lim r → ∞ u ( r ) = lim r → ∞ ∂ r u ( r ) = 0 . (27)
This has general solution given by a modified Bessel function: using [68] 8.494.9, the general solution is
u ( r ) = C ′ ( r - r 1 ) ( 2 - d ) / 2 K ( 2 - d ) / 2 ( r - r 1 ) 2 s / σ , (28)
where C′ and r1 are chosen to match boundary conditions. Asymptotics of Bessel functions ([68], 8.451.6) then imply that
q ( r ) ≈ C r ( 1 - d ) / 2 exp - r 2 s / σ + O ( 1 / r ) , (29)
where C is a different constant.
Here we make a more precise argument to back up Expression (11) for the migration rate. The argument made above in section Heuristics applies to general migration mechanisms, since it relies only on a decomposition of the migrant families upon hitting the new patch; but it is also imprecise in subtle ways that are difficult to formalize. Here we take a somewhat different tack, supposing that it suffices to model the spatial movement of a migrant family by following only the motion of the “trunk” (i.e., the red line in Fig 3), and supposing this motion is Brownian, with variance σ2 per generation. (Recall σ is the dispersal distance.) We then use facts about Brownian motion and branching processes, to compute more precise versions of Eqs (4) and (5).
We are approximating the dynamics of the focal allele in the region further away than r0 from the patch as the sum of independent migrant families, each of whose dynamics are given by a spatial branching process (as depicted in Fig 3). Call B(r0) the region closer than r0 to the patch, and ∂B(r0) its boundary. Denote by γ(x) the mean rate of outflux of migrant families from a point x ∈ ∂B(r0), i.e., the time-averaged density of individuals near a point x in ∂B(r0) that are the founders of new migrant families. Expressions (4) and (5) are a simple product of the “outflux of families”, when in fact they should be an integral of γ(x) over possible locations. However, it will turn out that the integrand of Eq (5) is well-approximated by a constant multiple of the integrand of Eq (4).
First consider Eq (4) for the equilibrium frequency. Suppose that Z is one such spatial branching process as above in section The genealogy of migrant families, started at time 0 with a single individual at x, and write S for the new patch. The mean occupation measure of Z in the region S, which we denote u(x, S), can be thought of informally as the expected total number of offspring of a family beginning at x that ever live in S. Let q(S) = ∫S q(y)dy denote the total equilibrium frequency in S. This is decomposed in Expression (4) as the sum of mean occupation densities of a constant outflux of branching processes from ∂B(r0):
ρ q ( S ) = ∫ ∂ B ( r 0 ) γ ( x ) u ( x , S ) d x . (30)
Now we will decompose u(x, S) under the assumption that the marginal distribution of the spatial motion of a single lineage is Brownian. Let Bt be a Brownian motion with variance σ2, and τ† an independent Exponential(sm) time. The mean occupation time of Z spent in a region S is,
u ( x , S ) = E ∫ 0 ∞ Z t ( S ) d t (31)
= ∫ 0 ∞ E [ Z t ] p t ( x , S ) d t (32)
= ∫ 0 ∞ e - s m t p t ( x , S ) d t (33)
= ∫ 0 ∞ P x { τ † > t & B t ∈ S } d t (34)
= E x ∫ 0 τ † 1 S ( B t ) d t , (35)
where P x gives probabilities for Brownian motion begun at x (i.e., B0 = x), and likewise E x. If we define τS to be the hitting time of S by the Brownian motion B, and μS(x) to be the hitting distribution of ∂S by BτS conditioned on τS < τ†, by the strong Markov property this is equal to
u ( x , S ) = P x { τ S < τ † } E μ S ( x ) ∫ 0 τ † 1 S ( B t ) d t (36)
= P x { τ S < τ † } g ( A ) , (37)
where now E μ denotes expectations for Brownian motion for which the distribution of B0 is μ, and g(A) is defined to be the latter expectation, which does not depend on x if S is circular (with area A).
This form we can now compare to the Expression (5) for the outflux of successful migrants. Consider the probability that a migrant family beginning with a single individual at x will ever establish in the new patch. It would be possible to analyze this probability directly, as in [29]; but here we take a simpler route, approximating this by the chance that the trunk hits the new patch, multiplied by the chance that at least one member of the family escapes demographic stochasticity and successfully establishes in the new patch. Write h(x, S) for the probability that the Brownian trunk hits the patch S before the family dies out, and f(S) for the chance that the family manages to establish in the new patch, given that it successfully arrives. (This is approximately independent of x.) As for q(S) above, Expression (5) is properly an integral against the outflux γ(x):
λ mig ( S ) = ∫ ∂ B ( r 0 ) γ ( x ) h ( x , S ) f ( S ) d x . (38)
If we make the approximation that Z hits the new patch only if the trunk of Z does, and recall that 1 − ke(t) is the chance that Z survives for t generations, then
h ( x , S ) ≈ ∫ 0 ∞ ( 1 - k e ( t ) ) P x { τ S ∈ d t } (39)
= ∫ 0 ∞ e s m t ( 1 - k e ( t ) ) e - s m t P x { τ S ∈ d t } (40)
= ∫ 0 ∞ e s m t ( 1 - k e ( t ) ) P x { τ S ∈ d t & t < τ † } (41)
≈ 1 E [ K ] ∫ 0 ∞ P x { τ S ∈ d t & t < τ † } (42)
= 1 E [ K ] P x [ τ S < τ † ] . (43)
Therefore, we have that
h ( x , S ) E [ K ] ≈ P x [ τ S < τ † ] = u ( x , S ) / g ( A ) . (44)
Since the integrands in Expressions (30) and (38) only differ by a factor that (at least asymptotically) does not depend on the distance between x and S, we can obtain the migration rate by multiplying the equilibrium frequency by this factor. This factor will not depend on A, because although f(A) and g(A) in principle depend on the patch size (and geometry), the dependence is very weak. For instance, the width of a circular patch only very weakly affects the chance of establishment of a new migrant that appears on its edge, as long as the patch is large enough.
By Eqs (30) and (38),
ρ q ( S ) = ∫ ∂ B ( r 0 ) γ ( x ) u ( x , S ) d x (45)
≈ g ( A ) E [ K ] ∫ ∂ B ( r 0 ) γ ( x ) h ( x , S ) d x (46)
= g ( A ) E [ K ] f ( A ) λ mig ( S ) . (47)
Once we show that g(A) ≈ 1/(2sm), we will have arrived at the result, Eq (10).
The function g(A) is the expected amount of time that a Brownian motion begun on the edge of a disk of area A is expected to spend inside the disk before τ†. This is integral of the Green function for the Bessel process of the appropriate order, so using [45], and letting w be the width of the patch, in d = 1,
g ( A ) = ∫ 0 2 w / σ e - y 2 s m / σ 2 s m d y = ( 1 - e - 2 w 2 s m / σ ) / ( 2 s m ) (48)
and in d = 2, by [68] 5.56.2,
g ( A ) = ∫ 0 2 w / σ 2 y K 0 ( y 2 s m ) d y (49)
= 1 s m ∫ 0 2 w 2 s m / σ y K 0 ( y ) d y (50)
= 1 2 s m 1 - 2 w 2 s m σ K 1 ( 2 w 2 s m / σ ) , (51)
where K0 and K1 are modified Bessel functions of the second kind. In either case, g(A) ≈ 1/(2sm), which is the approximation we use in the main text.
In the development above we need to approximate the integral of Eq (3) across the area occupied by the new patch, q(S) = ∫S q(x)dx. The precise answer depends on the shape and orientation of the patches; but we aim for a usable approximation, primarily in terms of the shortest distance from the old patch to the new patch, denoted R, and assuming R is large enough we can take Eq (3) as an equality.
Since q(R) decreases as ∣x∣ increases,
q ( S ) ≤ A q ( R ) (52)
where A is the area of S. This will be a good approximation if S is small.
In one dimension, if S has length ℓ,
q ( S ) = ∫ R R + ℓ C e - u 2 s m / σ d u (53)
= C e - R 2 s m / σ σ 2 s m 1 - e - ℓ 2 s m / σ (54)
≤ C e - R 2 s m / σ σ 2 s m . (55)
In two dimensions, suppose that T is a rectangle enclosing S with one axis aligned towards the original patch and transverse width w. Then
q ( S ) ≤ ∫ T q ( x ) d x (56)
≤ w ∫ R ∞ q ( r ) r d r (57)
= w C ∫ R ∞ ( r 2 s m σ ) − 1 / 2 e − r 2 s m / σ d r (58)
≤ w σ 2 s m C π ( R 2 s m σ ) − 1 / 2 e − R 2 s m / σ d r . (59)
If we absorb π into the constant C, this is of the form ( width ) × ( σ / 2 s m ) × q ( R ), i.e., roughly the area, after replacing the length by σ / 2 s m.
In both cases, the upper bound is q(S) ≤ A′ q(x), where A′ is the area of the parts of S that are no more than R + σ / 2 s m away from the original patch. Lower bounds could be obtained along similar lines.
|
10.1371/journal.ppat.1001171 | Platelet-Activating Factor Receptor Plays a Role in Lung Injury and Death Caused by Influenza A in Mice | Influenza A virus causes annual epidemics which affect millions of people worldwide. A recent Influenza pandemic brought new awareness over the health impact of the disease. It is thought that a severe inflammatory response against the virus contributes to disease severity and death. Therefore, modulating the effects of inflammatory mediators may represent a new therapy against Influenza infection. Platelet activating factor (PAF) receptor (PAFR) deficient mice were used to evaluate the role of the gene in a model of experimental infection with Influenza A/WSN/33 H1N1 or a reassortant Influenza A H3N1 subtype. The following parameters were evaluated: lethality, cell recruitment to the airways, lung pathology, viral titers and cytokine levels in lungs. The PAFR antagonist PCA4248 was also used after the onset of flu symptoms. Absence or antagonism of PAFR caused significant protection against flu-associated lethality and lung injury. Protection was correlated with decreased neutrophil recruitment, lung edema, vascular permeability and injury. There was no increase of viral load and greater recruitment of NK1.1+ cells. Antibody responses were similar in WT and PAFR-deficient mice and animals were protected from re-infection. Influenza infection induces the enzyme that synthesizes PAF, lyso-PAF acetyltransferase, an effect linked to activation of TLR7/8. Therefore, it is suggested that PAFR is a disease-associated gene and plays an important role in driving neutrophil influx and lung damage after infection of mice with two subtypes of Influenza A. Further studies should investigate whether targeting PAFR may be useful to reduce lung pathology associated with Influenza A virus infection in humans.
| Influenza virus causes disease that affects people from different age, gender or social conditions. The illness spreads easily and affects millions of people every year. Vaccines are effective preventive approaches, but the high degree of viral antigenic drift requires annual formulation. Anti-viral drugs are used as therapy, but are only effective at the very early stages of disease. The main symptoms that lead to hospitalizations and deaths are associated with the severe inflammatory host immune response triggered by the virus infection. Our approach was to decrease the inflammatory events associated with the viral infection by targeting a molecule, Platelet Activating Factor receptor (PAFR), known to induce several inflammatory events, including leukocyte recruitment and leakage. We found that PAFR deficient mice or wild type mice treated with a PAFR antagonist had less pulmonary inflammation, pulmonary injury and lethality rates when infected by two subtypes of Influenza A virus. In contrast, the immune response against the virus, as assessed by viral loads and specific antibodies, were not decreased. Our findings concur with the idea that severe inflammation plays an important role in flu morbidity and mortality and show that PAFR is a major driver of the exacerbated inflammation in mice infected with Influenza A virus.
| Influenza A viruses belong to the Orthomixoviridae family of RNA single-stranded, negative-sense viruses and cause epidemics, leading to about 250,000 to 500,000 deaths and 3 to 5 million severe cases annually worldwide [1], [2]. The new Influenza A H1N1 pandemic and the warning of a possible avian H5N1 pandemic increased the search for new vaccines and therapies [2], [3]. Antiviral drugs are an attractive possibility [4]. However, the need for the initiation of treatment very early in the course of infection [5] and the possibility of resistance suggest that novel alternatives are necessary [6]. A promising approach to reduce flu morbid is targeting immune molecules and cells related to disease severity (reviewed in [7]).
The immune system is activated shortly after respiratory epithelial cells have been infected by Influenza A. The single stranded RNA of Influenza virus is sensed inside endosomes by Toll like receptor 7 (TLR7) and TLR8 [8] and in cytoplasm by the helicase RIG-I (retinoic acid inducible gene-I) [9] and the inflammasome protein NLRP3 [10], [11]. TLR3 recognizes a intermediate of double strand RNA during Influenza replication [12]. The recognition of infection leads to alveolar macrophage recruitment and release of cytokines and chemokines with antiviral and proinflammatory actions (reviewed by [13]). NK cells are recruited in the first days of infection and are important for the initiation of adaptive immune responses against Influenza virus via IFN-γ production [14]. Neutrophils are another important leukocyte population involved in Influenza control [15]. Influenza A virus is a potent stimulus for neutrophil activation in the lungs and airways [16]. In addition to the antiviral action of neutrophils, excessive lung inflammation may result in lung damage, disruption of alveolar epithelial barrier and fluid leakage that limits respiratory capacity [17], [18].
Platelet Activating Factor (PAF) is a phospholipid mediator involved in many physiological and pathological conditions. The synthesis of PAF under inflammatory conditions is mediated by an acetyl-CoA:lyso-PAF acetyltransferase, named LysoPAFAT/LPCAT2 [19]. PAF acts through a single G protein-coupled receptor (PAFR) expressed in the plasma and nuclear membranes of leukocytes, endothelial cells and platelets [20]. Several inflammatory events have been associated with the administration of PAF, including increase of vascular permeability and lung edema [21], and leukocyte recruitment and activation [20], [22]. Moreover, blockade of the PAFR has been shown to decrease edema formation and/or leukocyte recruitment in several models of inflammation [23], [24], [25]. Phosphatidilcoline oxidation may also lead to the generation of PAF-like lipids that can activate the PAFR [26] and that have been reported to trigger lung injury after Influenza infection [27], [28]. Furthermore, upregulation of PAFR mRNA is seen during Influenza A virus infection [29], which is consistent with its expression on leukocytes and increase of these cells during Influenza A virus infection. Because of the involvement of PAFR during inflammatory responses and its expression during Influenza A virus infection, we hypothesized that PAFR activation may play an important role in driving pulmonary inflammation and injury in the context of Influenza A virus infection. To test our hypothesis, PAFR deficient mice or mice treated with PAFR antagonists were infected with two subtypes of Influenza A virus: a mouse-adapted Influenza virus A/WSN/33 H1N1 or a less virulent Influenza virus, H3N1. Our studies demonstrate that absence or antagonism of PAFR protects against Influenza A related lethality and inflammatory injury.
In order to determine the lethal inoculum of Influenza A/WSN/33 H1N1 virus in C57BL6/J mice, we infected the animals with four different inocula – 103, 104, 105 and 106 PFU – and accompanied them for 21 days after infection. Weight loss occurred over time after infection with 105 and 106 PFU and culminated with 100% death in 9 or 7 days, respectively. There was substantial weight loss in mice infected with 104 PFU until the eighth day of infection; thereafter, mice which survived (45%) recovered weight gradually. There were no weight loss and deaths associated with infection with 103 PFU (data not shown).
Based on the previous findings, we chose 104 as the mild inoculum and 106 PFU as the lethal inoculum to assess pulmonary inflammation associated with the infection. Since weight loss after infection with 104 PFU was slower, we chose to evaluate lung specimens at 3-day intervals from the first to the tenth day of infection. On the other hand, shorter 2-days intervals were chosen in the experiments with 106 PFU, from day one to five of infection.
We found substantial accumulation of neutrophils in BALF and in lung tissue, as assessed by myeloperoxidase (MPO) activity and histology, following infection (Figure 1). After 104 PFU, influx of neutrophils in the lungs (Fig 1A) and airways (Fig 1B) was first observed at day 4 after infection and returned to basal levels at day 10. When infected with the lethal inoculum (106 PFU), neutrophil infiltration in the lungs and BAL fluid was already detectable at day 1 and was more pronounced than with the lower inoculum (Figure 1 D,E). Pulmonary infiltration tended to resolve by day 5 after lethal infection. Consistent with the early peak of neutrophil accumulation at one day after infection with the lethal inoculum, mRNA expression of LPAFAT/LPAFAT2, the enzyme responsible for PAF synthesis in inflammatory conditions [19], was upregulated 2.5 fold at the first day of lethal infection (Fig 1 G).
Quantification of protein in BALF may be used as a marker of plasma leakage and, consequently, of lung injury [30]. There was progressive protein accumulation in the airways after infection with both inocula (Fig 1 C, F). Lung injury could also be observed by histological evaluation in animals infected with Influenza A (Fig 1 H, I). There was peribronchiolar and perivascular infiltration at 10 days after infection with 104 PFU, but, as seen by the normal thickness of alveoli, the infiltration was not present in the majority of alveolar walls (Fig 1H). Inflammation appeared to be more pronounced after infection with 106 PFU with significant peribronchial, perivascular and perialveolar inflammation, with thickening of alveolar walls, edema and mucus production on bronchioles at day 5 (Fig 1I).
There was a good correlation between the levels of CXCL1 and CXCL2 in the lungs (Fig S1A, B and Fig S2A, B) and the recruitment of neutrophils (Fig 1) after infection with 104 and 106 PFU. This is in agreement with studies showing the role of these chemokines and their receptors for neutrophil recruitment during Influenza A virus infection [31], [32]. Levels of the chemokine CCL2 were increased above baseline from day 4 after infection with 104 PFU and from day 1 after 106 PFU (Fig S1C, S2C). Thereafter, levels of CCL2 remained high throughout the observation period after infection with the lethal inoculum. Levels of the pro-inflammatory cytokine TNF-α were only increased at the beginning of the infection, i.e. at days 1 and 4 after 104 PFU and day 1 after 106 PFU (Fig S1D, S2D).
Aiming to evaluate a possible role of the inflammatory mediator PAF in Influenza A virus infection, PAFR and WT mice were infected with 104 PFU or 106 PFU of Influenza A/WSN/33 H1N1 virus. Infection of WT mice with 104 PFU resulted in 35% lethality rate. In contrast, only 7% of PAFR KO mice died after infection with the same inoculum (Fig 2A). Inoculation of 106 PFU caused 100% death by day 9 after infection of WT mice. In contrast, 23% PAFR KO infected with 106 were alive at day 21 after infection (Fig 2A).
As the absence of PAFR resulted in partial protection from Influenza A-associated lethality and in an attempt to seek for mechanisms of protection, we evaluated several parameters of inflammation in the lungs of mice after infection with 106 PFU. After infection with 106 PFU, there was decreased recruitment of total leukocytes (Fig 2B), mononuclear cells (Fig 2C) and neutrophils (Fig 2D) in the airways of PAFR KO mice when compared to WT mice. Neutrophil accumulation in lungs of WT and PAFR KO infected mice was similar at the evaluated time point (Fig 2E), as assessed by MPO quantification. The concentration of Evans' blue in BALF, a marker of vascular permeability, was reduced by approximately 35% in PAFR KO when compared to WT mice infected mice (Fig 2F). To examine in greater detail pulmonary changes and inflammation induced by infection with Influenza A, lung sections stained with H&E were analyzed and graded by a pathologist blind to the experimental situations. Leukocyte infiltration into bronchioles and alveoli, hyperemia, exudation, edema and bronchial mucus production were observed in most tissue sections from infected WT mice (Fig 3E, F). Despite of similar myeloperoxidase levels (Fig 2E) and neutrophilic infiltrates (Fig 3I) in lungs of infected groups, in PAFR KO mice, cellular infiltrates were restricted to conducting airways, in peribronchiolar and perivascular areas and did not reach distal airways like alveoli and the lung parenchyma, where gas exchange occurs (Fig 3G, H). Grading scores confirmed the limitation of infiltrates around the airways, the vessels and showed that these infiltrates were present in smaller areas of the parenchyma of PAFR KO infected mice – ranging from 1 to 29% of affected parenchyma in PAFR KO infected mice against 10 to 69% of affected parenchyma in WT infected mice (p<0.05, Fig 3I). Thus, the inflammatory infiltration in the larger airways was similar in WT and PAFR KO mice infected with the lethal inoculum; however, there was more infiltration of respiratory alveoli, with thickening of alveolar walls and pneumonitis in lungs of WT mice. Experiments in animals infected with 104 PFU showed decreased neutrophil influx and less protein leakage, hence less pulmonary injury, in PAFR KO mice when compared to WT infected mice after 8 days of infection (data not shown). Pulmonary inflammation after 5 days of infection with 104 PFU was discrete and restricted to neutrophil infiltration and compromised a percentage of lung parenchyma between 10 and 29%, which is much less severe than lesions developed after lethal infection. Pulmonary inflammation in PAFR KO mice infected with the low inoculum was also discrete and similar to that found in lungs of WT mice (data not shown).
Because there was a significant change in the pathological score, we conducted a more detailed analysis of leukocyte populations in BALF and in lungs of infected mice. In the airways, there was an increase in the number of all leukocyte populations evaluated at day 5 after infection – CD4+ T cells, CD8+ T cells, NKT cells, NK cells, macrophages and neutrophils. CD8+ and CD4+ lymphocytes, NK and NKT cells were equally increased in WT and PAFR KO infected mice (data not shown). Number of granulocytes (Fig 4A) and macrophages (Fig 4B) were increased after infection of WT mice but this increase was lower in PAFR KO mice infected with the same inoculum.
We performed Annexin V binding assay to evaluate whether increased apoptosis could account for the decreased accumulation of neutrophils and macrophages in the airways after flu infection of PAFR KO mice. Influenza A induced a marked increase in the number of apoptotic cells in the airways at day 5 after infection. Overall, the number of apoptotic cells was greater in infected WT than PAFR KO mice (Fig 4C). The number of apoptotic granulocytes was similar in both infected groups (Fig 4D). Therefore, the lower number of neutrophils in the airways of PAFR KO infected mice was due to decreased recruitment of these cells and not due to a higher degree of cellular death. In contrast, there was decreased number of apoptotic macrophages and lymphocytes in infected PAFR KO than in WT mice (Fig 4E, F).
Leukocyte populations were also evaluated in lung homogenates. The percentage of lung CD3+CD8+ T cells did not increase after infection (data not shown). CD3+CD4+ T cells increased after infection in WT but not in PAFR KO mice (Fig 4G). F4-80+ CCR5+ (macrophage) and GR1+ CXCR2+ (neutrophils) cells in lung homogenates increased after infection in both WT and PAFR mice to a similar extent (data not shown). NK populations (CD3+NK1.1+ and CD3− NK1.1+) enhanced after infection in WT mice but the enhancement was significantly greater in PAFR KO mice (Fig 4H, I, J).
The concentrations of the cytokines IL-1β, IL-6, TNF-α, CXCL1, CXCL2, CCL5, IFN-γ and IL-12p40 were evaluated in lung homogenates of control and Influenza infected mice. As seen in Figure S2, levels of TNF-α and CXCL2 were not increased in the lungs at day 5 after infection with 106 PFU infection in both WT and PAFR KO groups (data not shown). Levels of IL-6 (Fig 5A) and CXCL1 (Fig 5B) were increased after infection but there was no difference between WT and PAFR KO mice. Levels of IL-1β were also enhanced by infection but the increase was more pronounced in PAFR KO than WT infected mice (Fig 5C). Levels of IL-12p40 (Fig 5D), IFN-γ (Fig 5E) and CCL5 (Fig 5F) increased after infection and the increase was more pronounced on WT that PAFR KO mice.
Because pulmonary inflammation and levels of some cytokines known to be important for host resistance to viral infection, including IFN-γ and IL-12p40, were decreased in PAFR KO mice, we investigated whether this reduction was sufficient to impair viral clearance and adaptive responses. As seen in Figure 6A, there was no difference in viral load at day 5 in the lungs of WT and PAFR KO mice infected with 106 PFU (Fig 6A) or 104 PFU (Fig 6B). In animals infected with 104 PFU, viral load was significantly lower in PAFR KO than WT mice at day 8 after infection (Fig 6C).
Next we investigated humoral adaptive immune responses in the presence or the absence of PAFR. In a re-infection protocol, animals were initially infected with a non-lethal (103 PFU) inoculum of Influenza A/WSN/33 H1N1. After 14 days, animals were subjected to another infection with the same virus, at this time using the lethal inoculum, 106 PFU. No weight loss or lethality was observed in WT or PAFR KO mice using this protocol (data not shown). Measurement of anti-Influenza A WSN/33 IgG revealed no differences in antibody titer between the groups at day 21 after re-infection. The titer was 1/250 in both groups and the optical densities at that dilution are shown in Fig 6D.
Because there was decreased survival and pulmonary inflammation in PAFR-deficient mice and this was associated with decreased viral load after infection with the lower inoculum at day 8 after infection, we assessed whether virus propagation was altered by the absence of PAFR. To this end, we infected A549 cell line, a human alveolar basal epithelial cell, with Influenza A virus expressing red fluorescent protein (RFP). Using this methodology, we found no difference in the ability of influenza virus to infect epithelial cells in vitro in the absence or presence of a PAFR antagonist PCA 4248 (50 µM) (data not shown).
We also measured the height of bronchiolar epithelium after infection with 106 PFU to determine whether infection of epithelial cells, the primary target of influenza infection, was similar in the absence or presence of PAFR. Lethal infection promoted pronounced reduction in epithelial height of bronchiole and absence of PAFR did not alter the susceptibility of epithelial injury (Fig S3). Therefore, virus entry and replication and subsequent cell injury in its primary target, epithelial cells, is not affected when PAFR is absent.
TLR3, TLR7/8 and NALP3 are thought to be major intracellular recognition molecules used by the host to detect Influenza A infection [14]. To investigate whether these activation of these pathways could lead to inflammation in a PAFR-dependent manner, we studied the effects of synthetic agonists of TLR7/8 (R848) and TLR3 and NLRP3 [poly(I∶C)] in WT and PAFR KO mice.
Intratracheal instillation of R848, a TLR7/8 agonist, induced LPAFAT/LPAFAT2 mRNA in lungs of WT mice (Fig 7A). This was accompanied by infiltration of neutrophils in lungs (Fig 7B) and airways (Fig 7C), and increased levels of IL-12p40 (Fig 7D), IL-6 (Fig 7E) and CXCL1 (Fig 7F) in lungs of WT mice. In PAFR KO mice, R848 induced similar amount of neutrophil influx in lungs (Fig 7B), but greatly reduced neutrophil recruitment to the airways (Fig 7C). Levels of IL-12p40 (Fig 7D) and CXCL1 (Fig 7F) were reduced in lungs of PAFR-KO mice instilled with R848 when compared to WT.
Poly(I∶C) stimulation did not induce expression of LPAFAT/LPAFAT2 mRNA in WT mice (Fig S4A). Instillation of poly(I∶C) induced neutrophil influx in the airways (Fig S4C), but not lungs (Fig S4B) and CXCL1 production in lungs (Fig S4D) of WT mice, a response that was similar or slightly greater in PAFR KO mice.
To test the potential use of PAFR as a pharmacological target against Influenza A-associated disease, we used the selective PAFR antagonist PCA 4248 in a therapeutic protocol. The treatment was started 3 days after infection of mice with 104 or 106 PFU and was continued until day 10. Day 3 was chosen because this is the time at which weight loss and neutrophil influx peaked in the higher inoculum (Fig 1B, G). Treatment with PCA 4248 significantly enhanced survival after infection with 106 PFU of Influenza A/WSN/33 H1N1 virus (Fig 8A). Akin to the experiments in PAFR KO mice, higher survival was accompanied by reduction in total leukocyte (Fig 8B), mononuclear cells (Fig 8C) and neutrophil (Fig 8D) accumulation in the airways. Neutrophilic accumulation in lung parenchyma after Influenza infection was not affected by PCA 4248 treatment (Fig 8E) as observed in PAFR KO (Fig. 2E). Plasma leakage in the airways, as assessed by measuring total protein in BAL was significantly lower in PCA 4248-treated than vehicle-treated animals (Fig 8F).
PCA treatment failed to alter significantly (from 55% to 90% survival, p = 0.10) the lethality rate caused by 104 PFU infection (Fig 8A), but there was a significant decrease in weight loss from the ninth to the eleventh day of infection, when compared to the vehicle-treated group (Table 1).
To assess whether the protection observed in PAFR deficient mice and WT mice treated with a PAFR antagonist infected with Influenza A H1N1 was virus-specific, we used another virus strain, a reassortant Influenza A H3N1 subtype to infect WT and PAFR KO mice. Since mouse infection with different Influenza virus subtypes requires adaptation [33], H3N1 virus was subjected to three lung passages in C57BL/6J in order to cause disease in these animals. After three lung passages the virus was able to multiply and increased viral titer (data not shown), a sign of adaptation. As a result, the inoculum of 106 PFU of H3N1 virus was enough to cause 15% of weight loss on average after three days of infection. PAFR KO mice were protected from the disease caused by Influenza A H3N1, since they start to regain weight, from day ten of infection, faster than WT animals (Fig 9).
Severe inflammation caused by highly pathogenic Influenza A strains was described to be an important cause of death during 1918 pandemics and highly pathogenic avian Influenza H5N1 infections [34]. Using a mouse-adapted Influenza A/WSN/33 H1N1 strain we show a correlation between viral load, pathogenesis and lethality – i.e. the higher the viral load, there is greater lung injury and death. More importantly, the present work demonstrates that the course of Influenza A virus infection is less severe in the absence of PAFR. Mechanistically, it appears that activation of TLR7/8 by Influenza A explains the induction of LPAFAT/LPAFAT2 mRNA and consequent activation of PAFR. Further, absence or blockade of PAFR during infection is associated with decreased neutrophil and macrophage influx into airspaces, decreased production of certain pro-inflammatory mediators and decreased lung edema, parameters which are commonly increased after pulmonary administration of PAF to rodents or humans [35], [36]. Lack of PAFR did not increase viral load or prevent specific anti-Influenza antibody production. Finally, we demonstrate that administration of PAFR antagonist 3 days after infection also causes similar protection as observed in PAFR-deficient mice.
Neutrophils and neutrophil-active chemokines peaked very early in the course of Influenza A/WSN/33 H1N1 infection and decreased thereafter. On the contrary, lung edema and injury was progressive. Thus, it seems that inflammation is self-resolving but causes lung damage which is more pronounced in the last days of infection, leading to progressive weight loss and death. As inflammation is important to trigger lung injury in Influenza A infected mice and PAF attracts neutrophils into the lung [36], we evaluated whether the receptor for PAF was involved in H1N1-associated lung inflammation and injury, and death. A previous study demonstrated up-regulation of PAFR mRNA in lungs following Influenza A/PR8/34 H1N1 infection [29]. This is consistent with the expression of PAFR on neutrophils and other leukocytes [20] and the influx of these cell types during infection. In our experiments, we found increased expression of the enzyme involved in inflammatory synthesis of PAF – LPAFAT/LPAFAT2 – in the first day of lethal infection. Therefore, the release of PAF is an early event that could be involved in the pathogenesis of Influenza A infection. In fact, PAFR deficient mice or mice treated with PAFR antagonist PCA 4248 were protected from death or weight loss caused by Influenza A virus infection. PAFR signaling was also important for pulmonary inflammation and injury.
Neutrophils are required for the clearance of Influenza virus during the early stages of infection [37]. The antiviral actions of neutrophils are clearly demonstrated through the use of RB6-8C5 antibodies to deplete these cells. In the absence of neutrophils, mice are more susceptible to virus growth and associated lethality [15], [38]. Therefore, direct ablation of neutrophils cannot be used therapeutically. On the other hand, Influenza A virus is known to be a potent activator of neutrophil respiratory burst, apoptosis and subsequent deactivation in front of a second stimulus [16]. The neutrophil activation that occurs during infection, despite its role in controlling viral replication, is thought to be very harmful to the host [37]. PAFR deficiency or antagonism increased survival after Influenza A virus infection and this was associated with lower neutrophil recruitment to the airways. There was, however, no inhibition of neutrophil into pulmonary parenchyma. Therefore, reduction instead of ablation of neutrophil recruitment into the airways appears to be an approach which is sufficient to maintain control of viral burden, but, at the same time, avoid excessive neutrophil activation and lung damage.
Bacterial pneumonia, especially by Streptococcus pneumonia, is an important complication following Influenza A virus infection [39]. Pro-apoptotic effects of Influenza A virus on neutrophils are one of the explanations for the increased susceptibility to bacterial pneumonia after Influenza infection [40]. In our model, absence of PAFR did not influence neutrophil apoptosis. The role of PAFR in altering bacterial pneumonia after Influenza infection is controversial. While the study of van der Sluijs and co-workers described protection in PAFR KO mice after Streptococcus pneumoniae following flu infection [29], McCullers and colleagues showed no correlation between the increased pathology after secondary infection and the antagonism [41] or absence of PAFR [42]. Both groups focused on lethality and harm caused by the bacteria, not the virus, and the overall message was that absence or blockade of PAFR was either without effect or protective, but never harmful. Here, we show that targeting PAFR during Influenza A virus infection is protective against viral-induced pneumonia, regardless of the risk of bacterial pneumonia after viral infection.
In addition to neutrophils, endothelial cells express PAFR and are affected by PAFR signaling [22]. Protein leakage to the airways reflects the increased vascular permeability following inflammatory events associated with Influenza A virus infection [30]. In our system, lower protein amounts or Evans' blue extravasation were found in BALF of PCA treated or PAFR KO animals. As mentioned above, neutrophil accumulation in the parenchyma was not affected by PAFR absence or antagonism. Thus, neutrophil transmigration into the alveolar space seems to play an important role in the pathogenesis of Influenza infection. During the transmigration, neutrophils may release proteinases or oxygen reactive species that lead to increase in vascular permeability that is accompanied by significant protein leakage [43]. Hence, lower protein amounts found in lethally infected mice treated with PCA or PAFR KO mice infected with lower infection inoculum compared with their controls corroborates with this hypothesis.
Pulmonary inflammation and injury are common histopathological findings in humans with mild or severe Influenza A infection. Viruses of low pathogenicity affect basically the proximal airways (bronchi and bronchioles), whereas highly pathogenic viruses or infection of immunocompromised people infected is associated with inflammation and injury in the distal airways (alveoli and parenchyma). When distal airways are affected, lung injury is more severe and lead to loss of respiratory capacity and gas exchange [44]. In our system, the extent of distal airways involvement was decreased in PAFR KO mice, suggesting these results may bear relevance to humans with severe Influenza A infection. Evaluation of gas exchange in our mice was not possible as animals needed to be anesthetized for the procedure (data not shown).
NK cells are recruited to the lungs very early in the course of Influenza infection, are involved in initial viral clearance and provide initial signals to the development of protective adaptive responses [45]. NK cells recognize Influenza A virus through its NKp46 receptor and, stimulated by IL-12 released from dendritic cells, mediate cytotoxicity to infected cells and release of IFN-γ that, in turn, mediates adaptive responses [46]. PAF seems to be important for cytotoxic functions of NK cells [47], [48] and PAFR antagonists reduced cytotoxic activity of NK cells [47]. Jin and colleagues showed that inflammation-released PAF recruits NK cells activated by IL-2, IL-12, IL-15 and IFN-α [48]. Despite these studies showing a potential effect of PAF on NK cell function and recruitment, we actually observed that NK cell recruitment to the lungs following Influenza A virus infection was increased in the absence of PAFR. Our studies do not provide a mechanism to this particular finding but do suggest that enhanced NK cell recruitment could be functionally relevant as seen by lower viral loads in the lung in some of the experiments.
In the absence of PAFR, there were reduced pulmonary levels of IFN-γ, IL-12p40 and CCL5. The reduction in these cytokines which are preferentially produced by or activate Th1 lymphocytes is related to the reduced number of CD4+ T cells in the lungs of PAFR-deficient mice infected with influenza A. These results are consistent with previous experiments in mice infected with a different microorganism, Leishmania amazonensis, where we also demonstrated that absence of PAFR decreased CCL5 expression and Th1-associated function [49]. However, in the case of the protozoan infection, the absence of PAFR was deleterious, because of an inability of the host to control the infection [49]. These results suggest that PAF may be important regulator of CCL5 production and consequent infiltration of effectors T cells with Th1 phenotype in vivo. Despite decreased IFN-γ expression and decreased accumulation of CD4+ T cells in the lungs of infected PAFR-deficient mice, viral load in the lung was decreased or at worst similar to that found in WT mice. Reduction of viral load was not associated with decreased propagation in epithelial cells in vitro. Furthermore, the capacity of Influenza virus to cause epithelial cell injury in vivo was unchanged by absence of PAFR. It is unclear why there was a small decrease in viral load in the lungs on day 8 after infection with the low inoculum. It is possible that this may reflect the better clinical status of animals and the decreased pulmonary inflammation in the absence of PAFR. Absence of PAFR was not associated with a decreased specific antibody release, as shown by re-infection studies and measurement of specific IgG titers. Therefore, changes in CD4+ T cell recruitment and related cytokines were not sufficient to modify negatively the course of infection in this model of experimental Influenza A virus infection.
Macrophages are also recruited to the lungs and airways after Influenza infection where they are thought to play an important role in the production of pro-inflammatory cytokines and in the phagocytosis of Influenza virus-induced apoptotic cells [50]. In vivo abrogation of phagocytosis of apoptotic infected cells by macrophages increases lethality rates in Influenza A infected mice [51]. We found a partial reduction of macrophage recruitment to the airways in PAFR deficient mice infected with Influenza A. There was also, a proportional reduction in number of apoptotic macrophages in PAFR deficient mice, showing that the proportion of live active macrophages was similar in both groups. These macrophages appear to be sufficient for clearance of apoptotic infected cells and helping in the resolution of the inflammatory phase of the infection [52], as seen by the reduced lung damage and survival in infected PAFR-deficient mice.
The suggestion that macrophage function in infected PAFR-deficient mice was sufficient to keep immune responsiveness is strengthened by findings that IL-6 expression was maintained and IL-1β expression was actually increased in these mice. IL-1β is produced mainly by macrophages after infection with Influenza A [13]. The role of IL-1β for viral clearance and pathology during Influenza infection is very complex [53]. Indeed, a previous study has shown that IL-1R1 deficient mice had greater lethality rates, but decreased lung damage caused by Influenza A virus infection [53]. Maines and colleagues showed a clear inverse correlation between viral virulence and IL-1β release in lungs [54]. While highly virulent H5N1 virus induces a small increase in IL-1β secretion, low virulent H5N1 stimulates high levels of this cytokine [54]. Therefore, the cytokine IL-1β does appear to contribute to viral clearance, even at the cost of increasing pulmonary recruitment of leukocytes in some models. In our system, enhanced release of IL-1β in PAFR-deficient mice was associated with maintained or decreased viral load, but no enhancement of leukocyte infiltration or pathology. The inability of IL-1β to enhance inflammation in the system may be explained by the role of PAFR in mediating IL-1β-associated inflammation [55], [56].
The recognition of Influenza virus RNA is the main trigger of antiviral and inflammatory responses [14]. Using the synthetic TLR agonist R848, we found that activation of this ssRNA sensor induces the expression of LPAFAT/LPAFAT2 and inflammation that is directly dependent of PAFR. It has been previously shown that LPAFAT/LPAFAT2 expression and enzyme activity may be induced by the TLR9 ligand ODN1826 and the TLR4 ligand LPS [19]. Poly(I∶C), a classical TLR3 ligand and recently described as a NLRP3 activator [10], did not induce LPAFAT/LPAFAT2 mRNA expression and cause pulmonary inflammation which was PAFR-independent. Therefore, of the known major recognition receptors for Influenza, it appears that TLR7/8 is the one capable of inducing LPAFAT/LPAFAT2 expression and triggers pulmonary inflammation in PAFR-dependent manner. Results with R848 were qualitatively similar (LPAFAT/LPAFAT2 mRNA expression, PAFR-dependency of inflammation) to that observed after Influenza infection, suggesting that activation of TLR7/8 is a main mechanism by which Influenza infection leads PAF release and PAFR activation. One alternative possibility to explain the activation of PAFR in our system derives from recent findings that oxidized phospholipids (Ox-PL) were produced during acute lung injury induced by inactivated H5N1 [27]. Ox-PL are PAF-like molecules known to induce PMN migration and protein leakage in pleural cavity, effects which are PAFR-dependent [57]. Ox-PLs levels were also correlated with the reduced severity of flu in IL-17RA KO mice infected with Influenza A/PR/8/34 (H1N1) [28]. Thus, production of Ox-PL may contribute to the activation of PAFR during influenza infection.
The present model of Influenza A/WSN/33 H1N1 infection mimics infections with highly pathogenic strains during pandemics or in immunocompromised people. We used a low inoculum of H1N1 and a low pathogenic Influenza A strain to model the common features of the seasonal flu in humans. Using a low pathogenic Influenza A H3N1 reassortant virus, we showed that PAFR deficient mice had decreased weight loss in comparison to WT mice. Therefore, PAF may be important to the pathogenesis of Influenza A virus infection, regardless of subtype or strain. This feature could be explained because PAFR targeting modulates the inflammatory response to the virus, without affecting, or also improving viral clearance by the host. Recently, our group published similar results in a murine model of Dengue virus infection. In that work, the absence or pharmacological blockade of PAFR resulted in protection against the main symptoms of dengue virus infection and lethality, regardless of viral titer change [58].
In conclusion, our studies clearly show that PAFR-mediated inflammatory events that follow Influenza A virus infection are important for disease pathogenesis and lethality. Mechanistically, protection in PAFR deficient mice was associated with decreased infiltration of neutrophils and macrophages into the airways and decreased lung damage. Importantly, PAFR deficiency tended to enhance the ability of the murine host to deal with the virus and antibody and adaptive response were maintained. Importantly, treatment with PAFR antagonists starting 3 days after infection also protected against Influenza A morbidity and lethality. These studies show that PAFR is a disease-associated gene after Influenza A virus infection and suggest that PAFR antagonism could be a useful therapeutic target to interfere with inflammatory damage that follows infection. It remains to be determined whether this will be a useful therapeutic strategy in humans and whether association with antiviral will enhance benefits provided by each individual strategy, as suggested elsewhere [59].
All the experiments were conducted under prior CETEA/UFMG animal ethics committee approval (203/08), according to Brazilian guidelines on animal work.
The strains used in the study were Influenza A WSN/33 H1N1 and Influenza A (H3 of equine/Cordoba/18/1985 and N1 of Yamagata/32/1989) H3N1. Briefly, Influenza A WSN/33 H1N1 was produced in chicken eggs and passed once more in eggs and then cultured in MDCK (Madin-Darby Canine Kidney) cells. The stocks in a final concentration between 4×107 and 2×108 PFU/mL were diluted in sterile phosphate buffered saline prior to infections.
Influenza A H3N1 strain was adapted to mice through three lung passages. Briefly, 104 PFU of Influenza A H3N1virus was inoculated via intranasal in five animals and after 5 days lungs were collected an tittered. The sample that had the higher titer (105 PFU) was passed through a 0.45 µm filter and used to infect another 5 animals (100 PFU/animal). The process of selection was repeated once and the final titer achieved was 5×107 PFU. The stock was diluted in sterile phosphate buffered saline prior to infections.
Male 8–10 weeks C57BL/6J and PAFR deficient mice (PAFR KO) on a C57BL/6J background, maintained in pathogen free conditions at Laboratorio de Imunofarmacologia (UFMG/Brazil) facilities, were used in the infection experiments. Mice anesthetized with ketamine/xylazine received 25µL of Influenza A/WSN/33 H1N1 virus, Influenza A H3N1virus, or sterile phosphate buffered saline (PBS, Mock group), intranasally. The H3N1 virus was previously adapted to mice, through three lung passages as described above.
The PAFR antagonist PCA 4248 (Tocris Bioscience), 5mg/kg, diluted in 5% of ethanol in PBS was given twice a day, via subcutaneous injections. The control group (vehicle) received the same volume of the solution used to dilute PCA 4248.
MDCK and the human lung epithelial cell line A549 (ATCC CCL-185) were cultured in Dulbecco's Modified Eagle Medium (DMEM) supplemented with 6% of fetal bovine serum were cultured at 37°C with 5% of CO2. Cells were seeded at a density of 4×104 cells/well in a 24-well plate and after 24 hours of growth were incubated with 50 µM of PCA 4248 in DMEM or vehicle for 30 minutes. Cells were infected with a multiplicity of infection of 2 with a red fluorescent protein (RFP) labeled Influenza A virus and incubated for 16 hours. Virus propagation was observed through a fluorescence microscopy.
At indicated time points, mice were euthanized with an overdose of ketamine/xylazine solution. Subsequently, a 1.7mm catheter was inserted into the trachea and two 1mL aliquots of PBS were inserted into the lungs and collected three times to acquire leukocytes recruited to the airway space. After centrifugation, the pellet was used to total and differential leukocytes counts of stained slides.
At day 5 after infection with 106 PFU of Influenza A WSN/33 or PBS instillation, WT and PAFR were injected with Evans blue dye (50 mg/kg, i.v.) 2 h before they were killed. Animals were subjected to BAL with 1mL of PBS. BALF was then centrifuged and the optical density was determined at 620 nm. The concentration of extravasated EBD (microgram of EBD per gram lung) in lung homogenates was calculated against a standard curve.
After performing BAL, lungs were perfused with 5 mL of PBS to remove circulating blood and frozen. A hundred µl of tissue was homogenized to perform ELISA and MPO assay, as previously described [60].
Lung tissues were homogenized in a PBS-buffer containing antiproteases, as previously described [60], to assess the concentrations of the cytokines IL-1β, IL-6, IL-12 p40, IFN-γ and TNF-α and the chemokines CXCL1, CXCL2, CCL2, CCL5 and serum IL-6 levels by ELISA DuoSet kits (R&D Systems), in accordance to the manufacturer's instructions.
Total RNA from diaphragmatic lung lobe tissue conserved at −70°C in RNA later (Applied Biosystems, California, USA) was extracted using Trizol (Invitrogen), as described by the manufacturer. The total RNA obtained was suspended in RNAse-free water and stocked at −70°C. Real-time PCR was performed on an ABI PRISM Step-One sequence-detection system (Applied Biosystems) by using SYBR Green PCR Master Mix (Applied Biosystems) after a reverse transcription reaction of 2 µg of RNA by using M-MLV reverse transcriptase (Promega). The relative expression level of LysoPAFAT/LPCAT2 gene was determined by the 2 (−delta delta Ct) method, normalized by ribosomal subunit 18S and expressed as fold change compared with constitutive gene expression. The following primer pairs were used: LysoPAFAT/LPCAT2 forward 5′ GTCCAGCAGACTACGATCAGTG 3′; LysoPAFAT/LPCAT2 reverse 5′ CTTATTGGATGGGTCAGCTTTTC 3′ as described by [19]; and 18S forward 5′ CTCAACACGGGAAACCTCAC 3′; 18S reverse 5′ CGTTCCACCAACTAAGAACG 3′.
We analyzed lung injury following Influenza A virus infection assessing protein leakage to the airways, histological changes in lung architecture. A protein quantification assay (Bio-Rad Protein Assay) was performed in bronchoalveolar lavage fluid (BALF) supernatant according to manufacturer's instructions. Formalin-fixed lung left lobes were dehydrated gradually in ethanol, embedded in paraffin, cut into 4-µm sections, stained with H&E and examined under light microscopy and scored by a pathologist blinded to the experiment. The score of 18 points was based on Horvat and colleagues paper [61], which evaluates airway, vascular and parenchymal inflammation, added to a 5 points score evaluating general neutrophilic infiltration (0, absent; 1, minimal; 2, slight; 3, moderate; 4, marked; and 5, severe).
Photographs of areas containing bronchioles in H&E stained slides were taken under 200 fold of magnification and were analyzed with an AxioVision software. Epithelial height of bronchioles in areas of inflammatory infiltrates had their length measured. The number of bronchioles analyzed in each slide varied according to their length, to totalize 1500 µm per slide. The mean epithelial height for each animal was used to construct the graph.
Leukocytes recovered from BALF and from lungs processed with collagenase IV (Sigma) [62] were stained with fluorescent-labeled monoclonal antibodies CD3, CD4, CD8, NK1.1 and GR1 (BD Pharmigen TM); CXCR2 (R&D Systems); F4/80 and CCR5 (Biolegend). Stained cells were acquired in FACScan cytometer and analyzed in FlowJo (Tree Star) software (Text S1).
We performed Annexin V binding assay on BALF leukocytes using Annexin V/Propidium Iodide (PI) kit (Caltag Laboratories) according to manufacturer's instructions. Flow cytometry was carried out in FACScan and analyzed in FlowJo software.
In order to determine viral load in lungs, we collected the organs in sterile conditions and performed a plaque assay using MDCK cells. Lungs were weighted and homogenized in PBS, plated in MDCK monolayer and after incubation and staining it was possible to count the plaques. The viral titer was expresses as Plaque Forming Units (PFU) per gram of tissue.
WSN/33 H1N1 influenza virus stocks were used as antigens in an indirect ELISA to detect specific antibodies in serum samples of reinfected animals (Text S1).
All data are presented as the mean ± SEM. All data were tested for normality and found to have a normal distribution. Normal data were tested for significance using ANOVA followed by use of Newman-Keuls post-test, which corrects for multiple comparisons. Unpaired t test was used to compare two groups and Log-rank test for lethality experiments, where appropriate. Statistical significance was set as P<0.05 and all graphs and analysis were performed using Graph Pad Prism 4 software.
|
10.1371/journal.pgen.1001086 | Allelic Selection of Amplicons in Glioblastoma Revealed by Combining Somatic and Germline Analysis | Cancer is a disease driven by a combination of inherited risk alleles coupled with the acquisition of somatic mutations, including amplification and deletion of genomic DNA. Potential relationships between the inherited and somatic aspects of the disease have only rarely been examined on a genome-wide level. Applying a novel integrative analysis of SNP and copy number measurements, we queried the tumor and normal-tissue genomes of 178 glioblastoma patients from the Cancer Genome Atlas project for preferentially amplified alleles, under the hypothesis that oncogenic germline variants will be selectively amplified in the tumor environment. Selected alleles are revealed by allelic imbalance in amplification across samples. This general approach is based on genetic principles and provides a method for identifying important tumor-related alleles. We find that SNP alleles that are most significantly overrepresented in amplicons tend to occur in genes involved with regulation of kinase and transferase activity, and many of these genes are known contributors to gliomagenesis. The analysis also implicates variants in synapse genes. By incorporating gene expression data, we demonstrate synergy between preferential allelic amplification and expression in DOCK4 and EGFR. Our results support the notion that combining germline and tumor genetic data can identify regions relevant to cancer biology.
| Cancer is a disease of two distinct, but related, genomes: the inherited genome and the tumor genome. Despite the fact that the tumor genome arises from the germline, the genomes are typically studied as separate entities. For example, germline genetic studies focus on how inherited variation is related to a particular trait such as disease risk, whereas tumor genetic studies focus on areas of recurrent aberrations such as amplifications to identify genes involved in tumor biology. In this study, we integrated both germline and tumor genetic information to pinpoint areas of the human genome that are likely undergoing selection during the evolution of the tumor. Our results support the notion that combining germline and tumor genetic data can identify regions relevant to cancer biology.
| Cancer is a disease of two related, but karyotypically distinct genomes: germline and somatic. Researchers typically focus on identifying genetic alterations by exclusively studying either the germline genome or the somatic genome. Germline genetic variants that play key roles in tumor biology (e.g., risk alleles) have typically been discovered using linkage and, more recently, association studies. On the other hand, somatic genetic elements important for tumor biology, such as amplifications, deletions, and point mutations, are usually identified by patterns of recurrence across tumor samples. Given the kinship between these two genomes, however, studies of cancer biology should be amenable to population genetic analysis, since the tumor cells can be considered descendants of a progenitor cell. In the population of tumor cells, lineages are subject to somatic versions of mutation, drift and selection [1]. We hypothesize that integrating germline allelic (i.e., genotypic) information with somatic amplification events could yield novel insights into the alleles that undergo selection during tumor evolution.
Associations between cancer risk alleles and somatic patterns are beginning to appear in the literature with increasing frequency. Preferential allelic amplification at candidate risk loci has been convincingly demonstrated in several mouse studies [2], [3] as well as in the analysis of the AURKA oncogene in humans [4], [5]. Additionally, a germline risk allele for colorectal cancer was demonstrated to be preferentially amplified (relative to the wild type allele) in tumors that were heterozygous for this single nucleotide polymorphism (SNP) [6]. More recently, a somatically acquired mutation in JAK2 for myeloproliferative disorders was shown to arise preferentially on a particular haplotypic background [7]–[9]. These targeted studies of specific loci provide compelling evidence for the relationships between the germline and somatic genomes. One of the goals of our study is to perform a genome-wide query for such relationships.
We have developed a battery of statistical methods to query tumor DNA data for preferential allelic amplification [10]. These methods are designed to identify alleles that have likely been positively selected during tumor evolution within areas of copy number gain (Figure 1; Materials and Methods). One of these statistical tests, termed the amplification distortion test (ADT), is closely related to a well-known genetic test of association and linkage, the transmission disequilibrium test (TDT) [11]. Consider an example where N = 100 tumors harbor an amplification and are heterozygous at a particular SNP locus (whose alleles are arbitrarily labeled A and B). Under the null hypothesis, on average 50 tumors will amplify one allele, and the other 50 will amplify the alternate allele since tumors typically amplify one of the two parental chromosomes [12]. In this sense, amplification is a somatic analog of Mendelian 50∶50 transmission in germline genetics. Significant deviations from a 1∶1 ratio (Figure 1B) are inconsistent with this null hypothesis and suggest that the particular allele – or a variant linked to that allele within the same amplified region – is advantageous to the tumor when amplified. Formally, we compare the observed number of germline heterozygotes amplifying the A allele to the Binomial(N, p = 0.5) distribution to obtain a two-sided P-value (Materials and Methods and [10]). Similar to the TDT, the ADT is robust to population stratification because the non-amplified homolog provides a perfectly matched control.
The recent National Cancer Institute-directed initiative, the Cancer Genome Atlas (TCGA; [13]), provides an ideal resource to test our hypothesis, furnishing SNP array data from multiple platforms across hundreds of glioblastoma multiformae tumors and matched normal samples. Since SNP arrays contain both the allelic and amplification information at hundreds of thousands of loci across the genome, the data are well-suited for our allelic distortion analysis. For each patient in the study, we made use of TCGA-generated SNP genotypes in the germline, as well as amplification status (generated on three separate probe hybridization-based platforms) and allelic amplification status for matched tumor DNA.
TCGA recently published (TCGA, 2008) the first report from a pilot study of over 200 human glioblastoma samples. As part of that study, the tumor DNA was interrogated at some 1.8 million loci using the Affymetrix SNP array 6.0, over 236,000 loci using the Agilent CGH microarray 244A, and at about 550,000 loci using the Illumina HumanHap550 array. We restricted our analysis to the 178 individuals for whom both germline and tumor array data from the Illumina platform were available. From these data, we extracted allelic copy number on a SNP-by-SNP basis; that is, for each individual, we inferred amplification status at each SNP locus, also identifying the amplified alleles in amplicon SNPs. As shown in Figure S1 (and observed in the TCGA manuscript), commonly (>5% of samples) amplified loci are restricted to several discrete – but wide – genomic regions. Such regions have a median length of 166 kb, and regions amplified in at least 15% of samples are usually even longer (median length 382 kb). Such broad regions of recurrent amplification can make it difficult to identify the target of these amplifications.
From a statistical standpoint, although we apply our test statistics across the genome, the ADT is not a genome wide test in the conventional sense because statistical power is expended only in a fraction of the genome. In practice, we only test loci with an amplification frequency sufficiently large to detect allelic selection. The power to detect selected allelic amplification of a SNP depends on its amplification frequency as well as its heterozygosity rate in the sample set [10]. For example, in the ADT, only SNPs with at least nine heterozygote calls in amplified samples have a chance of achieving a nominal two-tailed P-value <0.005. By deciding a priori that P-values above this level will not be considered significant in downstream analysis, we dramatically reduced the candidate loci under interrogation (Figure S2), decreasing the de facto number of SNPs to be tested by 91.9% from 547,458 to 44,132. This represents a far smaller multiple testing burden than in germline genome wide association studies (GWAS). Such reduction in testing burden improves our power to detect true effects.
Figure 2A presents the amplification distortion signals for SNPs tested along the genome. The statistical association for all but the top-scoring SNPs closely follows the distribution expected under the null hypothesis (Figure 2B), attesting to the validity of the assumptions. Although no single SNP achieves genome-wide significance, our results do yield a larger number of SNPs with lower P-values than would be expected by chance. Specifically, given the distribution of amplified heterozygotes in our data, we would expect an average of 114 SNPs to attain a P-value below 0.005 (95% confidence interval 98–132) under the null hypothesis of no random allelic amplification (Materials and Methods). In the actual data, 139 SNPs surpass this threshold (Table S1). This suggests that a subset of the SNPs among these top 139 is likely subject to selective allelic amplification. We checked the level of linkage disequilibrium (LD) these SNPs possess in HapMap CEU data (Table S1); 40 of these 139 SNPs are in strong LD in HapMap (r2≥0.7) with at least one other SNP within this set of 40 (Table S1, Figures S3, S4, S5). Note that our permutation scheme preserves LD structure, so that blocks of SNPs in LD can jointly contribute much of the signal not only in the actual data, but also during each permutation. Therefore, the 139 SNPs are still indeed more than expected by chance.
We should point out that one potential artifact arises from the fact that a germline copy number variant (CNV) gain might appear to be a somatic amplification when compared with the signal intensities from pooled normal samples. However, our methodology guards against this artifact in two ways (Materials and Methods). First, we call amplification in the tumor only if the intensity is greater than that of all normal samples in the study (Materials and Methods). Second, we call the amplified allele only if its allelic intensity is considerably larger in the tumor than in the matched normal. Finally note that, of our top 139 SNPs, only 23 (16.5%) are harbored in gains reported in the Database of Genomic Variants (DGV) (http://projects.tcag.ca/variation/), and of these only 8 are reported in more than three individuals in the database. Therefore, it is quite unlikely that germline CNVs significantly contribute to the ADT signal.
To investigate whether certain classes of genes may be driving our signals, we mapped each of the top 139 SNPs to the nearest gene (within at most 100 kb), which yielded 73 unique genes (Table S1). All but 22 of the 139 SNPs fell within 100 kb of a transcribed region, and 53 of the genes had single SNPs mapping to them. The largest number of significant SNPs mapping to the same gene was seven, all lying within the transcribed region of NSPR1 (see Table S1 for r2 LD values). We performed a gene ontology (GO) analysis [14] on the gene set to query for enrichment in specific annotations (Materials and Methods). The control set of genes for such analysis deserves special attention in this study, as gene sets may be over represented among our signals simply because they are over represented in amplified regions. To distinguish the signal driven by genes undergoing allelic selection from that driven by more general (non-allelic) amplification, this analysis was conducted by comparing our gene set with genes harboring (or near) SNPs that are recurrently amplified in our data. Thus, any observed enrichment in GO terms is above and beyond that which is due solely to general somatic amplification. This analysis allows us to query for signals from the allele-specific selection, controlling for those due to somatic amplification alone. The results are shown in Table 1 and Figure S6.
Among our five significant (FDR q-value <0.1) GO enrichments is the cellular component term synapse (P = 0.0006). Of the 73 genes harboring (or very near) SNPs below the ADT threshold of P-value<0.005, six (CADPS2, CHRM2, CHRNA4, GRM8, MAGI2, and SNAP25) possess this annotation. Notably, the brain-related enrichment is independent of synapse related genes undergoing amplification in brain tumors, since general amplification is controlled for in this analysis. Therefore, the synapse annotation emerged strictly from the ADT selection signal among SNPs already in regions amplified in this brain-tissue tumor. This may be indicative of tumor selection for particular variants in these specific synapse-annotated genes.
Interestingly, the most significantly enriched GO terms (Table 1) were positive regulation of kinase activity (P = 9.03×10−5; Benjamini-Hochberg corrected q-value 0.0471) and positive regulation of transferase activity (P = 0.000132; Benjamini-Hochberg corrected q-value 0.0471). The six genes in our gene set associated with these GO terms are AGK, DGKB, EGFR, INSR, KIT, and RELN. Each of these genes is the closest to a single significant SNP, with the exceptions of EGFR with two such SNPs, and RELN with three such SNPs (see Table S1 for r2 LD values). Furthermore (Table S1), each harbors one or more of the 139 SNPs within its transcribed region, with the exception of DGKB whose associated SNP is 66 kb downstream. To investigate the relative dependencies between the amplifications of these six genes, we examined the frequencies of their co-amplifications on a sample-by-sample basis. Of the six, four (AGK, DGKB, EGFR, and RELN) are located on chromosome 7. As expected, amplifications of these genes tend to co-occur far more often than would be expected by random assortment (Fisher's exact test P <10−20), largely due to the fact that amplicons often encompass most or all genes on the chromosome. The other two genes are located on chromosomes 4 (KIT) and 19 (INSR). Surprisingly, INSR is co-amplified with chromosome 7 genes in a statistically significant manner (Fisher's exact test P <10−6 for co-amplification with EGFR, odds ratio 14.2). On the other hand, KIT amplification is anti-correlated with that of the genes on chromosome 7 (P = 0.05 for anti-correlation with EGFR, odds ratio 0.5). Figure 3 provides an overview of the amplification association structure among these six genes. These correlation patterns may point to interdependent and/or alternative pathways that a tumor engages.
Given the preferential allelic amplification observed in some SNPs, we reasoned that an allele undergoing selection when amplified in a tumor may have an effect on disease risk as a germline predisposition variant. This principle has been previously demonstrated in mice [3] and in humans [6]–[9]. We therefore compared our list of the top 139 SNPs from our ADT analysis with the 406 SNPs reported in a recent GWAS for glioblastoma [15]. The rationale is that the variant that is selectively amplified in glioblastoma tumors may actually predispose the carrier to the initiation of the tumor, and thereby occur at a higher frequency in cases as compared to controls.
A pair of SNPs, rs4367471 and rs4132013 (r2 = 0.78 in CEU HapMap population, Table S1) within a single haplotype block (in the European populations) in an intron of the LHFPL3 gene, appears on both lists. Permutation analysis shows that an overlap of two or more SNPs between lists of these sizes (Materials and Methods) would be expected by chance only 2.1% of the time (P = 0.021). This is remarkable since the ADT makes use of no population control genotypes, while the GWAS study does not take tumor DNA into account. For both SNPs, the minor allele is overrepresented on amplified chromosomes (in the present study) and among glioblastoma cases (in the Wrensch et al GWAS). Among 24 amplified heterozygotes for rs4367471 in our study, 20 amplify the minor allele (P = 0.0015), while 27 of 32 amplified heterozygotes for rs4132013 amplify the minor allele (P = 0.00011). In the GWAS study, the rs4367471 minor allele frequency was 0.28 for glioblastoma cases, as compared to 0.23 for the disease-free controls (P = 0.00022). Similarly, the rs4132013 minor allele frequency was 0.24 for cases as compared to 0.19 for controls (P = 0.00042). The odds ratios were 1.28 for both SNPs after adjusting for population structure with the Eigenstrat software [16].
The selective advantage gained by a cell amplifying a specific allele of a gene may be acting through direct changes in a gene product (e.g., a missense SNP) or by regulatory changes that modulate the quantity of gene product. The latter option is a testable hypothesis – it predicts that the amplification of the selected-for variant will be associated with elevated transcript expression levels. To investigate whether any of the detected signals of selected allele-specific amplification associates with expression, we integrated the expression data from the tumor samples with the genotype and amplification status. We considered Affymetrix U133A expression array data from the 154 individuals in our sample set for which the data was available from the TCGA website. Our list of top SNPs includes 65 whose nearest (as measured by base pair distance to transcribed region) gene is represented on the expression array. Of the 65 SNPs, only 28 had at least 5 examples of each SNP allele being amplified among the 154 samples, and were thus available for testing this association. Topping this list of 28 SNPs were rs6959338 and rs13222385, intronic variants in DOCK4 (chr 7q31.1) and EGFR (chr 7p11.2), respectively: rs6959338 shows amplification of the T allele over the C allele in 33 of 41 amplified heterozygotes (P = 1.1×10−4); rs13222385 amplifies the G allele over the A allele in 35 of 45 amplified heterozygotes (P = 2.5×10−4). Intriguingly, the expression data shows statistically significantly higher expression in samples amplifying the selected-for allele than in those amplifying the other allele (Figure 4) in both DOCK4 (P = 0.027) and EGFR (P = 0.015). To pursue this idea further, we tested the expression levels of the genes immediately flanking EGFR and DOCK4 for association with amplification of the selected-for alleles. We were interested to discover that LANCL2, a gene 158 kb downstream from EGFR, has statistically significantly higher expression in rs13222385 heterozygotes amplifying the G allele than those amplifying the A allele (P = 0.0371). Taken together with the SNP's association with EGFR expression levels, this finding could point to a regulatory element, such as an enhancer for the allele or a linked variant.
We have presented a novel genome wide approach to identify genetic variants that are preferentially selected, via amplification, during tumor evolution. The ADT approach is statistically rigorous and is robust to the confounding effects of population stratification. The non-amplified chromosomal homolog provides the ideal matched control for the amplified homolog, as it comes from the same individual. Although no single SNP individually achieves genome-wide significance under the ADT (likely due to a lack of power owing to limited sample size [10]), our data does show enrichment in strong ADT signals as compared to chance. Currently, we are performing similar analyses in areas that have undergone copy number loss.
Our integrated analysis of genes harboring (or near) SNPs undergoing apparent allelic selection has revealed intriguing pathways and annotations. As revealed by the GO analysis, many of the variants showing ADT signals with P-values <0.005 are located within genes related to kinase activity. The fact that both the EGFR and KIT kinases reach statistical significance is of particular interest. KIT expression is often observed in gliomas, and imatinib (which is known to inhibit c-Kit) is currently being evaluated in clinical trials [17]–[21]. The correlation/anti-correlation relationships among these implicated genes may highlight glioblastomas that utilize different pathways and may therefore represent distinct subtypes of tumors that may be clinically relevant as has been recently described [22].
We also observed particular instances of the selectively amplified alleles driving higher expression in DOCK4 and EGFR. DOCK4 was originally isolated in a screen to identify homozygous genomic deletions during tumor progression in a mouse model and is part of a larger family of atypical guanine exchange factor (GEF) for Rho family GTPases [23]. Rho GTPases are highly conserved molecular regulators of cytoskeletal dynamics and influence many cellular processes including cell polarity and migration [24]. Interestingly, it has been previously shown that suppression of DOCK4 RNA reduces dendritic growth and branching in hippocampal neurons, while overexpression enhances these processes [25]. Moreover, increases in Dock180 levels, another Dock family member, enhanced migratory and invasive capacity in vitro, while inhibition of expression significantly reduced glioma cell invasion in vitro [26]. Therefore, we speculate that DOCK4 influences the invasive potential of gliomas and that the DOCK4 alleles may differentially modulate this potential. The role of EGFR in glioma biology is well established [27] (and references therein). Somatically acquired mutations of EGFR are commonly (∼40%–50%) observed in gliomas, and the EGFR pathway is commonly targeted in this disease [13], [28]–[32]. Our results further substantiate the importance of EGFR and demonstrate that particular alleles play important roles in determining EGFR expression levels. It will be of interest to study if expression differences in this gene lead to amplified or diminished phenotypic consequences. Indeed, a recent article demonstrates that subtle alterations in expression levels can lead to dramatic phenotypic consequences [33].
The apparent selection of specific inherited alleles when amplified is consistent with several biological interpretations. The data can be considered in the context of Knudson's two-hit hypothesis [34] in that the associated SNP alleles are inherited variants (or capture variants via linkage disequilibrium) that provide a selective advantage when amplified. Indeed, it has been demonstrated that inherited alleles of a locus (e.g., the Arg72 and Pro72 variants of TP53) can have differential mechanistic effects (e.g., apoptotic potential) [35]. Another explanation is that cis- acting germline determinants influence the acquisition of somatic mutations, which are subsequently acted on by selection. Elegant experiments supporting this hypothesis in mice and humans have recently been published [7]–[9], [36]. Third, one may hypothesize that a somatic mutation provides a selective advantage only when amplified on a specific haplotypic background, or is selected against if the mutation arises on other allelic backgrounds; that is, only certain alleles will tolerate the somatic mutation.
Since selection implies function, the loci identified in this study are high-priority candidates for further investigation. The results may provide a way to rationally identify subtypes of cancers that are driven by distinct risk loci. If this is the case, then genome wide association studies for cancer risk may benefit from typing matched tumor DNA samples, in addition to germline DNA, and performing an integrative analysis. Alleles that do not affect risk predisposition may still yield important clues with respect to acquired tumor traits, such as angiogenesis, tissue invasiveness, evasion of apoptosis, etc. Functional studies, such as allele specific RNA interference for protein coding regions or somatic cell knock-in of alleles, may shed light on the mechanistic consequences of the alleles.
In summary, we demonstrate that integrating information from germline and tumor genomes can reveal aspects of tumor biology that are not readily identified by studying each genome in isolation.
We obtained glioblastoma array data (GBM Publication Data Freeze) from the ftp site of TCGA. We utilized three different data types – germline genotypes, amplification status, and allelic imbalance – from various hybridization-based platforms, downloaded from the TCGA ftp site. First, germline SNP genotypes (Illumina platform) for all normal samples were obtained. Second, we accessed copy number segmentation data (from Affymetrix SNP array 6.0, Illumina HumanHap550, and Agilent CGH array 244A) for tumor samples, providing genomic regions for each individual that are inferred to have constant copy number along with the estimated “raw” (non-integer) copy number of that segment. Third, we obtained the raw allelic A and B signals for all samples (Illumina BAF measure), tumor and normal. This provides a raw measure of allelic imbalance, commonly termed the “B allele frequency” (BAF), defined as
We also obtained Supplementary Table 7 from the Wrensch et al GWAS [15], which lists 406 SNPs with p<0.001 for association with high grade glioma comparing cases from San Francisco
Bay Area Adult Glioma Study, 1997–2006 (AGS) and the Cancer Genome Atlas (TCGA) to AGS and Illumina controls (iControls).
For each of the three platforms (Affymetrix, Agilent, and Illumina), we first inferred amplification at all 1.3 million autosomal SNPs represented by the Affymetrix and Illumina arrays combined, as follows. First, for each sample, all SNPs harbored in each genomic segment from the sample's copy number segmentation file (see above) are assigned that segment's raw copy number. A SNP is called amplified by the platform in a tumor sample if its raw copy number in that sample exceeds its raw copy number in all normal samples. This conservative amplification calling procedure accounts for local probe intensity effects, and avoids miscalling germline copy number variants as somatic amplifications. Note that this procedure, while conservative, is designed to include single-copy gains as well as high-level amplification events. Finally, for all downstream analyses, a sample is considered to harbor an amplification at a SNP if it is called amplified by at least two of the three platforms.
For each SNP, we restrict the remainder of our analysis to individuals that are both heterozygous in the germline and amplified in the tumor at the SNP site. For each of these samples, we aim to determine which of the two alleles is amplified. Towards this end, we exploit the BAF measure described above. Since each sample is heterozygous in the germline, we expect the SNP's BAF measure to be near 0.5 in the germline. A tumor BAF larger than 0.5 is indicative of B allele amplification, and a BAF smaller than 0.5 is indicative of A allele amplification. However, bias in A and B intensity measures can result in deviations from these expectations. We therefore rely on the deltaBAF measure, defined as
The expectation here is that A (respectively, B) allele amplification will result in a negative (respectively, positive) deltaBAF value. To avoid erroneous deltaBAF calls due to noisy probe intensities, we only have confidence in allele calls where |deltaBAF| >0.05. That is, for heterozygous (in the germline) samples that are amplified (in the tumor), we call A allele amplification if deltaBAF <−0.05 and B allele amplification if deltaBAF >0.05.
The procedure described above yields sample counts for A amplification and B amplification at each SNP. Let nA and nB, respectively, denote these counts. Under the null hypothesis of random allelic amplification, nA follows a Binomial(nA + nB, 0.5) distribution. In other words, if there is no causal allele or site within an amplified region, the distortion signature of each SNP within the amplified region should conform to the null signature on the binomial distribution. Therefore, a (two-sided) P-value testing preferential allelic amplification may be performed by comparing nA with this distribution in the obvious manner. The chance of a non-causal/non-associated allele within an amplified region being randomly selected enough times to result in a distortion (i.e. false positive) is α, where α represents a chosen level of significance as described in the following section.
Although not a genome-wide association scan, our approaches comprise many tests whose correlations are manifold and complicated. Furthermore, some regions harbor more amplifications than others and therefore have a higher a priori likelihood of displaying allelic distortion even under the null hypothesis. Therefore, analytically determining genome-wide significance from the test statistics is not straightforward. To address this, we developed a permutation procedure that assesses the significance of our results. For each run of the procedure, we first randomly determined – at each sample chromosome pair – whether to swap amplification status from the amplified allele to the non-amplified allele at all amplified (in the tumor) SNPs on the chromosome. This preserves haplotype and amplicon structure while destroying correlation between the two. We then recomputed the test statistics across the genome. In this manner, the amplification status of the samples is preserved, and we are randomly sampling from the null situation of non-preferential (random) amplification. For the ADT test, these simulations produced an average of 114 SNPs (95% confidence interval 98–132) surpassing the 0.005 threshold. For the qq-plot, the qth null P-value quantile was estimated by averaging the qth quantiles of P-values from 1000 permutations. Finally, the significance of the overlap between the ADT SNPs and the Wrensch et al GWAS SNPs was assessed by permuting, 1000 times, and retaining the 139 most significant SNPs for each permutation (since our actual data generated 139 top SNPs). These number of SNPs in each permutation that intersected with the Wrensch list was tallied for each permutation, which yielded the expected distribution expected by chance.
For our GO analyses, we compared the gene list with a “gene universe” comprised of all genes that had any a priori chance of demonstrating preferential allelic amplification, at the P<0.005 nominal level, in our data. For a given gene, this depends upon many factors, including amplification frequency and allele frequencies of nearby array SNPs. We restricted the gene universe to genes that were within 100 kb of a HumanHap550 array SNP that is heterozygous and amplified in at least nine of our samples. This is reasonable, as these are the only genes (by definition) that have an a priori chance of having an associated SNP with ADT P-value below 0.005. This left 2696 genes as a reference set. Using the Ontologizer [37] software, we assessed our gene lists for enrichment in GO terms, as compared with this reference gene universe, using the Term-For-Term method and Benjamini-Hochberg correction.
For each pair of genes, we constructed a 2×2 table of counts for number of samples in each category of amplification/non-amplification status for each gene. Using this table, we computed the odds ratio estimate for correlation between amplification of the genes, and assessed its significance using Fisher's exact test. For genes X and Y, this corresponds to the ratio of the odds of gene X being amplified in a sample with gene Y amplified to the odds of gene X being amplified in a sample without gene Y amplified.
Expression levels from the Affymetrix 133A array were downloaded from the TCGA website. For each gene/SNP combination, expression differences between samples expressing each of the two alleles were computed using the non-parametric Wilcoxon rank sum test. The test was one-sided, since there was an a priori hypothesis that the preferentially amplified allele would result in a higher expression level.
|
10.1371/journal.ppat.1007031 | Plasmodium falciparum dipeptidyl aminopeptidase 3 activity is important for efficient erythrocyte invasion by the malaria parasite | Parasite egress from infected erythrocytes and invasion of new red blood cells are essential processes for the exponential asexual replication of the malaria parasite. These two tightly coordinated events take place in less than a minute and are in part regulated and mediated by proteases. Dipeptidyl aminopeptidases (DPAPs) are papain-fold cysteine proteases that cleave dipeptides from the N-terminus of protein substrates. DPAP3 was previously suggested to play an essential role in parasite egress. However, little is known about its enzymatic activity, intracellular localization, or biological function. In this study, we recombinantly expressed DPAP3 and demonstrate that it has indeed dipeptidyl aminopeptidase activity, but contrary to previously studied DPAPs, removal of its internal prodomain is not required for activation. By combining super resolution microscopy, time-lapse fluorescence microscopy, and immunoelectron microscopy, we show that Plasmodium falciparum DPAP3 localizes to apical organelles that are closely associated with the neck of the rhoptries, and from which DPAP3 is secreted immediately before parasite egress. Using a conditional knockout approach coupled to complementation studies with wild type or mutant DPAP3, we show that DPAP3 activity is important for parasite proliferation and critical for efficient red blood cell invasion. We also demonstrate that DPAP3 does not play a role in parasite egress, and that the block in egress phenotype previously reported for DPAP3 inhibitors is due to off target or toxicity effects. Finally, using a flow cytometry assay to differentiate intracellular parasites from extracellular parasites attached to the erythrocyte surface, we show that DPAP3 is involved in the initial attachment of parasites to the red blood cell surface. Overall, this study establishes the presence of a DPAP3-dependent invasion pathway in malaria parasites.
| Malaria remains one of the most devastating infectious diseases and its clinical manifestation is caused by the exponential multiplication of parasites in patients. This asexual replication cycle consists of red blood cell invasion, intracellular parasite multiplication, and release (also known as egress) of daughter parasites for further red blood cell invasion. Host cell invasion is therefore essential for parasite replication and the only moment in this cycle when parasites are exposed to the immune system. Understanding the molecular mechanisms that control red blood cell invasion might not only lead to the identification of novel antimalarial targets but also to the development of better invasion blocking vaccines. DPAP3 is a druggable cysteine protease that was previously believed to be essential for parasite egress. In this study, we show that parasites lacking DPAP3 activity are unable to efficiently invade red blood cells but escape the confines of the host cell normally. Overall, this study increases our understanding of the proteolytic pathways that govern host cell invasion by the malaria parasite.
| Malaria is a devastating infectious disease caused by Apicomplexan parasites of the Plasmodium genus and is transmitted by Anopheles mosquitoes during a blood meal. After an initial asymptomatic liver infection, parasites are released into the blood stream where they replicate within red blood cells (RBCs). This asexual exponential growth is responsible for all the pathology associated with malaria, causing close to half a million deaths every year[1]. Over the last 15 years, the world has seen a significant decrease in malaria incidence mainly due to the distribution of insecticide-impregnated bed nets and the introduction of ACT (artemisinin-based combination therapy) as the standard of care for uncomplicated malaria[2]. However, the recent emergence of artemisinin resistance[3] has made the identification of viable therapeutic targets extremely important[4,5].
P. falciparum is the most virulent Plasmodium species accounting for most of malaria mortality. Its 48 h asexual erythrocytic cycle consists of RBC invasion, intraerythrocytic parasite growth and division into 16–32 daughter merozoites, followed by parasite egress for further RBC invasion. Parasite egress and RBC invasion are key for parasite replication and blocking either one of these processes would lead to a quick drop in parasitemia and malaria pathology. Proteases have been shown to play essential roles in both processes and might therefore be viable therapeutic targets[6].
RBC invasion is a multistep process involving initial recognition of RBC receptors by adhesin proteins on the surface of the merozoite (invasive extracellular parasite form), tight attachment to the RBC membrane (RBCM), reorientation of the merozoite apical end towards the RBCM, active invasion driven by an actin-myosin motor with invagination of the RBCM and formation of the parasitophorous vacuole (PV), and finally, sealing of the RBCM and PV membrane (PVM)[7,8]. The PV is a membrane-bound vacuole within which the parasite grows and replicates isolated from the RBC cytosol. Rupture of the PV and RBC membranes is required for parasite egress and is mediated by proteases. In particular, subtilisin-like protease 1 (SUB1), an essential serine protease residing in apical secretory organelles known as exonemes, is released into the PV right before egress where it processes several proteins important for egress and invasion[9–13]. These include cleavage and likely activation of serine repeat antigen 6 (SERA6)[14], an essential papain-fold cysteine protease[15–17].
In a forward chemical genetic approach, P. falciparum dipeptidyl aminopeptidase 3 (DPAP3) was identified as a potential regulator of parasite egress acting upstream of SUB1[18]. DPAPs are papain-like cysteine proteases that cleave dipeptides off the N-terminus of protein substrates[19]. In that study [18], the vinyl sulfone inhibitor SAK1 was shown to preferentially inhibit DPAP3 over other cysteine proteases. This compound arrests parasite egress at mid-micromolar concentrations, blocks processing of SUB1 substrates, and prevents proper expression and maturation of SUB1 and apical membrane antigen 1 (AMA1), a micronemal protein secreted onto the merozoite surface that is essential for RBC invasion[9,20,21]. These results led to the hypothesis that DPAP3 might act as a general maturase of secretory proteins involved in egress and invasion. However, while the function (or essentiality) of P. falciparum DPAP3 has not been validated genetically, in the rodent parasite P. berghei, DPAP3 knock out (KO) parasites are viable but replicate significantly slower[22–24].
Here, we combine chemical, biochemical and conditional genetic approaches to show that DPAP3 is an active protease that resides in apical secretory organelles, and that its activity is critical for efficient RBC invasion. We also provide very strong evidence showing that DPAP3 does not play a significant function in parasite egress.
Using single homologous recombination we were able to replace the endogenous catalytic domain of dpap3 with a C-terminally tagged (GFP, mCherry or HA) version. However, our multiple attempts to replace the DPAP3 catalytic Cys with a Ser failed despite using the same homology region upstream of the catalytic domain (Fig 1A and 1B and S1A and S1B Fig). Our attempts to KO dpap3 by double homologous recombination also failed. These results strongly suggest that DPAP3 activity is important for parasite development. Parasites containing differently tagged dpap3 were cloned by limited dilution, and the clones selected for this study will be referred to as DPAP3-GFP, DPAP3-mCh, and DPAP3-HA. Analysis of parasite extracts by western blot (WB) using a polyclonal antibody that targets the C-terminal half of DPAP3 showed a shift in migration pattern in accordance with the tag molecular weight (Fig 1C).
In all our tagged lines, DPAP3 consistently localizes to the apical pole of merozoites (Fig 1D). To determine when it is expressed, tightly synchronized DPAP3-HA and DPAP3-mCh parasites were collected every 3 h throughout the erythrocytic cycle and were either lysed and analyzed by WB (Fig 1E), or fixed for immunofluorescence analysis (IFA, Fig 1F and S1C Fig). Consistent with its transcription profile[25], DPAP3 is most abundant in late schizonts and merozoites, but could also be detected in rings and trophozoites by WB. To confirm that the small amount of DPAP3 observed by WB at ring stage is not due to schizonts contamination in our cultures, we analyzed 50 fields of five Giemsa-stained thin blood smears at 1 hour post invasion (h.p.i.), i.e. around 75,000 RBCs. We did not observe any schizonts in these slides, only ring stage parasites at around 5% parasitemia. Also, WB analysis using an MSP1 (merozoite surface protein 1) antibody as a marker of schizogony showed no significant staining at ring or trophozoite stages (S1D Fig). Although we could detect multiple processed forms of DPAP3 by WB, these processed forms are rarely observed in live parasites (Fig 1C) and are likely an artefact of parasite lysis (see below).
By IFA, DPAP3 was first detected in young schizonts (6–8 nuclei) and seems to be expressed at the same time as rhoptry (rhoptry neck protein 4, RON4, and high molecular weight rhoptry protein 2, RopH2) and inner membrane complex (glideosome-associated protein 45, GAP45) proteins (Fig 1F and S1 and S2 Figs). During schizont maturation, DPAP3 and rhoptry proteins localization changes from a diffuse and granular cytosolic staining to a clear punctuated apical staining in daughter merozoites (Fig 1F and 1G and S1E and S2 Figs), probably reflecting protein trafficking and organelle biogenesis. By contrast, exonemal (SUB1) and micronemal proteins (AMA1) are expressed and localize to their apical organelles at a later stage (Fig 1F and S2 Fig). Note that the diffuse staining observed at 39 h.p.i. for DPAP3, RON4 and RopH2 is not background fluorescence signal given that no staining was observed in these same slides for the few parasites that were lagging behind in development, i.e. infected RBCs (iRBCs) with a single nucleus (S2A Fig).
Using standard confocal microscopy, we could not observe consistent colocalization of DPAP3 with any apical organelle marker tested: AMA1 and EBA175 (erythrocyte binding antigen 175) for micronemes, SUB1 for exonemes, RON4 and RopH2 for rhoptries, and Exp2 (exported protein 2) for dense granules (Fig 1F and S2B Fig). We therefore decided to increase the resolution of our images by using structured illumination microscopy (SIM). By SIM we could clearly observe DPAP3 staining in small but well-defined dot-like structures at the apical pole of each merozoite in the DPAP3-HA (Fig 1H and S3A Fig) and DPAP3-GFP (S3B Fig) lines. Although we did not observe colocalization with exoneme (SUB1), microneme (EBA175) or rhoptry (RON4, RopH2) protein markers (Fig 1H and S3A and S3B Fig), DPAP3 seems to be closely associated with RON4, suggesting that it resides in apical organelles that surround the neck of the rhoptries (Fig 1H).
Timely discharge of proteins from the different apical organelles is crucial to regulate parasite egress and RBC invasion[26]. Secretion of exonemal and micronemal proteins is mediated through activation of cGMP-dependent protein kinase G (PKG), which takes places 15–20 min before egress[9]. Release of SUB1 into the PV leads to parasite egress. Secretion of micronemal proteins, such as AMA1 or EBA175, onto the parasite surface is essential for merozoite attachment to the RBCM and invasion. To determine whether DPAP3 is secreted, we used the DPAP3-HA line to compare the level of DPAP3 present within parasites, in the PV and RBC cytosol, and in the culture supernatant at three different stages of egress: schizonts arrested before exoneme/microneme secretion with the PKG reversible inhibitor compound 2 (C2), schizonts arrested between PVM and RBCM breakdown using the general cysteine protease inhibitor E64, and free merozoites collected after egress. In E64-arrested schizonts, the erythrocyte is still intact but the RBCM is highly porated allowing leakage of RBC and PV proteins into the culture supernatant[17,27]. Parasite pellets, PV and RBC cytosol proteins, and proteins secreted in the culture supernatant under these three conditions were collected after treating the cultures with the fluorescent activity-based probe FY01[28]. FY01 is a cell-permeable probe that covalently modifies the catalytic Cys of cysteine proteases including DPAP3[18]. Fluorescently labelled DPAP3-HA can then be visualized by in-gel fluorescence in a SDS-PAGE gel as a band running around 130 kDa that matches the WB band observed with a HA antibody (Fig 2A and S4A Fig). DPAP3 was mainly detected in the culture supernatant and saponin soluble fraction of E64-arrested or rupturing schizonts, but not in C2-arrested schizonts, indicating that it is secreted downstream of PKG activation but before merozoites become extracellular. The presence of DPAP3 in E64-arrested schizont pellets and free merozoites suggests only partial secretion before egress. To confirm proper fractionation of our samples, we used SERA5, BiP (binding immunoglobin protein) and Hsp70 (heat shock protein 70) as PV, ER, and cytosol protein markers, respectively, in WB analysis (Fig 2A). Processing of SERA5 was also used to confirm that C2 or E64 treatments arrested parasite development at the expected stage. Upon exoneme secretion, SUB1 sequentially cleaves SERA5 into a 73 and 56 kDa forms. This processing is blocked by C2 since exoneme secretion is regulated by PKG. SERA5 is further processed into a 50 kDa form by an unknown cysteine protease that is inhibited by E64. Two additional biological replicates showing secretion of DPAP3 are shown in S4B and S4C Fig.
To determine the timing of DPAP3 secretion, DPAP3-mCh schizonts were arrested with C2 and parasite egress monitored by live microscopy after C2 wash out. Using this assay, PVM breakdown is clearly observable on the DIC channel when merozoites become more spread out within the RBC and their shape is better defined[12,27]. This is followed by RBCM rupture and merozoites dispersal (Fig 2B and S1–S3 Videos). Quantification of fluorescence signal in these egress videos shows a 40% decrease in mCherry signal right after PVM breakdown but before RBCM rupture (Fig 2C). Interestingly, conditional KO (cKO) of SUB1 has recently been shown to prevent breakdown of the PVM but not AMA1 secretion, suggesting that microneme secretion takes place before PVM breakdown [17]. This implies that DPAP3 secretion probably takes place downstream of microneme secretion and coincides with PVM breakdown.
To confirm that DPAP3 is secreted into the RBC cytosol after PVM breakdown, we performed immunoelectron microscopy (IEM) on DPAP3-GFP schizonts collected at the time of egress (Fig 2D and S3D and S3E Fig). In schizonts containing an intact PVM, DPAP3-GFP staining was observed at the apical end of merozoites, and in some sections, in close proximity to the rhoptries or enclosed within membrane bound vesicles, thus supporting our IFA results. In IEM images collected on schizonts after PVM breakdown, immunostaining was mainly observed in the RBC cytosol. Almost no DPAP3 staining was detected in the PV of schizonts with an intact PVM.
DPAPs are generally processed from a zymogen form (full-length protein after removal of the signal peptide) into an active form through removal of an internal prodomain and cleavage of the catalytic domain into two polypeptides[29,30] (S5A Fig). Three different isoforms of DPAP3 (p120, p95, and p42) consistent with the canonical processing of DPAPs were previously shown to be labelled by FY01 in merozoite lysates[18]. However, we have now shown that this processing is an artefact of parasite lysis (S5B–S5D Fig and S1 Text). In addition, our WB and FY01-labelling experiments show that full length DPAP3 (p120) is the predominant form found in live parasites (Figs 1 and 2 and S5 Fig).
To determine whether this full-length p120 form is active, we recombinantly expressed wild type (WT) and mutant (MUT, replacement of the catalytic Cys504 to Ser) DPAP3 in insect cells using the baculovirus system[31]. Expression and purification of recombinant DPAP3 (rDPAP3) from insect cells culture supernatant yielded predominantly the p120 and p95 forms (Fig 3A and S5B Fig). While WT DPAP3 is able to efficiently cleave the VR-ACC fluorogenic DPAP substrate[32], no activity was observed with MUT DPAP3 (Fig 3B). We also show that rDPAP3 is active under mild acidic conditions with an optimal pH of 6 (Fig 3C). Importantly, in one of our purifications we were able to separate the p120 form from a fraction containing a mixture of the p95 and p120 forms. We used these fractions to show that the p120 form is fully active (S5E Fig). This result strongly suggests that the predominant DPAP3 form present in live parasites (p120) is the one performing a biological function.
Since we were unable to directly KO DPAP3, we generated DPAP3 cKO lines on the 1G5 parasite line background that endogenously expressed DiCre[33]. In the DiCre system, Cre recombinase is split into two domains fused to rapamycin (RAP) binding domains. Addition of RAP triggers dimerization and activation of DiCre, leading to rapid recombination of specific DNA sequences known as loxP sites[34,35]. We used this system to conditionally truncate the catalytic domain of DPAP3 rather than excising the full gene to prevent potential episomal expression of DPAP3 after excision.
Two independent strategies, which differ in how the first loxP site was introduced within the dpap3 open reading frame (ORF), were used (Fig 4A and S6A Fig). In both cases, one loxP site was introduced downstream of the 3’-UTR. The other was inserted either within an Asn-rich region of DPAP3 predicted not to interfere with folding or catalysis, or within an artificial intron (loxPint), which does not alter the ORF of the targeted gene and has been shown to be well tolerated in several P. falciparum genes[12,17,36,37]. In both instances, the recodonized catalytic domain was tagged with mCherry such that RAP-induced truncation would result in the loss of mCherry signal. A control line containing only the 3’-UTR loxP site was also generated. After transfection of 1G5 parasites, drug selection, and cloning, three DPAP3cKO clones (F3cKO and F8cKO with loxPint, and A1cKO with loxP in Asn-stretch), and the E7ctr line (only one loxP) were selected for further studies. Evidence of integration by PCR is shown in S6B Fig. Analysis of genomic DNA of DMSO- or RAP-treated cKO lines by PCR showed highly efficient excision (Fig 4B) resulting in the loss of DPAP3-mCh expression in mature schizonts (Fig 4C and 4D). Although we consistently achieved more than 95–99% excision efficiency (Fig 5A), a fraction of non-excised parasites was always present after RAP treatment, which explains the presence of a non-excision DNA band after RAP treatment (Fig 4B).
To confirm that any phenotypic effect observed upon conditional truncation of DPAP3 is due to the loss of DPAP3 activity, A1cKO and F8cKO parasites were transfected with plasmid expressing WT or MUT DPAP3-HA under the control of the dpap3 or ama1 promoters (S6A Fig), resulting in the following complementation lines: F8cKO+WTdpap3, F8cKO+MUTdpap3, F8cKO+WTama1, F8cKO+MUTama1, A1cKO+WTdpap3, A1cKO+WTama1, and A1cKO+MUTama1. All complementation lines grew normally before RAP treatment and showed no apparent delay in parasite development.
IFA analysis confirmed colocalization between chromosomal DPAP3-mCh and episomal DPAP3-HA (Fig 5B and S6C Fig). Episomal expression was only high enough to be detected by IFA in 60–80% of schizonts, but was independent of RAP treatment (Fig 5C). This is probably due to different levels of episomal expression and plasmid segregation in schizonts. Efficient, but not complete, truncation of DPAP3-mCh was observed in all our complementation lines (Fig 5A). Labelling of DPAP3 with FY01 in parasite lysates from these lines show clear labelling of chromosomal DPAP3-mCh and episomal WT DPAP3-HA but not MUT DPAP3-HA (Fig 5D). As expected, RAP treatment results in a decrease of labelling of chromosomal DPAP3-mCh but not of episomal WT DPAP3-HA.
To measure the effect of DPAP3 truncation on parasite proliferation, we used the recently published plaque assay[16] where the wells of a 96-well flat bottom plate containing a thin layer of blood were seeded with ~10 iRBCs/well. After 10–14 days, microscopic plaques resulting from RBC lysis can be detected with an inverted microscope. RAP treatment of our cKO lines resulted in 90% less plaques, an effect that could be partially rescued through episomal complementation with WT but not MUT DPAP3 (Fig 6A, S1 Table). After RAP treatment of the F3cKO and F8cKO lines, some wells contained a single plaque, suggesting that only one clonal parasite population grew in these wells. Parasites present in 12 of these wells were propagated and dpap3 excision checked by PCR. All contained non-excised parasites but dpap3 excision was detected in three cultures (Fig 6B). The presence of excised parasites in some of the wells suggests that DPAP3KO parasites replicate less efficiently and might not have had enough time to form a visible plaque within the 14 days of the assay. That said, the presence of non-excised parasites in all samples indicates that WT parasites quickly outcompeted DPAP3KO ones.
To test this hypothesis, we perform standard parasite multiplication assays after RAP or DMSO treatment. To prevent parasite overgrowth, cultures were diluted 10-fold in fresh blood and media whenever parasitemia reached 5%. RAP treatment of DPAP3cKO parasites results in a 10- to 15-fold decrease in parasitemia after 3–4 cycles, corresponding to an overall 50% decrease in multiplication rate per cycle compared to DMSO treatment (Fig 6C). However, these values underestimated the importance of DPAP3 on parasite replication since 5 cycles after RAP treatment, 60% of iRBC are non-excised parasites expressing DPAP3-mCh (S7A Fig). This result proves that the small fraction of non-excised parasites quickly outcompetes the excised ones. After RAP treatment, parasites complemented with WT DPAP3 grew significantly faster than those complemented with MUT DPAP3 (Fig 6D). Importantly, our multiple attempts to clone DPAP3KO parasites after RAP treatment failed, indicating that DPAP3 activity is required for parasite proliferation under our culturing conditions.
To determine which point of the erythrocytic cycle is disrupted by the loss of DPAP3, a tightly synchronized culture of A1cKO parasites at ring stage was treated with DMSO or RAP for 3 h, and the culture monitored for the following 80 h. Samples were collected every 2–4 h, fixed, stained with Hoechst, and analyzed by FACS. No significant difference in DNA staining was observed between WT and KO parasites, suggesting that DPAP3 is not required for intracellular development (Fig 6E, left graph). Quantification of iRBCs belonging to the first or second cycle after treatment shows that DPAP3KO parasites egress at the same time as WT but produce ~50% less rings (Fig 6E, right graph). This suggests that DPAP3 is only important for RBC invasion, which is in direct contradiction with its previously suggested role in egress[18].
Despite being expressed early during schizogony (Fig 1F), we did not observe any delay in parasite development between 36–48 h.p.i. (Fig 6E). This was confirmed by IFA by counting the number of mature schizonts at the end of the cycle after RAP treatment. Apical localization of SUB1 to the exonemes was used as a marker for schizont maturity. No significant difference was observed between DMSO and RAP treatment of cKO or complementation lines (S7B Fig). Importantly, proper localization of MSP1, SUB1, AMA1, EBA175, RON4, and RopH2 was observed in all RAP-treated cKO lines (Fig 4D and S7C and S7D Fig).
Previously published work using the SAK1 inhibitor showed arrest of egress upstream of SUB1 activation[18]. This result suggested that DPAP3 might be important for parasite egress. Although we have been able to reproduce these results, we show that SAK1 treatment of schizonts 6 h before egress arrests schizogony upstream of SUB1 and AMA1 expression rather than parasite egress (S8 Fig). This explains why no SUB1 or AMA1 could be detected by WB in the previous study[18]. In addition to SAK1, we also synthesized a more selective DPAP3 inhibitor by replacing the nitro-tyrosine N-terminal residue of SAK1 with L-Trp (L-WSAK) and its diastereomer negative control containing D-Trp (D-WSAK). To test the specificity of these inhibitors, merozoite or schizont lysates were pre-incubated with a dose response of compound followed by 1 h labelling with FY01 (Fig 7A). Although SAK1 blocks labelling of DPAP3 at lower concentration than L-WSAK, it also inhibits all the falcipains (FP1, FP2, and FP3) above 5 μM. L-WSAK only inhibits other targets above 200 μM. As expected, the D-WSAK control compound is unable to inhibit any of the labelled cysteine proteases and is at least 100-fold less potent than L-WSAK at inhibiting DPAP3 (Fig 7A). We then compared the effect of these inhibitors in parasite egress on the A1cKO line. Surprisingly, all compounds blocked egress independently of RAP treatment (Fig 7B), and no difference in potency between L- and D-WSAK was observed. These results prove that these vinyl sulfone compounds do not act through inhibition of DPAP3 but rather through off-target or toxicity effects.
As a final proof to show that DPAP3 is not involved in parasite egress, we arrested DMSO or RAP treated DPAP3cKO schizonts with C2 and monitored egress by live microscopy after removal of the PKG inhibitor. Analysis of these videos showed no significant difference in the number of schizonts that ruptured, nor on how fast merozoites egressed after C2 wash out (Fig 7C and S4 Video). Also, we could not detect differences in the levels and/or processing of AMA1[38] or SUB1 substrates (SERA5[10] and MSP1[13]) between DMSO and RAP treated parasites (Fig 7D). These findings together with the lack of colocalization between DPAP3 and SUB1 (Fig 1H and S3 Fig) clearly demonstrate that DPAP3 is not responsible for proper processing and activation of SUB1, and that DPAP3 does not play a role in egress. However, its localization in an apical secretory organelle (Figs 1 and 2) and the time-course analysis of the A1cKO line (Fig 6E) strongly suggest a function in RBC invasion.
Mature schizonts obtained after DMSO or RAP treatment of our different parasite lines were incubated with fresh RBCs for 8–14 h, fixed, and the population of schizonts and rings quantified by FACS (Fig 8A). On average, we observed a 50% reduction in the number of iRBCs after RAP treatment of our cKO lines. This invasion defect could be rescued by episomal expression of WT but not MUT DPAP3 independently of the promoter used (ama1 or dpap3), thus indicating that DPAP3 activity is important for RBC invasion (Fig 8B).
To determine which invasion step is impaired by the loss of DPAP3, DMSO- or RAP-treated cKO parasites were arrested at schizont stage with C2, incubated with fresh RBCs after C2 washout, and samples collected at different time points for FACS analysis. Fluorescent wheat germ agglutinin (WGA-Alexa647) binds to lectins on the RBC surface and when combined with Hoechst staining allows us to differentiate free merozoites from iRBCs (Fig 8C). Quantification of the different parasite stage populations over time clearly shows a decrease in the number of rings upon RAP treatment, with an inversely proportional increase in the number of free merozoites (Fig 8C and S9A Fig).
The RBC population showing similar levels of DNA signal as free merozoites is mainly composed of ring stage parasites but likely contains a small proportion of extracellular merozoites tightly attached to the surface of the erythrocyte. These two populations were differentiated by staining the samples with a monoclonal MSP1 antibody (m89.1), whose epitope is within the portion of MSP1 that is shed during invasion (Fig 8D). Using this assay, we did not observe a significant difference in the number of attached merozoites relative to the number of rings between DMSO and RAP treatment of our cKO lines. This suggests that after DPAP3KO parasites tightly attach to the RBC surface they invade as efficiently as WT parasites. Therefore, DPAP3 is likely important for the initial recognition of and attachment to RBCs.
This study provides the first characterization of the biological function of DPAP3 in parasite development. We have shown that DPAP3 is important for efficient RBC invasion and parasite proliferation. DPAP3 is expressed early during schizogony, it localizes to small apical organelles that are closely associated with the neck of the rhoptries, and it is secreted at the time of PVM breakdown but before parasite egress. We have also demonstrated that DPAP3 has dipeptidyl aminopeptidase activity but contrary to other DPAPs, removal of the prodomain is not required for activation. Our cKO and complementation studies provide strong evidence that DPAP3 activity is only required for efficient RBC invasion, but not for intracellular parasite development or parasite egress. Importantly, we have proven that the block in egress phenotype previously reported using vinyl sulfone inhibitors is not due to DPAP3 inhibition but rather to off-target or toxicity defects. Two cysteine proteases have been shown to play an important role in egress: human calpain-1 [39] and SERA6 [17]. It is therefore possible that the block in egress phenotype might be due to inhibition of either one of these two proteases. This study illustrates the importance of using negative control compounds when trying to associate a specific phenotype to the inhibition of a particular target, as well as the need to genetically validate functional information obtained through chemical methods [6]. Interestingly, in a recently published study, the Koning-Ward lab used the glmS riboswitch system to conditionally knockdown DPAP3 in P. falciparum [40]. However, they did not observe any significant effect in parasite proliferation, nor in egress or RBC invasion. We think that this is likely due to the fact that they only achieved partial knockdown of DPAP3, and that the residual level of DPAP3 activity was sufficient to perform its function. That said, their localization studies are consistent with ours and confirm that DPAP3 resides in novel apical secretory organelles, and that it does not colocalize with rhoptry, microneme, or dense granule markers.
Our initial characterization of the invasion defect associated with the loss of DPAP3 suggests that this protease might play a role in the attachment of merozoites to the RBC surface. Indeed, in our invasion assays we observed a significant increase in the number of free merozoites upon cKO of DPAP3. Moreover, while DPAP3 KO results in a significant decrease in the number of rings, the ratio between merozoites attached to the RBC surface and those that have successfully invaded RBCs is the same between WT and KO parasites, suggesting that the decrease in invasion efficiency is likely upstream of merozoite attachment and tight junction formation, and that DPAP3 is likely important for the initial recognition of RBCs. We think it is unlikely that DPAP3KO parasites might be less efficient at forming a tight junction since we predict that such a defect would result in an increase of attached parasites relative to the total number of invaded RBCs. However, this is a possibility that we cannot completely rule out at this moment.
Interestingly, we observed a wide variation in the decrease of invasion efficiency upon cKO of DPAP3, ranging from 25 to 75% inhibition. This variation is much larger than the experimental variation observed with our E7 control line, and it is likely due to heterogeneity among the different batches of blood used to perform invasion assays. This observation suggests that DPAP3 activity might be important to recognize host cell receptors that are differentially expressed in the human population, and is consistent with our propose role of DPAP3 in RBC attachment.
The facts that DPAP3 is expressed early during schizogony, that it localizes in secretory apical organelles, and that it is active under mild acidic conditions (pH 5–7, maximum activity at pH 6), suggest that DPAP3 might process its substrates within the apical organelle where it resides. However, at this stage we cannot discard the possibility that DPAP3 might process its substrates at neutral pH, either during trafficking (in the ER or Golgi), extracellularly after secretion, or even during secretion of proteins from apical organelles. Micronemes are trafficked to the merozoite apex underneath the inner membrane complex before releasing their protein cargo, thus coming in very close proximity to the neck of the rhoptries [41,42]. It is therefore possible that DPAP3 might interact with proteins of other secretory organelles as they are secreted, similarly to how micronemal CyRPA (cysteine rich protective antigen) and Ripr (Rh5 interacting protein) come together with rhoptry Rh5 (reticulocyte binding protein homologue 5) at the merozoite apex after egress [43]. Finally, we think it is unlikely that DPAP3 acts extracellularly on RBC surface proteins because co-culturing equal amounts of WT and DPAP3KO parasites did not rescued the DPAP3KO invasion defect (S9B Fig).
It is difficult to speculate about the nature of DPAP3 substrates since we could not colocalize this protease with any of the tested apical markers. However, its substrates are likely to be proteins in the secretory pathway that are directly or indirectly important for invasion. Also, given that DPAPs cleave N-terminal dipeptides from protein substrates, DPAP3 likely recognizes the N-terminus of its substrates after they have been cleaved by another protease. Most proteins secreted into the PV or onto the merozoite surface are processed during traffic or after secretion, thus exposing one or multiple N-termini that might be potential DPAP3 substrates. For example, DPAP3 might trim the N-terminus of its substrates after signal peptide removal, thus potentially affecting their localization or stability. Indeed, the N-terminal sequence downstream of the signal peptide has been shown to be important for proper localization of rhoptry and micronemal proteins both in P. falciparum [44,45] and T. gondii [46,47]. Interestingly, two recent studies have shown that the aspartyl protease plasmepsin IX (PMIX) is essential for RBC invasion and that it acts as a maturase of rhoptry proteins[48,49]. In both studies, PMIX was shown to localize at the apical end of merozoites either within or, similarly to DPAP3, in close proximity to the rhoptries. This similar localization raises the possibility that DPAP3 might trim the N-terminus of rhoptry proteins after being processed by PMIX. Finally, most surface proteins that are involved in RBC invasion (EBAs, Rhs, MSPs, AMA1, RON2, etc) expose one or more extracellular N-termini[7,50], and the N-terminal domains of some of these proteins have been shown to be important in mediating protein-protein interactions and biological function. Interestingly, in T. gondii extracellular trimming of the N-terminus of surface proteins has been well documented [51,52]. T. gondii also expresses DPAPs in secretory organelles[53], and a recent proteomic study has shown evidence that the N-terminus of certain secreted proteins, such as TgSUB1 and TgMIC11, is trimmed through the removal of dipeptides[54]. It is therefore possible that DPAP3-mediated trimming of the N-terminus of certain parasite surface proteins might modulate their affinity towards host cell receptors.
Our biochemical studies have shown that DPAP3 is an unusual papain-fold protease since removal of its prodomain is not required for activation. Most proteases prodomains act as endogenous inhibitors and internal chaperones. Although the prodomain of DPAP3 might be required for proper folding of this large protein (941aa), we cannot discard the possibility that it might have other biochemical functions, such as recognizing substrates or binding to cofactors that modulate DPAP3 activity. Interestingly, the pro-form of P. falciparum DPAP1 has been shown to localize to the PV in mature schizonts[30], and processing of recombinant DPAP1 from its ‘zymogen’ form to is fully processed form only increases its activity 2–3 fold [32], suggesting that similarly to DPAP3, the ‘zymogen’ form of DPAP1 is active and might be able to process the N-terminus of PV or merozoite surface proteins. Therefore, both DPAP1 and DPAP3 are secreted before egress, raising the possibility that they might play redundant or complementary functions during RBC invasion.
Synthesis of inhibitors and production of DPAP3 antibodies are described in S1 Text. Primers and antibodies used in this study are listed in S2 Table and S3 Table, respectively.
Three synthetic genes codon-optimized for insect cells were synthesized by Genewiz and cloned into the puc57 vector backbone: puc57-rDPAP3-Nt, puc57-rDPAP3-Ct-wt, and puc-rDPAP3-Ct-mut. The first one codes for the N-terminal portion of DPAP3 (Met1-Asp469) and the other two for the C-terminal portion (Lys455-Stop941) containing the catalytic domain of DPAP3 and harboring either the catalytic cysteine Cys504 or the C504S inactivating mutation. All synthetic sequences contained a C-terminal His6-tag and were flanked with the BamHI and HindIII restriction sites at the 5’ and 3’ end, respectively. A ClaI restriction site is present in the 45 bp overlapping sequence (Lys455-Asp469) between puc57-rDPAP3-Nt and puc57-rDPAP3-Ct-wt/mut. Digestion of these plasmids with BamHI, HindIII, and ClaI, followed by ligation of the N- and C-terminal products into the puc57 backbone yielded puc57-rDPAP3-WT and puc57-rDPAP3-MUT. The BamHI and HindIII restriction sites were used to clone full length dpap3 into the pFastBacHT vector (Thermo Fisher Scientific) for expression of WT or MUT DPAP3 (pFB-rDPAP3wt and pFB-rDPAP3mut) in insect cells.
rDPAP3 was expressed in Sf9 insect cells using the baculovirus system. E. coli DH10Bac cells (Invitrogen) were transformed with pFB-rDPAP3wt and pFB-rDPAP3mut following the manufacturer recommendation. Baculovirus DNA was extracted using the BACMAX DNA purification kit (Epicentre) and transfected into a 5 mL culture of Sf9 cells (2x106 cells/mL) using Cellfectin (Thermofisher). After 3 days, the culture supernatant containing baculovirus particles was collected (P1 stock). To increase the viral load of our stocks, 1 mL of culture supernatant was serially passage twice into 25 mL of Sf9 cultures at 2x106 cell/mL for 3 days to obtain P2 and P3 viral stocks, which were stored at 4°C or frozen in liquid N2 in the presence of 10% glycerol. Insect cells were grown in SF-900-II serum free medium (Gibco) at 27°C under shaking conditions.
For rDPAP3 expression, Sf9 cells at 2x106 cells/mL were infected with 0.4 mL of P3 viral stock per liter of culture. The supernatant containing rDPAP3 was collected 72 h after infection, supplemented with protease inhibitors (1 mM PMSF, 0.5 mM EDTA, 1 μM pepstatin, 1 μM bestatin, and 10 μM E64), and its pH adjusted by adding 50 mM TrisHCl from a 1 M solution at pH 8.2; 10% glycerol was added before storage at -80°C. Note that despite being a general Cys protease inhibitor, E64 does not inhibit DPAP3.
A three steps purification consisting of ion exchange, affinity, and size exclusion chromatography was used to purify rDPAP3. First, culture supernatant was passed 3 times through 0.05 volumes of Q-sepharose (GE Healthcare) pre-equilibrated with Buffer A (50 mM Tris pH 8.2 containing the above-mentioned protease inhibitors). The resin was washed with 5 volumes of Buffer A and 2.5 volumes of 50 mM NaCl in Buffer A, and rDPAP3 eluted with 400 mM NaCl in Buffer A. Fractions containing rDPAP3 were pooled, diluted 1:1 into Buffer B (100 mM sodium acetate, 100 mM NaCl, pH6, and the protease inhibitors mentioned above), and passed through 0.05 volumes of Ni-NTA resin (Qiagen). The resin was washed with 10 volumes of Buffer B, and rDPAP3 eluted by lowering the pH of Buffer B to 5. Fractions containing rDPAP3 were pooled, concentrated using a Centricon Plus-70 Centrifugal Filter Unit (Millipore), loaded on a Superdex 200 10/300 GL size exclusion column, and run on an AKTA FPLC with Buffer A. Fractions containing rDPAP3 were pooled, concentrated, and stored at -80°C in the presence of 10% glycerol.
DPAP3 activity was measured either using the DPAP fluorogenic substrates VR-ACC[32] or FR-βNA (Sigma), or with the FY01 activity-based probe. When using FY01, samples (intact parasites, parasite lysates, insect cell supernatant, or rDPAP3 purification fractions) were labelled with 1 μM FY01 for 1 h, boiled in loading buffer, run on a SDS-PAGE gel, and the fluorescence signal measured on a PharosFX (Biorad) flatbed fluorescence scanner [18]. To determine the potency and specificity of inhibitors against DPAPs and the falcipains, parasite lysates diluted in acetate buffer (50 mM sodium acetate, 5 mM MgCl2, 5 mM DTT, pH 5.5) were pretreated for 30 min with a dose response of inhibitor followed by FY01 labelling.
When using VR-ACC (10 μM) or FR-βNA (100 μM), substrate turnover was measured on a M5e Spectramax plate reader (λBex = 355 nm/λem = 460 nm or λex = 315 nm/λem = 430 nm, respectively) in 50 mM sodium acetate, 20 mM NaCl, 5 mM DTT, and 5 mM MgCl2, pH5.5. The pH dependence of rDPAP3 was determined at 10 μM VR-ACC using a 20 mM sodium acetate, 20 mM MES and 40 mM TRIS triple buffer system containing 5 mM DTT, 0.1% CHAPS, 20 mM NaCl, and 5 mM MgCl2.
All constructs designed to integrate into the dpap3 locus by single-crossover recombination (tagged or cKO lines) contained either a 1065bp C-terminal homology region fused to GFP, or a 1210bp homology region upstream of the catalytic Cys504 (Asp39-Glu392) fused to a recodonized C-terminal region (Lys393-Stop941) tagged with mCherry or triple HA tag (HA3). The recodonized region contained either WT Cys504, or the C504S mutation. The construct designed to generate the DPAP3-HA, DPAP3-mCh, and DPAP3-GFP tagged lines we obtained as shown in S10A and S10B Fig. Briefly, the dpap3 C-terminal homology region was inserted into the pPM2GT plasmid to generate pPM2GT-DPAP3Ct-GFP. The N-terminal region of dpap3 in the pFB-rDPAP3-wt/mut vector was replaced with the homology region to generate pFB-chDPAP3-wt/mut. The dpap3 ORF was then introduced into the pHH1-SERA5ΔCt-HA—obtained after removal of the C-terminal part of SERA5 in the pHH1-SERA5-loxP-DS_PbDT3’[33]—resulting in plasmids pHH1-chDPAP3-wt/mut-HA, which harbor a C-terminal HA3 tag and a loxP site downstream of the Pb3’ UTR. The HA3 sequence of pHH1-SERA5ΔCt-HA was replaced with mCherry—amplified from the pREST-B plasmid[55]—to generate the pHH1-SERA5ΔCt-mCh, and the ORF of WT or MUT DPAP3 introduced into this plasmid to generate the pHH1-chDPAP3-wt/mut-mCh constructs.
To conditionally truncate dpap3 we used two different approaches. The first approach introduced a loxP site within the ORF of dpap3, in an Asn-rich region (Asn414-Asn444) upstream of the catalytic domain, resulting in replacement of Asn430-Asp434 with a loxP coding peptide (ITSYSIHYTKLFTG). To make the pHH1-chDPAP3_loxP-mCh construct (S10C Fig), the N- and C-terminal portions of DPAP3 were amplified from the pHH1-chDPAP3-wt-mCh, which contains a 3’UTR loxP site, and ligated into the plasmid backbone. A loxP site was introduced in the backward primer used to amplified the N-terminal region. The second approach inserted a loxPint[36] between the homology and recodonized regions. A synthetic 1600bp sequence (GeneWiz) containing the loxPint fragment flanked by targeting sequences was introduced into construct pHH1-chDPAP3-wt-mCh to generate the pHH1-chDPAP3_loxPint-mCh plasmid.
Finally, for episomal complementation of the cKO lines, plasmids pHH1-chDPAP3-wt/mut-HA (S10A Fig) were modified in order to express full-length WT or MUT DPAP3 under the control of the dpap3 or ama1 promoters. Firstly, to select for parasites containing the complementation plasmids after transfection, the puromycin N-acetyltransferase (pac) gene, which confers resistance to puromycin, was amplified from mPAC-TK (a kind gift of Alex Maier) and subsequently ligated into pHH1-chDPAP3-wt/mut-HA plasmids. The homology region of this plasmid was replaced with a recodonized N-terminal dpap3 amplified from puc57-rDPAP3-Nt, resulting in pHH1-rDPAP3-wt/mut-HA plasmids (S10D Fig). The dpap3 and ama1 promoters (970 and 1456bp upstream of the start codon, respectively) were amplified from genomic DNA and ligated into these plasmids to generate the pHH1-dpap3-rDPAP3-wt/mut-HA, and pHH1-ama1-rDPAP3-wt/mut-HA complementation constructs.
All final construct sequences were verified by nucleotide sequencing on both strands. Primers for PCR amplification and restriction sites used to generate these plasmids are listed in S2 Table and indicated in S10 Fig, respectively.
All constructs were transfected at schizont stage using a 4D-Nucleofector electroporator (Lonza) as previously described[12]. DPAP3-tagged and DPAP3-cKO lines were obtained through multiple on and off drug selection cycles with WR99210 (Jacobus Pharmaceuticals) and cloned by limited dilution as previously described[56]. To generate the DPAP3-GFP, DPAP3-HA and DPAP3-mCh lines, pPM2GT-DPAP3Ct-GFP, pHH1-chDPAP3-wt-HA and pHH1-chDPAP3-wt-mCh were transfected into P. falciparum 3D7 parasites. Plasmid pHH1-chDPAP3-mut-mCh was transfected multiple times into 3D7 in an attempt to swap the endogenous catalytic domain of DPAP3 with one containing the inactivating C504S mutation, but no integration was observed even after five drug cycles. The A1cKO and F3cKO & F8cKO lines were obtained after transfection of pHH1-chDPAP3_loxP-mCh and pHH1-chDPAP3_loxPint-mCh, respectively, into P. falciparum 1G5 parasites that endogenously expressed DiCre. Finally, the E7ctr line containing only the 3’UTR loxP site was generated by transfecting 1G5 parasites with pHH1-chDPAP3-wt-mCh. Complementation lines were obtained after transfection of A1cKO or F8cKO with pHH1-dpap3-rDPAP3-wt-HA, pHH1-dpap3-rDPAP3-mut-HA, pHH1-ama1-rDPAP3-wt-HA, or pHH1-ama1-rDPAP3-wt-HA and selection with WR99210 and puromycin. All parasite lines were maintained in RPMI 1640 medium with Albumax (Invitrogen) containing WR99210 (plus puromycin for the complementation lines) and synchronized using standard procedures[57].
To conditionally truncate the catalytic domain of DPAP3, tightly synchronized ring-stage parasites were treated for 3–4 h with 100 nM RAP (Sigma) or DMSO at 37°C, washed with RPMI, and returned to culture. Schizonts purified at the end of the cycle were used to determine the excision efficiency at the DNA (PCR) or protein (IFA, WB) level.
To determine exactly when DPAP3 is secreted in relation to PVM and RBCM breakdown, 20 μL of purified schizonts in 4 mL of RPMI were arrested with C2 or E64, or were allowed to egress for 1 h in the presence of 1 μM FY01. These cultures were centrifuged at 3000 rpm to separate intact schizonts from free merozoites and the culture supernatant. Merozoites were isolated from the culture supernatant by centrifugation (10 min at 13000 rpm), and proteins in the culture supernatant precipitated with 10 volumes of ice cold methanol and overnight incubation at -80°C. The schizont fractions were treated with 30 μL of 0.15% saponin in PBS to lyse the PVM and RBCM, and thus separate PV and host cytosolic components (saponin soluble fraction) from the parasite pellets (saponin insoluble fraction), which were washed once with PBS. Each fraction was then dissolved into PBS to a final volume of 50 μL, boiled for 10 min after adding 17 μL of 4X loading buffers. Equal volumes of each fraction (20 μL) were run in a SDS-PAGE gel under reducing conditions for WB and FY01 labelling analysis.
Thin films of P. falciparum cultures were air-dried, fixed in 4% (w/v) formaldehyde (PFA) for 20 min (Agar Scientific Ltd.), permeabilized for 10 min in 0.1% (w/v) Triton X100 and blocked overnight in 3% (w/v) bovine serum albumin (BSA) or 10% (w/v) goat serum (Invitrogen) in PBS. Slides were probed with monoclonal antibodies or polyclonal sera as described previously[58] (See S3 Table for antibodies used in this study), subsequently stained with Alexa488-, Alexa594-, Alexa647-labelled secondary antibodies (Molecular Probes) and DAPI (4,6-diamidino-2-phenylindole), and mounted in ProLong Gold Antifade (Molecular Probes). Images were collected using AxioVision 3.1 software on an Axioplan 2 Imaging system (Zeiss) using a Plan-APOCHROMAT 100x/1.4 oil immersion objective or LAS AF software on an SP5 confocal laser scanning microscope (Leica) using a HCX PL APO lamda blue 63x/1.4 oil immersion objective. Super-resolution microscopy was performed using a DeltaVision OMX 3D structured illumination (3D-SIM) microscope (Applied Precision). Images were analyzed with ImageJ (NIH), Adobe Photoshop CS4 (Adobe Systems) and Imaris x64 9.0.0 (Bitplane) software.
For IEM, mature schizonts from the DPAP3-GFP and 3D7 control lines were concentrated using a magnetic activated cell sorting (MACS) LD separation column (Miltenyi Biotec). Briefly, iRBCs were loaded onto an LD column attached to Midi MACS pre-equilibrated with media. The column was washed twice with media and schizonts eluted with media after detaching the column from the magnet. Parasites were then fixed in 4% paraformaldehyde/0.1% glutaraldehyde (Polysciences) in 100 mM PIPES and 0.5 mM MgCl2, pH 7.2, for 1 h at 4°C. Samples were embedded in 10% gelatine and infiltrated overnight with 2.3 M sucrose/20% polyvinyl pyrrolidone in PIPES/MgCl2 at 4°C. Samples were trimmed, frozen in liquid nitrogen, and sectioned with a Leica Ultracut UCT cryo-ultramicrotome (Leica Microsystems). Sections of 70 nm were blocked with 5% FBS and 5% NGS for 30 min and subsequently incubated with rabbit anti-GFP antibody 6556 (Abcam) at 1:750 overnight at 4°C. Colloidal gold conjugated anti rabbit (12 nm) IgG (Jack Imm Res Lab) was used as secondary antibody.
Plaque assays were performed as previously described[16]. Briefly, the 60 internal wells of a flat-bottom 96 well plate were filled with 200μL of DMSO- or RAP-treated parasite culture at 10 iRBCs/well and 0.75% hematocrit, incubated for 12–14 days at 37°C, and the number of microscopic plaques counted using an inverted microscope.
For all FACS-based assays, samples were fixed for 1 h at RT with 4% PFA and 0.02% glutaraldehyde, washed with PBS, and stored at 4°C. Samples were stained with SYBR Green (1:5000) or Hoechst (2 μg/mL), run on a FACScalibur or FortessaX20 flow cytometers (Becton-Dickinson Bioscience), and the data analyzed with CellQuest Pro or FlowJo.
For replication assays, cultures at 0.1% parasitemia (ring stage) and 2% hematocrit were grown for 4 cycles after DMSO or RAP treatment. Aliquots were fixed every 48 h at trophozoite stage, stained with SYBR Green, and parasitemia quantified by flow cytometry. To avoid parasite overgrowth, cultures were diluted 10-times whenever they reached 5% parasitemia. The cumulative percentage parasitemia (CP) over 4 cycles was fitted to an exponential growth model: CP = Pt0.MRN, where Pt0 is the initial parasitemia, MR the multiplication rate per cycle, and N the number of cycles after treatment.
To look at the effect of DPAP3 truncation on the full erythrocytic cycle, A1cKO parasites were synchronized within a 2 h window, treated with DMSO or RAP for 3 h, and put back in culture for 76 h. Sample were collected every 2–4 h, fixed, stained with Hoechst, and analyzed by flow cytometry. DNA levels were quantified as the median fluorescence signal of iRBC divided by the background signal for uninfected RBCs (uRBCs).
For standard invasion assays, schizonts purified from DMSO or RAP treated cultures were incubated with fresh RBCs for 8–14 h, fixed, and stained with Hoechst. The population of uRBC, rings and schizonts was quantified based on DNA content. The invasion rate was determined as the ratio between the final population of rings and the initial population of schizonts. Invasion time courses were performed by arresting purified schizonts with 1 μM C2 for 4 h, washing twice with warm media, and culturing with fresh RBCs under shaking conditions. Samples were collected at different time points, fixed, and split into two aliquots: One was stained with Hoechst and WGA-Alexa647 and run on a flow cytometer at a high forward scattering voltage in order to detect free merozoites. The populations of uRBCs, free merozoites, rings, and schizonts were quantified with FlowJo. The other aliquot was blocked with 3% BSA in PBS overnight at 4°C, and stained without permeabilization with the MSP1 monoclonal antibody 89.1 (1:100), and subsequently with anti-mouse Alexa488 (1:3000). The population of uRBCs, schizonts, rings, and merozoites attached to the RBC surface were quantified using FlowJo.
Time lapse video microscopy of egress was performed as previously described[9]. Briefly, tightly synchronized schizonts were percoll-enriched and arrested with 1 μM C2 for 4 h. After C2 wash out, DIC and mCherry images were collected every 5 and 25 s, respectively, for 30 min using a Nikon Eclipse Ni-E wide field microscope fitted with a Hamamatsu C11440 digital camera and a Nikon N Plan Apo λ 100x/1.45NA oil immersion objective. For each experiment, videos of the RAP- and DMSO-treated parasites were taken alternately to ensure that possible differences in the rate of egress were not a result of variation in the maturity of the parasite populations. The images were then annotated using Axiovision 3.1 software and exported as AVI movie or TIFF files. Individual egress events were annotated by detailed visual analysis of the movies, and the delay to the time of egress was recorded for each schizont for subsequent statistical analysis. Mean fluorescence intensity values of individual mCherry-expressing schizonts right before and after PVM breakdown were determined from exported raw image files (TIFF format) as described previously[9] and using the elliptical selection tool and ‘Histogram’ options of ImageJ/Fiji V1.0. DPAP3KO parasites were analyzed to determine the residual background fluorescence derived from the hemozoin.
Saponin pellets of parasites from different erythrocytic stages as well as samples from culture supernatant harvested during egress and invasion were syringe filtered (Minisart, 0,2 μm, Sartorius), boiled in SDS-PAGE loading buffer under reducing conditions, run on a SDS-PAGE gel, and transferred to Hybond-C extra nitrocellulose membranes (GE Healthcare). The membranes were blocked with 5% (w/v) nonfat milk in PBS, probed with monoclonal or polyclonal antibodies (see S3 Table for antibodies used in this study), and followed by application of horseradish peroxidase-conjugated secondary antibodies (Pierce). The signal was detected using SuperSignal West Pico chemiluminescent substrate (Thermo Scientific) and a ChemiDoc MP imager (BioRad).
PCR was performed using GoTag (Promega), Advantage 2 (Clontech), or Q5 High-Fidelity (NEB) polymerases. Diagnostic PCR to detect integration of targeting constructs was performed using extracted genomic DNA as template. Primer pairs specific for detection of integration, namely P21 and P22 for integration of pHH1-chDPAP3-mCh and pHH1-chDPAP3-HA, and II-inte_F and II-wt_R for integration of pPM2GT-DPAP3Ct-GFP, were designed such that the forward primer hybridized in a genomic region upstream of the plasmid homology region, and the second in a region unique to the introduced plasmid. Primer pairs P21 and P23 were designed to detect presence of the unmodified dpap3 locus. To assess whether dpap3-mCh had been excised after RAP treatment, diagnostic PCR was performed using extracted genomic DNA as template. Primers P24 and M13 were used to detect non-excision dpap3 at the genomic locus and hybridize upstream and downstream of the second loxP site, respectively. Primers P25 and SP6 were used to detect presence of excision and hybridize upstream and downstream of the first and second loxP sites, respectively.
|
10.1371/journal.pcbi.1005233 | Dopamine Neurons Change the Type of Excitability in Response to Stimuli | The dynamics of neuronal excitability determine the neuron’s response to stimuli, its synchronization and resonance properties and, ultimately, the computations it performs in the brain. We investigated the dynamical mechanisms underlying the excitability type of dopamine (DA) neurons, using a conductance-based biophysical model, and its regulation by intrinsic and synaptic currents. Calibrating the model to reproduce low frequency tonic firing results in N-methyl-D-aspartate (NMDA) excitation balanced by γ-Aminobutyric acid (GABA)-mediated inhibition and leads to type I excitable behavior characterized by a continuous decrease in firing frequency in response to hyperpolarizing currents. Furthermore, we analyzed how excitability type of the DA neuron model is influenced by changes in the intrinsic current composition. A subthreshold sodium current is necessary for a continuous frequency decrease during application of a negative current, and the low-frequency “balanced” state during simultaneous activation of NMDA and GABA receptors. Blocking this current switches the neuron to type II characterized by the abrupt onset of repetitive firing. Enhancing the anomalous rectifier Ih current also switches the excitability to type II. Key characteristics of synaptic conductances that may be observed in vivo also change the type of excitability: a depolarized γ-Aminobutyric acid receptor (GABAR) reversal potential or co-activation of α-amino-3-hydroxy-5-methyl-4-isoxazolepropionic acid receptors (AMPARs) leads to an abrupt frequency drop to zero, which is typical for type II excitability. Coactivation of N-methyl-D-aspartate receptors (NMDARs) together with AMPARs and GABARs shifts the type I/II boundary toward more hyperpolarized GABAR reversal potentials. To better understand how altering each of the aforementioned currents leads to changes in excitability profile of DA neuron, we provide a thorough dynamical analysis. Collectively, these results imply that type I excitability in dopamine neurons might be important for low firing rates and fine-tuning basal dopamine levels, while switching excitability to type II during NMDAR and AMPAR activation may facilitate a transient increase in dopamine concentration, as type II neurons are more amenable to synchronization by mutual excitation.
| Dopamine neurons play a central role in guiding motivated behaviors. However, complete understanding of computations these neurons perform to encode rewarding and salient stimuli is still forthcoming. Network connectivity influences neural responses to stimuli, but so do intrinsic excitability properties of individual neurons, as such properties define synchronization qualities and neural coding strategy. We investigated the excitability type of the DA neuron and found that, depending on the synaptic and intrinsic current composition, DA neurons can switch from type I to type II excitability. In short, without synaptic inputs or under balanced excitatory and inhibitory inputs DA neurons exhibits type I excitability, while excitatory AMPAR inputs can switch the neuron to type II. Type I neurons are best suited for coding the stimulus intensity due to their ability to decrease the firing rate smoothly. Type I excitability might be important for achieving low a basal DA concentration necessary for normal brain functioning. Switching to type II excitability further enables robust transient DA release of heterogeneous DA neuron population in response to correlated inputs, partially due to evoked population synchrony.
| Midbrain dopamine (DA) neurons predominantly fire in a low frequency, metronomic manner (i.e. tonic) and display occasional, yet functionally important, high frequency, burst-like episodes [1,2]. While regular tonic firing is observed in isolated preparations (i.e. slices), tonic firing pattern in vivo is somewhat more variable due to active synaptic inputs [3,4]. Such tonic activity is important for maintaining a constant basal level of dopamine in projection areas. Accordingly, abnormal basal DA levels are linked to psychiatric disorders from depression to schizophrenia [5,6]. While the maintenance of basal DA levels seem to be critical for normal brain function, a consistent picture has not yet emerged regarding how changes in firing patterns of the DA neuron facilitates this important biological function.
Background activity of the DA neuron appears to rely on the intrinsic pacemaking mechanism that generates tonic firing. The current composition producing low-frequency pacemaking in DA neurons is of vibrant debate among researchers in the field. A number of experimental [7–19] studies suggests that the maintenance of tonic firing in at least a subpopulation of DA neurons relies on the interactions of the voltage gated calcium (Ca2+) and SK-type Ca2+-dependent potassium (K+) currents Slow pacemaking in our model relies on a subthreshold Ca2+-K+ oscillatory mechanism, similar to a number of well-established models [4,20–24]. Interaction between Ca2+ and Ca2+-dependent K+ currents periodically brings the neuron to the spike threshold and generates a metronomic firing pattern. In our model, spike-producing currents (fast sodium and the delayed rectifier potassium) play a mostly subordinate role in this dynamic, adding a spike on top of the oscillations without significant changes to the period or shape of voltage and calcium oscillations, as in the study by Wilson and Callaway 2000 [4]. A number of studies suggest an additional Ca2+-independent oscillatory mechanism [25–27]. In particular, they emphasize the contribution of sodium currents to pacemaking. We review the literature on the mechanisms of DA neuron pacemaking in the discussion section in more detail. The specific composition of currents contributing to oscillations determines the response of the DA neurons to stimuli, their synchronization properties and, ultimately, the computations they perform. In this paper, we use recent experiments to calibrate the dynamical properties of the DA neuron and determine its excitability type.
A standard method to classify neuronal excitability is via characterizing the frequency-to-input relationship, or F-I curve. Two major types of excitability can be determined based on how the onset of tonic firing occurs as the applied current increases and the neuron is released from quiescence at the hyperpolarized rest state [28]. A type I-excitable neuron can fire at an arbitrary low frequency near the onset of firing, whereas a type II neuron shows a discontinuous jump to a minimal frequency above a certain current threshold and fires only in a limited range of frequencies [21]. Generally, the onset of repetitive firing occurs through one of two mechanisms: 1) a saddle-node bifurcation on invariant circle or SNIC (type I excitability) or 2) an Andronov-Hopf bifurcation (type II excitability) [29]. Type II neurons, such as fast-spiking inhibitory interneurons in the cortex, display precise spike timing even in the presence of noise and are therefore suitable for the implementation of spike time coding [30,31]. A type I neuron, such as a weakly adapting cortical pyramidal neuron, was shown to relay the stimulus rate by modulating its own frequency, and, therefore, displayed rate coding [31]. Further, type II neurons display resonance and controlled synchronization in networks [31–33]. For neurons that are tonically active without any injected current, such as DA neurons, the transition to the non-spiking rest state occurs when a sufficiently strong hyperpolarizing current is injected. For these neurons, the excitability type would be defined by the transition from tonically firing to quiescent/excitable: again type I would show a smooth frequency decrease to zero, while type II should show an abrupt transition to quiescence. At this point, there is no direct evidence defining to what type of excitability DA neurons belong, and how the different intrinsic conductances and the different synaptic inputs influence their type. Determining the type of excitability will allow us to predict the behavior of the DA neuron during application/blockade of different currents and better understand computations it performs in different input conditions (e.g. rate coding vs. resonance at a particular input frequency).
A number of experimental studies provide indirect evidence of excitability type of the neuron in control conditions and during activation of synaptic inputs. It has been shown that the firing rate of DA neurons increases linearly in response to a ramping depolarizing current until it goes into depolarization block (e.g. [3,34]). Further, injection of a tonic hyperpolarizing current to the regularly firing DA neuron in vitro increases its interspike intervals [7]. The firing properties of the neuron in response to a combination of tonic inhibitory and excitatory synaptic conductances were investigated by Lobb and colleagues [35,36]. Using the dynamic clamp technique, they injected inhibitory γ-Aminobutyric acid (GABA) and excitatory N-methyl-D-aspartate (NMDA) receptor conductances in SNc DA neurons. Injection of tonic GABAR conductance decreased the firing rate of the neuron several-fold. Furthermore, the neuron fired at low frequencies when NMDAR and GABAR conductances balanced each other. Thus, NMDAR activation, which strongly increases the firing frequency [37–40], can be effectively compensated by GABAR activation. Such compensation would be impossible if the inhibition produced an abrupt transition to quiescence and the neuron jumped from a high frequency to zero. The experimentally observed compensation suggests, again, a smooth frequency decrease upon GABAR activation rather than an abrupt transition to the resting state at hyperpolarized potentials. Together, these data resemble the tonic firing/quiescence transition in type I neurons with two distinctions. First, the transition parameter is not an injected current, but an ohmic GABAR conductance. In experiments, a conductance has already been used instead of an injected current to determine the neuronal excitability [41]. Second, the co-activation of the NMDA receptor introduces an additional parameter (its maximal conductance). Both of these extend the definition of excitability into the space of synaptic conductances. Formally, the excitability type is an intrinsic property of a neuron, yet viewing synaptic inputs as changing excitability of a neuron is a powerful concept used to understand neuron dynamics in vivo [31,42]. We used the experiments described above to parameterize a model of the DA neuron, determine its type of excitability, and determine how intrinsic and synaptic currents shape the excitability type and, therefore, the computational properties of the neuron.
These experiments suggest that the DA neuron exhibits type I excitability in isolation from synaptic inputs and under the balanced influence of excitatory and inhibitory synaptic conductances. However, the excitability type has been shown to vary depending on the intrinsic currents and network connectivity [42,43]. For example, modeling results suggest that changes in the intrinsic currents, e.g. L-type Ca2+ current, can switch the excitability type of the DA neuron [44,45]. Here we address the variability in the excitability type under different conditions by studying the contribution of intrinsic and synaptic currents to regulation of the low-frequency DA neuron firing.
We investigated the behavior of a simulated dopamine neuron in response to irregular asynchronous GABA and glutamate (Glu) inputs to mimic temporal structure of neural firing in in vivo conditions. The Glu input was produced by Poisson distributed spike trains and GABA inputs was explicitly modeled as activity of a population of GABA neurons (detailed description of the inputs and equations are given in the methods section). We quantified changes in the firing rate and the regularity of DA neuron firing in response to synaptic inputs of different strengths (Fig 1A1 and 1A2). We identified a parameter region where the excitatory and inhibitory inputs balance to produce low frequency DA neuron firing at rates similar to background firing (Fig 1A1, between the black lines). This happens because asynchronous GABA and Glu inputs (see rasters in Fig 1B1 and 1B2) activate GABARs and NMDARs nearly tonically (Fig 1B3 and 1B4) and provide quasi-constant levels of inhibition and excitation to the DA neuron respectively. Under the influence of these two inputs, the DA neuron fires similarly to the in vitro-like conditions (tonic inputs), but with less regularity, which is typical of the background firing in vivo. An example voltage trace of the DA neuron in response to synaptic inputs formed by asynchronous Glu and GABA populations is shown in Fig 1B5. Fluctuations in the firing of neural populations innervating the DA neuron can produce irregular spiking as observed in in vivo experiments.
Considering that asynchronous inputs produce nearly constant receptor activation (see Fig 1B3 and 1B4), for further analysis we substituted these Glu and GABA inputs by tonic currents. Moreover, tonic synaptic currents mimic long-lasting injection of the conductances in dynamic clamp experiments [46,35,36], or iontophoresis of the agonists [39,40], or bath application of the agonists. Transition from asynchronous inputs to tonic currents is described in the methods section.
Our next goal was to reproduce the experimentally-observed compensatory influence of tonic NMDAR and GABAR conductances [35]. Using the dynamic clamp technique, it was shown in vitro that a balanced injection of GABAR and NMDAR conductances leads to DA neurons firing at frequencies comparable with background frequencies (1–5 Hz). Removal of inhibition in such conditions evokes a classical disinhibition burst (the disinhibition model of burst generation is well known and described in e.g. [35,47–49]). Fig 2A reproduces the voltage traces obtained in the experiments by Lobb et al. 2010 [35]. In this example, the simulated DA neuron is tonically active at 1.5 Hz during tonic co-activation of NMDA and GABA receptors (gNMDA = 16.9 m S/cm2, gGABA = 5mS/cm2). Removal of the GABAR conductance produces an episode of high-frequency firing (Fig 2A). Removal of the NMDAR conductance produces a pause in firing (Fig 2A).
We explored the range of NMDAR and GABAR conductances that produce tonic firing in the DA neuron model (Fig 2B). Compensation of NMDAR and GABAR activation can be readily achieved near the upper boundary of the firing region (Fig 2B, blue). When both receptors are activated, low frequency tonic activity is observed (Fig 2A). The dot labeled as A on the heat plot indicates conductances taken for this simulation. As in the experiments [35], the balanced region is stretched linearly on the conductance plane with NMDA/GABA slope around 3.4. Moving to the left on the diagram corresponds to deactivation of the NMDAR current and blocks DA neuron firing due to the remaining GABAR activation (Fig 2A). A pause may also be produced by stronger activation of the GABAR (Fig 2C). Conversely, moving down on the diagram corresponds to deactivation of the GABAR and evokes high-frequency firing (Fig 2A). The firing frequency also increases by moving from the upper boundary of the firing region to the right (increasing NMDAR conductance; Fig 2D). These two directions correspond to two ways of eliciting a DA neuron burst: strengthening NMDA excitation or removing inhibition, respectively.
Excessive tonic NMDAR activation leads to a depolarization block, as shown in Fig 2B and 2E at high NMDA and low GABA receptor conductances. Interestingly, application of a tonic GABAR conductance in combination with an excessive NMDAR conductance may rescue high-frequency firing in the model (Fig 2E). Thus, the compensatory influence of GABAR activation removes depolarization block induced by an excessive NMDAR activation and restores the intrinsic oscillatory mechanism required for tonic firing.
The smooth frequency decrease to zero as the neuron transition to quiescence when GABAR conductance increases suggests type I excitability for the DA neuron both in in vivo and in vitro like cases (Figs 1A1 and 2B). However, excitability is classically defined by the structure of the transition between spiking and hyperpolarized rest state induced by an injected current, as opposed to a synaptic conductance. We show that the DA neuron exhibits type I excitability by standard definition with a continuous F-I curve and place it in Supporting Information (S1 Fig) as this case has less physiological significance than the influence of synaptic currents. We further investigate the influence of intrinsic and synaptic currents on the excitability type of the DA neuron.
In vivo, the type of excitability may change due to tonic synaptic inputs [42], and next we explore how this change occurs in the DA neurons. AMPA receptor may co-activate together with NMDA and GABA receptors in vivo. By contrast to NMDAR, conductance of which is voltage-dependent, AMPA and GABA receptor currents are purely ohmic. Their combination is also an ohmic current:
IAMPA+IGABA=Ieff=geff(Eeff−v),
Where geff = gAMPA + gGABA is a combined synaptic conductance, and
Eeff=gGABAEGABA+gAMPAEAMPAgGABA+gAMPA|EAMPA=0=gGABAgGABA+gAMPAEGABA
is a synaptic reversal potential. Fig 8 shows the frequency distribution and the type of bifurcation at the transition to the rest state on the plane of these two parameters: conductance and the reversal potential of the ohmic synaptic current. For instance, if the AMPA receptor is blocked, the reversal potential coincides with the GABAR reversal potential, which is in the range of -90 mV to -70 mV [55]. In this range, an increase in the conductance Eeff leads to a gradual decrease in the frequency to zero and a transition to the rest state via a SNIC bifurcation. By contrast, at higher reversal potentials, the frequency drops to zero abruptly and the transition corresponds to an Andronov-Hopf bifurcation. This suggests a transition to type II excitability for the DA neuron. Thus, elevated GABAR reversal potential or tonic activation of AMPAR leads to a switch in the excitability to type II.
AMPAR-mediated input induces depolarization block in the DA neurons and firing cannot be restored. This is consistent with the previous modeling studies showing that NMDA can elicit bursting [56–58] or burst envelope [59], while AMPA abolishes high frequency firing. The dynamical explanation is that AMPAR activation shifts the minimum of the voltage nullcline across the Ca2+ nullcline, so that for high AMPAR conductance values (as well as positive applied currents), voltage oscillations decrease in amplitude and depolarization block occurs (S3B Fig). Thus, DA neuron firing does not exceed the frequency of ~10 Hz when driven with AMPAR activation, similarly to the experimental results (see e.g. [37,38,40]), Application of GABAR-mediated input shifts the voltage nullcline even further to the right and makes the stable equilibrium more robust. Therefore, the region of parameters for which spiking is produced is much smaller for combined application of AMPA and GABA than for combined NMDA and GABA activation (compare Figs 2B and 9A). Therefore, the prediction of our model is that a disinhibition burst can be supported by tonic background activation of NMDA but not AMPA receptors. Please note that nullclines can be produced only for the model without the fast sodium current (see S3 Fig for the illustration of depolarization block in the model with spike-producing currents).
Together with AMPA and GABA receptors, the NMDAR may be also co-activated, since glutamate binds to both AMPA and NMDA receptors. To make the analysis of the excitability type possible in the parameter space of all three synaptic currents, we further perform 2-dimensional bifurcation analysis. The point marked Bogdanov-Takens bifurcation in Fig 8 is a good predictor of the type of excitability at the boundary between spiking and the rest state. Mathematically, it is defined as a junction of the SNIC bifurcation and the Andronov-Hopf bifurcation, as it appears in the figure. In Fig 10, we plot this point as a function of the NMDA receptor conductance. As in the previous figure, the transition to quiescence occurs as the combined conductance of the ohmic synapses geff grows. The information about the specific value of the conductance is omitted in Fig 10 because the transition occurs in a dimension orthogonal to the plane of the figure. For example, the transition in Fig 8 is represented by one line at gNMDA = 0. The diagram in Fig 10 shows that the separation between the types of excitability shifts to lower values of the combined reversal potential for the AMPAR and GABAR currents as the NMDAR conductance increases. However, this shift quickly saturates and is restricted to the range of GABAR reversal potentials. Thus, similar to Ih, NMDAR activation may switch the type of excitability of the DA neuron from type I to type II in a certain window of other synaptic currents received by the neuron.
To illustrate the importance of changing DA neuron type of excitability, we simulated heterogeneous populations of DA neurons under two conditions: 1) in control (in the absence of the tonic synaptic inputs), and 2) during the tonic influence of AMPA and GABA inputs. The DA population is electrophysiologically heterogeneous [60–62], and its uncoordinated activity produces a homogeneous low-level DA concentration. In order to have similar firing rates and basal DA levels in both cases, we balanced the increase in firing rate produced by the application of AMPAR conductance with GABAR conductance (note that GABA can balance AMPA only for a very limited range of values). In both cases, DA neurons received correlated fluctuating NMDA inputs (Fig 11A, see methods for the detailed description of the inputs). Our simulations show that the population of DA neurons that receive the background synaptic tone produces higher DA levels in response to bursty correlated NMDA input than a population without the synaptic tone (Fig 11D). As described above, DA neurons are type I excitable in the absence of synaptic tone, while AMPAR activation switches DA neuron excitability to type II. Thus, the transition from type I to type II excitability of the DA neurons is accompanied by higher dopamine release in response to a correlated synaptic input. The higher responsiveness is partially due to a greater synchronization of the DA neurons receiving the synaptic tone, as evident by the higher number of peaks in their summed activity in Fig 11C. Type II neurons display more robust synchrony when they receive a common input, even in the presence of independent noise [33,63]. Thus, in vivo background synaptic tone might be important not only for regulating basal DA levels, but also for the responsiveness of the DA neurons, so that they are more ready to produce coincident bursts in response to correlated synaptic inputs.
The type of excitability for a neuron determines the neurons’ responses to stimuli and their dynamics in a population (phase response curve, synchronization, resonators vs. integrators). The type of bifurcation determines the type of excitability: neural oscillations that arise via an Andronov-Hopf bifurcation have type II excitability, while those appearing via SNIC have type I excitability [52,65]. Based on the bifurcation analysis and frequency responses to hyperpolarizing inputs (negative injected current and GABAR conductance), we have shown that in control conditions, the simulated DA neuron is type I-excitable. It’s known that the type of excitability in vivo may be different from in vitro [42]. In vivo, DA neurons display irregular low-frequency activity occasionally interrupted by high-frequency bursts. This low-frequency regime may reflect the balance of inhibitory and excitatory inputs. We found that, in the most prominent low-frequency regime, the DA neuron is type I excitable, in either high or low synaptic conductance states.
The baseline level of dopamine is important for the normal function of the brain. The level is determined by the background activity of the DA neurons. This activity is intrinsically generated by the neuron and controlled by its synaptic inputs [66], reviewed in [67]. Thus, the capacity of the DA neuron to adjust its firing rate according to the inputs is vital. The graded response curve of a type I-excitable neuron, as opposed to an on-off response of a type II neuron, provides this capacity. Accordingly, at every level of excitation provided by NMDAR input, inhibitory GABA synaptic conductance can balance it and bring the frequency down to an arbitrary low value. A similar hyperpolarizing current activated by dopamine D2 receptors on the DA neuron has also been shown to slow down its firing rather than abruptly block it all together [68]. These are very important autoregulatory functions of the DA system that allow it to adjust basal DA levels in target areas.
Persistent sodium current is known to amplify subthreshold oscillations [69] and increases neural excitability of DA neurons by contributing to spontaneous depolarization in between the spikes [25]. Consistent with experimental observations, the subthreshold sodium current increases the firing rate of the DA neuron in the model. Additionally, we found that the current is necessary for achieving gradual frequency decrease upon application of hyperpolarizing current, thus, maintaining type I excitability of the DA neuron. The type of excitability is determined by the internal properties of the currents contributing to pacemaking in the DA neuron. L-type Ca2+ and SK-type Ca2+-dependent K+ currents are the core currents that traditionally constitute this mechanism [7–18] (but see the section on the mechanisms of DA neuron pacemaking). However, our model shows that the mechanism results in type II excitability, in which a hyperpolarizing current blocks the voltage oscillations without restoring a low frequency. Our explanation is that, without the contribution of further currents, the steep part of the Ca2+ nullcline is very close to the minimum of the voltage nullcline (Fig 4A), because they reflect the same event: opening of the Ca2+ channel. This positions the system close to the Andronov-Hopf bifurcation, which occurs whenever the minimum of the voltage nullcline moves across the Ca2+ nullcline. When the subthreshold sodium current is included into the mechanism, the minimum of the voltage nullcline reflects opening of this current, and it is shifted away from the steep part of the Ca2+ nullcline (Fig 6B). This shifts the system away from the Andronov-Hopf bifurcation. Now, a downward shift of the voltage nullcline following inhibitory inputs moves its minimum across the flat part of the Ca2+ nullcline and produces a SNIC bifurcation. In this transition, the firing frequency gradually reduces to zero, and this allows a balance between NMDAR and GABAR conductances and restores the background firing frequency.
The influence of NDMA and GABA receptor conductances on the DA neuron have been studied in several papers [23,22,70]. The compensatory influence of NDMA and GABA receptor activation on the firing frequency has been predicted in modeling studies by Komendantov et al. [23]. Lobb et al. [70] modified our previous model [24] to capture the balance of the inhibition and excitation and disinhibition bursts. In these models, a high maximal frequency (>10 Hz) can be achieved by tonic activation of the AMPAR or in response to depolarizing current injection, which contradicts experimental observations [7,34]. To the contrary, a number of experimental studies suggest that stimulation of NMDA receptors evokes a burst of high-frequency firing, whereas AMPA receptor activation evokes modest increases in firing [37–40] (but see [61,71]). This is an important distinction, which impacts the excitability of the neuron. Further, in Komendantov et al. [23] and Canavier & Landry [22], the NMDAR conductances were restricted to dendrites, whereas GABAR conductance was somatic. The mechanism of frequency rise during dendritic application of NMDA is different from the mechanism of response to somatic NMDAR stimulation [56,24]. Somatic NMDAR stimulation has been shown to elicit high-frequency firing in earlier experiments [46,40] and used to achieve the NMDA-GABA balance [35]. Here, we base a new model on our previous model [56] that presented a mechanism for somatically-induced high-frequency firing in a reconstructed morphology first and reduced it to a single compartment. In the current model, we have integrated the mechanism for high-frequency firing together with the balance of NMDAR and GABAR activation.
The mechanism of low frequency pacemaking in the DA neurons has been extensively studied. However, it is still a matter of on-going debate in the literature since different experimental results lead to contradicting conclusions, proposing that different currents are critical for the DA neuron spontaneous firing. In a number of experimental and modeling studies it has been shown that spontaneous tonic firing relies on the interactions between the voltage gated calcium (Ca2+) and SK-type Ca2+ -dependent potassium (K+) currents [7–19]. Wilson and Callaway [4] and later Chan et al. [19] showed that calcium-driven slow oscillatory potentials (SOPs) drive the spiking rate of the SNc DA neurons. Chan et al 2007 [19] also showed that dependence of pacemaking on Ca2+ oscillations changes with the age. Particularly, TTX blocks slow oscillations in juvenile neurons, but not in adult neurons, which is related to the change in density of Ca2+ channels with age. Ping and Shepard 1996 showed that the frequency of SOPs after the application of TTX is approximately the same as the frequency of spiking. In contrast, in a more recent study, Guzman et al. [26] demonstrated that, in a number of DA neurons, SOPs and spiking frequencies are weakly correlated, and that TTX inhibits spontaneous oscillatory potentials pointing to the importance of sodium currents for pacemaking. A number of other studies also suggest that sodium channels are highly involved in controlling spontaneous DA neuron frequency [26,25], especially in the VTA DA neurons [27].
The sources of the apparent discrepancy in the experimental results were investigated by Drion and colleagues [21]. Based on the combination of experimental and modeling approaches, authors suggested that calcium and sodium currents likely cooperate to produce pacemaking and prevailing mechanism depends on the density of the ion channels in the neuron. Further, authors showed that the lack of correlation between spikes and SOPs does not lead to a conclusion that generating mechanisms are different. The complementary role of the two mechanisms, notably if they co-exist in the same cell or represent pacemaking in distinct populations, is a matter of on-going research in the field.
For the sodium-based pacemaking mechanism, it is still unclear what hyperpolarizing current provides the long interspike interval when the SK current is not functional. This renders the Ca2+-independent oscillatory mechanism incomplete and is a matter of a future investigation. Control of the firing by the SK current has been shown to be stronger in the DA neurons positioned more laterally in the midbrain [15]. Thus, we focus on the subpopulation of DA neurons that are more abundant in the substantia nigra pars compacta (SNc) than the ventral tegmental area (VTA).
Changes in intrinsic currents can affect the excitability type and, thus, computational properties of the DA neuron. For instance, we observed that potentiation of Ih current promotes type II excitability of the simulated DA neuron (Fig 7). Further, we show that Ih current can induce bistability in the DA neuron, and the bistability region increases with the increase in Ih conductance (Fig 7B). This can affect the behavior of the neuron near the boundary between spiking/resting states, particularly, if the current reached the value necessary to induce spiking onset, a small perturbation in the current will not silence the neuron. This makes spiking more robust near the threshold. In addition to the contribution of Ih current to pacemaker activity, has been shown in DA neurons [72], as well as in the other neuronal types that Ih induces intrinsic subthreshold resonance [73–75]. Thus, augmentation of Ih current increases oscillatory behavior of the DA neurons, as well as their synchronization in response to excitatory pulses. However, low-frequency tonic firing could not be maintained at high conductances of Ih current, likely affecting background DA levels.
Further, the influence of tonic synaptic inputs can also change the transition to the rest state and, therefore, be described as altered excitability. Tonic activation of AMPA receptors or an elevated reversal potential of the GABAR conductance may make the low-frequency balanced state unreachable. The reason for that is a transition to type II excitability: firing is blocked at higher frequencies. A model prediction that follows from this result is that tonic AMPAR activation induces depolarization block and firing cannot be rescued by application of GABA. Our explanation is that shunting is so strong that opening of the subthreshold sodium current cannot sustain the voltage growth. In other words, these changes unfold the voltage nullcline and bring its minimum close to the steep part of the Ca2+ nullcline (Fig 9B). This primes the system for the Andronov-Hopf bifurcation responsible for type II excitability. Further, we found that NMDAR activation also biases the neuron towards type II excitability (Fig 10). Although the type may change as parameters shift away from the boundaries of the firing region [76], together, these results suggest that in high-frequency regimes the DA neuron displays type II excitability. This switch in excitability type may play a role in a transient increase in DA concentration in response to salient stimuli as it is easier to synchronize type II neurons by an excitatory input. Fig 11 supports this hypothesis by showing higher transient DA release produced by heterogeneous population of DA neurons receiving synaptic AMPA and GABA tones than in the absence of synaptic tone. Thus, correlated excitatory synaptic inputs are more likely to evoke robust coincidence DA release when DA neurons display type II excitability.
A growing body of literature links the type of excitability to neural coding [29,31,41,65,77–79]. For instance, Prescott et al. [80] suggested that type I neurons are best suited for coding stimulus intensity. Hence, the DA neuron is designed for encoding the intensity of the tonic depolarizing and hyperpolarizing inputs by its smooth frequency dependence. This further supports and augments a recently found unique computational role for the DA neuron: it performs subtraction of inhibitory and excitatory inputs [81]. The operation is optimal to calculate unpredicted value of an event but rarely observed and hard to implement in the brain. The first type of DA neuron excitability is necessary to quantitatively encode the level of input by the firing rate and perform the subtraction.
Several studies, for example Eshel et al. 2016 [82] and Tobler et al. 2005 [83], showed that activation of DA neurons gradually increases with the increase in the reward value. We attempted to reproduce this experimental result in our model. For the simulations, we assumed that the reward value is proportional to the strength of NMDA input coming to the DA neurons (gNMDA). GABA inputs do not seem to be a plausible candidate because Eshel et al. 2016 [81] showed that firing of the GABA neurons, as opposed to the DA neurons, does not vary consistently with the reward value. We calculated firing frequency dependence of the type I/ type II DA neurons on the input strength (Fig 12). Type II DA neurons were modeled by applying tonic AMPA along with NMDA conductance (additional excitation was compensated by increasing GABA conductance). Type II DA neurons are unable to encode low reward values as their firing abruptly drops to zero with the decrease in the input strength. In contrast, frequency dependence of type I DA neurons resembles the experimental curve of the DA neuron frequency dependence on the reward value shown in [82]. The difference can also be seen in the raster (Fig 12A and 12C) and frequency responses (Fig 12B and 12D) to a transient increase in the input strength, representing the reward value. First peak in the frequency response represents salience, and was simulated by applying constant level of NMDA for all the values, while second peak represents the reward value, and was simulated by scaling NMDA input accordingly. There is a gap in a frequency response to low reward values of type II DA neurons (Fig 12D). Thus, type I DA neurons best encode the value of an event, i.e. the difference between predicted and received reward.
On the other hand, spike timing of type I neurons in response to weak or noisy transient inputs is not reliable [84,85], while temporal precision of type II neurons is much higher. The ability of the DA neuron to switch excitability type from type I to type II under certain synaptic inputs might play a significant role in producing enhanced transient DA release, since it likely relies on the precise coordinated activity of the DA neurons. Multiple drugs of abuse, including EtOH (e.g. [86]) evoke transient increases in the DA concentration in nucleus accumbens.
Using our model, we show that EtOH shifts excitability of the DA neurons to type II and induces higher DA release. EtOH acts on multiple intrinsic and synaptic currents. Mainly, it enhances Ih current [87,88] and increases AMPA/NMDA ratio [89]. Moreover, it increases GABA release onto DA neurons [90]. Thus, we modeled EtOH action by increasing Ih, AMPA and GABA receptor currents. Ih and AMPAR currents switch DA neuron excitability to type II and, therefore, promote synchronization in the population of DA neurons in response to noisy excitatory inputs (Fig 13). This is one of the mechanisms that can produce higher DA transients. By switching DA neuron to type II, EtOH can increase the motivational potential of a stimulus because the same excitatory input produces enhanced DA signal under EtOH. In other words, neutral stimulus can become salient after EtOH exposure. Our modeling prediction regarding the excitability switch after EtOH could be tested in-vitro by studying how the shape of F-I curve changes after EtOH. In in-vivo conditions it could be checked by stimulating DA neurons before and after EtOH exposure with a chirp pattern signal in order to check for spiking resonance. To test whether DA neurons better synchronize after EtOH, a phase response curve (PRC) or a spike triggered average (STA) could be calculated. Presence of the negative PRC component and narrow STA indicates that neurons are more amenable to synchronization by common synaptic noise.
In conclusion, our results predict that DA neurons can exhibit traits of both integrators and resonators and these traits are modulated by intrinsic and synaptic conductances. Depending on the current constitution, DA neurons can perform rate coding by integrating slow variations in the inputs and adjust basal DA concentration or they can detect transient coherent changes in the inputs and synchronize for producing robust DA transients.
The biophysical model of the DA neuron is a conductance-based one-compartmental model modified from [91]
cmdvdt=gCa(v)(ECa−v)︷ICa+(gKCa([Ca2+])+gK(v)+g¯DRn4)(EK−v)︷IKCa+IK+IDR++(gsNa(v)+g¯Nam3h)(ENa−v)︸IsNa+INa+gl(El−v)︸Ileak+ghq(Eh−v)︸Ih+INMDA+IAMPA+IGABA︸Isyn,d[Ca2+]dt=2βr((gCa(v)+0.1gL)zF(ECa−v)−PCa[Ca2+]),dqdt=q∞−qτq(v),dhdt=αh(v)(1−h)−βh(v)h,dndt=αn(v)(1−n)−βn(v)n,
(1)
where v is the voltage and cm is the membrane capacitance. There are eight intrinsic currents of the DA neuron: a calcium current (ICa), a calcium-dependent potassium current (IKCa), a potassium current (IK), a direct rectifier current (IDR), a subthreshold sodium current (IsNa), a hyperpolarization-activated current (Ih), a fast sodium current (INa), and a leak current (Ileak). The first subgroup of intrinsic currents: ICa, IKCa, IK, IsNa and Ih constitute a pacemaking mechanism of the DA neuron. The second subgroup of the intrinsic currents (INa, IDR) is responsible for spike generation. The last subgroup includes synaptic currents: the excitatory α-Amino-3-hydroxy-5-methyl-4-isoxazolepropionic (AMPA) and N-Methyl-D-aspartate (NMDA) receptor currents (IAMPA and, INMDA respectively) and the inhibitory γ-Aminobutyric acid (GABA) receptor current (IGABA). Synaptic inputs can produce bursts and pauses in firing.
The main currents of the model that produce pacemaking activity of DA neuron are an L-type voltage-dependent calcium current (ICa) and an SK-type calcium-dependent potassium current (IKCa). Gating of the calcium current is instantaneous and described by the function:
gCa=gCa¯⋅αc4(v)αc4(v)+βc4(v)
(2)
Calibration of the calcium gating function reflects an activation threshold of an L-type current, which is significantly lower in DA cells than in other neurons (~ -50mV; [4]). Calcium enters the cell predominantly via the L-type calcium channel. Contribution due to the NMDA channel is minor [92]. Thus, calcium concentration varies according to the second equation of the system (1). It represents balance between Ca2+ entry via the L channel and a Ca2+ component of the leak current, and Ca2+ removal via a pump. In the calcium equation, β is the calcium buffering coefficient, i.e. the ratio of free to total calcium, r is the radius of the compartment, z is the valence of calcium, and F is Faraday’s constant. Pca represents the maximum rate of calcium removal through the pump. A large influx of Ca2+ leads to activation of the SK current, which contributes to repolarization as well as afterhyperpolarization of the DA cell. Dependence of the SK current (IKCa) on calcium concentration is modeled as follows:
gK,Ca=gK,Ca¯⋅[Ca2+]4[Ca2+]4+[K+]4
(3)
The neuron is repolarized by the activation of a large family of voltage-gated potassium channels. In addition to the already described potassium current, the model contains voltage-dependent K current (IK). Conductance of this current is given by a Boltzmann function:
gK=gK¯⋅11+exp(−(v+10)7)
(4)
The DA neuron expresses voltage-gated sodium channels that carry a large transient current during action potentials (a spike-producing sodium current) and a non-inactivating current present at subthreshold voltages (a subthreshold sodium current). Even though the persistent subthreshold sodium current is much smaller than the transient spike-producing current, it influences the firing pattern and the frequency of the DA neuron by contributing to depolarization below the spike threshold [25]. We modeled the voltage dependence of the subthreshold sodium current as follows:
gsNa=g¯sNa11+exp(−(v+50)5)
(5)
The kinetics and the voltage dependence of the subthreshold sodium current were taken from [93].
The majority of DA neurons express a hyperpolarization-activated nucleotide-gated (HCN) inward cation current (Ih). The HCN current contributes to spontaneous firing of subpopulations of DA neurons [94]. The activation variable of Ih is governed by a first-order ordinary differential equation (the third equation of the system (1)).
The maximal activation of Ih current is described by the following voltage-dependent equation [88]
q∞=11+exp(v+958)
(6)
The voltage-dependent time constant is described by
τq=625⋅exp(0.075(v+112))1+exp(0.083(v+112))
(7)
The leak current (Ileak) in the model has the reversal potential of -35 mV, which is higher than in the majority of neuron types. In DA neurons, several types of depolarizing, nonselective cation currents are expressed, which likely contribute to depolarization during interspike intervals.
DA neurons receive glutamatergic (Glu) excitatory drive through AMPA and NMDA receptors and inhibitory drive through GABA receptors. Changes in the membrane potential induced by synaptic conductances are described by the following equation
Isyn=gNMDA(v)sig(sNMDAact)(ENMDA−v)︷INMDA+gAMPAsig(sAMPAactsAMPAdes)(EAMPA−v)︷IAMPA+gGABAsGABA(EGABA−v)︷IGABA
(11)
where gNMDA(v), gAMPA, gGABA are the maximal conductances of NMDA, AMPA and GABA receptor currents accordingly, sNMDA, sAMPA, sGABA are gating variables that depend on the input spike trains.
The AMPA and GABA conductances are voltage-independent, but the NMDA conductance has voltage sensitivity as in [95]
gNMDA(v)=g¯NMDA1+0.1[Mg2+]e−mev
(12)
where [Mg2+] denotes the amount of magnesium, taken to be 0.5μM. The low slope of the voltage dependence (me = 0.062) is critical for the increase in the frequency of spikes or subthreshold oscillations during NMDA application [56].
A multidimensional system can be analyzed with two-dimensional nullcline methods only after its reduction to a two-dimensional system. The description of this method is provided by Strogatz [104] and Rinzel and Ermentrout [29]. The reduction was done by eliminating the spike-producing currents, which do not significantly change the firing frequency of the neuron [4]. Nullclines are the curves where either dvdt=0 or d[Ca2+]dt=0. Accordingly, nullclines of our system were obtained by numerically solving the following equations:
dvdt=1cmgCa(v)(ECa−v)+(gKCa([Ca2+])+gK(v))(EK−v)+gsNa(v)(ENa−v)+gl(El−v)=0;d[Ca2+]dt=2βr((gCa(v)+0.1gl)zF(ECa−v)−PCa[Ca2+])=0.
(16)
The type of bifurcation was determined by systematically varying parameters of the system (in our case gGABA) until the behavior of the system qualitatively changes.
The model of DA release is adopted from Wightman and Zimmerman (1990) [105] and is described by the following equation
d[DA]dt=[DA]maxδ(t−tspike)−Vmax[DA]Km+[DA]
(17)
The first term describes the release from spiking activity of the DA neuron. Dirac delta function δ(t − tspike) represents the release at time of a spike. Maximum amount of DA released per spike is [DA]max = 0.1μM. The second term represents DA uptake described by Michaelis-Menten equation, where Vmax = 0.004μM/ms is the maximal rate of uptake by a transporter and Km = 0.2μM is the affinity of the transporter for dopamine.
Heterogeneity in the population of DA neurons was putatively introduced by varying the leak conductance. Further, neurons received correlated fluctuating NMDA inputs. NMDAR conductance to each DA neuron was given by a linear summation of Ornstein-Uhlenbeck (OU) processes [106] described as following:
gNMDA(t)=μ+σ(1−cxi(t)+cxc(t))
(18)
where μ = 1.5mS/cm2 and σ = 0.5mS/cm2 are the values of the mean and the standard deviation of the NMDAR conductance used for the simulations. xc(t) is the common component of the NMDA input that was applied to all of the DA neurons, whereas xi(t) is the independent component, which was generated individually for each neuron. A shared fraction of the input is determined by the input correlation c and was set to c = 0.5. Each OU process was formed by the following equation:
dx=−xτdt+Nτxdt
(19)
where x(t) is Gaussian white noise with zero mean and unit variance. Nτ = (2/τ)1/2 is a normalization constant that makes x(t) have unit variance. A correlation time τ = 5ms was used [33,64].
In the model with fast sodium and the delayed rectifier potassium spike-producing currents, a spike was registered whenever voltage oscillation reached the threshold of 0 mV. In the reduced model (without spike-producing currents), a spike was registered every time voltage oscillations crossed the threshold of -40 mV, as experimentally it was shown that a DA neuron action potential is triggered when the voltage is depolarized to approximately -40 mV [3]. Voltage oscillations that were below these thresholds in the models with and without the spike-producing currents respectively were not counted as spikes and did not contribute to the firing frequency. To analyze firing pattern of simulated DA neuron in the presence of different synaptic currents, we quantified its firing rate and bursting. Mean firing rate of the simulated DA neuron was calculated as an inverse of the mean interspike interval (ISI). To calculate bursting we used ISI coefficient of variation (CV), calculated as the SD/mean of 200 ISIs.
|
10.1371/journal.pcbi.1000175 | Preferentially Quantized Linker DNA Lengths in Saccharomyces cerevisiae | The exact lengths of linker DNAs connecting adjacent nucleosomes specify the intrinsic three-dimensional structures of eukaryotic chromatin fibers. Some studies suggest that linker DNA lengths preferentially occur at certain quantized values, differing one from another by integral multiples of the DNA helical repeat, ∼10 bp; however, studies in the literature are inconsistent. Here, we investigate linker DNA length distributions in the yeast Saccharomyces cerevisiae genome, using two novel methods: a Fourier analysis of genomic dinucleotide periodicities adjacent to experimentally mapped nucleosomes and a duration hidden Markov model applied to experimentally defined dinucleosomes. Both methods reveal that linker DNA lengths in yeast are preferentially periodic at the DNA helical repeat (∼10 bp), obeying the forms 10n+5 bp (integer n). This 10 bp periodicity implies an ordered superhelical intrinsic structure for the average chromatin fiber in yeast.
| Eukaryotic genomic DNA exists as chromatin, with the DNA wrapped locally into a repeating array of protein–DNA complexes (“nucleosomes”) separated by short stretches of unwrapped “linker” DNA. Nucleosome arrays further compact into ∼30-nm-wide higher-order chromatin structures. Despite decades of work, there remains no agreement about the structure of the 30 nm fiber, or even if the structure is ordered or random. The helical symmetry of DNA couples the one-dimensional distribution of nucleosomes along the DNA to an intrinsic three-dimensional structure for the chromatin fiber. Random linker length distributions imply random three-dimensional intrinsic fiber structures, whereas different possible nonrandom length distributions imply different ordered structures. Here we use two independent computational methods, with two independent kinds of experimental data, to experimentally define the probability distribution of linker DNA lengths in yeast. Both methods agree that linker DNA lengths in yeast come in a set of preferentially quantized lengths that differ one from another by ∼10 bp, the DNA helical repeat, with a preferred phase offset of 5 bp. The preferential quantization of lengths implies that the intrinsic three-dimensional structure for the average chromatin fiber is ordered, not random. The 5 bp offset implies a particular geometry for this intrinsic structure.
| Eukaryotic genomic DNA exists in vivo as a hierarchically compacted protein-DNA complex called chromatin [1]. In the first level of compaction, 147 bp lengths of DNA are wrapped in 1 3/4 superhelical turns around protein spools, forming nucleosomes [2]. Consecutive nucleosomes are separated by short stretches of unwrapped “linker” DNA. Most chromatin in vivo is further folded into shorter, wider fibers, ∼30 nm in diameter. Despite much effort, the structure of the 30 nm fiber remains unresolved [3],[4].
Here we report that an analysis of the relative locations of nucleosomes along the DNA sheds new light on chromatin fiber structure. The connection arises from the helical symmetry of DNA itself [5]–[8]. Each base pair increase in separation between two consecutive nucleosomes moves them apart by 0.34 nm along the DNA - a potentially minor change relative to the 30 nm fiber's width. However, because of the 10.2–10.5 bp per turn helical symmetry of DNA, this 0.34 nm translation is coupled to a ∼35° rotation about the DNA helix axis, rotating the second nucleosome to an entirely different location in space, creating an entirely different intrinsic chromatin structure.
In vivo, attractive nucleosome-nucleosome interactions [9],[10] might overwhelm this intrinsic structure for the chromatin fiber, and impose a particular folded structure that is independent of exact linker DNA lengths. In that case, changes in the fiber's intrinsic structure would be manifested instead as changes in the folded fiber's stability [8]. Because of the high torsional stiffness of DNA and the short lengths of linker DNAs, such changes in stability would be of great energetic significance.
While steps of one or several bp profoundly alter the intrinsic fiber structure, steps of 10–11 n bp (integer n) do not: instead, the next nucleosome rotates n complete turns around the DNA helix axis, ending up rotationally near where it began, but translated along the DNA by ∼3.4–3.7 n nm. If linker DNA lengths varied randomly about an average value, the resulting intrinsic chromatin structure would be a random flight chain. But if linker DNA segments instead were equal in length modulo the DNA helical repeat, this would define an intrinsically ordered (but possibly irregular) superhelical structure for the chromatin fiber, with the detailed intrinsic structure highly depending on the phase offset d0 (integer) for linker DNAs of length 10n+d0 bp.
There are many hints in the literature for a ∼10 bp-periodicity in lengths of linker DNAs [5]–[8], [11]–[13]; however, the results are inconsistent. An early analysis of oligonucleosome DNA lengths suggested that linker DNAs in the yeast S. cerevisiae preferentially occur in lengths of 10n+5 bp, while those in human HeLa and chicken erythrocyte cells have no periodicity [5]. Analogous studies on rat liver chromatin first did not [14], but later did [6], reveal periodic linker DNA lengths, again of the form 10n+5 bp. Later genome-wide correlation analyses of AA and TT dinucleotides (which favor particular locations within the nucleosome [15],[16]) similarly yielded variable results, suggesting preferences of the form 10n+5 [11] or ∼10.6n+8 bp for yeast [12], ∼10.6n+8 for Caenorhabditis elegans and Drosophila melanogaster, [12], and ∼10n+8 for human k562 cells [13].
These conflicting conclusions of existing studies motivated us to develop two new independent computational methods and new experimental data, to define the probability distribution of linker DNA lengths in yeast. Our results from both approaches show that linker DNA lengths in yeast are indeed preferentially periodic, implying that the yeast genome encodes an intrinsically ordered three-dimensional structure for its average chromatin fiber.
A well-known characteristic of nucleosome DNA sequences is the ∼10 bp periodicity of key dinucleotide motifs, particularly AA, TT, TA, and GC. AA/TT/TA steps occur in phase with each other, and out of phase with GC [16]–[18]. These facts allow one to test for genomically encoded preferences in linker DNA lengths. Consider a set of experimentally mapped nucleosome sequences S = {s1, …, si, … sI}, strictly aligned at their dyad (2-fold rotational symmetry) axes. Extend each aligned sequence in both directions along the genome by roughly one nucleosome length (Fig 1A, note: positions from 1 to 147 stand for the nucleosome region). Occurrences of AA/TT/TA motifs as a function of position in the flanking regions of S would then exhibit collective patterns that are determined by the distribution of linker DNA lengths around the central (mapped) nucleosomes. If the central nucleosomes were perfectly aligned, and linker lengths were a constant, d0, then the nucleosomes in the up-/downstream region of S would also be strictly aligned. A significant periodic signal from AA/TT/TA motifs would then occur at up-/downstream positions dependent upon d0 (Figure 1B). If instead the ith linker length downstream of sequence si, li, equals 10ni+d0 for some integer ni and a fixed d0 (0≤d0<10), then the nucleosomes immediately downstream of S would not be strictly aligned, but would instead be offset by a multiple of 10 bp relative to each other. In this case, the ∼10 bp periodicity of dinucleotide motif signals would be roughly maintained in the extended regions, but more weakly, since end regions of the adjacent nucleosomal DNA in some sequences would be partially aligned over linker DNA in other sequences. Alternatively, if linker DNA lengths were random, these dinucleotide motifs would lack any significant periodicity in the extended regions.
We used this approach to test for intrinsically encoded linker DNA length preferences in the yeast genomes. Our in vivo yeast nucleosome sequence collection (filtered for nonredundant sequences of length 142–152 bp) contains 296 sequences. We focus the analysis on the AA/TT/TA signal because this is the most strongly periodic in aligned nucleosome sequences [16]. Alignment of these sequences (Figure 2) reveal several striking features. Sharp signals at positions −1 and 149, and systematic differences in average AA/TT/TA frequency between the original mapped nucleosomes and the extended regions, may reflect sequence preferences of the micrococcal nuclease [19] which is used to biochemically isolate the nucleosomes, or may reflect sequence preferences intrinsic to nucleosomes and linker regions [20].
Most importantly, the plot reveals hints of a ∼10 bp periodicity in the extended regions, implying that the yeast genomes intrinsically encode preferentially quantized linker DNA lengths of the form ∼10n+d0. The value of d0 can be deduced from the positions of the AA/TT/TA peaks in the extended region. Assume the AA/TT/TA signal appears periodically at positions 8, 18, …79, … 139 within a nucleosome region [15],[16]. If the linker length is 5 bp (or more generally 10n+5), then the expected positions of AA/TT/TA peaks (as indexed in Fig. 1B) in the downstream nucleosome region would be 160, 170,…, 231,… 291 (or these indices+10n if the linker length is 10n+5). In accord with this analysis, the AA/TT/TA peaks in Figure 2 are roughly positioned at 10's or 1's in the downstream region. Therefore we conclude that the preferential linker length value in the yeast data obeys the rule 10n+5(d0 = 5).
To test the significance of the observed 10 bp periodicity, we first calculated the Fourier transform of the AA/TT/TA signal in the extended region from position 147+d to 147+d+180 (Figure 2) for a given d. We chose d to be greater than 10 to avoid the sharp peaks observed at the boundaries of mapped nucleosome sequences (at positions −1 and 149 in Figure 2) that likely owe to the micrococcal nuclease enzyme specificity. We then varied d from 11 to 20. The amplitude spectrum (square root of spectral power) averaged over all d's is plotted as a function of period in Figure 3 (red curve). As a control, we constructed 500 randomly shifted samples of the extended regions by choosing a random di value between 11 and 20 for each sequence si, i = 1, … I (see details in Methods). As another control, we obtained 500 random genomic samples, each sample containing the same number of sequences of the same length (180 bp) randomly selected from the genome. The mean and 95% percentile of the Fourier transform amplitude at each periodicity value derived from these two sets of control samples are also plotted in Figure 3.
A significant peak at the 5% level (i.e., where the average amplitude from the extended samples with fixed d's (red curve) exceeds the 95% percentile line from the randomly shifted samples (green dashed line) or random genomic samples (blue dashed line)) is observed at period ∼10.2 bp. Because multiple peaks exist around 10 bp periodicity, we use the total power corresponding to period between 9–11 bp as the test statistic. The total power at 9–11 bp averaged over different d's was compared with that of the random shifted and random genomic samples. The resulting p-values are 0.008 and 0 respectively, refuting the hypothesis that linker DNA lengths within any 10 bp range are uniformly preferred in the genome. Instead, these significant ∼10 bp periodicities are consistent with the hypothesis that linker lengths in the yeast genome prefer values of the form ∼10n+d0, for some constant d0.
Information about preferred values for d0 is contained in the phase of the corresponding ∼10 bp periodicity peak in the Fourier transform. In Table 1 we present the location of the Fourier transform amplitude peak (T*) around 10 bp periodicity, and the corresponding phase angle in radians, for all d's. If the experimentally obtained nucleosome sequences were perfectly aligned, and if linker DNA lengths were genuinely 10n+d0 for a constant d0, then shifting the downstream region leftward by d0 bp will synchronize the extended region's AA/TT/TA motif signal with that in the original mapped nucleosome region. For example, suppose the true linker length is 15 bp (i.e., 10n+d0 with n = 1 and d0 = 5). As indexed in Figure 1, the downstream nucleosome AA/TT/TA peaks will be positioned at 170, 180, … 301. By shifting the downstream region leftward by 5 bp, we will extract the region comprising basepairs 153 ( = 147+5+1) and above. The AA/TT/TA peaks in the extracted region now are positioned at basepairs 18, 28,…149 (relative to the first basepair of the extracted region). We hence expect the phase of Fourier transform at period ∼10 bp from this extracted region to be close to that from the original mapped nucleosome sequences, which had AA/TT/TA peaks positioned at 8, 18,…139 relative to their own first basepairs. Consistent with this analysis, the phase of the AA/TT/TA signal when d = 15 (i.e., when the extended region begins on the 16th basepair after the end of the mapped nucleosomes) is 2.56 in radians, closest among all d's to 2.48, the phase from the original mapped nucleosomes. Based on this criterion therefore, we conclude the optimal d0 is 5 bp.
This phase analysis for detecting the preferred quantized linker DNA lengths (i.e., the preferred d0) assumes that the AA/TT/TA motif maintains the same periodicity in the extended region as in the mapped nucleosomes. This is true: the periodicity having maximum Fourier amplitude (T*) equals 10.20 bp for both the mapped nucleosomes and the extended regions (Figure 3 and Table 1). Hence this analysis implies that linker DNA lengths in yeast are preferentially quantized, with the form ∼10.2n+5 bp. The amplitude in the extended region however is much lower than in the original core region. This may suggest that that the linker length distribution is not strictly quantized at (odd) multiples of 5's, but rather in a form possessing non-degenerate peaks centered around (odd) multiples of 5's.
To test the conclusions of the Fourier analysis described above, and to better define the preferred phase offsets d0, we developed a duration hidden Markov model (DHMM, [21]) and used it to analyze a new collection of DNA sequences of dinucleosomes from yeast which we isolated for this purpose. Dinucleosomes are two nucleosomes connected by their linker DNA. The DHMM models the dinucleosomes as an oscillating series of two “hidden” states: a fixed-length (147 bp) nucleosome and a variable length linker DNA. A technical detail is that, as isolated biochemically, dinucleosomes may come with additional short partial linker DNA segments at either end, or alternatively, may be over-digested so as to have an incomplete nucleosome at either end. We generalize our DHMM to allow for this possibility. The algorithm predicts the positioning of two nucleosomes (complete or partial) in each sequence, and then uses the predicted results to update parameters in the model that describe the length and sequence preferences of the linker DNA. In particular, as the algorithm proceeds iteratively, the linker length distribution is updated using the kernel smoothing method (see details in Methods).
We isolated and fully sequenced 335 non-redundant dinucleosomes from yeast, with lengths ranging from 280 to 351 bp. Some of the dinucleosome sequences were shorter than 2*147 bp, meaning that they have been over-digested on at least one of their two ends. For such sequences the optimal path is more difficult to predict because of loss of information in either end. Hence we restricted our analysis to 214 sequences whose lengths are ≥300 bp.
At convergence of the model, the results (Figure 4A) confirm the results from the independent Fourier analysis of extended mononucleosome sequences (Figure 2 and Table 1, above): the linker length distribution function FL(d) obeys the rule 10n+d0 with the phase offset d0 = 5 bp, such that the most probable linker lengths (in the kernel-smoothed distribution) are around 5, 15, 25, 35, and 45 bp.
The noise reflected in Figure 4A comprises two chief components: individual major peaks can be slightly offset from 5's; also small peaks arise at seemingly random positions. This variability in the estimated density curve is not surprising, since we are estimating a distribution in an infinite-dimensional space. To reduce the dimensionality of the problem, we next consider a parametric approach in which we impose a periodicity on the linker length distribution function, FL(d), while allowing variability around each period. Such a distribution can be characterized by a mixture of Gaussian distributions with means that are equally spaced by 10 bp, and a common unknown variance (see Methods). The algorithm proceeds in the same way as the kernel smoothing method above, except that the linker length distribution is estimated using an EM algorithm.
The results at convergence of the model (Figure 4B) imply that, if the linker lengths do indeed prefer the form specified in this model, then the optimal d0 value is ∼5 bp. The estimate of the common standard deviation of the Gaussian components was 1.43, indicating a modest uncertainty of the linker length distribution around the quantized values. We further generalized the linker length model by treating the period as an unknown parameter and assuming heterogeneity in the variance of Gaussian components. The resulting maximum likelihood estimator of the period is 9.8 bp and the linker length distributions closely resemble those of Figure 4B (results not shown). Taken together, these results confirm the results of the Fourier analysis and of the DHMM kernel smoothing analyses. All of these analyses imply that linker DNA lengths in yeast obey the form 10n+d0 with d0 equal to ∼5 bp.
One possible concern in the DHMM analyses is whether the ∼10 bp periodicity in the linker length distribution could somehow arise from the model itself, especially given the ∼10 bp periodicity of motif signals inherent in the nucleosomes. Two simulation studies tested and disproved this possibility. One test simulated random sequences based on a product multinomial model with base composition and length distribution identical to that in the true dinucleosome sequences; the second test shuffled the natural dinucleosome sequences while keeping the dinucleotide frequency fixed within each sequence. The DHMM-kernel procedure was followed exactly as before. In both simulations, the resulting linker length distribution varied between trials, and the ∼10 bp periodicity disappeared in general (Figure S1). The DHMM-mixture method imposes the 10 bp periodicity on the linker length, but the peak positions often moved little away from their initial values of μ as the algorithm proceeded - presumably because, unlike for the real sequences, the randomized sequences lack signals that spur the movement of μ in the true nucleosome sequences. Thus, the real data are distinguished from the random data in both versions of the DHMM. We conclude that the linker length patterns deduced by these analyses reflect true signals of nucleosome organization present in the dinucleosome sequences.
To evaluate the robustness of these DHMM analyses to over- or under-digestion of the biochemically isolated nucleosomes and dinucleosomes, we carried out a simulation of the entire combined experiment. We simulated 2000 nucleosome sequences based on the experimentally obtained yeast nucleosome profile (a heterogeneous Markov chain model). Both ends of each simulated nucleosome were subjected to a random truncation or addition to the 147 bp-long nucleosome core by up to 3 bp, creating a set of simulated yeast nucleosome sequences having lengths in the range 141–153 bp, slightly greater than the 142–152 bp range of lengths in the real nucleosome sequences. Similarly, we simulated 2000 dinucleosome sequences, each starting and/or ending with a (simulated) nucleosome that was subject to a random truncation or addition of up to 20 bp. The linker DNAs were simulated using the homogeneous Markov chain model obtained from the yeast dinucleosome data, while the true linker length distribution followed a periodic distribution with peaks at 15,30,…105 (Figure 5A). The length range of resulting dinucleosome sequences is ∼250–440 bp, which we again filtered to retain lengths greater than or equal to 300 bp. We followed the same center alignment and model training procedure as for the real data. The periodic linker length distribution was successfully recovered using both the kernel smoothing and mixture model approaches (Figure 5B and 5C). Similar results were obtained with a small subset (300) of the dinucleosome sequences (Figure 5D), where the superior performance of the mixture method on this smaller dataset is evident (Figure 5C vs. Figure 5B and 5E vs. Figure 5D). In another check, we simulated another 2000 dinucleosome sequences having a uniform distribution for the linker length on [1,…120]. The resulting predicted linker length distribution (Figure 5F) lacks significant periodicity; peaks formed randomly, and their positions varied from sample to sample.
Classic experimental measurement of the nucleosome repeat length provide several additional checks on the results from the DHMM analyses. Experiments using gel electrophoresis to analyze the DNAs that result from random partial nuclease digestion of chromatin routinely reveal ladder-like patterns of DNAs fragments, which reflect a repetition of a (relatively) discrete sized structural unit comprising a nucleosome plus one average linker DNA length. The length of DNA in this repeating unit is referred to as the nucleosome repeat length. Specifically, the nucleosome repeat length may be defined, and measured, as the average length difference in base pairs between DNA fragments containing n+1 and n nucleosomes. In one test of our analyses, we find that the average length of linker DNA for yeast predicted from the kernel smoothing method is 20.2 bp (21 bp from the mixture model), in good agreement with the experimental value of ∼18 bp for yeast [22], and in good agreement with subsequent studies suggesting that ∼20 bp may be a more-accurate value than ∼18 bp [3],[8]. As a second check on our analyses, we simulated the complete experimental measurement of nucleosome repeat length. We first simulated the chromatin fiber itself, given our linker length distribution function deduced from the DHMM analysis, then simulated the random partial nuclease digestion, and then finally simulated the gel electrophoresis analysis of the resulting DNA fragments. The simulated chromatin fibers comprised 50,000 nucleosomes with linker DNA lengths distributed as from the mixture model results, i.e., μ = (5,15,25,35,45), σ = 1.43 and η = (0.3271,0.1682,0.1636,0.2243,0.116) (μ was rounded to integers for convenience). The simulated nucleosome chain was then subjected to a simulated nuclease digestion. To mimic the partial nuclease digestion conditions used experimentally, each linker was subject to zero or one enzyme cut, at a random position and with probability proportional to its length, such that the resulting DNA fragments had a wide range of numbers of nucleosomes, with a mean of 5 nucleosomes. The simulated gel intensity profile (Figure 6) resembles those observed experimentally. Thus, the complex linker DNA length distributions deduced in our DHMM analyses are consistent with the experimental observations of ladder patterns in nuclease digestions of chromatin. Finally, as a third check on our analyses, we used these simulation-derived plots of frequency versus fragment size to recover an apparent nucleosome repeat length. The average repeat length based on 50 simulations was 168.5 bp with standard deviation 1.0, which accurately recovered the true theoretic repeat length for this modeled distribution, 147+21.3 = 168.3 bp.
We conclude from all of these tests that the complex linker DNA length distribution functions deduced with our DHMM analyses represent true features in the dinucleosome DNA sequences, and that they are compatible with available experimental data on nucleosome repeat lengths.
In this paper, we developed and applied two different methods to investigate the distributions of linker DNA lengths in yeast. Despite being fully independent, and applied to different kinds of experimental data (genomic DNA sequences adjacent to experimentally mapped nucleosomes, and, separately, sequences of biochemically isolated dinucleosomes), both methods lead to the same conclusion: linker DNA lengths are not described by a uniform distribution, but instead are preferentially quantized, obeying the form 10n+5 bp.
Our results accord with some, but not others, of the previous experimental studies of linker DNA lengths in yeast. Surprisingly, our Fourier analysis could not detect evidence of periodic higher order structure in the recent genome-wide map of yeast H2A.Z-containing nucleosomes [23], using either their coarse-grained or fine-grained calls. Using a nonredundant subset comprising 1617 of their best-mapped nucleosomes (those which reveal the nucleosome's periodic AA/TT/TA signature [23] our Fourier analysis of dinucleotide frequency in the corresponding extended regions did reveal a ∼10 bp periodicity, with a phase offset d0 = 5 bp, equivalent to that observed with our smaller number of conventionally sequenced yeast nucleosomes; however this periodicity did not pass a test for significance at the 0.05 level. We suspect that the mapping accuracy of that genome-wide nucleosome collection, which includes nucleosome DNA fragments ranging in length from ∼100–190 bp that are sequenced at only one end, may simply be inadequate to reveal the fine-scale structure revealed by analysis of our conventionally mapped and sequenced nucleosomes.
It is possible that our yeast nucleosome collection may be enriched for an especially stable subset of nucleosomes due to sampling bias imposed by nucleosome mapping technology, and thus could reflect a particular chromatin structure that is enriched in such genome regions. That said, however, our ongoing analysis of more than 50,000 newly mapped unique yeast nucleosome sequences (accounting for ∼67% of the entire genome) leads to exactly the same conclusions regarding linker DNA lengths in yeast (unpublished results), suggesting at least that this linker length form 10n+5 is representative of much of the yeast genome.
Nevertheless, we note that our present analysis reveals only a single average most probable linker length distribution. It remains possible that the detailed distribution of linker DNA lengths (and corresponding intrinsic chromatin fiber structures) may vary with location throughout the genome. It is also possible that different species could have different most-probable linker DNA length distributions. Indeed, our ongoing study suggests that linker DNA in human k562 cells human may preferentially occur at lengths that are quantized at 10's. This result however is preliminary and requires further investigation.
Several aspects of our findings are significant. The existence of preferred linker DNA lengths that are constant, modulo the DNA helical repeat, implies an ordered superhelical structure for the average intrinsic chromatin fiber. The spread of detailed linker DNA lengths around the preferred quantized values (Figure 4) could reflect random disorder about an intrinsically ordered structure; or it could actually reflect the opposite of that, namely, a tendency to improve the local structural order by compensating for inevitable sequence-dependent differences in the intrinsic helical twist of DNA [24]. The 5 bp phase offset means that, on average, consecutive nucleosomes in the yeast genome tend to start from opposite faces of the DNA double helix.
Our work also introduces two approaches for the analysis of linker DNA lengths in any eukaryote for which the needed experimental data are available. In the Fourier analysis, an implicit assumption we made is that the nucleosome cores in the extended regions have the same features as the mapped ones, including the periodicity and relative phases of AA, TT, TA, and GC signals. The justification for this assumption is that these features of nucleosome DNA sequences are thought to reflect the requirement of DNA wrapping in the nucleosome, and to be generic to all nucleosomes [17],[25]. The success of the Fourier method highly depends on both the alignment quality and on the extent to which the linker DNA lengths are actually quantized. A bad alignment tends to degrade the 10 bp periodicity of AA/TT/TA signal in the extended region, just as occurs in the randomly shifted samples (i.e. a randomly shifted sample can be regarded as resulting from randomly aligned nucleosome sequences). In reality the center alignment is not perfect due to various factors such as sequence specificity of the nuclease which is used to biochemically isolate the nucleosomes. Hence we believe that the AA/TT/TA periodicity in the extended region based on a “true” alignment would be even stronger than as obtained in Table 1.
The DHMM provides a general framework for analysis of the linker length distribution function. The components of the DHMM (e.g., the model for the nucleosome sequences or the lengths and sequences of the linker DNA) are not limited to what have been used in this paper: any probabilistic models for the two states can be readily adapted into this framework. The legitimacy of the conclusion regarding the linker DNA length distribution, which is drawn based on the DHMM model, depends on the validity of the model assumptions. Markovian models have proved exceedingly successful in modeling natural DNA or protein sequences in various important problems. In this paper, we proposed a first-order inhomogeneous Markov chain model for the nucleosome state. This model is explicitly designed to characterize the sequential dependence of nucleosomal DNAs in the form of dinucleotides. In addition, it accounts for the variation of signal intensity as a function of positions within the nucleosome region. The need for representing dinucleotides instead of just mononucleotides was explicitly demonstrated in our earlier study [16]. Similarly, the distinction of transition probabilities among positions in the nucleosome region is essential in the prediction of nucleosome positioning, given that the dinucleotide signals are known to be periodic [17],[25]. As expected from these past studies, our training data show that the transition probabilities are NOT homogeneous at different positions across the nucleosome core region. The resulting nucleosome model contains a large number of parameters in the transition matrices (see Methods) because of this time-dependence. Nevertheless, from this perspective, over-fitting is not a big concern in this model. In addition even if this assumption were mis-specified, the trained transition probabilities are still unbiased and consistent estimates of the true parameters. The only loss incurred would be some asymptotic efficiency of the estimates from a statistical point of view. One limitation is that the DHMM is a complex machinery, involving many parameters. Thus we are unable to provide a measure for uncertainty in terms of the entire distribution of linker length, other than the variability around the quantized values quantified by the DHMM-mixture approach. This remains as an open problem.
We obtained 296 nonredundant 142–152 bp long in vivo nucleosome DNA sequences from yeast as described [18]. These sequences were mapped to the genome using BLAST [26]. Dinucleosomes (experimentally isolated chromatin oligomers containing just two nucleosomes) were purified from yeast as described [18], except using less micrococcal nuclease and then gel purifying, cloning, and sequencing protected dinucleosomal DNAs instead of mononucleosomal. We isolated and fully sequenced 335 non-redundant dinucleosomes, with lengths ranging from 280 to 351 bp. These were subsequently filtered (see Results) to yield 214 sequences whose lengths are ≥300 bp. We compared the 296 mononucleosome sequences with the 214 dinucleosome sequences that were at least 300 bp, and found only 4 of them were overlapped. Therefore these two collections can be regarded as two independent sets.
The center of each experimentally mapped nucleosome DNA sequence was treated as the dyad symmetry axis and was indexed as position 74. We then extended the genomic DNA sequence on both strands in the 3′ direction for 200 bp. The resulting extended sequences were aligned according to the center of the mapped nucleosome sequences (Figures 1A, and 2). We denote the extended sequence as S = s1, …sI. We sequentially obtained the aligned chunk of DNA of length L0 from position (147+d+1) to (147+d+L0) for d = 11, …, 20 bp in the downstream region. d is chosen to be greater than 10 bp to avoid sharp peaks observed at the nucleosome boundaries. (Differing values d do not influence the observed periodicities, rather, they lead to slight perturbations of amplitude, because of variation of base composition. We then average the results obtained over the set of d values.) The average linker DNA length in yeast is ∼20 bp [1]. We therefore chose L0 = 180 bp such that the extended block roughly covers three full nucleosomes for each sequence. We further generated 500 randomly shifted samples as follows. For each sample, we first generated random shift values di∈{11, …, 20} for i = 1, … I. For each sequence si, we extracted the region from position (147+di+1) to (147+di+L0). These randomly shifted extended regions were center aligned.
Let f(t) be the AA/TT/TA frequency (smoothed with a 3 bp moving window to reduce noise from codons) in the tth column of the alignment of 296 nucleosome sequences or in the extended regions. We calculated the discrete Fourier transform of f(t) using N = 2000, i.e. . Let be the amplitude spectrum of f(t) from the downstream region from position (147+d+1) to (147+d+L0) as described in last paragraph. We averaged the amplitude spectrum over d's, i.e. , and plotted A̅k as a function of period in the range 6–20 bp, compared to the amplitude from the original center aligned nucleosomes, and to the average and 95% percentile of the amplitudes from random samples.
For clarity, we first describe our generic duration hidden Markov model (DHMM), which is appropriate for analysis of infinitely long chromatin fibers. We then consider refinements of the model that are necessary for analysis of dinucleosomes.
We model a long chromatin sequence as an oscillating series of two “hidden” states: nucleosome (N) and linker (L). At the end of each state, the chain must transit to the other state. The nucleosome state has a fixed duration of 147 bp, while the linker state duration (denoted as d) has an unknown probability distribution FL(d). We further define probabilistic models for the emission of events within each state. Let r = r1r2…rm be a DNA sequence. Suppose the N state has a model PN(r) : = P(r1, …rm| r is in a nucleosome), and the L state has PL(r) : = P(r1…rm| r is in a linker) (a subscript “N” or “L” of P is hereafter reserved for the conditional probability given that the sequence is a nucleosome or linker respectively, while a “P” without sub-/superscript denotes the probability in general). Note thatthus the probability distribution of linker length is explicitly modeled here.
For the nucleosome state, we use a first-order time-dependent (inhomogeneous) Markov chain model as in [18], motivated by two facts about nucleosomes: (1) the base composition is sequentially dependent, as revealed by strong patterns of dinucleotide motifs; and (2) the pattern varies as a function of position in the nucleosome (referred to as time here) [16],[17]. A time-dependent Markov chain captures the sequential dependence while allowing heterogeneity across different positions. More explicitly, let be the probability of observing the letter “a” at position 1 (a = A, C, G or T); and let be a 4×4 transition matrix specifying probabilities of observing a at position (i+1) given b at position i for i = 1, …, 146. Then for any given nucleosome sequence = e1…e147,We model the linker state with a homogeneous Markov chain, which can be fully defined by the initial base composition denoted as qe for (e = A,C,G,T) and a single transition matrix v : = [va|b] (defined analogously to ti above). For any linker DNA sequence e = e1…em,
For a DNA sequence r = r1r2…rm (Watson strand) and its reverse complementary (Crick) strand , let π = π0π1…πk…πKπK+1 be the path of underlying hidden states. The states π0,πK+1 are the initial and ending states without emission (“silent” states). The state πk, equal to N or L, is associated with a duration dk, such that (if πk = N, then dk = 147). Let the probability that a random sequence starts with the N state be τ, and that of ending in N be γ (i.e., P(π1 = N|π0) = τ, P(πK+1|πK = N) = γ). We set γ equal to τ throughout this paper, assuming balanced digestion at the two ends of the dinucleosomes during their biochemical isolation. Now suppose that π partitions r into m1 nucleosomes on the Watson strand: , and m2 interwoven linkers: , corresponding to and on the Crick strand (note: |m1−m2|≤1). Then(1)where I()is an indicator function. The exact value of m1 or m2 depends on the length of DNA sequence under modeling. For the dinucleosome data, the value for m1 is restricted to 2, while the value for m2 can be 1, 2, or 3. Using dynamic programming (e.g., [27]), one can find the optimal path π* that maximizes the probability, i.e.(2)
For dinucleosomes, the standard DHMM needs to be modified to reflect that the first and last non-silent states, i.e., π1 and πK, are in general truncated due to the extensive nuclease digestion used to maximize the yield of these short chromatin fragments. In other words, the duration for N and L at π1 and πK are different from internal πK's. Let and be the duration distribution of N and L states at π1 or πK. If π1 = N and the corresponding emission is (d1≤147), then , where is the occurrence probability of letter e1 at nucleosome position 147−d1+1. If π1 = L, then PL(e) must be calculated using in equation (1). Analogous modifications apply to πK.
Based on the center alignment of the 296 mononucleosome sequences, we trained a nucleosome model as follows. The probability for letter e at position j, i.e. (e = A/C/G/T), was estimated as the fraction of e at the jth column of the alignment of both strands for j = 1,…147. Likewise the transition probability in tj was estimated as the fraction of occurrence a at position j+1 given b at position j. The transition probabilities for the nucleosome model were again smoothed using a 3 bp moving window.
Let si, i = 1,…I be the dinucleosome sequences. We set to be a uniform distribution on [147−α, …, 147] and to be a uniform distribution on [1, 2, …, β], where α measures the maximum possible nucleosome truncation (i.e. maximal over-digestion into a nucleosome at either end of the dinucleosome) and β the maximum extra linker DNA length at either end of the dinucleosome. The ideal values of α and β should be chosen according to the real extent of truncation or extra linker DNA at two ends in the population of dinucleosomes, as isolated biochemically. If α or β is set too small, systematic biases in the resulting linker length distribution will result, as the path π predicted by the model must satisfy these constraints. On the other hand, choosing over-large values of α and β will inflate path space, degrading the precision of the predictions. We chose α and β to be 20 and 30 bp, respectively. The linker length distribution is initialized as uniform in [1, …, c], where c = 50. The linker DNA initial base composition (qe) and the transition matrix v are initialized equal to the corresponding average probabilities from the mononucleosome data. The linker length distribution is estimated iteratively, as follows (see the flow chart in Figure 7):
The empirical distribution of d from step 1 is noisy prior to convergence, therefore we employed a standard kernel smoothing technique with bandwidth of 1.5 bp to improve the estimation under a Gaussian kernel [28].
Our results using the DHMM kernel smoothing method suggest that linker DNAs preferentially occur according to the form 10n+ε where ε is a random term whose density has mode at a constant d0, such that 0≤d0<10, but with variability, i.e., that ε can take values d0, d0±1, ±2 etc.. If this model holds, Fig. 4 suggests that d0≈5. In the extreme case, where Var(ε) = 0, the linker lengths would be strictly quantized with the form 10n+d0.
We characterize such a distribution with a location mixture model. Let μ = (μ1,…,μK) be the peak positions in a projected range [1,2,…c], where each μk indexes a location distribution with density f(d;μk), i.e.: f(d;μk)≡f(d−μk) for k = 1,…K. The linker length value around μk is locally distributed as f(d;μk). Suppose that the probability of observing a linker length d from the kth component is ηk (and hence ). Let η = (η1,…,ηK). Then the marginal distribution of d can be written as .
Based on the results of the Fourier and DHMM-kernel smoothing analyses, we impose the constraint that these components are equally spaced with 10 bp period, i.e. μk = μ1+10*(k−1) for k = 1,…K. Under this model, the between-component distance is fixed throughout the algorithm. The absolute positions of components μ1,…μK are varied as a whole to maximize the likelihood as the weights η are simultaneously updated.
We modeled f with a Gaussian mixture distribution with a common standard deviation σ. The Gaussian density is normalized since the linker length distribution is discrete. The initial value of μk was set as 10*k+2.5 or 10*(k−1)+7.5, with equal weight 1/K for each k = 1,…K, and with K = 5. (We confirmed that the peak positions resulting from this analysis are consistent under different starting values). We follow the same four steps as in the kernel smoothing method, except that step 2 is replaced by a standard expectation-maximization algorithm (EM, [29]), in which (μ,η,σ) are updated using predicted linker lengths.
The mono- and dinucleosome sequences and some codes used in the paper will be available at http://bioinfo.stats.northwestern.edu/jzwang.
|
10.1371/journal.pgen.1000930 | Genetic Evidence for Hybrid Trait Speciation in Heliconius Butterflies | Homoploid hybrid speciation is the formation of a new hybrid species without change in chromosome number. So far, there has been a lack of direct molecular evidence for hybridization generating novel traits directly involved in animal speciation. Heliconius butterflies exhibit bright aposematic color patterns that also act as cues in assortative mating. Heliconius heurippa has been proposed as a hybrid species, and its color pattern can be recreated by introgression of the H. m. melpomene red band into the genetic background of the yellow banded H. cydno cordula. This hybrid color pattern is also involved in mate choice and leads to reproductive isolation between H. heurippa and its close relatives. Here, we provide molecular evidence for adaptive introgression by sequencing genes across the Heliconius red band locus and comparing them to unlinked wing patterning genes in H. melpomene, H. cydno, and H. heurippa. 670 SNPs distributed among 29 unlinked coding genes (25,847bp) showed H. heurippa was related to H. c. cordula or the three species were intermixed. In contrast, among 344 SNPs distributed among 13 genes in the red band region (18,629bp), most showed H. heurippa related with H. c. cordula, but a block of around 6,5kb located in the 3′ of a putative kinesin gene grouped H. heurippa with H. m. melpomene, supporting the hybrid introgression hypothesis. Genealogical reconstruction showed that this introgression occurred after divergence of the parental species, perhaps around 0.43Mya. Expression of the kinesin gene is spatially restricted to the distal region of the forewing, suggesting a mechanism for pattern regulation. This gene therefore constitutes the first molecular evidence for adaptive introgression during hybrid speciation and is the first clear candidate for a Heliconius wing patterning locus.
| Hybrid speciation challenges our view of biodiversity as a branching tree and is considered rare or absent in animals. A possible route by which it may occur is establishment of a novel “magic trait,” influencing both ecological adaptation and mating preference, via hybridization. We provide, to our knowledge, the first molecular genetic evidence for this process in the tropical butterfly Heliconius heurippa. We sampled molecular markers both linked to the locus controlling red color pattern and across the genome of Heliconius heurippa and its putative parents, H. cydno and H. melpomene. We found evidence of genetic introgression from H. melpomene into the hybrid H. heurippa only at the genomic region of the forewing red-band locus. This signature of introgression corresponds to the 3′ end of a kinesin gene that also shows a pattern of expression restricted to the distal region of the forewing. As the wing color pattern in these butterflies is crucial in maintaining the isolation of this species through mate choice, this study provides molecular support for the hybrid origin of a new adaptive trait that can lead to speciation.
| Identifying the genetic mechanisms involved in speciation is an important challenge in the study of evolution [1]–[3]. Empirical studies have shown that species differences can be localized in just a few genomic regions [3]–[5], and that reproductive isolation is more easily achieved when traits causing assortative mating are also subject to natural selection [6], [7]. Such characteristics have been termed ‘magic traits’ [6] and can facilitate speciation as a side-effect of ecological divergence in the presence of ongoing gene flow [8], [9]. Likely examples of such magic traits include body size and color in sticklebacks, flowering time in edaphic plants, host shifts in phytophagous insects, color patterns in Heliconius butterflies, beak size in Darwin finches, development time in melon flies and color patterns in Hypoplectrus fish [9]–[15].
If ‘magic traits’ were acquired by introgression from related lineages, adaptation and speciation could proceed without the requirement for novel mutations [16], [17]. Recent studies in plants and animals have shown that introgression can provide the raw material for adaptation [18]–[20]. Hence, it is plausible that if introgression produces new adaptive phenotypes that also generate reproductive isolation, for example through mate choice, habitat colonization or asynchronous emergence, then hybrid speciation can occur without geographic isolation [17], [21]. We have called such a scenario ‘hybrid trait speciation’, as a special case of speciation through hybridization without a change in chromosome number or homoploid hybrid speciation (HHS) [2], [22]. Hybrid trait speciation contrasts with what we have termed mosaic genome speciation, documented in Helianthus sunflowers, where the hybrid species genome is composed of blocks derived from both parental species [23]. Rapid establishment of incompatibilities between parental and daughter species can result due to the large number of genes causing epistatic hybrid breakdown in hybrids [23]. The two scenarios contrast in their genomic signature, with hybrid trait speciation potentially involving introgression of just a few adaptively important loci into the genetic background of one of the parental species, making it much more difficult to detect using traditional approaches based on ‘neutral markers’ [21]. This is in addition to the fact that detecting hybrid speciation at the molecular level is difficult anyway, due to incomplete linage sorting and historical gene flow, which can leave similar signatures of shared variation [24].
There is evidence that hybridization has played an important role in the adaptive radiation of Heliconius butterflies [21]. Heliconius have aposematic wing coloration, mimicry between divergent species and frequent hybridization, providing an excellent opportunity to study the genetics of adaptation and speciation [25]–[28]. In particular, studies in the closely related species H. melpomene and H. cydno, that occur sympatrically throughout Central America and in the west Andes, show that mimicry shifts are coupled with assortative mating and lead to speciation [29]. In addition, differences in host plant use, microhabitat preferences and partial hybrid sterility also contribute to reducing genetic interchange between these species [30]. The species hybridize in both the field and the laboratory, although natural interspecific hybrids are collected at a very low rate (one in a thousand or less) [28], [30]. Nonetheless, introgression of color pattern alleles has been observed in natural hybrid zones and the same phenotypes have been recreated in experimental crosses [31], [32]. Furthermore, studies using neutral markers reveal that introgression between the species has been frequent throughout their evolutionary history [33], [34].
Occasionally, novel color pattern variants produced through hybridization appear to have produced stable hybrid populations. The best studied example is Heliconius heurippa, found in the eastern slopes of Colombian Andes, which has a color pattern that can be recreated in the laboratory through crosses between H. c. cordula and H. m. melpomene, the races of the melpomene group adjacent to its current geographical range [35], [36]. H. heurippa is abundant and its color pattern is stable along several hundred kilometers of the Andean slopes, although is not mimetic with any other species [30], [37]. This wing pattern stability across a broad geographic area contrasts with the transient production of hybrid forms in narrow Heliconius hybrid zones [25].
Surprisingly, only three generations of crosses are needed to obtain a homozygous H. heurippa like color pattern [35]. Two tightly linked loci controlling the red forewing band (B allele, hereafter HmB) and the absence of brown forceps marks in the ventral surface (br allele) from linkage group 18 are introgressed into an H. c. cordula genetic background that includes the yellow forewing band allele at the HmN locus on linkage group 15. The resultant pattern contributes to reproductive isolation through assortative mating, and therefore plays a direct role in speciation [35]. In particular, mating experiments revealed strong pre-mating mating isolation between H. heurippa and H. melpomene (≈90%) and between H. heurippa and H. cydno (≈75%). Furthermore, assays with wing models showed that H. heurippa males use the combined red and yellow bands to discriminate females [35]. Even first-generation backcross hybrids between H. m. melpomene and H. c. cordula, resembling H. heurippa, showed a strong preference for their own color pattern over that of either parental species, implying that mate preference can be established directly through hybridization [38]. This result implies, first that assortative mating preferences would facilitate the initial establishment of a homozygous hybrid color pattern by increasing the probability that early generation hybrids mate among themselves. Second, once the new hybrid population was established, it would immediately possess the assortative mating preferences that generate partial reproductive isolation from the parental species. Thus, H. heurippa is an excellent candidate for speciation through adaptive introgression [38].
Despite extensive support for the hybrid origin of H. heurippa from biogeography, crosses, mate choice experiments and mathematical simulations [39], molecular evidence for a hybrid origin is inconclusive. Neutral markers reveal extensive gene flow between all three species, H. heurippa, H. melpomene and H. cydno [40]. To definitively test the hybrid origin hypothesis we need to study the loci controlling the adaptive traits responsible for speciation, in this case wing patterns [21], [40]. Here, we take advantage of the recent cloning of the HmB locus controlling the red forewing band of H. melpomene, in order to carry out such a test [41], [42].
Previous work on H. heurippa has used a panel of largely intronic markers and shown evidence for ongoing gene flow between H. heurippa and its close relatives [40]. Here, we directly addressed the adaptive introgression hypothesis by sampling 18 contigs based on 24 amplicons representing 18,629bp across the HmB locus for ten individuals of each species (60 alleles; Table S1) [41], [42]. In addition, we improved our broader genome sampling by developing a larger panel of unlinked molecular markers based on single copy genes with large exons. We analyzed around 15,000 contigs assembled from H. melpomene GSS (BAC-end) sequences, 484 of which showed strong homology with single copy genes in B. mori. From these unigenes, 27 had exons longer than 700 bp (Table S2). Additionally, we used two previously published markers, CAD and GAPDH [43]. Thus, we used a set of 29 markers that were putatively distributed in at least 17 of the 21 chromosomes in H. m. melpomene (Table S2). Sequences of these markers were obtained for the same ten individuals per species used in the HmB locus analysis (60 sequences per gene).
Among the 29 loci sampled from across the genome, most SNP polymorphisms were shared among the three species and therefore did not associate H. heurippa with one or other of the two parental species. Only 8 SNPs (over six genes) were fixed polymorphisms shared by H. heurippa and H. c. cordula relative to H. m. melpomene (Figure 1A), consistent with previous data from mitochondrial genes that related H. heurippa with H. cydno [40], while no SNPs were fixed in H. heurippa and H. m. melpomene relative to H. c. cordula (Figure 1A).
Our prediction derived from the adaptive introgression hypothesis was that within the HmB locus there should be a region introgressed from H. melpomene into the H. heurippa genome. This prediction was upheld. From nearly 19Kb of sequence analyzed across the HmB locus, there was a 6,493 bp region corresponding to the 3′ end of a putative kinesin gene (hereafter, 3′ kinesin) showing a strong association between H. heurippa and H. m. melpomene (Figure 1B). Across the remaining HmB region there was either shared variation among the three species, unique polymorphisms in one of the three species, or nearly fixed changes unique to H. heurippa. In the case of two genetic markers, kin_2 and sdp, H. heurippa was most strongly associated with H. c. cordula (Figure 1B).
The HmB locus was a significant outlier relative to the rest of the genome. We calculated a likelihood statistic that estimated the relationship of H. heurippa to the parental species, where positive values indicate sites linking H. heurippa with H. m. melpomene and negative values sites where H. heurippa is similar to H. c. cordula. The distribution of mean likelihood values across unlinked loci gives a distribution of expected values for the genome. Comparison of the 3′ kinesin region showed values that fell outside this distribution derived from unlinked markers (Figure 2; p<0.05), demonstrating that this genetic region has a significantly stronger association with H. melpomene than any other region of the H. heurippa genome. The kin_2 region was also an outlier, but with a significantly stronger association with H. cydno (Figure 2; p<0.05), implying that the kinesin gene is in fact a chimera derived from two parental species. As H. heurippa is most closely allied to H. c. cordula, we analyzed linkage disequilibrium (LD) for these two species combined. Across the HmB region, the highest r2 values (p<0.001) were observed within the 3′ kinesin and nearby genes, with LD decaying in surrounding regions, suggesting a haplotype structure across the kinesin gene resulting from strong selection on wing pattern (Figure 3). Nonetheless, consistent with the wing patterning alleles being relatively ancient, there was no evidence for a reduction in diversity or deviation from neutrality that might indicate a recent selective sweep at 3′ kinesin (Table 1). Similarly, there was no evidence for adaptive amino acid substitution at this locus, with Ka/Ks ratios not significantly greater than 1 (p>0.05; Figure S1) and McDonald-Kreitman tests showing no deviation from neutrality (p>0.05). H. heurippa also had five private amino acid substitutions not observed in the other two species (Figure S2). Only one amino acid replacement was shared between the red-banded species, H. heurippa and H. m. melpomene, representing a putative causative site for this phenotype (Figure S2).
An alternative hypothesis to adaptive introgression is that the H. heurippa pattern might be ancestral. However we could rule this out by reconstructing a rooted genealogy of the 3′ kinesin with either nucleotides or amino acid sequences (Figure 4B; data not shown for AA). Gene genealogies for kin_2, sdp and 3′ kinesin confirmed the results obtained with the SNP association tests (Figure 4). In the 3′ kinesin, H. heurippa was monophyletic and branched from within the H. m. melpomene clade, with H. c. cordula an outgroup to both species (Figure 4B, Table 2). In contrast, genomic regions surrounding 3′ kinesin showed H. heurippa alleles more closely related to H. c. cordula (Figure 4A and 4C; Table 2). Thus, the SNP analysis, LD patterns and gene genealogies revealed that a clearly delimited genomic portion of the HmB region closely relates H. heurippa with H. m. melpomene. This result is in contrast to the rest of the genome where no other gene showed a similar association.
When the genealogy of 3′ kinesin was used to estimate divergence times, we found that H. heurippa alleles were derived from H. m. melpomene around 0.43 Mya (0.12–0.84) ago, subsequent to the splitting of the H. c. cordula/H. m. melpomene alleles at 2.82 Mya (1.03–5.22) ago. A coalescent expansion model [44], [45] (SSD>0.05 in all the cases) similarly suggested that H. heurippa haplotypes radiated more recently (∼0.385 Mya; 0.176–1.428) than either H. m. melpomene or H. c. cordula (∼2.055 Mya (1.175–4.219) and ∼2.745 Mya (1.584–4.489) respectively), giving a confirmation of the relative ages of the alleles found in each species, independent of tree topology. Thus, the H. heurippa 3′ kinesin alleles diverged subsequent to the split between H. m. melpomene and H. c. cordula at this locus.
To examine the role of the putative kinesin gene in specification of wing pattern, we visualized spatial localization of kinesin transcripts in developing wings using in situ hybridization. In two red-banded forms H. melpomene cythera and H. melpomene rosina, a probe from exon 13 of the 3′ kinesin showed localization to the distal portion of the developing wing in early pupal stages (72–96 hrs after pupation; Figure 5A; Figure S3). No such spatial localization was seen either in individuals of H. cydno that do not express a red band phenotype (Figure 5B), nor in H. melpomene forewings treated with riboprobes for a different gene (Figure S3). This spatial localization suggests a model for pattern specification whereby the kinesin gene interacts with another as yet unidentified gene product to specify proximal and distal boundaries of the forewing band (Figure 5C), leading to upregulation of pigmentation genes such as cinnabar [46].
Homoploid hybrid speciation has been considered controversial in animals. We here provide the first molecular support for this hypothesis derived from sequence analysis of a gene region directly implicated in controlling a hybrid trait. H. heurippa was originally proposed as a hybrid species based on its unusual color pattern. The main evidence in support of this hypothesis are crossing experiments demonstrating experimental introgression of the H. m. melpomene red color forewing band into the H. c. cordula genomic background [35]. Such experiments demonstrate a plausible route for the origin of H. heurippa, and make a clear prediction: the region controlling the red forewing band should show a pattern of introgression from H. melpomene into H. heurippa. Here, we provide support for this hypothesis at a molecular level, by demonstrating a 6.5Kb region in the HmB locus that is introgressed from H. melpomene into H. heurippa.
The majority of SNPs (634) sampled in 29 coding genes, located on 17 of the 21 H. melpomene linkage groups, showed shared polymorphism among the three species (Figure 1A). H. heurippa and H. c. cordula shared fixed polymorphism relative to H. m. melpomene at only 8 SNPs, and there were no fixed SNPs in H. heurippa and H. m. melpomene relative to H. c. cordula (Figure 1A). This agrees with previous genetic data showing extensive allele sharing in the nuclear genome between the three species, but H. heurippa somewhat closer to H. c. cordula [40]. As we have argued previously [40], these data do not strongly support a hybrid speciation scenario, but are more consistent with either recent gene flow among the three species or shared ancestral polymorphism.
Here we have taken advantage of the recent cloning of HmB, the key locus underlying the speciation of H. heurippa. Our hypothesis derived from previous crossing experiments and sequence surveys is that the H. heurippa genome is most closely related to H. cydno, but with the introgression of the red forewing band, controlled by HmB, from H. melpomene. Here we directly test this hypothesis by sampling markers across the 721 Kb HmB locus [41]. From 13 genes evaluated in this region (comprising 24 molecular markers), we found a 6,493 bp region, corresponding to the 3′ end of the kinesin locus, where H. heurippa is strongly related to H. melpomene (Figure 1B). The likelihood values for species relationships at this locus differs significantly from that seen among unlinked genes, implying that this relationship cannot be explained by chance (Figure 2). The high long-range LD at 3′ kinesin relative to the H. c. cordula genetic background is also expected under the introgression hypothesis (Figure 3). The pattern is comparable to that seen across the same region in a Heliconius melpomene hybrid zone, where long-range LD is observed between sites showing significant genotype-by-phenotype association [41]. A rather surprising observation is that across the HmB locus there is also shared variation between the three species, at sites interspersed between those generating a strong phylogenetic signal. These could be due to gene flow and recombination subsequent to speciation, recurrent mutations or alternatively a hybrid founding event for H. heurippa that transferred significant polymorphism from the parents to the hybrid species. In the data, there was no marked difference in the transition/transversion ratio among fixed and shared polymorphisms, which might be indicative of recurrent mutation (data not shown).
An alternative hypothesis to be considered is that H. heurippa pattern might be ancestral and have given rise to H. melpomene and H. cydno lineages that inherited different aspects of the ancestral wing pattern. However, this is not supported by the rooted gene genealogy for the 3′ kinesin that shows H. heurippa monophyletic, forming a well supported and derived clade within H. melpomene (Figure 4B). Furthermore, none of the dating approaches showed H. heurippa older than the other two species. Other genomic regions have shown a genealogical pattern whereby H. heurippa was similarly nested within an H. cydno clade, also arguing that H. heurippa is not an ancestral taxon [47].
In addition, kinesin in situ hybridizations on developing wing tissue (72–96h post-pupation) showed localized gene expression in the distal region of the wing, supporting a likely functional role in specification of the proximal boundary of the forewing band. In combination with previous analyses showing parallel differences in expression levels of the kinesin gene between color pattern races of both H. melpomene and H. erato [41], [48], these data strongly suggest that a regulatory change in the kinesin gene is functionally required for pattern determination. The kinesin gene appears to be chimeric, with the 3′ region derived from H. m. melpomene and the 5′ end more strongly related to H. c. cordula. Since crossing experiments suggest that introgression of the HmB allele from H. melpomene is sufficient to generate the H. heurippa pattern, the implication is that the functionally important sites are located at the 3′ end of the gene. We have identified one amino acid change and 11 synonymous changes shared between red-banded H. melpomene and H. heurippa, representing candidate functional sites for red band specification.
We also observed five amino acid differences between H. melpomene versus H. heurippa (Figure S2), perhaps reflecting adaptive change subsequent to formation of H. heurippa, although there was no significant evidence for selection on the locus (Figure S1). Perhaps more likely, these changes may represent fixation of nearly-neutral variation due to a population bottleneck during the origin of H. heurippa.
The implication of this gene in phenotypic control at HmB is also consistent with previous population genetic analysis of phenotypic races of H. melpomene, which showed a region of high genetic differentiation corresponding to a genomic region including the kinesin [41]. Members of the kinesin superfamily (KIFs) are key players in cellular functioning and morphology that interact with cargo molecules such as proteins, lipids or nucleic acids [49], [50]. In both vertebrates and invertebrates, kinesin molecules are implicated in pigment transport, however in Heliconius melpomene upregulation of pigment pathway genes occurs later in development relative to the localized kinesin expression observed here. This would suggest a likely upstream role in scale cell fate specification, rather than pigmentation per se.
Although adaptive introgression has recently been demonstrated at a molecular level, for example between species in the genus Senecio [19], H. heurippa is unusual in the fact that the hybrid trait contributes directly to reproductive isolation and hence speciation. We have also recently demonstrated that first generation backcross hybrids resembling H. heurippa also exhibit mate preferences very similar to that of wild H. heurippa. This implies that mate preferences could also have been produced by introgression in addition to color pattern [38]. A possible mechanism for this is suggested by the recent demonstration of a genetic association between the red band and male preference for red mates, in interspecific hybrids between H. m. rosina and H. c. chioneus (Merrill et al. pers. comm.). Thus, the derived color pattern and mate preferences of H. heurippa could potentially have arisen from introgression of the same gene or tightly linked genes.
Several cases of animal homoploid hybrid species have been recently proposed, such as Rhagoletis sp., Lycaeides sp., Cottus gobio group, cichlid fishes, Xiphophorus clemensiae and Pogonomyrmex sp. [17], [51] where ecological divergence, sexual selection or both promote reproductive isolation of a hybrid taxon. However, to our knowledge, this is the first time that molecular evidence for introgression has been established for an adaptive trait that also contributes directly to reproductive isolation and hence speciation. We feel that our results therefore represent the most convincing molecular evidence to date for homoploid hybrid speciation in animals. Similar molecular evidence also supports the hybrid origin of sunflower species in the genus Helianthus, although the pattern is very different in this case. Hybrid sunflower genomes are a mosaic of genomic blocks inherited from one or other parent [23], in contrast to H. heurippa which shares polymorphism with both parental species across most of the genome. Although mathematical simulation has suggested that the origin of H. heurippa probably involved an initial period of allopatry, during which the hybrid pattern became established [39], the contemporary genetic pattern supports our model of ‘hybrid trait speciation’ whereby localized introgression of key traits can promote the origin of hybrid species [21].
Butterflies were collected from Colombia and Venezuela: H. m. melpomene in Morcote (5°37′0.52″N; 72°18′0″W, Casanare-Colombia) and Chirajara (4°12′48″ N; 73°47′70″W, Cundinamarca-Colombia), H. c. cordula in San Cristobal (7°47′56″N; 72°11′56″W, Merida-Venezuela) and H. heurippa in Buenavista (4°10′30″N; 73°40′41″W, Meta –Colombia). Wings of 10 individuals of each species were removed and stored in glassine envelopes and are lodged in the Natural History Museum of the Universidad de los Andes. The bodies were preserved in 20% DMSO-0.25M EDTA salt saturated solution. DNA was isolated with DNeasy Blood & Tissue Kit (QIAGEN) following manufacturer's instructions. Quality of genomic DNA was confirmed by visualisation in a 0.8% agarose gel.
In order to confirm the species relationships, genealogical topologies were reconstructed for three fragments within the HmB region, rooted for the 3′ kinesin (6,493 pb) using H. numata as outgroup and unrooted for 5′ kinesin partial sequence (kin_2; 1,100 pb) and sorting nexin (sdp; 402 pb) (Figure 4). Maximum Parsimony analysis was carried out in PAUP*v4.0b10 [65] using a heuristic search with TBR branch swapping; bootstrap values were calculated with 5,000 replicates using the same search conditions. Modeltest v3.7 [66] was used to determine the most appropriate model for nucleotide substitution based on corrected Akaike information criterion (AICc). For the 3′ kinesin data set Modeltest identified the HKY+I+G model, for 5′ kinesin the K81uf+I+G and for nexin the K80+I. Likelihood reconstructions were also made in PAUP*v4.0b10 [65] based on selected evolutionary models. Heuristic search and bootstrapping were carried out as for parsimony.
The 3′ kinesin genealogy was used to enforce a molecular clock hypothesis. When the likelihoods were compared, constant rate evolution was rejected (x2 = 96.92, df = 48; P<0.001). Then a Bayesian framework, implemented in BEAST v1.4.8 [67], was employed to obtain an approximate time for the 3′ kinesin introgression. We applied the HKY+I+G model of evolution with four rate categories and assumed a relaxed lognormal clock. Based on the calibration proposed by Wahlberg et al. for Nymphalidae, with Heliconius and Eueides diverged from their common ancestor 18.4 Mya [68]. This date was used as a prior for a probabilistic calibration to determine the splitting time between H. cydno and H. melpomene alleles and between H. heurippa and H. melpomene alleles. The rest of the parameters were sampled keeping the default prior distributions. Two independent runs were implemented, with 50 million steps and burn-ins of 5,000,000. Tracer v1.4 was used to combine runs and observe parameter convergence [69]. Divergence time standard deviations were calculated from 95% confidence/credibility intervals using a normal approximation. We also computed the time of 3′ kinesin introgression under the assumption of species expansion. To perform this calculation, we first tested the fit of the observed mismatch distribution to the theoretical expectation as implemented in Arlequin v. 3.0 [70]. The calculations were made with a neutral mutation rate of ∼2.99×10−10 per base per generation for this region and 10 generations per year.
kinesin RNA in situ hybridizations were performed on H. m. cythera, H. m. rosina and H. cydno 72 to 96h pupal forewings. The specific races involved in the rest of the study were not available as live tissue for this analysis. A 303bp region of exon 13 in the H. melpomene kinesin gene was cloned into the vector pSPT19 (linearised with NheI). RNA probes were prepared with the DIG RNA labeling kit (SP6/T7) (Roche, Cat. 11 175 025 910) according to the manufacturer's instructions. Tissue fixation and in situ hybridization were carried out following a procedure modified from Ramos and Monteiro, 2007 [71].
|
10.1371/journal.ppat.1001003 | Activation of Akt Signaling Reduces the Prevalence and Intensity of Malaria Parasite Infection and Lifespan in Anopheles stephensi Mosquitoes | Malaria (Plasmodium spp.) kills nearly one million people annually and this number will likely increase as drug and insecticide resistance reduces the effectiveness of current control strategies. The most important human malaria parasite, Plasmodium falciparum, undergoes a complex developmental cycle in the mosquito that takes approximately two weeks and begins with the invasion of the mosquito midgut. Here, we demonstrate that increased Akt signaling in the mosquito midgut disrupts parasite development and concurrently reduces the duration that mosquitoes are infective to humans. Specifically, we found that increased Akt signaling in the midgut of heterozygous Anopheles stephensi reduced the number of infected mosquitoes by 60–99%. Of those mosquitoes that were infected, we observed a 75–99% reduction in parasite load. In homozygous mosquitoes with increased Akt signaling parasite infection was completely blocked. The increase in midgut-specific Akt signaling also led to an 18–20% reduction in the average mosquito lifespan. Thus, activation of Akt signaling reduced the number of infected mosquitoes, the number of malaria parasites per infected mosquito, and the duration of mosquito infectivity.
| For malaria transmission to occur, a mosquito must ingest and harbor the parasites for approximately two weeks while the parasites complete multiple developmental stages. Until development is complete and the malaria parasites invade the mosquito salivary glands, transmission to another host cannot occur. Upon completion of parasite development, transmission is possible with every subsequent bite. In this study we demonstrate that tissue-specific overexpression of a single activated protein kinase that is essential to insulin signaling in the mosquito can dramatically reduce parasite development. This kinase – Akt – has been described as a critical cell signaling node that regulates a range of physiological processes. In addition to the impact on parasite development, increased Akt signaling also reduced the average mosquito lifespan relative to controls, thereby limiting the window of opportunity for successful parasite transmission. Thus, we demonstrate that genetic manipulation of one key signaling protein directly reduces parasite development in the insect vector as well as the duration of mosquito infectivity.
| Malaria is one of the world's most severe public health concerns, killing nearly one million people annually [1]. The disease is caused by infection with parasites of the genus Plasmodium that are transmitted by female anopheline mosquitoes. Shortly after an infective bloodmeal is consumed by the mosquito, motile ookinetes develop and attempt to invade the mosquito midgut. Ookinetes that successfully traverse the midgut epithelium form non-motile oocysts and develop on the midgut for a minimum of 12 days before rupturing and releasing sporozoites capable of invading the salivary glands. Following salivary gland invasion by sporozoites, and within 16 days after ingestion of an infectious bloodmeal, the mosquito becomes infective to humans and remains so for the duration of its life. Midgut invasion by the parasite is highly risky and a majority of the parasites perish before developing into oocysts [2], [3]. Further, Anopheles stephensi mosquitoes – the leading vector of malaria in India, parts of Asia and the Middle East and the focus of our work – rarely survive more than two weeks in the field [4]–[6]. These observations suggest that only the oldest mosquitoes in a population are capable of transmitting malaria and that even a modest reduction in lifespan could significantly impact parasite transmission.
The insulin/insulin-like growth factor 1 signaling (IIS) cascade plays a critical role in the regulation of innate immunity and lifespan in a wide range of vertebrate and invertebrate organisms [7], [8]. IIS is initiated through the binding of insulin-like peptides (ILPs) to the insulin receptor, leading to a series of downstream phosphorylation events that include the key signaling protein Akt. Activation of IIS results in translocation of Akt to the cell membrane where it is phosphorylated and activated by phosphoinositide-dependent kinase-1 (PDK1). Activated Akt then phosphorylates the forkhead transcription factor FOXO1, preventing it from entering the nucleus and activating transcription of target genes [9].
In model invertebrates, the IIS cascade has been linked to both innate immunity and lifespan regulation. In the nematode Caenorhabditis elegans, disruption of the insulin receptor orthologue daf-2 leads to decreased IIS, extension of lifespan [10] and increased resistance to bacterial infection [11]. In contrast, loss of function mutations in the FOXO1 orthologue daf-16 result in nematodes that are sensitive to infection [11] and short-lived [12]. As in C. elegans, disruption of the IIS can lead to lifespan extension in the fruit fly Drosophila melanogaster [13]–[15]. Recent work has also demonstrated that activation of the Toll cascade, a key pathway in fly immunity, inhibits IIS in the fly [16]. These observations confirm that the connections observed in humans between innate immunity, metabolism and aging are evolutionarily conserved (reviewed in [17]).
Lifespan extension due to IIS disruption is tissue-dependant, although the tissues involved can vary within and across genera. In C. elegans [18] and D. melanogaster [15], the nervous system is a key IIS center. In D. melanogaster, disruption of IIS in the fat body can also lead to lifespan extension [15]. Overexpression of the transcription factor daf-16 in the C. elegans intestine extends lifespan [19]. Our previous work with A. stephensi suggests that the analogous mosquito tissue – the midgut – is also a center of IIS. In particular, we have shown that ingested human insulin can activate IIS in midgut epithelial cells and significantly decrease the lifespan of A. stephensi mosquitoes [20], implying a direct relationship between exogenous insulin from the mammalian bloodmeal, activation of the midgut IIS, and lifespan. Therefore, we predicted that genetic manipulation of key IIS components in the midgut would offer a unique strategy for disrupting P. falciparum development while simultaneously decreasing the lifespan of the mosquito below the extrinsic incubation period (EIP) or the time required for malaria parasite development.
We genetically engineered A. stephensi to express an active variant of the mosquito Akt under the control of the midgut-specific carboxypeptidase (CP) promoter. As predicted, increased Akt signaling in the midgut significantly reduced malaria parasite development and mosquito lifespan. Both the number of infected mosquitoes and the average number of parasites per mosquito were reduced in transgenic mosquitoes relative to controls. In addition, transgenic mosquitoes had significantly shorter lifespans than non-transgenic siblings reared under identical conditions. These results demonstrate that manipulation of one signaling protein, Akt, in the mosquito midgut can affect both mosquito innate immunity and lifespan.
We generated a transgenic A. stephensi line overexpressing an activated form of Akt under the control of the midgut-specific CP promoter. Activated Akt was generated by a myristoylation sequence encoded at the amino terminus. An HA epitope at the carboxy terminus (myr-AsteAkt-HA) facilitated protein identification. The construct was inserted into the pBac[3XP3-DsRedafm] plasmid vector [21] for transformation into the A. stephensi genome (Fig. 1A). We injected approximately 4400 embryos with a mixture of the pBac[3XP3-DsRedafm]CP-myr-AsteAkt-HA donor plasmid and the phsp-pBac helper plasmid, resulting in approximately 176 adult mosquitoes whose progeny were screened for DsRed eye fluorescence (Fig. 1B). We isolated three F1 progeny with stable DsRed eye fluorescence, from which we established a stable line (Fig. 1C). Transgenic mosquitoes were maintained as a heterozygous line by outcrossing the mosquitoes in each generation to non-transgenic colony A. stephensi. A homozygous line was generated after approximately 20 generations of outcrossing and used to verify the viability of homozygous mosquitoes and to test the effect of increased Akt signaling on P. falciparum development in the midgut.
We assessed myr-AsteAkt-HA transcript expression levels during mosquito development and found that myr-AsteAkt-HA is primarily expressed in pupae and adult female stages (Figure S1, Text S1). In adult females, the myr-AsteAkt-HA transcript and protein were only detected within the midgut of transgenic mosquitoes (Figs. 1D and 1E). No expression was observed in the carcass of transgenic mosquitoes or in the midgut or carcass of non-transgenic mosquitoes. Surprisingly, we observed high levels of transcript (Fig. 2A) and protein (Fig. 2B) in the midguts of both non-bloodfed and bloodfed transgenic mosquitoes. Transcript expression increased slightly 2–6 h after bloodfeeding and increased dramatically between 24–48 h after the bloodmeal (Fig. 2A). Protein expression increased 2–12 h after the bloodmeal as would be expected for the CP promoter, but was reduced during the latter half of the reproductive cycle (24–48 h) (Figs. 2B and 2C).
The myristoylation sequence at the amino terminus was expected to target myr-AsteAkt-HA to the cell membrane to be phosphorylated and activated by PDK1, eliminating the need for Akt binding to the upstream IIS component phosphoinositide (3,4,5)-trisphosphate and endogenous insulin signaling in general. To assess membrane localization of myr-AsteAkt-HA, we performed immunocytochemistry on both midgut sections (Fig. 3A) and whole midguts (Fig. 3B) of transgenic and non-transgenic mosquitoes using an anti-HA-fluorescein antibody. Strong staining of midgut epithelial cells was observed only in transgenic mosquitoes and no expression was observed in non-transgenic mosquitoes. A majority of the staining in midguts from transgenic mosquitoes was localized to the cell membrane as expected with the myristoylation sequence (Fig. 3A and 3B – white arrows). To confirm this result, we isolated the nuclei, cell membranes, and cytoplasm from midgut epithelia of transgenic mosquitoes and compared transgene protein levels in these fractions. The transgene protein was detected only in the cell membrane fraction at levels similar to those observed in an intact midgut (Fig. 3C).
FOXO1 is key transcription factor in the IIS cascade that is directly phosphorylated by Akt. Human insulin induced FOXO1 phosphorylation in the midguts of bloodfed, non-transgenic A. stephensi (Fig. 4A). In CP-myr-AsteAkt-HA-expressing mosquitoes, we also observed a marked increase in midgut FOXO1 phosphorylation relative to non-transgenic sibling mosquitoes even though a bloodmeal was not provided (Fig. 4B). This indicates that myr-AsteAkt-HA is active and capable of phosphorylating downstream IIS effectors. In sum, both human insulin and myr-AsteAkt-HA induced FOXO1 phosphorylation in vivo.
Increased Akt activity in the midgut epithelium led to major reductions in both the percentage of mosquitoes infected with P. falciparum and the number of oocysts in the midguts of infected mosquitoes (Fig. 5). The percentage of mosquitoes with one or more oocysts decreased from an average of 58.5% (36–86%) in non-transgenic controls to 10.5% (2–14%) in heterozyogous myr-AsteAkt mosquitoes (Fig. 5A, p<0.0001; pooled across replicates). Similarly, the intensity of infection was reduced by 95.6% from an average of 3.9 oocysts/midgut (0–45; n = 200) in non-transgenic controls to 0.18 (0–6; n = 200) in myr-AsteAkt mosquitoes (Fig. 5B). This rate of inhibition is higher than rates reported for other anti-parasite effector molecules, including SM1 (81.6%), PLA2 (87%), anti-HAP2 (81.1%), and anti-chitinase (91.3%) [22]–[25].
We also assessed the effect of doubling myr-AsteAkt expression by establishing a homozygous transgenic line. P. falciparum infection in the homozygous line was completely blocked, with no viable oocysts observed in any of the midguts (Fig. 6; n = 90). In contrast, 62% of control mosquitoes had at least one oocyst, with an average of 6.6 parasites per midgut (0–76; n = 150). A recent study demonstrated that the combination of two effector molecules, defensin A and cercropin A, was capable of completely blocking the development of the avian malaria parasite Plasmodium gallinaceum in Aedes aegypti [26]. However, our data constitute the first example of a single effector molecule in a transgenic mosquito completely blocking invasion by the human malaria parasite.
We hypothesized that increased activation of IIS due to expression of myr-AsteAkt-HA in the midgut would alter the lifespan of sugarfed and bloodfed mosquitoes relative to non-transgenic controls. In contrast to bloodfeeding, mosquitoes provided with sugar only do not enter a reproductive cycle or produce eggs. In replicated assays, sugarfed transgenic mosquitoes lived an average of 18.85 (17.16–20.29) days compared to 23.02 (22.17–24.04) days for non-transgenic control siblings, a decrease of nearly 20% (Fig. 7A; p<0.0001 to p = 0.0001). This same trend was observed in transgenic mosquitoes provided with weekly bloodmeals and given the opportunity to produce a weekly clutch of eggs. Bloodfed transgenic mosquitoes survived an average of 17.44 (16.54–19.24) days compared with 21.32 (17.94–25.75) days for non-transgenic control siblings, a reduction of more than 18% (Fig. 7B; p = 0.0058 to 0.0482).
An important measure for malaria control is the percent of the population that survives long enough to transmit the malaria parasite. If one assumes that (1) a female mosquito finds a mate on the first day after adult eclosion, (2) she acquires an infective bloodmeal on the second day, and (3) the parasite develops and invades the salivary glands 14 days after taking the infected bloodmeal, then that mosquito must survive a minimum of 16 days to successfully transmit P. falciparum (blue areas Figs. 7A and 7B). Under our conditions, an average of 59% of the non-transgenic mosquitoes given weekly bloodmeals were still alive at day 16 compared to 44% of the myr-AsteAkt transgenic mosquitoes. Comparing the area under the lifespan curves of the transgenic and non-transgenic siblings after 16 days, we observed a 53% reduction in sugarfed mosquitoes and a 48% reduction in bloodfed mosquitoes. This indicates that the population of competent malaria vectors can be reduced by half with a modest 20% reduction in lifespan.
Fitness costs due to the generation of the transgenic line or the transgene itself were likely minimal due to transgene insertion into non-coding sequence (Figure S2, Text S1) and repeated out-crossing to non-transgenic mosquitoes. In addition, lifespan studies were performed using sibling transgenic and non-transgenic mosquitoes to minimize genetic differences and were performed with sibling mosquitoes reared as larvae in the same pans of water and separated as pupae to minimize environmental differences.
Insulin signaling regulates reproduction in a wide range of organisms. In insects, including mosquitoes, IIS has been shown to regulate steroidogenesis in the ovaries and vitellogenesis in the fat body [27], [28]. Although IIS in the midgut has not previously been implicated in the regulation of reproduction, we examined whether any differences in egg production between the transgenic and non-transgenic siblings could be detected. For the five replicates in which zero egg counts were recorded, there were no significant differences between transgenic and non-transgenic females in the number of eggs laid (Table 1). Among those that laid eggs, only one replicate of the six indicated a significant difference between non-transgenic and transgenic females. The remaining replicates indicated no difference between genotypes (Table 1). There was no difference between genotypes in whether or not females laid eggs.
To ensure that differences in egg production were not due to the size of the blood meal, the amount of blood ingested was also compared between transgenic and non-transgenic females. For all five replicates, there was no significant difference between genotypes in the amount of blood ingested (Figure S3A and S3B, Text S1). In addition, no obvious differences were observed between transgenic and non-transgenic sibling mosquitoes in the amount of undigested BSA remaining at 24 h after blood ingestion (Figure S3C, Text S1).
Mosquitoes require a bloodmeal to initiate a reproductive cycle and produce eggs. Within this bloodmeal are insulin, insulin-like growth factor 1, and various other factors that circulate in the blood of the human host. Our previous work demonstrated that some of these factors, including human insulin and human TGF-β1, activate mitogen-activated protein (MAP) kinase and phosphoinositide-3 kinase (PI3K) signaling cascades in the mosquito midgut [20], [29]. Here, we used transgenesis to overexpress a key component of the IIS cascade, Akt, in the A. stephensi midgut to induce signaling independent of exogenous insulin. We observed significant reductions in both the prevalence and intensity of P. falciparum infections in transgenic mosquitoes following the consumption of an infective bloodmeal. We also observed a reduction in lifespan consistent with that observed in insulin-fed A. stephensi [20], indicating that the mosquito midgut plays a central role in regulating lifespan.
Myristoylated Akt localized to the midgut epithelial cell membrane in transgenic A. stephensi (Fig. 3) where it was activated to subsequently phosphorylate the downstream effector protein FOXO1 (Fig. 4C). This parallels FOXO1 phosphorylation in the midguts of mosquitoes fed bloodmeals containing insulin (Fig 4A). Taken together, these results suggest that the mechanisms of parasite and lifespan reduction observed in CP-myr-AsteAkt-HA transgenic mosquitoes are dependent on the activation of the PI3K/Akt/FOXO arm of the IIS cascade. It is noteworthy that Akt has been defined as “a critical signaling node within all cells of higher eukaryotes and one of the most important and versatile protein kinases at the core of human physiology and disease [30].” Akt has more than 100 experimentally verified substrates and broad crosstalk between a variety of biologically important signal transduction pathways. Thus, the mechanisms through which tissue-specific Akt overexpression regulates innate immunity and lifespan are likely to be complex [30].
A carboxypeptidase promoter drives the myr-AsteAkt-HA transgene, so we expected expression to rise shortly after a bloodmeal was consumed and to be midgut-specific. Expression of myr-AsteAkt-HA was indeed specific to the midgut (Fig. 1D and E), but the timing of expression was unexpected since both transcript and protein were observed even in the absence of a bloodmeal (Fig. 2). As expected for a gene regulated by a carboxypeptidase promoter, however, protein expression increased following ingestion of the bloodmeal. Leaky transgene expression has been observed with this promoter, resulting in expression prior to bloodfeeding [31] or late in the reproductive cycle [32]. The process of generating a transgenic mosquito strain could also explain the unexpected expression patterns. For example, the transgene may have inserted near an enhancer DNA sequence, resulting in greater gene and protein expression [33]. Although this pattern of myr-AsteAkt-HA expression was unexpected, it was ultimately advantageous because increased insulin signaling is maintained for the apparent duration of adult female life and does not depend on consumption of a bloodmeal for activity. Thus, the anti-parasite activity and lifespan effects of myr-AsteAkt-HA will occur regardless of the timing and quantity of bloodmeals that are consumed by a transgenic mosquito.
Increased insulin signaling in the mosquito midgut, whether through ingestion of exogenous insulin [20] or overexpression of active IIS proteins such as Akt, can significantly reduce mosquito lifespan and inhibit P. falciparum development. Importantly, we observed that increased AsteAkt expression in the homozygous line increased parasite resistance to the point that oocyst formation on the midgut was completely blocked. Although it will likely be necessary to deploy heterozygous mosquitoes for any future transmission blocking strategy, our data suggest that an increase in myr-AsteAkt expression, possibly through manipulation of the promoter or transgene insertion site, could yield heterozygous mosquitoes that are resistant to P. falcipaurm infection. Lifespan reduction can also impact malaria parasite prevalence based on the combined effects of a relatively short natural lifespan of A. stephensi [4]–[6] and a relatively lengthy parasite development time. In particular, models of vector competence routinely demonstrate that the daily probability of survival is the single most important factor in determining how effectively a mosquito transmits a pathogen [34]. All else being equal, even modest reductions in lifespan will have significant effects on disease transmission.
In summary, we have developed a novel mechanism to reduce the transmission of the human malaria parasite P. falciparum. This approach is based on the manipulation of two key physiological parameters, lifespan and innate immunity, through activation of a single signaling protein, Akt. Increased Akt activity significantly reduced infection prevalence in the mosquito host at the same time that it reduced the infective period of the mosquito lifespan. A multi-component approach to transgenesis focused on manipulation of the IIS cascade could be combined with overexpression of additional anti-parasite effectors to effectively block parasite transmission, reduce lifespan, and increase fecundity. Perhaps more importantly, a multi-component approach could prevent the escape of adaptive parasite variants, providing a powerful new tool for malaria control.
Anopheles stephensi mosquitoes were maintained at 28°C, 75% RH, on a 16∶8 light∶dark photoperiod. Larval mosquitoes were fed cat food pellets (Purina). Adult mosquitoes were fed ad libitum on a 10% dextrose solution. Porcine blood (UA Meat Science facility) supplemented with sodium citrate (0.38%) and warmed to 37°C was used for colony maintenance and bloodfeeding experiments. For feeding experiments, engorged females were separated from unfed and partially fed mosquitoes and maintained on 10% dextrose until needed. Females used for post-oviposition studies were transferred 48 h post-blood meal to a new container and allowed to oviposit on moistened filter papers overnight.
The A. gambiae carboxypeptidase (CP) promoter was kindly provided by Dr. Luciano Moreira [32]. The 5′ promoter was amplified with primers designed to remove the signal peptide, start methionine and Kozak consensus sequence, and to add XhoI and NotI restriction digest sites. The modified 5′ CP promoter was ligated into the phsp-pBac shuttle plasmid using XhoI and NotI sites [21]. The SV40 3′ UTR was ligated into the EcoRI site of phsp-pBac. A Kozak consensus sequence (CCAACCATGG) and Src myristoylation sequence (MGSSKSKPKDPSQR) were added to the 5′ end of AsteAkt, while an HA epitope (YPYDVPDYA) was added to the 3′ end. This construct was inserted it into the phsp-pBac shuttle containing the CP promoter and SV40 3′ UTR. Finally, the CP-myr-AsteAkt-HA-SV40 construct was ligated into the pBac[3XP3-DsRedafm] construct to generate the pBac[3XP3-DsRedafm]CP-myr-AsteAkt-HA plasmid for injection into A. stephensi embryos.
Donor (500 ng/µl) and helper (200 ng/µl) plasmids were injected into newly oviposited embryos, which were then reared to adulthood and screened for transgene insertion as described by Lobo et al [35]. Lifespan and reproduction experiments were initiated only after five generations of outcrossing the transgenic line to a non-transgenic lab strain. Crosses between heterozygous transgenic and non-transgenic mosquitoes produced a 50/50 ratio of transgenic to non-transgenic siblings.
Midguts and carcasses (whole body without midgut) were collected from ten transgenic females prior to bloodfeeding and at 2, 6, 12, 24, 48, and 72 h PBM. Total RNA was extracted using RNeasy kit (Qiagen), treated with DNase 1 (Fermentas) and cDNA was synthesized using High Capacity cDNA ReverseTranscription Kit (Applied Biosystems) with random hexamer primers. Quantitative real-time PCR (qRT-PCR) was performed using Maxima SYBP Green/ROX qPCR master mix (Fermentas) and an ABI 7300 real-time PCR system. Myr-AsteAKT-HA-specific primers (forward: 5′-TTACCGGTGAAAGTGTGGAGCTGA-3′; reverse: 5′-AAGCGTAATCTGGCACATCGTATGG-3′; efficiency - 98%) were used to detect myr-AsteAKT-HA in midguts and carcasses. Myr-AsteAKT-HA expression was normalized to ribosomal protein S7 expression. qRT-PCR reactions were performed in triplicate and the experiment was replicated twice with separate cohorts of mosquitoes.
Immunoblots were performed with one midgut equivalent of protein as previously described [36]. Myr-AsteAkt-HA protein levels were detected using an anti-HA antibody (1∶20,000 dilution; Roche). RT-PCR and immunoblot assays were replicated three times with separate cohorts of mosquitoes.
A total of 100 3- to 5-day-old female A. stephensi mosquitoes were fed artificial bloodmeals supplemented with 1.7×10−3 µmol of human insulin or an equivalent volume of insulin buffer as described in Kang et al [20]. Immunoblot analyses of protein phosphorylation from 60 midguts per treatment group were conducted as previously described [20]. Midgut samples were probed with anti-phospho FOXO1A/FOXO3A antibody (1∶1000 dilution, Millipore) or an anti-GADPH antibody (1∶10,000 dilution, Abcam) to assess protein loading.
In the CP-myr-AsteAkt transgenic mosquitoes midguts were subjected to immunoblot analysis as described above, and were probed with anti-phospho-FOXO1A antibody (1∶10,000 dilution; Millipore). Five midgut equivalents of protein were used per lane. Blots were stripped and re-probed with an anti-GADPH antibody (1∶40,000 dilution, CST) to assess protein loading.
For whole mount immunocytochemistry studies, midguts were dissected from 10 transgenic and 10 non-transgenic mosquitoes in 1× Aedes saline (125 mM NaCl, 5mM KCl, 1.85 mM CaCl2, pH 6.5) and opened into a midgut sheet. Immunocytochemistry was performed as described by Riehle and Brown [37], except that an anti-HA antibody conjugated to fluorescein (1∶1000, Roche) was used without a secondary antibody. All samples were imaged at identical settings to facilitate comparison. Experiments were replicated a minimum of three times with separate cohorts of mosquitoes. For immunocytochemistry using paraffin embedded sections, midguts were dissected from 10 transgenic and 10 non-transgenic mosquitoes in 1× Aedes saline and 10× Complete protease inhibitors (Roche). Midguts were immediately transferred to 4% paraformaldehyde in PBS for 2 h at RT and then stored in 70% EtOH at 4°C until embedded. The midguts were embedded in paraffin at the University of Arizona histology center and cut to obtain 4.5–5 µM sections. The paraffin was removed by two xylene washes of 10 min and the samples were hydrated through a series of solutions of decreasing ethanol concentration (100, 95, 70, 50 and 30%). The tissues were washed in PBS with 0.1% Tween 20 (PBS-T) and then blocked in a solution of 2% BSA/PBS-T for 2 h at RT. The slides were incubated overnight in a humid chamber with a 1∶500 dilution of the anti-HA antibody conjugated to fluorescein. The tissues were washed 3× in PBS-T for 15 min at RT and observed under a Nikon Eclipse E600 fluorescent microscope. Images were acquired using a SPOT camera system (Diagnostic Instruments Inc) at identical settings for all fluorescent images.
To verify the subcellular localization of myr-AsteAkt-HA, we prepared midgut cell membranes, nuclei and cytoplasm from midguts from transgenic and non-transgenic A. stephensi as described by Brown et al [38]. The three sub-cellular fractions were subjected to immunoblot analysis using the anti-HA antibody as described above with replicated samples from three separate cohorts of mosquitoes.
Cultures of P. falciparum NF54 were initiated at 1% parasitemia in 10% heat-inactivated human serum, and 6% washed human RBCs in RPMI 1640 with HEPES (Gibco) and hypoxanthine. Stage V gametocytes were evident by day 15 and exflagellation was evaluated on the day prior to and the day of mosquito feeding. For our assays, 5-day old female transgenic and non-transgenic A. stephensi were fed on mature gametocyte culture diluted with human erythrocytes and heat-inactivated serum. On day 10, midguts from fully gravid females were dissected in PBS and stained with 1% mercurochrome/PBS to visualize P. falciparum oocysts. Oocysts were counted for each midgut and mean oocysts per midgut (infection intensity) and percentages of infected mosquitoes (infection prevalence; infection = at least one oocyst) were calculated from all dissected mosquitoes.
Transgenic mosquitoes heterozygous for the CP-myr-AsteAkt-HA construct were mated with non-transgenic mosquitoes to generate 50% transgenic and 50% non-transgenic mosquitoes. The resulting larvae were reared together under identical conditions and separated based on DsRed fluorescence in the eyes of pupae under a fluorescent stereomicroscope. Female mosquitoes were separated into four treatment groups: transgenic bloodfed, transgenic sugarfed, non-transgenic bloodfed, and non-transgenic sugarfed. Bloodfed mosquitoes were given weekly bloodmeals throughout their entire adult life in addition to 10% dextrose ad libitum, while sugarfed mosquitoes were only provided 10% dextrose ad libitum. Daily mortality for each treatment was recorded and dead mosquitoes were removed until all mosquitoes had perished. These experiments were replicated twice. A third experiment was conducted using approximately 500 mosquitoes per treatment to verify the initial results.
Transgenic CP-myr-AsteAkt-HA females and their non-transgenic siblings were mated with colony males shortly after emergence. At 5–7 days post-emergence, females were starved overnight and then fed a blood meal. Fully engorged females were placed into individual cages and provided with an oviposition site and 10% dextrose ad libitum. Oviposition sites were removed 72 h after bloodfeeding and the numbers of eggs were counted. The experiment was repeated six times with separate cohorts of mosquitoes. In the first experiment, data were recorded only for those mosquitoes that laid eggs. In subsequent replicates, the number of individuals that did not lay eggs was recorded. For each replicate, the non-normally distributed egg counts were first analyzed using a Wilcoxon test to determine if there was a significant difference between transgenic and non-transgenic females.
Parasite prevalence and oocyst numbers were analyzed to determine whether transgenic mosquitoes were more resistant than their nontransgenic siblings. The data were analyzed in two ways, first by determining whether genotype was an important predictor of resistance within replicates and also pooled across replicates. This allowed us to infer, in part, why replicates within the same experiment differed. In contrast, for the pooled data sets, we included replicate as a random effect to control for inter-replicate variation without explicitly estimating their mean values.
Parasite prevalence data were analyzed to determine whether infection status (infected or not) depended on genotype. The data were analyzed for each replicate separately using a logistic regression with genotype as a fixed effect. Data for all replicates were then combined and analyzed using a generalized linear mixed model with replicate and genotype included as a random and fixed effect, respectively, in the model. Significant differences were detected using a Wald χ2 statistic.
Oocyst counts were square-root transformed to correct for overdispersion prior to using a generalized linear mixed model analysis. Data were first analyzed for each replicate separately to test for the fixed effect of genotype. The data were then combined across replicates and analyzed using replicate as a random effect and genotype as a fixed effect. Significant differences were detected using Wald's F statistic.
Analysis of survival curves was conducted using the Kaplan Meier method [39] and significant differences were detected using the Wilcoxon test as previously described [20].
|
10.1371/journal.pbio.1000544 | Control of Directed Cell Migration In Vivo by Membrane-to-Cortex Attachment | Cell shape and motility are primarily controlled by cellular mechanics. The attachment of the plasma membrane to the underlying actomyosin cortex has been proposed to be important for cellular processes involving membrane deformation. However, little is known about the actual function of membrane-to-cortex attachment (MCA) in cell protrusion formation and migration, in particular in the context of the developing embryo. Here, we use a multidisciplinary approach to study MCA in zebrafish mesoderm and endoderm (mesendoderm) germ layer progenitor cells, which migrate using a combination of different protrusion types, namely, lamellipodia, filopodia, and blebs, during zebrafish gastrulation. By interfering with the activity of molecules linking the cortex to the membrane and measuring resulting changes in MCA by atomic force microscopy, we show that reducing MCA in mesendoderm progenitors increases the proportion of cellular blebs and reduces the directionality of cell migration. We propose that MCA is a key parameter controlling the relative proportions of different cell protrusion types in mesendoderm progenitors, and thus is key in controlling directed migration during gastrulation.
| Cell migration, like any event involving shape changes, is a mechanical process controlled by complex biochemical pathways. Here, we examine cell migration in developing embryos with a combination of cell biological tools and atomic force microscopy, so as to investigate how cellular mechanical properties control migration. A fundamental step during migration is the formation of a protrusion at the leading edge of the cell. In three-dimensional environments, and particularly in vivo, cells use different protrusion types: spike-like filopodia and flattened lamellipodia, whose growth is driven by actin polymerization, and spherical blebs, which grow because of intracellular pressure pushing on the membrane. It is important to understand how the formation of different protrusion types is mechanically and molecularly controlled, and how the different protrusions specifically contribute to migration. We have addressed this using the zebrafish embryo as a model system. We show that reducing the strength of the attachment between the plasma membrane and the underlying cortical network of actin filaments, or increasing intracellular pressure, increases the proportion of cellular blebs and reduces the directionality of cell migration. Our work reveals that blebs, lamellipodia, and filopodia are not interchangeable and that the relative proportion of each type of protrusion, under the control of mechanical parameters, determines migration directionality during zebrafish gastrulation.
| During development of the vertebrate body, progenitor cells must migrate from the site at which they are specified to the site where they will eventually form the different body parts. Cell migration is the direct result of mechanical forces mediating cell shape changes and cell-substrate translocation [1]. Thus, the study of cellular mechanics is a prerequisite for understanding cell migration [2]–[4]. In recent years, most studies of cell migration have focused on its molecular control [5]. To fully understand migration, the molecules controlling cell migration must be linked to the mechanics underlying this process.
The attachment of the plasma membrane to the cytoskeleton (membrane-to-cortex attachment [MCA]) has been proposed to be an important mechanical parameter involved in cell shape changes, such as protrusion formation [6]. MCA is thought to modulate the protrusive activity of cells by providing resistance to the flow of plasma membrane into the expanding protrusion [7].
Several molecules are involved in the regulation of MCA, including Ezrin/Radixin/Moesin (ERM) proteins and class 1 myosins [8],[9]. Studies in mice, Drosophila melanogaster, Caenorhabditis elegans, and cultured cells have shown that ERM proteins are critical for cell shape control during mitotic cell rounding, cell polarization, cell migration, and cell-cell adhesion [10]–[14]. Likewise, class 1 myosins in both single-celled eukaryotes and metazoans have been implicated in various morphogenetic processes, ranging from actin polymerization and microvilli formation to cell motility [15]. In zebrafish, ERM proteins, and in particular Ezrin, are essential for tissue morphogenesis during gastrulation ([16] and Figure S1; for methods see Text S1), while the role of zebrafish class 1 myosins has not yet been studied. It remains unclear whether the functions of ERM proteins and class 1 myosins in cell and tissue morphogenesis are the direct consequence of MCA modulation, or are linked to other functions of these proteins [15],[17].
To analyze the role of MCA in cell protrusion formation and migration in vivo, we turned to zebrafish anterior axial mesendoderm progenitor cells (prechordal plate progenitors), which during the course of gastrulation migrate from the germ ring margin, where they are specified, towards the animal pole of the gastrula using a combination of different protrusion types [18],[19]. Several signaling pathways, including PDGF/PI3K and Wnt/PCP signaling, have been suggested to control protrusion formation and migration of prechordal plate progenitors [18],[19]. Recently, we showed that ERM proteins are phosphorylated and thus activated in prechordal plate progenitor cells, and are required for prechordal plate morphogenesis ([16] and Figure S1; for methods see Text S1), alluding to the possibility that ERM proteins modulate prechordal plate cell morphogenesis by regulating MCA.
Here, we show that MCA is a critical mechanical parameter determining the proportion of different protrusion types formed by prechordal plate progenitors, and thereby controlling directed migration during zebrafish gastrulation.
To test whether MCA can be modulated in prechordal plate cells by interfering with ERM protein activity, we developed an assay for measuring MCA using atomic force microscopy (AFM), and compared MCA in isolated control and ERM-deficient prechordal plate cells. Control cells were obtained from embryos expressing the Nodal-ligand Cyclops (Cyc), previously shown to induce prechordal plate progenitor cell fate and activate ERM proteins [16],[20]. ERM-deficient cells were obtained from embryos expressing Cyc in combination with either a dominant negative non-phosphorylatable version of ezrin (DNEzrin T564A; [21]) or a combination of morpholinos (MOs) targeted against ezrin and moesin-a to inactivate ERM protein function ([16]; details about MO and controls in Materials and Methods). To quantify MCA, we estimated the adhesion energy density between the plasma membrane and the subjacent cytoskeleton (W0; Figure 1A and 1B; [22],[23]) by measuring via single cell force spectroscopy [24] the force needed to extrude single lipid-membrane nanotubes (or tethers) from the cell plasma membrane. Various models of tether extrusion have shown that the force required to hold a tether at a constant height (static tether force, F0; see Figure 1A and 1B) depends on the membrane bending rigidity (κ), the plasma membrane surface tension (σ), and the energy density of MCA (W0; [22],[23]):(1)where σ+W0 is also called apparent surface tension of the membrane (Tapp; [22]). By extruding tethers from control and ERM-deficient prechordal plate cells, we found that the static tether force F0 was significantly reduced in ERM-deficient cells (Figure 1C; Table S1). We then used F0 to calculate the reduction of apparent tension Tapp in ERM-deficient cells (Tapp = 18 µN•m−1, median) compared to control cells (Tapp = 46 µN•m−1, median), using a previously determined value for κ, which we assumed was unchanged upon ERM depletion ([25]; details in Materials and Methods). To estimate the corresponding decrease in MCA energy (W0), we then measured the plasma membrane tension σ by extruding tethers from cells treated with Latrunculin A (LatA) to depolymerize the actin cortex, where W0 is negligible and thus Tapp≅σ [26]. We found Tapp to be strongly reduced in LatA-treated cells (Tapp≅2.5 µN•m−1≅σ), indicating that σ is small compared to W0 and contributes very little to Tapp (W0≅Tapp). Using this value of σ, we calculated W0 and found it to be strongly reduced upon ERM inactivation in prechordal plate cells (Figure 1D).
Inactivating ERM proteins is expected to result in a decrease in the number of molecules cross-linking the cortex to the membrane (cross-linkers). To analyze whether the density of cross-linkers is indeed reduced in ERM-deficient prechordal plate cells, we extruded tethers at varying velocities in control and ERM-deficient cells (Figure 1E and 1F; [22],[27]). The tether pulling force has to counteract the friction of the cross-linkers against lipid bilayer flowing into the tether, and increases with increasing pulling velocities (Figure 1G). A recent model has related pulling force–velocity profiles to the density of cross-linkers and the lipid bilayer viscosity ([23]; details in Materials and Methods). By measuring the diffusion of a palmitoyl-anchored GFP (GAP43-GFP) within the plasma membrane as a reporter of lipid mobility [28], we first verified that the viscosity of the plasma membrane remains unchanged between control and ERM-deficient cells (Figure S2A–S2F; details in Text S1). Using a published value for membrane viscosity (details in Materials and Methods), we then deduced the density of membrane-to-cortex cross-linking molecules from the fits of the force–velocity profiles. We found that control cells displayed about 600 cross-linking molecules per square micrometer, which corresponds to a 41-nm lateral separation between molecules on average (Figure S2G). In ERM-deficient cells, the density of cross-linking molecules was strongly reduced (Figures 1H and S2G), indicating that the reduction of W0 in ERM-deficient cells is caused by a decrease in the density of active cross-linking molecules.
Mechanical coupling of the plasma membrane to the underlying actin cortex has been proposed to influence the formation of cellular blebs [29]. Bleb-like protrusions are a common alternative to lamellipodia during migration in various cell types ranging from primordial germ cells in zebrafish to cancer cells in culture [30],[31]. We thus compared protrusion formation in isolated control and ERM-deficient prechordal plate cells expressing membrane-anchored RFP to mark the plasma membrane and Lifeact-GFP to label F-actin [32]. Isolated control cells on nonadhesive substrates formed only blebs, recognizable by the local detachment of the plasma membrane from the underlying actin cortex (Figure 2A; Video S1). Some of these blebs propagated around the cell circumference by asymmetric assembly of the actin cortex at the bleb neck, a behavior previously described as “circus movements” [33]. In contrast, ERM-deficient cells exhibited less coordinated circus movements and formed significantly larger blebs with a higher frequency (Figure 2A–2C; Videos S2 and S3). These findings indicate that reduced MCA in isolated ERM-deficient prechordal plate cells correlates with increased blebbing activity.
To determine whether similar changes in cell blebbing occur in ERM-deficient prechordal plate cells in vivo, we analyzed prechordal plate progenitor cell protrusion formation in wild type (wt) and ERM-deficient embryos expressing membrane-anchored RFP and Lifeact-GFP to distinguish between protrusion types (Videos S4, S5, S6). Three types of cellular protrusions were found in both wt and ERM-deficient prechordal plate progenitors (Figure 2D and 2E): (i) spherical protrusions initially devoid of actin, a characteristic of blebs [34], (ii) sheet-like protrusions containing actin throughout their expansion, resembling lamellipodia, and (iii) long, thin, actin-containing protrusions resembling filopodia. To quantify the formation of these different cellular protrusions in prechordal plate progenitors, we determined the frequencies of their formation, their respective proportions, and the mean time spent by the cell forming each type of protrusion. We found that in ERM-deficient prechordal plate progenitors, the frequency and size of blebs, the mean time spent blebbing, and the proportion of blebs were significantly increased, at the expense of lamellipodia and filopodia (Figures 2F–2H and S3). These observations indicate that, similar to isolated cells in culture, ERM-deficient prechordal plate progenitors with reduced MCA in vivo exhibit increased blebbing and that increased blebbing is accompanied by reduced filopodium and lamellipodium formation.
Both cortical contractility and MCA have been previously shown to be key mechanical properties controlling bleb formation [35],[36]. To exclude that changes in cortical tension rather than in MCA are responsible for the increased blebbing phenotype, we compared tension between control and ERM-deficient cells by colloidal force microscopy using AFM [37]. We found no significant differences in cell cortex tension between control and ERM-deficient cells (Figure S4), indicating that increased blebbing of ERM-deficient prechordal plate progenitors is not due to altered contractility.
We next asked whether increased blebbing activity in ERM-deficient prechordal plate progenitors with reduced MCA changes their migratory behavior. To analyze the migratory activity of prechordal plate progenitors, we tracked the nuclei of individual progenitors at the leading edge of the prechordal plate marked with Histone-Alexa-488 from mid to late gastrulation stages (8–10 h post-fertilization [hpf]; Figure 3A; Video S7). While the instantaneous speed of the cells remained largely unchanged, we found a significant decrease in the directional persistence and thus net speed of prechordal plate progenitor cell migration in ERM-deficient embryos (Figure 3B–3D). This suggests that increased blebbing activity in ERM-deficient prechordal plate progenitors with reduced MCA leads to reduced net movement speed and directionality.
To determine whether ERM proteins function cell-autonomously in mesendoderm progenitors to modulate cell migration, we co-transplanted single mesendoderm control cells (expressing Cyc, which activates ERM proteins; [16]) with ERM-deficient cells (expressing Cyc in combination with ezrin-MO to inactivate ERM proteins) into the lateral side of MZoep mutant embryos lacking most of their endogenous mesendoderm progenitors [38]. Under these conditions, transplanted cells only rarely interact with their neighbors and mostly undergo single cell migration [39]. We then tracked the movement of the cell nuclei from mid to late gastrulation stages (6–10 hpf; Figure 3E; Video S8). Similar to the behavior observed in the prechordal plate, transplanted ERM-deficient mesendoderm cells displayed a reduced directional persistence and slower net migration speed when compared to co-transplanted control cells, while their instantaneous speed was unchanged (Figure 3F–3H). This suggests that ERM proteins cell-autonomously modulate mesendoderm progenitor cell migration.
We found that in ERM-deficient cells, reduced MCA correlates with increased blebbing and that increased blebbing correlates with reduced movement directionality, which suggests that these phenotypes are functionally linked. To test whether the observed changes in cell blebbing and migration are caused by the reduction in MCA rather than by potential changes in other ERM-controlled activities, we reduced MCA independent of ERM proteins. To reduce MCA in prechordal plate progenitors, we injected a MO targeted against myosin1b-like2 (details about MO and controls in Materials and Methods) to interfere with the activity of Myosin1b, which has been previously associated with regulating MCA [9]. Similar to ERM-deficient cells, Myosin1b-deficient mesendoderm cells exhibited reduced MCA, increased blebbing, and reduced movement directionality and net speed both within the prechordal plate and as single cells transplanted in MZoep mutant embryos (Figures 4A–4G and S5; Video S9). This supports our suggestion that reducing MCA is sufficient to enhance mesendoderm cell blebbing and interfere with movement directionality and net speed, and that these phenotypes are functionally linked.
We next sought to test whether increased cell blebbing leads to the observed reduced movement directionality or whether these phenotypes are independent consequences of reduced MCA. To do so, we analyzed prechordal plate progenitor cell movement directionality when cell blebbing is increased but MCA is not reduced. To increase cell blebbing without reducing MCA, we injected a MO targeted against myosin phosphatase, target subunit 2 (myop-MO), which has previously been shown to promote the formation of bleb-like protrusions in mesendoderm cells by activating Myosin2 ([40]; details about MO and controls in Materials and Methods). MyoP-deficient prechordal plate progenitor cells showed increased cortex tension, as well as increased blebbing activity, reduced formation of lamellipodia and filopodia, and increased MCA (Figures 4H–4K and S6; Video S10). As in ERM- and Myosin1b-deficient cells, enhanced blebbing activity of MyoP-deficient mesendoderm cells, both within the prechordal plate and as single cells transplanted in MZoep mutant embryos, was accompanied by a significant reduction in the directional persistence and net speed of their migration, while the instantaneous speed of the cells did not change (Figure 4L–4N). This indicates that increased cell blebbing leads to reduced movement directionality and net speed in mesendoderm progenitors.
We have shown that reducing MCA in prechordal plate progenitors by interfering with the function of ERM proteins and class 1 myosins leads to increased bleb formation, at the expense of filopodia and lamellipodia, and that this increased proportion of blebs leads to less directed migration during gastrulation. These findings indicate that MCA is a key mechanical parameter controlling the protrusive and migratory activity of prechordal plate progenitor cells during gastrulation. The mechanical coupling of the plasma membrane to the underlying actin cortex has been proposed to regulate various cellular processes ranging from endocytosis to cell spreading [6]. Although MCA has been directly measured in cultured cells [6],[22],[23], very little is known about its actual regulation and function in cell morphogenesis in vivo, in particular in a developmental context. To directly evaluate the function of MCA in migrating prechordal plate progenitors in vivo, we developed a highly sensitive assay system based on AFM and high resolution confocal microscopy. We showed that changes in MCA lead to alterations in prechordal plate progenitor cell protrusion formation and migration. Moreover, to establish a causative relationship between MCA strength and prechordal plate progenitor cell morphogenesis, we showed that similar reductions in MCA due to inactivation of different proteins (ERM and Myosin1b) lead to comparable changes in cell morphogenesis. These experiments strongly support a critical function of MCA in cell protrusion formation and directed migration.
Our finding that reducing MCA in prechordal plate progenitor cells leads to an increase in the formation of blebs, as compared to lamellipodia and filopodia, suggests that MCA is an important mechanical parameter determining the proportion of different protrusion types formed by migrating cells. Decreasing MCA has previously been suggested to promote the formation of cellular blebs in cultured cells [36]; however, the mechanisms of bleb formation are still poorly understood [30]. Our finding that in both ERM- and Myosin1-deficient prechordal plate progenitors, reduced MCA leads to enhanced blebbing provides direct experimental evidence for a critical function of MCA in bleb formation during prechordal plate progenitor cell migration. MCA has also been proposed to modulate the extension of lamellipodia [7], although the role of MCA in this process in not yet clear. The observation that in Myosin1-deficient prechordal plate progenitor cells, reduced MCA increases blebbing but leaves the mean time spent forming lamellipodia unaltered (Figures 4 and S5) argues against a major function of MCA in lamellipodium formation in our system. However, as the frequency of lamellipodium formation is reduced in both ERM- and Myosin1-deficient cells (Figures S3 and S5), a role of MCA in controlling certain aspects of lamellipodium extension cannot be ruled out.
The observation that not only decreasing MCA, but also increasing cortical tension, which raises intracellular pressure, enhances blebbing in prechordal plate progenitors (Figure 4) suggests that the balance between MCA and intracellular pressure controls bleb formation. Interestingly, lowering MCA and/or elevating cortical tension increases not only the frequency but also the size of blebs (Figures 2 and 4). We have previously shown that cortical tension, and the resulting intracellular pressure, regulate bleb size by directly determining the force driving bleb expansion [35]. MCA, on the other hand, might control bleb size by regulating the size of the bleb base, which has been shown to correlate with bleb size [35] and is enlarged upon treatments reducing MCA (Figure 2). In addition, MCA might influence bleb size by setting the mechanical resistance to membrane flow into the expanding bleb, which in turn may control bleb expansion. Future studies addressing the contribution of cytoplasmic streaming, bleb base opening, and membrane flow to the dynamics of bleb growth will help to elucidate the mechanisms by which cortical tension and MCA together control bleb size and frequency.
We found that changing the proportions of blebs versus lamellipodia and filopodia by reducing MCA leads to less directed migration of prechordal plate progenitors. This finding indicates that the correct proportion of different protrusion types is critical for directed migration in these cells. Blebs are required for the directed migration of various cell types, including zebrafish primordial germ cells and cancer cells [34],[41],[42]. Studies in the teleost Fundulus heteroclitus have demonstrated that germ layer progenitor cells can also undergo directional migration by blebbing locomotion, suggesting that blebs are sufficient for directional migration [43],[44]. Interestingly, these cells change from bleb- to filopodium- and lamellipodium-driven migration during the course of gastrulation, resulting in individual progenitors often simultaneously forming different protrusion types [45]. While this suggests that both blebs and lamellipodia/filopodia function in directed progenitor cell migration, it remains unclear whether these different protrusion types are interchangeable or specifically contribute to directed migration. Our finding that the proportion of different protrusion types is critical for the directed migration of prechordal plate progenitors argues against interchangeability and points to specific functions for different protrusion types in this process.
Nodal/TGFβ signals are thought to be key regulators of mesendoderm cell fate specification and morphogenesis [20]. Since Nodal signaling is required for ERM phosphorylation and hence activation in mesendoderm progenitors [16], it is conceivable that Nodal proteins control mesendoderm protrusion formation and migration by regulating ERM-dependent MCA. Future studies analyzing the function of Nodal signaling in MCA will be needed to elucidate the specific contribution of MCA in Nodal-mediated mesendoderm progenitor morphogenesis.
The regulation of MCA is also likely to be important for cell migration in processes other than zebrafish gastrulation. Notably, ERM deregulation has been implicated in tumor metastasis [46], raising the possibility that the regulation of MCA is critical for cell protrusion formation and migration during tumor progression and metastasis.
Zebrafish maintenance was carried out as described in [47]. Embryos were grown at 31°C in E3 medium and staged as described in [48].
mRNA was synthesized as described in [49]. For tether force measurements wt TL embryos were injected with 100 pg of cyc alone (control) or together with a combination of 4 ng of ezrin-UTR-MO [16] plus 4 ng of moesin-a-MO (TGGTCTCTTCCTTCACGAATGTGTC) or 300 pg of DNEzrin to generate ERM-deficient cells, 2 ng of myop-MO [40] to generate MyoP-deficient cells, and 8 ng of myo1b-UTR-MO (CGAGCAGTGATGTTTTCACCTCCAT) to generate Myo1b-deficient cells. For in vitro confocal microscopy, an additional 50 pg of lifeact-GFP plus 100 pg of GPI-RFP were injected in control and ERM-deficient embryos. For in vivo confocal microscopy, wt TL embryos were injected with 50 pg of lifeact-GFP plus 100 pg of GPI-RFP alone (control) or together with 250 pg of DNEzrin (ERM-deficient), 4 ng of ezrin-UTR-MO (ERM-deficient), 3 ng of myop-MO (MyoP-deficient), or 8 ng of myo1b-ATG-MO (Myo1b-deficient). For tracking of prechordal plate cell nuclei, wt embryos were injected with Alexa Fluor-488 conjugated histone H1 (H13188, Invitrogen) and 100 pg of GPI-RFP. For tracking of cell nuclei in the transplantation experiments, wt donor embryos were injected with 100 pg of cyc together with Alexa Fluor-488 conjugated histone H1 (H13188, Invitrogen) (control), 100 pg of histoneH2A-zf::mcherry plus 4 ng of ezrin-UTR-MO (ERM-deficient), 100 pg of histoneH2A-zf::mcherry plus 3 ng of myop-MO [40] (MyoP-deficient), or 100 pg of histoneH2A-zf::mcherry plus 8 ng of myo1b-ATG-MO (Myo1b-deficient). MZoep host embryos were injected with Dextran Alexa Fluor-647 (D22914, Invitrogen).
The ezrin-UTR-MO and myop-MO were used and controlled as described in [16]. As a further control for the ezrin morphant phenotype, we expressed a dominant negative non-phosphorylatable zebrafish version of ezrin [21], resulting in a phenotype similar to that observed in ezrin morphant embryos. The myo1b-ATG-MO was designed according to Gene Tools targeting guidelines against myosin1b-like2 gene. To control the myo1b morphant phenotype, we tested a second myo1b-UTR-MO and a zebrafish dominant negative myosin1b-like2 version truncated as in [50], which produced similar prechordal plate progenitor cell blebbing phenotypes as observed with the ATG-MO. We also rescued the myo1b-UTR-MO prechordal plate progenitor cell blebbing phenotype by co-expressing mouse full-length myosin1a mRNA [50] (data not shown).
For in vivo experiments, images were obtained with an Andor spinning disc system equipped with a 63×/1.2 objective using 488-nm and 563-nm laser lines. Frames were captured at 10-s intervals for 15 min between 8 and 10 hpf. The temperature was kept constant at 28°C. For in vitro experiments, cells from Lifeact- and GPI-RFP-expressing embryos were seeded on a BSA-coated glass slide to prevent attachment and imaged using a Leica SP5 inverted microscope equipped with a 63×/1.2 lens using 488-nm and 561-nm laser lines for 2 min at 2-s intervals. For bleb size measurements, the projected area of the bleb at its maximal extension was measured using ImageJ and normalized to the projected area of the whole cell.
Wt TL and MZoep mutant donor and host embryos were dechorionated with Pronase (2 mg·ml−1 in E2) and transferred onto an agarose plate with E3 medium. Two to three cells were taken from control and experimental donor embryos at dome stage (5 hpf) and transplanted into the emerging lateral mesendoderm of a MZoep dharma::GFP host embryo labeled with Dextran Alexa Fluor-647 at shield stage (6 hpf). Time-lapse images were obtained with an upright Leica SP5 confocal microscope equipped with a 20× water immersion lens using 488-nm Argon, DPSS 561-nm, and 633-nm HeNe laser lines. Frames were captured at 90-s intervals for 3.5 h (7–10 hpf). The temperature was kept constant in all videos (28°C).
Cell/nuclei tracking in three dimensions (x, y, and z) was performed with Imaris 6.2.0 software. The instantaneous and net speeds, as well as directional persistence (ratio of the net displacement to the distance actually traveled by the cells), were extracted from the tracks.
Tethers were extruded as described in [24] using a JPK Instruments Nanowizard equipped with a CellHesion module. In short, Olympus Biolevers (k = 6 mN·m−1) were plasma-cleaned and incubated in 2.5 mg·ml−1 Concanavalin A (Sigma) for 4 h at room temperature. Before the measurements, cantilevers were rinsed in PBS plus Ca2+ and calibrated using the thermal noise method. For the measurement, cells were seeded on a glass slide in a home-built fluid chamber filled with DMEM-F12 cell culture medium and not used longer than 1 h for data acquisition. To depolymerize actin, cells were treated with 1 µM LatA for 10 min. Approach velocity was set to 5 µm·s−1, contact time was minimized to yield an interaction in 30% of all contacts (between 0.0 and 0.6 s), and contact force was set to 100 pN. For static tether force measurements, the cantilever was retracted for 6 µm at a speed of 10 µm·s−1, and the position was kept constant for 30 s. Resulting force–time curves were analyzed using IgorPro. For dynamic tether force measurements, each cell was probed with different speeds ranging from 1 to 50 µm·s−1 in a random order. Tethers were allowed to retract completely between successive pulls. Raw data were analyzed using a home-written IgorPro procedure adapted from the Kerssemakers algorithm.
Static and dynamic tether pulling experiments were used to measure the MCA energy density W0 and the density of plasma-membrane-to-cortex cross-linking molecules ν, respectively. For static tether pulling, Equation 1, described in the Results, was used to extract W0 from the static tether force (F0). For dynamic tether pulling experiments, the force (f)–velocity (ν) profiles were analyzed using the model described in [23], where the pulling force depends on the surface viscosity of the plasma membrane η and on ν:(2)where F0 is the static tether force, Rc is the radius of the cell, and Rt is the radius of the tether. The model was fitted to the data using a home-written least squares minimization procedure. This yielded the static tether force F0 and the coefficient characterizing the dynamics of extrusion, a. Values for cell radius were measured with light microscopy (Figure S2H), and the tether radius was calculated from static tether forces according to Rt = 2πκ/F0 [22].
The other parameters of the model (Equations 1 and 2) are the plasma membrane bending rigidity κ, the membrane tension σ, and the membrane surface viscosity η. All three are properties of the plasma membrane that change only if the composition of the membrane itself changes, which is unlikely to happen upon perturbations affecting proteins lying within the cortex under the plasma membrane [51]. Supporting this assumption, FRAP experiments showed that the diffusion coefficient of lipids within the plasma membrane was not changed between ERM-deficient and control cells (Figure S2A–S2F; for methods see Text S1), suggesting that membrane surface viscosity η was unchanged. Moreover, we measured the tether force (F) and membrane tension (σ) in isolated control and ERM-deficient prechordal plate progenitor cells that were treated with LatA to disassemble their actin cortex. The tether force in LatA-treated cells is determined by σ and κ only (see also Equation 1). Both the tether force F and the membrane tension σ remained unchanged in LatA-treated ERM-deficient cells relative to control LatA-treated cells (Table S1 and data not shown), suggesting that κ is also unchanged. The values of κ, σ, and η were thus kept constant for all the experimental conditions. κ and η were taken from the literature with κ = 2.9×10−19 N•m [22],[25],[36] and η = 1.5×10−7 Pa•m•s [26]. Plasma membrane tension σ was calculated from tether pulling experiments using cells treated with LatA (Tapp = σ = 2.5 µN•m−1). During tether extrusion, the model assumes that the lipids flow past the cytoskeleton-bound transmembrane molecules as they are dragged into the tether (permeation regime). This is true for intermediate velocities up to several 100 µm•s−1 (Figure S2I), while transmembrane molecules unbind from the cortical cytoskeleton if tethers are extruded faster or membrane viscosity becomes greater [23]. Since the tether pulling velocities in our experiments were ≤50 µm•s−1, we were most likely within the permeation regime, allowing us to investigate the density of binding molecules. No history effect was observed when sequential tethers were extruded from one cell (Figure S2J).
Cortex tension measurements were carried out as described previously [37]. In short, an AFM cantilever was modified with a glass bead (diameter D = 5 µm) and coated with heat-inactivated FCS to prevent unspecific binding with the cell during the contact measurement. The colloidal force probe was then brought into contact with the cell with 500 pN contact force at 1 µm•s−1. A fit to the cortical shell liquid core model [37] between 125 pN and 250 pN yielded cortex tension. To depolymerize actin, cells were treated with 1 µM LatA for 10 min.
Analysis of variance (ANOVA) and t tests were performed after data were confirmed to have normal distribution and equal variance; otherwise, Kruskal–Wallis tests or Mann–Whitney U tests were applied. p-values were computed in R. For cell transplantation experiments, ttest2 from Matlab was used, which compared our data points with a random distribution of numbers around one with the same standard deviation as our data.
|
10.1371/journal.pntd.0000224 | A Simplified 4-Site Economical Intradermal Post-Exposure Rabies Vaccine Regimen: A Randomised Controlled Comparison with Standard Methods | The need for economical rabies post-exposure prophylaxis (PEP) is increasing in developing countries. Implementation of the two currently approved economical intradermal (ID) vaccine regimens is restricted due to confusion over different vaccines, regimens and dosages, lack of confidence in intradermal technique, and pharmaceutical regulations. We therefore compared a simplified 4-site economical PEP regimen with standard methods.
Two hundred and fifty-four volunteers were randomly allocated to a single blind controlled trial. Each received purified vero cell rabies vaccine by one of four PEP regimens: the currently accepted 2-site ID; the 8-site regimen using 0.05 ml per ID site; a new 4-site ID regimen (on day 0, approximately 0.1 ml at 4 ID sites, using the whole 0.5 ml ampoule of vaccine; on day 7, 0.1 ml ID at 2 sites and at one site on days 28 and 90); or the standard 5-dose intramuscular regimen. All ID regimens required the same total amount of vaccine, 60% less than the intramuscular method. Neutralising antibody responses were measured five times over a year in 229 people, for whom complete data were available.
All ID regimens showed similar immunogenicity. The intramuscular regimen gave the lowest geometric mean antibody titres. Using the rapid fluorescent focus inhibition test, some sera had unexpectedly high antibody levels that were not attributable to previous vaccination. The results were confirmed using the fluorescent antibody virus neutralisation method.
This 4-site PEP regimen proved as immunogenic as current regimens, and has the advantages of requiring fewer clinic visits, being more practicable, and having a wider margin of safety, especially in inexperienced hands, than the 2-site regimen. It is more convenient than the 8-site method, and can be used economically with vaccines formulated in 1.0 or 0.5 ml ampoules. The 4-site regimen now meets all requirements of immunogenicity for PEP and can be introduced without further studies.
Controlled-Trials.com ISRCTN 30087513
| All human deaths from rabies result from failure to give adequate prophylaxis. After a rabid animal bite, immediate wound cleaning, rabies vaccine and immunoglobulin injections effectively prevent fatal infection. Immunoglobulin is very rarely available in developing countries, where prevention relies on efficacious vaccine. WHO approved vaccines are prohibitively expensive, but 2 economical regimens (injecting small amounts of vaccine intradermally, into the skin, at 2 or 8 sites on the first day of the course) have been used for many years in a few places. Practical or perceived difficulties have restricted widespread uptake of economical methods. These could largely be overcome by introducing a new, simpler regimen, involving 4 site injections on the first day. We vaccinated volunteers to compare the antibody levels induced by the 4-site intradermal regimen with those induced by the current 2-site and 8-site regimens and the “gold standard” intramuscular regimen favoured internationally. All the economical intradermal regimens were at least as immunogenic as the intramuscular method. The results provide sufficient evidence that the 4-site regimen meets the criteria necessary for its recommendation for use wherever the cost of vaccine is prohibitive and especially where 2 or more patients are treated on the same day.
| Rabies is a neglected disease affecting particularly tropical developing countries [1]. Estimates of the Global use of rabies post-exposure prophylaxis (PEP) are rising. In China, it was 8 million in 2005 [2], yet rabies currently kills more people than any other infectious disease there. Rabies deaths are underreported and misdiagnosed, for example as cerebral malaria [3]. As the obsolete nervous tissue-based rabies vaccines are replaced by expensive tissue culture vaccines, there is increasing need to reduce the cost of post-exposure prophylaxis. In Africa, the average cost of a standard intramuscular (IM) course of vaccine is $39.6, equivalent to 50 days wages [1]. There is a shortage of affordable rabies vaccines of reliable quality in the developing world [4].
Economical PEP regimens employ multiple site intradermal (ID) injections, saving 60% of the vaccine used in the standard IM method (Table 1). Increasing the number of sites of injection is designed to stimulate several different groups of lymph nodes to initiate antibody production. Two economical regimens are now recommended [5], an 8-site [6] and a 2-site [7] method (Table 1). The urgency of PEP demands a rapid induction of neutralising antibody using minimal amounts of vaccine in all recipients including the ∼3% of the population who are ‘low responders’ [8] and the many others whose immune response is impaired [9]–[12].
The original IM PEP vaccine regimen is the most widely used globally. In Asia only 3% of tissue culture rabies vaccine treatments use economical ID regimens [1], and they are rarely used in Africa . The reasons are misgivings about reducing the vaccine dosage in prevention of a fatal disease [13], confusion over regimens, and the competence of staff giving ID inoculation. Economical regimens require sharing of ampoules between patients, but rabies vaccines have no added preservative and so the reconstituted ampoule of vaccine should be used within a day. The use of economical regimens is therefore mainly confined to large treatment centres, yet 90% of rabies deaths occur in rural areas [4].
Evidence to date indicates that the 8-site regimen is more immunogenic than the 2-site regimen [14],[15]. However the 8-site method is not economical when used with one of the two major vaccines, purified vero cell rabies vaccine (PVRV) (Verorab™ Sanofi Pasteur), because this vaccine is relatively concentrated: an IM dose is 0.5 ml, in contrast to the equivalent 1 ml dose of the other widely used vaccine, purified chick embryo cell vaccine (PCECV) (Rabipur™; Novartis) [15]. Although the 8-site regimen has some advantages and was recommended by some authorities for use when rabies immunoglobulin (RIG) was not available [15], the 2-site regimen is more acceptable and convenient. The total dose of vaccine should be the same with the two regimens. The only difference between the two schedules is that with the 8-site a large dose of vaccine is given on the first day, whereas with the 2-site regimen this is divided between days 0 and 3, entailing an extra treatment visit [14]. Ambrozaitis et al. [16] demonstrated that the 4-site regimen was apparently immunogenic with both PVRV and PCEC vaccines, but there was no comparison with any current PEP method and historical controls are unreliable.
For all these reasons, a single, simple, acceptable, immunogenic and economical PEP regimen is needed, suitable for use with all vaccines fulfilling WHO requirements. We tested a 4-site PEP regimen which allows the 8-site regimen principle to be used economically with PVRV. We also investigated whether injecting the same amount of vaccine between 4 instead of 8 sites affected immunogenicity. The new 4-site regimen and the currently used ID regimens were compared with the standard IM method in a single blind, randomised, controlled trial.
The CONSORT checklist for this study is available in Supporting Information as Checklist S1.
Healthy volunteers were recruited in Oxford and Bristol UK, between June 2002 and April 2005. The exclusion criteria were: previous rabies vaccine treatment; pregnancy; a recent blood transfusion; taking immunosuppressive drugs; receiving another killed vaccine or chloroquine treatment [17] within 2 weeks, or any live virus vaccine within 3 weeks of a rabies vaccine dose. The Oxfordshire Clinical Research Ethics Committee approved the project (ref. C01.078), conducted in accordance with GCP regulations (EU Directive 2001/20/EC).
Each participant was allocated to one of four rabies PEP regimens according to a computer generated list with fixed blocks of 12. Group A received the 4-site regimen; group B, the 8-site regimen; group D, the 2-site regimen and group E, the IM regimen. Allocations were concealed in opaque serially numbered sealed envelopes, opened once written informed consent had been obtained. All laboratory staff were blinded to the treatment allocation.
The vaccine used was PVRV, (Verorab™ Sanofi Pasteur) Lot no XO291-1 potency 5.3 IU/dose in 165 subjects, and Lot no. U0271 potency 8.4 IU/dose in 64 subjects. The Medicines and Healthcare products Regulatory Agency granted exemption from a licence.
The 2-site and IM regimens were according to standard methods (Table 1) [15]. For the 4-site regimen, on day 0 the entire contents of the 0.5 ml PVRV vial are injected ID, divided between 4 sites over the deltoids and thighs (approximately 0.1 ml per site).
On day 7, 0.1 ml is injected ID at 2 sites (deltoids). Single site injections are given on days 28 and 90. The 8-site regimen is the exact equivalent of the current 8-site method [6],[15], using a vaccine containing 0.5 ml/ampoule. The entire contents of the vial are divided between 8 ID sites on day 0: (deltoids, thighs, suprascapular, lower anterior abdominal wall). The dose per site is approximately 0.05 ml. All the ID regimens use the same total amount of vaccine. There is a little inevitable wastage in syringes. Opened ampoules were refrigerated and used or discarded within 8 hours. See Table 1 for the timing, doses, routes and sites of inoculation of all the regimens.
Blood samples were taken at days 0, 7, 14, 90 and 1 year. Serum aliquots were coded, stored at −70°C and assayed blind. Neutralising antibody levels were measured by an adaptation of the rapid fluorescent focus inhibition test (RFFIT) for 96 well plates [18],[19], at the Institut Pasteur, Paris. Briefly, a constant dose of challenge virus standard (CVS) is incubated with diluted test sera. An in-house reference serum (SHR2 31/03/06 = 22 IU/ml), is calibrated against an international standard (RAI = 30 IU/ml). Serum/virus mixtures are incubated, and BSR cells (a clone of BHK-21 cells) were added. After 24 hours incubation, the monolayer is acetone-fixed and stained with a fluorescent anti-nucleocapsid antibody (Chemicon). The result in IU/ml was the mean of independent duplicate tests.
Selected sera were also assessed using the fluorescent antibody virus neutralisation (FAVN) assay at the Veterinary Laboratories Agency ,Weybridge [20],[21]. This test is the same in principle as the RFFIT, using the same challenge virus. The FAVN and RFFIT vary in that they use a different dilution series (3 fold versus 5 fold); the FAVN runs samples in quadruplicate; BHK-21 cells (ATCC, USA) are used, and the internal serum standard is the WHO human positive control (NIBSC, UK). The antibody titre is based on 100% virus neutralisation for the FAVN and 50% reduction of fluorescent foci in the RFFIT.
Protocol deviations were not permitted on days 7 and 14, but flexibility was allowed if necessary: on day 28±1 day; on day 90 - 7 to +10 days, and at one year −2 weeks to +4 weeks. All records were kept in strict confidence. Volunteers kept a health record diary for a week after each vaccine dose.
The aim was to demonstrate that the 4-site test regimen was at least as immunogenic as the standard regimens. The primary outcome is the proportion of participants reaching the WHO criterion for post-exposure regimens: a minimum neutralising antibody level of 0.5 IU/ml by day 14. The failure rate for the current regimen in meeting this threshold is less than one in 1000. At this rate, the expected number of failures in the control group is likely to be zero. The sample size calculation was based on the assumption that the new regimen was just as effective (i.e. rate of less than 1 in 1000) and was computed by simulation method using exact methods for estimating the confidence interval (CI) for the difference.
The initial protocol envisaged the recruitment of 75 participants per group to make 5 comparisons over 7 regimens expecting zero events to be observed, giving 90% power to show that the difference in failure rates was at most 6.2% (adjusting for pre-planned multiple comparisons). Because the trial failed to recruit at an adequate rate, the revised sample size of 55 participants per group (Protocol S1) was calculated for a total of 6 comparisons among four groups giving 90% power to show that the difference in failure rates was at most 9% by day 14.
Proportions and 95% CI for the difference in proportions were calculated using the method based on Wilson's score [22]. Agreement between the results of the two antibody tests was assessed by the Bland-Altman method [23]. Titre concentrations were log transformed and groups were compared using analysis of variance. Results were deemed statistically significant at P<0.05. Fisher's exact test was used to compare side effects between groups. Post-hoc pairwise comparisons were also carried out on any local reactions (redness, swelling hardness, or tenderness/pain) and on any local or generalised signs or symptoms. P-values were adjusted if multiple comparisons were performed.
Two hundred and fifty four subjects were recruited. Data from 229 were complete up to day 90, and used in the final analysis. Twenty five were excluded, usually because they failed to keep appointments (for details see Figure 1). Three sera taken on day 14 were lost during storage.
Subjects were aged between 18 and 50 years. Ages and sex ratios were similar between the groups (Table 2).
One person withdrew within the first week because of transient arthralgia, possibly related to the vaccine.
Local reactions to the vaccine observed by 229 volunteers are shown in Table 3. Redness (erythema), swelling (inflammation) and hardness (induration) were more frequent in all ID groups than in the IM group (E) (P<0.0001). The incidence of local tenderness or pain was similar in all groups. Itchiness and local lymphadenopathy (tenderness at regional lymph nodes) was not solicited, but was volunteered more often in the ID groups (P<0.0001).
Volunteers were asked to report any generalised symptoms, whether or not listed in their reaction diary. Some were attributable to causes unrelated to vaccination. The incidence of each of the generalised symptoms was lower with the IM regimen but this only reached significance when compared with all three ID groups together (P<0.001) (Table 4).
The lower limit of detection of antibody was 0.06 IU/ml, while the threshold for a positive result, was 0.3 IU/ml, as naïve sera can range between 0 and 0.3 IU/ml.
Two sera gave pre-vaccination results above this threshold (the means of two tests were 0.38 IU/ml and 0.46 IU/ml). These subjects denied previous rabies immunisation and subsequent titres did not suggest a secondary immune response, but they were excluded from the analysis. Data for the remaining 227 people were analysed. Undetectable titres were assigned the value of 0.02.
Geometric mean titres (GMTs) on day 7 for the 4 treatment groups (Table 5, Figure 2), showed that group E (IM) had a lower GMT than group D (2-site) (P<0.001) and group B (8-site) (P = 0.01). Group A (4-site) was lower than group D (P = 0.01). The percentage of people with detectable antibody >0.3 IU/ml was 60%, 77.6%, 86.2% and 62.5% for groups A, B, D and E respectively. The day 7 results were no different with the two batches of vaccine (data not shown).
On day 14 all subjects had antibody levels >0.5 IU/ml (Table 5, Figure 2). The 95% confidence intervals for the differences in proportions between any two regimens indicated that differences could at most be between 6% and 7% (Figure 3). The only significant difference between the GMTs is that Group E (IM) was lower than group D (2-site) (P = 0.04).
On day 90, GMTs were similar (Table 5, Figure 2). At 1 year all ID recipients had detectable antibody, but two people in group E (IM) had <0.3 IU/ml. Eight had levels between 0.3 and 0.5 IU/ml: 2 in group A, 2 in group B and 6 in group E. All the ID regimens induced more persistent antibody than the IM group (P<0.001 for groups B and D, P<0.02 for A). The 2-site (D) GMT was greater than the 4-site (A) (P<0.04).
Before the serological data were decoded, some unusual results were identified. Antibody levels were unexpectedly high on days 7 and 14, compared with other clinical trials [6],[7],[16]. Two subjects were excluded because they had pre-vaccination antibody levels above the 0.3 IU/ml threshold. Two people had antibody levels >3000 IU/ml on day 14 (both of them later proved to be in group D). The next highest were seven subjects with levels between 1000 and 1500 IU/ml. The other results for these people were well within the range of the rest. After decoding, none of the high results was found to be among group A subjects. They were individual high titres, without any suggestion of an anemnestic response. To confirm the results of the trial, 224 (of the original 229) day 7 samples available, and a few others (see below) were tested blind in another laboratory which uses the FAVN method.
The FAVN lower limit of detection of antibody was 0.05 IU/ml. The threshold for a positive result is >0.13 as naïve sera can range up to 0.1 IU/ml.
The day 7 results for 224 subjects, including the 2 excluded because of high initial RFFIT titres, showed GMTs between 1.044 IU/ml for group D (2-site) and 0. 573 IU/ml for E (IM) (P<0.01) (Table 6, Figure 4). There were no other significant differences and GMTs were in the same order as the RFFIT results for all 4 treatment groups. The percentage of people with detectable antibody, >0.13 IU/ml, was 96.3 %, 93.0%, 96.6 % and 87.3 % for groups A, B, D and E respectively.
This comparison showed general consistency but considerable individual variation, as demonstrated graphically in a Bland-Altman plot (Figure 5). Further analysis was not appropriate in such a small sample. All the FAVN results were <6 IU/ml, except one of 13.5 IU/ml (the RFFIT result was 2.8 IU/ml). For the RFFIT, all titres were <7 IU/ml, except two of 8.39 and 9.36 IU/ml (the FAVN results were 1.14 and 3.42 IU/ml respectively).
The day 0 sera with RFFIT results >0.3 IU/ml, and day 14 sera with RFFIT results >3000 IU/ml, were included in a group of 38 otherwise randomly selected sera to be tested by the FAVN method. The day 0 results of 0.38 and 0.46 IU/ml were both 0.06 IU/ml by the FAVN. RFFIT results of 3711.5 and 3021.5 IU/ml were 53.3 and 121.2 IU/ml respectively by the FAVN.
This study demonstrates that ID rabies vaccination is at least as immunogenic as the standard IM regimen and induces greater persistent immunity. ID regimens are therefore recommended anywhere in the world where the cost of PEP is critical. All three ID regimens required the same total amount of vaccine and proved equally immunogenic, but the 4-site ID regimen has several key advantages.
First, the 4-site needs one less clinic visit (omitting day 3). WHO now recommends omitting the day 90 dose of ID regimens, and doubling the day 28 dose [24],[25],[26]. The 4-site regimen would then require only 3 visits (days 0, 7 and 28) the same as the current 3 dose IM pre-exposure regimen, but using only about half the amount of vaccine.
Secondly, the 4-site regimen is safer than the 2-site as it uses a whole ampoule of vaccine divided between intradermal sites on the crucial first day. If some vaccine were inadvertently injected subcutaneously, the wide margin of safety would ensure an adequate immune response [27]. Thirdly, sharing of ampoules of vaccine between patients is only necessary on days 7 and 28. The 4-site regimen can therefore be started in a rural clinic with referral a week later. It is economical anywhere if two or more people are treated on the same day.
The 4-site regimen can be used economically with current vaccines formulated in 0.5 and 1.0 ml ampoule sizes. Our results show that there is no need to divide the initial dose between 8 sites, because it was equally immunogenic in 4 sites. We injected over the deltoid and thigh areas, whereas Ambrozaitis et al. [16] used deltoid and suprascapular sites. The choice might be important in cultures where there is reluctance to expose the thighs.
The efficacy of the 8-site regimen has been demonstrated in patients bitten by proved rabid animals, with and without concomitant RIG [6]. Since the 4-site method has the same timing of doses and amount of vaccine, and is equally immunogenic, it can be inferred that RIG treatment would not be significantly immunosuppressive. All authorities recommend the combination of RIG with vaccine for PEP, especially for high risk exposure to rabies. Treatment failures are inevitable in severe cases (bites on the head, neck or hands or multiple bites) if vaccine is given alone. However RIG is not generally available or affordable in developing countries where it is given to <1% of PEP patients for whom it is recommended [28]. The 4-site regimen fulfils WHO requirements for immunogenicity for PEP and so could be introduced without further studies.
WHO recommendations have changed since 1997, when the difference in dilution was recognised [15], to the latest rule that an ID dose of either vaccine is 0.1 ml [5],[26]. Other studies of 8-site PVRV have used 0.1 ml per ID site [29], as recommended by WHO [5], which almost doubles the amount of vaccine used. The results for the 2-site regimen we report here apply to PVRV, the equivalent dose for PCECV would be 0.2 ml per site.
Ambrozaitis et al. [16] have tested this 4-site regimen to compare different doses of vaccine. Using PCECV, which is formulated in 1 ml ampoule, they showed that 0.1 ml per ID site, a lower dose, was as immunogenic as 0.1 ml per ID site of PVRV. This confirms the safety of our 4-site method, in which 0.25 ml of PCECV would be injected at each ID site on day 0, and 0.2 ml per site subsequently. Using the lower dose of 0.1 ml per site would sacrifice the advantages of using a whole ampoule on the first day, but would be more economical in large treatment centres [13].
The FAVN and the RFFIT tests are identical in principle but differ in the way their results are read. A comparison between these tests, performed within the same laboratory, showed close correlation [20], but there has been no report of inter-laboratory comparisons. Our data were too few for substantial analysis. In this study, at least one unusually high level was seen with one test, but not confirmed by the other. These results were used in the analysis but did not affect the overall findings or conclusion. Similarly high individual results have been reported previously, but not explained [30],[31],[32]. Rabies immunisation is expensive and unusual in the UK. Thorough investigations excluded previous immunisation in the group analysed and so the high titres cannot be dismissed as an anamnestic response.
Antibody GMTs on days 7 and 14 were much higher, both by the RFFIT and FAVN than in some other recent studies [16],[30]. Over 30 years, no difference has been reported in serological responses to tissue culture rabies vaccines between people in America, Europe and Asia. The higher levels found here remain unexplained. In a 2-site ID vaccine trial in Thailand, antibody levels varied 2.2 fold between different hospitals [30].
Economical rabies PEP regimens using 2, 4 or 8 initial ID sites are as immunogenic as the standard IM regimen, but they use 60% less vaccine. The 4-site regimen has several practical advantages over both currently used regimens, and is the most economical since only 3 or 4 clinic visits are needed (on days 0, 7 and 28 with optional day 90). Our finding that ID regimens were at least as immunogenic as the “gold standard” 5 dose IM regimen should increase confidence in multiple-site ID techniques. The 4-site regimen is suitable for use anywhere in the world where there are financial constraints, and especially where 2 or more patients are likely to be treated on the same day. |
10.1371/journal.pgen.1006682 | Wnt/Tcf1 pathway restricts embryonic stem cell cycle through activation of the Ink4/Arf locus | Understanding the mechanisms regulating cell cycle, proliferation and potency of pluripotent stem cells guarantees their safe use in the clinic. Embryonic stem cells (ESCs) present a fast cell cycle with a short G1 phase. This is due to the lack of expression of cell cycle inhibitors, which ultimately determines naïve pluripotency by holding back differentiation. The canonical Wnt/β-catenin pathway controls mESC pluripotency via the Wnt-effector Tcf3. However, if the activity of the Wnt/β-catenin controls the cell cycle of mESCs remains unknown. Here we show that the Wnt-effector Tcf1 is recruited to and triggers transcription of the Ink4/Arf tumor suppressor locus. Thereby, the activation of the Wnt pathway, a known mitogenic pathway in somatic tissues, restores G1 phase and drastically reduces proliferation of mESCs without perturbing pluripotency. Tcf1, but not Tcf3, is recruited to a palindromic motif enriched in the promoter of cell cycle repressor genes, such as p15Ink4b, p16Ink4a and p19Arf, which mediate the Wnt-dependent anti-proliferative effect in mESCs. Consistently, ablation of β-catenin or Tcf1 expression impairs Wnt-dependent cell cycle regulation. All together, here we showed that Wnt signaling controls mESC pluripotency and proliferation through non-overlapping functions of distinct Tcf factors.
| Studying how to safely expand stem cells in culture is essential for regenerative medicine applications. Hence there is a clear need to decode how the cell cycle of mouse embryonic stem cells (mESCs) is regulated. Tcf3 and Tcf1 belong to the Tcf family of proteins. Tcf/Lef are effectors of the Wnt/β-catenin pathway and Tcf3 controls mESC pluripotency. Here we identified a recruitment site for Tcf1 embedded into a number of cell cycle repressor genes such as p15Ink4b, p16Ink4a and p19Arf. Tcf1-mediated activation of these genes drastically slows down proliferation of mESCs. In conclusion, here we showed that the Wnt pathway, besides controlling mESC pluripotency via Tcf3, also regulates mESC cell cycle through the recruitment of Tcf1 to the regulatory sites of key cell cycle genes.
| Wnt/β-catenin signalling plays an essential role in development, tissue homeostasis and cancer [1]. In addition, activation of the Wnt pathway maintains pluripotency in mouse embryonic stem cells (mESCs) [2] and controls somatic cell reprogramming [3,4]. On the other hand, deregulation or constant activation of Wnt signalling may lead to cancer formation [5].
In the absence of Wnt ligands, β-catenin is recruited by the destruction complex, where it is phosphorylated by GSK3 and subsequently degraded by ubiquitin-mediated proteolysis. Binding of Wnt ligands to their receptors results in the inactivation of the destruction complex, thereby allowing hypophosphorylated β-catenin accumulation [6]. Small molecules such as 6-bromoindirubin-3'-oxime (BIO) [7] or CHIR99021 [8] can also be used to inhibit GSK3 and thus to stabilize β-catenin.
Stabilized β-catenin can enter the nucleus, where it interacts with members of the T cell factor/lymphoid enhancer factor (Tcf/Lef) family. While a single Tcf/Lef gene is found in Drosophila melanogaster and Caenorhabditis elegans, four Tcf genes, Tcf1, Lef1, Tcf3 and Tcf4 exist in mammals [9]. An important issue that warranted investigation is if the complexity of Tcf factors has also evolved with specialized or redundant functions of the distinct Tcf/Lef factors. Tcf1 and Tcf3 are the most expressed Tcf/Lef factors in pluripotent mESCs [10,11]. Tcf3 acts as a transcriptional repressor of Wnt target genes regulating the pluripotent gene network in mESCs [12,13]. Activation of Wnt/β-catenin pathway reduces the Tcf3 transcriptional repression thereby reinforcing the stability of the core pluripotency network. However, the function of the Wnt transcriptional activator Tcf1 [14] and its target genes in pluripotent mESCs are unknown. Here we show that Tcf/Lef factors regulate distinct target genes showing gene target specialization determining context-specific responses to Wnt signaling.
In somatic stem cells, activation of the canonical Wnt pathway stimulates cell proliferation [6,15] mainly by inducing expression of c-Myc and Cyclin D1 genes [16]. However, even if the mitogenic effects of the Wnt pathway on somatic cells are well known, whether Wnt signalling regulates the cell cycle of pluripotent cells remained unknown.
Pluripotent mESCs, differently to their somatic stem cell counterparts, display a unique and singular cell cycle defined by a fast proliferation rate, characterized by a long S phase and very short G1 and G2 phases [17–20]. The high proliferative rate of mESCs is due to the absence or low expression of Cyclin-Dependent Kinase Inhibitors (CDKIs) such as the Ink4 family members p15Ink4b, p16Ink4a, p18Ink4c and p19Ink4d, the CIP1/KIP family members p21Cip1, p27Kip1 and p57Kip2 [19,21–24], and p19Arf [25]. The Ink4/Arf locus encodes for p15Ink4b, p16Ink4a and p19Arf, which are considered strong tumor suppressors. p15Ink4b and p16Ink4a, along with the other Ink4 and Cip/Kip family members can slow down cell proliferation by binding to and inhibiting CDK-cyclin complexes. On the other hand, expression of p19Arf inhibits the Mdm2 E3 ubiquitin ligase to activate and stabilize p53, which induces expression of the CDKI p21Cip. Therefore, the Ink4/Arf locus controls the two main cell cycle inhibitors and tumor suppressor pathways [26,27].
The biological significance of a short G1 phase in mESCs is yet unclear. It has been hypothesized that a short G1 phase might be essential in actively sustaining the pluripotent state. Accordingly, it has been shown that the longer mESCs stay in G1, the more likely they could be subject to signals for cell differentiation [20,28–31]. However, on the other hand, accumulation of mESCs in G1, by inhibition of Cdk2 [32] or by overexpression of p21 or p27 [33] reduces mESC proliferation but does not affect cell pluripotency.
Here, we show that the activation of the canonical Wnt pathway has a dual role in mESCs. Wnt induces the expression of negative regulators of cell cycle; leading to a reduction of cell proliferation and an increase in the number of cells in G1. Furthermore, activation of the Wnt pathway results in the downregulation of some cell differentiation genes, while the expression of pluripotency genes remains unperturbed. The cell cycle effects are dependent on β-catenin and the downstream transcription factor Tcf1 but independent of Tcf3, indicating specialized and non-overlapping functions of Tcf/Lef factors in mESCs. Tcf1 recruitment was enriched at the promoters of cell cycle genes such as in the Ink4/Arf locus. Activation of the Wnt pathway induces therefore an increased expression of negative regulators of the cell cycle such as the tumour suppressors Cdkn2a (p16Ink4a, p19Arf) and Cdkn2b (p15Ink4b). All together our results show that, in contrast to its mitogenic effect in somatic cells, the Wnt/β-catenin pathway triggers an anti-proliferative effect in mESCs via Tcf1 activity.
We performed comparative gene target analysis of the two most expressed Tcf/Lef factors in mESCs, Tcf1 and Tcf3 [10,14], by chromatin immunoprecipitation combined with DNA sequencing (ChIP-Seq).
Tcf3 was found to be associated with the canonical Wnt/Tcf DNA binding motif named the Wnt Response Element (WRE: 5′-CTTTGWW-3; W = A or T), as previously reported [34,35] (Fig 1A). We found that Tcf3 binds to ±5 kb of the transcription start site (TSS) of more than 1000 annotated genes (S1 Table). Accordingly with previous reports [13,34], we found that Tcf3 associates with the promoter regions of known Wnt targets (Axin2, Lef1), pluripotency transcription factor genes (Oct4, Tbx3, Nanog) as well as specific pluripotency miRNAs (miR302) (S1 Table) in line with its role as regulator of ESC pluripotency network, of lineage priming and of ESC exit from pluripotency [8].
Interestingly, we found that Tcf1 is recruited to a palindromic DNA binding site (Fig 1A) different from the already described WRE. Most of the peaks (95%) were within 3.5 kb distance from the TSSs (Fig 1B). Among these regions (within ±3.5kb region from TSS), 40% corresponded to promoter regions and 27% to intronic regions (Fig 1C). We found around 1800 annotated genes containing a Tcf1 recruitment site at ±3 kb distance from the TSS. The number of Tcf1 target genes increased up to 2100 when the sequence analysis was extended to ±5 kb from the TSS (S2 Table). Importantly, known Wnt target genes in mESCs such as Axin2, Lef1 and Cdx1 were identified as Tcf1 targets (Fig 1D and S1A Fig) and some targets were validated by independent ChIP-qRT-PCR (Fig 1E).
Next we followed a reverse strategy to link a list of genes with the transcriptional machinery. We used the Enrichr Analysis Tool [36] to determine which transcription factors regulate the genes that are Tcf1 targets at ±3 kb distance from the TSS (S2 Table). Interestingly, an unknown transcription factor with a “TMTCGCGA” DNA binding sequence was identified as best candidate, which matched the newly identified Tcf1 DNA binding sequence (S1B Fig and S4 Table).
These results show that in the majority of cases Tcf1 and Tcf3 bind to distinct DNA binding motifs in mESCs in different promoter regions (S1C Fig and S5 Table), suggesting that they might control different cellular programs and functions.
To explore the biological processes regulated by Tcf1, enriched Gene Ontology (GO) categories associated with Tcf1 target genes were identified and displayed using EnrichNet [37] (S3 Table). Genes associated with the category “Negative Regulators of Cell cycle” (GO:0045786) were highly enriched in Tcf1 targets, indicating that some of the Tcf1 direct target genes might be negative regulators of mESC proliferation (Fig 1F). Analysis of KEGG enriched terms also produced “cell cycle” as the first category of Tcf1-binding genes (S1D Fig).
To date, the crosstalk between signal pathways and transcription factors regulating cell cycle in pluripotent cells is still unexplored. Furthermore, mESCs have a unique cell cycle defined by the absence of the expression of CDK inhibitors [20]. Surprisingly, we found recruitment of Tcf1 to the promoter of genes in the Ink4/Arf tumour suppressor locus (also known as Cdkn2 locus) (Fig 1D and S1E Fig).
To investigate a possible novel function of Wnt/Tcf1 in the regulation of mESC proliferation, we focused on the expression and activity of CDKIs as a new direct target of the Wnt/Tcf1 pathway. We activated the Wnt pathway by treating mESCs with the GSK3 inhibitor BIO. We observed significant upregulation of the transcript level of p15Ink4b, p16Ink4a and p19Arf (Fig 2A) after BIO treatment, being p19Arf the most abundant expressed transcript among them when compared to Gapdh levels (S2A Fig). In addition, we also observed an increase in p15Ink4b, p16Ink4a and p19Arf protein level along with β-catenin stabilization in mESCs treated with BIO or CHIR99021 for 48h (Fig 2B). Importantly, short-term activation of the pathway did not change the expression of pluripotency markers such as Nanog, as expected [3] (Fig 2A). Genes like c-Myc and Cyclin D1, which are regulated by Wnt signalling and increase the proliferation of a variety of adult stem cells [16], were not upregulated in mESCs after BIO treatment. Axin2, a known Wnt target gene, increased, as expected (Fig 2A).
Ink4 family members have a direct role in regulating G1 to S transition [27] and expression of p19Arf is known to stabilize p53 [26,27]. Interestingly, in agreement with the increased levels of p19Arf, we observed increased levels of p53 together with its downstream target p21Cip in protein nuclear extracts of mESCs treated with BIO for 5 days (S2B Fig) and a reduction of c-Myc protein levels (S2C Fig). Overall, these data indicate that the Tcf1 transcriptional targets belonging to Ink4 and Arf families are upregulated after Wnt pathway activation in mESCs.
Somatic cells slow down their cell cycle and reduce proliferation upon increased expression of any of the genes of the Ink4/Arf locus, such as p16Ink4a or p19Arf. In addition mESCs are believed to be refractory to the action of some CDKi as p16Ink4a [21,24]. However, it has recently been shown that an increased expression of p21 or p27 can increase the length mESC G1 and reduce their proliferation [33]. Thus we examined if the Ink4/Arf locus could regulate mESC proliferation. We infected WT mESCs with retroviruses expressing mouse p16Ink4a or mouse p19Arf (Fig 2C and S2D Fig). Overexpression of p16Ink4a or p19Arf did not significantly changed pluripotent marker expression (S2G Fig). Cell proliferation was analysed by EdU staining and cell counting. We observed a reduction of EdU+ cells in all clones overexpressing p19Arf and in 4 out of 6 clones expressing exogenous p16Ink4a (Fig 2D and S2E Fig). These results were confirmed by cell counting (Fig 2E and S2F Fig).
All together these data show that cell cycle inhibitors of the Ink4/Arf locus, are transcriptional targets of the Wnt/Tcf1 activation and reduce mESC proliferation.
Activation of the canonical Wnt pathway is necessary to maintain self-renewal and pluripotency of mESCs [2,38]. However, Wnt is also a proliferative signal for intestinal, hair follicle and hematopoietic adult stem cells [39–41], and an oncogenic initiator when aberrantly activated in cancer cells [1,16]. Having identified cell cycle inhibitors as novel Tcf1 target genes in mESCs, we assessed the effect of Wnt pathway activation on mESC morphology, proliferation and cell cycle progression.
mESCs were cultured under feeder-free conditions with Leukemia inhibitory factor (LIF) and serum and were treated with Wnt3a or with BIO. This successfully stabilized β-catenin (S3A and S3B Fig) and induced increased expression of Axin2 and Sp5 target genes (S3C Fig) in a dose-dependent manner. Treatment of mESCs with 0,15% DMSO, used as BIO and CHIR99021 vehicle, did not induced mESC differentiation, as shown by alkaline phosphatase (AP) staining (S3D Fig), neither affected mESC proliferation (Fig 3B). After 3 days of treatment with Wnt3a or BIO, mESCs formed packed colonies with mainly smooth boundaries (Fig 3A and 3B), a characteristic morphology induced by stabilized β-catenin in mESCs [42]. mESC colonies formed in Wnt3a or BIO containing medium were smaller as compared to the colonies formed without these drugs. We observed a significant reduction in total cell number upon 300 ng/ml Wnt3a treatment of mESCs for 48 and 72 hours but not upon 150 ng/ml Wnt3a treatment (Fig 3A). Accordingly, BIO treatment also reduced cell number at concentrations of 2 and 3 μM (Fig 3B) suggesting either Wnt-dependent inhibition of mESC proliferation or Wnt-induced apoptosis.
No significant differences in cell viability and Annexin-V staining were observed in BIO-treated compared to DMSO-treated cells, therefore excluding occurrence of Wnt-induced apoptosis (S3E and S3F Fig). In contrast, after culturing BrdU-labelled mESCs in BIO or DMSO for 72 hours, we observed a reduction of BrdU+ cells in mESCs treated with 2 and 3 μM BIO compared to DMSO-treated cells (Fig 3C). When mESCs were cultured in serum+LIF+DMSO, ~ 55% of cells were in the S phase. However, administration of 2 and 3 μM BIO significantly reduced the number of cells in the S phase, and caused accumulation of cells in G1 (Fig 3D and S3G Fig). The frequency of cells in G2 was similar by comparing BIO-treated and untreated cells. The increased number of cells in G1 was reflected by the significantly lower cycling index, [(S+G2M)/G0G1], of BIO-treated cells compared to untreated cells (Fig 3E). To further validate these results we introduced the Fluorescence Ubiquitination Cell Cycle Indicator (FUCCI) into mESCs [43]. The FUCCI system provides for direct fluorescent visualization of mESCs in G1 phase. As expected, the number of fluorescent mESCs in basal conditions was low in accordance with a very short G1 phase [44]. Interestingly, the number of mESCs activating the G1 phase reporter largely increased upon Wnt pathway activation with Wnt3a or CHIR99021 (Fig 3F and S3H Fig).
All together, these results show that Wnt pathway activation and β-catenin stabilization in mESCs determines an increased number of cells in G1 and a reduced number in the S phase, suggesting an unexpected activity of the canonical Wnt pathway as a negative regulator of proliferation in mESC.
The increased number of mESCs in G1 induced by Wnt activation (Fig 3D and 3F) might be detrimental for pluripotency given that a short G1 phase has been associated with pluripotent state [30,31]. To determine the long-term effects of BIO treatment and Wnt activation on mESC pluripotency and self-renewal we cultured mESCs with 2 and 3 μM BIO for 8 passages and analysed them at population level. At each passage, the cells were counted and the same number of cells was re-plated to calculate the growth rate and the cell doubling time. BIO-treated cells showed reduced cell proliferation during the 8 passages (Fig 4A). The doubling time was increased from 13,2 hours in untreated cells to 18 and 26 hours observed in in 2 and 3 μM BIO-treated cells, respectively (Fig 4B). After 8 passages in BIO-containing medium, the number of mESCs in G1 increased while those in the S phase decreased, as compared to untreated cells (S4A Fig).
At passage 8, pluripotency and differentiation markers were analysed by quantitative RT-PCR (qRT-PCR). No significant changes in the expression level of pluripotent markers (Oct4, Nanog and Rex1) were detected between cells cultured in BIO or DMSO containing media. However, expression of lineage differentiation markers, such as, Fgf5, Pax6, Otx2, Foxa2 or Sox7 were significantly decreased in BIO-treated cells (Fig 4C). We further confirmed these results by growing 8 independent mESC colonies in DMSO or BIO for 8 passages. mESC clones treated with BIO showed and increased expression of Wnt target genes (Sp5, T, Axin2, Cdx1 and Eomes) [38,45] and equal or reduced expression of many differentiation markers analyzed. Pluripotent markers such as Nanog or Esrrb did not show differential expression after the treatment. We observed a slight reduction of Oct4 expression, which has been demonstrated to correlate with a robust pluripotent state [46] and slight Sox2 increase (S4B Fig).
To investigate whether the effects of BIO on cell cycle length and expression of differentiation markers were reversible, mESCs cultured in BIO containing medium for eight passages were cultured for an additional 8 passages without BIO (No-BIO). After 16 passages (8 BIO + 8 No-BIO) mESCs reverted to a cell cycle and gene expression profile of control cells, i.e. those which had never been treated with BIO (S4C and S4D Fig).
These results indicate that BIO treatment increases the fraction of mESCs in the G1 phase of the cell cycle, thereby significantly increasing the cell doubling time, but it does not affect the expression of pluripotency markers. In addition, while BIO treatment increases the expression of direct Wnt-target genes associated with mesoendoderm differentiation (such as T and Eomes), most of analysed differentiation marker genes, which are not direct Wnt target genes, appeared as unchanged or show reduced expression. This result is in agreement with the previously demonstrated Wnt activity to maintain pluripotency and limit lineage priming [13].
To validate that the results obtained by inhibition of GSK3 were due to β-catenin stabilization and not to other GSK3-dependent cellular substrates [47], we generated three different mESC lines expressing stable β-catenin (ESCs-β-cat–OE). These lines displayed increased levels of stabilized β-catenin protein as well as increased levels of known Wnt/β-catenin target genes such as Axin2 and Sp5 (Fig 5A and 5B).
Like Wnt or BIO treated mESCs, β-cat-OE mESC clones formed smaller and more densely packed colonies as compared to WT cells (Fig 5C). Also the number of cells was significantly reduced over three passages (Fig 5C). Furthermore, after culturing both control and β-cat-OE mESC with EdU, we observed a reduction of EdU+ cells in β-cat-OE mESC clones compared to the control cells (Fig 5D). Cell cycle analysis of β-cat-OE mESC clones showed a significant reduction of the number of cells in S phase and in parallel an increase of the cells in G1 phase (S5A Fig). In addition, in ESCs-β-cat-OE clones compared to WT lines we observed a significant increase in transcript and protein levels of the Ink4 and Arf family members (Fig 5A and 5B) and no change in the expression of markers of cell pluripotency (S5B Fig). Thus, constitutive induction of β-catenin is correlated with an increased expression of Ink4 and Arf family members and a reduced mESC proliferation.
To validate and assess if the effects of Wnt activation on mESC cell cycle require β-catenin, we performed β-catenin loss of function experiments by using β-catenin Knock-out (KO) and Knock-down (KD) cells. In basal conditions, control (β-catfl/fl) and β-catenin KO cells (β-catΔ/Δ) [48] showed a comparable proliferation rate (Fig 5E). As expected, the known Wnt targets Axin2 or Sp5 were not activated in β-catΔ/Δ cells after Wnt3a or CHIR99021 treatment (S5C Fig). Interestingly, we observed a decreased proliferation of Wnt3a- or CHIR99021- treated β-cat fl/fl cells but not of β-catΔ/Δ cells (Fig 5F).
In addition, short hairpins against different regions of β-catenin were used to generate three distinct mESC lines wherein β-catenin was knocked down (shβcat) (S5E Fig). We treated shControl and shβcat mESCs with 1, 2 and 3 μM BIO and counted the cells after 72 hours by FACS. KD-shβcat cells displayed only a small decrease in cell number when treated with 2 and 3μM BIO while shControl cells showed a drastic reduction (S5E Fig). These results indicate that β-catenin is essential to regulate cell number upon GSK3 inhibition.
Next, we investigated the expression of Ink4 and Arf family members in β-catenin KO and KD cells. In β-catenin KO and KD cells, there was no increase in the expression of p16Ink4a, p19Arf, p15Ink4b and p18Ink4c as compared to respective control cells after GSK3 inhibition (S5D and S5F Fig). In addition, we observed significant upregulation of the protein level of p16Ink4a and p19Arf (Fig 5G) after BIO treatment in control β-catfl/fl but not in β-catΔ/Δ cells. All together these results indicate that the expression of the Ink4/Arf genes is dependent on β-catenin in mESCs.
As indicated above, Ink4 and Arf family members are targets of Tcf1 but not of Tcf3 in mESCs (S1 Table and S2 Table). Interestingly, following treatment with BIO, the proliferation of mESCsTcf3-/- cells was comparable to the proliferation of wild type (WT) cells, suggesting that the activity of Tcf3 is not required for the regulation of cell proliferation in response to canonical Wnt signalling (Fig 5H). Moreover, the expression of Ink4 and Arf family members increased in BIO-treated mESCsTcf3-/- (S5G Fig), excluding that Tcf3 is required for Wnt-dependent cell cycle regulation of mESCs.
All together these data demonstrate that GSK3 inhibition or β-catenin stabilization both transcriptionally regulate the expression of the Ink4 and Arf family members in a Tcf3-independent manner.
To further investigate the role of Tcf1 in Wnt-dependent cell cycle regulation in mESCs, we used Tcf1 KD cell lines (shTcf1 mESC) [49], wherein Tcf1 RNA levels were reduced by 70% (S6A Fig). Decreased expression of Tcf1 did not impaired the expression of pluripotent markers [49]. We treated shScrmbl and shTcf1 mESCs with BIO for three passages. We observed no difference in the proliferation of shScrmbl and shTcf1 mESCs when cultured with serum+LIF+DMSO. However, addition of BIO induced a reduction in the proliferation of shScrmbl compared to shTcf1 mESCs (S6B Fig). Importantly, upon addition of BIO, Ink4 and Arf genes were not activated in shTcf1 compared to shScrmbl cells (S6A Fig).
We then generated Tcf1 KO mESCs (S6C Fig) using CRISPR/Cas9 technology to further investigate the role of Tcf1 in regulating the expression of Ink4 and Arf family members in mESCs. Deletion of Tcf1 did not affect pluripotent gene expression in mESCs (S6D Fig) as expected [49], and in contrast with another report [50].
BIO treatment for 24h and 48h reduced the cell number in WT mESCs but not in mESCsTcf1-/- (Fig 6A). In WT cells, BIO enhanced expression of Axin2 as well as that of p15Ink4b, p16Ink4a and p19Arf. In contrast, after BIO treatment of three different mESCTcf1-/- clones we observed no increase in the expression of Ink4 and Arf family members (Fig 6B and S6E Fig). p16Ink4a and p19Arf protein levels also increased after BIO treatment in WT mESCs but not in mESCsTcf1-/- (Fig 6C).
Finally, we increased the expression levels of Tcf1 in mESCs using CRISPRa technology [51] (S7 Fig). The three-fold increase of Tcf1 in sgRNATcf7 cells (Fig 6F and S6F Fig) did not have any effect on cell number or expression of pluripotency markers when cells were cultured in serum+LIF (Fig 6D and S6F Fig). However, activation of the Wnt pathway by BIO further increased endogenous Tcf1 expression in sgRNATcf7 and controls (Fig 6F and S6F Fig). The combination of Wnt pathway activation with Tcf1 overexpression induced a strong increase in the expression of Ink4 and Arf family members compared to both DMSO-treated samples as well as to control cells (Fig 6F and S6F Fig). Interestingly, Wnt pathway activation along with Tcf1 overexpression resulted in a strong reduction in cell number (Fig 6E).
To investigate if the genes encoded by the Ink4/Arf locus, p16Ink4a and p19Arf, were the main downstream players of the Wnt-dependent reduced proliferation of mESCs, we infected mESCs with retroviruses carrying the KD for p16Ink4a or p19Arf. Specific KD for p16Ink4a or p19Arf reduce their protein levels in mESCs after treatment with BIO (Fig 7A). Cell number was reduced significantly after 24 and 48 hours of BIO treatment in control cells, however no differences in the cell number were observed in BIO treated shp16Ink4a and shp19Arf mESCs compared to control cells (Fig 7B). A rescue of the proliferative phenotype was observed in p19Arf KD infected cells at 24 and 48 hours after BIO treatment, while p16Ink4a KD can rescue the phenotype only 48 hours after BIO treatment (Fig 7B).
These results show that the Knock-down of p16Ink4a or p19Arf abolish the Wnt-dependent inhibition of cell proliferation in mESCs indicating that p16Ink4a or p19Arf are the major players downstream to the Wnt pathway to regulate cell cycle in mESCs.
Our findings show that Tcf1 is required to regulate Wnt-dependent Ink4 and Arf family expression, and this results in the decrease of mESC proliferation. However, reduced expression of Tcf1 has no effect on mESC pluripotency or differentiation [49], in contrast to Tcf3 [12,13], indicating that different Tcf family members mediate divergent functions of the Wnt pathway in mESCs (Fig 7C).
The Wnt pathway has important roles during early development, being activated from the two-cell stage [52] until the pre-implantation blastocyst, and becoming inhibited during post-implantation [38]. In morula stage bovine embryos, ectopic activation of the Wnt pathway inhibits development to the blastocyst stage and this is associated with a significant reduction in total cell number [53]. This observation is in accordance with our findings in mESCs. Given that the activation of the Wnt pathway maintains the pluripotency of mESCs [2], and as we have demonstrated here that it also reduces mESC proliferation, it is reasonable to hypothesise that inhibition of Wnt during the progression of the pre- to post-implantation blastocyst [38] is required to allow cells of the inner cell mass to exit the pluripotent state and start proliferation to produce lineage committed cells.
In the experiments carried out, we activated the Wnt/β-catenin pathway in mESCs using purified Wnt3a and GSK3 specific inhibitors as CHIR99021 and BIO. While these treatments can reduce proliferation of mESCs and increase G1 phase, Wnt3a treatment used at a concentration of 150ng/ml does not induce a significant effect on cell number as also previously reported [14,38]. This is likely due to the fact that high levels of β-catenin stabilization are needed in order to increase p16Ink4a and p19Arf protein levels and therefore to reduce mESC proliferation (S3B Fig). Low levels of purified Wnt3a, as a concentration of 100–150 ng/ml, might be sufficient to maintain mESC pluripotency [14,38] but not enough to induce effects on the cell cycle.
The activity of CDK/cyclin complexes, which are controlled by the expression of CDKI, regulate the transition from one cell cycle phase to another. Contrary to the generally accepted believe that a short G1 phase in mESCs could act as a brake for differentiation [20], we here found that upon Wnt activation, key cell cycle regulators are expressed in mESCs with a consequent increase in the number of cells in G1, which show a prolonged doubling time. Importantly, this is not coupled with a reduced expression of pluripotency genes. We here show that activation of Wnt pathway can increase expression of both p16Ink4a and p19Arf at transcriptional and protein level. Previously, it has been shown that mESCs are refractory to p16Ink4a regulation when overexpressed [21,24]. In the present study we overexpressed p16Ink4a or p19Arf in mESCs and selected resistant individual clones. We have shown that all the p19Arf overexpressing clones and four out of the six p16Ink4a overexpressing clones have a reduced proliferation rate. The fact that two out of six p16Ink4a clones were resistant to the overexpression of p16Ink4a and did not reduce their proliferation might suggest that, under certain circumstances, mESCs activate mechanisms to become insensitive to cell cycle regulation as also reported in other studies [21,24].
Finally, the activation of Wnt pathway induces somatic cell proliferation by activating transcription of c-Myc and CyclinD1 [16], while it restricts the cell cycle in pluripotent cells by activating negative cell cycle regulators and reducing c-Myc transcript and protein levels. It is therefore clear that the activation of the Wnt pathway results in opposite outcomes on the proliferation of somatic and pluripotent stem cells. Wnt-induced p19Arf expression in mESCs leads to increased expression of nuclear p53 protein levels. On the other hand, it has been previously shown that p53 binds to the c-Myc promoter and repress its transcription [54]. This might be the reason of the reduced expression of c-Myc in mESCs after Wnt pathway activation. Moreover, it has been recently shown that absence of c-Myc and n-Myc expression induces dormant state in mESCs [55] pointing out the important role of Myc family members on the cell cycle regulation of pluripotent cells.
The identified Tcf1 recruitment DNA motif in mESCs is not a canonical WRE motif. However, it was previously shown that some Tcf1 isoforms could be recruited to alternative C/G-rich DNA binding motifs [9,56]. Furthermore, the Tcf1 motif we identified in mESCs shares the same sequence of KAISO/ZBTB33 in adipocyte-specific promoters [57] or DYRK1A in glioblastoma cell line [58]. Furthermore, Tcf3 in hair follicle cells and Tcf4 in oligodendrocytes were shown to be able to be recruited to the KAISO binding site [35,59] indicating that Tcf/Lef factors can associate with a number of distinct DNA binding domains to regulate gene expression. KAISO was shown to regulate the cell cycle in preadipocytes [57]. Taking into account all these previous observations, it will be important to pursue further investigation on a possible KAISO/Tcf1 coordinated activity in mESCs.
The Wnt pathway acts during evolution starting from metazoans. However, higher organisms present an expanded number of the components of the pathway having four Tcf/Lef members differently from invertebrates that only have one Tcf [9]. Here we show that two different Tcf/Lef factors regulate distinct target genes and control distinct cellular functions in mESCs. Tcf3 regulates self-renewal, potency and lineage priming in mESCs. The expression of Tcf1 does not affect pluripotency. However, Tcf1 regulates mESCs proliferation while Tcf3 does not. All these observations indicate that Tcfs might not be redundant and can regulate context-specific responses of Wnt signalling by activating the expression of different target genes (Fig 7C). Our observations in embryonic stem cells open the path to investigate whether Tcf/Lef factors exert specialized functions also in adult stem cells. Indeed, the Wnt pathway was shown to control both potency and proliferation in hematopoietic and intestinal stem cells, however, whether this is due to the activity of different Tcf factors is not clarified.
Activation of the Wnt pathway as well as transcriptional repression of Tcf1 has been broadly associated with tumour formation [60–62]. Finally, whether the Wnt/Tcf1 pathway also directly controls the regulation of cell cycle and tumor suppressor genes in cancer stem cells will need further investigation. However, it has already been demonstrated that knock-out of the TCF1 gene in mice leads to intestinal tumors as well as highly metastatic thymic lymphomas [60–62], suggesting that Tcf1 is a tumor suppressor gene per se. In line with this notion, activation of the Wnt pathway reduces cell proliferation in melanocytes and melanoma [63,64].
R1 and E14Tg2 mouse ESCs were maintained feeder-free on gelatin- (EmbryoMax 0.1% Gelatin Solution, ES-006-B; Millipore) coated plates in DMEM (41965–039 Gibco), 15% fetal bovine serum (Sigma), 2 mM L-glutamine (25030–024; Gibco), 1X minimal essential medium non-essential amino acids (Gibco), penicillin (100 U/ml) /streptomycin (100U/ml) (15140122; Gibco), 100 μM β-Mercaptoethanol (Gibco) and 1,000 U/ml recombinant mouse leukemia inhibitory factor (ESG1107; ESGRO, Chemicon International). mESCs were treated at indicated concentrations to activate the Wnt pathway: purified Wnt3a (315–20; Peprotech); BIO (361550; Calbiochem); CHIR99021 (361571; Calbiochem). BIO and CHIR99021 were resuspended in DMSO (Sigma) at a stock concentration of 2 mM (BIO) and 6 or 10 mM (CHIR99021), Wnt3a (Peprotech) was resuspended at a stock concentration of 50ng/μL following manufacturer instructions.
For mESC infection, lentiviral particles were produced following the RNA interference Consortium (TRC) instructions for lentiviral particle production and infection in 6-well plates (http://www.broadinstitute.org/rnai/public/). Briefly, 5 ×105 HEK293T cells/well were seeded in 6-well plates. The day after plating, the cells were co-transfected with 1 µg specific lentivirus construct, 750 µg pCMV-dR8.9, and 250 µg pCMV-VSV-G, using Polyfect reagent (Qiagen). The day after transfection, the HEK293T culture medium was substituted with the ESC culture medium. Then 5 ×104 ESCs/well were plated onto gelatin-coated 6-well plates the day before transduction. The lentiviral-containing medium was harvested from HEK293T cells at 48, 72 and 96 h after transfection, filtered, and added to the ESC plates. The day after infection, these mESCs were washed twice in PBS and cultured with normal medium.
Lentiviral constructs for mouse p16Ink4 and p19Arf Knock-Down (PIGΔRI-p16Ink4a and PIGΔRI-19Arf) were generously provided by Scott Lowe and Manuel Serrano laboratories. Retroviral constructs for mouse p16Ink4a and p19Arf overexpression (pLPC-puro-p16Ink4a, pLPC-puro-p19Arf) were generously provided by Manuel Serrano laboratory.
ChIP was carried out as described in [67]. Briefly, ESCs were trypsinised and crosslinked in 1% formaldehyde at room temperature for 10 min. Crosslinking was quenched with 0.125 M glycine for 5 min. The pelleted cells were lysed in 1 ml ChIP buffer and sonicated in a Bioruptor sonicator (Diagenode) for 10 min. The soluble material was quantified using Bradford assays. To immunoprecipitate the transcription factors, 500 μg protein was used. Antibodies were incubated with the chromatin overnight. The immunocomplexes were recovered with 30 μl protein A or G agarose bead slurries. The immunoprecipitated material was washed three times with low-salt buffer and one time with high-salt buffer. DNA complexes were decrosslinked at 65°C for 3 h, and the DNA was then eluted in 200 μl water using the PCR purification kit (QIAGEN). Two microliters DNA were used for each qPCR reaction. Antibodies used were: Tcf1 (C46C7, Cell Signalling); Tcf3 (sc-8635, Santa Cruz); rabbit IgG (Sigma) and Goat IgG (Santa Cruz).
ChIP-seq reads were barcode-sorted, checked for quality control using Fastqc (http://www.bioinformatics.babraham.ac.uk/projects/fastqc/) and Chance (https://github.com/songlab/chance).
Quality-controlled reads were aligned to the latest mouse genome available (mm10, Genome Reference Consortium GRCm38) using the Bowtie2 version 2.2.0 (http://bowtie-bio.sourceforge.net/bowtie2/index.shtml). Reads with low mapping quality (minimum mapping quality (–q) > = 10) and PCR duplicates were removed using Rmdup from the Samtools (http://samtools.sourceforge.net/) suite. Finally, SAM files were checked and converted to BAM files using Picard (http://broadinstitute.github.io/picard/).
ChIP-seq raw data were submitted to ArrayExpress:
Experiment ArrayExpress accession: E-MTAB-4358
Peaks were called using GEM (http://groups.csail.mit.edu/cgs/gem/) [65] high resolution peak calling algorithm with significance level for q-value 1, specified as -log10(q-value) and without the default noise distribution model. We included 1,5 fold enrichment over the control as significant. This allowed us to exclude regions with low signal-to-noise ratios, while including regions that proved reproducible based on ChIP-qPCR even if their overall enrichment was only low to moderate.
Annotated mouse REfSeq genes with a peak at their promoter proximal (±2kb of the transcription start site, TSS) were considered as target. ChIP-seq signal track were visualized by IGV (The Integrative Genomics Viewer).
Gene ontology was analysed using Enrichnet.
RNA was extracted and purified using Maxwell Total RNA purification kits (Promega), according to the manufacturer’s instructions.
The cDNA was produced with SuperScript III Reverse Transcriptase kits (Life Technologies) starting from 300 ng to 1 μg mRNA. Real-time quantitative PCR reactions from 8,3 ng of cDNA were set up in triplicate using a DNA SYBR Green I Master Mix (Roche) or Platinum SYBR Green qPCR SuperMix-UDG (Thermoscientific) on a LightCycler machine (Roche) or ViiA 7 Real-Time PCR System (Thermoscientific) respectively. The sequences of the oligonucleotides used in this study are provided on request. Expression levels were normalized to PCR amplification with primers for Gapdh.
Statistical analyses were determined by two-tailed Student’s t-test. The 0.05 level of confidence (P<0.05) was accepted for statistical significance.
Cells were harvested and washed twice with PBS. Cell lysis was performed on ice for 25 min, in RIPA buffer (150 mM NaCl, 1% Nonidet P40, 0.5% sodium deoxycholate, 0.1% sodium dodecyl sulphate, 50 mM Tris-HCl, pH 8.0) containing a protease inhibitory cocktail (Roche). Insoluble material was pelleted by centrifugation at 16,000× g at 4°C for 3 min. Protein concentrations were determined using the Bradford assay (Bio-Rad). Thirty micrograms extract was mixed with 4× Laemmli buffer (40% glycerol, 240 mM Tris/HCl, pH 6.8, 8% SDS, 0.04% bromophenol blue, 5% β-mercaptoethanol), denatured at 96ºC for 5 minutes, separated by SDS-PAGE, and transferred to nitrocellulose membranes (PROTRAN-Whatman, Schleicher&Schuell). The membranes were blocked with 5% non-fat dry milk in TBS-T for 60 min, incubated with primary antibodies overnight at 4°C, washed three times with TBS-T for 10 min, incubated with the peroxidase-conjugated secondary antibody (1:2000; Amersham Biosciences) in TBS-T with 5% non-fat dry milk for 60 min, and washed three times with TBST for 10 min. Immunoreactive proteins were detected using Supersignal West Dura HRP Detection kits (Pierce). The primary antibodies used were: p16Ink4a (Santa Cruz); p19Arf (Ab80 Abcam); p15Ink4b (Santa Cruz); β-catenin (clone 14, BD Biosciences); p53 (sc-6243 Santa Cruz); p21 (BD Pharmigen); c-Myc (sc-764 Santa Cruz); β-actin (ab8226, abcam).
Two single-guide RNAs (sgRNA+1 [5’-TGCCGCAGCTGGACTCGGG-3’] and sgRNA+1027 [5’-GCTCCGGAGGCCGGTGGGTA-3’]), targeting the first and the third exon of Tcf1 (+1 nd +1027bp from ATG), respectively were cloned separately into pSpCas9(BB)-2A-Puro (PX459). The constructs were co-transfected transiently in mESCs and 24 hours after transfection puromycin selection was applied for an additional 48 hours. Cells were then split and seeded at clonal density. Clones from single cells were manually picked, and analyzed by Western blot for the expression of Tcf1.
pSpCas9(BB)-2A-Puro (PX459) was a gift from Feng Zhang (Addgene plasmid # 48139) [66].
Activation of endogenous Tcf1 promoter was achieved by using a catalytically inactive Cas9 (dCas9) fused to 4 repetition of Vp16 (Vp64) transcriptional activator.
sgRNA were designed using E-crispr online software (http://www.e-crisp.org/) against a region of DNA spanning -400 to -50 bp from TSS of Tcf1.
One sgRNAs was selected: sgRNA Tcf1a1 (5’-GAAGCCTCCAGATTGAGCAA-3’) at -310 from TSS. sgRNA was cloned into pLKO u6 Puro.
Briefly, E14Tg2 mESCs were infected with dCas9Vp674_eGFP and GFP+ cells were FACS-sorted 72hrs after infection to obtain a stable pool expressing dCas9-Vp64. Control cells (infected with dCas9-Vp64+pLKO sgRNA empty vector) and Tcf1 overexpressing cells (infected with dCas9+Vp64+sgRNA Tcf1A1) were selected with puromycin for 72hrs and assessed by qRT-PCR for Tcf1 expression levels.
pLKO.1-puro U6 sgRNA BfuAI large stuffer was a gift from Scot Wolfe (Addgene plasmid # 52628) and dCAS9-VP64_GFP was a gift from Feng Zhang (Addgene plasmid # 61422).
pCF823, pLenti hEF1a-βcatenin4A//SV40-PuroR construct (E[beta]P), containing an unphosphorylatable form of β-catenin (S33A, S37A, T41A and S45A), and vector backbone pRRLSIN-(E(i)P) were used to produce lentivirus particles to infect R1 mESCs. The day after infection, cells were tripsinized and replated to single-cell confluency. Puromycin selection was applied for 4 days and resistant clones were selected and grown individually. Clones displaying high levels of stabilized β-catenin were selected and used for gene expression analysis and growth curve experiments.
E[beta]P was a gift from Roel Nusse (Addgene plasmid # 24313).
For cell counts by hemocytometer, cells were seeded at a uniform density (usually between 25,000 to 40,000 cells per 6 well plate) in the appropriated media. Treatment with Wnt3a, BIO or CHIR99021 was initiated 24 hours after seeding. Cell proliferation of mESCs was assessed by counting the respective cell number in 10μl cell suspension stained with 0,4% trypan blue solution (Sigma) in a Neubauer chamber. For cell counts by FACS cells were trypsinized, diluted in serum containing media and propidium iodide (PI) to detect dead cells. Diluted cells were plated in 96-well plates and counted using FACScanto. For cell growth analysis during several days, mESCs were counted at 48 or 72 hours and replated at the same number for the following days. The total number of cells at each passage was calculated multiplying the number of cells by the product of the previous dilution factors. Exponential growth curves were calculated setting the intercept equal to the number of cells plated at day 0 (pc) and the growth rate (gr) was used to calculate the doubling time (dt).
Statistical analyses were determined by the unpaired two-tailed Student’s t-test unless indicated in figure legend. The 0.05 level of confidence (P<0.05) was accepted for statistical significance.
ES cells were pulse-labeled with 10μM BrdU for 60 min before harvest. Cells were fixed with absolute Ethanol for at least 2 hours. Cells were then washed with PBS+0,5%BSA followed by 15’ incubation of freshly prepared denaturing solution (1ml = 700μl of 0,7%BSA in PBS+ 300μl 25%HCL). After another washing, cells were incubated with (PBS+0,5%BSA+0,5%Tween-20) for 5’. Next, cells were incubated with anti-BrdU antibody conjugated with FITC or with isotype control antibody (BD Pharmingen, 556028) in the dark for 60 minutes. Cells were washed twice with PBS+0,5%BSA and incubated with Propidium Iodide for 30 minutes at RT. Cells were then analyzed by flow-cytometry. ModFit was used as analysis software.
The cycling index was calculated by adding the percentages of cells in S and G2/M phases and dividing them by the percentage in G0/G1 phase (S+G2M)/G0G1.
Statistical analyses were determined by the two-tailed Student’s t-test. The 0.05 level of confidence (P<0.05) was accepted for statistical significance.
Non-synchronized ES cells were pulse-labeled with 10μM 5-ethynyl-2′-deoxyuridine (EdU, Life Technologies) for 40–60 min. Cells were fixed with 4%PFA for 15 minutes, washed with PBS+2%BSA followed by 15’ permeabilization with PBS+0,5% Triton. Cells were further processed using the Click-IT EdU 555 Imaging kit to reveal EdU incorporation, according to the manufacturer’s instructions, and stained with Propidium Iodide (Life Technologies).
24h after plating, mESC cells were treated with DMSO 0,15% or 3μM BIO or Puromycin 0,4μg/mL. Cells were collected and analyzed every 6h after treatment for 48h. Supernatant and trypsinized cells from each time-point were collected, washed (2x DPBS) and counted. 1x106 cells/mL were stained with 1uL of the BD Horizon Fixable Viability Stain 660 (stock solution) for 12 minutes at room temperature in the dark. Cells were washed twice with 1x DPBS+2%FBS, and fixed (4%PFA) for 15 minutes at room temperature. Viable cells were analyzed in the FACS Canto I. Dot plots and histograms were analyzed by FlowJo v.10 software. As positive technical control of cell death cells were incubated at 65°C for 15 minutes before staining.
At the indicated incubation time, floating cells were collected together with the supernatant and adherent cells were harvested by trypsinization. Cells were sedimented by centrifugation, counted and 1x106 cells were resuspended in 1 ml of 1x binding buffer (BD Bioscences). Subsequently, 3 μl Annexin-V-APC (BD Biosciences) was added to 100 μl of cell suspension followed by gently vortexing and incubation for 10 min at room temperature in the dark. Thereafter, DAPI was added. Cells were analyzed immediately using a FACS flow cytometer for Annexin-V and DAPI binding. Dot plots and histograms were analyzed by FlowJo v.10 software.
|
10.1371/journal.pgen.1003952 | Deletion of an X-Inactivation Boundary Disrupts Adjacent Gene Silencing | In mammalian females, genes on one X are largely silenced by X-chromosome inactivation (XCI), although some “escape” XCI and are expressed from both Xs. Escapees can closely juxtapose X-inactivated genes and provide a tractable model for assessing boundary function at epigenetically regulated loci. To delimit sequences at an XCI boundary, we examined female mouse embryonic stem cells carrying X-linked BAC transgenes derived from an endogenous escape locus. Previously we determined that large BACs carrying escapee Kdm5c and flanking X-inactivated transcripts are properly regulated. Here we identify two lines with truncated BACs that partially and completely delete the distal Kdm5c XCI boundary. This boundary is not required for escape, since despite integrating into regions that are normally X inactivated, transgenic Kdm5c escapes XCI, as determined by RNA FISH and by structurally adopting an active conformation that facilitates long-range preferential association with other escapees. Yet, XCI regulation is disrupted in the transgene fully lacking the distal boundary; integration site genes up to 350 kb downstream of the transgene now inappropriately escape XCI. Altogether, these results reveal two genetically separable XCI regulatory activities at Kdm5c. XCI escape is driven by a dominant element(s) retained in the shortest transgene that therefore lies within or upstream of the Kdm5c locus. Additionally, the distal XCI boundary normally plays an essential role in preventing nearby genes from escaping XCI.
| Early in mammalian female development, one X chromosome is largely silenced to equalize X-linked gene expression between the sexes. Nevertheless, some genes “escape” this silencing and therefore are expressed from both X chromosomes. Understanding how these escape genes are regulated, particularly when they closely juxtapose silenced genes, may give important insight into regulatory transitions throughout the genome. To evaluate sequences that are essential for appropriate inactive X expression we analyzed large transgenes that integrated on the X chromosome in mouse embryonic stem cells. Transgenes that include an escape gene, Kdm5c, but lack all or part of the downstream sequences, including the X-inactivation boundary, still escape X inactivation. Nevertheless, downstream genes at the transgene insertion site are misregulated and now inappropriately escape X inactivation as well. These data identify two important regulatory activities at this locus. First, sequences retained within the truncated transgene are sufficient to direct the Kdm5c gene to escape X inactivation. Further, we have uncovered a function for an X-inactivation boundary in protecting adjacent genes from escape.
| Recent annotation of the human and mouse genomes has revealed chromosome domains that are distinguished by sequence and gene content, regulatory-factor binding, replication dynamics, chromatin composition, or nuclear location. Many of these domains overlap and can functionally segregate active and inactive transcripts [1]–[4]. What regulates such extensive genome compartmentalization is not fully understood. Intriguingly, many boundaries share common features including opposing chromatin marks, active transcription, or binding by the CCCTC binding factor, CTCF [1], [4]–[6]. Whether these elements are essential for segregating domains has not been thoroughly examined, yet boundary deletion can lead to misregulation (e.g. [7]).
An interesting example of partitioned, closely juxtaposed, active and inactive transcripts is found on one X chromosome in female mammals. This X is largely silenced during early embryonic development in order to balance dosage between the sexes. X-chromosome inactivation (XCI) is mediated by the cis-limited action of Xist, a structural RNA that coats the X chromosome and recruits inactive chromatin modifiers [8]. Nevertheless, XCI is not chromosome-wide, as some genes “escape” inactivation [9]. Current understanding of how genes escape XCI on an otherwise silenced chromosome is incomplete, but the answer may reveal novel insights about regulatory sequences not only at XCI boundaries but also at other expression transitions throughout the genome.
Escape and X-inactivated genes are epigenetically and structurally distinct [9]. Escape genes are depleted in Xist RNA and promoters are marked by active histone modifications and lack silent epigenetic marks associated with X-inactivated transcripts (e.g. [10]–[12]). However, long-range regulation is likely involved, as many escape genes, particularly in humans, are physically clustered [13], [14]. Further supporting this idea, unique sequence composition distinguishes these domains relative to the rest of the X [15], [16]. Distant escapees also frequently interact on the inactive X [17] and can be spatially separated from silent inactive X regions [18].
To functionally delimit sequences sufficient to confer XCI escape, we previously developed a transgene approach in female mouse embryonic stem (ES) cells, a well established ex vivo XCI model [19]. X-linked BAC transgene lines were isolated that carry the escapee Kdm5c (previously Jarid1c) that encodes a histone H3K4 demethylase [20], [21]. The BACs also included an adjacent long non-coding RNA (lncRNA) AK148627 that escapes XCI [14], [22] and flanking X-inactivated genes [16], [19], [23]. Endogenous expression patterns examined were maintained including transgenic Kdm5c (Kdm5c-tg) escape at four ectopic X-chromosome locations. Therefore, these BACs must include sequences necessary for Kdm5c to escape XCI. What features at this locus direct XCI escape? Plausible candidates include CTCF and the AK148627 lncRNA, as both CTCF and lncRNAs are found at a number of XCI boundaries [14], [22], [24]. Further, such elements are enriched at other boundaries throughout the genome (e.g. [1], [4], [5]), and can function to regulate adjacent genes in cis [25], [26]. Intriguing associations notwithstanding, both candidates lack functional validation. To better understand the role of boundary sequences in inactive X regulation we now extend our analysis of Kdm5c BAC transgenes. We further narrow sequences necessary for XCI escape and identify a novel role for XCI boundary sequences in regulating inactive X expression.
Previous studies focused on four full-length BAC transgenes that were derived from two overlapping BACs [19] (Figure 1). However, by PCR analysis of BAC-backbone sequences, six additional female ES lines carry X-linked integrants of the BAC RP23-391D18 with partial deletions. We turned to these truncated transgenes to further delimit sequences that dictate XCI states at the Kdm5c locus.
To determine transgene content and copy number, we exploited allele differences between the 129 and M.m. castaneous (CAST) X chromosomes in the ES cell line and assayed for the presence or absence of an additional BAC-transgene allele. Allele ratios for up to 18 SNPs across the region were measured using a quantitative primer-extension assay, qSNaPshot [13], [19], with primers that abut each SNP. The approach was validated with allele ratios of 2∶1 ((129+BAC)/CAST) for all SNPs mapping within full-length, single-copy BACs (e.g. B202) [19] (Figure 1, Figure S1). Similar analysis excluded three lines with multi-copy inserts (Figure S1). Further, the transgene in the B176 line is severely truncated and deletes the entire Kdm5c-tg.
Breakpoint analysis for two other transgene lines revealed deletions of distal XCI boundary sequences. ES lines C048 and C138 carry single-copy inserts that retain all or most of Kdm5c-tg (Figure 1). The C048 transgene contains the AK148627 lncRNA but deletes a large portion of non-transcribed XCI boundary sequence. The transgene in C138 is more extensively deleted as all sequences downstream of Kdm5c are removed including the lncRNA. Additional SNPs narrowed the C138 transgene breakpoint to a small ∼900 bp window and indicate that at least 90% of the Kdm5c genomic locus remains intact. Further, by RNA fluorescence in situ hybridization (FISH) a stable nascent Kdm5c-tg transcript is detected in pre-XCI undifferentiated ES cells (not shown). Therefore, the C048 and C138 transgenes lack all or part of the intervening region between the 3′ end of escapee Kdm5c and the closest X-inactivated gene and allow the role of sequences within an escape domain and at an XCI boundary to be evaluated.
Prior to examining transgene expression we surveyed the local chromosomal environment flanking the C048 and C138 BAC transgenes. By inverse PCR and subsequent analysis of an adjacent SNP, the C048 transgene inserted on the CAST X, upstream of the first coding exon of the Mid1 gene (166,290,616 bp, mm9). Importantly, Mid1 is normally X inactivated on the CAST X [19]. Additionally, FISH and SNP screening indicate that this transgene insertion was accompanied by a large and likely terminal deletion that removes the entire pseudoautosomal region (Figure S2A).
Similar characterization of the C138 transgene revealed that the BAC integrated on the CAST X at 98,065,555 bp (mm9) (Figure S2B). DNA FISH and SNP analysis near the C138 transgene integration site ensured that the BAC insertion was not accompanied by a larger chromosomal rearrangement or deletion (Figure S2B). This places Kdm5c-tg in an intron near the 3′ end of Tex11, a gene that functions in male meiosis [27], [28]. Although predominantly expressed in testis [29], we detected a low level of Tex11 expression in somatic tissues by RT-PCR; monoallelic expression of a transcribed polymorphism in female fibroblasts with non-random XCI confirms that Tex11 is normally X inactivated (Figure S2C). Therefore, both transgenes integrated into regions that are normally silenced by XCI, enabling direct testing of BAC sequence influences on Kdm5c-tg expression.
Will Kdm5c-tg still escape XCI in the absence of distal boundary sequences? Expression was examined by sequential RNA and DNA FISH upon ES cell differentiation and concomitant XCI. Non-denatured cells were hybridized with a Kdm5c BAC probe to detect nascent transcripts from the endogenous and transgenic loci. Following probe fixation, cells were denatured and hybridized for DNA FISH to demarcate all Kdm5c loci. In C138 and C048, three expressed foci were detected in most cells (Figure 2). Importantly for each line, nuclei with two RNA signals colocalizing with Xist RNA demonstrate that both endogenous and transgenic loci are expressed on the inactive X. Additional FISH for C138 directly confirmed Kdm5c-tg escape, as one inactive X transcript colocalizes with a DNA signal from a probe at the integration site (Figure S3A). RNA FISH using a smaller Kdm5c-specific probe ensured results reflect Kdm5c expression (Figure S3B). Because of genetic background differences in the ES cells, XCI is skewed and the transgene is on the inactive CAST X in ∼25% of cells [19], [30]. For both C138 and C048, the proportion of cells with two expressed Kdm5c foci from the inactive X closely mirrors the frequency that cells inactive the CAST X chromosome (Figure 2B, Figure S3). Therefore, these data indicate Kdm5c-tg escapes XCI at a frequency similar to the non-transgenic locus.
To better estimate the level of Kdm5c-tg escape in C138, we isolated a clonal line that carries the transgene on the inactive X chromosome. Allelic expression, measured by qSNaPshot, is consistent with Kdm5c-tg and the non-transgenic locus each partially escaping XCI, at levels that are ∼34% of active X expression (see methods). Such levels are in good agreement with previous reports of partial escape for the endogenous locus [18], [19], [31], [32]. These data indicate that despite BAC truncation, Kdm5c-tg is expressed from the inactive X chromosome. Altogether, we conclude that Kdm5c escape does not require distal sequences.
Previous studies of Kdm5c indicate that escape genes preferentially assume an exterior location on the Xist-coated inactive X in interphase nuclei [18]. This positioning likely facilitates more frequent long-range associations with other escape genes than with X-inactivated genes [17]. To further confirm the active state of Kdm5c-tg, we asked if transgenes establish similar interactions with distant escapees. Interactions were evaluated in differentiated post-XCI cells by FISH using three-dimensional deconvolution microscopy (Figure 3A). Inactive X distances were initially measured between the escapee Ddx3x and a probe detecting either escapee Kdm5c or an X-inactivated gene (Figure 3A,B). For each comparison, cumulative frequency plots indicate the proportion of nuclei in which two loci are closer than a given nuclear distance (normalized for area) (Figure 3B). This approach was first validated in a non-transgenic line and confirmed that profiles differ for the active and inactive X [17]; distant loci are more frequently in close proximity on the inactive X relative to their distance on the active X (Figure S4A). Further, inactive X escapee associations are also consistent with previous observations [17]. A higher proportion of nuclei have two escape loci in close proximity as the cumulative frequency plot of nuclear distances between escapees Ddx3x and Kdm5c is significantly shifted to the left relative to profiles comparing Ddx3x and either X-inactivated gene, Tex11 or Mecp2 (Figure 3B, Figure S5A). All differences were readily apparent regardless of whether or not probe distances were normalized to nuclear area (Figure S4B).
Similar probe comparisons were performed in the transgene lines. All profiles in line C048, with the Mid1-integrated transgene, were indistinguishable from the non-transgenic line (Figure 3B) indicating that a transgene at a location unrelated to the genes tested is insufficient to alter gene localization and interaction. In contrast, while C138 cumulative frequency curves comparing Ddx3x to active and inactive non-transgenic loci mirrored the other lines tested, comparison to the Tex11 BAC revealed a significant left shift (Figure 3B, Figure S5A). Tex11 lies at the C138 transgene integration site and proxies for the transgene in cells that inactivate the transgenic X. Indeed, the Tex11 BAC is frequently located near Ddx3x on the inactive X, with a profile that is more similar to plots comparing two escapees than to curves for genes with differing XCI states, e.g. Ddx3x and Mecp2. These data suggest that a transgene can reconfigure associations on the inactive X.
To more directly visualize transgene interactions we specifically scored transgenic inactive X associations between Kdm5c-tg and the endogenous Kdm5c locus. Compared to interactions with X-inactivated Tex11 (measured on non-transgenic inactive Xs), Kdm5c more frequently lies in close proximity to the transgene in C048, C138, and the full-length B202 transgene (Figure 3C, Figure S5B). In contrast, profiles for the severely truncated transgene in B176 resemble those with X-inactivated locus Tex11 (Figure 3C, Figure S5B). Such a profile likely reflects the absence of Kdm5c-tg transcript in this line and indicates that the partial proximal boundary sequences retained in B176 are insufficient to direct interactions with escape loci. Importantly, these studies demonstrate that Kdm5c-tg in C138 and C048 structurally interacts in a manner similar to the endogenous locus, further confirming the active state of the transgenes on the inactive X. Therefore, despite truncating the endogenous escape domain, retained sequences are sufficient to induce an altered inactive X conformation even when inserted at a different chromosomal location.
We previously established that the full-length BAC transgenes retain intact XCI boundaries as Kdm5c-tg is expressed, but adjacent transgenic Tspyl2 or Iqsec2 properly undergo XCI [19]. Therefore, we next sought to determine if transcripts near the integration site would remain silent despite the absence of distal boundary sequences (Figure 4A). Given the orientation and close proximity of the C048 transgene to the pseudoautosomal boundary (Figure S2A) we focused on the C138 line. C138 proximal transgene sequences and XCI expression boundary are intact and therefore, adjacent genes are predicted to remain X inactivated. Consistent with this expectation, robust mono-allelic expression from the active X was detected by RNA FISH in both C048 (used to control for a non-transgenic Tex11 locus) and C138 (Figure 4B). These data further establish that transcripts in this region are normally X inactivated and are not altered upon transgene integration.
To examine effects at the C138 distal boundary we queried transcripts included in BAC RP23-263O9 because low Tex11 expression was undetectable on either X by RNA FISH (Figure S6). Monoallelic expression from only the active X in C048 confirms that RP23-263O9 transcripts are normally X inactivated (Figure 4B). However, a heterogeneous pattern was seen in C138, with inactive X expression in 22% of cells. This proportion closely approximates the percentage of cells that inactivate the transgenic CAST X (Figure 4B), and argues that distal genes on the transgenic X escape XCI at a high frequency. Aberrant XCI regulation does not extend further, as adjacent transcripts detected by BAC RP23-295G17 are properly X inactivated (Figure 4B).
To confirm and extend these results, we determined the XCI status of proximal and distal transcripts in differentiated clonal lines that carry the C138 transgene only on the active X or only on the inactive X chromosome. First, allele-specific expression of cDNA from the C138-derived clonal lines confirmed that the proximal gene Dlg3 is X inactivated (Figure 4C). Next, Tex11 at the integration site was tested. While Tex11 is X inactivated in the clonal line that carries the transgene on the active X (Figure 4C), the gene now escapes XCI when interrupted by Kdm5c-tg. To determine the extent of XCI misregulation, we queried additional genes downstream of Tex11. Two additional transcripts, Slc7a3 and Snx12, aberrantly escape XCI on the transgenic X (Figure 4C). By qSNaPshot, the level of inactive X escape relative to active X expression is quite similar for all three genes. However, it is unlikely that absolute inactive X expression is equivalent given that RNA FISH suggests significantly higher Snx12 transcription on both Xs (Figure 4B, Figure S6). Altogether these results argue that absence of the distal XCI boundary results in 350 kb expansion of an escape domain.
Recent genome-wide studies have made tremendous strides in uncovering long-range organization and predicting functional domains [33]. Direct annotation of the inactive X is more challenging, in part because it is masked by its active X counterpart. Despite recent efforts to catalogue allele-specific epigenetic features (e.g. [10], [11], [34]), current understanding of the pivotal sequences and modifications that regulate how a gene responds to XCI remains incomplete. While inactive X profiling has identified intriguing candidates, functional dissection can reveal unexpected regulatory modes, such as uncovered here at Kdm5c.
These studies have expanded our understanding of the Kdm5c locus. Because our BAC transgenes carry large inserts encompassing X-chromosome genes that normally are influenced by XCI, effects are expected to recapitulate endogenous regulation and identify candidate sequences that are highly likely to be relevant. Our previous full-length BAC transgene studies allowed us to conclude that an element(s) within the BAC is sufficient to initiate Kdm5c-tg escape [19]. Such a regulatory element could also explain XCI escape of a human autosomal transgene [35]. For the Kdm5c locus, this activity was mapped to a 112 kb region defined by BAC overlap (Figure 1) [19]. Here we examine additional transgenes that further narrow this interval, as Kdm5c-tg still escapes XCI from BAC transgenes lacking distal boundary sequences (Figure 2). Because the truncated BACs integrated into X-inactivated regions, we conclude that the remaining transgene sequences must include a dominant element(s) sufficient to initiate Kdm5c escape and to structurally remodel the X in a manner that allows preferential association with escape genes (Figure 3). Further, our studies of the C138 transgene reveal an additional role for distal XCI boundary sequences, since in contrast to the full-length BACs [19] XCI regulation of adjacent X-inactivated genes was disrupted (Figure 4).
What sequences are necessary for XCI escape and do these elements also facilitate long-range escapee interactions? Sequences orchestrating these activities must map within the C138 transgene and likely reside within the proximal XCI boundary (Figure 5A). Therefore, the complete escape domain, including the escapee lncRNA, cannot be necessary for directing inactive X expression. Retained BAC sequences include the Kdm5c promoter and CTCF-binding sites that are proposed to delimit this proximal XCI boundary [24] (Figure 5A). Nevertheless, CTCF binding alone is not sufficient to confer XCI escape [36]. Further, whether specific promoter elements alone can drive escape is untested, but large-scale transgenesis likely excludes promoter strength as a sole property [35]. Sequences within C138 also enable long-distance association with other escape genes. Yet, the region may be further narrowed as the short B176 transgene, lacking Kdm5c-tg and its promoter, fails to preferentially interact.
Deletion of distal transgene sequences in C138 reveals additional regulation at Kdm5c. In the absence of an XCI boundary, three normally X-inactivated genes near the BAC integration site now escape XCI (Figure 4). We asked whether aberrant distal expression is due to permissive chromatin propagated by read-though transcription from the truncated Kdm5c-tg. This possibility seems unlikely, as transcription does not extend across the entire escape domain (Figure S6). Further any read-through is at most minimal, as no transcription across the Tex11 locus is seen by RNA FISH, even when the transgene is on the active X. Nevertheless, strand-specific RT-PCR within Tex11 detects low-level sense and antisense transcripts from both non-transgenic and transgenic undifferentiated ES cells (Figure S6). That these transcripts are not unique to the Kdm5c-tg locus argues that low levels of transcription alone cannot enable escape. Therefore, while the extent that XCI is disrupted is likely dependent on integration site characteristics, the C138 transgene must lack a regulatory element that normally has an essential role in establishing an XCI boundary at the endogenous Kdm5c locus (Figure 5B). How this element functions is not clear, but could actively prevent heterochromatin encroachment into active domains or instead block escapee regulators from influencing adjacent silenced genes. Consistent with the former, a chromatin barrier could act as a boundary if upon deletion other distal elements reposition the XCI boundary (Figure 5C). CTCF could perform such a function, as sites are found near the distal Kdm5c boundary and are normally present at locations that could delimit the expanded escape domain (Figure S7). Moreover, CTCF frequently binds at chromatin boundaries throughout the genome [37], and can organize and reorganize chromatin loops [38]–[40]. This would suggest plasticity at XCI boundaries and could explain tissue differences in some escape genes [11], [17], [41].
Sequences at the distal XCI boundary could instead actively block adjacent genes from escape in a manner that is directional and in cis (Figure 5C). Deletion of such a boundary could appear as euchromatin spreading, although, to our knowledge, similar effects have not been described elsewhere in the genome. Yet, elements at other loci could explain this observation. CTCF functioning as an enhancer-blocking element fits this model [42]–[44], particularly since deletion at other epigenetically regulated loci can induce gene reactivation [45]. Alternatively, transcripts near escape genes may require additional elements to be properly X inactivated [46]. In this role, the lncRNA could silence by transcriptional interference [47], although effects extending such distances are not reported. Further, lncRNAs can recruit chromatin-modifying enzymes in cis (e.g. [47], [48]). Supporting recruitment, it is intriguing the AK148627 lncRNA is amongst transcripts immunoprecipitated by the PRC2 polycomb-complex component EZH2 [49].
Finally, we considered the role that inactive X topological structure plays in determining XCI states. Distant escapee contacts are maintained for Kdm5c-tg at all three ectopic locations tested. Therefore, long-distance interaction is another inherent property of an escape locus, yet its mechanistic relationship to active transcription remains undefined. Transgenic loci are likely repositioned at the exterior of the Xist compartment, similar to endogenous Kdm5c [18]. Such rearrangement would also impact genes adjacent to the transgenes. While positioning on the inactive X could influence distal gene escape in C138, it cannot be sufficient since proximal genes remain X inactivated. Additional factors must be necessary to direct XCI fates.
Epigenomic features may refine the XCI boundary and localize key regulatory sequences. Using available data sets, H3K27me3 profiles in non-transgenic female lines mirror inactive X expression, with depletion clearly characterizing the expressed Kdm5c locus (Figure S7A). Intriguingly, while the proximal H3K27me3 transition is quite distinct, the distal boundary appears more diffuse (Figure S7A). Both H3K27me3 patterns occur at domain boundaries throughout the genome [50] and the distal profile may be indicative of an expression transition [9]. That this moderate H3K27me3 region contains critical regulatory sequences is supported by our current studies, since the shortest transgene breakpoint directly abuts this region. Nevertheless, the nature of the boundary makes regulatory element localization more difficult. If boundary repositioning expands the escape domain, it is intriguing that the novel boundary appears demarcated by H3K27me3 even on non-transgenic chromosomes (Figure S7B). However, further conclusions will require chromatin profiling on transgenic chromosomes. We next turned to DNaseI hypersensitivity that demarcates many regulatory elements [51]. At both the endogenous Kdm5c locus and C138 transgene integration site available data only identify hypersensitive sites at gene promoters and CTCF-binding sites (Figure S7). Perhaps this strengthens CTCF as a candidate. A caveat is that such a function may be developmentally regulated and no female lines have been profiled upon the onset of XCI.
Altogether, work here has defined two separable functions at the Kdm5c locus. We narrowed sequences required for directing escape and for the first time have assigned a function to an XCI boundary in actively delimiting expression domains. By defining and demarcating regions responsible for each activity, future experiments can be directed to examine specific candidate elements.
The parental ES line SA13 was derived from a (129×CAST)F1 female [19]. ES cell lines carrying X-linked BAC transgene RP23-391D18 were described previously [19]. All cells were cultured using established conditions and were maintained in the absence of drug selection [19]. For post-XCI experiments, cells were differentiated for ten days following LIF removal.
Clonal C138 lines were isolated by first differentiating ES lines for 10 days. Cells were replated using conditions that further enrich for differentiated cells [19] and after two days were infected with SV40-VA4554 [52]. Cells were passaged as required and after >20 days plated at very low cell density and allowed to clonally expand. Monoallelic expression of SNPs within Hprt and/or Pctk1 [32] confirmed clonality.
Due to the location of the selectable marker within the RP23-391D18 BAC vector [19], truncated transgenic lines surviving initial drug selection lack genomic sequences at the distal XCI boundary. Informative SNPs to delimit these transgene breakpoints were identified (http://cgd.jax.org/cgdsnpdb) and are listed in Table S1. Allelic ratios were evaluated using a quantitative primer extension assay, qSNaPshot [13], [19]. Samples were run on an ABI 3130XL sequencer and peak heights measured using GeneMapper 4.0 software with SNaPshot default settings. Allele ratios in transgenic lines were normalized by comparison with the non-transgenic ES line. Results were further adjusted as allele ratios for a non-transgenic SNP rs29296320 deviated slightly from an expected ratio of 1.0 (ranging from 0.84 to 1.07), likely reflecting loss of an X in a small proportion of cells.
Precise transgene integration sites were determined by inverse PCR [53]. For C048 and C138, genomic DNA was digested with XbaI or PstI respectively. Purified DNA was self-ligated in dilute conditions and used as template for PCR with BAC-derived primers. PCR products were cloned and sequenced. Similar efforts for B176 failed to isolate integration sequences, consistent with a more complex vector rearrangement upon insertion. To determine if C138 and C048 transgene integrations resulted in large-scale deletions, genomic SNPs distal to the integration site were analyzed by qSNAPshot (Table S1 for primers and SNPs).
To identify the strain origin of the transgenic Xs in C138 and C048, SNP alleles were assayed from transgenic X specific PCR products that were generated by anchoring one primer to the BAC backbone. For C048, the closest informative SNP was >6 kb away and required initial amplification from a self-ligated template, similar to inverse PCR (Table S1 for SNP and primer information). Strain origin of the transgenic X in additional lines was inferred by determining the frequency that the BAC is on the inactive X since XCI skewing results in inactivation of the CAST X in 25% of cells [19].
The normal XCI status of transcripts at the transgene integration site was assayed using qSNaPshot to measure allelic expression in the non-randomly X-inactivated mouse fibroblast lines B120 or B119 [13], [14]. Mid1 was tested previously in a similar manner [19]. Mid1 has a unique gene organization and XCI pattern; it straddles the pseudoautosomal (PAR) boundary in some strains, but is X-specific in others [54], [55] (Figure S2A). Mid1 escapes XCI in domestic mouse strains [10], [54], but is X inactivated on the CAST X [19].
Allelic expression was similarly assayed in the C138 clonal lines. For genes flanking the transgene, inactive X expression was measured relative to the active X allele and normalized to DNA. Kdm5c, with three expressed loci in C138, required the expression ratio to be normalized to non-transgenic DNA (to account for dye incorporation differences) and additionally to DNA from the clonal line (to account for loss of an X in a small subset of cells). However, both Kdm5c-tg and the endogenous locus on the active X are derived from domestic strains and are not distinguishable. Therefore, levels of Kdm5c-tg escape were estimated from the normalized allele ratios as if equivalent to the endogenous CAST inactive X allele. This estimate appears justified since both inactive X alleles (CAST and Kdm5c-tg) are predicted to partially escape at levels similar to those previously reported [18], [19], [31], [32]. Further, given the measured allele ratios, estimates of lower Kdm5c-tg escape require concomitant reduction in the endogenous CAST allele to levels below that been previously seen.
FISH probes included Xist (7.2 kb of exon 1) [19], Kdm5c (19 kb spanning exons 5–12 [19]), DXWas70, an X-specific repeat [56], and BACs RP23-391D18 (includes Kdm5c), RP23-330G24 (Kdm5c), RP23-67G4, RP23-459H14 (Tex11), RP23-263O9, RP24-255O24, RP23-295G17, RP23-459P19 (Ddx3x), and RP23-378I14 (Mecp2). Probes were directly labeled with Alexa Fluors 488, 594, or 647 by nick translation using either ARES DNA labeling kits (Invitrogen) or ChromaTide Alexa Fluor dUTPs (Invitrogen) as indicated by the manufacturer.
Slides were prepared and FISH performed for each specific experiment as follows. For DNA FISH studies, metaphase spreads were prepared and FISH performed as previously described [19], [57]. For all other studies, embryoid bodies were plated on slides at day 3 of differentiation and cultured to day 10. RNA FISH was performed on non-denatured slides as described [18], [58]. For sequential RNA and DNA FISH, slides were initially processed as for RNA FISH. Subsequently, signals were fixed in 4% paraformaldehyde in PBS prior to denaturing (75°C for 5 minutes) and processing for DNA FISH [19]. For association studies, cells were fixed in 4% paraformaldehyde before permeablization to preserve nuclear morphology [17]. Slides were denatured at 85°C for 4′ or 75°C for 7′, which allowed sufficient retention of Xist RNA to identify the inactive X chromosome.
Slides were imaged on Nikon TE2000-U microscope with Roper Scientific CCD camera and NIS elements software equipped with a 60× objective. Alternatively, a DeltaVision Elite microscope was used that is equipped with 60× or 100× objective and CoolSnap HQ2 Photometrics camera. Deltavision images were acquired across 0.2 µm Z stacks, deconvolved, and analyzed using softWoRx software version 5.5.5. In all cases, wavelengths were captured separately and merged and pseudocolored in Adobe Photoshop. Image manipulation was restricted to overall fluorescent level adjustment applied uniformly across the image.
To ensure optimal hybridization, we adopted specific scoring criteria for each experiment. For all FISH expression studies, we required hybridization patterns for scored cells to at least reflect known endogenous XCI expression. That is, for a gene that normally escapes XCI (Kdm5c), all cells included had at least one active X and one inactive X signal; for normally X-inactivated genes, only cells with at least one robust active X signal were scored. Additional RNA signals then reflect transgene expression (Kdm5c escape) or aberrant escape (for normally X-inactivated genes). Multiple planes were examined to ensure that out-of-focus signals were not excluded. Assignment of Kdm5c signals was facilitated by colocalization with BAC DNA FISH signals to pinpoint the Kdm5c locus or the integration site locus. Unless noted, each experiment scored at least 100 nuclei that met criteria, with each scored cell selected from an independent field of vision. Statistical significance was evaluated by Chi square analysis.
To evaluate probe association slides were viewed on a DeltaVision Elite microscope (100× objective). X,Y,Z coordinates were recorded for each signal and the 3D distance between probes calculated [17]. Nuclear area was calculated by averaging polygon areas (demarcating the nucleus) across all in focus Z sections and was used to normalize for differences in nuclear size and morphology. Analysis was limited to ∼95% of cells with nuclear area <175 µm2 to ensure overlapping distributions across all cell lines. For each probe set 100–150 nuclei were scored per cell line. Significance was assessed using a Kolmogorov-Smirnov two-sample statistic [59].
|
10.1371/journal.pbio.1001731 | Escherichia coli Ribosomal Protein S1 Unfolds Structured mRNAs Onto the Ribosome for Active Translation Initiation | Regulation of translation initiation is well appropriate to adapt cell growth in response to stress and environmental changes. Many bacterial mRNAs adopt structures in their 5′ untranslated regions that modulate the accessibility of the 30S ribosomal subunit. Structured mRNAs interact with the 30S in a two-step process where the docking of a folded mRNA precedes an accommodation step. Here, we used a combination of experimental approaches in vitro (kinetic of mRNA unfolding and binding experiments to analyze mRNA–protein or mRNA–ribosome complexes, toeprinting assays to follow the formation of ribosomal initiation complexes) and in vivo (genetic) to monitor the action of ribosomal protein S1 on the initiation of structured and regulated mRNAs. We demonstrate that r-protein S1 endows the 30S with an RNA chaperone activity that is essential for the docking and the unfolding of structured mRNAs, and for the correct positioning of the initiation codon inside the decoding channel. The first three OB-fold domains of S1 retain all its activities (mRNA and 30S binding, RNA melting activity) on the 30S subunit. S1 is not required for all mRNAs and acts differently on mRNAs according to the signals present at their 5′ ends. This work shows that S1 confers to the ribosome dynamic properties to initiate translation of a large set of mRNAs with diverse structural features.
| Gene expression is regulated at multiple levels, including the decision of whether or not to translate a mRNA. This phenomenon, known as translational regulation, allows rapid changes in cellular concentrations of proteins and is well suited to the adjustment of cellular growth in response to stress and environmental changes. Many bacterial mRNAs adopt structures in their 5′ untranslated regions that modulate the accessibility of the mRNA to the small ribosomal 30S subunit and so are directly involved in this regulatory process. Structured mRNAs must interact with the 30S subunit in a two-step pathway whereby the docking of a folded mRNA is followed by an accommodation step that involves unfolding of these structures. However, it is not known how the ribosome unfolds mRNA structures to promote translation initiation, nor which ribosomal factors are responsible for this activity. We demonstrate that the first three domains of ribosomal protein S1 endow the 30S subunit with an RNA chaperone activity that is essential for the binding and unfolding of structured mRNAs, allowing the correct positioning of the initiation codon for translation. However, ribosomal protein S1 is not required for all mRNAs and acts differently depending on the type of regulatory elements present in a given mRNA. In all, we have shown that ribosomal protein S1 provides an RNA-melting activity to the exit site of the 30S decoding channel and confers some plasticity on the ribosome to initiate translation of mRNAs.
| Translation initiation, which ensures the formation of the first codon–anticodon interaction into the peptidyl (P)-site of the small ribosomal subunit in the correct frame, is the rate-limiting step of protein synthesis. Binding of the mRNA to the 30S subunit takes place at any time during the assembly of the 30S initiation complex (30SIC) and the kinetics is independent of the initiation factors, relying uniquely on 30S and mRNA features (e.g., [1],[2]). Crystal structures of the ribosome containing an unstructured mRNA and tRNA showed the mRNA path, refered as the mRNA channel, occupied by around 30 unpaired nucleotides forming numerous interactions with the 30S subunit [3]–[5]. The formation of a short duplex between the Shine–Dalgarno (SD) sequence of the mRNA and the 3′ end of the 16S rRNA (aSD) locks precisely the 5′ end of mRNA at the exit site of the mRNA channel, a specific place of the 30S known as the platform [3],[6]. The SD/aSD interaction is sufficient for unstructured model mRNAs to bind to the 30S, however most of the natural mRNAs contain additional sequences or structure motifs in their 5′ untranslated regions (UTRs) that have been exploited by bacteria to regulate translation initiation [7]–[11]. Structures in the 5′UTR of mRNAs are thought to represent a kinetic barrier that could lower translation initiation rates because the 30S must disrupt first the structures it encounters in the ribosome binding site (RBS) to allow the mRNA to reach its decoding site [2]. Several studies have revealed that mRNA structure motifs located upstream of the initiation codon bind to the 30S in a two-step process [2],[12]–[14]. A typical example is Escherichia coli rpsO mRNA encoding ribosomal protein (r-protein) S15, which carries a pseudoknot structure within the RBS, and which is recognized by the 30S for translation and by S15 for autoregulation [15]. Structure analysis of several ribosomal complexes identified intermediates of the initiation pathway of rpsO mRNA [13]. It revealed that the pseudoknot structure is first docked on the 30S platform where it forms the SD/aSD helix and interacts with r-proteins S2, S7, S11, and S18. In a second step, the pseudoknot unfolds to promote the formation of the codon–anticodon interaction at the P-site. This activity is carried out by the ribosome, but the mechanism is yet unknown.
Recent studies have shown that the 30S is endowed with an RNA helicase activity at the mRNA entry site. This helicase activity is due to the r-proteins S3, S4, and S5, which unwind mRNA structures during translation elongation [16],[17]. Are both the extremities of the mRNA channel endowed with a similar RNA unfolding activity? In other words, is the platform of the 30S able to unfold mRNA structures to promote mRNA accommodation during translation initiation? The protein environment of the 30S platform consists of several essential r-proteins, namely S1, S2, S7, S11, and S18 [5],[13],[18]. Among these proteins, S1 is an atypical r-protein because it is the largest and most acidic one that is weakly and not always associated with the 30S subunit [19]. The protein consists of six imperfect OB-fold repeats, which is an RNA-binding module specific for single-stranded regions, and is found in many proteins involved in RNA metabolism [20]. Although the structure of the protein has not yet been solved, a cryo-EM analysis suggested that the protein may adopt an elongated shape on the 30S and may bind 11 nts upstream of the SD of a model RNA [18],[21]. E. coli r-protein S1 is essential for the translation of many mRNAs and for viability [22]. Particularly, S1 forms an essential component of the mRNA binding site for mRNAs lacking or bearing weak SD sequences [23]–[26]. Furthermore, isolated S1 is able to melt RNA duplexes or helices independently from the 30S [27]–[31]. These works led to the hypothesis that S1 would confer to the 30S an RNA melting activity to facilitate translation of structured mRNAs, although these studies were not carried out on S1 bound to the ribosome and with natural mRNAs. Finally, S1 has been implicated in many other functions [20]. The versatility of the RNA–S1 interaction and the existence of multiple OB-fold domains might explain the diverse biological functions of S1 outside or on the ribosome.
In the present work, we demonstrate that r-protein S1 confers to the 30S an RNA chaperone activity, which is modulated by the ribosomal environment and essential for the binding and the accommodation of structured mRNAs into the decoding channel. We have analyzed the S1 dependence on three different mRNAs from E. coli, which all contain specific binding sites for translational repressors located close to or within the RBSs and which are repressed at the translation initiation step by various mechanisms. Using these natural mRNAs, we show that S1 on the ribosome interacts transiently with structured mRNAs and promotes a metastable folding state to create new interactions with the 30S subunit. The melting process is slow and represents most likely the rate-limiting step of translation of structured mRNAs. In contrast, an mRNA bearing optimal SD sequence and weakly structured RBS does not need S1 to form active ribosomal initiation complex. Our study reveals the mechanism of action of r-protein S1 on natural mRNAs and how S1 modulates the activity of the 30S dependent on the mRNA context.
We first monitored the effect of r-protein S1 on the formation of the 30SIC using three different natural mRNAs (Figure 1A). These mRNAs have been selected because they have evolved specific structural features to be well translated and regulated at the initiation step of translation. They also all carry an unpaired SD sequence. E. coli sodB mRNA (SD AAGGAG, ΔG −8.48 kcal/mol predicted for the SD/aSD helix), encoding superoxide dismutase, contains a weakly structured RBS [32] and the binding sites for the translational repressor RyhB sRNA and Hfq [33]. E. coli thrS mRNA (SD UAAGGA, ΔG −5.96 kcal/mol), encoding threonyl-tRNA synthetase (ThrRS), contains a bi-partite unstructured RBS interrupted by a hairpin structure recognized by ThrRS for translation repression [34]. Both RyhB and ThrRS hinder the ribosome binding to repress translation. Finally, E. coli rpsO mRNA (SD GGAG, ΔG −5.85 kcal/mol) contains a pseudoknot structure, which sequesters part of the coding sequence. Binding of r-protein S15 stabilizes the pseudoknot on the 30S platform to prevent the start codon from reaching the P-site [13].
Toeprinting assays were used to analyze the formation of a simplified 30SIC, composed of the 30S, the mRNA, and the initiator tRNA [35]. A toeprint is observed at position +16 (+1 is the adenine of the start codon) if the mRNA occupies the decoding channel and if the codon–anticodon interaction takes place at the P-site. To monitor the action of S1, the assays were performed with wild-type (WT) 30S, S1-depleted 30S (30S−S1) (Figure S1A), or the 30S−S1 complemented with purified r-protein S1 (30S+S1). Quantification of the data showed that the 30S efficiently recognizes and accommodates sodB mRNA into the decoding channel in the presence or in the absence of S1 (Figure 1B). Thus, S1 is dispensable for mRNA carrying an unstructured RBS with a strong SD sequence. Conversely, the formation of the 30SIC performed with the 30S−S1, thrS mRNA, or rpsO mRNA are strongly perturbed, showing that S1 has a role for activating these mRNAs (Figure 1B).
Because thrS and rpsO mRNAs have a weak SD, we addressed the question of whether S1 would be required for the docking and/or for the accommodation process of these mRNAs by introducing an enhanced SD (AGGAGGU, ΔG −12.53 kcal/mol) to reduce the S1 dependence for mRNA docking. Translation of thrSSD and rpsOSD mRNAs was indeed significantly enhanced in vivo [36],[37]. Formation of the 30SIC with thrSSD mRNA was similar with WT 30S and 30S−S1, indicating that S1 becomes dispensable (Figure 1C). However, for rpsOSD mRNA, the yield of 30SIC was still low when formed with 30S−S1. Concomitantly, several other reverse transcriptase (RT) pauses in rpsOSD mRNA were observed when WT 30S and 30S−S1 were bound to the mRNA. These stops located at positions −5 and +10 correspond to the entrance of the SD/aSD helix and to the pseudoknot structure, respectively (Figure 1C). They represent signatures of the stalled 30S pre-Initiation complex (30S-preIC) in which rpsOSD mRNA binds to the 30S but remains folded onto the 30S platform [13].
We then performed filter binding assays to monitor the direct binding of 5′ end-labeled mRNAs to WT 30S or 30S−S1 in the absence of the initiator tRNA—that is, before the accommodation step. The binding saturation curves show that S1 strongly enhances the docking of WT thrS and rpsO mRNAs on the 30S, while sodB, thrSSD, and rpsOSD mRNAs bind to the 30S independently of S1 (Figure S1B). The three WT mRNAs bind the WT 30S (containing S1) with a similar Kd value (around 1 µM), although the SD sequence of sodB mRNA is stronger than the SD sequence of thrS and rpsO mRNAs. However, the absence of S1 on the 30S strongly decreases the recognition of WT thrS and rpsO mRNAs (Figure S1B). This S1-specific effect was completely alleviated when the SD was enhanced in thrSSD and rpsOSD mRNAs, and the binding affinity for the 30S increased 5-fold (Figure S1B). Therefore, a strong SD sequence compensates the lack of r-protein S1 to anchor the mRNAs onto the 30S subunit. However, the ability of rpsOSD mRNA to bind the 30S independently of S1 is not sufficient for its translation because S1 is still required to promote the formation of the active 30SIC as evidenced by the toeprinting assays (Figure 1C). Hence, these data indicate that S1 is directly involved in the accommodation of rpsO mRNA into the decoding channel.
Together, the data show that two S1 functions can be distinguished: (i) promotion of mRNA binding and (ii) mRNA accommodation. The various activities of r-protein S1 reflect the diversity of RBS architectures. The data support the following schemes where mRNAs with weakly structured RBSs (i.e., sodB and thrS) form 30SICs in a single step—that is, binding directly leads to active 30SIC formation. Instead, with structured mRNA (i.e., rpsO) two distinct steps have been identified, where mRNA binding (influenced by their SDs) precedes its accommodation into the decoding channel. In both cases, the need of S1 for the binding is exclusively dictated by the strength of the SD sequence, whereas S1 is essential for the accommodation of structured mRNAs.
Because our data suggest that r-protein S1 promotes the accommodation of rpsO mRNA on the 30S that would require unfolding of its pseudoknot structure, we used fast kinetics to analyze the structural changes of the pseudoknot on the ribosome using 2-aminopurine (2-AP) modifications. The fluorescent nucleobase 2-AP, which can interact with uracil in a Watson–Crick pair or with cytosine in a wobble pair, is known to quench its fluorescence emission in a quantifiable manner [38], depending on local changes of the RNA structure when it stacks with other bases while fluorescence increases when it is fully exposed to solvent [39],[40]. Two modifications were introduced in a mRNA fragment encompassing the rpsO pseudoknot called psk-rpsOSD (containing nucleotides −56 to +12) at the strategic positions A-40 and A-42 involved in long-range interactions of the pseudoknot structure (Figure 2A). Melting of the pseudoknot on the 30S is expected to enhance fluorescence due to an increased accessibility of A-40 and A-42 towards the solvent. The kinetic of the pseudoknot melting on the 30S was analyzed by stopped-flow fluorescence experiments. The formation and stabilization of the pseudoknot structure was evidenced during the renaturation process. The addition of Mg2+, known to greatly stabilize the pseudoknot structure [15], causes significant quenching of the fluorescence (Figure S2A). To analyze the effect of S1 isolated or 30S-bound, we added either S1 alone, 30S containing S1, or 30S−S1. The time course of the increase in 2-AP fluorescence as the result of the pseudoknot melting was reproducibly observed when the RNA was incubated with the WT 30S (Figure 2B). Conversely, the addition of 30S−S1 to the 2-AP modified RNA only slightly changed the fluorescence emission as compared to the controls (Figure 2B). Noteworthy, binding and toeprinting experiments showed that rpsOSD mRNA is well recognized by the 30S−S1 but does not form an active 30SIC (Figures 1C and S1B), demonstrating that mRNA binding to the 30S is not sufficient per se to change the fluorescence. Therefore, our data indicate that the increased fluorescence is mediated through S1 due to an increased accessibility of A-40 and A-42 towards the solvent. Our data are consistent with previous findings showing that G-39 and G-41 of rpsO were highly accessible to RNase T1 in 30SIC, while these residues were not cleaved in the stalled 30S complex where the pseudoknot structure is stabilized [41]. The analysis of the stop-flow data required fitting a double exponential function, revealing two kinetic phases for the melting of the pseudoknot structure, a fast (kfast 0.9 s−1) and a slow (kslow 0.08 s−1) process. The kfast value corresponded to the majority of the fluorescence increase (73.3%). The addition of the initiator tRNA had no effect on the kinetics, suggesting that the S1-dependent melting process does not require the formation of the anticodon–codon interaction (unpublished data). The same experiment performed with r-protein S1 alone shows marginally enhanced fluorescence emission. Saturation could not be attained even with long recording times, so that the fitting of the S1 curves could not be performed accurately (Figure S2B–C). As a control, we demonstrate that S1 deleted of the OB-fold domains 1, 5, and 6, a mutant with impaired mRNA and 30S binding (see below), did not enhance the fluorescence emission (Figure S2B–C). These experiments show that the RNA melting activity of S1 is strongly stimulated when the protein is bound to the 30S as compared to the isolated protein, indicating that S1 is primarily acting on the 30S subunit.
Thus r-protein S1 is endowed with a 30S-stimulated melting activity that leads to the unfolding of the pseudoknot structure required for the relocation of the mRNA into the decoding channel.
Because ribosomal protein S15 stabilizes the pseudoknot conformation of rpsO onto the 30S to repress its own translation, we analyzed whether r-protein S1 interferes with the regulatory function of S15 (Figure S2D). Toeprinting reveals that in the absence of initiator tRNA, formation of the trapped ribosomal complex involving S15, WT 30S, and rpsO mRNA causes several RT pauses around position +10, corresponding to the entrance of the pseudoknot. Identical patterns were observed with 30S−S1 or 30S+S1, indicating that the pseudoknot is stabilized by S15 regardless the presence or not of S1 on the 30S (Figure S2D). Therefore, S1 did not affect the formation of the trapped complex, while in the absence of S15, the formation of the active 30SIC was strictly dependent on S1 (Figure 1B). These data illustrate that the mRNA unfolding activity of S1 can be counterbalanced by regulatory factors such as S15, which stabilizes the mRNA in the structured form onto the 30S platform.
To gain more insight into the mechanism of interactions between S1 and the pseudoknot of rpsO mRNA, we analyzed deletion mutants of S1 (Table S1, Figure S3A) lacking one or more OB-fold domains based on sequence and structural information available for domains 4 and 6 [42]. To avoid possible structural heterogeneity of the rpsO mRNA fragment forming the pseudoknot, we also studied the mutant (C-14 to G, mut psk-rpsOSD; Figure 2A), which was shown to exclusively form the pseudoknot structure [43]. We first show that WT S1 binds similarly to the two RNAs (rpsOSD and mut-rpsOSD) and that the protein concentration around 400–500 nM causes a shift of almost 50% of the 5′ end-labeled RNAs (Figure 3A–B). These data were also well correlated with surface plasmon resonance (SPR) experiments (Kd≈350 nM; Figure S3B). The contribution of each OB-fold domain in recognizing wt or mut psk-rpsOSD mRNAs was defined using gel retardation assays (results not shown and Figure 3C, respectively). The deletion of domain 1 or of the two first N-terminal domains (Δ12) in S1 caused a complete loss of RNA binding even at a concentration of 5 µM. The removal of domains 4 to 6 (S1Δ4–6) decreased the stability by 5-fold, while the additional deletion of domain 3 (S1Δ3–6) abolished mRNA binding. Deletion of domains 5 and 6 affected RNA binding only slightly. These data correlate well with the SPR experiments, which show that the truncated protein S1Δ12 interacts weakly with psk-rpsOSD (Figure S3B). Taken together, these data demonstrate that the six OB-fold domains of S1 are not functionally equivalent, with the first three N-terminal domains of r-protein S1 being essential for the recognition of rpsO pseudoknot structure.
Because the domains of S1 are not equivalent for RNA binding, we then analyzed the importance of each OB-fold domain for cell growth in vivo. We constructed a set of strains with the chromosomal copy of rpsA (the gene for S1) carrying deletions of increasing length as well as a control allele with the kan cassette inserted downstream of WT rpsA, called rpsA1 (Figure S4A). The growth of the control strain and the levels of S1 were identical to that of the WT strain (Figure 4A–B). Two other mutant alleles carry either deletion of domain 6 (rpsAΔ6) or of domains 5 and 6 (rpsAΔ56). The alleles rpsA1, rpsAΔ6, and rpsAΔ56 were obtained with high yields as haploids, indicating that they are viable (Figure S4B), although the growth of the two mutant strains was slower than the WT strains (Figure 4B). In addition, rpsAΔ6 and rpsAΔ56 alleles confer a cold-sensitive phenotype (Figure S4C). Larger replacements such as rpsAΔ4–6 (deletion of domains 4 to 6), rpsAΔ3–6 (deletion of domains 3 to 6), and rpsAΔ2–6 (deletion of domains 2 to 6) were only obtained as diploids carrying both the WT and the mutant copy of rpsA (Figure S4D). We then transduced these three mutant alleles to strains transformed with the complementing plasmid pNK34, which carries the rpsA gene under the control of an IPTG-inducible promoter. In the absence of IPTG, the strain carrying rpsAΔ4–6 was able to grow, whereas the strains carrying the larger deletions (rpsAΔ3–6 and rpsAΔ2–6) did not grow, indicating that they are lethal alleles (Figure 4C). In summary, the in vivo experiments showed that the successive deletions of the OB-fold domains gradually affect cell growth: the two last C-terminal domains are dispensable, the additional deletion of domain 4 still allows growth but at extremely slow rates, and the further deletion of domain 3 causes lethality.
Because some of the domains of r-protein S1 were dispensable in vivo, we analyzed the implication of each OB-fold domain in binding to the 30S. The WT and mutant proteins were incubated with the 30S at a ratio of 3∶1, and the excess was removed by size exclusion chromatography. The S1-occupancy of the 30S was quantified by Western blot and revealed that a minimal protein containing domains 1 to 3 fully retains 30S binding (Figure 5A). Only the deletion of the two first N-terminal domains (1 and 2) totally abolishes 30S binding.
Formation of the active 30SIC using thrS, rpsO, or rpsOSD mRNAs, the initiator tRNA, and the 30S−S1 pre-incubated with the different S1 variants was monitored by toeprinting (Figures 5B–D and S5). For thrS mRNA, which bind the 30S in a single step process and for which unfolding is not necessary, the domains 1 to 3 of S1 are essential and sufficient to promote the formation of the active 30SIC (Figures 5B and S5B). Indeed, 70% of the 30SIC is formed with S1Δ4–6, whereas the additional deletion of domain 3 causes a strong reduction to 40%. Thus, the ability of S1 to stimulate the binding step of thrS mRNA is sustained by the three first N- terminal domains of S1. Similar data were obtained for rpsOSD (enhanced SD) and rpsO mRNAs (Figures 5C–D and S5A and S5C) where the structure of the pseudoknot needs to be unfolded on the 30S to be positioned into the decoding channel. The deletion of domains 5 and 6 only slightly affect the formation of 30SIC, while the additional deletion of domain 4 decreases the 30SIC yields to 50% and 60% for rpsO and rpsOSD, respectively. The removal of domains 3 to 6 completely abolished the formation of 30SIC for rpsO mRNA, whereas a residual signal of 30% was observed for rpsOSD. The enhanced SD compensates the lack of S1 for the binding step as demonstrated by filter binding assays (Figure S1B), but a minimal core of S1 (domain 1–3) is still important to promote the unfolding/accommodation second step. Noteworthy, S1Δ3–6 binds efficiently to the 30S but with impaired functions, suggesting that domains 1 to 3 are essential for all the steps including the binding of thrS and rpsO mRNAs, and the accommodation of rpsO mRNA. All in all, these data show that both 30S-dependent activities of S1, the docking of mRNA carrying weak SD (as for thrS and rpsO) and the unfolding of structured mRNA and accommodation into the mRNA channel (as for rpsO and rpsOSD), require the first three OB-fold domains of r-protein S1. Hence, domains 1 to 3 constitute the minimal protein that retains most of the S1 functions with respect to structured mRNAs.
The ability of isolated r-protein S1 to unwind model RNA duplexes or helices has been well documented [2],[27]–[31]. However, it was not yet demonstrated that S1 would be the key r-protein to unfold mRNA structures on the ribosome. In this study, we have monitored the action of r-protein S1 on the natural structured and regulated E. coli rpsO mRNA encoding r- protein S15 during the formation of the 30SIC. This mRNA carries a pseudoknot structure within the RBS, which is recognized by the 30S for translation [15]. It sequesters the beginning of the coding sequence through base pairings that need to be melted for the formation of the codon–anticodon interaction [13],[15].
We demonstrate here that r-protein S1 and primarily its three OB-fold domains 1 to 3 are essential for the accommodation process allowing rpsO mRNA to unfold and to relocate its initiation codon into the decoding center. Using 2-AP–modified rpsO mRNA, we were able to follow the S1-dependent melting of the pseudoknot directly on the ribosome. We could also compare the S1 RNA melting activity isolated or on the ribosome (Figure 2). Using a combination of approaches, we show that the fluorescence emission does not result from the interaction of the mRNA on the 30S but is primarily due to the melting of the pseudoknot structure (Figures 2A and 6C). In its natural ribosomal context, the melting activity of S1 is clearly more pronounced and is independent of the presence of the initiator tRNA. This enhanced activity on the 30S could be explained by different conformations of S1 when free in solution or anchored to the ribosome where the OB-fold domains 1 to 3 would be orientated in an optimal way to interact with rpsO mRNA. An alternative explanation is the possible contribution of other ribosomal components to the S1-dependent unfolding process.
A recent single-molecule study demonstrated that isolated r-protein S1 is able to melt in a multistep process a large artificial 274 bp stem-loop structure by binding to an upstream single-stranded RNA region [31]. This elegant study showed that S1 binds to the transient open form of the helix-unpaired junction region and stabilizes the open form to promote the local melting of the base pairs. This model is consistent with our data that are obtained on a natural structured mRNA. We propose that the three first domains of S1 bind successively to the A/U-rich connecting loop next to the long-range interaction allowing S1 to bind to the transiently opened base pairs. This mechanism would then lead to pseudoknot unwinding. The rate (0.9 s−1) at which the pseudoknot conformational change takes place on the 30S is rather slow as compared to the rates determined for other events of the translation initiation pathway [1]. It could thus represent the rate-limiting step of the initiation of structured mRNA as it was previously proposed [44]. Our data also indicate that the initiator tRNA is not essential for the RNA melting process. However, the formation of the anticodon–codon interaction is critical to stabilize rpsO mRNA into the channel of the 30S.
Using three E. coli natural mRNAs (sodB, thrS, rpsO), we demonstrate that r-protein S1 acts differently according to the nature of the signals present in the 5′UTR of mRNAs to form the 30SICs. Indeed, S1 is dispensable for the formation of the initiation complex involving sodB mRNA, which contains a strong SD and a weakly structured RBS (Figure 6A). In the second example, S1 is required for the docking of thrS mRNA onto the ribosome in a single step process and the replacement of its weak SD with a stronger one alleviates the requirement of S1 for the formation of 30SIC (Figure 6B). Finally, S1 is required for the recruitment of rpsO mRNA through its pseudoknot structure and for the accommodation process allowing the mRNA to occupy the decoding channel (Figure 6C). Noteworthy, pseudoknots were preferentially selected as strong binders of E. coli ribosomes or of free r-protein S1, while SD-containing unstructured mRNAs were selected against S1-depleted 30S ribosomes [45]. Hence, the complexity of mRNA structure within the RBS would direct the choice of the S1 actions to promote the formation of active 30SIC (Figure 6). In addition, we show that the action of S1 can be prevented by repressor proteins such as r-protein S15, which binds to rpsO pseudoknot and prevents its melting onto the ribosome to repress translation (Figure 6). One can predict that other translational regulatory proteins would interfere with the action of S1 onto the ribosome.
This variety of mechanisms is consistent with the fact that S1 is weakly associated to the 30S subunit. In agreement with this observation, a subpopulation of ribosomes lacking S1 was suggested to co-exist in E. coli under normal growth conditions [46]. Furthermore, the overexpression of rpsA led to the dissociation of leaderless mRNAs from the ribosomes [47]. This was supported by the fact that the overproduction of S1 slightly enhanced the occupancy of the ribosomes, suggesting that the WT levels of the protein did not saturate the ribosomes [48]. Under stress conditions, subpopulations of ribosomes were recently isolated in vivo, which selectively translated leaderless mRNAs [49],[50]. Altogether, it is tempting to speculate that the absence of S1 on the ribosome might confer selectivity for specific mRNAs with strong SDs and unstructured RBSs, such as sodB mRNA or leaderless mRNAs. Thus, S1 confers to the ribosome the ability to dynamically adapt to the sequence and structure of mRNAs, increasing ribosome plasticity. This might help the ribosome to coordinate and fine-tune the rate of protein synthesis.
S1 belongs to the family of RNA-binding proteins composed of multiple RNA-binding motifs. It contains six OB (oligonucleotide/oligosaccharide-binding) fold domains that are connected by short linkers (Figure 5A). We show here that these domains exhibit distinct but also synergistic functions. We first demonstrated that the two N-terminal domains are critical to anchor S1 onto the 30S subunit (Figure 5A). Numerous studies supported the localization of S1 on the 30S platform where it makes contacts with mRNAs and r-proteins [18],[51]–[58]. More precisely, domain 1 of S1 was shown to interact with the coiled-coil domain of r-protein S2 [59]. In addition, we show that domain 2 and to a much lesser extent domain 3 enhance binding of S1 to 30S (Figure 5A). This would suggest that other ribosomal components contributed to precisely position S1 on the 30S platform so that domains 4 to 6 would be exposed to the solvent to recruit specific mRNAs at the initiation step.
Domains 1 to 3 of S1 are essential and sufficient to promote the formation of active 30SIC involving either thrS or rpsO, while domain 4 exerted a stimulating effect only on rpsO, providing additional interactions required for full biological function. This is well correlated with the in vivo data since successive deletions of the OB-fold domains had an increasing effect on cell growth. Indeed, the two last C-terminal domains 5 and 6 affected growth rate in a limited way as it was previously shown [60], while the deletion of domains 4 to 6 permitted growth at extremely slow rates and any further deletions (Δ3–6, Δ2–6) caused complete lethality (Figure 4). This effect on cell growth can be explained by the fact that the truncated proteins are still able to bind to the ribosome, while the recruitment and/or the accommodation of essential mRNAs is presumably strongly perturbed. Although domain 1 has been mainly described as the 30S binding site, we show here that this first N-terminal OB-fold domain is also critical for rpsO mRNA binding (Figure 3C). Other studies revealed that various RNA substrates bind to the same surface area of a protein carrying domains 3 to 5 [61]. In addition, domain 3 with either domain 2 or domain 4 of S1 confer high affinity through cooperative contacts with RNAs [48]. Directed evolution of S1 to enhance translation of GC-rich mRNAs in E. coli selected mutations primarily in domains 3 and 4 [62]. Hence, the flexibility of the domains respective to each other might confer to S1 a high adaptability to bind a large variety of RNA substrates.
The work presented here provides the notion that the six domains of S1 are not functionally equivalent, although they are structurally related with respect to a common fold. The deletion of the two last C-terminal domains of S1 had no major effects on cell growth, indicating that they are not required for translation [60]. In addition, deletion of domain 6 did not affect the translation and autoregulation of rpsA [63]. However, the absence of the C-terminal domain causes a cold-sensitive phenotype most likely due to an impaired ability to melt RNA structures stabilized at low temperature. The fact that mutations could alter the chaperone activity preferentially at low temperatures is not so surprising. Indeed, S1 r-protein does not use energy like other RNA helicases, and therefore at the permissive temperature, the thermal energy may help the protein to melt RNA secondary structures. It could also be possible that domains 5 and 6 contribute to the translation of specific mRNAs as it was previously proposed [60],[64].
In conclusion, this study shows that r-protein S1 confers a chaperone activity to the 30S subunit that promotes the active docking and accommodation of structured mRNAs into the decoding channel. In addition, the data are indicative of a hierarchy of mRNA targets with respect to S1 recognition on the ribosome. Because S1 is essential in E. coli, phylogenetic analysis may shed light on how the S1 functions have evolved among bacteria. A phylogenetic study has been carried out on r-protein S1 based on structural signatures present within each OB-fold domain [42]. This analysis revealed that S1 from Gram-negative bacteria (proteobacteria, chlamidiae, spirochates, bacteroides, aquificae), thermotogae, chloflexi, and high G+C content Gram-positive bacteria (actinobacteria) contained at least the four first domains, suggesting that most of the activities of S1 would be preserved in these organisms. Although the actinobacteria, such as Micrococcus luteus, contained an additional fifth domain different from E. coli S1, M. luteus S1 was able to substitute E. coli S1 on the ribosome to translate mRNAs with weak SD [23],[65]. Another group of bacteria including the firmicutes, tenericutes, and cyanobacteria contained shorter forms of the protein with a first N-terminal domain that differs greatly from E. coli S1, questioning the ability of these proteins to bind to the ribosome. Two S1 homologues containing three OB-fold domains were identified in Synechococcus. One of these homologues was able to bind the ribosome and was found to be essential for the translational initiation of several mRNAs [66]. In B. subtilis, S1 protein is not essential [67]–[69], consistent with the fact that the protein plays no major role in translation [23],[42],[70]. In these Gram-positive bacteria, most of the mRNAs carry a strong SD sequence, and the low G+C content of their genomes may disfavor the formation of very stable mRNA structures, which might obviate the need for S1 melting activity on the 30S. Whether these truncated forms of S1 act as RNA chaperones outside the ribosome remains to be studied. It would also be of interest to analyze how the functions of S1 have evolved, and what are the strategies used by the ribosomes to translate structured mRNAs, in the low GC content Gram-positive bacteria.
All strains and plasmids, which have been used and constructed in this study, are given in Table S1; the oligonucleotides (oligos) used for cloning and for mutagenesis are given in Table S2. Experimental details for the constructions of the strains are given in the Text S1.
WT thrS, thrSSD (−195 to +65 nts, +1 being the A of the thrS translational initiation codon), WT rpsO and rpsOSD (−120 to +65) transcripts were prepared in vitro by T7 transcription of linearized plasmids (see [36] for thrS and [37] for rpsO constructs). WT sodB mRNA (−55 to +64 nts) was transcribed from the PCR product on the genomic DNA of E. coli MG1655 using the appropriate oligonucleotides (Table S2). The psk and mut-psk (−56 to +12, +1 being the A of the rpsO initiation codon) RNA fragments have been transcribed from linearized plasmids (Table S1). The 5′ end-labeling of dephosphorylated RNA or of the chemically synthesized RNA was performed with T4 polynucleotide kinase and [γ-32P]-ATP [71]. All RNAs were purified on 8% polyacrylamide-8 M urea slab gel electrophoresis (PAGE). Before use, mRNAs were renatured as follows: incubation at 90°C for 1 min in RNase-free water, cooled in ice for 1 min, and at 25°C for 30 min in the appropriate buffer containing monovalent ions and MgCl2. Predictions of the SD/aSD stabilities were obtained using RNAcofold of the Vienna package [72].
Wild-type and mutant rpsA genes were cloned in vectors pET23a or pDEST14, and the plasmids were transformed into E. coli strain BL21 (Table S1). The proteins carrying six histidines at their C-terminus were purified using an affinity chromatography followed by a monoQ (for details, see Text S1). We have verified by mass spectrometry that S1 was homogeneous and was not contaminated by E. coli Hfq (see Text S1).
E. coli 30S subunits were purified on sucrose gradients after dissociation of the 70S [73]. Ribosomal protein S1 was removed from the 30S using a polyU-sepharose 4B column (Text S1).
The formation of a simplified translational initiation complex with mRNA (toeprinting assays) was done according to Fechter et al. [73]. Experimental details are given in Text S1.
RNA fragments (psk-rpsO, G-56 to U12) containing two 2-AP nucleotides (A-40 and A-42) were synthesized on Pharmacia Gene Assembler or Applied Biosystems instrumentations using 2′-O-TOM protected phosphoramidite nucleoside building blocks [74]. All experiments were measured on a Kintek SF-2400 stopped-flow device at 37°C. The renatured psk-rpsOSD mRNA (50–100 nM) present in 20 mM Tris-HCl pH 7.5, 60 mM NH4Cl, 1 mM DTT, 7.5 mM MgCl2 was placed in one of the two syringes just before the experiment. The r-protein S1, 30S, 30S−S1, or 30S/fMet-tRNA (1 or 2 µM) in the same buffer were introduced in the second syringe. The protein and the 30S were incubated at 37°C for 20 min before their injection in the mixing chamber. The melting of the pseudoknot psk-rpsOSD was monitored by measuring the increment of the fluorescence signal after passing the samples through KV408 filters (Schott) at 405 nm, generated by the 2-APs excited at 308 nm. The kfast and kslow values obtained by double exponential fitting were obtained with the Prism Graphpad software.
Purified WT and mutant proteins S1 (150 pmoles) were incubated with 30S−S1 (50 pmoles) for 15 min at 37°C in 20 µl of 20 mM Tris-HCl pH 7.5, 60 mM KCl, 40 mM NH4Cl, 10 mM MgCl2, 3 mM DTT, and 0.02 mg/ml BSA. After purification on a Superdex 200 HR 10/30, the fractions containing the 30S or S1 were analyzed on a 4%–12% SDS-PAGE, and visualized by Western blots using antibodies against the His-tag of each S1 (Text S1).
Protein S1 was pre-incubated for 15 min at 37°C in the S1 buffer containing 20 mM Tris-HCl pH 7.5, 10 mM MgCl2, 60 mM KCl, 40 mM NH4Cl, 3 mM DTT, and 0.02 mg/ml BSA. Complex formation was performed at 37°C for 15 min with the renatured 5′ end-labeled RNA (12,000 cpm) and increasing concentrations of r-protein S1 in 10 µl of S1 buffer.
|
10.1371/journal.pntd.0001196 | Population Genetics of Trypanosoma evansi from Camel in the Sudan | Genetic variation of microsatellite loci is a widely used method for the analysis of population genetic structure of microorganisms. We have investigated genetic variation at 15 microsatellite loci of T. evansi isolated from camels in Sudan and Kenya to evaluate the genetic information partitioned within and between individuals and between sites. We detected a strong signal of isolation by distance across the area sampled. The results also indicate that either, and as expected, T. evansi is purely clonal and structured in small units at very local scales and that there are numerous allelic dropouts in the data, or that this species often sexually recombines without the need of the “normal” definitive host, the tsetse fly or as the recurrent immigration from sexually recombined T. brucei brucei. Though the first hypothesis is the most likely, discriminating between these two incompatible hypotheses will require further studies at much localized scales.
| Trypanosomiasis due to Trypanosoma evansi is a widely distributed disease of livestock, affecting especially camelids and equines and is transmitted by biting flies. The disease is of great concern to many developing countries such as Sudan, where its large camel population estimated at over 4.6 million heads is at risk. It is generally believed that T. evansi has evolved when camels infected with T. brucei moved to tsetse-free areas, but only a few studies have been carried out to elucidate the genetic make-up of T. evansi. Therefore, in the current study, 15 microsatellite markers from non-coding loci on 38 isolates of T. evansi originating from different locations in Sudan were analyzed. Three reference strains from Sudan and Kenya were additionally analyzed and compared to the recent isolates. The results of this study revealed a highly significant isolation by distance pattern with rather small neighborhood sizes. It also suggested that T. evansi is either purely clonal with numerous problems of allelic dropouts or that it often sexually recombines without the need of the definitive host, the tsetse fly, or as the result of recurrent immigration from sexually recombined T. brucei brucei mutants.
| Trypanosoma evansi is the most widely distributed of the pathogenic animal trypanosomes, affecting domesticated livestock in Asia, Africa and Central, South America, Canaries Island and recently in Europe (Spain and France) [1–2–3–4]). Until recently, countries were obligated to declare to the OIE surra outbreaks only in equine species, while infection in other animal species was excluded. Recently, this limitation was modified at the OIE Executive Committee meeting held in May 2008: surra is now considered an “OIE listed disease – multi-species” – to be reported to the OIE in the same way as the previously listed trypanosomes (Dourine, surra in horses, tsetse transmitted trypanosomoses) (see: OIE Manual of Diagnostics for terrestrial animals, edit. 17 July 2008, online). Recently human infections have been reported in India making it a potential human pathogen [5].
Camel trypanosomiasis caused by T. evansi is of great concern to countries like Sudan, which possesses the second largest camel population in the world, estimated at nearly 4,623,000 heads (Annual Report of Federal Ministry of Animal Resources and Fisheries, Sudan, 2010). The existence of carrier animals in the vicinity of susceptible camels makes transmission by biting flies possible. The old paradigm according to which T. evansi evolved, via T. equiperdum, when camels infected with T. brucei moved to tsetse free areas [6] is now challenged by the consensus of opinion that both evolved from T. brucei brucei [7]. It was even more recently suggested that T. evansi should be considered as a subspecies of T. brucei complex [8–9–10].
Microsatellite-length polymorphisms using PCR (microsatellite loci) have been recently and widely used for molecular typing of genetically distinct parasite populations such as Plasmodium spp. [11]–[12], Theileria parva [13], Cryptosporidium parvum [14], Toxoplasma gondii [15]–[16], Leishmania spp. [17–18–19], Trypanosoma cruzi [20]–[21] and Trypanosoma brucei groups [22–23–24]. Microsatellite markers have been shown to be polymorphic enough to highlight the existence of genetic diversity within the very homogeneous T. evansi [25]–[26].
In Sudan, a few molecular studies have been carried out on T. evansi using isoenzyme characterizations [8] or on drug resistance of T. evansi [27]–[28]. Parasite prevalence and infection pattern were also performed with varying estimates of prevalence 5.4% using parasitological examination [29], 31.3% with ELISA [29]. The overall prevalence estimated by Salim et al., using molecular epidemiological tools, ranged between 33.9 to 42.1% [30]. However basic genetic analyses of the parasite populations in the country using multilocus neutral markers have not been reported so far.
In this study we selected 15 microsatellite markers from non-coding loci (a priori not subjected to selective forces) on 38 isolates from different sites from Sudan and three reference strains from Sudan and Kenya. This is the first report on microsatellite markers from non-coding loci in population analysis of T. evansi in east Africa.
A total of 685 samples were collected during a survey conducted in March 2008 and during the period between September and November 2009 from seven geographically distinct zones in Sudan (Figure 1). These regions were grouped as “West Nile and East Nile regions” their names and location coordinates are shown in Table 1. Samples were collected from different camel herds mostly nomads that perform transhumance northwards migration in the wet season and southwards in the dry seasons. Sixty two T. evansi positive samples were included in this study in addition to three reference samples collected previously from Sudan and Kenya (Table 1). To maintain anonymity of subject and owners' confidentiality and to adhere to the International Ethical Guidelines for Biomedical Research involving animal subjects, no owner names were recorded within the database or as part of the data collection process. The owners of the sampled camels provided consent to have their animals included in the study. Research on samples from animals was conducted adhering to guidelines of the Institutional Animal Care and Use Committee of the Graduate School of Veterinary Medicine, Hokkaido University. The study protocol has also been approved by the Faculty of Veterinary Medicine, University of Khartoum, according to their guidelines for sampling domestic animals in Sudan.
Fourteen trypanosome isolates were purified from blood cells using DE-52 columns (Whatman, UK) and trypanosome DNA was extracted using Qiamp DNA mini kit (Qiagen, Australia) following manufacturer's instructions then stored at −20°C. Forty eight more samples were collected in FTA cards (Whatman FTA Classic Cards, Whatman,UK), and the DNA was eluted from blood spotted onto the FTA cards using a modified methanol fixation method as described by [31]. Isolated DNA was stored at −20°C until used.
All samples were first subjected to a PCR test, which amplified the ITS1 region of rRNA gene of all African trypanosomes according to [32]. To exclude other T. brucei subspecies, samples were further analyzed with a PCR test specific for T. evansi, using a primer set that amplified 151bp of the T. evansi RoTat gene fragment [33]. The primer set used was TeRoTat920F 5′-CTG AAG AGG TTG GAA ATG GAG AAG-3′ and TeRoTat1070R, 5′-GTT TCG GTG GTT CTG TTG TTG TTA-3′ and the reaction conditions were set in a final volume of 20 µl, which contained 50 ng/µl templates DNA, 10 mM primer “forward and reverse” added to AmpliTaq Gold DNA Master Mix. The cycling conditions consisted of initial step at 94°C for 5 min, followed by 35 cycles of 94°C for 40 s, 58°C for 40 s, 72°C for 1 min, and final extension at 72°C for 5 min. All 62 samples were verified as T. evansi.
Out of the 62 samples examined, 41 were successfully analyzed for 15 microsatellite markers. These markers were selected from the T. brucei genome project release 4. The markers used here were originally designed in silico and computed by [34]. The 15 markers are distributed across all 11 chromosomes of the T. brucei and no two loci are closely physically linked. Markers 10/1, 10/5 and 10/19 on chromosome 10, are separated by distance of 84 Kb and 2.2 Gb respectively, while 11/13 and 11/29, on chromosome 11, are separated by distance of 48 Kb. Recombination rate is in average 15.6 kb/cM in T. brucei according to reference [34], making that a distance of 48 kb is unlikely to create a strong physical linkage. Markers used and different loci are shown in Table 2. The loci were amplified using the following PCR conditions: initial denaturation step at 94°C for 10 min, followed by 35 cycles of 94°C for 30 s, annealing for 30 s, 72°C for 1 min, and final extension at 72°C for 5 min. The annealing temperatures used for different microsatellite amplification are indicated in Table 2. PCR products were electrophoresed in 2% agarose in TAE buffer and stained using GelRed dye (Biotium, USA) before being visualized under UV light.
All forward primers were labeled with fluorescent dye at the 5′ termini. Following multiplex PCRs with the forward fluorescent labeled primer and the unlabeled reverse primer (reverse primers are modified to add A to PCR-products, check the ABI website), as described above, 1 to 5 µl of each product was mixed with 9 µl formamide and 0.5 to 1 µl of GeneScan 600 LIZ size standard (Applied Biosystems, USA). The samples were denatured at 95 °C for 3 min and cooled prior to electrophoresis on an ABI PRISM 3130 Genetic Analyzer under denaturing conditions on a 50 cm capillary column with performance-optimized pop 7 polymer (Applied Biosystems). The instrument was previously calibrated with DS-33 matrix standards (Applied Biosystems). The electrophoresis data were analyzed with GeneMapper software v4.0 (Applied Biosystems).
For most of analysis, we used Create v1.1 [35] from a text file general spreadsheet to convert it into the appropriate format. Because genetic differentiation occurs at a spatial and at a temporal scale in Trypanosomatidae [23–19], we distinguished each of the seven year×site combinations as separated subsamples (see Table 1). Reference strains from Sudan 1976, Kenya 1978 and Kenya 1980 were only considered for the Neighbor-Joining Tree (NJTree) analyses (see below). Number of strains, GPS coordinates and year of sampling can be seen in Table 1.
Linkage disequilibrium was tested with G-based randomization test implemented in Fstat 2.9.3.2. [36], updated from [37] per pair of loci and overall subsamples as recommended [38]. There are as many tests as possible pairs of loci (here 15×14/2 = 105). We thus expect 0.05×105∼5 significant tests under the null hypothesis of no linkage with α = 0.05. We thus tested if there was significantly more than 5% significant tests in the 105 tests series with a unilateral exact binomial test under R [39] to test for the existence of a global signal across the whole test series (hence the genome) and used the sequential Bonferroni procedure [40]–[41] by multiplying the smallest P-value by the number of remaining tests to identify which tests are significant (e.g. [42]).
Wright's F-statistics [43], the parameters most widely used to describe population structure [44], were initially defined for a three-level hierarchical population structure (individuals, sub-populations and total). In such a structure, three fixation indices or F-statistics can be defined. FIS is a measure of the inbreeding of individuals (hence I) resulting from non-random union of gametes within each sub-population (hence S). FST is a measure of the relatedness between individuals resulting from non-random distribution of individuals among sub-populations, relative to the total population; FST quantifies the differentiation between sub-populations in the total population (hence S and T). These F-statistics are classically estimated by Weir and Cockerham's unbiased estimators f (for FIS) and θ (for FST) [45]. These statistics were estimated with Fstat 2.9.3.2. FIS is particularly convenient to measure departure from panmixia and in particular clonal reproduction that is expected to generate strongly negative values at all loci when predominant [46–47–48]. FST is a convenient measure of differentiation between the different sub samples of a data set. Its estimator is expected around 0 under the null hypothesis of random distribution of genotypes across sub samples and positive values, up to 1, in case of genetic differences. The significant deviation from 0 was tested through randomization of alleles between individuals within sub samples for FIS and the statistic used was f; the estimator of FIS. In all cases, randomisation number was set to 10000. These tests are unilateral. For FIS, because in partially clonal organisms it can display positive and negative values, a bilateral test, testing if FIS values are not significantly above or below 0, must be implemented. This is simply implemented by using the P-values obtained with Fstat when testing for FIS>0 and FIS<0 and computing Pbilateral = Pmin+(1-Pmax) and where Pmin and Pmax are the two P-values obtained and Pmin is the smallest one. When individual tests were needed, to identify which loci are in departure from the FIS expected under random union of gametes (panmixia), we also used the sequential Bonferroni procedure. 95% confidence intervals of F-statistics were computed by bootstrap over loci for the mean, and by jackknife over populations for individual loci undertaken with Fstat 2.9.3.2 [49].
A convenient way to represent the genetic composition of clonal organisms is to draw a dendrogram based on genetic distances. We computed Cavalli-Sforza and Edwards [50] chord distance matrices between all individual strains with MSA [51] (and drew a Neighbour-Joining Tree (NJTree) with Mega 3 [52] as recommended [53].
Isolation by distance was assessed following Rousset's method (Rousset, 1997)[54] using FST/(1-FST) as the genetic distance to regress against geographic distance DG. For a two dimensional model of population structure, neighborhood size is related to the slope b of the regression between geographic distances natural logarithm Ln(DG) (computed out of the georeferenced coordinates of each isolate) with the equation FST/(1-FST) = b×Ln(DG)+Constant, with b = 1/4πDeσ2, and where Deσ2 is the product of the effective population density (i.e. ∼density of reproducing adults per square meter) by the dispersal surface that separates them from their parents [54]. In a two dimensional framework the product Nem of the effective population size times the migration rate, which corresponds to the number of migrants arriving in one neighbourhood from the other neighboring sites, is equal to Nem = 2Dσ2 = 1/(2πb) [54]. The significance of the regression was tested by a Mantel test [55] with 1000000 randomizations (Markov chain method) and 95% confidence intervals (CI) by bootstrap over loci. Isolation by distance procedures and testing were all implemented using Genepop version 4 [56]. For this test, only subsamples from year 2009 were considered in order to avoid temporal Wahlund effects [23–19].
Among the 105 possible pairs of loci, 13 were in significant linkage at the 5% level of significance. This if far above the 5 expected under the null hypothesis (P-value = 0.0023). None of these pairs stays significant after Bonferroni correction, but this is probably due to the low power of individual tests and the extreme severity of correction (first lowest P-value must be under or equal to 0.00048 to stay significant here) [see [49]). We can nevertheless note that some loci tended to appear in significant linked pairs more often than expected if linkage was due to the reproductive system or population structure (leading to balanced genome wide signatures). For instance, loci TB10/5 and TB6/7 were both found five times in significant linkage. Here with 105 possible locus pairs, 14 possible pairs per locus (P1 = 14/105) and 13 significant tests (P2 = 13/105), we expect each locus to be in significant linkage with probability P3 = P1×P2 = 0.017. An exact binomial test with five successes among 105 attempts, mean probability = 0.017 and alternative hypothesis “greater” gives a significant result for loci TB10/5 and TB6/7 (P-value = 0.031). Physical linkage is not a really satisfactory explanation as other loci involved in linkage with these two loci should also be linked which is not often the case here, but this cannot be excluded. Moreover, and unless chromosomal changes occurred between T. brucei and T. evansi, all markers are distributed across all 11 chromosomes of the T. brucei and no two loci are closely physically linked (see Materials and Methods). It is also noteworthy that no multilocus repeated genotype (MLG) could be found across the whole data set, which is unexpected if T. evansi is clonal. This unusual genetic structure can be illustrated by the NJTree in Figure 2. Huge effective population size could maintain a great diversity of MLGs but would have prevented isolation by distance pattern to emerge (see below). Mutation rate would need to be extreme also to totally hinder repeated MLGs clonal signatures over all subsamples.
In the dendrogram of Figure 2, West Nile strains seem reasonably together, while East Nile strains appear more heterogeneous. Despite one outlier, Darfur strains are well gathered, and then Kurdofan. Most Showak strains are gathered in Eastern Nile cluster, but again with strong heterogeneity, some being clustered in the “wrong” side of Nile. Kassala and Halfa strains can be found almost everywhere. Globally, considering that Cavalli-Sforza and Edwards chord distance is bounded to 1, the differentiation between strains of the present data is very strong and the distance between some isolates could have been measured between different taxa.
There is a significant isolation by distance (P-value = 0.008) with a slope b = 0.0668 with 95% CI = [0.03, 0.14] (Figure 3). This would correspond to a strongly viscous population (at the scale investigated) with neighborhood size Nb = 15 individuals and small number of immigrants Nem = 2.4 individuals per generation. Obviously, even if camels are known to migrate much over the entire zone and even from other countries of the region [29], infected camels either do not migrate and/or end with poor local transmission to autochthonous hosts when immigrating in new sites and/or insects vector transmit the parasites mainly within herds and not between herds.
FIS analysis results are presented in Figure 4. Over all loci, there is an apparent agreement with panmixia (FIS∼0, P-value = 0.23). This, with an apparent absence of MLG, might be the signature of frequent sex in T. evansi. Nevertheless, some loci display strongly and significant heterozygote excesses or deficits, some of which stay significant at the sequential Bonferroni level. This is not expected under panmixia. From here, several questions arise. First, T. evansi has lost the ability to transform into specific stages that colonize tsetse midgut and salivary glands where sexual recombination occurs and the existence of sexual recombination is thus highly unlikely in this species [57]–[58]. Moreover, other population genetic studies [22]–[26] reveal that levels of heterozygosity of some markers can be very high, that missing allele may be frequent and MLGs numerous. In fact, missing DNA can be found in diploid organisms that are not constrained to frequent meiosis, like it is the case in the yeast Candida albicans [59] or in Leishmania [60]. These DNA losses, if frequent enough, may lead to frequent haploid-like sequences that are in fact heterozygous “DNA/Missing DNA” and, combined with Wahlund effects, could have led to the odd results obtained during linkage disequilibrium and FIS analyses. Allelic dropout and gene conversion can have similar consequences and do occur in kinetoplastid parasites [61]–[62]. The fact that primers were designed from another species, T. brucei, could indeed have lead to discrepancies in microsatellite loci flanking sequences and hence to allelic dropouts and/or null alleles.
We undertook a quick and incomplete simulation study with Easypop v2.01 [63] to check if a Wahlund effect combined with allelic dropout could lead to our observations. It appeared that this indeed can be achieved with certain parameter sets (see Supplementary Material S1) and thus that our interpretation regarding a combined effect of some Wahlund effects and allelic dropouts is rather reasonable. Another source of Wahlund effect would come from the frequent migration of infected camels from different areas. Camels are indeed known to move a lot in this part of Africa [29]. Nevertheless, this hypothesis is extremely hard to reconcile with the clear isolation by distance found that would surely be destroyed by such massive migration of parasite strains. This hypothesis is also self contradictory as even such Wahlund effect would quickly vanish due to the homogenization of parasite populations that such migration pattern would produce, unless one admits these migrating strains are unable to survive locally for unknown reasons. Given the formidable spread this parasite has experienced across the world [4], such an explanation appears poorly convincing.
Another interpretation would come from [10] and would mean T. evansi stocks are continuously filled with recombining T. brucei brucei that would recurrently loose their maxicircle kDNA. This hypothesis requires that this phenomenon be enough frequent so that “brucei” and “evansi” stocks display exactly the same isolation by distance pattern. If this was true “evansi” stocks would never happen to cluster with any kind of marker. This is apparently not obviously the case as T. evansi strains can cluster together pretty well [64–65–66–58]. Moreover, we need a rather strong Wahlund effect to explain the shape of the NJTree under such a recurrent recombination events hypothesis. Though we cannot reject totally this hypothesis, it is not the most parsimonious and we will not consider it further.
If we do not assume full clonality for T. evansi our genetic data are difficult to interpret and in contradiction with what is expected from this strictly mechanically transmitted trypanosome [58]. Moreover, the shape of the NJTree obtained is compatible with frequent allelic dropouts, Wahlund effects or both. Sampling at much smaller scales and redesigning primers would offer the opportunity to test these hypotheses. If true, this would mean much smaller (more negative) FIS to be considered. In the opposite case, the possibility of recombination for T. evansi freed from the tsetse obligate stages or recurrently coming from mutant T. brucei brucei could be considered though we do not consider it as a likely explanation. This remains to be investigated further including T. brucei brucei strains in sites where both species can be found, which is hardly the case here as most our samples are North of the tsetse belt. This is an important issue as these different hypotheses (full clonality and strong viscosity versus sexual reproduction) have not the same consequences in term of spread of parasite resistance to trypanocidal drugs. Nevertheless, isolation by distance was evidenced with a relatively small number of migrants between neighboring sites at each generation. Given the strong structuring power of stepping stones, this means quite a strong viscosity as regards to parasite propagation of T. evansi across camel herds in Sudan.
|
10.1371/journal.ppat.1002446 | The Membrane Fusion Step of Vaccinia Virus Entry Is Cooperatively Mediated by Multiple Viral Proteins and Host Cell Components | For many viruses, one or two proteins allow cell attachment and entry, which occurs through the plasma membrane or following endocytosis at low pH. In contrast, vaccinia virus (VACV) enters cells by both neutral and low pH routes; four proteins mediate cell attachment and twelve that are associated in a membrane complex and conserved in all poxviruses are dedicated to entry. The aim of the present study was to determine the roles of cellular and viral proteins in initial stages of entry, specifically fusion of the membranes of the mature virion and cell. For analysis of the role of cellular components, we used well characterized inhibitors and measured binding of a recombinant VACV virion containing Gaussia luciferase fused to a core protein; viral and cellular membrane lipid mixing with a self-quenching fluorescent probe in the virion membrane; and core entry with a recombinant VACV expressing firefly luciferase and electron microscopy. We determined that inhibitors of tyrosine protein kinases, dynamin GTPase and actin dynamics had little effect on binding of virions to cells but impaired membrane fusion, whereas partial cholesterol depletion and inhibitors of endosomal acidification and membrane blebbing had a severe effect at the later stage of core entry. To determine the role of viral proteins, virions lacking individual membrane components were purified from cells infected with members of a panel of ten conditional-lethal inducible mutants. Each of the entry protein-deficient virions had severely reduced infectivity and except for A28, L1 and L5 greatly impaired membrane fusion. In addition, a potent neutralizing L1 monoclonal antibody blocked entry at a post-membrane lipid-mixing step. Taken together, these results suggested a 2-step entry model and implicated an unprecedented number of viral proteins and cellular components involved in signaling and actin rearrangement for initiation of virus-cell membrane fusion during poxvirus entry.
| Poxviruses are large DNA viruses that cause diseases in humans and other animals. To initiate infection, the core of the large, membrane-enveloped particle must penetrate into the cytoplasm where replication occurs. For most enveloped viruses only one or two proteins are needed for attachment and penetration. However, at least sixteen poxvirus proteins are dedicated to entry: four for attachment and twelve for penetration. The latter proteins form the entry fusion complex (EFC) and are conserved in all poxviruses indicating that the entry mechanism has been retained since the origin of the family. The purpose of the present study was to determine the cellular processes and poxviral proteins needed for fusion of the viral and cellular membranes. We found that a variety of inhibitors that interfered with cell signaling and reorganization of the actin cytoskeleton prevented membrane fusion as determined by lipid mixing, whereas others targeted the subsequent stage in entry. In addition, seven viral protein components of the EFC were required for the initial membrane fusion step, whereas three were not. A neutralizing monoclonal antibody to one of the latter also did not interfere with membrane lipid mixing but still prevented core entry supporting a 2-step poxvirus entry model.
| Entry of enveloped viruses into cells can be divided into three steps: (i) close apposition of viral and cellular membranes, (ii) lipid mixing of the outer membrane leaflets leading to formation of a hemifusion intermediate, and (iii) formation and expansion of a fusion pore allowing entry of the viral nucleoprotein or core into the cytoplasm [1]. One or two glycoproteins that provide cell binding and membrane fusion are sufficient to mediate entry of many enveloped viruses [2]. The process is more complex for members of the herpesvirus family, which employ four to five glycoproteins for entry [3]. Poxviruses represent an extreme case, as at least sixteen unglycosylated vaccinia virus (VACV) proteins participate in this process (referenced below). The large number of poxvirus proteins and the absence of any that resemble conventional membrane fusion proteins by sequence suggest a novel entry mechanism. For mature virions (MVs), the basic and most abundant infectious VACV particle, entry can occur by fusion at the plasma membrane [4], [5] or in a low pH-dependent manner from within an intracellular vesicle, depending to some extent on the virus strain [6], [7] and cell type [7]–[9]. Endocytosis of MVs is believed to occur by macropinocytosis [10]–[15] or dynamin-mediated fluid phase uptake [16], consistent with a role for actin dynamics and cell signaling. Progeny virions that depart the cell by exocytosis contain an additional membrane that helps escape antibody neutralization and is ultimately ruptured to allow fusion of the enclosed MV with the plasma membrane or endocytic vesicle [17], [18].
Four VACV proteins are involved in attachment of MVs [19]–[22] and twelve, conserved in all members of the poxvirus family, participate in subsequent entry steps [23]–[34]. Initial binding to target cells occurs via interactions of the MV attachment proteins with cell surface glycosaminoglycans or laminin. A cellular protein, referred to as VACV penetration factor, appears to be important for entry but exactly how is not yet understood [16]. The twelve conserved VACV entry proteins are mostly small, ranging in size from 35 to 377 amino acids, and have a N- or C-terminal transmembrane domain. The proteins are all components of the MV membrane, which is formed within the cytoplasm by incompletely defined mechanisms rather than by budding as typically occurs with other viruses [35]. This feature, as well as the association of most or all the proteins in a complex [31], makes it difficult to investigate the roles of individual entry proteins. A useful approach has been to construct conditional lethal mutants, with one putative entry gene controlled by the Escherichia coli lac operator/repressor system and positively regulated by ß-D-isopropylthiogalactopyanoside (IPTG) inducer, or with an analogous tetracycline-inducible system. These mutants share similar phenotypes: in the presence of inducer, replication proceeds normally and the progeny virions contain the protein product of the inducible gene and are infectious; in the absence of inducer, progeny virions appear indistinguishable from wild type by electron microscopy and protein analysis (except for the missing entry protein) but have very low infectivity. Although the non-infectious virions bind to cells, immunofluorescence microscopy studies show reduced numbers of cores in the cytoplasm. With the exception of I2 [30], repressed expression of the individual proteins does not significantly reduce the trafficking of the others to the MV membrane. However, when expression of an individual component is repressed, the formation or stability of the complex is reduced, as determined by detergent extraction and immunoaffinity purification [31]. The proteins A16, A21, A28, G3, G9, H2, J5, L5 and O3, make up the central components of the so-called entry fusion complex (EFC). The L1 and F9 proteins are also required for entry; although they physically interact with the EFC, they are not required for assembly or stability of the complex, and consequently have been referred to as EFC-associated proteins [26], [32]. The overall structure of the EFC has not been elucidated, though several pair-wise protein interactions have been identified [36]–[38].
The mechanisms involved in poxvirus entry are poorly understood. Previous studies have depended on post-membrane fusion assays and a specific role of the EFC in fusion could only be inferred from the inability of cells infected with the mutant viruses made in the absence of IPTG to undergo low pH-induced syncytia formation. Thus, direct evidence for a role of EFC proteins in membrane fusion during entry of virions has been lacking. Here, we used a variety of approaches including cell binding, membrane lipid mixing, core entry and reporter gene expression (Figure 1) to evaluate the roles of host components and individual MV membrane proteins.
Fusion of viral and cellular membranes involves lipid mixing, which can be studied by loading a self-quenching fluorescent probe such as octadecylrhodamine (R18) into viral membranes (Figure 1). Fusion of viral and cell membranes results in dilution of the probe and increased fluorescence [39]. Dequenching does not require full fusion of the viral and cell membrane but can occur at the initial step in which only the outer leaflets of the viral and cellular membranes fuse, known as hemifusion [1]. Therefore, dequenching could signify the occurrence of hemifusion alone or full fusion with pore formation. In a 2-step membrane fusion model (see Discussion), inhibitors that prevent dequenching must operate at or prior to the hemifusion step, which precedes full fusion.
In the present experiments, sucrose gradient purified VACV MVs were incubated with R18 at room temperature for 20 min. Incorporation of R18 into MVs minimally affected infectivity as shown in Figure 2A. After removal of excess R18, the MVs were incubated with HeLa cells for 1 h at 4°C to allow adsorption and then the temperature was raised to permit fusion. R18 fluorescence was more rapid at the physiological temperature of 37°C than at 20°C (Figure 2B), consistent with an active transfer process. We used WRvFire, a recombinant VACV that expresses firefly luciferase (LUC) regulated by an early promoter, to compare the kinetics of fusion and reporter gene expression. Whereas fusion occurred within a few minutes after incubation of virus-bound cells, LUC expression was detected at 40 min (Figure 2C) and was routinely assayed after 1 or 2 h.
The above results supported the use of the fluorescent R18 probe for analyzing VACV-cell membrane fusion. In subsequent experiments we compared the effects of inhibitors on binding of virions to cells, fusion, and core entry as measured by LUC expression and in some cases by transmission electron microscopy.
An earlier study had shown that fusion of VACV strain WR was not enhanced at low pH [40], which in retrospect seemed surprising in view of the subsequent demonstration of low pH enhancement of core entry and reporter gene expression [6]. Nevertheless, we confirmed the similar rates of VACV WR fusion following a brief incubation with a pH 7.4 or pH 5.0 buffer and return to neutral pH (Figure 2D). Furthermore, we found that bafilomycin A1, which prevents endosomal acidification and reduces firefly LUC expression, had little effect on binding of MVs containing a Gaussia LUC core protein chimera or membrane fusion (Figure 3A), similar to previous findings of membrane fusion in the presence of ammonium chloride and chloroquine [40]. Thus, low pH promotes an entry step beyond membrane lipid mixing.
Depletion of cellular cholesterol reversibly prevents the accumulation of VACV cores in the cytosol at a post-attachment step [41]. Treatment of HeLa cells with methyl- ß-cyclodextrin (mßCD) resulted in up to a 74% reduction in total cellular cholesterol levels (Figure S1A) without reducing cell viability over the time-course of the experiment (Figure S1B), although some cell rounding occurred. Nevertheless, MVs efficiently bound to cholesterol-depleted HeLa cells and R18 fluorescence was only mildly reduced, whereas LUC expression was greatly inhibited (Figure 3A). These data indicated that the lowered level of cellular cholesterol was sufficient for membrane lipid mixing but impaired a later step in entry or reporter gene expression.
Inhibitors targeting membrane blebbing, dynamin function, actin dynamics, and the activities of certain protein kinases have been shown to reduce VACV entry to varying extents as measured by reporter gene expression or detection of cytoplasmic cores [11]–[13], [16], [42]. In the present experiments, HeLa cells were preincubated for 30 min with inhibitors at previously used concentration ranges and the drugs were maintained in the medium during and after virus adsorption. Infection with VACV induces actin-enriched protrusions or cellular blebs [42] and entry can be partially reduced by blebbistatin, a small molecule specific inhibitor of myosin-II-dependent blebbing, virus movement along filopodia and macropinocytosis [11], [43], [44]. Blebbistatin was without effect on virion attachment but reduced LUC reporter expression by about 50% (Figure 3A), similar to the value previously reported for a GFP reporter assay [11]. However, we found little or no effect on dequenching of the R18 probe (Figure 3A), indicating that membrane fusion can occur independently of cell membrane blebbing.
Dynasore is a small molecule inhibitor of the GTPase activity of dynamin1, dynamin2 and the mitochondrial dynamin and is a rapid and potent inhibitor of dynamin-dependent endocytic pathways [45]. Dynamin also directly interacts with actin and regulates the actin cytoskeleton [46]–[48]. The effect of dynasore on VACV entry is ambiguous as it was reported not to influence entry in some studies [11] but to inhibit entry in another [16]. We found that dynasore had no effect on virion binding to HeLa cells but severely decreased LUC expression (Figure 3A). Moreover, dynasore potently inhibited membrane fusion (Figure 3B). These results implicated cellular dynamin as a critical factor in promoting VACV entry into HeLa cells at the membrane fusion step.
We also tested several specific inhibitors of actin dynamics: CK-636 and CK-548 bind to the Arp2/3 complex and prevent actin nucleation whereas latrunculins and cytochalasins bind actin and inhibit polymerization [49], [50]. These drugs had little effect on virion attachment but severely blocked LUC expression (Figure 3B). CK-548 and CK-636 were also very effective inhibitors of membrane fusion, whereas latrunculin A and cytochalasin D inhibited fusion by approximately 50% at the concentrations used (Figure 3B). These studies indicated a role for actin rearrangement in membrane fusion and raised the possibility that the effect of dynasore was related to its influence on the actin cytoskeleton rather than endocytosis.
Cell signaling has been reported to have a role in VACV entry at the stage of blebbing and macropinocytosis [11]. Genestein, gefitinib (Iressa) and 324674 (PD153035) are small molecule tyrosine kinase inhibitors [51], [52]. These drugs did not reduce virion binding but profoundly inhibited LUC expression (Figure 3C). Moreover, they also greatly inhibited membrane fusion (Figure 3C). The results could be related to the relative specificity of gefitinib and 324674 for epidermal growth factor receptor signaling, which causes rapid actin polymerization and rearrangement [53].
Based on a previous report [11], we attempted to bypass the effects of inhibitors of actin remodeling and signaling on entry by brief low pH treatment of cells with attached virions. However, in our hands, such treatments only alleviated the effects of drugs such as bafilomycin A1, concanamycin and monensin that prevented endosomal acidification [6] but did not bypass the effects of several other inhibitors on entry as measured by LUC expression or R18 dequenching (Figure S2).
Core entry steps were also analyzed by transmission electron microscopy. The results cannot be precisely compared to the above assays because a high virus multiplicity and spinoculation were used to allow counting of a sufficient number of virus particles in thin sections of infected cells. Hemifusion cannot be detected by this procedure and the earliest recognizable entry step consisted of full fusion of the viral and plasma membranes with an open pore allowing core entry (Figure 4A). Although MVs can be readily detected in vesicles, full fusion of viral and vesicle membranes are rarely seen (5). Cores that accumulate in the cytoplasm (Figure 4B) could have entered through the plasma membrane or an endocytic vesicle. In the absence of inhibitors, the number of plasma membrane full fusion images decreased and cores in the cytoplasm increased between 30 and 90 min (Figure 4C, D). At both times, the numbers of plasma membrane full fusion images (Figure 4C) and cytoplasmic cores (Figure 4D) were reduced when the cells were treated with blebbistatin, dynasore, latrunculin A or cytochalasin D. These observations confirmed the results obtained with the LUC assay for measuring core entry.
In summary, our data are generally consistent with other studies showing the importance of cell signaling and remodeling of the actin cytoskeleton on VACV entry [10]–[15], and importantly further demonstrate that these activities are necessary for the membrane fusion step. Low pH, cholesterol and membrane blebbing appear to be more important for entry steps beyond membrane lipid mixing.
Most or all of the MV membrane proteins required for entry, as distinguished from cell attachment, are components of the EFC (A16, A21, A28, G3, G9, H2, J5, L5, O3) or physically associated with the EFC (L1, F9). We employed conditional lethal mutants for all EFC and EFC-associated proteins except J5, for which a stringent mutant was unavailable. As a control, we tested a mutant with a deletion of the gene encoding the I5 MV membrane protein that is not required for entry [54]. The recombinant viruses were replicated in the presence or absence of the IPTG inducer and the MVs were purified by sucrose gradient sedimentation. For each mutant, the number of purified virions was determined from the optical density. In some cases, virions were inactivated at 56°C prior to adsorption to cells as an additional control [55]. Equivalent numbers of particles were loaded with R18 and washed by sedimentation to remove excess dye. Dye transfer to HeLa cells was determined by increased fluorescence as in the preceding sections. In addition parallel cultures were maintained for 48 h and the yield of infectious virus determined by plaque assay. As expected, R18-loaded MVs lacking the I5 protein (I5−) promoted R18 probe transfer as efficiently as wild type MV (I5+), whereas transfer was reduced with the heat-inactivated MVs (Figure 5A). Virions deficient in individual EFC and EFC-associated proteins had very low infectivity and except for A28, L1 and L5 mutants exhibited severely reduced R18 dequenching as well (Figure 5B-K), providing the first evidence of a direct role of EFC proteins in the membrane fusion step of virus entry. Previous studies had only shown that the EFC was required for fusion of infected cells.
We used transmission electron microscopy to monitor core entry steps, following attachment of H2+, H2−, A28+ and A28− virions. We chose H2 and A28 as examples of mutants that reduced and allowed R18 dequenching, respectively (Figure 5G, I). As indicated earlier, a high multiplicity and spinoculation was needed because of the thin cell sections. The lower numbers of full fusions with pore formation at the plasma membrane and cytoplasmic cores in cells infected with H2− virions compared to H2+ virions were expected in view of the inability of the former to mediate R18 dequenching (Figure 6A,B). However, there was a similar reduction in full fusion images at the plasma membrane and cytoplasmic cores after infection with A28− virions compared to A28+ virions (Figure 6C, D) despite the greater ability of the former to allow membrane fusion as determined by lipid mixing. Inhibition of core entry was previously shown using a confocal microscopy assay for virions deficient in L1 [32] and L5 [34] confirming an entry block despite their ability to allow lipid mixing as shown here.
The above results showing that L1-deficient virions allowed membrane fusion but not core entry led us to investigate the effect of a potent L1-neutralizing monoclonal antibody (MAb) [56]. We found that a concentration of L1 MAb that severely inhibited core entry as determined by LUC expression and formation of infectious virus had minimal effect on membrane fusion as determined by R18 dequenching (Figure 7). This result was confirmed by a flow cytometry-based 1,1′-dioctadecyl-3,3,3′,3′-tetramethylindodicarbocyanine (DiD) lipid mixing assay using a wide-range of MAb concentrations (Figure S3).
We still needed to consider the possibility that the role of the EFC is to activate the cell for virion entry rather than to directly participate in the entry step per se. In this context, Mercer and Helenius [11] had reported that very few VACV particles are needed to induce widespread blebbing and actin rearrangement. To further investigate the role of the EFC in entry, we coinfected cells with wild type VACV and either A28+ or A28− virions that expressed firefly LUC. We used a particle/cell multiplicity of approximately 200 for the A28+ and A28− virions and varied the multiplicity of the wild type virions from 9 to 1840 particles/cell (equivalent to 0.1 to 20 plaque forming units (PFU)/cell). Coinfection with wild type virions caused a two-fold increase in LUC expression by A28+ virions and raised expression about four-fold for A28− virions (Figure 8A). However, the latter was still only 3% of the value for A28+ virions indicating that efficient trans-complementation had not occurred. We also determined that soluble A28 protein [57] mixed with virions had no effect on entry of either the A28+ or A28− virions (Figure 8B).
Viral and cellular membranes each consists of two leaflets and in principal membrane fusion could occur by two different pathways as discussed by Chernomordik [1]. The direct fusion model posits that pores form in each of the apposing membranes and the pore rims join forming a fusion pore that allows lipid and content mixing in a single step. In contrast, the 2-step model posits fusion of the outer leaflets of the apposing membranes to form a hemifusion intermediate followed by merging of the inner leaflets to form the fusion pore. In the latter model, lipid mixing and content mixing occur sequentially. Evidence to support the second model involving a hemifusion intermediate has been obtained for several different viruses by demonstrating membrane lipid mixing without content mixing by mutation of viral fusion proteins, slowing or interrupting fusion with inhibitors and decreasing the surface density of viral fusion proteins [58]–[61]. In the present study of VACV, we showed that membrane lipid mixing could occur without core entry under three circumstances: depletion of certain EFC proteins (A28, L1 or L5), neutralization of VACV with a MAb to the L1 EFC-associated protein, and partial cholesterol depletion of the cell membrane. These findings are consistent with a 2-step entry model with a hemifusion intermediate for VACV.
In the first part of the Results, we described the effects of inhibitors of cell processes on virion attachment, membrane fusion and core entry. Most of the inhibitors had previously been shown to reduce entry as determined by reporter gene expression or detection of cytoplasmic cores [11]–[13], [16], [42]. We found that none of these inhibitors prevented binding of virions to cells, many reduced membrane fusion, while others only acted at the core entry step (Figure 9). The membrane fusion inhibitors were either directly involved with actin polymerization or remodeling (CK-636, CK-548, latrunculin A, cytochalasin D) or blocked tyrosine kinases that can modulate actin cytoskeletal changes (genestein, Iressa, 324674). The action of dynasore, a specific inhibitor of dynamin GTPase, could be due to its known effect on actin since there is evidence against a role for caveolae-mediated endocytosis in VACV entry [16]. Further evidence for dynamin2 in VACV core entry has been obtained with siRNA [16]. Extensive actin remodeling and mobilization has been observed during MV binding to cell surfaces [11], [16], [42] suggesting that actin-enriched membrane protrusions increase the intimacy of membrane contact and promote virus-cell membrane fusion. Actin remodeling has been suggested to facilitate fusion by forcing membranes together and enlarging pores in a variety of systems [62]–[64] including virus entry and viral protein-induced cell-cell fusion [65]–[70]. With human immunodeficiency virus, actin remodeling appears to have a more important role in pore expansion and content mixing than in hemifusion [71], [72]. We found that cytochalasin D and latrunculin A had a greater inhibitory effect on core entry (determined by LUC expression) than membrane fusion as determined by lipid mixing, suggesting that actin dynamics may be required for multiple steps in VACV entry.
In contrast to the role of actin rearrangement, inhibitors that prevented membrane blebbing involved in virus surfing and macropinocytosis or that interfered with the reduction in pH of endosomes, had a much greater effect on core entry than membrane lipid mixing (Figure 9). It will be important to determine whether lipid mixing is occurring at the plasma membrane or in endosomes at neutral pH. Similarly, a 74% reduction of cellular cholesterol with mßCD had little effect on membrane fusion but had a major effect on core entry as measured by LUC expression. A previous study had shown that MVs associate with cholesterol-rich regions of the plasma membrane and that cholesterol depletion reduced VACV entry as measured by visualizing cores in the cytoplasm [41]. In studies with influenza virus and Semliki Forest virus in insect cells, which can be more stringently depleted of cholesterol than mammalian cells, both hemifusion and pore widening were affected [73], [74]. The cell surface receptors for certain viruses reside in cholesterol-rich lipid rafts, but receptors for VACV have not been identified.
The VACV EFC proteins were previously shown to be required for virus core entry and cell-cell fusion but evidence for a role in the fusion of viral and cell membranes had been indirect. Of the ten EFC or EFC-associated mutants tested in the present study, all were blocked in core entry as determined by infectivity or transmission electron microscopy and seven of these were unable to mediate membrane fusion. The three proteins apparently not required for membrane fusion were A28, L1, and L5. It is possible that these proteins have a specific role at a later step in entry such as pore formation. However, in other systems it has been shown that the density of activated fusion proteins has to be higher for the formation and expansion of a fusion pore than for hemifusion [1]. Although these three mutants each display stringent repression of EFC protein expression as shown by Western blotting, undetectable differences could affect the sensitive lipid-mixing assay. Therefore, our main conclusion is that the EFC is required for membrane fusion and that additional studies are required to conclude that A28, L1 and L5 have a specific role at a later step of entry such as pore formation.
The L1 protein is a target of potent neutralizing and protective antibodies [56], [75]. The structure of L1 alone and in association with a conformation-specific MAb has been solved to high resolution [76], [77]. The Fab fragment binds to a discontinuous epitope containing two loops that are held together by a disulfide bond. Here we showed that the MAb prevents VACV entry at a step beyond lipid mixing, consistent with the effect on entry of virions deficient in the L1 protein.
Since our inhibitor studies had shown that actin dynamics are required for membrane fusion and core entry, we considered the possibility that the EFC indirectly promotes entry by inducing cell signaling. Indeed, such a role could contribute to the need for multiple EFC proteins. Since Mercer and Helenius [11] had shown that cell signaling requires few virus particles, we tried to rescue EFC protein-deficient virions in trans by coinfecting with wild type VACV. Although wild type virus enhanced core entry by four-fold as measured by LUC expression, this value was still only 3% of that achieved by the control virus, suggesting that the EFC proteins have a direct role in membrane fusion and entry. Nevertheless, whether EFC protein interactions also cause signaling is an interesting question for future studies.
Why so many different proteins are needed for poxvirus entry remains an enigma. None of the proteins resemble type I or type II viral fusion proteins by sequence so that determination of the 3-dimensional structure of the VACV EFC may be needed to define putative fusion loops, if the mechanism of entry involves such structures. At this time, only the structure of the L1 EFC-associated protein has been solved [76].
African green monkey kidney BS-C-1 and human HeLa cells were maintained in minimum essential medium with Earle's salts (EMEM) supplemented with 2.5% fetal bovine serum (FBS), 2 mM L-glutamine, 100 U/ml penicillin, and 100 µg/ml streptomycin (Quality Biological). The recombinant VACV WRvFire expressing firefly LUC under a synthetic early/late VACV promoter was described previously [6]. Recombinant VACVs in which expression of individual EFC or EFC-associated proteins are IPTG-inducible have been previously constructed and characterized: A16 [23], A21 [24], A28 [25], G3 (A. Townsley and BM, unpublished), G9 [28], H2 [29], J5 [31], L5 [34], O3 [33], L1 [32], and F9 [26]. The recombinant VACV in which the I5L gene was deleted has been described [54]. The recombinant VACV Gauss-A4 (parental strain WRvFire), which expresses the Gaussia LUC enzyme fused to the A4 core protein was generated as follows. Overlap polymerase chain reaction (PCR) was utilized to generate a construct in which the Gaussia LUC gene (New England Biolabs) was appended to the N-terminal codon of the VACV A4L gene and the EGFP coding region (and accompanying synthetic early/late VACV promoter sequence) was placed downstream of the Gaussia-A4L region. To achieve homologous recombination, flanking genomic sequences of A4L (approximately 500 bp in length) were appended to the termini of the PCR product. HeLa cells were infected with 0.05 PFU of WRvFire per cell and at 2 h post infection were transfected with 400 ng of purified PCR product using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol. At 24 h post infection, the infected cells were lysed by five freeze/thaw cycles and clonally purified five times by picking GFP positive plaques on BS-C-1 cells. The recombinant VACV in which A28L is IPTG-inducible and expresses firefly LUC under a synthetic early/late VACV promoter has been described [25].
BS-C-1 cells were infected with VACV in the presence or absence of the inducer IPTG (Calbiochem) and at 48 to 72 h post infection MVs were isolated as described [78], [79]. Briefly, infected cells were subjected to Dounce homogenization and MVs were purified by sedimentation through two 36% (wt/vol) sucrose cushions followed by one sedimentation on a 25 to 40% (wt/vol) continuous sucrose gradient; the visible virus band was collected, and virus was pelleted and stored at −80°C. Upon thawing, virus was sonicated on ice for 1 min. The infectious viral titer (PFU per ml) for each purified MV stock of recombinant VACV was determined by plaque assay on BS-C-1 cells as described [80]. Additionally, the number of total virus particles obtained for each purified MV stock of recombinant VACV was estimated from the optical density at 260 nm [80].
Purified MVs (approximately 9.0×109 particles) were labeled with 3 ml of 1 mg/ml of R18 (Molecular Probes) in phosphate-buffered saline (PBS; Quality Biological) + 0.2% bovine serum albumin (BSA; Sigma-Aldrich) for 20 min at room temperature in the dark. Non-incorporated R18 was removed by pelleting virions (16,000 x g for 10 min at 4°C) and washing several times in PBS + 0.2% BSA. R18-labeled virions were re-suspended in PBS + 0.2% BSA, vortexed, and sonicated for 15 sec on ice. Virions sufficient to achieve a multiplicity of 1 to 5 PFU (or the equivalent number of non-infectious particles) per cell were then incubated with approximately 1.5×106 HeLa cells in suspension for 1 h at 4°C in cold fusion medium comprised of EMEM without phenol red and with 10 mM N-2-hydroxyethylpiperazine-N'-2-ethanesulfonicacid (HEPES) and 10 mM 2-(N-morpholino)ethanesulfonic acid (pH 7.4) in the dark. Virus-bound cells were washed twice with cold fusion medium following low-speed centrifugation (750 x g for 3 min at 4°C). Virus-bound cells were injected into a cuvette containing fusion medium pre-warmed to 37°C and kept in suspension utilizing a magnetic stir bar. R18 fluorescence (560 nm excitation and 590 nm emission) was monitored by use of a Fluoro-Max3 spectrofluorometer (Horiba Jobin Yvon) outfitted with a Peltier sample cooler (Horiba Jobin Yvon) and a temperature control unit (Wavelength Electronics model LFI-3751) to maintain the desired temperature within the chamber housing the sample cuvette. For graphical presentation, the raw fluorescence data were plotted versus time. For quantitative comparisons, we determined the percent fluorescence by dividing the value obtained at 40 min by the value obtained following addition of Triton X-100 (1% [wt/vol] final concentration).
HeLa cells seeded in 24-well plates (2.0×105 cells per well) were chilled to 4°C before virus adsorption. WRvFire MVs were adsorbed in cold EMEM + 2.5% FBS for 1 h at 4°C. Cells were washed with cold PBS to remove unbound virions and incubated with pre-warmed EMEM + 2.5% FBS for 2 h (unless indicated otherwise) at 37°C. Cells were washed with PBS and then incubated with Cell Culture Lysis Reagent (Promega) for 30 min at room temperature with gentle agitation. LUC activity in cellular extracts was measured according to the manufacturer's protocol (Promega) and quantified on a Berthold Sirius luminometer (Berthold Detection Systems).
HeLa cells seeded in 24-well plates (2.0×105 cells per well) were left untreated or treated with 10 mM mßCD (Sigma-Aldrich) for 30 min in EMEM at 37°C. Cells were then washed with cold PBS and cold EMEM was added to cells prior to virus adsorption at 4°C for R18 hemifusion or LUC entry assays as described above. Cholesterol levels in HeLa cells were determined using the Amplex Red Cholesterol Assay Kit (Molecular Probes) and was performed according to the manufacturer's protocol. The viability of mßCD-treated cells was assayed using the CellTiter 96 Aqueous One Solution Cell Proliferation Assay (Promega) and was performed according to the manufacturer's protocol.
HeLa cells were left untreated or pre-treated with the indicated concentrations of inhibitors: Sigma-Aldrich: blebbistatin (75 µM), dynasore (100 µM), bafilomycin A1 (50 nM), latrunculin A (10 µM), cytochalasin D (10 µM), CK-548 (100 µM), CK-636 (100 µM), genistein (100 µM); LC Laboratories: Iressa (40 µM); EMD4Biosciences: 324674 (40 µM) for 30 min at 37°C. Cells were then chilled to 4°C prior to virus adsorption for virus-cell binding, R18 hemi-fusion, or LUC assays as described. The indicated drug concentrations were maintained throughout the assay.
Equivalent amounts of VACV Gauss-A4 virions (5 PFU per cell) were incubated with untreated or inhibitor-treated HeLa cells in 24-well plates at neutral pH for 1 h at 4°C. Cells were washed twice with cold PBS to remove unbound virus. Cells were then incubated with LUC assay lysis buffer (Promega) for 30 min at room temperature with gentle agitation. Gaussia LUC activity in cellular extracts was measured according to the manufacturer's protocol (Promega) and quantified on a Berthold Sirius luminometer (Berthold Detection Systems).
Low pH stimulation of virus entry was performed as described previously [6]. Following a wash to remove unbound virions, cells were incubated for 3 min in 37°C PBS with Ca2+ and Mg2+ at pH 7.4 or PBS with Ca2+ and Mg2+ supplemented with 1 mM 2-morpholinoethane-sulfonic acid adjusted to pH 5.0 with HCl. After removal of buffers, the pH was neutralized by one wash with EMEM + 2.5% FBS. Cells were incubated in pre-warmed EMEM + 2.5% FBS for 2 h at 37°C and then prepared for the LUC entry assay as described above.
BS-C-1 cells in six-well tissue culture plates (1.0×105 cells per well) were pre-chilled at 4°C for 30 min prior to virus spinoculation. Purified MVs (350 PFU per cell or equivalent number of particles) in cold EMEM + 2.5% FBS were sedimented onto the BS-C-1 cells at 4°C for 1 h at 650 x g in a Legend RT centrifuge (Sorvall). Cells were washed with cold PBS to remove unbound virions and incubated with pre-warmed EMEM + 2.5% FBS for varying amounts of time at 37°C. At the indicated time, the samples were fixed on ice with 4% paraformaldehyde (Electron Microscopy Sciences) in 0.1 M phosphate buffer for 10 min and processed for transmission electron microscopy as described previously [6]. For quantitation of virus entry events, ninety randomly selected cell sections were visualized and particles therein counted.
Equivalent numbers of R18-loaded MV particles (recombinant strain WRvFire) were incubated with 100 µg/ml of anti-L1 mouse MAb 7D11 [56] or control anti-HA mouse monoclonal (clone 16B12, Covance) for 30 min at room temperature. Virion and antibody mixtures were then divided and used for R18-based fusion, LUC core entry, or plaque formation assays as described above.
Purified MVs (approximately 9.0×109 particles) were labeled with 3 µl of DiD (Molecular Probes) in phosphate-buffered saline (PBS; Quality Biological) + 0.2% bovine serum albumin (BSA; Sigma-Aldrich) for 20 min at room temperature in the dark. Non-incorporated DiD was removed by pelleting virions (16,000 x g for 10 min at 4°C) and washing several times in PBS + 0.2% BSA. DiD-labeled virions were re-suspended in PBS + 0.2% BSA, vortexed, and sonicated for 15 sec on ice. Virions sufficient to achieve a multiplicity of 1 to 5 PFU per cell were then incubated with approximately 8.0×104 HeLa cells in a 48-well plate for 90 min at 37°C in minimum essential medium with Earle's salts (EMEM) supplemented with 2.5% FBS, 2 mM L-glutamine, 100 U/ml penicillin, and 100 µg/ml streptomycin. Cells were washed with PBS, trypsinized, spun and fixed in 4% paraformaldehyde/PBS for 2 h at 4°C. DiD-positive cells were quantified using a FACSCalibur (BD Biosciences). DiD loading had minimal effect on virus infectivity as measured by plaque assay.
|
10.1371/journal.pgen.1000955 | Ablation of Whirlin Long Isoform Disrupts the USH2 Protein Complex and Causes Vision and Hearing Loss | Mutations in whirlin cause either Usher syndrome type II (USH2), a deafness-blindness disorder, or nonsyndromic deafness. The molecular basis for the variable disease expression is unknown. We show here that only the whirlin long isoform, distinct from a short isoform by virtue of having two N-terminal PDZ domains, is expressed in the retina. Both long and short isoforms are expressed in the inner ear. The N-terminal PDZ domains of the long whirlin isoform mediates the formation of a multi-protein complex that includes usherin and VLGR1, both of which are also implicated in USH2. We localized this USH2 protein complex to the periciliary membrane complex (PMC) in mouse photoreceptors that appears analogous to the frog periciliary ridge complex. The latter is proposed to play a role in photoreceptor protein trafficking through the connecting cilium. Mice carrying a targeted disruption near the N-terminus of whirlin manifest retinal and inner ear defects, reproducing the clinical features of human USH2 disease. This is in contrast to mice with mutations affecting the C-terminal portion of whirlin in which the phenotype is restricted to the inner ear. In mice lacking any one of the USH2 proteins, the normal localization of all USH2 proteins is disrupted, and there is evidence of protein destabilization. Taken together, our findings provide new insights into the pathogenic mechanism of Usher syndrome. First, the three USH2 proteins exist as an obligatory functional complex in vivo, and loss of one USH2 protein is functionally close to loss of all three. Second, defects in the three USH2 proteins share a common pathogenic process, i.e., disruption of the PMC. Third, whirlin mutations that ablate the N-terminal PDZ domains lead to Usher syndrome, but non-syndromic hearing loss will result if they are spared.
| Usher syndrome is a devastating genetic disorder affecting both vision and hearing. It is classified into three clinical types. Among them, type II (USH2) is the predominant form accounting for about 70% of all Usher syndrome cases. Three genes, USH2A, USH2C, and USH2D, underlie the development of USH2; and they encode usherin, Very Large G protein-coupled Receptor-1 (VLGR1), and whirlin, respectively. In this study, we show that the long whirlin isoform organizes the formation of a multi-protein complex in vivo that includes usherin and VLGR1. Targeted disruption of whirlin long isoform abolishes the normal cellular localization of the two partner USH2 proteins in the retina and in the inner ear and causes visual and hearing defects. We present the first definitive evidence that the USH2 proteins mark the boundary of the periciliary membrane complex, which was first described in frog photoreceptors and is thought to play a role in regulating intracellular protein transport. We propose that defects in all USH2 proteins share a common pathogenic pathway by disrupting the periciliary membrane complex in photoreceptors.
| Usher syndrome manifests as both retinal degeneration and hearing loss [1], [2]. It is classified into type I, II, and III based on clinical features of the hearing defects [3]–[8]. Usher syndrome type I (USH1) presents with profound congenital deafness and vestibular dysfunction. USH2 is the most common form and is characterized by moderate non-progressive hearing loss without vestibular dysfunction. USH3 is distinguished from USH2 by the progressive nature of its hearing loss and occasional vestibular dysfunction. There is further genetic heterogeneity within each clinical type of Usher syndrome. For example, three distinct gene loci, referred to as USH2A, USH2C and USH2D, are known to underlie USH2. These three genes encode the USH2A protein (also known as usherin), Very Large G protein-coupled Receptor-1 (VLGR1) and whirlin, respectively. Among these, mutations in USH2A account for over 70% of USH2 patients whereas USH2C and USH2D are responsible for the remainder. A previously proposed USH2B locus was subsequently shown to be in error and has been withdrawn [9].
Genetic defects in the whirlin gene have long been known as a cause of nonsyndromic deafness DFNB31 [10], [11] and, more recently, were found to underlie USH2D [12]. Whirlin R778X and c.2423delG mutations (Figure 1A) that truncate the protein close to its C-terminus cause profound prelingual hearing impairment in humans. In the naturally occurring whirler mouse, from which the name whirlin was derived, a large deletion was found in the middle of the whirlin gene (Figure 1A). Similar to human patients with DFNB31, the whirler mouse suffers from inner ear defects [10]. Neither patients with DFNB31 nor the whirler mouse manifest any retinal deficits. The whirlin gene defect underlying USH2D arises from compound heterozygosity of a Q103X mutation and a c.837+1G>A mutation [12], which are positioned in the first and second exon of the whirlin gene, respectively (Figure 1A). Therefore, different mutations of the whirlin gene account for a spectrum of hearing and vision defects although the mechanism underlying the variable disease expression of different mutations in the whirlin gene is not known.
Multiple whirlin transcript variants were found in the inner ear [10], [13], [14]. They are conceptually translated into two groups of proteins, the long and short isoforms (Figure 1A). The whirlin long isoform contains two N-terminal PDZ domains, a proline-rich domain and a third PDZ domain near the C-terminus. Heterogeneity in the whirlin short isoform arises from use of alternative transcriptional start sites and/or splicing sites of the whirlin gene, which generates several variants with different N-termini. The short isoform has no N-terminal PDZ domains but retains the proline-rich region and the third C-terminal PDZ domain. Both the PDZ domain and proline-rich region are modular protein interaction domains. PDZ domains bind to a short conserved sequence, known as a PDZ-binding motif, present at the C-terminus of proteins or found as an internal motif [15]. A proline-rich region usually binds to WW and SH3 domains [16]. With these two types of protein interaction domains, whirlin is believed to be engaged in the assembly of supramolecular complexes at specific subcellular locations. A series of in vitro analyses have found that whirlin is able to interact with usherin [17] and VLGR1 [14], the two causative proteins for other forms of USH2 [18]–[20]. A recent report demonstrates that these interactions probably exist at the ankle-link complex in developing hair cells [21].
A few reports have been published which examined the localization of whirlin in photoreceptors [14], [22], [23]. Whirlin has been reported to localize to the apical inner segment collar, the ciliary apparatus, the adherens junctions and the synaptic region of photoreceptors [14], [23]. However, there is no consensus from these reports on where the whirlin protein is localized in photoreceptors. As the photoreceptors are highly polarized neurons and are well organized into stratified layers of the retina, whether a protein is localized to the apical inner segment vs. the synaptic layer has completely different implication for its putative functions. More importantly, there has been no in vivo study of any kind on the association among the three USH2 proteins in photoreceptors. To fill in this knowledge gap, we carried out targeted disruption of the whirlin gene in mice at the 5′-terminal region. This disruption abolishes the long isoform and simulates the human mutations that cause USH2D. This mutant line of mice reiterated the vision and hearing defects of human USH2 patients. Using this mouse line and the Ush2A and Ush2C mutant mouse lines that had been previously generated, we analyzed the expression, localization and function of whirlin in the retina and compared them with those in the inner ear cochlea. We further analyzed the interaction among the USH2 proteins using those mouse lines as in vivo model systems. Our data provide new insight into the function of whirlin and other USH2 proteins and point to a possible disease mechanism for USH2. The data also help to explain the molecular basis for the variable disease expression caused by mutations in different regions of the whirlin gene.
A whirlin mutant mouse line was generated by replacing a portion of exon 1, which included the translation start codon for the whirlin long isoform, with a Neor expression cassette (Figure 1B). The targeted allele was confirmed by amplifying the genomic DNA fragments containing the junctional sequences between the whirlin gene and the Neor expression cassette. To determine if expression of whirlin was abolished in the mutant mice, we conducted RT-PCR and western blotting analyses in the retina. RT-PCR analysis verified that the first exon of whirlin transcripts was absent in the homozygotes (Figure 1D). Western blotting analysis showed that the whirlin long isoform, normally migrating at an apparent molecular weight of about 110 kDa, was completely absent in the retina of homozygous mice (Figure 1E). Thus, this targeted allele of whirlin is a null allele for the whirlin long isoform. To distinguish it from the existing whirler mice, we refer to this line of mutant mice as the whirlin knockout mouse. Whirlin knockout mice appeared viable and comparable to their wild-type littermates in growth characteristics, reproductive performance and general health.
To examine the normal expression of whirlin isoforms at the protein level in the retina, we generated a series of antibodies against whirlin and used two whirlin mutant mouse lines, whirlin knockout and whirler mice, as negative controls. In whirlin knockout mice, deletion of the first exon ablates the long isoform, while mutation in whirler mice eliminates the short isoform [10] (Figure 1A). Rabbit PDZIE, chicken PDZIE, and CIP98 [24] antibodies are directed against epitopes common in both the whirlin long and short isoforms (Figure 1A). Western blotting using these antibodies detected only the whirlin long isoform in the wild-type (WT) retina (Figure 2A), suggesting that the short isoform was either not expressed or was expressed at such a low level that was beneath the threshold of detection by this assay. To confirm this result, we enriched the whirlin protein(s) from the retinal lysate by immunoprecipitation using the rabbit PDZIE antibody, and then performed western blotting analysis of the precipitates using the chicken PDZIE antibody. While we found significant enrichment of the whirlin long isoform, we again did not detect the short isoform. As a positive control, we found both isoforms were enriched and readily detectable in the cochlear immunoprecipitate (Figure 2B). Therefore, the whirlin short isoform in the retina is a rare variant if expressed at all. In addition to the long and short variants reported previously, we found a distinct N-terminal transcript of whirlin in the retina by screening a mouse retinal cDNA library. This transcript terminates in the middle of the second PDZ domain such that if translated, it would produce a whirlin protein that includes only the first N-terminal PDZ domain. This transcript is therefore not affected by the whirler mutation or by the corresponding human mutations causing DFNB31 (Figure 1A). To examine whether this N-terminal whirlin isoform was abundant at the protein level in the retina, we performed immunoprecipitation using the rabbit PDZ320 antibody, whose antigen is the N-terminal 320 amino acids of whirlin (Figure 1A). Again only the whirlin long isoform was detected by western blotting using the chicken PDZ320 antibody (Figure 2B), suggesting this N-terminal whirlin isoform is not an abundant variant either. Nevertheless the presence of an N-terminal whirlin variant may be of functional significance. Taken together, these results clearly demonstrate that the whirlin long isoform is the predominant variant expressed in the retina.
Photoreceptors are highly polarized sensory neurons consisting of three major subcellular compartments, the outer segment, the inner segment and the synaptic terminus. Linking the light sensing outer segment and the biosynthetic inner segment is a thin bridge known as the connecting cilium. By immunofluorescence whirlin was found at the vicinity of the connecting cilia (Figure 3A and 3B) but not in the photoreceptor synaptic layer (the outer plexiform layer, data not shown). RPGR (retinitis pigmentosa GTPase regulator) and RP1 (retinitis pigmentosa 1) are proteins known to be localized at the connecting cilia and at the axonemal microtubules distal to the connecting cilia, respectively [25] (see Discussion). Double staining of whirlin with either RPGR or RP1 showed whirlin to localize adjacent to RPGR (Figure 3A) but proximal to RP1 (Figure 3B). However, unlike RPGR, RPGRIP1 and other ciliary proteins, immunostaining of dissociated photoreceptors, which include the outer segments and the connecting cilia, could not detect any whirlin signals at the connecting cilia (data not shown). This indicated that whirlin was not a core component of the connecting cilia. Immunoelectron microscopy from both longitudinal (Figure 3C) and cross (Figure 3D) sections found the immunogold labels of whirlin at a plasma membrane microdomain in the apical inner segment, which wraps around the connecting cilium and is usually destroyed in the dissociated photoreceptors. Thus, data from immunofluorescent staining and immunoelectron microscopy were consistent with whirlin localizing to a membrane microdomain that surrounds the connecting cilia, a location that is identical to that of usherin [26] (see Discussion).
The distribution pattern of whirlin in mouse photoreceptors was reminiscent of a structure called the periciliary ridge complex (PRC) found in frog photoreceptors [27]. The PRC was defined by a morphological feature, which includes a set of ridges and grooves with a nine-fold symmetry, seen by scanning electron microscopy. It marks a specialized domain on the plasma membrane of the inner segment that surrounds the base of the connecting cilium. To examine whether whirlin was localized at this structure in frogs, we generated an antibody against the C-terminus of frog whirlin. Double staining of whirlin with γ–tubulin and acetylated α–tubulin, markers of basal bodies and axonemal microtubules, respectively, showed that whirlin was localized immediately above the basal bodies (Figure 3E) and beneath the axonemal microtubules (Figure 3F). This is similar to the findings in mouse photoreceptors. In a cross sectional view, the signals of whirlin appeared as circles surrounding the basal bodies (Figure 3G). The diameter of these circles was approximately 2 µm, which is in the range of the previously determined diameter of the PRC [27]. Both rod and cone photoreceptors had the same distribution of whirlin (Figure 3F and 3H). These data suggest that whirlin is a resident protein at the PRC in frogs. The PRC as a morphologically distinct structure is not present in mammalian photoreceptors [28]. However, the conserved whirlin distribution in frog and mouse photoreceptors suggests that a functionally equivalent structure, delineated by the presence of whirlin, exists in the latter. We refer to this PRC-homologous membrane microdomain as the periciliary membrane complex (PMC). Thus, whirlin is a marker of the mammalian photoreceptor PMC.
The distribution of whirlin in photoreceptors was similar to that of USH2A protein (usherin), which was previously reported by our laboratory [26]. Usherin is predicted to have a PDZ-binding motif at its C-terminus [19]. We investigated whether whirlin and the cytoplasmic C-terminus of usherin interacted with each other. Yeast two-hybrid analysis demonstrated their interaction and the involvement of the first and second PDZ domains of whirlin in this protein binding (Figure 4A). We then sought further confirmation of their interaction by performing GST pull-down assays. We generated frog and mouse usherin-GST fusion protein constructs using either intact or mutant versions of the usherin C-terminal (intracellular) domain. The mutant usherin C-terminal domain lacked a functioning PDZ-binding motif. The expressed GST fusion proteins were incubated with mouse retinal lysate in an attempt to pull down whirlin. The results showed that the intact but not the mutant usherin C-termini were able to pull down endogenous whirlin from retinal lysate (Figure 4B). Therefore, our studies demonstrated that whirlin and usherin directly interacted with each other through the two N-terminal PDZ domains of whirlin and the C-terminal PDZ-binding motif of usherin. Our data support the findings of others reported in recent publications [14], [17].
We next evaluated the in vivo interaction between whirlin and usherin by double labeling immunofluorescence. In WT mouse photoreceptors, these two proteins colocalized fully at the PMC (Figure 5A). Examination of their distribution in the retinas of whirlin knockout, whirler and Ush2a knockout mice revealed profound perturbation of their localization pattern. In Ush2a knockout mice whirlin disappeared from the PMC. In whirlin knockout mice, usherin signals was largely absent from the PMC. In whirler mice, usherin staining was greatly reduced though not extinguished; trace amount of usherin staining was seen uniformly distributed at the PMC of all photoreceptors (Figure 5B). These results suggest that the normal localization of whirlin and usherin at the PMC depends on each other. Thus, ablation of usherin disrupts the normal localization of whirlin, and vice versa. The observation that usherin localization at the PMC was only partially disrupted in whirler mice is consistent with the lack of an overt retinal phenotype in these mice, and can be explained on the basis that the N-terminal PDZ domains of whirlin is not disrupted by the whirler mutation (see Discussion). Loss of binding partners also appeared to destabilize these two proteins. Western blotting analysis showed a reduction in the amount of usherin by 80% in the whirlin knockout mice (Figure 5C), and a reduction in whirlin by 50% in the Ush2a knockout mice (Figure 5D).
VLGR1 is the third known protein to be implicated in the USH2 etiology, and was previously reported to interact with whirlin in vitro [14]. Therefore, we studied whether VLGR1 was in the complex of whirlin and usherin in photoreceptors. Double staining of VLGR1 with either whirlin or usherin in the retina found VLGR1 to colocalize with both whirlin and usherin at the PMC in photoreceptors (Figure 6A and 6B). VLGR1 localization at the PMC in mouse rod and cone photoreceptors was further verified by immunoelectron microscopy (Figure 6C). Moreover, immunostaining demonstrated a decrease in VLGR1 signals at the PMC in whirlin and Ush2a knockout retina (Figure 6F), and an absence of whirlin (Figure 6D) and usherin (Figure 6E) proteins at the PMC in the Vlgr1 knockout retina. These results indicate that whirlin, VLGR1 and usherin form a multi-protein complex in vivo at the PMC in photoreceptors and that functional deficits in any of these three known USH2 proteins destabilize this complex and disrupt its function.
Along the cochlear spiral, there are one row of inner hair cells and three rows of outer hair cells. The inner hair cells are responsible for mechanoelectric transduction, whereas the electromotile outer hair cells also perform an electromechanical transduction, thereby amplifying the sound-evoked vibrations of the entire sensory epithelium. Both types of hair cells have stereocilia on their apical surfaces, which are modified microvilli filled with bundles of actin filaments. The tips of the stereocilia are for the sites of the mechanoelectric transduction channels. Because of the involvement of USH2 proteins in hearing impairment in humans, we studied their localization in the cochlea. Double staining of the cochleas from mice aged at postnatal day (P) 3–6 showed VLGR1 colocalized with whirlin and usherin in the stereocilia bundles of both inner (data not shown) and outer hair cells (Figure 7A and 7B). The three USH2 proteins are localized to the ankle-link complex of the hair cell stereocilia [21]. This ankle-link complex appears as fine extracellular fibers at the base of the stereocilia bundle during development (P2–P12) [29]; however, its exact function is not clear. To study whether the three USH2 proteins are interdependent at the ankle-link complex as at the PMC, we examined their distribution in hair cells in whirlin and Ush2a knockout mice at P3–P6. The signals of whirlin and VLGR1 were decreased in Ush2a knockout mice and the signals for usherin and VLGR1 were decreased in whirlin knockout mice (Figure 7C). These data are consistent with the reported findings of mislocalization of USH2 proteins in whirler mice and one line of the Vlgr1 mutant mice [21], and support the notion that whirlin, usherin and VLGR1 also form a multi-protein complex at the ankle-link complex of the stereocilia in hair cells, and the normal subcellular localizations of these three proteins are, to some extent, dependent on one another in the cochlea.
Retinal function tested by electroretinogram (ERG), a recording of the retinal electrical response to flashes of light, and histology examined by light microscopy did not reveal overt retinal degeneration in whirlin knockout mice up to 24 months of age (data not shown). However, morphological defects were evident at the ultrastructural level as early as 5 months of age. Examination by electron microscopy found membrane fusion between the apical inner segment and the connecting cilium and accumulation of vacuoles next to the PMC in the apical inner segment (Figure 8). The synaptic terminus of photoreceptors appeared normal (data not shown). Further analysis of whirlin knockout mice aged from 28 to 33 months found that the amplitudes of both a- and b-waves of dark-adapted ERG recordings significantly decreased compared to their heterozygous littermate controls. The light-adapted ERG amplitudes also decreased although the difference did not reach statistical significance (Figure 9A and Table 1). Histological examination of the eyes from this cohort of animals found that the photoreceptor nuclear layer was significantly thinner and the outer segments shortened in the whirlin knockout mice (Figure 9B and 9D), which are signs for retinal degeneration. Thus both functional and morphological assays indicate that the whirlin knockout mice develop late-onset retinal degeneration. In contrast, histological examination of the whirler mouse retina from 28 to 33 months of age did not find any abnormalities compared with age-matched wild-type controls (Figure 9C and 9E).
We measured distortion product otoacoustic emissions (DPOAE) to assay cochlear function in two groups of whirlin knockout mice at 2 and 9 months of age, respectively (Figure 10A). At both ages, knockouts showed no cochlear responses (i.e. thresholds were above the measurement ceiling at which the system produces its own distortion components), thus demonstrating a profound congenital hearing loss across all cochlear frequencies. Light microscopic evaluation of whirlin knockout ears at 2 months of age (data not shown, n = 2) showed only sporadic loss of hair cells in the mid-basal turn. All other accessory structures of the inner ear, including spiral ligament and stria vascularis, appeared normal. There did not appear to be a substantial loss of cochlear neurons. Scanning electron microscopy was performed to examine the morphology of cochlear stereocilia. Throughout the cochlear spiral, hair bundles on outer hair cells were abnormally compressed in the spiral dimension, i.e. the angle between the two limbs of the “V” shaped formation was smaller in the knockout ears (Figure 10B). In general, the hair bundle formation exhibits a “U” shape, which is a morphological defect characteristic of USH2 mutant mice [21], [26], [30]. Closer examination (Figure 10C) showed that, although some outer hair cell stereocilia bundles were normal with obvious interstereocilia links, many showed a patchy loss of stereocilia from the innermost (shortest) row of stereocilia. The inner hair cell stereocilia were normal in appearance throughout the cochleas. Given the critical role of outer hair cells in cochlear amplification and the production of DPOAEs, these stereocilia abnormalities in whirlin knockouts could explain the cochlear dysfunction.
The whirlin knockout mice characterized in this study have a late-onset retinal degeneration and a congenital, non-progressive hearing impairment. This phenotype reiterates the clinical features of USH2D disease in humans [12]. Therefore, this whirlin knockout mouse line is an appropriate animal model for studying the pathogenesis of this disease. In this study, we have provided definitive evidence on the in vivo interaction of whirlin with usherin and VLGR1 in both the retina and the inner ear. Because these three proteins are all involved in USH2, this finding suggests that the USH2 proteins function coordinately as a multi-protein complex in vivo. Usherin and VLGR1 are proteins with an extremely large extracellular region containing multiple repeats of a number of known cell adhesion motifs. It is believed that usherin and VLGR1 may participate in the linkage to various extracellular matrix proteins and/or cell adhesion proteins. Thus, it is essential that they be anchored at specific plasma membrane microdomains of the cells to fasten these linkages. Their interaction with whirlin appears to provide this anchorage by binding them to a submembrane protein supramolecular complex. Moreover, the localization of the protein supramolecular complex at the plasma membrane of the PMC requires binding of whirlin to usherin and VLGR1. As a result, the three proteins are interdependent for their normal subcellular localization and stability in photoreceptors (Figure 11) and hair cells. In the absence of one USH2 protein, the other two USH2 partner proteins are dispersed and destabilized, and are presumed to no longer function normally. This observation has important implications for understanding the disease mechanisms of USH2. First, all three USH2 subtypes, despite their genetic heterogeneity, affect the same subcellular target in photoreceptors and hair cells. Second, loss of one USH2 gene function is predicted to be functionally close to loss of all three. Third, the photoreceptor degeneration in USH2 disease arises from dysfunction of the PMC, a subcellular structure that is conserved from amphibian to mammalian photoreceptors.
The PRC, the analogous structure of the PMC in frog photoreceptors, is a set of nine symmetrically arrayed ridges and grooves, seen by scanning electron microscopy, at the apical inner segment membrane surrounding the connecting cilium. Originally discovered over 20 years ago [31], [32], the molecular components of the PRC had remained unknown. In the present study, we show that whirlin is a component of the PRC, the first identified marker for this complex in frogs. Although a morphologically apparent PRC structure in mammalian photoreceptors has not been seen, the similar localization pattern of whirlin in frog and mouse photoreceptors strongly suggests that a functional equivalent structure of the PRC exists in mammalian photoreceptors. Hence, we propose that the mammalian equivalent of the PRC be designated the periciliary membrane complex (PMC). Our group was the first to propose the concept of a PRC equivalent structure in mammalian photoreceptors based on the subcellular distribution of whirlin [33]. Our findings in the present study of the subcellular localization and functional interaction among whirlin, usherin and VLGR1 in mouse photoreceptors further strengthen this argument. In frogs, numerous rhodopsin-containing vesicles are present in the surrounding cytoplasm of the PRC, suggesting that the PRC may be a docking site of vesicles transporting newly synthesized rhodopsin from the Golgi [32]. Consistent with this theory, we found accumulation of vacuoles around the PMC in a small proportion of photoreceptors in whirlin knockout mice. However, in both whirlin knockout and Ush2a knockout mice [26], polarized distribution of rhodopsin to the outer segments was not measurably disrupted as shown by immunofluorescence. This observation suggests that either loss of these proteins is not sufficient to abolish completely the organization and function of the PMC, or alternative routes exist in mammalian photoreceptors for targeting rhodopsin to the outer segments. It is also possible that the USH2 protein complex is not involved directly in protein trafficking but plays only a structural role. Interestingly, we have found that the spacing between the PMC and the connecting cilium became irregular in whirlin knockout mice and there was frequent occurrence of membrane fusion between the PMC and the connecting cilium. These findings indicate that the USH2 proteins are important in maintaining the integrity of the spatial relationship between the PMC and the juxtaposing connecting cilium.
A series of in vitro analyses have found that whirlin is able to interact with calmodulin-dependent serine kinase [34], NGL-1 [24], SANS [23], myosin VIIa [24] as well as usherin [17] and VLGR1 [14]. Among these proteins, SANS and myosin VIIa are involved in human Usher syndrome type I [20], [35]. They have been reported to localize at or in the vicinity of the connecting cilium in photoreceptors [23], [36]. In inner ear hair cells, immunostaining, biochemical and cellular analyses suggest that the interaction between whirlin and myosin XVa through the PDZ domains of whirlin is required for delivery of whirlin to the tip of stereocilia [13], [24], [37]. Additionally, whirlin has been shown to interact with p55 in hair cells [38]. Therefore, some of these proteins might be candidate components of the PMC, although further studies are necessary to verify their presence in the PMC in vivo, Such studies could lead to a more comprehensive understanding of this specialized membrane domain.
In the inner ear hair cells, the interaction among whirlin, usherin and VLGR1 plays a similar role in localizing the USH2 protein complex at their normal subcellular location, i.e., the stereocilia. Here, the interactions among the three proteins may be subtly different from those in photoreceptors. The three proteins may not be completely dependent on one another for their normal localization in hair cells, as indicated by the incomplete loss of the complex from stereocilia in whirlin and Ush2a single knockouts. Usherin and VLGR1 have been demonstrated in this study and a recent study [14] to bind to the N-terminal two PDZ domains in whirlin. In the inner ear, the high abundance of whirlin short isoform, which lacks these two PDZ domains, may make the interaction of whirlin long isoform with usherin and VLGR1 partially redundant. Additionally, there may be different proteins participating in the formation and localization of the multi-protein complex containing whirlin, usherin and VLGR1 between photoreceptors and hair cells. For example, a unique exon in the cytoplasmic region of usherin in hair cells, which is missing in photoreceptors [17], may provide a platform for binding yet unidentified proteins in hair cells.
In contrast to previous studies on the localization of USH2 proteins in photoreceptors [14], [22], [23], we localized whirlin and VLGR1 only to the PMC in photoreceptors as what we have found for usherin in one of our recent publications [26]. To further confirm this finding, we rigorously exploited two approaches, double immunostaining of whirlin with different subcellular structure markers in two different species and immunoelectron microscopy. We determined the specificity of our antibodies of USH2 proteins in western blotting and immunostaining using USH2 mutant mice as valid negative controls. Additionally, ultrastructural examination of whirlin knockout mice found various defects only around the PMC but not in other regions, such as the synaptic terminus, in photoreceptors. Therefore, our study presents strong evidence that the USH2 proteins are only located at the PMC in photoreceptors.
Comparison of whirlin knockout mice generated in this study with the whirler mice demonstrates that whirlin long isoform plays an essential role in photoreceptors. In our whirlin knockout mice, whirlin long isoform including the first and second PDZ domains, which bind to usherin, have been disrupted. By immunofluorescence, usherin is lost from the PMC (Figure 5B). Since usherin is required for maintaining the long term viability of photoreceptors [26], the absence of usherin from the PMC could be responsible, at least in part, for the late-onset retinal degeneration in whirlin knockout mice. In whirler mice, a large deletion in the whirlin gene (Figure 1A) removes all predicted translational start codons of the short isoform and a portion of the proline-rich region. This mutation, therefore, is believed to completely ablates the short isoform and truncates the long isoform leaving only the N-terminal PDZ domains intact. Furthermore, an N-terminal whirlin transcript that we have found in abundance by cDNA library screening is predicted to produce a protein that retains the first PDZ domain. These N-terminal whirlin protein variants appeared to partially compensate for the loss of the intact whirlin long isoform. Indeed, in whirler mice a reduced amount of usherin is still found at the PMC in photoreceptors (Figure 5B). This residual whirlin/usherin function appears to be sufficient in maintaining photoreceptor viability, and hence no photoreceptor degeneration was found in whirler mice. Our finding that whirlin long isoform protein alone is expressed in the retina further supports the notion that the long but not the short variant of whirlin is required in photoreceptors.
The differences in hearing and vestibular dysfunction and in hair cell stereocilia defects between whirlin knockout and whirler mice suggest that whirlin long and short isoforms may function differently in hair cells. In whirlin knockout mice, only the outer hair cell stereocilia exhibit an abnormal ‘U’ shape formation, while the inner hair cell stereocilia appear normal. These whirlin knockout mice are partially deaf and have no circling behavior (no vestibular defect). But in whirler mice, besides the abnormal ‘U’ shape stereocilia formation in the outer hair cells, the inner hair cells have significantly shortened stereocilia [10], [39], [40]. These mice are completely deaf and exhibit a vestibular balance problem.
The position-dependent outcome of whirlin gene mutations observed in mice is also apparent in humans. In a German USH2 family, compound heterozygosity of a nonsense mutation p.Q103X and a mutation in the splice donor site, c.837+1G>A, which are in the 5′-terminal region of the whirlin gene, was found to cause USH2 [12]. In addition, a nonsense mutation, p.R778X, and a single nucleotide deletion, c.2324delG, leading to a deletion of the C-terminus of the whirlin protein were found responsible for deafness DFNB31 [10], [11]. Therefore, in both humans and mice, mutations at the N-terminus of the whirlin protein cause both vision and hearing impairments (our study and [12]), while mutations at the C-terminus of the whirlin protein cause more severe hearing defects only [10], [11]. These data support our conclusion that the long isoform plays an essential role in photoreceptors, while the short isoform functions primarily in hair cells. In summary, this study provides strong evidence that USH2 proteins form a multi-protein complex in which the whirlin long isoform plays a key role. This complex is localized at the PMC in photoreceptors and the stereocilia in hair cells. Disruption of this USH2 protein complex could be the common pathogenic mechanism underlying all three subtypes of human USH2 disease.
Two genomic DNA fragments (2.8 and 6 kb) flanking the 3′ portion of the first exon of whirlin were amplified from 129/Sv mouse genomic DNA by PCR, and inserted separately as the short and long arm into a modified pGT-N29 vector, which contained a diphtheria toxin expression cassette as a negative selection marker (Figure 1B). The targeting vector was linearized and electroporated into R1 embryonic stem (ES) cells. An ES clone was found to have the partial replacement of the first exon of whirlin by the Neor gene, and was microinjected into C57BL/6 blastocysts. The resulting chimeras were crossed with C57BL/6 mice. Heterozygous and homozygous knockout mice were identified with respect to the targeted allele by PCR (Figure 1C). The MEEI institutional guidelines were followed on all animal procedures.
A tiny piece of the mouse tail (about 2 mm long) was lysed by proteinase K at 50°C overnight in tissue lysis buffer (100 mM Tris-HCl pH 8.0, 200 mM NaCl, 5 mM EDTA, and 0.2% SDS). The genomic DNA was precipitated from the resulting lysate by adding the same volume of isopropanol and centrifugation. The pellet was finally dissolved in TE buffer. The total RNA was isolated using TRIzol Reagent (Invitrogen) according to the manufacturer's instruction. RT (ThermoScript™ RT-PCR system, Invitrogen) and PCR (Expand long template PCR system, Roche Diagnostics) reactions were performed following the manufacturer's instructions.
Mouse whirlin cDNA fragments (PDZ320, 1–320 aa; PDZIE, 315–580 aa, accession number, NP_082916) and frog whirlin cDNA fragment (analogous to mouse whirlin 816–907 aa) were inserted into the expression vector pET28 (Novagen). Recombinant proteins were expressed as His-tagged fusion proteins in Escherichia coli host BL21-CodonPlus (DE3)-RIPL. The recombinant proteins were purified through a Ni2+-charged nitriloacetic acid agarose column and were used to immunize rabbits and chickens. Whirlin-specific antibodies were affinity-purified from antisera or egg yolk extracts. Usherin antibodies used in this study were raised against the N-terminal and C-terminal domains of the protein [26]. RP1, RPGR and CIP98 antibodies were as described previously [25], [34], [41], [42]. VLGR1 antibody was kindly provided by Dr. Perrin C. White (University of Texas Southwestern Medical Center, Dallas, Texas). Mass1 (C20) antibody was purchased from Santa Cruz Biotechnology, Inc. Monoclonal anti-γ-tubulin and anti-acetylated α-tubulin antibodies were obtained from Sigma-Aldrich. Alexa fluorochrome-conjugated phalloidin and secondary antibodies, and Hoechst dye 33342 were obtained from Molecular Probes, Inc.
Mouse whirlin and its fragments (full-length, 3–907 aa; N-terminus, 3–472 aa; C-terminus, 438–907 aa, accession number, NP_082916) and mouse usherin fragment (5053–5193 aa, accession number, NP_067383) were amplified from the retina and individually cloned into both pGBKT7 and pGADT7 vectors. Yeast two-hybrid analysis was performed as described previously [43]. Briefly, a protein/peptide in pGBKT7 vector was co-transformed with its putative interacting protein/peptide in pGADT7 vector. Empty pGBKT7 and pGADT7 vectors were used as negative controls. Co-transformants were grown on both SD-4 (-Leu, -Trp, -Ade, and -His) and SD-2 (-Leu and -Trp) plates. The growth on SD-4 plates indicated an existence of interaction between the two co-transformed proteins/peptides. In our experiments, all co-transformants were able to grow on SD-2 plates indicating a successful co-transformation.
GST pull-down assay: cDNA fragments of intact (mouse: 5053–5193 aa, NP_067383; frog: analogous to mouse usherin 5053–5193 aa) and mutant (without PDZ-binding domain, mouse: 5053–5186 aa, NP_067383; frog: analogous to mouse usherin 5053–5189 aa) C-terminal usherin were amplified from frog and mouse retinas and cloned into the pGEX4T-1 vector. The GST-fused intact and mutant usherin were expressed in BL21-CodonPlus (DE3)-RIPL cells and then incubated with mouse retinal lysate and glutathione sepharose beads for 2 hours at 4°C. Subsequently, the sepharose beads were washed with lysis buffer (50 mM Tris-HCl pH8.0, 150 mM NaCl, 0.5% TritonX-100, 5 mM EDTA, 0.5 mM PMSF, 1× protease inhibitor, and 1 mM DTT) for three times and boiled in Laemmli sample buffer for 10 minutes. All the procedures were performed at 4°C. Retinal lysates incubated with glutathione sepharose beads and GST or only with GST were used as negative controls.
Immunoprecipitation: Dissected retinal or inner ear tissues were homogenized and incubated for about 60 minutes in lysis buffer. After centrifugation at 18,000 g for 10 minutes, the supernatants were precleared by incubation with protein G sepharose (Amersham Biosciences) for 1 hour. Subsequently, they were incubated with the primary antibodies for 3 hours and then centrifuged at 18,000 g for 10 minutes. The resulting supernatants were incubated with protein G sepharose for an additional 1 hour. After a brief centrifugation at 2000 g, the pellets were washed with lysis buffer for four times and then boiled in Laemmli sample buffer. All the procedures were performed at 4°C. A non-immune rabbit IgG served as a negative control. Western blotting was carried out as described previously [43].
Immunofluorescence: Eyes were enucleated, frozen immediately and sectioned at 10-µm thick. Sectioned tissues were fixed in 4% formaldehyde/PBS for 10 minutes (for usherin staining, 2% formaldehyde/PBS for 5 minutes), and permeabilized by 0.2% Triton X-100/PBS for 5 minutes at room temperature. Pup heads on postnatal day 3–6 were fixed in 4% formaldehyde/PBS for about 36 hours, switched to 30% sucrose/PBS for several days, and sectioned at 30-µm thick. The subsequent steps of blocking and incubation with primary and secondary antibodies were as described previously [43]. Alexa 488- and 594-conjugated secondary antibodies were routinely used for tissue double-labeling. Stained sections were viewed and photographed on a fluorescent microscope (Olympus, model 1X70) equipped with a digital camera (Carl Zeiss MicroImaging, Inc.) or on a confocal laser scanning microscope (Leica, model TCS SP2).
Immunoelectron microscopy: Eyes were enucleated. Their anterior segments and lens were removed. Dissected retina was fixed with 4% formaldehyde/PBS (whirlin) or 2% formaldehyde/0.1% glutaraldehyde/PBS (VLGR1) for 30 minutes, washed with TTBS buffer (Tween/Tris-buffered saline), blocked in 5% goat serum/TTBS for 1 hour, and incubated with the primary antibodies at 4°C overnight. After rinses, the retina was incubated with Nanogold goat anti-rabbit antibody (Aurion, Wageningen, The Netherlands), post-fixed sequentially with 1% formaldehyde/2.5% glutaraldehyde/0.1 M cacodylate buffer and 2% osmium tetroxide. Later, it was silver-enhanced, dehydrated, embedded in Epon, and sectioned at 70 nm thickness. In an alternative protocol, the retina was fixed in 2% formaldehyde/0.1% glutaraldehyde/PBS for 30 minutes, frozen and cut to 10 µm sections prior to staining with primary and secondary antibodies. Staining was done while floating in a dish. After final wash, the sections were post-fixed and processed for immunoEM as above. ImmunoEM for whirlin were studied with both methods which yielded the same results. ImmunoEM for VLGR1 used the alternative protocol.
Transmission electron microscopy and scanning electron microscopy was performed as described previously [26], [43].
Measurements of photoreceptor outer segment length and outer nuclear layer thickness were made along the vertical meridian (superior to inferior) at five locations to each side of the optic nerve head separated by approximately 600 µm each. Measurements began at approximately 200 µm from the optic nerve head and ended at approximately 200 µm from the retinal periphery. For the analysis in the whirlin knockout mice, seven whirlin knockouts and five whirlin heterozygous littermates from 28–33 months of age were included. For the analysis in the whirler mice, three whirler mice aged from 28–33 months and two age-matched wild-type mice were included.
Photoreceptors with abnormal morphology around the PMC were counted at the retinal mid-periphery. Abnormal morphology was defined as membrane fusion between the apical inner segment and the distal connecting cilium or vacuole accumulation in the apical inner segment around the PMC. The presence of at least 3 large vacuoles (diameter is larger than 200 nm) or 4 small vacuoles (diameter is about 100 nm) was considered as vacuole accumulation. Four wild-type and six whirlin knockout mice at the age from 5 to 24 months were included in this experiment.
The Student's t-test was conducted to compare the measured values of whirlin knockout and control mice. A P value of less than 0.05 was considered to indicate a significant difference between the two groups.
ERG and DPOAE recordings were performed as described previously [26], [44].
|
10.1371/journal.pntd.0006499 | The spleen microbiota of small wild mammals reveals distinct patterns with tick-borne bacteria | Wild mammals serve as reservoirs for a variety of microbes and play an important role in the enzootic cycles of these microbes. Some of them are vector-borne bacteria in the genera Anaplasma, Ehrlichia and Rickettsia of the order Rickettsiales, which can cause febrile illnesses in human beings as well as animals. Anaplasma spp., Ehrlichia spp. and many spotted fever group (SFG) Rickettsia spp. are transmitted to mammalian hosts by tick vectors during blood meals. As a powerful sequencing method, the next generation sequencing can reveal the complexity of bacterial communities in humans and animals. Compared with limited studies on blood microbiota, however, much fewer studies have been carried out on spleen microbiota, which is very scarce in wild mammals. Chongming Island is the third biggest island in China. It was unclear whether there were any vector-borne bacteria in Chongming Island. In the present study, we explored the bacterial microbiota in the spleens of wild mice and shrews from the rural areas of Chongming Island and investigated the prevalence of vector-borne bacteria.
Genomic DNAs were extracted from the spleen samples of 35 mice and shrews. The 16S rDNA V3-V4 regions of the DNA extracts were amplified by PCR and subjected to the 16S rDNA-targeted metagenomic sequencing on an Illumina MiSeq platform. All the 35 spleen samples obtained data with sufficient coverage (99.7–99.9%) for analysis. More than 1,300,000 sequences were obtained after quality control and classified into a total of 1,967 operational taxonomic units (OTUs) clustered at 97% similarity. The two most abundant bacterial phyla were Firmicutes and Proteobacteria according to the analysis of rarefied sequences. Among the bacterial communities detected in this study, Anaplasma, Rickettsia and Coxiella were adjacently clustered by hierarchical analysis. Significant differences in many bacterial features between Anaplasma-positive and Anaplasma-negative samples were identified by LEfSe analysis and Wilcoxon rank-sum test, suggesting that the Anaplasma-infection of small wild mammals was associated with a specific pattern of spleen microbiota.
Our study has comprehensively characterized the complex bacterial profiles in the spleens of wild mice and shrews from Chongming Island, Shanghai city. This work has revealed distinct spleen bacterial communities associated with tick-borne bacteria in wild animals. The detection of tick-borne bacteria highlights the risk of contracting pathogens with public health importance upon tick-exposure in the studied areas.
| In this study, the 16S rDNA-targeted metagenomic sequencing was used to determine the bacterial community and diversity in the spleens of small wild mammals from China. The 16S rDNAs were amplified from the spleen genomic DNAs of 35 small wild mice and shrews and sequenced by Illumina MiSeq technology. More than 1,300,000 sequences were obtained after quality control and classified into a total of 1,967 operational taxonomic units (OTUs) clustered at 97% similarity. The two most abundant bacterial phyla were Firmicutes and Proteobacteria according to the analysis of rarefied sequences. Within the bacterial communities detected in this study, vector-borne bacteria, Anaplasma, Rickettsia and Coxiella, were adjacently clustered by hierarchical analysis. Significant differences in many bacterial features between Anaplasma-positive and Anaplasma-negative samples were observed, suggesting that the infection of small wild mammals with Anaplasma is associated with a distinct pattern of spleen microbiota. This study has revealed the complex bacterial profiles in the spleens of wild mice and shrews. The detection of vector-borne bacteria highlights the role of wild mice and shrews as animal reservoirs with potential public health importance in the studied areas.
| Wild mammals serve as reservoirs for a variety of microbes and play an important role in the enzootic cycles of these microbes. Some of them are vector-borne bacteria in the genera Anaplasma, Ehrlichia and Rickettsia of the order Rickettsiales. Anaplasma spp., Ehrlichia spp. and many spotted fever group (SFG) Rickettsia spp. are transmitted to mammalian hosts by tick vectors. They are obligate intracellular bacteria, and their main target cells are white blood cells, erythrocytes, platelets and/or vascular endothelia [1–3]. These bacteria have evolved adapted strategies to evade and/or suppress host protective immune responses and can cause febrile illnesses in animals and/or humans [1–3]. They have been gradually recognized as emerging pathogens of public health importance around the world [3–7]. The prevalence of these tick-borne bacteria has been increasingly reported in China. For instance, Anplasma phagocytophilum, Anaplasma bovis, Anaplasma ovis, Anaplasma central, Anaplasma marginale, Anaplasma platys, Anaplasma capra, Ehrlichia chaffeensis, Ehrlichia canis, Candidatus Neoehrlichia mikurensis, Rickettsia heilongjiangiensis, Rickettsia sibirica, Rickettsia raoultii and Rickettsia conorii have been detected in ticks, animals or humans in many provinces of China [4,8–19]. However, the existence of these bacteria in Shanghai city, China is still unknown.
Mammals are ecosystems that are inhabited by niche-specific microbiota including bacteria, viruses and fungi etc. The commensal microbiota plays essential roles in the development of immune system, modulation of metabolism and maintenance of health [20]. The perturbation of symbiotic microbiota has been shown to be associated with various diseases such as infection, immunological disorders, metabolic diseases and cancer etc. [20–22]. It had been thought that the circulatory system was sterile in healthy organisms, and that bacteria were present in the circulation only due to sepsis. Nevertheless, the presence of bacteria in the blood of healthy humans began to be documented several decades ago [23,24]. With the advance in sequencing technology, blood microbiota has been gradually uncovered in healthy organisms in the past decade [25,26].
The spleen, a peripheral lymphoid organ in vertebrates, acts as a blood filter. It plays an important role in the modulation of immune responses and hematopoiesis [27]. The spleen can be infected by the tick-borne bacteria from the order Rickettsiales [28–30]. During the establishment of intracellular infection in the spleen, these bacteria may have impacts on their host cells and alter the spleen niche. Therefore, we hypothesize that the changed spleen niche due to the infection with tick-borne bacteria would lead to the formation of specific spleen microbiota. As important reservoirs for the tick-borne bacteria, mice also serve as model animals for human infection. The present study explores the spleen microbiota in wild mice and shrews from Chongming Island, Shanghai city, China. The blood microbiota of wild mice from Israel and the spleen microbiota of wild voles from France have been recently reported, respectively, [31,32]. Compared with limited studies on blood microbiota, however, much fewer studies have been carried out on spleen microbiota, which is very scarce in wild mammals.
Chongming Island is the third biggest island in China. Its major part belongs to an administrative county of Shanghai city. It locates at the mouth of the Yangtze River. There has been no description on tick species present in Chongming Island yet. Rhipicephalus sanguineus and Haemaphysalis longicornis, however, have been reported to be the ticks infesting on pet dogs in other areas of Shanghai [33]. It is unclear whether there are any tick-borne bacteria in Chongming Island. The aim of the present study is to explore the bacterial microbiota in the spleens of wild animals from the rural areas of Chongming Island and investigate the presence of tick-borne bacteria.
Animals were handled in accordance with National Guidelines for Ethic Review of Laboratory Animal Welfare. Animal treatment protocols were approved by the institutional animal ethics committee (the Animal Ethics Committee of Tongji University School of Medicine, Shanghai, China).
The Chongming Island has a humid subtropical monsoonal climate and a woodland habitat, which is suitable for tick infestation. Chongming Island has an area of around 1267 km2. The mouse collection sites mainly covered the middle and west parts of Chongming Island where there was more green coverage. Spring-loaded bar mousetraps with bait were used to trap mice. Traps were strategically placed in the environment such as crop fields, residential houses, bank of rivers and forests where wild mice were seen or expected living or traveling. These traps were set in the evening and checked in the next morning. The latitude and longitude of locations where mice and shrews were trapped were recorded through the Global Positioning System (GPS).
Trapped live animals were transported in cardboard containers to the institutional Animal Biosafety Laboratory where they were euthanized by CO2. Necropsies were conducted after euthanasia. The spleen samples of trapped animals were collected and stored at –80°C.
Genomic DNAs were extracted from the spleen samples using the E.Z.N.A. tissue DNA extraction kit (Omega Bio-tek, Norcross, GA, US) according to the manufacturer’s protocol. The quality and quantity of extracted DNAs were examined by 1% agarose gel electrophoresis and NanoDrop 2000 spectrophotometer (Thermo Scientific, MA, US). The V3-V4 regions of the 16S rDNA were amplified by PCR in a thermal cycler GeneAmp 9700 (Applied Biosystems Inc, Foster City, CA, US). The PCR condition was 95°C for 3 min, followed by 30 cycles at 95°C for 30 s, 55°C for 30 s, and 72°C for 45 s and a final extension at 72°C for 10 min. The primers used were 338F (5’-barcode-ACTCCTACGGGAGGCAGCAG-3’) and 806R (5’-GGACTACHVGGGTWTCTAAT-3’). The barcode is an eight-base sequence unique to each sample. The PCR reactions were performed in triplicate in 20 μL mixture containing 2 μL of 10 × PCR Buffer, 2 μL of 2.5 mM dNTPs, 0.8 μL of each primer (5 μM), 0.2 μL of rTaq DNA Polymerase (TaKaRa Bio, Dalian, China), and 10 ng of template DNA.
The PCR amplicons were extracted from 2% agarose gels and purified using the AxyPrep DNA gel extraction kit (Axygen Biosciences, Union City, CA, US) according to the manufacturer’s instructions. After being quantified using QuantiFluor-ST (Promega, Madison, WI, US), the purified DNAs were pooled in equimolar and paired-end sequenced (2 × 300) on an Illumina MiSeq platform according to the standard protocols (Majorbio, Shanghai, China). The raw reads were deposited into the NCBI Sequence Read Archive (SRA) database (Accession Number: SRP118742).
Raw fastq files were demultiplexed and quality-filtered using QIIME (version 1.9.1). The following criteria were met: (i) The 300 bp reads were truncated at any site with an average quality score < 20 over a 50 bp sliding window, discarding the truncated reads shorter than 50 bps. (ii) Exact barcode matching; maximal 2 nucleotide mismatches in primer matching. (iii) Only sequences overlapping longer than 10 bps were assembled according to their overlapped sequence. Reads containing ambiguous characters were removed. Reads that could not be assembled were discarded.
Operational taxonomic units (OTUs) were clustered with 97% similarity cutoff using UPARSE (version 7.1 http://drive5.com/uparse/) and chimeric sequences were identified and removed using UCHIME. The taxonomy of each 16S rRNA gene sequence was analyzed by RDP Classifier (http://rdp.cme.msu.edu/) against the SILVA (SSU123) 16S rRNA database using confidence threshold of 70% as previously described [34].
The small wild mammals trapped in the present study included mice and shrews. Thirty five samples submitted for 16S rDNA-targeted metagenomic sequencing were listed in Table 1. Except the four samples CS6, CS7, CS34 and CS88 from shrews, all the other 31 samples were from mice including Apodemus agrarius, Mus musculus and Rattus flavipectus. Shrews belong to the order Eulipotyphla.
A. phago: A. phagocytophilum. Abbreviated environmental types: A, agricultural area; F, forest; LB, lake bank; R, residential area; RB, river bank near residential area. The latitudes and longitudes of locations where animals were trapped were provided.
In this study, all the 35 spleen samples obtained data with sufficient coverage (99.7–99.9%) for analysis. A total of 1,323,308 16S rRNA gene sequences with a read length of 469 bps were identified with an average of 37,808 reads per sample. And a total of 1,967 OTUs were clustered at 97% similarity across all samples. The number of OTUs per sample ranged from 127–528. The inverse Simpson’s diversity indices were from 0.072 to 0.6551, which indicated a broad variation in the bacterial diversity between samples. Rarefaction (to 17,808) resulted in 85–520 OTUs per sample.
Firmicutes was the most abundant phylum, and Proteobacteria was the second among the total taxa of 35 samples tested in this study except that the scenario for sample CS97 was inverse (Fig 1). Firmicutes and Proteobacteria had mean abundances of 71.53% (SD 13.48%) and 22.45% (SD 11.85%), respectively, in the total taxa of 35 samples (Fig 1). The sum of their mean abundances accounted for greater than 90% of the total taxa. The next three following bacterial phyla were Bacteroidetes, Actinobacteria and Choloriflexi with a mean abundance of 2.33% (SD 1.88%), 1.91% (SD 1.94%) and 0.47% (SD 0.73%), respectively (Fig 1). At genus level, there were 11 major bacterial taxa with a mean abundance greater than 1%, which included Bacillus, Lactococcus, Peptoclostridium, Pseudomonas, Oceanobacillus, Clostridium_sensu_stricto_1, Acinetobacter, Psychrobacter, Brochothrix, Bartonella and Anaplasma (Fig 2). In some samples e.g., CS39, CS55, CS56, CS57 and CS63, the relative abundance of Peptoclostridium was exceptionally high, whereas the relative abundances of Bacillus and Lactococcus were quite low.
The prevalence of vector-borne bacteria in the tested samples was summarized in Table 1. Anaplasma (Anaplasma ovis and/or Anaplasma phagocytophilum), Ehrlichia, Rickettsia, Coxiella and Bartonella were detected in 11, 7, 21, 11 and 16 of 35 (31.43%, 20%, 60%, 31.43% and 45.7%) samples, respectively. Anaplasma detected in the present study included two species, A. ovis and A. phagocytophilum. The relative abundance of A. ovis was much higher than that of A. phagocytophilum (S1 Table). Ehrlichia detected in the present study was much less abundant than Anaplasma, Rickettsia, Coxiella or Bartonella (S1 Table). Coxiella detected in the present study consisted of only one member, Coxiella_endosymbiont_of_Rhipicephalus_turanicus (S1 Table). All Anaplasma-positive samples were co-infected with Coxiella and vice versa. All Ehrlichia-positive samples were co-infected with Anaplasma and Coxiella, and all Anaplasma/Coxiella-positive samples were co-infected with Rickettsia but not Bartonella. Anaplasma (A. ovis and A. phagocytophilum), Ehrlichia, Rickettsia, Coxiella and Bartonella were detected in both the male and female animals. And they were all detected in the mouse samples. Except A. phagocytophilum and Ehrlichia, A. ovis, Rickettsia, Coxiella and Bartonella were detected in the shrew samples.
The 35 samples were hierarchically clustered against the genera with top 50 relative abundances including Anaplasma, Rickettsia, Coxiella and Bartonella but not Ehrlichia (Fig 3). Notably, Anaplasma, Rickettsia and Coxiella were adjacently clustered to each other, whereas Bartonella were not adjacent to these three genera. This further reflected the closely related occurrences of Anaplasma, Rickettsia and Coxiella but not Bartonella in the wild mice and shrews.
Principal coordinate analyses (PCoA) based on the Bray-Curtis metrics (S2 Table) were performed to look at the overall differences in the spleen microbiota of the 35 samples considering the factors of animal genders, types, locations or infection with Anaplasma. Anaplasma had a relatively high mean abundance among the tick-borne bacteria detected in the present study. As shown in Fig 4A, two relatively dense groups and one relatively scattered group were observed and circled. One of the two relatively dense groups consisted of 22 samples including 10 male and 12 female animals (Fig 4A). Three of the 22 samples were shrews (Fig 4B). And the geographic sites of these 22 samples covered all the 8 sites in the present study (Fig 4C). All of the 22 samples were Anaplasma-negative (Fig 4D). The other relatively dense group consisted of 8 samples including 7 male and 1 female animals (Fig 4A). One of these 8 samples was shrew (Fig 4B). The geographic sites of the 8 samples were from the 3 sites, JZ, BH and DP (Fig 4C). And all of the 8 samples were Anaplasma-positive (Fig 4D). The 5 samples from the relatively scattered group were all mice and from the 2 sites, BH and HX (Fig 4B and 4C). Three of these 5 samples, CS39, CS55 and CS57, were close to each other, and they were all Anaplasma-positive (Fig 4D). One of these three samples was male, and 2 were female (Fig 4A). The remaining 2 samples from the relatively scattered group were more scattered, which were female animals and Anaplasma-negative (Fig 4A and 4D). Compared with the samples in the other relatively dense groups, these 5 samples in the relatively scattered group had exceptionally high percentages of Peptoclostridium and low percentages of Bacillus and Lactococcus (Fig 2), which contributed greatly to their straying away from the other two groups in the PCoA plots (Fig 4).
To further analyze the microbiota considering the factor of infection with Anaplasma, a hierarchical clustering using unweighted pair group method with arithmetic mean (UPGMA) was conducted to compare the microbiota similarities between Anaplasma-positive and Anaplasma-negative samples. The overall 35 samples were divided into two major clusters as shown in Fig 5. Eight of the 11 Anaplasma-positive samples were in the bigger cluster, and they were sub-clustered into an independent group. The other 3 of the 11 Anaplasma-positive samples were in the smaller cluster, and they were clustered in a consecutively order. Among the spleen microbiota of tested samples, the overall similarities indicated by hierarchical clustering (Fig 5) was consistent with the diversities revealed by the PCoA plot (Fig 4D).
A number of differentially abundant bacterial taxa between Anaplasma-positive and Anaplasma-negative samples were identified in the spleen microbiota by the linear discriminant analysis effect size (LEfSe) analysis as shown in Fig 6. The differentially enriched taxa in Anaplasma-positive samples were mainly from the phyla of Proteobacteria and Fusobacteria, whereas the differentially enriched taxa in Anaplasma-negative samples were mainly from the phyla of Actinobacteria, Acidobacteria, Choloroflexi, Nitrospirae and Proteobacteria. Although there was no significant difference in the abundance of overall Proteobacteria phylum, there were significant differences in the class α-Proteobacteria and in some genera from the classes of β and γ- Proteobacteria between Anaplasma-positive and Anaplasma-negative samples.
A comparison of the spleen microbiota between Anaplasma-positive and Anaplasma-negative samples to genus level revealed a list of differentially abundant bacterial features with absolute linear discriminant analysis (LDA) scores > 2 (Fig 7), suggesting that the Anaplasma-infection was associated with specific patterns of spleen microbiota from mice and shrews. The features with top 25 absolute LDA scores in Anaplasma-positive samples were Peptostreptococcaceae, Alphaproteobacteria, Rickettsiales, Clostridiaceae_1, Clostridium_sensu_stricto_1, Acinetobacter, Anaplasmataceae, Anaplasma, Stenotrophomonas, Dyadobacter, Xanthomonadaceae, Xanthomonadales, Rickettsia, Rickettsiae, Helcococcus, Escherichia_Shigella, Enterococcus, Enterococcaceae, Coxiella, Coxiellaceae, Leginellales, Ehrlichia, Roseateles, Moraxella and Arcobacter (Fig 7). The tick-borne bacteria, Anaplasma, Rickettsia, Coxiella and Ehrlichia, were all recognized within the top 25 differentially represented features of Anaplasma-positive samples. In contrast, the features with top 25 absolute LDA scores in Anaplasma-negative samples were Bacilli, Bacillales, Bacillaceae, Bacillus, Pseudomonodaceae, Pseudomonoas, Oceanobacillus, Actinobacteria, Actinobacteria, Psychrobacter, Micrococcales, Micrococcaceae, Arthrobacter, Brochothrix, Listericeae, Planococcaceae, Paenirhodobacter, Ruminococcaceae, Lysinibacillus, Aerococcaceae, Myroides, Chloroflexi, Bacteroidales_S24_7_group, Bacteroidales_S24_7_group_g_norank and Flavobacteriia. Bartonella were neither included in the differentially abundant bacterial taxa of Anaplasma-positive samples nor in those of Anaplasma-negative samples (Fig 7).
In addition, a few differentially abundant taxa in the spleen microbiota between male and female animals, mice and shrews or animals from multiple geographic sites at the ranks below phylum were observed by LEfSe analysis, respectively (S1 Fig). Nevertheless, no vector-borne bacteria, Anaplasma, Ehrlichia, Rickettsia, Coxiella or Bartonella, were identified among these differentially enriched taxa.
Wilcoxon rank-sum test was used to further compare the relative abundances of taxa between Anaplasma-positive and Anaplasma-negative samples at phylum level. Consistent with LEfSe analysis (Fig 7), Actinobacteria, Chloroflexi, Acidobacteria and Nitrospirae were significantly more abundant in Anaplasma-negative samples than in Anaplasma-positive samples, whereas Fusobacteria was significantly more abundant in Anaplasma-positive samples than in Anaplasma-negative samples based on the analysis of Wilcoxon rank-sum test (Fig 8). At genus level, 44 significantly different genera were identified between Anaplasma-positive and Anaplasma-negative samples (S2 Fig) by the analysis of Wilcoxon rank-sum test, which was consistent with the result from LEfSe analysis too (Fig 7). Anaplasma, Rickettsia, Coxiella and Ehrlichia were all significantly more abundant in Anaplasma-positive samples than in Anaplasma-negative samples by the analysis of Wilcoxon rank-sum test (S2 Fig).
The present study has comprehensively characterized the spleen microbiota in wild mice and shrews from Chongming Island, Shanghai City, which has advanced our understanding on spleen microbiota in wild animals. To our knowledge, this is the first time that bacterial profiles in the spleens of wild animals have been explored in China. It has provided a wealth for the comparative analyses of spleen microbiota across different types of mammalian hosts from different geographic areas. The present study is the first report on unique bacterial taxa associated with tick-borne bacteria in wild mammals. This study has further proved the application of 16S rDNA metagenomics as a powerful methodology to study the prevalence of bacteria in the circulatory system of wild life as suggested by Razzauti et al. [32].
In the present study, the two major bacterial phyla among the taxa of tested mouse and shrew samples were Firmicutes and Proteobacteria. Previous reports have shown that microbiota of blood and the spleen from humans or mice were mainly consisted of Proteobacteria and sometimes Firmicutes as well. For instance, the major prevalent bacterial genus in both gerbil rodent blood samples from Israel and mouse spleen samples from France was Bartonella, belonging to the phylum Proteobacteria [31,32]. Firmicutes and Proteobacteria were the two major phyla with comparable relative abundances in the blood microbiota of healthy human samples [26]. The predominant phylum, Proteobacteria, represented 90% of the overall microbiota in human blood samples [35]. Greater than 80% of the blood microbiota in 30 healthy blood donors was from the phylum Proteobacteria followed by the phyla of Actinobacteria, Firmicutes and Bacteroidetes [25]. The blood microbiota in nonalcoholic fatty liver disease (NAFLD) patients mainly consisted of Proteobacteria (87.9%), which was followed by Actinobacteria (7.3%), Firmicutes (3.7%) and Bacteroidetes (1.1%) [36]. Although there was variation in the proportions of Proteobacteria and Firmicutes between the spleen microbiota of wild mice and shrews in the present study and the blood and spleen microbiota of humans or mice from the aforementioned reports, the overall blood or spleen microbiota in humans or mice is different from the gut microbiota, which is dominated by Firmicutes and Bacteroidetes [37].
Intriguingly, the present study has shown that the infection of wild mice and shrews with Anaplasma has been associated with a specific spleen microbiota. A number of significantly differentially abundant bacterial taxa between Anaplasma-positive and Anaplasma-negative samples were revealed by both LEfSe analysis and Wilcoxon rank-sum test, respectively. As shown in Fig 5, the 35 tested samples fell into two major clusters based on the analysis using unweighted pair group method with UPGMA. Eight Anaplasma-positive samples were independently sub-clustered within the bigger major cluster, and 3 Anaplasma-positive samples were adjacently clustered within the smaller major cluster. It seemed that the formation of these two major clusters were resulted from some unknown factors, which were different from the factors analyzed in the present study i.e., animal genders, types, geographic sites or infection with Anaplasma (Fig 4). Additionally, it was interesting to notice that there were a few differentially enriched taxa identified in the spleen microbiota from animals with different genders, types or geographic locations at the rank levels below phylum by LEfSe analysis in the present study, suggesting that the factors of animal genders, types or geographic locations had impacts on the spleen microbiota.
The present study is the first report on the detection of vector-borne bacteria, Anaplasma, Ehrlichia, Rickettsia, Coxiella and Bartonella, in Chongming Island, which suggests that the wild mice and shrews serve as important animal reservoirs for these vector-borne bacteria in the studied areas. Anaplasma, Ehrlichia, Rickettsia, Coxiella and Bartonella were detected from multiple foci in Chongming Island in the present study, which reflected their relative wide distributions in this island. Among these bacteria, only Ehrlichia was not detected in shrews, which was probably due to the small quantities of shrew samples in the present study. Eight of 11 Anaplasma were male animals, which could be resulted from differences in the ecological behaviors between male and female wild mice and shrews tested in this study. This was less likely due to any potentially intrinsic differences in the spleen niche between male and female animals since Anaplasma, Ehrlichia, Rickettsia, Coxiella or Bartonella were not among the differentially enriched taxa identified in the spleen microbiota of animals with different genders. Furthermore, these bacteria were not among the differentially enriched taxa in the spleen microbiota of mouse versus shrew groups or different geographic site groups either, suggesting that infection with these bacteria was neither specific to mice or shrew nor to a single geographic site in Chongming Island.
The closely related occurrences of Anaplasma, Ehrlichia, Rickettsia and Coxiella in the present study suggested that Anaplasma, Ehrlichia, Rickettsia and Coxiella shared the transmission routes in the studied areas. Both of Ehrlichia and Anaplasma belong to the family Anaplasmaceae and are tick-borne bacteria. Many members in the SFG Rickettsia are transmitted by ticks. Coxiella can be transmitted by ticks too. It was very likely that Anaplasma, Ehrlichia, Rickettsia and Coxiella in the co-infected animals in the present study were transmitted by ticks instead of other vectors. Therefore, there was a high chance to get infected with multiple of them upon a tick exposure. A. ovis was the major prevalent Anaplasma sp. in the present study. A. ovis infects ruminants and causes ovine anaplasmosis. A. phagocytophilum, a zoonotic pathogen, can cause anaplasmosis in both humans and animals. Infection with Ehrlichia causes febrile diseases in mammalian hosts. Coxiella and Rickettsia were usually considered as vector-borne pathogens. However, with the advance of molecular biology, some members of Coxiella and Rickettsia are gradually recognized as non-pathogenic intracellular bacteria, which are actually endosymbionts to their hosts [38].
Coxiella_endosymbiont_of_Rhipicephalus_turanicus, also called Coxiella-like endosymbiont (Coxiella-LE), was the only member of Coxiella detected in this study. Coxiella-LE distributes in ticks worldwide [38]. Coxiella and Rickettsia were among the ten maternally inherited bacteria found in ticks summarized by Bonnet, i.e., Coxiella-LE, Rickettsiella, Arsenophonus, Francisella-LE, Cardinium, Spiroplasma, Lariskella, Midichloria, Rickettsia and Wolbachia [38]. Besides Coxiella and Rickettsia, Rickettsiella were detected in our study, too. Rickettsiella was transferred from the order Rickettsiales to the family Coxiellaceae in the order Legionellales based on the phylogenetic analysis of 16S rRNA sequences [39]. Nevertheless, Rickettsiella were only detected in sample CS57 and had a much less relative abundance in the present study. There may be new tick borne-bacteria present in the differentially abundant bacterial taxa of Anaplasma-positive samples revealed in the present study.
In the present study, the infection rate of Rickettsia was 60%, which was the highest among the vector-borne bacteria detected. The Rickettsia-positive samples covered all Anaplasma, Ehrlichia or Coxiella-positive samples but not all Bartonella-positive samples. Rickettsia was prevalent in mice and shrews from all types of environment investigated in the present study, i.e., forests, agricultural fields, residential areas and banks. In contrast, Anaplasma, Ehrlichia or Coxiella were detected in mice and shrews from the forests and agricultural fields rather than residential areas or banks. Compared with residential areas and banks, the forests and agricultural fields in the studied areas had more green coverage and were more suitable for tick survival. It was unclear whether the Rickettsia spp. from the animals co-infected with Anaplasma, Ehrlichia or Coxiella in the forests and agricultural fields were same as those prevalent in residential areas and banks in the present study. Furthermore, it was unclear whether the vectors transmitting Anaplasma, Ehrlichia, Coxiella or Rickettsia in the forests and agricultural fields were same as those transmitting Rickettsia in residential areas and banks in this study either.
Bartonella in the present study were probably transmitted by vectors other than ticks. Bartonella can be transmitted by several arthropod vectors such as fleas, keds, lice, sand flies and ticks, or direct bites by infected animals and often establish persistent infection in asymptomatic mammalian hosts [40]. Bartonella were the most frequently identified bacteria in the fleas collected from southern Indiana, USA [41]. Bartonella together with Mycoplasma were the dominant flea-borne bacteria detected in gerbil rodent blood samples from Israel [31]. Haemotrophic Mycoplasma has different subgroups and been detected in the spleen or blood samples of rodents [42]. In the present study, however, the relative abundance of Mycoplasma was less than 1%. Both Bartonella and Mycoplasma were among the 23 features detected in the rodent blood and/or flea samples summarized by Cohen et. al. [31]. Besides Bartonella, Mycoplasma and Rickettsia, other 14 of these 23 features i.e., Aquabacterium, Bifidobacterium, Bradyrhizobium, Cyanobacterium (phylum), Halomonas, Lactobacillus, Massilia, Methylobacterium, Neisseria, Ralstonia, Rhizobiales_unclassified (order), Staphylococcus, Streptococcus and Sphingobacteria (class), were detected in the present study too. The remaining 6 of the 23 features, Azovibrio, Catenuloplanes, Diaphorobacter, Saccharothrix, Spirosoma and Wolbachia, were not detected in the present study.
Bartonella were also the most prevalent genus in vector-borne bacteria detected in the spleen microbiota of wild voles from France [32]. There were 45 potential zoonotic bacterial genera in total detected by Razzauti M et al [32]. Eleven of the 45 genera, Anaplasma, Bacillus, Bartonella, Clostridium, Coxiella, Escherichia/ Shigella, Moraxella, Rickettsia, Staphylococcus, Stenotrophomonas and Streptococcus, were among the relatively abundant bacterial genera listed in Fig 2 in the present study. Twenty one of the 45 genera, Aeromonas, Burkholderia, Campylobacter, Corynebacterium, Ehrlichia, Enterococcus, Eubacterium, Granulicatella, Haemophilus, Helicobacter, Leptospira, Mannheimia, Micrococcus, Mycobacterium, Mycoplasma, Neisseria, Neochlamydia, Pasteurella, Rhodococcus, Treponema and Vibrio, were detected with relatively low abundance in the present study (S1 Table) and hence not listed in Fig 2. The left 13 genera, Bordetella, Borrelia, Brucella, Francisella, Klebsiella, Legionella, Listeria, Orientia, Salmonella, Ureaplasma and Yersinia were not detected in the present study. The prevalence of Rhodococcus, Legionella, Staphylococcus, Corynebacterium, Streptococcus and Stenotrophomonas, reported to contaminate laboratory reagents [43], were high in the spleen microbiota of wild voles from France, and the authors suspected that the presence of bacteria in the samples were due to contamination instead of real infection of the animals [32]. In the present study, however, Rhodococcus and Corynebacterium were the genera with relatively low abundance, and Legionella was not detected. And Staphylococcus, Stenotrophomonas and Streptococcus were the 37th, 14th and 20th abundant genera in the present study, respectively. Blank controls were set throughout the 16S metagenomics sequencing in our study, and the detection of these bacteria was very likely due to real infection of wild animals rather than contamination of samples. However, as emphasized by Razzauti et al. [32], caution should be taken when DNA-based techniques are used to detect microbes.
In future, it would be important to investigate the molecular characteristics of vector-borne bacteria prevalent in the studied areas. At the same time, it would be also important to characterize the vectors. These will contribute to the prevention and control of vector-borne bacterial infection in the studied areas. Furthermore, it would be interesting to study the interaction between tick-borne bacteria and their host cells in the spleen. Studies on the mechanism underlying the alteration of the spleen microbiota due to infection with tick-borne bacteria would not only advance the knowledge of the pathogenesis of tick-borne bacteria but also shed light on the function of spleen microbiota from the perspective of infection.
|
10.1371/journal.pgen.1002714 | Knowledge-Driven Analysis Identifies a Gene–Gene Interaction Affecting High-Density Lipoprotein Cholesterol Levels in Multi-Ethnic Populations | Total cholesterol, low-density lipoprotein cholesterol, triglyceride, and high-density lipoprotein cholesterol (HDL-C) levels are among the most important risk factors for coronary artery disease. We tested for gene–gene interactions affecting the level of these four lipids based on prior knowledge of established genome-wide association study (GWAS) hits, protein–protein interactions, and pathway information. Using genotype data from 9,713 European Americans from the Atherosclerosis Risk in Communities (ARIC) study, we identified an interaction between HMGCR and a locus near LIPC in their effect on HDL-C levels (Bonferroni corrected Pc = 0.002). Using an adaptive locus-based validation procedure, we successfully validated this gene–gene interaction in the European American cohorts from the Framingham Heart Study (Pc = 0.002) and the Multi-Ethnic Study of Atherosclerosis (MESA; Pc = 0.006). The interaction between these two loci is also significant in the African American sample from ARIC (Pc = 0.004) and in the Hispanic American sample from MESA (Pc = 0.04). Both HMGCR and LIPC are involved in the metabolism of lipids, and genome-wide association studies have previously identified LIPC as associated with levels of HDL-C. However, the effect on HDL-C of the novel gene–gene interaction reported here is twice as pronounced as that predicted by the sum of the marginal effects of the two loci. In conclusion, based on a knowledge-driven analysis of epistasis, together with a new locus-based validation method, we successfully identified and validated an interaction affecting a complex trait in multi-ethnic populations.
| Genome-wide association studies (GWAS) have identified many loci associated with complex human traits or diseases. However, the fraction of heritable variation explained by these loci is often relatively low. Gene–gene interactions might play a significant role in complex traits or diseases and are one of the many possible factors contributing to the missing heritability. However, to date only a few interactions have been found and validated in GWAS due to the limited power caused by the need for multiple-testing correction for the very large number of tests conducted. Here, we used three types of prior knowledge, known GWAS hits, protein–protein interactions, and pathway information, to guide our search for gene–gene interactions affecting four lipid levels. We identified an interaction between HMGCR and a locus near LIPC in their effect on high-density lipoprotein cholesterol (HDL-C) and another pair of loci that interact in their effect on low-density lipoprotein cholesterol (LDL-C). We validated the interaction on HDL-C in a number of independent multiple-ethnic populations, while the interaction underlying LDL-C did not validate. The prior knowledge-driven searching approach and a locus-based validation procedure show the potential for dissecting and validating gene–gene interactions in current and future GWAS.
| The catalog of genome-wide association studies (GWAS) [1] has collected to date over 1,194 publications since the end of 2008, for a total of over 5,697 single nucleotide polymorphisms (SNPs) that are associated with complex human diseases and other complex traits. However, most these associated SNPs exhibit a small effect size, and collectively only explain a relatively small fraction of additive variance [2], [3], [4], [5]. Specifically, a recent meta-analysis of several GWAS, studying a combined sample size between ∼20,000 to ∼100,000 individuals, identified 95 loci associated with the level of one of total cholesterol (TC), low-density lipoprotein cholesterol (LDL-C), triglyceride (TG), and high-density lipoprotein cholesterol (HDL-C) [6]. In aggregate, these loci explain only 25–30% of heritable variation for each trait [6]. Many hypotheses aiming to explain the missing heritability of GWAS have been proposed, including structural variants, rare variants, gene-environment interactions, epigenetics, and complex inheritance [2], [3], [4], [5]. Because gene-gene (epistatic) interactions may contribute to missing heritability to some extent [7], [8], [9], here we seek to find examples of pairs of loci that interact in their effects on any of the four lipid levels, which are important risk factors of coronary artery disease [10].
Epistasis has been investigated in order to understand the relationship between genotype and phenotype since Bateson [11] discovered in 1905 that some genes can suppress the effects of others. Thereafter, a number of epistatic interactions have been identified in QTL mapping studies or GWAS in humans [12], [13] and other organisms [14], [15], [16]. Studies of model organisms suggest that gene-gene interactions are a common phenomenon [17], [18], [19], [20]. However, they have proven difficult to detect in humans, chiefly due to the limited statistical power associated with the large combinatorial number of tests and the skew towards low minor allele frequencies [18], [21]. Hence, in order to increase power to detect gene-gene interactions in GWAS, a series of methods have been developed to prioritize candidate SNPs using prior knowledge of established GWAS hits [22], and recently also using knowledge of protein-protein interactions (PPIs) [23], [24] and pathway information [25].
Although some interactions affecting complex diseases and traits have been reported in humans [12], [26], replication of these interactions in independent samples has proven difficult [13]. He et al. [27] showed that this low replication is in part attributable to low power and small effect sizes of tag SNPs in GWAS. For two interacting causal loci, the observed interaction effect between two respective tag SNPs (each tagging one of the causal loci) is proportional to the underlying causal interaction effect multiplied by the product of the two linkage disequilibrium (LD) coefficients between each tag SNP and the respective causal variant. This decrease in the measured interaction effect reduces the statistical power of the interaction test and it also reduces the probability of replication of significantly identified interactions. This reduction is further exacerbated by heterogeneity in the LD structure between different populations and among population samples. These are the same problems that plague the power of single-marker GWAS tests, but they are exacerbated in interaction testing, with a quadratic dependence on LD between markers and causal loci, which lead to a much greater reduction in power. Motivated by this problem, Liu et al. [28] proposed a local validation analysis and successfully replicated the loci of a few interactions underlying common human diseases.
In this study, we aim to improve the power to detect gene-gene interactions in existing large-scale GWAS data sets by considering for interaction testing only a highly focused set of candidate SNPs extracted from prior information of known GWAS hits, PPIs, and pathway information. To improve the power of replicating gene-gene interaction signals in independent samples, we introduce an adaptive locus-based validation procedure that follows an approach similar to Liu et al. [28]. Applying these procedures for testing for gene-gene interactions underlying lipid levels, we discovered a significant interaction affecting HDL-C levels, which provides new insights into the genetic architecture of this complex trait. Using the adaptive locus-based validation procedure, we also successfully replicated this novel interaction in four independent cohorts, including two cohorts of different ethnicity.
We tested the statistical significance of gene-gene interaction between each pair of SNPs among 125 SNPs from 95 loci that have been previously individually associated with any of the four lipid levels [6] for a total of 7,750 tests, out of ∼3 trillion possible tests between each pair of SNPs in our data. Tests of interaction were conducted using genotype data or imputed genotypes in a sample of 9,713 European Americans (EAs) from the Atherosclerosis Risk in Communities (ARIC) study [29] (Materials and Methods). We used an F-test with four degrees of freedom within a linear model framework for interaction testing [30], [31]. This test considers the 3×3 table of genotype pairs for two SNPs and tests for significant interaction between the two SNPs on top of any additive or dominance effects that each of the SNPs might exhibit by itself. For consideration of statistical power and robustness, we discarded from testing pairs of SNPs for which one or more of the 9 genotype-by-genotype combinations appeared in fewer than 20 individuals in our sample (Materials and Methods).
Testing for interaction between 7,750 pairs of SNPs for each of four quantitative traits, we identified one significant interaction underlying each of LDL-C level and HDL-C level (Figure 1a). The interaction underlying LDL-C level is between rs2247056 and rs1030431 (Bonferroni corrected Pc = 0.003; Figure 1a). To explore the interaction between the two loci with better resolution, we tested for interaction between each SNP in the 100 kb surrounding rs2247056 and each SNP in the 100 kb surrounding rs1030431 and found that the interaction signal peaked between rs2853928 and rs1993453 (Pc = 0.01 after accounting for all additional pairs of SNPs tested; Figure S1). The discovery SNP pairs are in high LD with the fine-mapped SNP pairs, with an r2 value of 0.997 between rs2247056 and rs2853928 and 0.999 between rs1030431 and rs1993453. The former two reside near a pseudogene, LOC100133383, and the latter two are located near and in gene UBXN2B, respectively. However, this suggestive interaction underlying LDL-C did not replicate in independent cohorts.
Henceforth, we focus on the interaction between rs12916 and rs1532085 on HDL-C levels (Pc = 0.008), since its validation in additional cohorts is highly significant, as described below. We first tested for interaction between each SNP in the 100 kb surrounding rs12916 and each SNP in the 100 kb surrounding rs1532085. While many of these pairs show significant interactions (Figure 1b), as expected from LD, we observed the strongest signal between rs3846662 and rs2043085 (Pc = 0.002). The fine-mapped pair of SNPs is in high LD with the original pair of SNPs, with an r2 value of 0.88 between rs3846662 and rs12916 and an r2 value of 0.93 between rs2043085 and rs1532085 (Figure S2). rs3846662 is intronic in HMGCR (Table 1), which has not been previously associated with HDL-C, but has been associated with both TC and LDL-C levels [6]. rs2043085 is upstream of LIPC (Table 1), which has been previously found to be associated with HDL-C [6].
The interaction between rs3846662 and rs2043085 affects HDL-C twice as much as the effect of the polymorphism in LIPC alone: While individuals with TT genotype at rs2043085 already exhibit an average increase of 2.63 mg/ml in HDL-C (standard error (SE) = 0.014; Figure 2a), this genotype in combination with an AA genotype at rs3846662 leads to an average increase of 5.72 mg/ml (SE = 0.041; Figure 2b). The linear model with these two SNPs has an R-square value of 0.5% and the linear model with the two SNPs and their interaction has an R-square value of 0.8%, which indicates that the interaction explains additional 0.3% of the overall variation in HDL-C levels (Materials and Methods; Table S1). We tested whether rs3846662 and rs2043085 exhibit gene-gene interactions underlying any of the other lipid levels, and found a nominally significant interaction underlying LDL-C (P = 0.028), and almost significant interaction underlying TG (P = 0.08) in ARIC.
We performed a larger scale interaction analysis between all pairs of SNPs that (i) are found in interacting genes according to a curated human protein-protein interaction network (∼6 million pairs), or (ii) are involved in the pathway of metabolism of lipids and lipoproteins (∼27 million pairs). All SNPs in a gene were considered, as well as in the 5 kb regions upstream and downstream. This analysis detected no significant gene-gene interactions following Bonferroni correction (Pc≥0.58 for PPIs; Figure S3; Pc≥0.14 for pathway; Figure S4).
Considering the quadratic reduction in replication power as a function of LD between tag SNPs and causal loci, we aimed to increase power via an adaptive locus-based validation procedure that is related to that of Liu et al. [28]. In considering a replication dataset, the procedure follows three sequential stages that leverage the signals of proxy markers: (i) test for interaction between the original SNP pair between which gene-gene interaction has been detected; (ii) test for interactions between each of the two original SNPs and each SNP in the proximate region containing the other original SNP; (iii) test for interactions between each pair of SNPs in each of the two respective proximate regions containing the two original SNPs. This validation procedure proceeds sequentially and stops at any stage when significant interactions were detected after multiple-testing correction. Both the method of Liu et al. and our adaptive locus-based validation method focus on replicating the interaction between a pair of loci, rather than between a pair of SNPs, due to the power limitations of replicating an interaction between SNPs. The null hypothesis of the entire three-stage procedure is that there is no interaction between the pair of loci, rather than just between the pair of SNPs, thus the procedure continues sequentially as described to consider proxy SNPs from the loci containing each original SNP. Replication is successful if an interaction between any SNP pair from the two loci is significant after multiple-testing correction. Similar locus-based approach has also been used in the context of gene-based GWAS tests for single-marker association, which use an entire gene or locus as the testing unit of association, rather than a single SNP [27], [32].
To validate the gene-gene interaction affecting HDL-C, we performed replication analyses in two additional GWAS datasets from the Framingham Heart Study (FHS) [33] and the Multi-Ethnic Study of Atherosclerosis (MESA) [34], as well as in the African American (AA) cohort from the ARIC study [29]. Using our adaptive locus-based procedure, we tested for interaction sequentially between SNPs surrounding rs3846662 and SNPs surrounding rs2043085. We observed significant interactions in the two additional EA cohorts from FHS and MESA (Figure 1c), with Pc = 0.002 and Pc = 0.006 for the most significantly interacting SNP pair (Table 1). Replication was also significant in Hispanic Americans (HA) from MESA and AAs from ARIC (Figure 1c; Table 1). The R-square of linear model with the two interacting SNPs varies between 0.2–0.5% across the four replication cohorts, with the interaction term between the two explaining an additional 0.2–1.1% of the overall variation in HDL-C levels (Table S1). The replication procedure failed in a sample of AAs from MESA (Figure S5).
None of the successful replications were replicated at stage (i) of the adaptive locus-based validation procedure, which means that an interaction between the same SNP pair is not observed significantly in the additional samples. The interaction was successfully validated in stage (ii) of the three stages in the MESA EAs, with the same SNP in HMGCR (rs3846662) and a proxy SNP near LIPC exhibiting a significant gene-gene interaction after multiple-testing correction. The other three successful replications occurred at stage (iii) (Table 1), emphasizing the importance of a locus-based replication approach. The combined evidence from the discovery and four different validation cohorts for a gene-gene interaction between the two loci under study is overwhelmingly significant, even following a conservative Bonferroni correction (Pc = 9.0×10−8).
While the gene-gene interaction signal peaks for different pairs of SNPs across the different cohorts (Table 1), the type of interaction and effect patterns appear consistent across several sample sets (Figure S6). To test this formally, we partitioned the significant SNP-SNP interactions into the four possible interaction components on top of the marginal SNP effects, namely additive by additive (A×A), additive by dominance (A×D), dominance by additive (D×A), and dominance by dominance (D×D) components (Materials and Methods). Considering a nominal significance level of 0.01, D×A and D×D components are significant and underlie the significant interaction in the ARIC EA discovery set, between rs12916 and rs1532085 (Table S1). All four terms are significant between the pair of SNPs, rs3846662 and rs2043085, that resulted from fine mapping in the same discovery set, with D×A and D×D being of the same effect direction (sign) and similar effect sizes as between rs12916 and rs1532085 (Table S1). Examining the two replication cohorts of a similar (EA) ancestry, the interaction in the MESA cohort similarly shows significant D×A and D×D components, with same effect direction, though with larger effect sizes and a higher proportion of phenotypic variance explained (Table S1). None of the four terms is significant by itself in the EA FHS cohort. These results of consistent patterns of interaction across the EA cohorts support the possibility that they are all governed by the same (unobserved or partially unobserved) interacting variants.
To verify that our results are not an artifact of imputation errors, we compared imputed genotypes of the two SNPs (rs12916 and rs3846662) that were involved in significant interactions and for which we could obtain measured genotype data from an independent source, using the ITMAT/Broad/CARE (IBC) Vascular Disease 50 k SNP Array chip [35]. For these two SNPs, r2 between imputed and actual genotypes is 0.914 and 0.921 and the genotype concordance rate is 94.5% and 94.7%, respectively. Although the imputation is not perfect, the two interaction tests involving these two SNPs are at least as significant when replacing imputed genotypes with measured IBC genotypes, consistent with imputation errors adding noise and masking some of the signal, rather than biasing the statistical test.
Tests of gene-gene interactions are not as powerful as tests of single-marker association, so a judicious strategy is essential for successful interaction analysis in GWAS [9], [36]. The first step is to determine the size of the analysis, genome-wide or focusing on candidate SNPs. This step should consider the sample size, possible effect size of the underlying interaction, and the desired statistical power. Current single-marker GWAS have been successful in detection of single-marker associations for many complex diseases or traits using a stringent genome-wide significance level (P<5×10−8). To achieve a similar success for interaction analysis, we are limited to performing ∼1 million tests even if the interaction test and single-marker test had the same statistical power. This limitation means that we are not able to conduct an inclusive all-by-all pair-wise interaction analysis in current GWAS. Thus, in this study we only tested for interactions between candidate SNPs based on prior knowledge.
We used three types of prior knowledge, known GWAS hits, protein-protein interaction networks, and known functional pathways. These three analyses might be different in the enrichment of epistasis signals and are also different in the number of interaction tests, 7,750 based on known GWAS hits, ∼6.2 million using PPI, and ∼27 million with pathway information. We found significant interactions from the 7,750 interaction tests using known GWAS hits. As the sample size of ∼10,000 individuals is relatively large among existing GWAS, this indicates that the observed (tagged) effect size of any other underlying interactions is no larger than the marginal effects of single SNPs. It is also likely that the epistasis signals are better enriched between markers that are marginally associated with lipid traits such that testing interactions among known GWAS hits is more powerful in our study. Therefore, our results suggest that a small-scale interaction analysis of candidate SNPs driven by known marginal associations might be a good choice for detecting epistatic interactions in current GWAS.
Recently, the Population Architecture using Genomics and Epidemiology Study [37] found only ∼50% of the 125 reported associations with lipid levels [6] to replicate in three non-European cohorts. Due to the quadratic decrease in the interaction effect of tagged markers, gene-gene interactions are even less likely to replicate in diverse populations. Leveraging signals from proximate linked SNPs, our adaptive locus-based method successfully validated gene-gene interactions between HMGCR and LIPC in four additional, independent cohorts, including two of non-European ancestry. Although the most significant interaction in each cohort involves different SNPs, they are proximate across the cohorts, with stronger LD and smaller distances amongst the three EA cohorts and weaker LD and larger distances between them and the HA and AA cohorts (Figure S2 and Table 1). The differences in distance and LD between ethnicities could be due to differences in genetic background, demographic history, and natural selection, even if the different SNP pairs capture the same underlying causal interaction. However, the interaction shows similar patterns among some, but not all cohorts (Figure S6 and Table S1), while the different SNPs around HMGCR are in strong LD, and those around LIPC show weak LD (Figure S2). These results suggest that the five SNP pairs either capture separate causal interactions or are only in weak LD with the same pair of interacting, unobserved variants.
Another possibility is that the interaction is between relatively rare causal variants: Much like rare causal variants can lead to multiple independent associations of common variants, dubbed “synthetic associations” [38], an interaction between two rare causal variants can produce an even larger number of independent “synthetic interactions”, which can in principle explain almost-independent, yet proximate gene-gene interactions. Another possibility is that the underlying interaction is more complex and involves more than a pair of SNPs. In that case, in our analysis of pairs of SNPs, each pair might tag only certain aspects of the underlying interaction.
Both HMGCR and LIPC are involved in metabolism of lipids and lipoproteins. HMGCR, which has been associated with TC and LDL-C [6], regulates the rate of cholesterol synthesis via a negative feedback mechanism mediated by sterols and non-sterol metabolites [39]. LIPC encodes hepatic lipase which is an important enzyme in HDL metabolism [40] and has been previously associated with HDL-C levels [6]. The interaction between variants in these genes as discovered in this study can be possibly explained by an indirect interaction between cholesterol synthesis and the metabolism of LDL and HDL particles. HGMCR is the rate-controlling enzyme in the mevalonate pathway for cholesterol synthesis [41]. Much of this cholesterol will form cholesteryl esters that will be packaged into various lipoproteins including LDL, HDL, and TG-rich lipoproteins. There are a number of known lipoprotein interactions that result in the flow of cholesterol in the form of cholesteryl esters from LDL and VLDL to HDL-C [42]. This cholesterol is later processed with the HDL particle by either reabsorbing into the liver or excretion in the urine [43].
The rs2043085 SNP in the LIPC gene region, where our strongest signal has been observed in fine mapping in the discovery panel, was recently associated with elevated HDL-C in an additional cohort of individuals with mixed dyslipidemia [44]. Increased HDL-C may be related to modest inhibition of TG hydrolysis in the HDL particle by hepatic lipase, slowing its excretion in the urine along with its cholesterol content. Because HMGCR has a major effect on cholesterol synthesis, it will also indirectly affect the cholesterol content in the HDL particle through its interaction with LDL and TG-rich particles. In addition, LIPC has been reported to exhibit gene-gene interaction with other genes associated with lipid traits [45], [46], and HMGCR has been reported to interact with ABCA1 in Alzheimer's disease risk [47]. While these results increase the plausibility of a biological interaction between these two genes, we note that a statistical gene-gene interaction does not necessarily entail an underlying epistatic interaction in the biological sense [7]. We also note that while we refer to the interaction as being between HMGCR and LIPC, these two genes are implicated only by genomic proximity, and we presented no direct evidence that these genes are the interacting functional units.
We conclude that a focused study with higher enrichment of putative signals might have improved power to detect gene-gene interactions underlying complex diseases or traits. By focusing only on SNPs that were previously associated with the studied trait, HDL-C level, or any of a handful of related traits (other lipid levels), we successfully identified an interaction between SNPs in or near HMGCR and SNPs upstream of LIPC in European American samples. By using a locus-wide validation procedure to overcome the quadratic impact of partial SNP tagging on the observed interaction effect size, we further replicated the interaction between these loci in additional European American samples, as well as in African American and Hispanic American samples.
All work done in this paper was approved by local institutional review boards or equivalent committees.
We obtained Affymetrix 6.0 SNP array genotyping of samples from the ARIC study [29]. We obtained Affymetrix 6.0 SNP array genotyping of MESA samples and Affymetrix 500 K SNP array genotyping of FHS samples from the database of Genotypes and Phenotypes (dbGaP; MESA SHARe, downloaded in May 2011 and Framingham Cohort, downloaded in April 2010) [48], [49]. Genotype quality control (QC) steps included the exclusion of individuals with >10% missing data, and the exclusion of SNPs with call rates <90%, minor allele frequencies (MAF)≤1%, or Hardy-Weinberg Equilibrium (HWE) test with P<10−6. For the pairwise interaction test of each pair of SNPs we also required (i) sample size of each of the nine possible genotype-by-genotype combinations of the two SNPs being >20 in the discovery analysis and >10 in the validation analysis; and (ii) LD of r2<0.1 between the two SNPs between which interaction is tested. The first requirement is a generalization of the MAF requirement in single-marker analysis.
We used IMPUTE2 [50] with HapMap3 [51] and 1000 Genomes [52] reference haplotypes to impute untyped SNPs, resulting in the same set of SNPs across cohorts. We did not impute untyped SNPs in MESA HA samples since no appropriate reference panel was available at the time we conducted our analysis. We discarded imputed SNPs with information score less than 0.6. Following this QC stage, we considered the genotype with the maximum posterior probability, and discarded SNPs for which this probability is <0.8.
We considered four lipid measurements: TC, LDL-C, TG, and HDL-C. All measurements were done in the fasting state using standard enzymatic methods. In all three studies, each lipid level is measured at multiple time points and we considered the average level per individual of each lipid in all our analyses. We applied a log transformation to TG levels to normalize them in face of the skewness in the original distribution, as previously proposed [6]. We excluded individuals known to be taking lipid-lowering medications.
Gender, age, age squared, and body mass index (BMI) were included as covariates in all analyses, similarly to GWAS based on these phenotypes [6], [26]. We averaged values for age and BMI whenever multiple measurements were available, in line with the averaging of lipid levels [6]. The average age was also squared and included as a covariate. Plate is also included as a covariate in the ARIC data since it is correlated with some of the lipid levels (“plate effect”; data not shown).
Principal component (PC) analysis was conducted using EIGENSOFT [53]. Top 10 PCs were included in the analysis as covariates to account for potential population stratification in each of the ARIC and MESA cohorts. For FHS, we applied a mixed model method to account for relatedness by performing the interaction test on the residuals after removing familial structure [26], [54].
As described in [30], [31], we tested for interaction between two SNPs on a quantitative trait as follows. Assume Y is the trait of interest and Gi is the genotype of SNP i (i = 1, 2). Gi denotes the number of copies of the reference allele (0, 1, or 2). Two indicator variables xi and zi are defined for each SNP asTwo linear models were fitted. The first, model (1), allows for additive and dominance effects at each SNP, but is strictly additive (i.e. no interaction) over the two SNPs. The second, model (2), allows for the four possible forms of genotype-by-genotype interaction (additive×additive, additive×dominance, dominance×additive, and dominance×dominance) [55], as follows:(1)(2)Here, β0 denotes a vector of intercept and covariates as described above. ai and di denote the additive and dominance effects of SNP i, and iaa, iad, ida, and idd are the four interaction effects between the two SNPs.
We tested for the existence of an epistatic interaction of any type by an F-test with four degrees of freedom between models (1) and (2) [18]. The F-test with four degrees of freedom tends to be more powerful when little is known about the underlying epistatic effect in terms of the possible directions of the deviation from independence of the additive effects. This test is similar to the “–epistasis” option in PLINK [56], except that only additive effects and their interaction are considered in PLINK, and an F-test with one degree of freedom is hence applied. We also considered a test for “physiological epistasis” [7] under the same model and obtained very similar results (data not shown). Throughout the results, we report P values following a conservative Bonferroni correction. To compare the effects of the different SNP pairs detected in our discovery and validation analyses, we also estimated and tested the four interaction terms in model (2) for each pair of SNPs from different cohorts using a t-test.
Although we only focus on pairwise interaction analysis, the total number of potential pairwise interaction tests across 2.5 million SNPs is still huge, about 3 trillion tests. Due to the huge reduction in power entailed by multiple-testing correction for such a large number of tests, it is crucial to restrict the number of tests a priori. We aimed to enrich possible interaction signals in the limited number of tests we considered through the following three strategies.
Liu et al. [28] developed a local validation strategy and validated a few interactions affecting common human diseases. This strategy attempts to replicate the interaction between two loci rather than the interaction between the original pair of SNPs. To further improve power, we extended this local validation strategy to an adaptive locus-based validation procedure: For a detected interaction between SNP A and SNP B in the discovery panel we followed three stages in each of the validation panels. (i) First, test for interaction between SNP A and SNP B; (ii) Second, if the interaction in (i) is not significant by itself, test for interaction between A and each SNP<200 kb away from B, and similarly between B and each SNP surrounding A; (iii) Last, if no test in the second stage is significant following multiple-hypothesis correction, test for interaction between each SNP<100 kb away from A and each SNP<100 kb away from B. Assuming n1 and n2 SNPs in the locus surrounding A and B, respectively, the number of interaction tests performed is 1, n1+n2, and n1×n2 in the three stages, respectively, with n1 and n2 in stage (iii) being smaller than those in stage (ii) due to considering only 100 kb. To maintain power in light of multiple-testing correction, the validation process proceeds sequentially and stops once we find significant results after multiple-testing correction. The interaction between rs3846662 and rs2043085 on HDL-C was successfully validated in stage (ii) for MESA EA samples and in stage (iii) for the MESA HA, FHS EA, ARIC AA cohorts. It did not validate significantly after multiple-testing correction in any of the three stages in the MESA AA samples. We used the same procedure as in step (iii) for fine mapping within the discovery panel.
|
10.1371/journal.pgen.1002391 | Physiological IRE-1-XBP-1 and PEK-1 Signaling in Caenorhabditis elegans Larval Development and Immunity | Endoplasmic reticulum (ER) stress activates the Unfolded Protein Response, a compensatory signaling response that is mediated by the IRE-1, PERK/PEK-1, and ATF-6 pathways in metazoans. Genetic studies have implicated roles for UPR signaling in animal development and disease, but the function of the UPR under physiological conditions, in the absence of chemical agents administered to induce ER stress, is not well understood. Here, we show that in Caenorhabditis elegans XBP-1 deficiency results in constitutive ER stress, reflected by increased basal levels of IRE-1 and PEK-1 activity under physiological conditions. We define a dynamic, temperature-dependent requirement for XBP-1 and PEK-1 activities that increases with immune activation and at elevated physiological temperatures in C. elegans. Our data suggest that the negative feedback loops involving the activation of IRE-1-XBP-1 and PEK-1 pathways serve essential roles, not only at the extremes of ER stress, but also in the maintenance of ER homeostasis under physiological conditions.
| Proteins destined for secretion outside of eukaryotic cells are trafficked through the endoplasmic reticulum (ER). Protein folding in the ER involves the activity of chaperones, as well as catalysis of post-translational modifications such as disulfide bond formation and glycosylation. When the folding capacity of the ER is exceeded, the resulting accumulation of misfolded proteins activates the Unfolded Protein Response (UPR), a conserved signaling response that functions to restore protein folding homeostasis in the ER. Genetic studies have established that the UPR is required for the development of specific cell types in mammals, such as antibody-secreting plasma cells, and recent studies implicate a critical role for UPR signaling in the pathogenesis of metabolic and inflammatory diseases. In this paper we show that innate immunity and elevated physiological temperatures each necessitate UPR activity for C. elegans survival. Furthermore, we show that, under physiological conditions of larval development, basal activity of the UPR is required for the maintenance of ER homeostasis. Our data support the idea not only that the UPR functions as a “stress response” pathway, protecting against the extremes of unfolded protein accumulation, but also that the UPR plays a more general role in animal physiology and development.
| The accumulation of misfolded proteins in the endoplasmic reticulum (ER), also known as ER stress, activates the Unfolded Protein Response (UPR), which upregulates the synthesis of chaperones such as BiP and components of ER-associated degradation (ERAD), promotes ER expansion, and attenuates translation [1]–[3]. The UPR is conserved from yeast to humans and in metazoans is comprised of three branches, mediated by the transmembrane ER luminal sensors IRE-1, PERK/PEK-1, and ATF-6 [1]–[3]. In response to ER stress, IRE-1 oligomerizes, activating an endoribonuclease domain that splices the mRNA of xbp-1 to enable the generation of the activated form of the XBP-1 transcription factor [4]–[7]. PERK phosphorylates the translation initiation factor eIF-2α, causing global translational attenuation that diminishes the secretory load to the ER [8]. In addition, phosphorylation of eIF-2α selectively increases the translation of ATF4, a transcription factor that regulates stress responses [9]. ATF-6 undergoes proteolysis, releasing the cytosolic domain of ATF-6, which functions as a transcription factor that translocates to the nucleus and activates transcription of UPR genes [10].
Genetic studies suggest essential roles for UPR signaling in animal development. In mice, genetic studies focused on either the IRE-1-XBP-1 or the PERK pathway have shown that each functions in the development of specialized cell types, including plasma cells, pancreatic ß-cells, hepatocytes, and intestinal epithelial cells [3], [11]–[15]. In Caenorhabditis elegans, mutants deficient in any one of the three branches of the UPR are viable, but combining a deficiency in the IRE-1-XBP-1 pathway with loss-of-function mutations in either the ATF-6 or PEK-1 branch has been reported to result in larval lethality [6], [16]. These studies suggest that the UPR is required for animal development, but the specific essential role has not been defined. For example, UPR signaling may be required for a particular stage of development, or alternatively, constitutive UPR activity may be required. The experimental analysis of UPR signaling both in yeast and in mammalian cells has been greatly facilitated by the use of chemical agents that induce ER stress, such as the N-linked glycosylation inhibitor tunicamycin, the calcium pump inhibitor thapsigargin, and the reducing agent dithiothreitol (DTT). However, the activation of the UPR under physiological conditions is less well understood [17]. Constitutive IRE-1 activity has been observed in diverse types of mammalian cells, particularly with high secretory activity or in the setting of increased inflammatory signaling [11], [14], [15], [18]. These studies suggest critical roles for IRE-1-XBP-1 signaling in physiology and development, some of which have been proposed to be independent of its role in maintaining protein folding homeostasis in the ER [11], [19], [20].
Recently, we showed that XBP-1 is required for C. elegans larval development on pathogenic Pseudomonas aeruginosa, conferring protection to the C. elegans host against the ER stress caused by its own secretory innate immune response to infection [21]. Our study established that the innate immune response to microbial pathogens represents a physiologically relevant source of ER stress that necessitates XBP-1 function.
We sought to better understand the consequences of UPR deficiency under physiological conditions during C. elegans larval development. We describe our studies which suggest that even in the absence of ER stress induced by exogenously administered chemical agents, the IRE-1-XBP-1 pathway, in concert with the PEK-1 pathway, functions in a homeostatic loop that is under constitutive activation during C. elegans larval development. Our data implicate an essential role for the UPR in ER homeostasis, not only in the response to toxin-induced ER stress, but also under basal physiological conditions.
The detection of IRE-1 activity provides a sensitive and responsive measure of ER stress. Most methods used to measure IRE-1 activity require functional IRE-1-XBP-1 output, relying on either detection of the activated spliced form of the xbp-1 mRNA or the transcriptional activity of the resulting XBP-1 protein. In order to follow IRE-1 activity in the absence of a functional XBP-1 protein, we utilized the C. elegans xbp-1(zc12) mutant, which has a C→T mutation that results in an early premature stop codon [4]. We reasoned that we could detect IRE-1-mediated splicing of xbp-1(zc12) mRNA by quantitative RT-PCR (qPCR), as we have done previously for wild type xbp-1 mRNA, as a measure of IRE-1 activity [21].
We anticipated, however, that the xbp-1(zc12) mRNA might be degraded by nonsense-mediated decay (NMD) [22], which would reduce the abundance of xbp-1(zc12) mRNA (Figure 1A). Thus, we constructed a strain carrying xbp-1(zc12) and a null allele of smg-2, the C. elegans homolog of the NMD component Upf1 [23]. Indeed, we observed that the level of xbp-1 mRNA in the xbp-1(zc12) mutant was markedly diminished compared with the level of xbp-1(zc12) mRNA in the smg-2(qd101); xbp-1(zc12) mutant (Figure 1B). These data confirmed that xbp-1(zc12) mRNA is a substrate for the NMD pathway, but that inhibition of NMD permits detection of xbp-1(zc12) mRNA. As expected from the predicted truncated protein product made from translation of the xbp-1(zc12) mRNA (Figure 1A), loss of NMD had no effect on the null phenotype of the xbp-1(zc12) allele, as assessed by the effect of the smg-2(qd101) mutation on expression of an xbp-1-regulated gene, the C. elegans BiP homolog hsp-4 (Figure 1B). We next examined the level of WT xbp-1 mRNA in the smg-2(qd101) mutant, and we observed that NMD inhibition increased the level of xbp-1 mRNA 2-fold relative to WT C. elegans (Figure 1C), which suggests that the NMD complex may function to decrease the level of WT xbp-1 mRNA. This observation is consistent with a prior report suggesting that stress-induced genes may be NMD targets [24], although we hypothesize that the relatively early termination codon present in the xbp-1 mRNA prior to IRE-1-mediated splicing may also contribute to recognition and degradation by the NMD pathway. Consistent with this explanation, after exposing both the WT and smg-2(qd101) strains to tunicamycin for 4 h, the level of IRE-1-spliced xbp-1 mRNA was similar between the two strains (Figure 1C). Furthermore, the loss of NMD did not increase the lethality of either the WT strain or xbp-1(zc12) mutant when grown in the presence of tunicamycin (Figure 1D).
Comparing levels of IRE-1 activity between smg-2(qd101) and smg-2(qd101); xbp-1(zc12) animals, we observed a dramatic elevation in the level of spliced xbp-1 mRNA in the smg-2(qd101); xbp-1(zc12) strain (Figure 1E). To provide a measure for comparison, the basal elevation of spliced xbp-1 mRNA in the smg-2(qd101); xbp-1(zc12) mutant far exceeded the level of spliced xbp-1 mRNA in the smg-2(qd101) mutant even after administration of tunicamycin. Treating the smg-2(qd101); xbp-1(zc12) strain with tunicamycin resulted in only a minor additional increase in spliced xbp-1 mRNA compared with the magnitude of elevation in spliced xbp-1 in that strain under basal conditions (Figure 1E). These data show that XBP-1 deficiency results in a dramatic increase in IRE-1 activity, even in the absence of exogenously administered agents such as tunicamycin.
If the elevated level of IRE-1 activity observed in the smg-2(qd101); xbp-1(zc12) mutant were indicative of increased ER stress due to loss of xbp-1, we might anticipate compensatory activation of the PEK-1 and/or ATF-6 pathways in the absence of XBP-1. We therefore sought to determine levels of PEK-1 activity in an xbp-1 mutant through the detection of eIF-2α phosphorylation. In particular, these measurements would provide an additional measure of ER stress in the xbp-1 mutant that is not dependent on the inactivation of the NMD pathway and its aforementioned experimental caveats. Antibodies raised against mammalian eIF-2α and specifically phosphorylated eIF-2α (P-eIF-2α) cross-react with the highly homologous C. elegans protein [25], [26]. We detected a single band in immunoblots using these antibodies with lysates from WT C. elegans (Figure 2A). We observed that eIF-2α phosphorylation was induced by a 4 h exposure to a high dose of tunicamycin in a PEK-1-dependent manner (Figure 2A and Figure S1A). Of note, eIF-2α phosphorylation appears to be a less sensitive measure of ER stress than IRE-1-mediated xbp-1 mRNA splicing, as we did not observe a significant increase in eIF-2α phosphorylation in response to standard doses of tunicamycin sufficient to induce xbp-1 mRNA splicing.
We next determined PEK-1 activity under basal physiological conditions, specifically in the xbp-1 mutant. We saw induction of PEK-1-mediated eIF-2α phosphorylation relative to WT in the absence of exogenously administered agents to induce ER stress at 16°C (Figure 2B and Figure S1B). The magnitude of the effect of XBP-1 deficiency on PEK-1 activity was comparable to the induction of PEK-1 in WT by treatment with high-dose tunicamycin. We observe no increase in eIF-2α phosphorylation in the xbp-1; pek-1 mutant relative to the pek-1 mutant, confirming that the increase in eIF-2α phosphorylation in the xbp-1 mutant relative to WT is due to activation of PEK-1. In fact, we noticed a slight decrease in eIF-2α phosphorylation in the xbp-1; pek-1 mutant relative to the pek-1 mutant, but the mechanisms underlying this difference are unclear. These data corroborate our observations of increased xbp-1 mRNA splicing in the xbp-1 mutant. Taken together, the increase in levels of IRE-1 and PEK-1 activity in the xbp-1 mutant suggests that XBP-1 deficiency is accompanied by a marked increase in constitutive ER stress under basal physiological conditions.
Previously, we reported that the activation of innate immunity by infection with pathogenic P. aeruginosa induces ER stress, and that XBP-1 serves an essential role in protecting the host against the detrimental effects of immune activation [21]. Our prior ultrastructural analysis of the ER in xbp-1 mutants suggested that disruption of ER homeostasis contributes to this phenotype. One explanation for these observations is that ER homeostasis in the xbp-1 mutant might be minimally perturbed under basal physiological conditions but have a pronounced sensitivity to ER stress from endogenous (e.g. immune activation) or exogenous (e.g. tunicamycin) sources. However, the data in Figure 1 and Figure 2 suggest that even during physiological growth and development, XBP-1 deficiency results in a marked elevation in levels of basal ER stress. We hypothesized, therefore, that under these circumstances, the activation of innate immunity might further increase ER stress levels.
The smg-2(qd101); xbp-1(zc12) strain provided the opportunity to assess levels of ER stress caused by immune activation in the setting of XBP-1 deficiency. Whereas a 4 h exposure of the WT strain to P. aeruginosa PA14 causes a two-fold increase in spliced xbp-1 mRNA relative to exposure to the relatively non-pathogenic bacterial food Escherichia coli OP50 (Figure 3A and [21]), we observe a blunted response to P. aeruginosa infection in the smg-2(qd101) mutant (Figure 3A). This observation is likely due to the 5-fold elevation in spliced xbp-1 mRNA levels in the smg-2(qd101) mutant (Figure 1C), which may buffer the ER from the stress caused by pathogen-induced immune activation. Nevertheless, we observed that the level of spliced xbp-1 mRNA in the smg-2(qd101); xbp-1(zc12) mutant was increased by a 4 h exposure to P. aeruginosa relative to the smg-2(qd101); xbp-1(zc12) mutant treated in parallel with E. coli (Figure 3A). Specifically, under these treatment conditions, the level of spliced xbp-1 mRNA in the smg-2(qd101); xbp-1(zc12) mutant was 20-fold greater than that of the smg-2(qd101) mutant in the absence of additional stress, whereas exposure to P. aeruginosa increased the level of spliced xbp-1 mRNA to over 25-fold that of the smg-2(qd101) mutant (Figure 3A). The total amount of xbp-1 mRNA was unchanged between smg-2(qd101) and smg-2(qd101); xbp-1(zc12) strains, indicating that the increase in spliced xbp-1 mRNA is due to increased IRE-1 activation.
We observed persistent elevation of spliced xbp-1 mRNA after an 11 h exposure to P. aeruginosa, above the elevated basal levels of spliced xbp-1 mRNA in the xbp-1 mutant, suggesting that IRE-1 activity is not attenuated under conditions of physiological ER stress (Figure 3B).
Previously, we established that the ER stress induced by exposure to P. aeruginosa, as well as the lethality of the xbp-1 mutant during infection by P. aeruginosa, are suppressed by a loss-of-function mutation in pmk-1, which encodes a conserved p38 mitogen-activated protein kinase (MAPK) that regulates innate immunity in C. elegans [27]. Our interpretation of these data was that loss of PMK-1 activity diminished the secretory load on the ER by attenuating the innate immune response. In support of this interpretation, we found that the pathogen-induced increase in spliced xbp-1 mRNA in smg-2(qd101); xbp-1(zc12) was suppressed in the smg-2(qd101); xbp-1(zc12); pmk-1(km25) mutant, although the basal levels on E. coli OP50 nevertheless remained markedly elevated (Figure 3A). These data provide quantitative support for a model in which the activation of PMK-1-mediated innate immunity is a physiologically relevant source of ER stress, which in XBP-1-deficient animals exacerbates an already elevated level of ER stress to cause larval lethality.
What are the functional consequences of the elevated ER stress present in the xbp-1 mutant under standard growth conditions, in the absence of infection? The xbp-1 mutant, while viable, exhibits increased sensitivity to exogenously administered ER stress as well as physiological ER stress from immune activation [21], [28]. Inactivation of both xbp-1 and pek-1 was previously reported to result in larval arrest when propagated at 20°C [16]. Our observations of constitutive ER stress in the xbp-1 mutant and increased PEK-1 activity suggest a compensatory functional role for pek-1, and thus we sought to further characterize the larval arrest phenotype of the xbp-1; pek-1 mutant.
Surprisingly, we observed that the xbp-1(tm2482); pek-1(ok275) double mutant exhibited temperature-dependent viability over the physiological temperature range of C. elegans (Figure 4A). The larval development of the xbp-1(tm2482); pek-1(ok275) mutant was similar to that of WT at 16°C. At 20°C, however, approximately half of xbp-1(tm2482); pek-1(ok275) eggs developed to become gravid adults, while the remainder arrested during larval development in the L2 and L3 stages. These arrested larvae died over the course of several days with intestinal degeneration as previously described (Shen et al., 2005). At temperatures greater than 23°C, larval lethality was 100%. At 25°C, 100% of the population died in the L1 and L2 stages after just 2 days. The physiological temperature range for propagation of C. elegans in the laboratory is generally 15°C to 25°C, with optimal reproduction at 20°C. Thus, the observed temperature dependence is observed not at “heat shock” temperatures, but rather, well within the range of physiological temperatures for C. elegans.
The temperature dependence of xbp-1(tm2482); pek-1(ok275) lethality permitted the investigation of whether the larval lethality of the xbp-1; pek-1 mutant is due to a requirement for XBP-1 and PEK-1 at a specific stage of development, or whether the activities of XBP-1 and PEK-1 are required constitutively for viability at other life stages. Specifically, we propagated two xbp-1; pek-1 mutants comprised of different mutant alleles at 16°C until the animals reached the L4 larval stage, then either maintained the mutants at 16°C or shifted them to 25°C to monitor survival. When shifted to 25°C, the xbp-1; pek-1 double mutants exhibited a sharp decrease in survival as compared with the strains maintained at 16°C (Figure 4B). These observations suggest that the activity of either XBP-1 or PEK-1 is not specifically required at a particular developmental stage; instead, the constitutive activities of XBP-1 and PEK-1 are required for survival at physiological temperatures.
Although we previously observed that pek-1 and atf-6 single mutants did not exhibit larval lethality in the presence of pathogenic bacteria [21], our data presented in this paper suggest that PEK-1 functions in parallel to XBP-1 under physiological conditions in C. elegans to maintain ER homeostasis. Because the xbp-1; pek-1 mutant is viable through larval development at 16°C, we were able to ask whether PEK-1 contributes to protection against immune activation in the absence of XBP-1. Populations of synchronized eggs were grown at 16°C with P. aeruginosa as the only food source and development was monitored over time. P. aeruginosa has been shown to exhibit markedly diminished pathogenicity to C. elegans adults at 16°C relative to 25°C [29], and we found this to also be the case during larval development. Specifically, the pmk-1 mutant was able to complete larval development on P. aeruginosa at 16°C (Figure 5A), whereas only half of the pmk-1 eggs grown on P. aeruginosa develop to the L4 stage at 25°C [21], indicating that immune activation is less important for development in the presence of P. aeruginosa grown at 16°C than it is at 25°C. Likewise, the larval development of the xbp-1 mutant, which is severely compromised on P. aeruginosa at 25°C [21], was equivalent to that of WT at 16°C (Figure 5A). Both the diminished pathogenicity of P. aeruginosa at 16°C and the aforementioned temperature-sensitive requirement for UPR function may contribute the survival of the xbp-1 mutant at 16°C. Nevertheless, even under these conditions, the xbp-1(tm2482); pek-1(ok275) mutant exhibited complete larval lethality on P. aeruginosa at 16°C, reminiscent of the larval lethality of xbp-1 on P. aeruginosa grown at 25°C. Eliminating PMK-1-mediated immunity completely rescued this larval lethality (Figure 5A), demonstrating that PEK-1 functions with XBP-1 to protect against PMK-1-mediated immune activation during larval development.
We next asked whether the UPR is required for survival in the presence of pathogen during adulthood. In parallel with our observation that the xbp-1; pek-1 mutant exhibits temperature-sensitive lethality both during larval development and when shifted to a higher temperature from the L4 larval stage, we found that the xbp-1; pek-1 mutant exhibits enhanced lethality relative to the WT strain or either of the single mutants when shifted at the L4 stage to P. aeruginosa at 16°C (Figure S2). These data suggest that the UPR is required for survival during immune activation both in larval development and in adulthood.
Larval arrest of xbp-1; pek-1 mutants has been reported to be accompanied by evidence of intestinal degeneration, including the appearance of vacuoles and light-reflective aggregates in intestinal cells, degradation of intestinal tissues, and distention of the intestinal lumen [16]. We observe similar morphology not only in xbp-1; pek-1 larvae at 23°C on E. coli OP50, but also at 16°C on P. aeruginosa. The similar appearance between xbp-1; pek-1 larvae dying either at 16°C on pathogenic bacteria or at 23°C on E. coli OP50 led us to consider whether ER stress arising from intestinal innate immune activation might contribute in a similar manner to both conditions. We have previously characterized PMK-1-mediated innate immunity and observed both basal and induced components to immunity regulated by PMK-1 [30]. We therefore hypothesized that basal immune activity under standard, non-pathogenic growth conditions could present a low level of ER stress that is severely exacerbated in the absence of intact physiological UPR function, leading to larval lethality of the xbp-1; pek-1 mutants. Consistent with this hypothesis, we observed that pmk-1 loss-of-function was able to partially suppress the larval lethality of the xbp-1; pek-1 double mutant at 23°C and 25°C (Figure 5B).
One explanation for the temperature-sensitive lethality of the xbp-1; pek-1 mutant is that increased temperature leads to increased PMK-1 pathway activation, perhaps as the “non-pathogenic” E. coli OP50 becomes slightly pathogenic. However, the temperature-sensitive lethality is not abrogated by loss of PMK-1; instead, the xbp-1; pmk-1; pek-1 mutant exhibits larval lethality at a temperature several degrees higher than the xbp-1; pek-1 mutant (Figure 5B). Furthermore, the temperature-sensitive larval lethality of the xbp-1; pek-1 mutant on E. coli OP50 was not suppressed by the presence of the bacteriostatic drug ampicillin (Figure S3A). These data indicate that basal immune activation and temperature are distinct contributors to ER stress that function in parallel during growth on E. coli OP50.
The temperature-dependent larval lethality of the xbp-1; pek-1 mutant over a physiological temperature range suggested that UPR signaling might be required for survival in response to thermal stress. Indeed, we observed that the xbp-1 mutant exhibited larval lethality when grown at 27°C, an elevated temperature at which WT N2 C. elegans exhibits a reduced brood size and increased dauer formation (Figure 5C). Similar to our observation that depletion of basal immunity rescued the development of the xbp-1; pek-1 mutant when propagated on E. coli OP50, the temperature-sensitive lethality in the xbp-1 mutant was suppressed in the xbp-1; pmk-1 double mutant (Figure 5C), but not by the presence of ampicillin (Figure S3B).
Unlike the xbp-1 mutant, the development of the pek-1 mutant at 27°C was similar to WT. This is reminiscent of our previous observation that the pek-1 mutant did not exhibit the larval lethality found in xbp-1 when grown on P. aeruginosa at 25°C [21]. However, we next grew the pek-1 mutant on P. aeruginosa at 27°C, reasoning that the elevated temperature would not only increase the ER stress caused by basal growth, but also enhance the pathogenicity of the P. aeruginosa and thereby increase the immune response relative to that at 25°C. Indeed, the pmk-1 mutant exhibited 100% larval lethality on P. aeruginosa at 27°C (Figure 5D), as compared with the 50% lethality we have previously reported for the pmk-1 mutant on P. aeruginosa at 25°C (Richardson et al., 2010). The increased susceptibility of this immune-deficient mutant to P. aeruginosa at 27°C relative to 25°C indicates that the increased temperature causes an increase in P. aeruginosa pathogenicity. On P. aeruginosa at 27°C, the pek-1 mutant exhibited larval lethality relative to the WT strain grown 27°C (Figure 5D). These data further suggest that PEK-1 functions in parallel with XBP-1 to protect C. elegans against the ER stress caused by immune activation.
We showed in Figure 5B and 5C that loss of PMK-1 improves larval development of the xbp-1; pek-1 mutant and the xbp-1 mutant, respectively, in the absence of infection. We suggested that the mechanism behind this phenomenon is that the previously described basal immune activity through the PMK-1 pathway [31] contributes to ER stress. However, we also considered the possibility that the PMK-1 pathway might play an immunity-independent role in exacerbating ER stress in the setting of UPR deficiency. To test this possibility, we examined the ability of WT and UPR mutants to develop in the presence of tunicamycin with or without functional pmk-1. We found that the pmk-1 mutant actually exhibited increased sensitivity to tunicamycin during development. In fact, the pmk-1 mutant exhibited greater lethality at a lower dose of tunicamycin than either the xbp-1 or pek-1 single mutants (Figure 6). These data suggest that the PMK-1 pathway influences ER stress in two ways. First, during infection or under standard growth conditions in the setting of UPR depletion, activation of the PMK-1 pathway generates an increased secretory load that contributes to ER stress. However, when ER stress is induced exogenously with tunicamycin, the PMK-1 pathway activity serves a protective function.
We have shown that the IRE-1-XBP-1 and PEK-1 pathways function together to maintain ER homeostasis in C. elegans under physiological conditions. We found that XBP-1 deficiency results in marked activation of both IRE-1 and PEK-1, reflecting constitutive ER stress. Activation of innate immunity mediated by PMK-1 p38 MAPK further exacerbated the constitutive ER stress in the xbp-1 mutant. To investigate the physiological roles of UPR signaling as well as the compensatory activity between distinct UPR pathways, we examined both the individual and the combined effects of XBP-1 and PEK-1 deficiency in vivo. We found that the xbp-1; pek-1 double mutant exhibited temperature-sensitive lethality that was independent of developmental stage. Compared with the xbp-1; pek-1 mutant, the xbp-1; pmk-1; pek-1 mutant had moderately increased survival during larval development on non-pathogenic bacteria, when there is a low level of PMK-1-mediated basal immune activity, and dramatically increased survival on pathogenic P. aeruginosa, when the PMK-1-mediated immune response is induced. We further showed that both XBP-1 and PEK-1 are required for full protection against the combined stress of immune activation and that of growth at elevated physiological temperatures, confirming that these two branches of the UPR function together to protect against physiological ER stress.
Our observation of dramatically elevated levels of IRE-1 and PEK-1 activity in the setting of XBP-1 deficiency, under standard growth conditions in the absence of exogenous agents to induce ER stress, provides strong evidence for homeostatic activity of the IRE-1-XBP-1 signaling pathway under physiological conditions (Figure 7A), and not merely at the extremes of ER stress induced by pharmacological treatment or in specialized secretory cell types. Our data also reveal a dynamic requirement for UPR signaling in survival that increases with both temperature and increased secretory activity as is induced by immune activation (Figure 7B). Interestingly, the temperature-dependent role for the IRE-1 and PEK-1 pathways is manifest at physiological temperatures optimal for C. elegans development and fecundity, far from commonly utilized “heat shock” conditions (Figure 7A). We speculate that this temperature dependence may be due to altered secretory load at higher temperature or increased tendency for proteins to aggregate in the ER in the absence of intact chaperone production.
Importantly, our data suggest that PMK-1-mediated immune activation is one of many sources of the requirement for the UPR during larval development in the absence of infection. We found that, although loss of basal PMK-1 pathway activation partially suppressed the temperature-sensitive larval lethality of the xbp-1; pek-1 mutant, the xbp-1; pmk-1; pek-1 mutant nevertheless exhibited almost complete larval lethality at 25°C. Further, using our smg-2(qd101); xbp-1(zc12) strains, we observed high constitutive IRE-1-mediated xbp-1 splicing in the xbp-1; pmk-1 mutant that was similar under these experimental conditions to that of the xbp-1 mutant (Figure 3A). These results indicate that the UPR has an essential role during development in protection against immune activation as well as additional processes. Identification of these processes will likely lead to increased understanding of conserved physiological roles of the UPR.
We found that the PMK-1 pathway not only contributes to basal ER stress but also protects against exogenous ER stress induced by exposure to tunicamycin (Figure 6). We speculate that the mechanism underlying this dual function of the PMK-1 pathway may be differences in the PMK-1-activated transcriptional output under different circumstances. The importance of the PMK-1 pathway in protection against exogenous ER stress makes the role of the PMK-1 pathway in contributing to endogenous ER stress even more striking.
In mice, Xbp1 deficiency in intestinal epithelial cells (IEC) resulted in marked intestinal inflammation that may contribute to the observed activation of not only IRE1 but also of PERK, as measured by expression of one of its downstream effectors, CHOP [12]. In mammals, the transcription factor CHOP promotes apoptosis of mammalian cells that experience prolonged ER stress [32], and indeed, the majority of Paneth cells underwent apoptosis in the Xbp1−/− IECs. Our observations are consistent with the idea that Xbp1−/− IECs may be predisposed to detrimental consequences of additional ER stress caused by intestinal inflammation because of deregulation of basal ER homeostasis due to XBP-1 deficiency. In pancreatic ß-cells, another cell type that is specialized for high-level secretory activity, XBP1 deficiency has been observed to result in IRE1α hyperactivation, with increased degradation of mRNAs that encode insulin processing enzymes [33].
Our observations that PEK-1, in concert with XBP-1, functions to protect against ER stress from immune activation differ from observations in mouse macrophages, in which TLR stimulation was shown to activate IRE1, but PERK activation was reported to be suppressed rather than elevated [34], [20]. This difference may be due to roles for XBP-1 in macrophages that extend beyond its function in maintaining ER homeostasis. Indeed, when stimulated by TLRs in macrophages, the IRE1-XBP1 pathway was shown to induce expression of immune effectors rather than typical UPR genes, suggestive that the IRE1-XBP1 pathway may have been co-opted in macrophages to promote macrophage-specific function independent of the UPR [20].
Our data support the idea that UPR signaling does not function simply in response to the extremes of ER stress, as when induced by tunicamycin or by the elevated secretory load of specialized cells such as plasma cells, but instead, as a critical pathway in the maintenance of ER homeostasis during normal growth and development in C. elegans. The diverse and dramatic consequences of XBP-1 deficiency on development and disease, taken together with our observations on the effect of XBP-1 deficiency on basal ER stress levels, underscore the critical role of homeostatic UPR signaling in both normal physiology and disease.
C. elegans strains were constructed and propagated according to standard methods on E. coli OP50 at 16°C [35]. The smg-2(qd101) allele was isolated by K. Reddy and contains a C→T nonsense mutation at nucleotide 1189 of the spliced transcript. The following strains were used in the study: N2 Bristol, ZD627 smg-2(qd101), ZD607 smg-2(qd101);xbp-1(zc12), ZD605 smg-2(qd101);xbp-1(zc12);pmk-1(km25), KU25 pmk-1(km25), RB545 pek-1(ok275), ZD510 xbp-1(tm2482);pek-1(ok275), ZD524 xbp-1(zc12);pek-1(tm629), ZD496 xbp-1(tm2482);pmk-1(km25);pek-1(ok275). All of the alleles used are predicted to be null alleles. Specifically, xbp-1(tm2482) is a 202 bp deletion from nt 231 that causes a frame-shift. The xbp-1(zc12) allele is a nonsense mutation that changes Q34 to an ochre stop. The two alleles exhibit an equivalent phenotype in every assay tested ([21]; this work, and our unpublished data). The pek-1(ok275) allele is a 2013 bp deletion and the pek-1(tm629) allele is a 1473 bp deletion, both of which remove the PEK-1 transmembrane domain and are therefore likely null alleles [6]. Double mutants were made between xbp-1 and pek-1 by crossing strains marked with GFP: xbp-1(III);pT24B8.5::GFP(agIs220)(X) and pT24B8.5::GFP(agIs219)(III);pek-1(X). GFP-negative F2s were singled, propagated, and genotyped by PCR.
For the experiment in Figure 1B, 1C, and 1E, L1 larvae were synchronized by hypochlorite treatment, washed onto E. coli OP50 plates, and grown for 40 h at 16°C to the L3 stage, when they were washed in M9 to plates containing E. coli OP50 or E. coli OP50 with 5 µg/ml tunicamycin. For the experiments in Figure 3, strains were grown and treated as previously described [21]. Specifically, L1 larvae were synchronized by hypochlorite treatment, washed onto E. coli OP50 plates and grown at 20°C for 23 h, then washed in M9 onto treatment plates. After incubation at 25°C for indicated times, worms were washed off plates and frozen in liquid nitrogen. For P. aeruginosa treatment, P. aeruginosa strain PA14 was grown in Luria Broth (LB), and 25 µl overnight culture was seeded onto 10 cm NGM plates. Plates were incubated first at 37°C for 1 d, then at room temperature for 1 d. All RNA extraction, cDNA preparation, qRT-PCR methods and specific primers to detect xbp-1 mRNA were as described previously [21].
For all immunoblots, strains were synchronized by hypochlorite treatment and washed onto E.coli OP50 plates for growth until the L4 stage. For the experiment in Figure 2A, strains were grown at 20°C and L4 worms were then washed in M9 onto treatment plates for incubation at 25°C for 4 hours. For the experiment in Figure 2B, strains were grown at 16°C until the L4 stage and harvested without treatment. All strains were collected and rinsed 2 times in M9. Worm pellets were resuspended in an equal volume of 2× lysis buffer containing 4% SDS, 1oomM Tris Cl, pH 6.8, and 20% Glycerol. After boiling for 15 minutes with occasional vortexing to aid in dissolution, lysates were clarified by centrifugation. Protein samples (50 µg of total lysate loaded per lane) were separated by SDS-PAGE and transferred to a nitrocellulose membrane (Bio-rad). Western blots were blocked in 5% milk in PBST and probed with (1∶10,000) anti-eIF2α [26], (1∶1,000) anti-phospho-eIFα (Cell Signaling Technology), or (1∶10,000) anti-tubulin (E7 Developmental Hybridoma Bank, Iowa City). All primary antibodies were diluted in 5% milk in PBST. Following incubation with anti-rabbit or anti-mouse IgG antibodies conjugated with horseradish peroxidase (HRP) (Cell Signaling Technology), signals were visualized with chemiluminescent HRP substrate (Amersham). Quantification of immunoblots was preformed with ImageJ [36].
For all development assays, strains were egg laid on 4–5 prepared plates for no more than 3 h (at least 110 eggs for each strain and treatment). Development was monitored daily for 4 d for experiments conducted at 16°C and 3 d for experiments conducted at all other temperatures. Experiments monitoring development on E. coli OP50 were performed on 6 cm NGM plates. P. aeruginosa PA14 plates were prepared as described [27]. For the data presented in Figure S2, plates were prepared as described [37], except that ampicillin was used instead of carbenicillin.
To monitor L4 survival on E. coli OP50 or P. aeruginosa PA14, strains were incubated at 16°C to the L4 stage, when they were transferred to plates containing FUDR and incubated at either 16°C or 25°C. For each strain, 30 worms were transferred to each of 3–4 plates. Alive vs. dead worms were counted, and worms that died by exploding through the vulva or desiccating on the side of plates were censored.
|
10.1371/journal.ppat.1003468 | Ethnic Variation in Inflammatory Profile in Tuberculosis | Distinct phylogenetic lineages of Mycobacterium tuberculosis (MTB) cause disease in patients of particular genetic ancestry, and elicit different patterns of cytokine and chemokine secretion when cultured with human macrophages in vitro. Circulating and antigen-stimulated concentrations of these inflammatory mediators might therefore be expected to vary significantly between tuberculosis patients of different ethnic origin. Studies to characterise such variation, and to determine whether it relates to host or bacillary factors, have not been conducted. We therefore compared circulating and antigen-stimulated concentrations of 43 inflammatory mediators and 14 haematological parameters (inflammatory profile) in 45 pulmonary tuberculosis patients of African ancestry vs. 83 patients of Eurasian ancestry in London, UK, and investigated the influence of bacillary and host genotype on these profiles. Despite having similar demographic and clinical characteristics, patients of differing ancestry exhibited distinct inflammatory profiles at presentation: those of African ancestry had lower neutrophil counts, lower serum concentrations of CCL2, CCL11 and vitamin D binding protein (DBP) but higher serum CCL5 concentrations and higher antigen-stimulated IL-1 receptor antagonist and IL-12 secretion. These differences associated with ethnic variation in host DBP genotype, but not with ethnic variation in MTB strain. Ethnic differences in inflammatory profile became more marked following initiation of antimicrobial therapy, and immunological correlates of speed of elimination of MTB from the sputum differed between patients of African vs. Eurasian ancestry. Our study demonstrates a hitherto unappreciated degree of ethnic heterogeneity in inflammatory profile in tuberculosis patients that associates primarily with ethnic variation in host, rather than bacillary, genotype. Candidate immunodiagnostics and immunological biomarkers of response to antimicrobial therapy should be derived and validated in tuberculosis patients of different ethnic origin.
| Mycobacterium tuberculosis (MTB) is the causative agent of tuberculosis. Genetically distinct strains of MTB cause disease in particular ethnic groups, and these strains vary in their ability to elicit inflammatory responses from antigen-presenting cells in vitro. Circulating and antigen-stimulated concentrations of inflammatory mediators (‘inflammatory profile’) might therefore be expected to differ between tuberculosis patients of different ethnic origin; however, this question has not previously been addressed. We therefore conducted a study to characterise ethnic variation in inflammatory profiles in a cohort of 128 newly-diagnosed tuberculosis patients in London, UK. Patients of African vs. Eurasian ancestry had distinct inflammatory profiles at presentation; differences did not relate to MTB strain variation between groups, but they did associate with ethnic variation in host genotype. Moreover, immunological correlates of the rate of MTB clearance from sputum differed between patients of African vs. Eurasian ancestry. Our findings provide insight into the mechanisms underlying ethnic variation in inflammatory profile in tuberculosis patients, and indicate that candidate immunodiagnostics and immunological biomarkers of response to tuberculosis therapy should be derived and validated in tuberculosis patients of different ethnic origin.
| Mycobacterium tuberculosis (MTB), the causative agent of tuberculosis (TB), emerged as a pathogen in Africa and has co-evolved with humans following migration to Europe and Asia some 70,000 years ago [1]. Distinct phylogenetic lineages of MTB consistently associate with human populations of different genetic ancestry in a variety of settings [2]–[5] and elicit differing immune responses from antigen-presenting cells of healthy donors in vitro [6]–[11]. Antimycobacterial immune responses might therefore be expected to vary between TB patients of different ethnic origin; however, studies investigating this question have not been conducted. Demonstration of significant ethnic variation in inflammatory responses at presentation and after initiation of treatment would have implications for the development of immunodiagnostics and for the identification of surrogate endpoints for trials of antituberculous drugs.
We therefore conducted a study to characterise ethnic variation in circulating and antigen-stimulated concentrations of a panel of 43 soluble inflammatory mediators and 14 haematological parameters (collectively termed ‘inflammatory profile’) before and after intensive-phase antituberculous therapy in a multi-ethnic cohort of patients with pulmonary tuberculosis (PTB) who participated in a clinical trial of adjunctive vitamin D supplementation conducted in London, UK [12]. The primary comparison was between patients of African vs. Eurasian ancestry, on the grounds of the distinct genetic structure of these populations [13], and because TB patients of African ancestry are recognised to have delayed clearance of MTB from the sputum in comparison to non-African patients [14], [15] – a phenomenon that might be immunologically mediated_ENREF_17. We found that patients of African and Eurasian ancestry had significantly different inflammatory profiles at presentation, and that these differences associated primarily with variation in host, but not bacillary, genotype. Ethnic differences in inflammatory profile became more marked after intensive-phase treatment, and immunological correlates of time to sputum culture conversion between patients of African vs. Eurasian ancestry were distinct.
A total of 141 patients were eligible to participate in the study (Study Profile, Figure S1). Self-defined ethnic origin was used to attribute ancestry as African (n = 45), Eurasian (n = 83), East Asian (n = 9), Latin American (n = 3) or mixed (n = 1) according to Rosenberg's five-region classification [13]. Due to small numbers in other groups, analyses were confined to participants of African and Eurasian ancestry. These two groups had similar demographic and clinical characteristics at presentation, the only statistically significant difference being a slightly shorter duration of symptoms pre-diagnosis in patients of African vs. Eurasian ancestry (median 2.0 vs. 2.5 months respectively, p = 0.03; Table 1).
Forty-three soluble factors and 14 haematological parameters detailed in Table S1 were measured in samples of serum, plasma or whole blood taken at baseline. The median circulating concentrations of 7 soluble factors (Interleukin [IL]-2, IL-5, IL-13, IL-17, tumour necrosis factor [TNF], basic fibroblast growth factor [FGF-β] and matrix metalloproteinase-7 [MMP-7]) were below the limit of detection (LOD) at baseline and were excluded from further analyses; median values, ranges and LODs for these analytes at baseline are presented in Table S2.
The remaining 50 parameters were then analysed using the t-test for general linear models (GLM) with statistical adjustment for covariates with potential to influence inflammatory profile (age, sex, duration of symptoms pre-diagnosis, duration of antimicrobial therapy pre-sampling and baseline serum 25-hydroxyvitamin D [25(OH)D] concentration). Five parameters were identified as having concentrations which were significantly different (false discovery rate [q-value]≤0.05) between participants of African vs. Eurasian ancestry. Four (peripheral blood neutrophil count and serum concentrations of CC chemokine ligand [CCL] 2, CCL11 and vitamin D binding protein [DBP]) were lower in participants of African vs. Eurasian ancestry (p≤0.0018), and one (serum CCL5 concentration) was higher (p = 2.15×10−5; Table 2; Figure 1). These parameters were then assessed using principal component analysis (PCA), a well-established mathematical technique for reducing the dimensionality of complex datasets by transforming the data to a new coordinate system [16]. This provided a visual representation of how well the identified parameters differentiated individuals from the two ethnic groups (Figure 2).
Despite the relative homogeneity in genetic structure in populations of Asian and European ancestry [13], healthy Asians and Europeans have previously been reported to have differing inflammatory profiles that associate with increased risk of coronary heart disease [17]. In order to explore whether combining these groups concealed significant heterogeneity in inflammatory profile in patients with TB, we stratified the analysis above, subdividing the Eurasian group into European and Middle Eastern vs. Central and South Asian. The resultant PCA plot showed that inflammatory profiles of these two groups clustered together, and were separated from those of patients of African ancestry (Figure 2). In keeping with this observation, no significant differences in circulating concentrations of inflammatory mediators were found between Eurasian sub-groups (Figure 3). Our decision to combine data for Europeans and Asians in subsequent analyses was further justified by the finding that allele frequencies of two single nucleotide polymorphisms investigated in the DBP gene (rs 4588 and rs 7041) did not differ between participants of European/Middle Eastern vs. Central/South Asian ancestry (p≥0.32), but that they were different between Eurasians and Africans (p<0.001, Table 1)
In order to determine whether antigen-stimulated responses also differed between patients of African vs. Eurasian ancestry, whole blood samples taken from a sub-group of 42 patients (13 of African ancestry, and 29 of Eurasian ancestry) were stimulated ex vivo with the recombinant MTB antigen culture filtrate protein, 10 kDa (rCFP-10). The concentrations of 39 soluble factors listed in Table S1 were assayed in supernatants of whole blood samples taken at baseline and stimulated with rCFP-10 for 72 hours. The median concentrations of six soluble factors (IL-2, IL-5, IL-13, epidermal growth factor [EGF], FGF-β and MMP-7) were below the LOD at baseline and were excluded from further analyses; median values, ranges and LODs for these analytes are presented in Table S2. The remaining 33 parameters were analysed using the t-test for GLM with the same adjustment for covariates as conducted for circulating responses. Those that were different between groups were visualised by PCA. Two such parameters were found: antigen-stimulated concentrations of IL-1 receptor antagonist [IL-1RA] and IL-12 were both higher in participants of African vs. Eurasian ancestry (p≤0.0030; Table 2; Figure 1). As before, we conducted a sensitivity analysis to determine whether patients of European/Middle Eastern vs. Central/South Asian ancestry differed in their antigen-stimulated inflammatory profile: both the PCA plot (Figure 2) and scatter plots (Figure 3) showed similar patterns between these sub-groups. Moreover, conducting a t-test for GLM analysis did not identify any significant differences in inflammatory profile between the Eurasian sub-groups, further strengthening the rationale to pool data for patients of European/Middle Eastern and Central/South Asian ancestry together in subsequent analyses.
MTB has co-evolved with humans, and different bacillary strains associate with different ethnic groups [2]; moreover, MTB strains of different lineage elicit differing immune responses in vitro [6]–[11] _ENREF_8. Ethnic variation in inflammatory profile in PTB might therefore be explained by differential representation of MTB strain lineages between ethnic groups. To investigate this possibility, genetic lineages of isolates from sputum of study participants were determined using multilocus Mycobacterial Interspersed Repetitive Units – Variable Number of Tandem Repeats (MIRU-VNTR) analysis [18], and frequencies of isolates of different lineage were compared between patient groups. Isolates of Indo-Oceanic (Lineage 1), East Asian (Lineage 2) and East African-Indian (Lineage 3) lineages tended to be under-represented, and isolates of Euro-American (Lineage 4) lineage over-represented, in participants of African vs. Eurasian ancestry (Table 1; p = 0.08). We therefore repeated the analyses of ethnic differences in inflammatory profile above, including additional statistical adjustment for MTB strain lineage: the set of parameters identified as being significantly different between patients of African vs. Eurasian ancestry was unchanged (Table 2), suggesting that MTB strain lineage was not a determinant of ethnic differences in inflammatory profile that we had observed.
As a further test of the influence of MTB strain lineage on immune responses in the host, we conducted stratified analyses to compare inflammatory profiles associated with different MTB strain lineages in patients of African and Eurasian ancestry separately. Among patients of Eurasian ancestry, no statistically significant differences in either circulating or antigen-stimulated immune responses were observed between patients infected with organisms of different strain lineage. Among patients of African ancestry, serum concentrations of prostaglandin E2 (PGE2) were significantly lower in patients infected with MTB of East African-Indian lineage compared to those infected by other lineages (p = 0.0008; Figure 4), but no inter-lineage differences were seen for any other circulating parameter, or for any antigen-stimulated parameter investigated.
Since ethnic variation in the distribution of MTB strain lineages did not associate with differences in inflammatory profile observed between participants of African vs. Eurasian ancestry, we proceeded to investigate whether these differences might arise as a result of genetic variation in the host – a hypothesis suggested by results of human genome scans identifying chromosomal regions that influence immune responses to M. tuberculosis [19], [20]. To explore this possibility, we investigated two common functional single nucleotide polymorphisms in the DBP gene (rs4588 and rs7041), combinations of which form three haplotypes (Gc1F, Gc1S and Gc2). These polymorphisms were selected for investigation on the basis that they have been shown to influence antimycobacterial immune responses; that their frequency varies between people of African vs. Eurasian ancestry [21]; and that we had identified a significant difference in DBP concentration between ethnic groups. Rs4588 and rs7041 genotypes were determined, and haplotype frequencies were compared between ethnic groups: Gc1F carriers were over-represented, and Gc2 carriers under-represented, in patients of African vs. Eurasian ancestry (p<0.0001, Table 1). Moreover, serum DBP concentration in newly-diagnosed TB patients varied with DBP genotype, with those of Gc1F/1F genotype having the lowest concentrations and those with Gc1S/1S genotype having the highest concentrations, irrespective of ethnic group (p<0.0001 for comparison by genotype; p>0.05 for ethnic comparison within each genotype; Figure 5).
We therefore repeated the analysis of ethnic differences in inflammatory profiles, this time including statistical adjustment for DBP genotype in addition to the phenotypic characteristics previously incorporated in the model. Ethnic differences in neutrophil count, in serum DBP concentration, and in antigen-stimulated responses that had previously attained statistical significance in the original model were rendered non-significant by this adjustment (Table 2). The effect of the adjustment for DBP genotype is further illustrated in Figure 6, which shows a reduction in separation of samples from patients of African vs. Eurasian ancestry in a PCA plot after incorporation of DBP genotype in the model. We conclude that ethnic variation in DBP genotype associates with variation in inflammatory profiles observed between PTB patients of African vs. Eurasian ancestry.
We next proceeded to investigate whether ethnic differences in inflammatory profiles persisted after completion of intensive-phase antituberculous therapy. Concentrations of the same immunological parameters described above were assayed in samples of serum, plasma and whole blood taken after 8 weeks of antituberculous therapy from a cohort of 85 patients (30 of African ancestry and 55 of Eurasian ancestry) who fulfilled pre-defined criteria for inclusion in the per-protocol analysis of the clinical trial in which they were participating (Study Profile, Figure S1). Patients of different ethnic origin whose samples contributed to this analysis had similar demographic and clinical characteristics, the only statistically significant difference being a shorter duration of symptoms pre-diagnosis in patients of African vs. Eurasian ancestry (median 1.9 vs. 3.0 months respectively, p = 0.001). As before, parameters whose concentration was significantly different between participants of African vs. Eurasian ancestry were identified using the t-test for GLM, with adjustment for clinical and demographic covariates with potential to influence the effects of antimicrobial therapy on immune responses (age, sex, duration of symptoms pre-diagnosis, duration of antimicrobial therapy pre-sampling, isoniazid sensitivity vs. resistance, and allocation to vitamin D vs. placebo in trial). The effect of significant parameters was then assessed visually using PCA.
This analysis revealed that ethnic differences in neutrophil count and serum concentrations of CCL2, CCL5, CCL11 and DBP persisted at 8 weeks (p≤7.84×10−6, Figure 1) and that an additional parameter, serum C-X-C chemokine ligand 8 (CXCL8) concentration, was lower in participants of African vs. Eurasian ancestry at this time point (p = 0.0018; Table 2). PCA plots of circulating inflammatory parameters sampled at different time points show that samples from patients of African vs. Eurasian ancestry were more widely separated at 8 weeks compared to baseline (Figure 6), indicating that ethnic variation in circulating inflammatory profile was more marked at 8 weeks than at baseline. Ethnic variation in antigen-stimulated responses was also observed in 8-week samples, with supernatant concentrations of antigen-stimulated CCL11 and hepatic growth factor (HGF) being significantly lower in patients of African vs. Eurasian ancestry after completion of intensive-phase therapy (p≤0.0023; Table 2, Figure 1).
As an additional check to determine whether any of this variation could be attributed to the effects of adjunctive vitamin D supplementation, which we have previously shown to be immunomodulatory [22], we repeated the analyses above in the sub-group of 47 participants allocated to the placebo arm of the clinical trial in which they were participating. Near-identical results were obtained in this smaller cohort (Table S4, Figure S2), confirming that ethnic differences in 8-week inflammatory profile observed in the analysis of all participants did not arise as a result of confounding by differential allocation to vitamin D vs. placebo in patients of African vs. Eurasian ancestry.
Given that differences in immune response have been reported to associate with differences in microbiological clearance among patients with PTB [23]–[25], and that speed of sputum culture conversion has been reported to vary between TB patients of African vs. Eurasian ancestry [14], [15] we wished to determine whether ethnic differences in inflammatory response associated with variation in microbiological response to therapy. To this end, we compared time to sputum culture conversion between participants of African vs. Eurasian ancestry in our cohort, and found no significant difference (p = 0.41). Since ethnic differences in inflammatory profile persisted throughout intensive-phase treatment, we reasoned that profiles associated with fast vs. slow sputum culture conversion might therefore differ between ethnic groups. To test this hypothesis, we classified each participant for whom sputum culture conversion data were available (n = 82) according to their time to sputum culture conversion from positive to negative, denoting those with time greater than or equal to the median value of 37.25 days ‘slow converters’ (n = 41), and those with time less than this value ‘fast converters’ (n = 41). Clinical characteristics of patients having fast vs. slow sputum culture conversion, stratified by ethnicity, are compared in Table S3. Slow sputum culture conversion was associated with older age and higher baseline sputum bacillary load among Eurasians (p≤0.01); similar trends were seen among Africans (p≤0.45). We then compared inflammatory profiles measured during the course of intensive-phase therapy between fast and slow sputum culture converters for patients of African vs. Eurasian ancestry. The interaction analysis for ethnic group was conducted using rank regression on the interaction term ‘week of sampling*speed of sputum culture conversion’ with adjustment for the same covariates as for the analysis of 8-week samples above, plus week of sampling and subject ID. Baseline bacillary load was not adjusted for as this is likely to be a significant driver of differences in inflammatory profiles between fast vs. slow converters.
The kinetics of circulating inflammatory responses differed markedly between patients with fast vs. slow sputum culture conversion, and immunological correlates of speed of sputum culture conversion differed between patients of African vs. Eurasian ancestry. Circulating immunological correlates of fast vs. slow sputum culture conversion in patients of African vs. Eurasian origin are summarised schematically in Figure 7, and presented in detail in Table 3 and Figure 8. Of the 50 parameters investigated, 27 were associated with speed of sputum culture conversion in one or both ethnic groups. Twelve of these parameters associated with speed of sputum conversion in patients of Eurasian ancestry only; nine associated with speed of sputum culture conversion in patients of African ancestry only; four associated similarly with speed of response in both ethnic groups; and two were differentially associated with speed of sputum culture conversion in patients of African vs. Eurasian ancestry, i.e. elevated levels of these markers associated with slower sputum culture conversion in one ethnic group and faster conversion in another.
As a further step to validate our findings, we applied network PCA to the parameters listed in Table 3 in order to investigate the relationship between changes in inflammatory parameters observed during treatment (Figure 9). For both ethnic groups, acute phase reactants erythrocyte sedimentation rate (ESR) and C-reactive protein (CRP) were linked and red cell parameters were linked to each other. In the African network MMP-1 was linked to its inducer PGE2, and chemokines CCL2 and CCL11 were closely linked, while in the Eurasian network, monocyte and neutrophil counts were closely linked. These linkages are consistent with biological understanding of the regulation of these inflammatory mediators and cell populations, validating results of the PCA. Kinetics of antigen-stimulated responses were not compared between groups due to small numbers within each sub-group.
Clinically significant ethnic differences in immune responses to Plasmodium falciparum and human immunodeficiency virus have previously been described [26], [27], but to our knowledge, this study is the first to address the question of whether inflammatory responses vary between TB patients of different ethnic origin. We report that inflammatory profiles vary significantly between TB patients of African and Eurasian ancestry having similar clinical and demographic characteristics, and that these differences associate primarily with ethnic variation in host rather than bacillary genotype. We also show that ethnic differences in inflammatory profiles observed at presentation persist after completion of intensive-phase therapy, and that immunological correlates of speed of sputum bacillary clearance differ markedly between patients of African vs. Eurasian ancestry. These findings have important implications for the design of studies investigating immunological biomarkers of response to antituberculous therapy.
African patients living in Africa have previously been reported to have more extensive disease at diagnosis than Europeans living in Europe, and to have lower rates of sputum conversion after intensive-phase antimicrobial therapy [14], [15], but controversy remains as to whether this reflects ethnic variation in host-pathogen interactions or geographical variation in laboratory practice and/or access to effective therapy. In our study – where patients of different ethnic origin were recruited in a single city, and where all microbiological samples were analysed in a single laboratory – we observed no difference in rates of cavitation or 2-month sputum culture conversion between patients of African vs. Eurasian ancestry. Despite this, we did observe significant ethnic variation in inflammatory profile between groups. Many of these differences were associated with variation in host DBP genotype, supporting the findings of an in vitro study reporting that DBP has broad influences on the antimycobacterial response [28]. This is plausible, given that this protein modulates macrophage activation and neutrophil chemotaxis, as well as performing its classical role in transport of vitamin D metabolites in the circulation [29]. We also observed ethnic variation in circulating and/or antigen-stimulated concentrations of cytokines (IL-1RA, IL-12) and chemokines (CCL2, CCL5, CCL11, CXCL8), many of which play key roles in the antimycobacterial immune response. The genes encoding these mediators are all polymorphic, and in some cases, ethnic variation in frequency of alleles influencing antimycobacterial responses has been reported [30], [31]. Study of functional associations of polymorphisms in these genes might yield insights into the genetic basis for ethnic variation in immune responses to MTB. Further investigation in other populations is also required to validate the ethnic differences in inflammatory profile that we report, as the large number of analyses and relatively modest sample size of our study could have led to Type I and Type II errors regarding specific parameters. Nevertheless, our main conclusions regarding strong ethnic group differences appear solid given the highly statistically significant differences found after stringent adjustment for multiple comparisons.
In contrast to the variation in inflammatory response between patients of different DBP genotype, relatively little difference in circulating and antigen-stimulated responses was seen between individuals infected with MTB strains of different lineage when multivariate analysis of the full cohort of 128 patients was performed. Secondary stratified analyses within the two main ethnic groups were conducted as a ‘belt and braces’ validation, to ensure that multivariate analysis had been successful in adjusting for potential ethnicity-related confounders of the relationship between MTB strain and immune profile. This secondary analysis identified only one analyte which was affected by lineage, and only in one ethnic group. The fact that the main analysis and the validation analyses yielded the same result - i.e. minimal effect of MTB strain on immune profile - lends considerable weight to our conclusion that MTB strain is not a major determinant of immune profile in tuberculosis.
This finding complements that of Pareek and colleagues, who recently reported that ethnicity is a powerful determinant of clinical TB phenotype independently of mycobacterial lineage [32]. Other investigators have reported that ‘modern’ strains elicit lower inflammatory responses than ‘ancient’ strains in macrophages, but that no difference in responses was seen in peripheral blood leukocytes [11], the population of cells investigated here. Further study is required to determine whether macrophages isolated from TB patients of different ethnic origin vary in their response to different MTB strains. Nevertheless, our observation that ethnic differences in inflammatory profile persisted after the several log-fold reduction in bacillary load induced by intensive-phase therapy tends to support the hypothesis that host, rather than bacillary, factors are the major determinants of ethnic variation in inflammatory profile. Such variation in inflammatory responses to antimicrobial treatment might reflect ethnic differences in allele frequency of polymorphisms of drug transporter genes that have been shown to associate with pharmacokinetic response to rifampicin [33]. However, our observations that sputum conversion rates were similar in patients of African vs. Eurasian ancestry, and that ethnic differences in inflammatory responses post-therapy were qualitative rather than quantitative, does not support this hypothesis. It is more plausible that, as at baseline, ethnic differences in inflammatory profile after treatment represent ethnic variation in alleles encoding components of the inflammatory response. Such variation may have arisen as a result of differences in selective pressures on the immune response between populations that remained in Africa vs. those that migrated out of the continent some 70,000 years ago [1].
Whatever the underlying reasons for these differences, our observation that immunological correlates of fast vs. slow sputum culture conversion differ between patients of African vs. Eurasian ancestry has practical implications for the design of studies to identify immunological correlates of response to intensive-phase antituberculous therapy. Studies evaluating candidate biomarkers published to date have been relatively small, and have tended to investigate fewer parameters in smaller numbers of patients of homogeneous ancestry than in the current study. Our finding that high CRP and ESR associate with slow sputum culture conversion is in keeping with other reports [25], [34]. Larger studies are now needed; our findings indicate that the validity of candidate biomarkers of treatment response identified by such studies will need to be evaluated in patients of different ancestry, as the inflammatory response in TB is ethnically heterogeneous.
The patients included in this study were participants in the AdjuVIT study - a double-blind randomised placebo-controlled trial of high-dose vitamin D during intensive-phase antimicrobial treatment of pulmonary TB, conducted in London, UK. Recruitment commenced on January 25th 2007, and ended on July 3rd 2009. A detailed account of study design has previously been given [12]. Participants self-defined their ethnic origin using the UK Office of National Statistics classification [35] and this information was used to attribute ancestry into one of five groups: African, Eurasian (incorporating participants of European, Middle Eastern, Central or South Asian ethnic origin), East Asian, Oceanic and American [13]. Baseline assessment included collection of a sputum sample for microscopy and culture and a blood sample. Fresh whole blood was sent for determination of full blood count and ESR and ex vivo stimulation with a mycobacterial antigen as described below. Aliquots of serum, plasma and whole blood were also stored at −80°C until completion of the trial. Participants were reviewed at 14, 28, 42 and 56 days after starting antituberculous therapy to assess clinical status and to monitor for adverse events. Blood and sputum samples were collected at each time-point and processed as above. Full characterisation of inflammatory profile was performed in the sub-set of participants who fulfilled pre-defined criteria for per-protocol analysis (i.e. those infected with a rifampicin-sensitive isolate of M. tuberculosis who received at least three doses of study preparation, who were compliant with antituberculous therapy, who did not take second-line antituberculous therapy or oral corticosteroids, who completed all study visits and who were not HIV sero-positive). The study was approved by East London and The City Research Ethics Committee (ref 06/Q0605/83), and registered with ClinicalTrials.gov (NCT00419068). Written informed consent was obtained from all participants before enrolment.
For all participants recruited on or after May 15th 2008, fresh whole blood was diluted 1∶10 in RPMI 1640 medium (Sigma-Aldrich, Gillingham, UK) and duplicate 180 µl aliquots were stimulated in 96-well plates at 37°C in the presence of 5% CO2 with rCFP-10 (Rv3874, Proteix Biotechnologies, Vestek, Czech Republic; final concentration 2.5 µg/ml) or 2% bovine serum albumin in phosphate buffered saline (negative control). Plates were centrifuged after 72 hours' incubation, and cell-free supernatants were aspirated and frozen at −80°C pending immunological analysis. rCFP-10 was tested for presence of endotoxin: concentration was found to be 260 IU (EU)/mg, working concentration 63 pg/ml. Addition of this concentration of endotoxin to TB patients' whole blood in control experiments did not stimulate cytokine or chemokine secretion.
Immunological parameters were selected on the basis that they played a role in host defence against MTB and/or that they were recognised biomarkers of disease activity [36]. Concentrations of 43 soluble factors listed in Table S1 were determined in serum/plasma as follows. Serum CRP and albumin concentrations were assayed using an Architect ci8200 analyser (Abbott Diagnostics, Chicago, IL, USA). Serum concentrations of IL-1β, IL-1RA, IL-2, IL-2R, IL-4, IL-5, IL-6, IL-7, IL-10, IL-12 (p40/p70), IL-13, IL-15, IL-17, G-CSF, GM-CSF, IFN-α, IFN-γ, TNF, CXCL8, CXCL9, CXCL10, CCL2, CCL3, CCL4, CCL5, CCL11, EGF, FGF-β, HGF and vascular endothelial growth factor (VEGF) were quantified using a human 30-plex bead immunoassay panel (sensitivity [sens.] according to Lot #617361, Invitrogen, Camarillo, CA, USA). Serum samples required high dilution for accurate determination of CCL5 concentration and all were re-assayed using a single-plex bead assay (Invitrogen). Serum PGE2 concentration was analysed by high sensitivity competitive enzyme immunoassay (EIA; Assay Designs, Ann Arbo37.25r, MI, USA; sens. 13.4 pg/ml). Plasma concentrations of antimicrobial peptides (AMP) LL-37 (sens. 31 pg/ml), HNP1-3 (sens. 156 pg/ml) and NGAL (sens. 400 pg/ml) were analysed by ELISA (Hycult Biotechnology, Uden, The Netherlands). Plasma concentrations of MMP-1, -2, -3, -7 and -8 were determined by Fluorokine MAP multianlalyte profiling (sens. according to Lot #273379, R&D systems); plasma concentration of MMP-9 was determined by DuoSet ELISA (sens. 3 pg/ml, R&D systems). Serum concentration of DBP was determined by ELISA (sens. 0.65 ng/ml, R&D systems). Multi-plex bead assays were performed on a Luminex 200 anlayzer (Luminex Corporation, Austin, TX, USA). ELISA and EIA absorbances were measured using a Benchmark Plus microplate spectrophotometer (Bio-Rad Laboratories, Hertfordshire, UK). The concentrations of 39 of these analytes (all of the above except DBP, PGE2, CRP and albumin; listed in Table S1) were also determined in WBA supernatants. Antigen-stimulated AMP and MMP concentrations were corrected by subtraction of unstimulated values. For MMP-2, -3, -8 and HNP and NGAL, unstimulated values were generally greater than stimulated values and this was the case sometimes for MMP-9 and LL-37. Cytokine/chemokine values were generally undetectable in unstimulated samples and a correction was not applied.
Fourteen haematological parameters listed in Table S1 were also measured in fresh whole blood. Full blood counts were performed using a LH750 haematology analyser (Beckman Coulter, Brea, CA, USA). ESR was measured by the Wintrobe method using a s2000 analyser (Desaga, Wiseloch, Germany).
Human DNA was extracted from whole blood using the Promega Wizard SV 96 Genomic DNA Purification System on the Biomek FX robot (Beckman Coulter), quantified using the Nanodrop spectrophotometer and normalised to 5 ng/ml. 10 ng DNA was used as template for 5 ml pre-developed TaqMan assays (Applied Biosystems, Foster City, CA, USA) to type the StyI (rs4588) and HaeIII (rs7041) polymorphisms of the vitamin D binding protein. These assays were performed on the ABI 7900HT platform in 384-well format, and data were analysed with Autocaller software. DBP haplotypes were deduced from StyI and HaeIII genotypes as previously described [21]. Mycobacterial DNA was extracted and genotyped using automated 15 mycobacterial interspersed repetitive unit–VNTR as previously described [18].
Contingency tables were analysed using chi-square tests, unless more than 20% of cells in a table had an expected frequency of <5, when Fisher's exact tests were employed. Median serum DBP concentration was compared between groups using a Kruskal-Wallis test with Dunn's post hoc test to correct for multiple comparisons. Time to sputum culture conversion was compared between groups using a logrank test. Analyte concentrations were calculated from raw luminex, ELISA and EIA data using Masterplex ReaderFit software (Hitachi Solutions America, San Francisco, CA, USA) and these calculated values were plotted using GraphPad Prism 5 software (La Jolla, CA, USA). Linear modelling and PCA was conducted using Qlucore Omics Explorer 2.2 software (Qlucore AB, Lund, Sweden). Analyte concentrations were log2 converted and the variance was normalized to 1. For analytes that were undetectable in at least one sample, the ‘limit of detection’ value was added to every measured value for that analyte prior to log2 conversion. Missing values were imputed by K nearest neighbours (k-NN) [37]: for circulating parameters, 2.5% of data points were missing; for CFP-10-stimulated parameters, 3% of data points were missing. Parameters whose concentration differed significantly between patients of African vs. Eurasian ancestry were identified using the t-test for GLM with adjustment for covariates with potential to influence the inflammatory profile using the eliminated factors approach. This fits a multiple regression model to all covariates, and subtracts the expression values predicted by this model from the observed values in order to remove covariate effects between patients [38]. An F-test for GLM with adjustment for covariates with potential to influence the inflammatory profile was performed to identify parameters whose concentration varied according to MTB strain lineage within each ethnic group. Parameters associating with slow vs. fast sputum culture conversion within each ethnic group were identified by rank regression analysis on the interaction term ‘week of sampling*speed of sputum culture conversion’ with adjustment for covariates with potential to influence the inflammatory profile, week of sampling (to correct for effects of treatment duration alone) and subject ID (to correct for repeated measures). Rank regression was conducted by replacing the ordinal interaction categorical predictors with numerical predictors, followed by a normal linear regression. Samples were ordered alternately fast, then slow, with increasing time since treatment initiation [39].
These analyses yield t-statistics (calculated as the regression co-efficient for each parameter divided by its standard deviation) representing the magnitude of difference in concentration of a given parameter between groups being compared; p values, representing the probability that such differences could have arisen by chance alone; and q values, which define the lowest false discovery rate (FDR) for which the hypothesis would be accepted under the Benjamini-Hochberg procedure for multiple testing correction [40]. Thresholds of 0.05 were applied for p and q values throughout.
PCA networks were created using one connection, i.e. by connecting each analyte to the other analyte that it shares the most similar pattern of change with over time; the distance between analytes in the network represents their Pearson correlation coefficients. Points in the network are coloured according to the value of the R-statistic generated for each analyte from the rank regression interaction analysis, which identified variables that had a significantly different pattern of change between slow and fast responders over time. The value of the R-statistic indicates the proportion of the total variation of that variable which is explained by the model tested. It is calculated as the square root of the R2-statistic, and the sign indicates the direction of the observed effect. A positive R-statistic indicates a higher concentration of that analyte in slow vs. fast culture converters, and vice versa.
|
10.1371/journal.ppat.1005737 | Potent Allosteric Dengue Virus NS5 Polymerase Inhibitors: Mechanism of Action and Resistance Profiling | Flaviviruses comprise major emerging pathogens such as dengue virus (DENV) or Zika virus (ZIKV). The flavivirus RNA genome is replicated by the RNA-dependent-RNA polymerase (RdRp) domain of non-structural protein 5 (NS5). This essential enzymatic activity renders the RdRp attractive for antiviral therapy. NS5 synthesizes viral RNA via a “de novo” initiation mechanism. Crystal structures of the flavivirus RdRp revealed a “closed” conformation reminiscent of a pre-initiation state, with a well ordered priming loop that extrudes from the thumb subdomain into the dsRNA exit tunnel, close to the “GDD” active site. To-date, no allosteric pockets have been identified for the RdRp, and compound screening campaigns did not yield suitable drug candidates. Using fragment-based screening via X-ray crystallography, we found a fragment that bound to a pocket of the apo-DENV RdRp close to its active site (termed “N pocket”). Structure-guided improvements yielded DENV pan-serotype inhibitors of the RdRp de novo initiation activity with nano-molar potency that also impeded elongation activity at micro-molar concentrations. Inhibitors exhibited mixed inhibition kinetics with respect to competition with the RNA or GTP substrate. The best compounds have EC50 values of 1–2 μM against all four DENV serotypes in cell culture assays. Genome-sequencing of compound-resistant DENV replicons, identified amino acid changes that mapped to the N pocket. Since inhibitors bind at the thumb/palm interface of the RdRp, this class of compounds is proposed to hinder RdRp conformational changes during its transition from initiation to elongation. This is the first report of a class of pan-serotype and cell-active DENV RdRp inhibitors. Given the evolutionary conservation of residues lining the N pocket, these molecules offer insights to treat other serious conditions caused by flaviviruses.
| Dengue virus (DENV) is the world’s most prevalent mosquito-borne viral disease and nearly 40% of the world’s population is at risk of infection. Currently, no specific drugs are available to treat dengue or other flaviviral diseases. DENV NS5 is a large protein of 900 amino acids composed of two domains with key enzymatic activities for viral RNA replication in the host cell and constitutes a prime target for the design of anti-viral inhibitors. We performed a fragment-based screening by X-ray crystallography targeting the DENV NS5 polymerase and identified an allosteric binding pocket at the base of the thumb subdomain close to the enzyme active site. Potent inhibitors active in both DENV polymerase biochemical and cell-based assays were developed through structure-guided design. Resistant virus replicons grown in the presence of the inhibitor, harbored amino acid changes that mapped to the compound binding site. The proposed mode of action for this class of inhibitors is by impeding RdRp protein conformational changes during the transition from initiation to elongation phase of enzyme activity.
| Several flaviviruses, such as DENV, Japanese Encephalitis virus (JEV), West Nile virus (WNV), Yellow Fever virus (YFV) or Tick-borne encephalitis virus (TBEV) are major human pathogens, whilst Zika (ZIKV) is an emerging flavivirus of global significance causing severe neurological conditions in infected adults and newborn babies, most likely by mother-to-child transmission [1]. The mosquito-borne DENV causes widespread epidemics in over 100 countries, with ∼390 million infections each year [2]. Infection by any of the four DENV serotypes can lead to several outcomes, ranging from asymptomatic infection, dengue fever, to dengue hemorrhagic fever and dengue shock syndrome. After several decades of efforts, the first vaccine was recently licensed for use, but confers only partial cross protection for the four DENV serotypes [3, 4]. No antivirals have been approved to treat dengue or other flaviviral diseases [5].
Flavivirus RNA replication occurs in host cells on endoplasmic reticulum-derived membranes within a multi-protein replication complex (RC) consisting of viral NS proteins and host cofactors [6–8]. Comprising 900 amino acid residues, NS5 is the largest and most conserved protein component of the flavivirus RC. Its N-terminal domain (residues 1–265 in DENV3) is an S-adenosyl-L-methionine (SAM)-dependent methyltransferase (MTase) that methylates the viral RNA genome cap [9–15]. A guanylyltransferase activity was also proposed for the N-terminal domain of NS5 [16, 17]. Its C-terminal RdRp domain (residues 267–900) synthesizes the viral genomic RNA [18–22]. A potentially flexible linker region that connects the two catalytic domains of NS5 regulates RdRp activities and virus replication by modulating MTase-RdRp interactions [23–25]. In addition to its enzymatic functions, NS5 inhibits host interferon-mediated signaling by promoting degradation of STAT2 [26]. In DENV, NS5 localizes to the nucleus of infected cells in a serotype-dependent manner that modulates host processes [27].
Following DENV infection, the RdRp synthesizes viral RNA in the absence of a primer strand, via a de novo initiation mechanism, in which the (+) strand viral RNA template is transcribed into a complementary RNA strand of (-) polarity [18, 19]. This duplex in turn serves as a template for synthesis of additional RNA strands of (+) polarity that either act as mRNA for protein translation or are packaged into virions. DENV RdRp possesses a right hand-like architecture conserved across different polymerase families [21, 22, 25], with three subdomains termed ‘‘fingers”, ‘‘palm” and ‘‘thumb”. Within these subdomains, seven conserved amino-acid sequence motifs play key roles for binding RNA, NTPs and metal-ions and for catalysis [28, 29]. Structures of the apo-DENV RdRp were found to adopt a “closed” pre-initiation state conformation, with a well-ordered priming loop projecting into a narrow RNA binding tunnel. Disordered peptide segments were observed in motifs F, G and at the C-terminal end [21, 22, 25].
The importance of NS5 for viral replication makes it an ideal target for developing inhibitors to treat diseases caused by flaviviruses [30–32]. Although several high-throughput screening campaigns have been performed, only a few DENV RdRp non-nucleoside inhibitors have been described [33–36]. From these latter efforts, we previously identified two compounds that bind to the RNA tunnel but did not succeed in improving their lead-like properties [34, 35]. Here, using fragment-based screening via X-ray crystallography targeting the apo-DENV RdRp, we identified a fragment that bound to a pocket located in the thumb subdomain, close to the enzyme active site, which we term as the “N pocket” [37, 38]. Using a structure-guided approach that combines biochemical, biophysical and cell-based assays, we designed potent inhibitors that bound to this allosteric site, and inhibited DENV1-4 viral replication across various cell-based assays. Resistant DENV replicons with amino acid changes in the “N” pocket were raised with two compounds, confirming that the NS5 polymerase was the specific target for this class of inhibitors in DENV infected cells. To our knowledge, this is the first report of a Flavivirus RdRp allosteric pocket and the successful use of structure-guided approach for designing potent inhibitors targeting NS5. This work has major implications for the design of much-needed flavivirus anti-viral inhibitors.
Following fragment-based screening using X-ray crystallography, we identified 3, a bi-phenyl acetic acid fragment, that bound to a pocket in the DENV3 RdRp thumb subdomain (IC50≈ 734 μM; Fig 1 and Table 1; 37). Iterative rounds of structure-guided design led to compounds that inhibited both DENV polymerase activity and viral replication in cells (Fig 1 and Table 1; Fig 1A and 1B in S1 Text). Firstly, switching the distal unsubstituted phenyl ring in 3, with a thiophene ring (3i) improved compound potency by >12-fold in DENV1-4 polymerase de novo initiation (dnI) enzyme assays [38, 39]. Substitution of the methoxyl group on the outer phenyl ring with a second acid moiety increased potency in DENV-1 and -3 (compare 3i and 11). Replacement of the chloro-substituent on the thiophene ring in 11, with a propargyl alcohol, markedly increased compound inhibitory property. Compound 15, which bears this moiety, was >16-fold more active across DENV1-4 enzymes. Whilst subsequent derivatives, exemplified by compound 15, displayed low nano-molar potencies across DENV1-4 dnI polymerase assays, they failed to inhibit DENV replication in cells. This is probably due to unfavorable physicochemical properties that limited their cell permeability (likely due to the presence of bis-carboxylic acid groups in 15).
Successive design strategies produced compounds with acyl-sulfonamide derivatives (replacing the charged acid groups with the acyl-sulfonamide bio-isosteres increases lipophilicity) with EC50 ≥ 2 μM, in a HuH-7 DENV2 replicon cell-based assay (Fig 1 and Table 1; Fig 1B in S1 Text). The most active compounds in this series, 29 and 29i, bear the 8-quinolinol moiety, and demonstrated IC50 values ranging from 0.013 to 0.074 μM across DENV1-4 polymerase, with EC50 value of ~2 μM in the DENV2 replicon cell-based assay (Table 1).
To better understand the inhibition mode of this class of compounds, order-of-reagent addition experiments were performed using the DENV dnI FAPA assay (Table 2). The standard assay format involved compounds exposed to enzyme alone followed by reaction initiations with ssRNA template and NTPs [39]. In the first experiments, compounds 15, 27 and 29 were exposed to pre-formed enzyme-ssRNA complexes, followed by reaction initiation with NTPs. IC50 values generated for 15 and 27 were similar to the standard assay format, suggesting that these compounds do not discriminate between the apo-enzyme and the polymerase bound to ssRNA. Compound 29, showed about 3-fold reduction in potency. Next, compounds were exposed to elongated enzyme-dsRNA complexes, in which the active site was occupied by the ssRNA template and newly synthesized short RNA products AGAA or AGAACC. Resulting compound inhibitory potencies dropped by 8–15 fold. The change was most pronounced in compound 29 (10–15 fold decline). These findings imply that during transition from initiation to the elongation complex, to accommodate the growing dsRNA product, the N-pocket underwent conformational changes, leading to decrease in compound binding affinities. To verify these findings, we tested the compounds in the DENV elongation FAPA assay by exposing the compounds to enzyme alone, followed by reaction initiation with duplex hetero-polymeric RNA templates [25]. Similarly, compound potencies were markedly reduced. Their IC50 values were 10–23 folds weaker than in the standard dnI assay, with 27 showing the greatest change in potency. Nevertheless, compound 29 retained potent inhibitory activity, with IC50 values ranging from 0.023–0.427 μM across the different enzyme assays and formats. Control 3’dGTP showed similar IC50 values in order-of-addition reagent experiments and in the DENV elongation FAPA assay.
We proceeded to characterize the inhibition kinetics of compounds 15 and 29, in the dnI FAPA assay, using 3’dGTP, as a control (Fig 2; Fig 2 in S1 Text). As expected, kinetics studies using Lineweaver-Burk plots showed that 3’dGTP was a competitive inhibitor of GTP, but a non-competitive inhibitor of the viral RNA substrate. Both 15 and 29 exhibited uncompetitive inhibition profiles with respect to the viral ssRNA template. Results from kinetic competition experiments with GTP were more complex. Lineweaver-Burk plots of 15 and 29 were indicative of uncompetitive inhibition. However, at high GTP concentrations, a non-competitive mode of inhibition by these compounds was apparent. Both de novo initiation and elongation activities occur in the DENV polymerase dnI assay. For the rate-limiting de novo RNA synthesis step, the Km for GTP was found to be >20 μM [19], whilst a low Km value (0.2–0.4 μM; 18, 39) was reported for the elongation phase. It is possible that the mixed inhibition profiles for both compounds, reflect differential effects on the dnI and elongation phases of the enzyme activities. Indeed, the significantly weaker inhibitory capabilities of these compounds in the DENV elongation assay support this hypothesis. Furthermore, both compounds are not true un- or non-competitive inhibitors as they are also able to bind to the apo-enzyme with high affinity (see below).
DENV3 RdRp co-crystal structures with 27 and 29 solved to 2.45 Å and 1.88 Å resolution respectively (Table 3), show that the compounds occupy about 60% of the N-pocket volume and establish multiple polar contacts with several neighboring amino acid residues (Fig 3). The RdRp retains essentially the same structure as in its unbound form [21, 22] with RMSD of 0.25 Å for 612 superimposed α-carbon atoms (the RMSD is 0.18 Å between the two complexes). The compound binding mode is reminiscent of other closely-related analogs [38] with complete overlap in the positions of their most buried moieties: the thiophene ring and propargyl alcohol, whilst acyl-sulfonamide and the solvent-exposed ring: 8-quinolinol ring in 29 and methoxy-substituted phenol ring in 27, adopt different orientations. The sulfur of the thiophene ring makes a non-covalent interaction with the side-chain hydroxyl of S796, whilst the terminal propargyl alcohol arm extends deeply into a tunnel lined by residues W803, M761, and M765. Its terminal hydroxyl group forms H-bond interactions with the backbone amide of H800 and the side-chain of Q802, and displaced a single buried water molecule present in the RdRp apo structure [21, 22]. In addition, the acyl-sulfonamide carbonyl moiety forms hydrogen bonds with the side chain of T794, and additionally in 27, with the backbone amide of W795 via an intercalated water molecule. Co-crystallization of compounds 27 and 29 with DENV3 FL NS5 led to the same binding mode as observed for the polymerase domain (Fig 3 and Table 1 in S1 Text). Soaking of compound 27 in crystals of DENV2 (NGC strain) RdRp domain also generated the same binding mode, with the OH- moiety of the propargyl arm forming similar hydrogen bonds with residues K800 (backbone N) and E802 (carboxylic acid side chain), as H800 and Q802 in the DENV3 RdRp-29 co-crystal structure (Fig 3D and 3E).
The 10-fold higher binding affinity of 29 over 27 for DENV RdRp, observed in SPR analyses (Table 2) can be accounted for by formation of three additional hydrogen-bonds between the 8-quinolinol ring of 29 with the side chain of R729 (these favorable contacts are absent in 27 with the corresponding ring pointing towards the solvent away from R729). Thermo-denaturation studies using recombinant DENV FL NS5, RdRp and FL NS5 from DENV-replicon lysates further corroborated these findings. In these experiments, 29 consistently stabilized DENV polymerase better than 27, leading to 2.5–6 °C better increase in protein melting temperatures compared to 27 (Table 2; Fig 5 in S1 Text). Compound binding fits to a simple 1:1 binding model in SPR analyses, correlating well with the X-ray crystallography data (Fig 3F).
To assess the functional relevance of the N-pocket for DENV polymerase activity, we targeted RdRp residues interacting with 27 or 29 as well as residues lining the N-pocket (Fig 3G), and measured both de novo initiation (dnI) and elongation activities of the corresponding RdRp Ala mutants. All mutant proteins studied have similar melting temperatures as WT, indicating that stabilities of the protein structures were not compromised by the alanine substitutions (Table 4). Overall, the results indicate that N pocket residues play an important role in DENV polymerase dnI and have less impact on elongation. This is particularly evident in the S710A and R737A mutants, where dnI activities were substantially reduced, to 26.6 and 0%, respectively, compared to WT, whilst retaining about 72% elongation activity. Both residues are completely conserved across the Flavivirus family (Fig 3G). Thus, the N pocket conformation observed in the inhibitor-bound crystal structures is likely to correspond to the structural state adopted by the DENV RdRp during dnI [21, 22].
To further validate the mode of binding of this class of inhibitors in the N-pocket, we performed inhibition assays using mutant proteins S796A and W803A, both of which retain about 66% de novo initiation activity (Table 5). Residue S796 interacts with the sulfur-atom in the thiophene ring whilst W803A lines the propargyl alcohol tunnel. As a control, we used 3’dGTP, which retained the same IC50 when tested on the mutant enzymes. All four compounds exhibited more than 10 fold increases in their IC50 values, when assayed with S796A and W803A mutant RdRp. Compounds 27 and 29 gave the greatest IC50 shifts when tested with mutant W803A (107- and 70-fold respectively). In agreement with the X-ray crystallography data, 11, which bears a–Cl substituent instead of the extended propargyl alcohol arm on the thiophene ring, showed only 9–10 fold increase. Taken together these biochemical studies substantiate the binding modes observed in the X-ray co-crystal structures.
We next investigated the biological effects of alanine mutation of residues S710, R729, R737, Y766, T794, S796, H800, Q803 or W803 in the context of a DENV4 replicon (Fig 4, Table 4). Following electroporation into BHK-21 cells, WT replicon replicated robustly, generating renilla luciferase signals that were 422-fold above background levels (measured at 24 hr post-electroporation). Its growth rate subsequently plateaued at 48 hr (605-fold above background). Thereafter, luciferase levels dropped 25-fold at 72 hr post-transfection. In comparison, mutant replicons bearing R729A or R737A substitutions were non-replicative, a result that is in good agreement with their poor in vitro NS5 polymerase activity profiles (Table 4). R729A RdRp exhibited about 30–40% of both dnI and elongation activities whilst dnI activity of R737A was completely abolished. Mutant replicons S710A, Y766A, and W803A were also non-replicative, despite showing 27–98% de novo initiation and elongation activities in vitro. Mutant replicons H800A and Q802A were poorly replicative, although their polymerase activities were only moderately decreased in the enzyme assays. In contrast, mutant replicons with T794A, S796A with similar reductions in polymerase activities, were more replicative, and continued to expand during the three-day incubation period. At 24 hr post-electroporation, their luciferase activities were respectively 838- and 456-folds poorer than WT replicon activity. By day 3, the difference had narrowed to 4.5- and 5.1-fold, respectively, lower than WT replicon activity. The reason for the difference in the replicative profiles of these latter four DENV4 mutants is uncertain. It is possible that within the context of the replicative complex, mutating these residues produced subtle differential effects on NS5 polymerase activity that translate in large variation in terms of virus replicon fitness.
Using DENV cell-based assays, we next examined the inhibitory properties of four potent compounds in the series (Table 6). Compounds 27, 29 and 29i demonstrated similar EC50 values ranging from 1–5 μM in DENV2 replicon cells from HuH-7, BHK-21 and A549 backgrounds while 26i was slightly less active (EC50 = 12–15 μM). When inhibition studies were performed with clinical isolates of DENV1-4, 27 inhibited all four serotypes with EC50 ~2 μM. Compound 29i was the next most potent compound with EC50 values of 1–6 μM. Despite demonstrating significantly higher potency against DENV1-4 RdRp in the dnI enzymatic assays, 29 displayed relatively lower cellular inhibition (EC50 = 4–13 μM), particularly with DENV-2 and -4 (EC50 = 14.1 μM and 10.2 μM respectively). Similar to the observation with replicon cells, 26i was also the least active against DENV1-4 viruses. All four inhibitors were inactive against HCV replicon and human rhinoviruses (with EC50 values >25 and >50 μM. respectively). No cytotoxicity was observed in five different cell lines tested (EC50 >50 μM).
Both 27 and 29i have an extra methoxy moiety on their central phenyl ring compared to 26i and 29. The inhibition results indicate that the additional methoxy group present in the former compounds may be advantageous for inhibiting infectious DENV as they are consistently more active than the cognate analogs devoid of this moiety (26i and 29 respectively). It is possible that this additional substituent allows for the formation of an intra-hydrogen bond with the N-atom of the sulfonamide linker facilitating better cell permeability. Notably, whilst EC50 values of 29i with DENV-1, -3 and -4 were comparable with 27, its DENV2 EC50 value was about two-fold lower.
To confirm that the antiviral activity displayed by this series of compounds was due to the specific inhibition of RdRp, we raised resistant DENV2 EGFP-replicons using compounds 27 and 29 (Table 6). We first propagated DENV2-NGC EGFP replicon cells in 20 μM of 29 (≈1X EC90 value) and increased the compound concentration to 25 μM after 5 weeks. RNA was sequenced from individual colonies of resistant cells that grew in the latter compound concentration, as well as from a mixed population of cells kept in 20 μM of 29 (Fig 5). Two individual 29-resistant replicon clones harbored the same single nucleotide change in NS5 (GAA→GAC), resulting in E802D mutation (note that residue 802 is E in DENV2-NGC and Q in DENV3 used for structure determination). A third clone contained another single nucleotide change in NS5 (CTG→GTG), resulting in L511V mutation. A fourth clone contained a mixed profile in NS5, in the same position, with both the WT nucleotide (G) as well as mutation to C nucleotide present (GTG→G/CTG), giving rise to partial L511V mutation.
Similar attempts to raise 27-resistant cells by exposure to high concentrations of 27 (14–20 μM; ≈2X EC90 value) were not successful. We then exposed the DENV2-NGC replicon cells with gradually increasing concentrations of 27 (starting from 1.5 μM; ≈0.5X EC50). After about 6 weeks, cells in 28 μM of 27 propagated robustly, at similar rates to WT replicon cells. Further increases in compound concentrations resulted in cell death and no individual resistant colonies were obtained. RNA was sequenced from a mixed population of replicon cells maintained in 28 μM of 27. A partial E802D mutation profile (GAA→GAA/T) was observed (Fig 5).
The crystal structure of DENV3-RdRp bound to 29 (Fig 3C) shows that the polar side chain of residue Q802 (E802 in DENV2) hydrogens bond with the hydroxyl group of the propargyl alcohol of 29. E802D mutation results in the shortening of the amino acid side-chain by one methyl group and is likely to disrupt this H-bond formation. Residue L511 (in DENV-2 and -3) forms van der Waals interactions with the thiophene ring of 29. In this case, loss of a methyl group in L511V is likely to weaken the interaction with the thiophene ring of the inhibitor. As a result these mutations lower the binding affinity of 29 in the N pocket. These findings thus provide compelling evidence that compounds 27 and 29 inhibit DENV replication in cells by binding to the N-pocket in the DENV polymerase.
To better understand the molecular mechanism of resistance caused by amino acid changes observed in the DENV RdRp, we generated RdRp mutants bearing these amino acid changes both in serotype DENV2 (L511V and E802D) and DENV4 (L512V and Q803N). Both the single and double mutant NS5 proteins have similar thermo-stability as WT protein and comparable dnI and elongation activities (Figs 6 and 7 in S1 Text). We then examined the impact of these mutations on the inhibitory capabilities of compounds 27 and 29 (Table 7). Both compounds were significantly less active against mutant enzymes than WT protein, in the dnI FAPA assay: IC50 value of 29 declined by 4–12 fold in DENV2 and DENV4, single mutants, whilst potency was further reduced in double mutant enzymes, by 52–133 fold lower than WT enzyme. For compound 27, IC50 values dropped by 5-88-fold in DENV2 single and double mutants. Changes in potency against DENV4 single and double mutants were even more severe: a complete loss of inhibitory activity (IC50 >20 μM) was observed. Furthermore, these compounds were less effective in stabilizing the mutant enzymes compared to the corresponding WT proteins (Fig 7 in S1 Text). Taken together, both in vitro enzyme profiling and thermo-denaturation studies strongly corroborate the resistant replicon phenotype obtained with 29.
To evaluate the impact of L511V and E802D mutations on DENV replication, we introduced single (L511V or E802D) and double (L511V/E802D) amino acid changes into the DENV2 (strain NGC) replicon and its infectious full length virus genome. After electroporation into BHK-21 cells, replications of replicons (measured by renilla luciferase activity) or virus (measured by plaque assays) in the absence of compound, were monitored for 4 days (Fig 6; Table A in S1 Text). Electroporated cells harboring WT and mutant replicons showed similar multiplication rates and viability (Fig 6A). Compared to WT DENV2 replicon, all three mutant replicons replicated faster and generated higher levels of renilla luciferase signals (Fig 6B) as well as viral RNA (Fig 6C; Fig 8A in S1 Text) and NS5 levels (Fig 8A in S1 Text). Luciferase levels peaked at day 1 for mutants L511V and L511V/E802D, and at day 2 for WT and mutant E802D.
Experiments conducted with infectious WT and mutant DENV2 showed a different profile. Viral titers increased steadily from days 1–4 post-electroporation, unlike replicon growth curves (Fig 6D and 6E; Table 2B in S1 Text). L511V mutant produced the least infectious virus particles, compared to the other three viruses. Immunofluorescence staining of intracellular viral RNA and NS5 also revealed highest NS5 and dsRNA levels in mutant L511V (Fig 8B in S1 Text). The reason for the difference between extra- and intra-cellular viral RNA levels of mutants L511V is unclear. It is possible that mutation of this residue may have different impact on the replicon and virus.
Next, we examined the inhibitory effects of compounds 27 and 29 on the DENV2 single and double mutant replicons and viruses (Table 8). EC50 value of 29 was reduced by 3-6-folds in single and double mutant DENV2 replicons, compared to WT replicon. Similarly, its potency was also reduced by 5-6-folds in virus mutants compared against WT virus. These data further verify that the anti-DENV properties of 29 function through binding to the N-pocket in NS5 polymerase in DENV-infected cells.
Potency reduction of compound 27 was less pronounced. Its EC50 values were reduced by 2-4-folds in mutant DENV2 replicons and viruses. The observed weaker EC50 shifts for 27 are puzzling as its binding mode is similar to 29 and involves non-covalent interaction of the thiophene ring with L511 and H-bond formation between the propargyl alcohol and E802D (Fig 3). Additional 27-resistant DENV2 EGFP-replicons were raised and studied. EC50 values of compounds 27 and 29 shifted by 17- and 10-folds, respectively, in these cells, compared to control cells raised in DMSO. Full replicon genome sequence analyses revealed secondary mutations present in NS5 methyl-transferase and NS4B in the 27-resistant replicon cells. Further reverse genetics on DENV2 with these amino acids are ongoing to better understand their roles in overcoming 27-mediated DENV2 growth inhibition.
In this report, we characterized a novel allosteric pocket at the interface of the thumb and palm subdomains of DENV RdRp [21, 22]. This binding site, which we termed the “N pocket”, was found through a fragment-based screening approach, by X-ray crystallography, using the DENV3 apo-RdRp protein as a target [37]. It is located near the priming loop (aa782-809) of the enzyme and is lined by residues highly conserved across DENV1-4, as well as in other flaviviruses including ZIKV (Fig 3G). Alanine substitutions, demonstrated that several N-pocket residues are important for NS5 polymerase de novo initiation activity and also for virus replication. Accordingly, N-pocket inhibitors generated by rational design potently inhibited DENV1-4 polymerase de novo initiation activities and virus replication in various cell types. They bind with strong affinity to recombinant apo-enzyme as well as FL NS5 from DENV replicon cell lysates.
Compound 29, one of the most potent compounds in the series, binds DENV RdRp with single-digit nano-molar affinity and stabilizes the RdRp melting temperature by 7.5–14 °C. It inhibits de novo initiation activity of DENV1-4 polymerases with IC50 values ranging from 13 to 38 nM. Alanine substitutions of N-pocket residues diminished the inhibitory properties of this class of compounds. Resistant DENV raised against compound 29, harbored amino acid mutations (L511V and E802D; DENV2 numbering) that mapped to the N-pocket. Correspondingly, these amino acid alterations reduced compound potencies in DENV cell-based and RdRp enzyme assays.
Residue L511 is conserved across DENV1-4, WNV and YFV. Residues 800 (H) and 802 (Q) are conserved across DENV-1, -3 and -4 but not in DENV2 (Fig 3G). In the laboratory adapted DENV2 strain, NGC, these residues are respectively, K and E, whilst in the DENV2 clinical isolate, MY097-10340, they are T and E. Crystal structure of DENV2-NGC RdRp bound with 27 showed that the OH- moiety of the propargyl alcohol arm, made similar hydrogen bonds with residues K800 (backbone N) and E802 (carboxylic acid side chain), as H800 and Q802 in the DENV3 RdRp-27 co-crystal structure. Residue T800 in DENV2, MY097-10340, would also be expected to form the same interaction as H800 or K800.
Kinetic studies showed that N pocket inhibitors have a mixed inhibition profile in the de novo initiation assay. Competitive experiments performed with GTP suggest differential inhibitory modes (non- and un-competitive) during initiation and elongation phases. Indeed, whilst compounds such as 27 and 29 have nano-molar IC50 values in the de novo initiation assay, their inhibitory potencies drop dramatically by 10–23 fold in the elongation assay (IC50 = 5.5 and 0.43 μM respectively). Given that both de novo initiation and elongation events occur simultaneously in DENV-infected cells, we speculate that N pocket inhibitors block the first activity better than the second, giving rise to the observed DENV cell-based EC50 values (example 29; EC50 = 2–14 μM). Order-of-reagent addition experiments further corroborate this hypothesis. Compound potencies are reduced only when the enzyme is occupied with newly synthesized duplex RNA, and not by single-stranded viral RNA. Presumably, retraction of the priming loop (aa782-809) from the active site during enzyme elongation alters the conformation of the N-pocket, leading to weaker binding affinities of the RdRp for the compounds.
DENV RdRp N-pocket compounds discovered here, share some common features with Site III non-nucleoside inhibitors described earlier for the HCV polymerase [29]. Site II (thumb 1), III (palm 1) and IV (palm 2) HCV RdRp inhibitors are de novo initiation inhibitors that lock the thumb subdomain in a conformation that prevents de novo initiation. Several such HCV inhibitors possess sub-micromolar EC50 values in HCV replicon cell-based assays and progressed into late phase clinical trials [44]. Of note, Dasabuvir (ABT 333), a Site III inhibitor, has recently been approved for HCV therapy in combination with NS3/4A protease and NS5 inhibitors (http://hepatitiscnewdrugs.blogspot.sg/2015/07/fda-hepatitis-update-approval-of_24.html). HCV RdRp site III inhibitors bind at the interface of the thumb and palm subdomains, with one side comprising the “primer grip” and the opposite side formed by the β-hairpin loop from the thumb (equivalent to the priming loop in DENV RdRp). Inhibitor binding is promoted by interactions with both sides, in particular, with Y448 from the β-loop. The initiating nucleotide GTP, also binds to this site and forms key interactions with R386, S387 and R394 from HCV NS5B.
DENV RdRp N-pocket inhibitors also form several hydrogen bonds with residues from the priming loop that project from the thumb domain (aa794-802). Additionally, residues S710, R729 and R737 collectively form the mouth of the N-pocket and interact with the acyl-sulfonamide group from this class of inhibitors. This region of the binding pocket corresponds to the proposed i-1 site observed in other Flavivirus RdRps such as HCV, BVDV (discussed in [45]). The terminal aromatic rings of 27 and 29 protrude from the enzyme and are solvent-exposed. In the context of the replication complex, these compounds may differentially affect the interactions of NS5 with other viral or host proteins and further contribute to or contravene viral inhibition. Residues R729 and R737 in DENV RdRp are likely to play similar roles as R386 and R394 in HCV RdRp. They interact respectively with the γ- and β-phosphate of the GTP moiety bound in the DENV [21] and JEV RdRp active sites (corresponding to R734, R742, JEV numbering; [46]). Thus, N-pocket inhibitors could affect de novo initiation by interfering with the binding of the incoming +1 rNTP substrate.
Some differences with HCV NS5B are noteworthy: there is no equivalent of the HCV RdRp primer grip wall for DENV N-pocket. In addition, unlike HCV RdRp where the C-terminal loop penetrates the active site and participates in enzyme activity, the C-terminal end of flavivirus RdRp is disordered in most reported crystal structures. Interestingly, this segment was recently observed to interact with a neighbouring MTase domain in a DENV FL NS5 oligomeric structure [47]. We speculate that the absence of both regions in DENV RdRp active site, prevents formation of additional contacts with N-pocket inhibitors, and is the reason for the weaker binding affinities of N-pocket compounds, compared to HCV site III inhibitors. Design strategies that capture and order the C-terminal sequence of DENV RdRp or its G-loop [45], would likely further enhance inhibitor binding affinity and block de novo initiation.
Residue H798 in DENV priming loop was proposed to be the counterpart of Y448 in HCV RdRp and to be responsible for ATP-specific initiation [48]. Unfortunately, H798 is too distant to make contact with the acyl-sulfonamide moiety of the compounds and design strategies in this direction were not fruitful. However, given that the N-pocket is close to the enzyme active site, extensions of inhibitor towards the GDD motif may strengthen the compound affinity. High clearance was observed for acyl-sulfonamide propargyl alcohol compounds in vivo which rendered them unsuitable for mouse efficacy studies. Both the thiophene ring and the primary propargyl alcohol have potential metabolic liabilities in vivo. To develop N-pocket inhibitors with better pharmaco-kinetic properties, both groups would need to be replaced with more stable moieties, whilst retaining key hydrogen bond interactions, with residues such as 800 and 802.
Finally, compounds 27 and 29 were inactive when tested on the WNV replicon cell-based assay (Fig 9 in S1 Text). Previous comparisons revealed that the WNV RdRp priming loop is closer to the i-1 site, and prevents formation of a similar N-pocket [45]. In addition, whilst DENV N-pocket residues are mostly conserved across the flavivirus family, residues 799–802, which accommodate the propargyl alcohol arm, are more divergent (Fig 3G). Taken together, these may be the reasons for the lack of compound activity in WNV. Interestingly, residues 799–802 are more similar amongst JEV, MVEV WNV, YFV and ZIKV, compared to DENV1-4. In this light, it may not be plausible to develop pan-active N-pocket inhibitors that work on all flaviviruses. Rather, designing N-pocket inhibitors that specifically target different subgroups of the flavivirus family may be a more attainable goal.
Methods used in this study are briefly summarized below. Full descriptions are given in Supporting Information.
A549 cells (human alveolar epithelial cells; American Type Culture Collection (ATCC), USA) were maintained in Ham’s F-12K medium (LifeTech, USA) containing 10% fetal bovine serum (FBS), 1mM L-glutamine and 1% penicillin-streptomycin. BHK-21 cells (baby hamster kidney cells; ATCC, USA) were cultured in Dulbecco modified Eagle medium (DMEM; LifeTech, USA) supplemented with 10% FBS, 1 mM L-glutamine and 1% penicillin-streptomycin. C6/36 mosquito cells (ATCC, USA) were grown in RPMI 1640 medium containing 10% FBS, 1 mM L-glutamine and 1% penicillin-streptomycin. A549, BHK-21 and HuH7 (Japan Collection of Research Bioresources Cell Bank, Japan) cells containing a DENV2 (New Guinea C) renilla luciferase or EGFP sub-genomic replicon were maintained in F-12 and DMEM medium, respectively, containing 10% FBS, 1 mM L-glutamine 20 μg/ml puromycin, and 1% penicillin-streptomycin [40]. Huh-7.5 cells containing a firefly luciferase sub-genomic replicon of hepatitis C virus (HCV) genotype 1b were licensed from Apath LLC (St. Louis, MO, USA; 42) and were maintained in DMEM containing 10% FBS, 1 mM L-glutamine, 0.25 mg/ml Geneticin, and 1% penicillin-streptomycin. A549, BHK-21, DENV2 replicon, and HCV replicon cell lines were incubated at 37°C. C6/36 cells were cultured at 28°C. All compounds were synthesized in-house. DENV-specific mouse monoclonal antibody 4G2 against the DENV envelope (E) protein was prepared from a hybridoma cell line purchased from the ATCC (USA) and rabbit poly-clonal antibody against DENV2 NS5 was purchased from GeneTex (USA). Synthesis of N-pocket inhibitors are described in [38].
Site-directed alanine mutations of DENV4 FL NS5 cDNA were performed using pET28-D4-MY01-22713 NS5FL [23] as a template, according to the manufacturer’s protocol (Stratagene, USA). Protein expressions of DENV1-4 FL NS5 and their stability analyses by thermo-fluorescence were performed as described previously [23]. DENV de novo initiation and elongation FAPA assays were earlier described [25].
DENV3 RdRp protein expression and crystallization was as described previously [22]. Briefly, DENV3 RdRp at 12 mg/ml was mixed with 1 mM compound 27 or 29 (prepared from a 10 mM DMSO stock to give a final concentration of 10% DMSO) prior to setting up hanging-drop vapor-diffusion crystallization trials in 0.1 M Tris/HCl, pH 8.0 and 25% PEG 500 MME. DENV2 RdRp (Strain NGC, aa 266–900) protein expression was as described previously (23). Protein crystallization was performed at 8 mg/ml in a sitting-drop vapor-diffusion setup with a well solution of 0.1 M MES pH 7.0, 0.35 mM MgCl2 and 16% PEG 4000 with a drop ratio of 2:1 (protein:well). Crystals appeared in one day and were transferred to the well solution supplemented with 10 mM compound 27 (prepared from a 100 mM DMSO stock to obtain a final concentration of 10% DMSO) for overnight incubation. For cryo-protection, crystals were transferred to the crystallization solution supplemented with 10% glycerol and 10% compound/DMSO and cooled in liquid nitrogen. Diffraction data were integrated using autoPROC (DENV2) or XDS (DENV3) and scaled using SCALA or AIMLESS, both part of the CCP4 suite [49]. The structures were directly refined using BUSTER, part of the global phasing suite.
Biotinylated DENV3 and DENV4 RdRp were captured on flow cells 2 and 4 respectively in 50 mM Tris/HCl, pH 7.5, 200 mM NaCl, 2 mM DTT, 0.05% Tween 20, and 3% DMSO at 4°C. Flow cell 1 and 3 were left blank to serve as a reference [37]. Compounds were tested in a 7-point 2-fold serial dilution, from 2.5 μM and a zero-concentration sample was subtracted from each run. Compounds were injected at a flow rate of 30 μL/min, with 45 s contact time and 600 s dissociation, starting from the DMSO control and finishing with the highest concentration. The experiments were performed using a Biacore T200 instrument and the data were analyzed using Biacore T200 Evaluation software, version 2.0.
DENV1-4 FL NS5 dnI assays were performed as described previously [39]. Briefly, compounds from 0–20 or -100 μM concentrations are two-fold serially diluted into 384-well black opaque plates (Corning Costar), after which 100 nM DENV FL NS5 protein was added and the plates incubated at RT for 20 min. RNA and ATTO-CTP, ATP, GTP and UTP were then added and the plates incubated for another 120 min. Reactions were stopped with buffer containing 25 nM CIP, re- incubated at RT for 60 min and read on a Tecan Safire II microplate reader. For order-of-addition experiments, DENV4 FL NS5 was incubated for one hour at RT with RNA, ATP, and GTP or RNA, ATP, GTP and ATTO-CTP, followed by exposure to serially diluted compounds for 20 min at RT. The missing components (ATTO-CTP and UTP or UTP alone) were added and the reactions continued for 120 min after which STOP buffer was added as before. All datapoints were performed in duplicate wells. Each compound was tested at least twice.
BHK-21 DENV2 (strain New Guinea C) EGFP-replicon cells [40] resistant to compounds 27 and 29 were first obtained by serial passaging of the cells in 14 μM of 27 or 20 μM of 29 (1–2× EC90 values). Briefly, 1 X 105 cells were seeded over-night into 6-well plates, followed by addition of fresh media containing 2% FCS and compounds. Media was changed every 2–3 day. After 5 weeks, concentration of 29 was increased to 25 μM. Individual colonies or mixed populations of resistant cells were isolated, expanded and total cellular RNA extracted. Alternatively, native replicon cells were incubated with media containing 1.5 μM of 27 (0.5X EC50 value) and after 3 days, fresh media with 2-fold increase in compound concentration was added. The process was repeated until cells were exposed to 28 μM 27, after which total cellular RNA was extracted. Viral RNA was extracted by using QIAamp viral RNA minikit (Qiagen) and NS5 cDNA was amplified by SuperScript One-Step reverse transcription (RT)-PCR with Platinum Taq (Invitrogen) and subjected to DNA sequencing. Control cells were passaged in the presence of 0.5% DMSO.
Mutations in the DENV4 NS5 (GenBank accession number AF326825) sequence were engineered into the subclone, pACYC-DENV4-F shuttle, using the QuikChange II XL site-directed mutagenesis kit according to the manufacturer’s protocol (Stratagene). This plasmid harbours nucleotides 7564–10653 (from NS3-3’UTR) from the DENV4, MY01-22713 strain, linked at the 3’end to the Hepatitis D virus ribozyme (HDVr) sequence. Following sequence verification, the plasmids were digested with NotI and KpnI and inserted with a PCR product comprising the sequence comprising nucleotides 1–7563 downstream of the T7 promoter in which the region from nucleotides 217–2291 in this cDNA has been replaced by renilla luciferase and foot-and-mouth disease virus 2A protease cDNAs [50].
DENV2 (strain New Guinea C, NGC) replicons or full-length cDNA clones with NS5 mutations were constructed with a pACYC-NGC-RLuc replicon or pACYC-NGC FL, respectively and a TA-NGC (shuttle E) vector as previously described [50]. The pACYC-NGC FL plasmid contains the T7 promoter, the DENV2 NGC genome, and HDVr. The pACYC- NGC-RLuc replicon plasmid contains the same cDNA as pACYC-NGC FL except that cDNAs encoding structural proteins were replaced by renilla luciferase cDNA [40]. The shuttle E vector contains nucleotides 5427 to 10955 (from NS3 to 3’UTR and HDVr sequence). All NS5 mutations were engineered into the shuttle E vector, using QuikChange II XL site-directed mutagenesis kit (Stratagene) according to the manufacturer’s protocol. The mutants were cloned into pACYC-NGC replicon or FL plasmid at BspEI and MluI restriction sites. All constructs were verified by DNA sequencing.
A549, BHK-21 and Huh7 cells bearing stable DENV2 sub-genomic replicon [40] or transiently electroporated DENV2-NGC replicon or infectious virus cDNAs were seeded into 384-well microplate (3,000 cells per well). After over-night incubation at 37°C with 5% CO2, the cells were treated with 2-fold serially diluted compounds, starting from 20 or 50 μM. At 48 hr of post-incubation, renilla luciferase activities were measured with the ViviRen live-cell substrate (Promega, USA) according to the manufacturer’s protocol. CellTiter-Glo reagent (Promega, USA) was then added to determine cytotoxic effects of compounds. For the HCV replicon assay, Huh-7.5 cells harboring the HCV replicon [43] were seeded into a 96-well microplate (20,000 cells per well). At 48 hr after compound treatment, cells were assayed for firefly luciferase activity by using a Bright-Glo luciferase assay (Promega, USA). NITD-008, a nucleoside inhibitor of DENV and HCV was added as a control [33]. Compounds were tested up to 25 μM in HCV replicon cells due to limits in DMSO tolerability of these cells, and up to 50 μM in other cell types, based on compound solubility.
|
10.1371/journal.pcbi.1000964 | Role of Lipids in Spheroidal High Density Lipoproteins | We study the structure and dynamics of spherical high density lipoprotein (HDL) particles through coarse-grained multi-microsecond molecular dynamics simulations. We simulate both a lipid droplet without the apolipoprotein A-I (apoA-I) and the full HDL particle including two apoA-I molecules surrounding the lipid compartment. The present models are the first ones among computational studies where the size and lipid composition of HDL are realistic, corresponding to human serum HDL. We focus on the role of lipids in HDL structure and dynamics. Particular attention is paid to the assembly of lipids and the influence of lipid-protein interactions on HDL properties. We find that the properties of lipids depend significantly on their location in the particle (core, intermediate region, surface). Unlike the hydrophobic core, the intermediate and surface regions are characterized by prominent conformational lipid order. Yet, not only the conformations but also the dynamics of lipids are found to be distinctly different in the different regions of HDL, highlighting the importance of dynamics in considering the functionalization of HDL. The structure of the lipid droplet close to the HDL-water interface is altered by the presence of apoA-Is, with most prominent changes being observed for cholesterol and polar lipids. For cholesterol, slow trafficking between the surface layer and the regimes underneath is observed. The lipid-protein interactions are strongest for cholesterol, in particular its interaction with hydrophobic residues of apoA-I. Our results reveal that not only hydrophobicity but also conformational entropy of the molecules are the driving forces in the formation of HDL structure. The results provide the first detailed structural model for HDL and its dynamics with and without apoA-I, and indicate how the interplay and competition between entropy and detailed interactions may be used in nanoparticle and drug design through self-assembly.
| Cardiovascular diseases are the primary cause of death in western countries. One of the main causes is lipid accumulation and plaque formation on arterial walls, called atherosclerosis. The risk of being exposed to this condition is reduced by high levels of high density lipoprotein (HDL). The functionality of HDL has remained elusive, and even its structure is not well understood. Through extensive coarse-grained simulations, we have clarified the structure of the lipid droplet in HDL and elucidated its interactions with the apolipoprotein A-I (apoA-I) that surrounds the droplet. We have found that the structural and dynamic properties of lipids depend significantly on their location in the particle (core, intermediate region, surface). As for apoA-I, we have observed it alter the overall structure of the lipid droplet close to the HDL-water interface, with prominent changes taking place for cholesterol and other polar lipids. The nature of lipid-protein interactions is most favorable for cholesterol. Our results reveal that not only hydrophobicity but also conformational entropy are the driving forces in the formation of HDL structure, suggesting how the interplay and competition between entropy and detailed interactions may be used in nanoparticle and drug design through self-assembly.
| Cardiovascular diseases are the primary cause of death in western countries [1]. One of the main causes is the lipid accumulation and plaque formation on arterial walls, called atherosclerosis. This eventually leads to the narrowing of arteries, plaque rupture, clotting, and potential death. Generally speaking, high levels of low density lipoprotein (LDL) in blood have been found to increase the risk of atherosclerosis [2], [3], whereas high levels of high density lipoprotein (HDL) have been shown to reduce the risk [4], [5].
Despite more than a decade of extensive studies, LDL and HDL structures are not well understood. This is largely due to their small size which ranges from about 10 (HDL) to 25 nm (LDL) rendering experimental studies of the detailed lipoprotein structures extremely difficult. This challenge is further corroborated by the soft nature of lipoparticles whose structures are transient due to thermal forces driving molecular assembly processes in living matter. The challenge is to unravel the role and mechanisms of lipoproteins in the trafficking of cholesterol and in the cholesterol-based diseases. In this work, we focus on HDL.
Let us briefly summarize the main insight one has about high density lipopproteins. HDL particles are comprised of a lipid droplet surrounded by proteins [6], [7]. Apolipoprotein A-I (apoA-I) is the main protein associated with HDL, which is the main carrier of excess cholesterol from peripheral tissues to the liver, that is, for reverse cholesterol transport [5], [8]. After synthesization, the ATP-binding cassette transporter A1 (ABCA1) assembles lipid-poor apoA-I molecules and lipids into discoidal HDL particles [9], after which the enzyme lecithin∶cholesterolacyl transferase (LCAT) esterifies cholesterol molecules, leading to the formation of spheroidal HDL [10], [11]. The spheroidal HDL is the main form of HDL responsible for cholesterol transport to the liver.
Though a number of experimental studies have been carried out to unravel the structure and dynamics of apoA-I molecules in lipid-free form [12]–[15] and in discoidal HDL complexes [16], [17], the structure of the spheroidal HDL has remained unclear. As for the structure of apoA-I, a large amount of data is in favor of the so-called double belt model (see Ref. [18] and references therein), where the apoA-I proteins line along the lipid droplet. The composition of the droplet has been resolved [19], [20] (see Table 1), indicating free cholesterol (CHOL), cholesteryl esters (CE), triglycerides (TG), phospholipids, and lysolipids to be its main constituents, distributed such that there is a hydrophobic interior of triglycerides and cholesteryl esters and a surface covered by polar head groups of phospholipids [21]. This is essentially the so-called two-layer model for HDL [6], [7], [22]. Furthermore, parts of the apoA-I proteins have been proposed to interact with the acyl chains of the lipids [23]–[27].
Currently, the role of lipids for HDL functions are only vaguely understood. This is partly due to the transient time scales associated with and the nano-scale nature of HDL. Further issues include the poor understanding of lipid organization and interplay of lipids with apoA-I. Considering findings that lipids are an integral component of protein structures, e.g., in membrane proteins that are in constant interplay with lipids [28], it is obvious that clarifying the role of lipids for HDL properties is extremely important.
A number of computational studies have recently been conducted to complement experiments. Previous computational studies of HDL particles have focused on discoidal particles consisting of phospholipids and two apoA-I molecules [29]–[32]. These studies have provided some insight into the mechanisms of assembly and the dependence of the particle shape on the lipid/protein molar fractions. In a different approach, bulk melts of cholesteryl esters [33] and triglycerides [34], as well as combination of cholesteryl esters with POPCs [35] have recently been simulated. Catte et al. [36] reported the first computational approach towards understanding the structure of spheroidal HDL particles. Their study clarified the conformation of apoA-I in model spheroidal HDL particles using both all-atom (AA) and coarse-grained (CG) molecular dynamics (MD) simulations. This combination of AA and CG-MD simulations led to model spheroidal HDL particles with prolate ellipsoidal shapes having sizes consistent with experimental results and suggested that cholesteryl esters stabilize the conformations of apoA-I [37]. In a more recent work, Shih et al. also combined coarse-grained simulations with atomistic ones in a series of simulations where discoidal HDL was matured into spherical HDL upon incorporation of cholesteryl esters [38]. They found that maturation results from the formation of a dynamic hydrophobic core composed of cholesteryl esters, the core being surrounded by a layer of phospholipids and apoA-I proteins. Interestingly, Shih et al. also fine-grained the coarse-grained HDL particle to atomistic description and then used atomistic simulations to consider the structure of apolipoproteins around the lipid droplet, and in particular the importance of salt bridges in apoA-I.
The main limitations of previous simulations of HDL particles are two-fold. First, the lipid composition modeled in recent simulations has been somewhat unrealistic: instead of a many-component lipid mixture, the lipid droplet has been modeled as a single-component POPC melt, or as a mixture of POPCs and cholesteryl esters [36]. The role of the many different lipid species in HDL has therefore remained unknown. Second, the time scales of HDL simulations have been too short compared to the characteristic time scales of lipid mixing and structural deformations associated with HDL. As even the time scale of lipid mixing is of the order of 1 s (derived through the diffusion of lipids inside HDL), and the current state-of-the-art for atomistic simulations of HDL extends over 10–100 ns, it seems obvious that currently atomistic simulations are not the method of choice for dealing with HDL over large enough time scales.
Our objective is to overcome the above limitations. We have performed MD simulations of spheroidal HDL particles using the full lipid composition of human plasma HDL [19]. Instead of atom-scale simulations, we employ the coarse-grained MARTINI model [39], [40] that has performed exceptionally well in a number of studies dealing with lipids and proteins [36], [39]–[42]. We consider both the protein-free lipid droplets and the full HDL particles containing also two apoA-I molecules around the droplet, see Figures 1 and 2. Composition of the HDL system is described in Table 1 with abbreviations of all molecules included. By comparing the protein-free and the full HDL models, we can clarify the role of lipids and proteins in HDL. The principal objective is to fill the gap of detailed structural and dynamic information of lipids in spheroidal HDL particles. We also address questions related to the role of apoA-I proteins and their interactions with lipids in HDL structures. The currently incomplete understanding of the latter issue is largely due to the size heterogeneity of HDLs (diameters range from 7.2 () to 12 nm ()) and the large flexibility of apoA-I. The latter renders the prediction of the positioning of different alpha helices of apoA-I on a spherical surface very difficult. The distribution of lipids inside HDL and their interplay with apoA-I are of profound interest. From a more general perspective, knowledge of the structure of spheroidal HDL is crucial for understanding the conformational changes when HDL makes the transition from discoidal to spheroidal shape, and the trafficking of CHOL and CE through the action of cholesteryl ester transfer protein [43]. Additionally, to design nanoparticles with desired surface and bulk properties, e.g., for controlled transport and release of drugs and contrast agents, it is vital to understand the conformational changes as well as the underlying mechanisms in detail [44]–[51].
The radial density distributions shown in Figure 3 reveal the internal structure of the simulated lipoparticles. The hydrophobic CE and TG molecules are located in the core of the particle and have minimal overlap with water. The lipids with a polar head group, POPC and PPC, are mostly located at the surface region facing water, whereas most of CHOL is located just below these two lipids. Note that a small but significant concentration of CHOL is also found in the core of the particle.
Considering the size of HDL, the radii of gyration give an average of nm for the droplet and nm for the full HDL. Both particles are effectively spherical, as indicated by the moments of inertia (data not shown).
The apoA-I proteins are embedded onto the surface of the HDL particle, their density peaking just slightly below the headgroup region of POPC and PPC. The presence of the protein slightly disturbs the distribution of these lipids as revealed by the comparison of the lipid droplet with the full HDL particle. The distribution of hydrophobic lipids remains undisturbed. Most significant is the shifting of the distribution of CHOL, and partly PPC, towards water phase when the protein is present, while the distribution of POPC is shifted slightly towards membrane center, making room for CHOL and PPC. In the full HDL particle, water is found to distribute less to the particle compared to the droplet.
Our results clearly highlight the displacement of CHOL even further towards the interface in the full HDL particle. The data below shows that CHOL interacts prefentially with some of the protein residues, strongly promoting the partitioning of CHOL to the vicinity of apoA-I. CHOL further prefers to reside next to the water region, facilitating (hydrogen) bonding via the polar OH group. It has been proposed [52] that CHOL molecules can mediate the relief of membrane stress arising from chain-chain mismatches, since their dimerization is not favored in membranes with high surface curvature. This view is supported by the observations of Huang and Mason [53]. Their work suggests that high surface curvature requires CHOL to be at the interfacial region. Interestingly, Lemmich et al. have further found that very small amounts of CHOL (less than about 3 mol-%) may soften the interface and hence promote its fluidity [54]. In HDL, the average concentration of CHOL is about 10 mol-%, but at the interface it is about 5–10 mol-% depending on distance from the water phase (see Figure 3).
The minor but significant concentration of CHOL in the core of the particle calls for discussion. The usual assumption especially in studies of lipid membranes is that CHOL resides at the water-lipid interface due to its polar OH group. This is expected often to be the case, though there are also reported exceptions such as CHOL residing for short times in the middle of a polyunsaturated lipid bilayer [55], [56], and the suggestion of CHOL in the interior of LDL [57].
To start with, one gets an impression that the density plot adheres to the two-layer model [6], [7], [22] wherein one assumes almost full separation of hydrophilic and hydrophobic molecules into two separate regions. While the distribution of TG fits into this picture, the distribution of CE and CHOL does not. A rather significant amount of CHOL is also in the core of the particle as was discussed above. Detailed consideration further reveals that there is a significant overlap of CE with CHOL, POPC, and PPC: The radial density distributions shown in Figure 3 do not provide a sufficiently unique description of only two different structural regions inside HDL. Furthermore, by looking at the order parameters of CHOL and CE presented in Figure 4 it becomes evident that there are not only two regions but also the intermediate one between the hydrophobic and hydrophilic ones. The innermost core of the particle ( nm) is clear, as there the ring structures of both CHOL and CE are oriented in a completely random fashion. The situation changes as one approaches the lipid-water interface through the intermediate region (3 nm nm), which is characterized by significant ordering of the ring structures, in a manner where the principal axis of CE's and CHOL's ring moiety lies along the radial direction of HDL. This intermediate region overlaps with the distribution of the acyl chains of POPC and PPC, revealing that the sterol rings are also aligned with the acyl chains. Finally, at the HDL-water interface, one finds the region composed of hydrophilic headgroups of POPC and PPC that constitute the surface part of the lipid droplet interacting mostly with water.
The data clearly shows that instead of the two-layer model, the distribution of lipids in HDL is more complex. The structural nano-scale organization of CHOL and CE plays an important role in constituting the intermediate layer. However, there is no apparent reason to conclude that the lipid droplet in HDL would be described by a “three-layer” model either, since the intermediate region is narrow and represents a crossover from the hydrophobic to the hydrophilic environment rather than a clearly defined layer of its own. Our results for lipid dynamics are in favor of this view and will be discussed below in the context of diffusion. Meanwhile, while quantitative results have been missing, a three-layer model has earlier been proposed for LDL particles [3]. There the situation is different, though, since the diameter of LDL is roughly three times larger compared to HDL and the intermediate region can possibly be broader and more characteristic compared to HDL.
There are significant differences when the order parameters of CHOL and CE are compared (Figure 4). First, the height of the main peak is considerably lower for CE than for CHOL, indicating that the ring of CE has a lower tendency to orient itself along the acyl chains than CHOL. Second, unlike for CHOL, on the surface of the particle ( nm) the order parameter of CE obtains negative values. These indicate the ring of CE to lie along the surface, perpendicular to the radial direction. This obviously stems from entropic reasons, since while CE strives in part to organize its structure like CHOL, also directing its weakly polar ester bond region to the surface like CHOL does for the OH group, CE also has a long oleate chain. Previous atomistic simulations of CE in bulk conditions as well as in a combined system with POPCs have shown that the oleate chain of cholesteryl oleate has essentially three different conformations with respect to the ring of CE [33], [36]: one where the angle of the oleate chain (describing it as a semi-stiff rod) with respect to the principal axis of the ring is about 35 degrees, and two other modes with an angle of 90 or 150 degrees. Compared to CHOL with only one mode, CE inevitably aims to minimize free energy by promoting entropic degrees of freedom.
Another interesting observation is that apoA-I suppresses the main peak of both CHOL and CE molecules in Figure 4. The effect is an indication that the protein disturbs the ordering within the intermediate region (between the core and the surface), also facilitating the displacement of CHOL towards the water phase. This conclusion is supported by the broadening of the angle distributions of POPC in the presence of the protein (see Supporting Information (SI)).
An analysis of the internal conformations of CE molecules in Figure 5 provides a more detailed view of the situation. In the core of the particle, the most probable conformation of CE is the coil-like conformation (maximizing entropy), where the angle between the CE ring and the oleoyl chain is about 120 degrees. This is largely consistent with recent atom-scale simulations of CE in bulk conditions [33]. The behavior changes on the surface of the particle. The two peaks of the distribution on the surface correspond to two distinctly different conformations: one where the ester group of CE (corresponding to the OH-group of CHOL) points towards water and the oleate chain is extended towards the solvent, and another where the ester region is pointing towards the core of the particle, while the ring and the oleoyl chain form a small angle with each other.
Also for TG, we find a change of conformation when it is shifted from the core of the particle onto the surface. In the core, the three chains of TG place themselves to a similar conformation as in a bulk melt of TG [34]. When brought to the surface, the ester bond regions seek contact with water, which brings the three chains of TG closer to each other into a more tightly packed conformation (see Figure S2). Additional data for molecular conformations are presented in Figure S1, Figure S3, Figure S4, and Figure S5.
The large-scale dynamics within HDL and the lipid droplet are considered in terms of diffusion, characterized by the diffusion coefficient . The diffusion coefficients were determined by considering lipid displacement distribution functions over a fixed period of time (see SI). We found that the jump length distributions for lipids in the core of the particle (TG and CE) fitted well with the three-dimensional diffusion model, yielding . Meanwhile, the lipids on the surface (POPC, PPC, CHOL) fitted much better with the two-dimensional description for diffusion, yielding . For details, see SI.
Table 2 shows the long-time diffusion coefficients of the lipid components within the lipid droplet and the full HDL particle.
Figure 6 depicts how the diffusion rate varies significantly inside the lipid droplet and/or full HDL. The diffusion is the slowest in the middle of the particle, it speeds up as the molecules get closer to the interface, and the fastest diffusion is found at the interface. The influence of apoA-I on diffusion of lipids is modest. It turns out that the lipid diffusion coefficients in the protein-free lipid droplet and the full HDL particle are almost similar. The apoA-I proteins may slow down the diffusion of lipids slightly especially close to the interfacial regions. The effect is, however, weak (see Table 2).
The diffusion coefficients of POPC, PPC and CHOL in the surface region of the particles are about and in good agreement with experimental estimates of for two-dimensional lipid bilayers in fluid phase [58]. On the other hand, the diffusion coefficients for CE and TG are smaller by a factor of 10, about . To our knowledge, diffusion coefficients of lipids in HDL have not been experimentally determined. However, for LDL and LDL-like lipid droplets, Vauhkonen et al. used pyrene-linked PC lipids as probes to find that at the surface of lipoparticles [59], in good agreement with our findings. Massey and Pownall have further used single-chain cationic amphiphiles for considering lipid mobility at the surface regions of LDL and HDL, and while quantitative estimates for are missing, they concluded that the diffusion at the surface is about 2–3 times slower compared to cholesterol-free POPC vesicles [60]. Recent MARTINI-model simulations for single-component PC bilayers have yielded [61], which is about a factor of 2 larger than diffusion at the surface of HDL. While the comparison of our simulation data and experiments is suggestive rather than conclusive, the qualitative agreement is striking.
Our main result regarding diffusion is that diffusion at the surface region of HDL is largely similar to diffusion of lipids in cholesterol-containing lipid bilayers in the fluid-like phase, the cholesterol concentration being roughly 10 mol%. Figure 6 also shows convincingly that the effect of apoA-I on diffusion of lipids is not significant.
Additionally, Figure 6 provides compelling evidence that the dynamics of lipids in terms of their diffusion properties is not consistent with the two-layer model. Instead of two clearly different dynamic regions, we find the diffusion coefficients to increase monotonously: diffusion rates are clearly different in the core ( nm), in the ordered intermediate region ( nm), and at the surface ( nm).
Given the different proposed models for lipid distribution in HDL, the striking difference of the present findings compared to earlier studies is the role of dynamics: not only the structural and ordering properties of molecules in HDL differ across HDL, but also the dynamics in terms of molecular transport coefficients varies significantly in the different compartments. The biological relevance of this feature lies in the time scales of molecular trafficking inside HDL: while molecular transport between the surface and the intermediate region is relatively fast, the transport between the surface and the core of HDL is slower by a factor of 10.
The above results show that the apoA-I proteins do not induce large changes to the lipids' properties inside the droplet. Yet, while the protein collapses onto the surface of the droplet, it does disturb the packing, ordering and, although only slightly, also the dynamics of the lipids at the surface region. What remains to be explored is the nature of the lipid-protein interactions. In this work our primary interest is the lipid component of HDL, thus we have used the standard CG MARTINI model which does not enforce the full secondary structures in apoA-I. This optimizes computational efficiency and allows us to focus on generic issues such as the partitioning of lipids around apoA-I, and the influence of apoA-I on the lipid droplet. Meanwhile, we cannot address questions related to detailed atomistic phenomena at the lipid-apoA-I interface.
Data for the surface accessible surface areas (SASAs) of apoA-I hydrophilic and hydrophobic residues (data not shown) provide evidence for the low contribution of protein hydrophobic residues (11%) to the total SASA of the protein, the main contribution coming from protein hydrophilic residues (89%). The average value of SASA of protein hydrophobic residues () is in good agreement with that reported by Shih et al. [32] in a recent study on the assembly of lipids and proteins into lipoprotein particles.
The RMSFs of protein carbons are shown in Figure 7 and reveal the mobility of different protein domains. The -helical structure of the protein exhibits very little mobility for both chains. This rigidity of the protein is also in agreement with the observed slight disturbances produced on the lipid packing.
The number of annular lipids, as defined in the Method section, is given in Figure 8 for each lipid component. It is interesting to note that about 80% of CHOL molecules are annular (on average 40 out of 49) while only 10% of CE molecules (about 15 out of 122) are in close contact with the protein. An average of about 98 POPC molecules out of a total of 260 are annular. Overall, the results indicate that there is a preferential interaction between CHOL molecules and protein residues. This result is striking if one considers that the number of POPC molecules is larger than that of CHOL molecules.
It is known that the number of apolipoproteins in HDL depends on particle size. We characterized its role for lipid distribution through additional simulations with three apoA-Is. First, we performed a 20 microsecond simulation of the same lipid droplet with 3 apoA-I molecules placed 2 nm apart from each other. The protein molecules were found to insert themselves in the lipid droplet in the same way as was observed above, with hydrophobic moieties pointing towards the droplet. The only interesting difference was that in the structure with three apoA-Is, the C-terminus and the helix 9 of one protein molecule were not inserted in the lipid droplet. This situation is likely due to the crowded arrangement of apolipoproteins, or due to the limited time scale of the simulation. The addition of a third apoA-I molecule does also affect the interaction of cholesterol with apoA-I: Almost 100% of the cholesterol molecules ( out of 49) are in contact with the three proteins. That is, the addition of the third apoA-I molecule enhances the average number of annular cholesterol molecules from about 80% (observed with 2 apoA-I molecules) to about 96% of the total unesterified cholesterol in the particle.
The lipid-protein interactions of different moieties of each lipid component showed that POPC, PPC and TG molecules interact with apoA-I residues preferentially through the acyl chains (POPC and PPC glycerol backbone has also a high number of contacts with apoA-I), while CHOL and CE molecules interact with the protein mainly through the sterol ring (see Table S1). To better understand the nature of the interaction between CHOL and apoA-I we also measured the number of lipid-protein contacts per residue (hydrophobic and hydrophilic), shown in Figure 9. It is clear that there is a preferential interaction of CHOL molecules with apoA-I hydrophobic residues, in particular tryptophane (Trp) and phenylalanine (Phe) having aromatic side chains, but also valine (Val) and leucine (Leu). Highly preferred interaction with Trp and Phe is understandable through findings of aromatic ring pairing in e.g. known protein structures [62]. We also observe a relevant number of contacts with apoA-I hydrophilic residues with aromatic side chains such as tyrosine (Tyr) and histidine (His). This is not surprising, as Tyr has a hydrophobicity comparable to Phe as has been shown experimentally by Wimley and White [63] through the determination of a hydrophobicity scale for proteins at membrane interfaces. There are less contacts of CHOL molecules with the other apoA-I hydrophilic residues, namely serine (Ser), threonine (Thr) and asparagine (Asp) being the most attractive ones. These results are in good agreement with the observed large number of contacts of the sterol ring of CHOL molecules with protein residues.
The sterol ring of CHOL molecules can intercalate or interact with the aromatic side chains of protein residues as observed for CE in a recent study by Catte et al. [36]. This interaction between CHOL molecules and apoA-I was also observed experimentally by Dergunov et al. [64]. The authors estimated the degree of exclusion of CHOL molecules from the boundary lipid region in reconstituted discoidal HDL particles containing different apolipoproteins and observed an increase in the order A-I<E<A-II. The partial exclusion of CHOL molecules operated by apoA-I and the corresponding CHOL distribution among surface and bulk lipids are in good agreement with our findings as depicted through a series of snapshots in Figure 10 (see also SI).
The binding between CHOL molecules and apoA-I residues is quite weak, which permits exchange among apoA-I -bound and free CHOL molecules on the time scale of the simulation. We characterized this trafficking process by computing the distributions of lifetimes between CHOL-protein and CE-protein contacts. The average lifetime was found to be 146 ns for CHOL and 15 ns for CE. While the errors are of the same order as the lifetime due to a limited number of samples, and the fact that the distribution for CHOL is broad as there are cases where the CHOL-protein contact is maintained throughout the simulation, the results highlight the stability of CHOL-protein binding with respect to that of CE. The relatively large lifetime of the CHOL-protein binding also highlights that once CHOL has migrated to the vicinity of apoA-I, it remains there for a long period of time. For comparison, the average non-contact lifetime for CHOL-protein pairs, describing the characteristic time for CHOL to not be in contact with any parts of apoA-I was found to be about 175 ns. That is, CHOL molecules reside close to the water-HDL interface and on average spend half of their time in contact with apoA-I.
The above results are in good agreement with NMR experiments performed on human HDL, which indicate that CHOL molecules are present in two distinct environments [65]. More specifically, Lund-Katz et al. found that the cholesterol molecules dissolved in the core of HDL are relatively disordered and mobile, while the cholesterol molecules located among phospholipid molecules in the surface of the particle undergo relatively restricted, anisotropic motions. This view is in line with our simulation results discussed earlier in this article. Lund-Katz et al. thus proposed that cholesterol molecules are in two different microenvironments, undergoing fast exchange between these two regions, equilibrating between the surface and the core of HDL in the time scale of milliseconds or more. While the time scales proposed by Lund-Katz et al. are beyond those that are accessible via simulations, we have found that there is local exchange taking place at times up to microseconds. However, the time scales we have found via simulations should be regarded as the lower limit, since the diffusion coefficients we have found for the core of HDL imply that the exchange of cholesterols between the core and surface regions has to be larger than 1 s.
In this study, we elucidated the structure and dynamics of spheroidal high density lipoparticles with a realistic lipid composition corresponding to human serum HDL. We found that the traditional two-region model for HDL is not accurate enough. Instead, we found the distribution of the different lipid types in HDL to be more complex.
The innermost core of HDL is mainly occupied by TG and CE, which as hydrophobic lipids constitute a randomly oriented melt. However, in contrast to the common view, the inner core was also found to contain a rather significant fraction of free cholesterol partitioned into the disordered melt. The outermost surface region constitutes the interface with water, which is mostly occupied by the polar headgroups of POPC and PPC. Between these two is the intermediate region occupied by CHOL, partly also CE, and the acyl chains of POPC and PPC. However, there is no apparent reason to consider the intermediate region as a “third layer”, since it is narrow, unclear to define spatially, and represents a crossover from the hydrophobic to the hydrophilic environment rather than a true layer of its own. Yet it has properties that are distinct from those in the core and at the surface.
This is most obvious in two aspects: ordering of steroid moieties and molecular diffusion. Unlike in the core, in the intermediate region the bulky rings of CHOL and CE are strongly ordered along with the acyl chains. This ordering extends also to the surface region of HDL, highlighting the difficulty to define the intermediate region as a true layer of its own. This view is also supported by the diffusion data, which illustrates that the diffusion of molecules takes place at a clearly different pace in the different regions. In the core and in the intermediate region of the particle, diffusion was found to be three-dimensional, while the diffusion of lipids at the HDL-water interface turned out to be two-dimensional in nature. Quantitatively, diffusion in the core of the particle was observed to be slow, as in a polymer melt, and to speed up monotonously as one crosses the intermediate region and ends up in the interfacial region.
The perspective arising from our results is novel, providing the first molecular scale view to the nano-scale organization of lipids in HDL. The present results indicate that the spatial distribution of lipids within HDL provides only a narrow perspective to the complexity of lipid organization. To understand this issue, one has to pay considerable attention not only to density distributions but also conformational and orientational degrees of freedom of the lipids, and their dynamics within HDL.
Events where CE and TG penetrate to the surface were found to be rare. In the few observed cases when it occurred, their conformation was significantly changed. In the core of the particle, both CE and TG were observed to be capable of obtaining more coil-like conformations. This indicates that the formation of the HDL core is not only driven by the hydrophobic effect, but that conformational entropy has a significant effect.
When comparing the simulation of the full HDL (with apoA-I) to the lipid droplet (without apoA-I), we found that the overall structure of the lipid droplet was not significantly changed by the presence of the protein. Rather, we found a disturbance in the behavior of the surface lipids. In particular, the order of CHOL and CE molecules decreased and the conformations of the acyl chains of PC lipids got broader. Diffusion of the surface lipids was slightly perturbed by the protein, but the effect was minor.
The low contribution of the SASA of apoA-I hydrophobic residues to the total SASA of the protein and the large number of contacts of hydrophobic moieties of each lipid component with apoA-I evidence that the hydrophobic forces drive the insertion of the protein and contribute to the stability of the full HDL. Interestingly, we found that a large number of CHOL molecules interact with apoA-I, mainly through their sterol ring and especially with hydrophobic residues having an aromatic side chain. We also observed fast exchange among protein-free and protein-bound CHOL molecules. This result is in good agreement with experimental findings for human HDL particles [65].
It is tempting to discuss the physiological relevance of the above-discussed molecular level findings, especially the preferable interaction of the sterol ring moiety with the aromatic amino acids, and the observation that CHOL molecules spend about half of their time in contact with apoA-I, trafficking relatively rapidly back and forth in the vicinity of apoA-I. Such interaction is prone to have impact on the availability of sterols and lipids for related transfer proteins and enzymes, such as the cholesteryl ester transfer protein (CETP) and cholesteryl esterases. This interaction may be even more important in the process of cholesterol efflux, which is the critical part of reverse cholesterol transport, where the accumulated cholesterol is removed from macrophages.
For future purposes for characterizing the properties of HDL, as well as HDL under enzymatic reactions, our results bring about a useful view to consider. We have found that the interfacial region of HDL close to the water phase is rather well defined in terms of its molecular composition: it can be described as a model layer composed of PCs, lyso-PCs, CHOLs, and the apolipoproteins A-I. The diffusion results discussed in this study indicate that the lateral diffusion along the interfacial layer is largely consistent with diffusion taking place in model membranes, whose molecular composition is of the same type. These features suggest that both the physical and chemical properties of HDL could be explored with reasonable accuracy through studies of (planar) model membranes, which are considerably easier to characterize compared to nano-sized HDLs. Clearly, the insight gained in this manner would be limited, since a number of inherent features would be missing, such as the curved nature of the HDL-water interface and its effect on apoA-I. Nonetheless, there is reason for optimism, encouraging experiments and simulations to use model membranes for better understanding of lipoprotein properties, including both HDL and LDL.
The view presented in this article for HDL structure and dynamics paves way to extend the scope of computational studies for HDL, and to gain a much deeper understanding of HDL properties in a number of conditions related to health. For instance, there is reason to assume that the molecular composition in HDL depends to some extent on factors such as diet and lifestyle. In altered HDL the lipid composition can be abnormal due to e.g. dyslipidemia [66]. Characterization of molecular composition of HDL of subjects with varying degrees of health would allow coarse-grained simulation studies of HDL in these subject groups, using the present results as a reference. Preliminary studies in this spirit have very recently been reported and discussed by Yetukuri et al. [67], who found that an elevated triglyceride concentration in low-HDL subjects also affected its distribution in HDL, increasing the concentration of triglycerides markedly at the lipid-water interface next to apolipoproteins. Such results based on large-scale coarse-grained models can further be fine-grained to atomistic description to study the atom-scale features that are relevant e.g. in lipid-protein interactions, and the implications on HDL stability due to reactions of enzymes such as phospholipases. Work in this direction has already been initiated by Shih et al., who recently fine-grained coarse-grained models for matured HDL particles comprised of apoA-Is, phopholipids, and cholesteryl esters [38]. Similar work is in progress for the present HDL models.
Altogether, considering the complexity of HDL, our study highlights the importance of lipid-apoA-I interactions, and in particular the specificity of apoA-I for free cholesterol and its esters. The molecular-scale insight of HDL structure and dynamics found and confirmed in our study largely stems from the ordering and dynamical phenomena taking place close to the HDL-water interface, being in part driven by the interactions between cholesterol and apoA-I. Evidently, they have a prominent role to play in a number of transport processes dealing with cholesterol.
Construction of the models was implemented in two stages. First, we constructed lipid droplets (without the apoA-I proteins) using coarse-grained descriptions of lipids and water. Second, the studies of pure lipid droplets were complemented by models where the droplet was surrounded by two apoA-I proteins. Below, we describe the main stages of the model construction.
The initial structure for the lipid droplet was obtained by placing the set of lipid molecules (see Table 1) randomly into a three-dimensional simulation box without water or any other solvent. As CE, we used cholesterol oleate, while TG was chosen as trioleate. The system was then simulated under NpT conditions in order to reach proper density. The resulting molecular melt was next hydrated with 100,000 water particles and the energy was minimized, after which the system was equilibrated for 8 s. After equilibration, the system was simulated over a period of 4 s that was used for analysis. All time scales shown here represent the realistic effective time (simulation time multiplied by the scaling factor of four) [39].
For POPC, PPC, CHOL and water, we used standard components of the coarse-grained MARTINI force field [39]. The parameters for CE and TG are those corresponding to cholesteryl-oleate and trioleate respectively, and they come from a combination of standard MARTINI-components and careful adjustments of the key particle types and angle potentials (see SI). The adjustments were justified by comparing structural properties of the molecules in bulk with extensive atomic-scale simulations [33], [34], see SI for details. Sphingomyelin (SM) in ref. [19] has been included in the POPC contribution, as SM's properties in the MARTINI description do not differ considerably from those of POPC.
Next, all-atom apoA-I molecules were generated using as a reference the molecular belt model of apoA-I for discoidal HDL [68]–[72]. The hydrophobic faces of the amphipathic helices were oriented toward the interior of the alpha helical ring but for the N-terminal part of apoA-I; the first 32 residues of the N-terminus were rotated, as in the lipid-mimetic solution NMR structure of apoA-I [22], [73], in order to have their hydrophobic face oriented towards the lipid droplet surface. These all-atom models of apoA-I were coarse grained using a pre-released version of the MARTINI force field for proteins [40] for the assignment of beads to every amino acid residue (see also ref. [36]). To obtain the initial structure of the full HDL, two coarse grained apoA-I molecules were added to the equilibrated lipid droplet (discussed above) in a double-belt conformation at a distance of 4 nm from each other.
After energy minimization, the HDL particle was subjected to very short equilibration runs using different time steps to get a stable system for a simulation with a time step of 25 fs. Finally, the particle was simulated for a total of 19 s, of which the last 4 s was used for analysis.
To confirm the validity of the results, the simulations for HDL discussed in this article were complemented by several additional simulations that were started from different initial configurations. Each simulation covered a multi-microsecond time scale, and the results were found to be consistent with those discussed in this article.
The molecular dynamics simulations were performed with the GROMACS 3.3.1 package [74]. Time steps of 20 fs and 25 fs were used for integrating the equations of motion of the lipid droplet and of the full HDL, respectively. For production runs, the Nosé-Hoover thermostat [75], [76] and the Parrinello-Rahman barostat [77] were used to ensure proper NpT conditions ( K, atm). Water and the lipids were coupled to separate thermostats, and the whole system was coupled to the barostat isotropically. Time constant of ps was used for all couplings. For non-bonded interactions, we used the standard distance of 1.2 nm [36]. The Lennard-Jones interaction was shifted smoothly to zero after 0.9 nm.
The equilibration of the simulated lipoparticles was monitored through the numbers of intermolecular contacts between different lipid types and the radial density distributions as a function of time. The intermolecular contacts between different molecular groups were calculated using a 0.8 nm cutoff for all beads. The radial density distributions describe the number densities of the coarse-grained beads. The orientation order of CHOL and CE ring structures was measured by the order parameter , where is the angle between the molecular axis and the effective normal of the lipoparticle at the location of the molecule in question. Being more specific, the molecular axis in this definition for CHOL is drawn from the beginning of CHOL (carbon in the ring of CHOL attached to the short chain) to the carbon connected to the hydroxyl group. The effective normal is the vector from the center of mass (COM) of the lipoparticle to the center of the molecular axis. For CE, an additional measure is the angle between the molecular (ring, see above) axis and the vector from the beginning to the end of the oleoyl chain.
Diffusion was analyzed by measuring the jump-length distributions of the COM positions of the lipids over a time scale . Two types of Gaussian functions were fitted to the distributions, the two-dimensional (2d):and the three-dimensional (3d):
The diffusion coefficient from the best fitting function ( or ) is reported, which in practice means that lipids at the water-lipid interface were found to undergo 2d diffusion, while those under the interface diffused in a 3d manner. Also, different time scales were tested and the measured was observed to level off at long times, an indication of diffusive behavior in the hydrodynamic (long-time) limit. “Long” times here refer to times of the order of 100 ns, where is found to level off to a well defined constant value. Examples of data for and are shown in SI, including also a more detailed description of how to choose the diffusion time scale in the intermediate region under the lipid-water surface (see Text S1, and Figure S6).
In many-component systems such as the present one, the diffusion of different molecular components may take place at different rates, and it is not obvious that CG models account for this aspect correctly. For the MARTINI model used here, we have previously confirmed that this is not an issue. For instance, Niemela et al. [78] recently used atomistic and coarse-grained models to show that the protein diffusion coefficient was about 10 times smaller compared to that of lipids, and the diffusion mechanisms of lipids and proteins was similar in both models. Ramadurai et al. [79] studied the influence of membrane thickness (hydrophobic mismatch) with several peptides using both FCS measurements and coarse-grained simulations and found essentially quantitative agreement for the peptide diffusion coefficients after the MARTINI results had been scaled by a factor of 4. The simulations were also in agreement with experiments for the trend predicted with increasing hydrophobic mismatch. Further, Apajalahti et al. [61] considered the lateral diffusion of lipids in many-component protein-free membranes and found the diffusion of lipids in raft-like membrane domains (in the liquid-ordered phase) to be about 10 times slower compared to diffusion in domains that were in the liquid-disordered phase. Therefore, diffusion has been studied in several multi-component lipid systems, and the MARTINI models have been found to be consistent with experiments.
The solvent accessible surface area (SASA) of hydrophobic, hydrophilic and all-protein residues were measured using a radius of the solvent probe of 0.56 nm (the all-atom 0.14 nm radius of the solvent probe is converted to the coarse-grained one because one water bead corresponds to four water molecules) inside the GROMACS program gsas. The SASA values were averaged over the entire trajectory used for analysis. The root mean square fluctuations (RMSFs) of protein alpha carbons were measured for both apoA-I chains to monitor protein flexibility.
Lipid-protein interactions were monitored through the number of intermolecular contacts and their lifetimes for every lipid component with the protein. Annular lipid molecules, defined as those with any bead within 8 Å of any protein bead, were monitored over the analyzed trajectory. The average percentage of the number of contacts of each lipid component with the protein residues were estimated separately for each of the following moieties of every lipid component: POPC (polar head group, glycerol backbone, oleoyl and palmitoyl chains), PPC (polar head group, glycerol backbone and palmitoyl chain), CHOL (short acyl chain, sterol ring), CE (short acyl chain, sterol ring, and oleate chain) and TG (glycerol backbone, sn–1, sn–2, and sn–3 chains). Cholesterol-protein interactions were also tested by measuring the average number of contacts per protein residue of the cholesterol molecule with hydrophilic and hydrophobic protein residues. Additionally, for evaluation of CHOL-protein and CE-protein lifetimes, we accounted for cases where the distance between the molecules fluctuated around 8 Å: for an annular lipid, if its distance from apoA-I exceeded 8 Å temporarily for less than 10 frames (0.1 ns), the coupling was considered unbroken.
Here our primary interest is the lipid part of HDL, for which reason we have used the standard CG MARTINI model which does not enforce the full secondary structures in apoA-I. This computational efficient approach allows us to focus on generic issues such as the partitioning of lipids around apoA-I, as well as the influence of apoA-I on the disttributions of lipids in a droplet. By fine graining our equilibrated structures back to atomistic level, one could employ atom-scale simulations to elucidate the more detailed aspects of the system.
|
10.1371/journal.pgen.0030139 | Mutation in Mouse Hei10, an E3 Ubiquitin Ligase, Disrupts Meiotic Crossing Over | Crossing over during meiotic prophase I is required for sexual reproduction in mice and contributes to genome-wide genetic diversity. Here we report on the characterization of an N-ethyl-N-nitrosourea-induced, recessive allele called mei4, which causes sterility in both sexes owing to meiotic defects. In mutant spermatocytes, chromosomes fail to congress properly at the metaphase plate, leading to arrest and apoptosis before the first meiotic division. Mutant oocytes have a similar chromosomal phenotype but in vitro can undergo meiotic divisions and fertilization before arresting. During late meiotic prophase in mei4 mutant males, absence of cyclin dependent kinase 2 and mismatch repair protein association from chromosome cores is correlated with the premature separation of bivalents at diplonema owing to lack of chiasmata. We have identified the causative mutation, a transversion in the 5′ splice donor site of exon 1 in the mouse ortholog of Human Enhancer of Invasion 10 (Hei10; also known as Gm288 in mouse and CCNB1IP1 in human), a putative B-type cyclin E3 ubiquitin ligase. Importantly, orthologs of Hei10 are found exclusively in deuterostomes and not in more ancestral protostomes such as yeast, worms, or flies. The cloning and characterization of the mei4 allele of Hei10 demonstrates a novel link between cell cycle regulation and mismatch repair during prophase I.
| Human infertility and reproductive complications have devastating social and monetary costs. Errors in meiosis during reproduction may lead to birth defects, spontaneous abortion, or infertility. Many of the genes essential for meiosis function in DNA repair and mutations in several of these genes have been shown to contribute to cancer. The identification of the genes necessary for normal meiosis is an important goal and will potentially influence the fields of reproductive and cancer biology. In this study, genetic screens in mice have generated the mutation mei4. mei4 causes male and female sterility by disrupting meiosis and altering the function of the DNA repair system known as mismatch repair. We have identified the causative mutation behind the mei4 phenotype in a gene called Human Enhancer of Invasion 10 or Hei10. This work demonstrates that Hei10 is essential for the completion of meiosis and that it functions to coordinate the DNA repair system and the progression of the cell cycle during meiosis.
| The successful completion of meiosis I (MI) in the vast majority of multicellular eukaryotes requires replication of the genome, proper signals from somatic support cells, synapsis, and crossing over between homologous chromosomes. The progression of the distinct phases of MI is thought to be orchestrated by periodic synthesis and degradation of cyclins, the activity of cyclin dependent kinases (CDKs), the correct establishment of the synaptonemal complex (SC), and the proper functioning of the DNA repair machinery during replication and recombination. The accuracy of the meiotic cell cycle in mammals is monitored both during recombination via the pachytene checkpoint in prophase I or during the spindle checkpoint at metaphase I (reviewed in [1,2]). While meiocytes of both sexes are sensitive to early recombinational repair defects, the pachytene checkpoint in males is more sensitive to synapsis defects per se than in females [2]. The components of the mammalian pachytene checkpoint and the nature of the fundamental molecular communications triggering cell cycle arrest after recognition of DNA anomalies in prophase I remain elusive.
The DNA repair machinery is essential for meiotic progression. Genetically programmed double strand breaks (DSBs) are generated and processed by SPO11 [3,4]. DSB formation is followed by recombinational repair facilitated by the RecA homologs RAD51 and DMC1, whose protein products colocalize with ∼300 early recombination nodules on chromosome cores during leptonema [5,6]. Both are subsequently replaced by the single-stranded binding protein replication protein A (RPA) [5,6]. Foci containing the MutS homolog MSH4 colocalize with RAD51 foci during zygonema and persist into the onset of early pachynema [7,8]. Roughly half of MSH4 foci appear to be involved in reciprocal recombinations between homologous chromosomes, while the others result in non-crossover gene conversion events [9]. MSH4 and MSH5 form a heterodimeric complex thought to mediate recognition of the Holliday junction [10]. Both the Msh4−/− and Msh5−/− knockout mice fail to synapse appropriately and arrest in pachynema [7]. Temporally, Msh4 and Msh5 mark the first appearance of the mismatch repair (MMR) system in prophase I.
During pachynema, the protein products of the MutL homologs Mlh3 and Mlh1 [6,11,12] colocalize with the sites of reciprocal recombination and mark the second appearance of the MMR system in prophase I. MLH3/MLH1 foci appear in early pachynema and are present until early diplonema. Targeted deletion of Mlh3 causes arrest at metaphase I and prevents MLH1 recruitment to the chromosome [11]. Meiocytes deficient in MLH1 fail to complete meiosis owing to loss of bivalent maintenance and disorganized univalents at metaphase I [13–16]. As evidenced by their respective knockout phenotypes, the spatiotemporal progression from foci containing MSH4/MSH5 to those containing MLH3/MLH1 appears to be a critical event in the transition from susceptibility to a pachytene checkpoint to that of a spindle checkpoint. Interestingly, little is known about the cell cycle-related molecular mechanisms that are responsible for this transition.
In addition to the role of the DNA repair machinery, meiosis is regulated in part by the synthesis and ubiquitin-mediated degradation of cyclins and their functional relationship with CDKs. There are three cyclins (A1, B1, and B3) that are necessary for or are specifically expressed during mammalian meiosis (reviewed in [17,18]). Several CDKs have been implicated in the progression of meiosis, including CDK1, CDK2, CDK4, and CDK6 [19]. CDK2 and CDK4 are essential for proper meiosis as both Cdk2−/− and Cdk4−/− mice are sterile [20–22]. Cdk2−/− spermatocytes arrest in mid-pachynema, whereas Cdk2−/− oocytes are lost perinatally [20]. Cdk4−/− mice have reduced viability and display degeneration of primary spermatocytes [21]. Interestingly, CDK2 colocalizes with MLH1 at sites of reciprocal recombination [23]. Beyond this, the role of CDK2 during the progression of MMR in prophase I remains unclear.
Ubiquitin-mediated proteolysis is thought to ensure the properly timed involvement of the cyclins in the meiotic cell cycle. For example, limited proteolytic degradation of meiotic cyclin B1 (CCNB1) is necessary for homolog disjunction and release from MI and the exit from MII in yeast and mammals [17,24]. Ubiquitin-mediated proteolysis involves transfer of ubiquitin by a ubiquitin-activating enzyme (E1) to a ubiquitin conjugating enzyme (E2), which in turn assembles a ubiquitin chain onto a substrate recruited by a substrate-specific E3 ligase (reviewed in [25]). To date however, very few interactions have been identified between substrate-specific E3 ligases and meiotic cell cycle regulators. One such potential interaction involves the anaphase-promoting complex and CCNB1. The anaphase promoting complex is a large multiprotein complex containing E3-ubiquitin ligase activity thought to be responsible for degradation of meiotic cyclins, including CCNB1 [26].
Evidence is emerging that ubiquitin conjugation by non-anaphase promoting complex-related pathways occurs during gametogenesis. Male sterility results from targeted disruptions of the ubiquitin-like DNA repair gene Rad23b or the Ubr2 ubiquitin ligase via failure to initiate spermatogenesis and failure to complete MI, respectively [27,28]. Further, mutations in Hr6b (mouse RAD6 homolog) [29], Siah1a [30], and F-box protein family members result in MI defects [31], particularly in males. Despite these phenotypes, none of the above mentioned genes have been directly connected phenotypically or mechanistically to the progression of the cell cycle machinery during meiosis.
Previously, we reported the results of forward genetic screens to identify mutant alleles of genes necessary for the early stages of MI that lacked recognizable orthologs in other model organisms [32]. Here we report on the characterization of one of these alleles called mei4. Mutant spermatocytes are defective in maintaining interhomolog associations at MI due to a lack of crossing over. Positional cloning led to the identification of a mutation in the mouse ortholog of Human Enhancer of Invasion 10 (Hei10). This gene encodes a putative E3 ligase that has not previously been implicated in meiotic recombination. Defects in Hei10 result in aberrant MMR protein and CDK localization. We proposed that the function of Hei10 during MI reveals a link between cell cycle regulation and the accurate resolution of reciprocal recombination intermediates.
mei4/mei4 male and female mice are sterile and do not exhibit any other obvious somatic defects. Mutant males produce no sperm, have small testes (unpublished data), and spermatocytes undergo a metaphase I arrest (Figure 1A and 1B) [32]. In contrast to normal mice (Figure 1A), mutant testes (Figure 1B) contained no round, elongating (Figure 1A, white arrows), or further differentiated spermatids. Unlike the normal telophase and metaphase structures present in mei4/+ controls (Figure 1A, black arrow and black arrowheads, respectively), mutant metaphase spermatocytes contained abnormal metaphase and anaphase-like cells (Figure 1B, black arrows). The incidence of apoptosis was significantly higher in mei4/mei4 versus wild-type (+/+) tubules (black arrows, Figure 1D versus 1C) (p = 3.28 × 10−15). The gross morphology and histopathology of ovaries from mei4/+ and mei4/mei4 females did not differ (Figure 1E and 1F). Heterozygous and mutant oocytes matured in vitro resumed meiosis as characterized by metaphase plate formation at MI (Figure 1G, white arrow). However, like spermatocytes, most chromosomes in mei4/mei4 oocytes failed to congress to, and align at, the metaphase plate (Figure 1H, white asterisks). Bivalent alignment was rare but did occur (Figure 1H, white arrow).
Despite severe metaphase abnormalities, mei4/mei4 oocytes developed into two-celled embryos at a frequency similar to mei4/+ oocytes (63% versus 80.2% respectively; p = 0.1769). However, only 5.6% of mei4/mei4 two-cell embryos developed into blastocysts, versus 74% of the two-cell stage embryos from mei4/+ oocytes (p = 0.0001) (Table 1). Therefore, the developmental competency of embryos derived from mei4/mei4 oocytes was lost between the two-cell and blastocyst stage.
To determine the cause of the aberrant MI in mei4/mei4 spermatocytes, we prepared metaphase spreads. Whereas nineteen metaphase bivalents were present in mei4/+ spreads (Figure 2A), mei4/mei4 spreads contained primarily univalents (Figure 2B). Bivalents were rarely present (Figure 2B, white arrow). The total number of condensed chromosomes (univalents + bivalents) approached 40. These observations suggest that either homologous chromosomes never paired and/or synapsed or did so but failed to remain attached upon entry to metaphase.
To examine synapsis, we used antibodies to detect the distribution of SC proteins along mutant and wild-type prophase I chromosomes. SYCP3 marks the axial elements of the SC [33]. Wild-type (+/+) and mei4/mei4 spermatocyte chromosome spreads labeled with anti-SYCP3 antibodies revealed the presence of pachytene stage nuclei containing 19 paired chromosomes plus an apparent X-Y pair (Figure 2C and 2D, respectively). However, there was premature homolog separation in diplotene mei4/mei4 spermatocytes (Figure 2E) versus +/+ (Figure 2F). The distribution of SYCP1, a SC transverse element protein that is indicative of synapsis [6,34], was normal in mei4/mei4 versus mei4/+ spermatocytes (Figure 2H versus 2G).
Since mutant spermatocytes are capable of homolog synapsis but defective in maintaining interhomolog attachment, we examined the progress of recombination in these nuclei. RAD51 foci were distributed similarly along zygotene chromosome cores in control and mei4/mei4 spermatocyte nuclei (Figure 3A and 3B), indicating that DSBs formed and that recombinational repair was initiated. We then examined the distribution of two markers of recombination nodules, the MMR molecules MLH3 and MLH1 [6,11]. Mid-pachytene mei4/+ spermatocytes had 1–2 MLH3 foci on each synapsed chromosome (Figure 3C) and ∼23 per nucleus, typical for wild-type meioses [6]. However, no MLH3 foci were evident in mei4/mei4 (Figure 3B). Similar results were obtained with MLH1 (Figure 3E and 3F).
mei4 was previously mapped to a large region on Chromosome 14 between the markers D14Mit99 and D14Mit266 (Figure 4) [32]. In this study, we refined the genetic region containing mei4 to a 0.7-cM (four recombinants out of 572 meioses) region corresponding to 2.33 Mb between the markers D14Mit101 and D14Mit183 on Chromosome 14 (Figure 4A). The mouse Ensembl database (http://www.ensembl.org, version 43, NCBIM36) annotates 62 Refseq genes (Table S1) and 11 other Ensembl annotated genes in this region including a cluster of 28 olfactory receptors. Potential candidate genes for mei4 were prioritized on the basis of known or predicted meiotic function and/or germ tissue expression pattern. For novel genes or genes of unknown function, protein-protein BLAST (blastp) comparisons were made to identify functional domains or regions of homology to known genes or motifs. Three genes warranted further attention under these criteria: poly ADP ribosylase 2 (Parp2), apurinic/apyrimidic endonuclease 1 (Apex1), and the Ensembl novel gene Gm288 (Table S1, numbers 33, 35, and 41 in bold-face type). We performed reverse-transcriptase (RT)-PCR on testis RNA from mei4/+ and mei4/mei4 animals to detect possible expression variations in Parp2 and Apex1 (Figure 4B, rows 2 and 3, respectively). Primer pairs spanning multiple exons were used for both genes. Both genes appeared to be expressed normally in control and mutant testes. We sequenced all exons of both genes and no mutations were found.
Gm288 encodes the mouse ortholog of human CCNBP1IP1 (CCNB1 interacting protein 1), also called HEI10 [35]. Using RT-PCR, we found an altered length transcript present in both mei4/+ and mei4/mei4 mutants versus +/+ animals (Figure 4B, row 4). Using primers specific to exons 1 and 2, the expected 430-bp product was seen in +/+ and mei4/+ animals. In addition, a 358-bp fragment was also present in mei4/+ heterozygotes (Figure 4C). In mei4/mei4 testes, we observed only the 358-bp product (Figure 4C). RT-PCR analysis of multiple tissue types revealed that Hei10 is transcribed prominently in the testes and 17-d embryo (corresponding to prophase I in females) and to a much lesser degree in other tissues (Figure 4D).
Sequencing of Hei10 genomic DNA from control and mutant mice revealed a G>T transversion at bp 298 (from the start of the first coding exon, G298T) in mei4/mei4 animals (Figure 5A), which corresponds to the 5′ splice donor site at the 3′ end of exon 1 (Figure 5B, Transcript). Sequencing of cDNA from control and mutant animals showed that G298T causes the use of a cryptic splice donor site 72 bp upstream (GT in position 226) in exon 1. This results in an in-frame, 24 amino acid deletion in the predicted protein product (Figure 5B, Protein, amino acids 76–100) that removes a putative cyclin/CDK interaction domain, the RXL motif (Figure 5B, red RAL) [35–39]. We will refer to the above-described allele as Hei10mei4.
To investigate the possibility of an error in the progression of the cell cycle, we examined the localization of CDK2, a CDK essential for meiosis [20], on spermatocyte chromosomes (Figure 6). CDK foci are normally observed at the telomeres, at one to two interstitial sites on each synapsed bivalent, and on the asynapsed portions of the X and Y chromosomes in males [23]. Further, CDK2 colocalizes on pachytene chromosomes with MLH1 [23]. Mid-pachytene +/+ spermatocytes displayed CDK2 foci at interstitial sites (Figure 6A, white arrows), telomeres (Figure 6A, open arrowheads), and the X-Y body (Figure 6A, closed arrowheads). In contrast, CDK2 foci were seen at telomeric sites and on the X-Y body on mei4/mei4 pachytene chromosomes (Figure 6B, open arrowheads and white arrow, respectively) but were rarely seen at interstitial sites.
In this study, we have demonstrated that mei4/mei4 spermatocytes undergo a uniform early metaphase I arrest marked by aberrant chromosome congression (Figure 1). Mutant oocytes show a similar configuration but fail to arrest at early metaphase I and are competent, postfertilization, to reach the two-cell embryonic stage. Immunocytochemical labeling of microspread spermatocytes and 4′ , 6-diamidino-2-phenylindole, dihydrochloride (DAPI) labeling of metaphase chromosomes demonstrates that maintenance of bivalents is lost in all but a few chromosome pairs. Our data show that the high prevalence of univalents in late diplonema and metaphase I are due to a lack of chiasmata.
Failure to form chiasmata can arise from defects in the DSB system, the MMR system, or in the complexes required for proper SC formation [4,11,15]. Normal localization of RAD51 and SYCP3/SYCP1 in mei4/mei4 spermatocyte nuclei indicates that DSB repair and SC formation are not disrupted. However, MLH3 and MLH1 fail to form foci in mei4/mei4 spermatocyte nuclei, suggesting that defects in the MMR system cause chiasmata failure and subsequent metaphase arrest in males (Figure 3).
Sequencing of Hei10mei4 cDNA identified a shortened transcript produced in the testes. When translated, this mRNA would produce a protein missing 24 amino acids (residues 76–100) in a highly conserved portion of the N-terminal region of HEI10 (Figure 7). This missing region includes a RXL motif (asterisks, amino acids 91–93, RAL) with predicted cyclin/CDK-binding activity [35]. RXL domains are found in CDK substrates such as p107, p21CIP1, and the retinoblastoma protein [36,40–44]. Deletion or mutation of the RXL domain or peptides containing the RXL motif inhibit phosphorylation of the CDK substrate and/or eliminate binding. On the basis of data from a cyclin A/CDK2 crystal structure, the RXL domain contacts a highly conserved remote (relative to the active site) binding domain on the cyclin partner [45]. While we cannot rule out the contributions of the other residues in the region, deletion of the RXL domain in Hei10mei4 would abrogate binding of the E3 to its presumed target, a B-type cyclin.
Alternatively, HEI10 may function as a cell cycle regulator that acts between early and late pachynema. The loss of MLH3/MLH1 foci may be secondary to a possible cell cycle error such as premature exit from G2 before chiasmata are established. As mentioned, cyclins A1, B1, and B3 are essential for meiosis. Deletion of the germ-cell specific cyclin A1 results in a MI spermatogenic arrest in males [46,47]. Cyclin A1 mRNA and protein are expressed primarily from late pachynema through diplonema of prophase I [46,47]. CCNB1 is expressed from mid-pachynema through postmeiotic spermiogenesis [48], and regulation of partial proteolytic degradation of this cyclin is necessary for homolog disjunction in mouse oocytes [24]. In oocytes, CCNB1 is essential for the successful completion of MI as loss of Ccnb1 expression causes accelerated polar body extrusion and inability to enter G2–metaphase meiosis II [49]. Recently, a third mammalian B-type cyclin, cyclin B3 (Ccnb3) has been identified that is highly expressed in prepachytene spermatocytes and the fetal ovary [18].
HEI10 has been shown to associate with exogenously expressed human CCNB1 in mitotic cells and in yeast two-hybrid assays utilizing a somatic cell-derived library [35]. Golemis and colleagues suggested that the human protein product of HEI10 functions as a mitotic E3 ubiquitin ligase for CCNB1, ensuring its degradation and thus transition from metaphase I to interphase of meiosis II [35]. In the male germline, however, CCNB1′s activity is highest in postmeiotic spermatids [48], which is temporally inconsistent with a potential role of HEI10 as a CCNB1 E3 ligase. Further, since Ccnb1 and Hei10 are expressed in many somatic cell types (Hei10, Figure 4D) and mutation of Ccnb1 causes embryonic lethality [50], one might expect a more pleiotropic somatic phenotype in the Hei10mei4 mutant mice if CCNB1 were the primary target.
Koff and colleagues identified mammalian CCNB3, a meiotic B-type cyclin that is expressed maximally during the leptotene to zygotene transition [18], a point just prior to the first observed phenotypic defect in Hei10mei4 homozygotes. This raises the possibility that CCNB3 is the target of HEI10′s B-type cyclin E3 ligase activity in meiosis. Hei10mei4 has a deletion in the putative cyclin-B interaction domain that contains an RXL motif. We speculate that this mutation abrogates an association between HEI10 and CCNB3. This scenario may explain the observed defects in crossing over, homolog attachment, and congression to the metaphase plate. Alternatively, normal HEI10 may target any of a number of cyclin-B related sequences [51] found in the genome but as yet unstudied in meiosis.
In females, where anaphase I begins despite these errors, missegregation is the inevitable result. Interestingly, the Hei10mei4 phenotype in oocytes is similar, but not identical, to that of Mlh3−/− and Mlh1−/− mice. Only 7% of Mlh3−/− and 13.6% of Mlh1−/− oocytes reach the two-cell stage, and none reach the four-cell stage [11,15], whereas 63% of Hei10mei4 zygotes reach the two-cell stage with a small portion (5.6%) even capable of forming a blastocyst. This difference may suggest that HEI10 is not directly involved in MMR but rather serves to couple MMR to the cell cycle machinery. Further, in Mlh3−/− and Mlh1−/− mice, expression of both proteins is abrogated. Currently, we have no data suggesting that MLH3 or MLH1 protein levels are altered in Hei10mei4 mice, only that their localization with respect to the chromosome cores has changed. Noncore-associated MLH3 and MLH1 may be present and may affect the progression of late prophase or metaphase I in Hei10mei4 females.
The genetic and cytological data presented here are supportive of a key role for Hei10mei4 in mouse meiosis. Toby and colleagues previously demonstrated a role for (human) HEI10 in the mitotic cell cycle [35]. Our data show that Hei10 is transcribed in many somatic tissues but at considerably lower levels than in testes (Figure 4D). While the effect of the mutation in meiosis is dramatic, we have not identified any obvious defects in Hei10mei4 mice other than infertility. One possible explanation is that while Hei10 is expressed in many tissues at lower levels than in testis, its hypothesized in vivo target (CCNB3) is primarily expressed in germ cells. Thus, the effects in somatic cells may be subtle.
Alternatively, as a putative B-type E3 ligase, HEI10 may play an essential role maintaining accurate euploid segregation at the meiotic prophase I-metaphase I boundary and the mitotic G2–metaphase boundary. Errors in either, even if relatively rare, could lead to aneuploid segregation and/or neoplasia. Notably, Mine and colleagues have shown that HEI10 resides at a fusion breakpoint in some uterine leiomyomas, a benign tumor of the reproductive system [52]. We have not noticed an elevated tumor incidence in Hei10mei4 mice.
Several years ago, we initiated a forward genetic screen with the goal of identifying novel genes required for gametogenesis, with an emphasis on those required for meiosis [32]. Although meiosis is a highly conserved process, it is noteworthy that two (mei4 and Mei1 [53]) mutations cloned from these screens do not have direct orthologs in Saccharomyces cerevisiae, Drosophila melanogaster, or Caenorhabditis elegans. Hei10 is conserved among deuterostomes but has not been found in protostomes or unicellular organisms and may serve as a marker of the modification of sexual reproduction during a major transition in the animal lineage. Among deuterostomes, sequence identity is highest in the N-terminal portion of the protein containing the RING finger and a coiled coil domain (Figure 7, amino acids 1–197). The C-terminal portion (residues 198 to the end) of the protein exhibits reduced sequence identity when compared to the N terminus. Interestingly, Hei10 orthologs have not been found in Gallus gallus or Taeniopygia guttata (zebra finch). These results may reflect incomplete sequence coverage in the Hei10 syntenic region in chicken or zebra finch or may be the result of divergent meiotic evolution in avian taxa. The identification of Hei10mei4 and Mei1 validate the efficacy of N-ethyl-N-nitrosourea screens in mammals as a tool to identify novel genes required for meiosis.
The Hei10mei4 phenotype sheds light on the role of CDK2 in meiotic progression. Cdk2−/− mice fail to complete MI because of an arrest in early prophase I [20] by presumably triggering the pachytene checkpoint. CDK2 foci are normally found at telomeric sites from zygonema through diplonema but at interstitial crossover sites and the unsynapsed portion of the X and Y chromosomes only from mid-pachynema through early-diplonema [23]. In mei4/mei4 males, which arrest at metaphase I, CDK2 is normally localized to the telomeres and sex chromosomes but, interestingly, is absent at interstitial sites (Figure 6B). Therefore, in the mei4/mei4 background, the presence of CDK2 at the telomeres and/or the XY body, but not at the interstitial sites, is necessary to bypass the pachytene checkpoint. The correlation of the loss of CDK2 and MLH3/MLH1 foci at interstitial sites on mei4/mei4 spermatocyte chromosome cores suggests a role for CDK2 mediation of MMR during prophase I.
Hei10mei4 provides evidence of a functional connection between the cell cycle machinery via CDK2 (and possibly via B-type cyclins) and the DNA repair machinery during prophase I. The Hei10mei4 variant has a deletion in the proposed B-type cyclin interaction domain (RXL). CCNB3 has been shown to interact with CDK2 [18]. CDK2 also interacts with MLH1 at sites of recombination in mid-late pachynema [23] and is essential for meiosis [20]. We envision a mechanism by which HEI10 mediates the destruction of CCNB3, permitting CDK2 association with MLH1 or other MMR proteins in recombination nodules on the chromosome cores (Figure 8A). As suggested by Ashley and colleagues, this interpretation is consistent with CDK2 having a substrate in the recombination nodules [23]. In our model, HEI10mei4 fails to recruit CCNB3 to the E2 ubiquitin conjugating enzyme causing the accumulation or mislocalization of CCNB3 during early pachynema. Elevated levels of CCNB3 may preclude CDK2 association with the interstitial sites on chromosome cores and association with the MMR machinery leading to a failure of the MMR system during recombination (Figure 8B). We are currently testing these hypothesized interactions in vitro. The precise in vivo order of these events and molecular interactions involving CCNB3 are not well resolved and await the development of high-quality antibodies to CCNB3.
In conclusion, we have demonstrated a novel mouse mutation (Hei10mei4) that disrupts chiasmata formation and contributes to aberrant meiotic chromosome congression in mice via an interruption of both cell cycle machinery and the MMR system. We have linked the defects to Hei10, a putative E3 ubiquitin ligase with B-type cyclin specificity. They also suggest a role for CDK2 as regulator of recombination competence. These results provide new insights into the mechanistic control and the cell cycle regulation of mammalian meiotic DNA repair systems.
Mice were obtained from and maintained at The Jackson Laboratory and the Middlebury College Research Animal Facility according to the procedures outlined by the IACUC committees at both institutions.
Ovaries were removed from females 44 h after treatment of 3-wk-old female mice with equine chorionic gonadotropin (Organon, http://www.organon.com) and manually disrupted to release oocytes. The released oocytes were cultured as described to test for spontaneous resumption of meiosis [54,55]. The oocytes were examined using a stereomicroscope after 15 h of culture and assessed for germinal vesicle breakdown, indicative of the resumption of meiosis and the presence of a polar body, which is usually characteristic of progression to metaphase II.
Oocytes at metaphase II were washed three times in minimum essential medium (MEM) supplemented with 3 mg/ml BSA. In vitro fertilization and culture were performed as described previously [56,57]. Eggs were removed from fertilization drops after 4–6 h, rinsed twice in 2.5 ml MEM, and cultured overnight in a borosilicate tube with 1 ml of fresh medium. At 25 h postfertilization, cleavage-stage embryos were rinsed twice with KSOM medium supplemented with essential and nonessential amino acids and 1 mg/ml BSA (KSOM/AA) and cultured to the blastocyst stage in 1 ml KSOM/AA medium in borosilicate tubes.
Testes were fixed in Bouin's solution for >24 h before being embedded in paraffin. We cut 5-mm sections, and they were stained with hematoxylin and eosin (H & E). Digital images were obtained with either an Olympus BX51 upright microscope fitted with an Olympus MagniFire CCD (The Jackson Laboratory) or a Zeiss Axioskop 2 plus microscope fitted with a Zeiss Axiocam MRm digital camera using AxioVision 4.4 software (Middlebury College Imaging Facility).
Testes from prepubertal males 17–24 d postpartum (dpp) were microspread and immunolabeled as described previously [58,59] and counterstained with DAPI (Molecular Probes, http://probes.invitrogen.com) to visualize DNA. Antibodies used were anti-b tubulin (Sigma, http://www.sigmaaldrich.com), anti-SYCP3 (0.8mg/ml) (Abcam, http://www.abcam.com), anti-MLH3 (1:500) (gift from P. E. Cohen), anti-MLH1 (1:50) (BD Biosciences, http://www.bdbiosciences.com), anti-CDK2 (1:300) (Santa Cruz Biotechnology, http://www.scbt.com), anti-SYCP1 (1:500) (Abcam), and anti-RAD51 (1:600) (Calbiochem, http://www.emdbiosciences.com).
Oocytes were collected, cultured (described above), fixed, and stained for DNA and b-tubulin as previously described [60].
Testes were obtained from +/+, mei4/+, and mei4/mei4 mutant mice that were between 2 and 4 mo old. Apoptosis in paraffin embedded testes was assessed using ApopTag Plus Peroxidase In Situ Apoptosis Detection kit (Chemicon, http://www.chemicon.com). Tissue section images were taken using an Axiovert 200 microscope and AxioCamMRc camera using AxioVision 4.4 software. Apoptotic cells were imaged and counted in 0.146-mm2 fields of view (200× total magnification) that were representative of the highest level of apoptosis for each section. A total of six such images from each animal for each genotype were compared from two independent and replicable experiments.
Linkage was implicated by the association of phenotype with homozygosity for C57B6/J, the parental strain that was mutagenized. Additionally, heterozygous animals were crossed to wild-type CAST/Ei animals to take advantage of the higher degree of polymorphism between C57BL/6J and CAST/Ei.
Approximately 2-mm tail tips were lysed and directly used as template in subsequent standard PCR reactions. PCR products were separated by electrophoresis on a horizontal 4.0% MetaPhor (Cambrex, http://www.cambrex.com) gel.
C57BL/6J +/+ and mei4/mei4 mutant total RNA was extracted from homogenized testis tissue with RNeasy Mini kit (Qiagen, http://www1.qiagen.com). We then used 3 μg of total RNA as template for RT-PCR that was performed using First-strand cDNA Synthesis kit (Fermentas, http://www.fermentas.com). We amplified 1.5 μl of normalized wild-type and mei4 mutant cDNA in parallel 25-μl PCR reactions to see whether expression of these genes was different in wild-type and mei4 mutant animals. Expression analysis was performed using primers specific for Parp-2, Apex1, Hei10 (Table S2, MEI10cDNA1F.2 and MEI10cDNA2R.1), and beta-actin (Actb). Amplifications of Hei10 mutant transcript were then isolated using the QIAquick PCR Purification kit and bidirectionally sequenced on an Applied Biosystems 3130 Automated sequencer (Applied Biosystems, http://www.appliedbiosystems.com). Relative tissue expression was then analyzed by PCR and electrophoresis with a normalized Multiple Tissue cDNA Panel (BD Biosciences). PanelcDNAEx1F.1 and PanelcDNAExon3R primers (Table S2) were designed to match manufacturer's suggested product size and annealing temperature. Gapdh controls were amplified with 22 cycles while Hei10 transcripts were amplified with 37 cycles using manufacturers cycling parameters.
Sixteen PARP-2 exons, five APEX1 exons, and three Hei10 exons were amplified from genomic DNA template, and products were then isolated with the QIAquick Gel Extraction kit or PCR purification kit (Qiagen). Mutant sequences (in the case of Hei10) were identified and sequencing analysis was repeated with another mutant individual and one heterozygote and compared to sequence gathered from C57BL/6J, Cast/Ei, and C3HeB/FeJ control mice. MEI10Ex1F and MEI10Ex1R (Table S2) were used to amplify and sequence the first exon of Hei10.
To determine significance in our apoptotic cell counts we used a one-tailed, unequal variance Student's t-test. Significance during in vitro fertilization and blastocyst culture was also determined using a one-tailed t-test.
The Ensembl (http://www.ensembl.org) sequence accession information for HEI10 and its orthologs discussed in this paper are as follows: H. sapiens, ENSP00000337396; M. musculus, ENSMUSP00000093622; C. familiaris, ENSCAFP00000008029; B. Taurus, 1ENSBTAP00000012337; M. domestica, ENSMODP00000008960; ENSOANP00000001628; X. tropicalis, ENSXETP00000037001; G. aculeatus, ENSGACP00000025421; D. rerio, XP_691290; C. intestinalis, ENSCINP00000019943; and S. purpuratus, XP_780975. The Ensembl Gene ID for Gm288 is ENSMUSG00000071470. |
10.1371/journal.pcbi.1002771 | Internal Representations of Temporal Statistics and Feedback Calibrate Motor-Sensory Interval Timing | Humans have been shown to adapt to the temporal statistics of timing tasks so as to optimize the accuracy of their responses, in agreement with the predictions of Bayesian integration. This suggests that they build an internal representation of both the experimentally imposed distribution of time intervals (the prior) and of the error (the loss function). The responses of a Bayesian ideal observer depend crucially on these internal representations, which have only been previously studied for simple distributions. To study the nature of these representations we asked subjects to reproduce time intervals drawn from underlying temporal distributions of varying complexity, from uniform to highly skewed or bimodal while also varying the error mapping that determined the performance feedback. Interval reproduction times were affected by both the distribution and feedback, in good agreement with a performance-optimizing Bayesian observer and actor model. Bayesian model comparison highlighted that subjects were integrating the provided feedback and represented the experimental distribution with a smoothed approximation. A nonparametric reconstruction of the subjective priors from the data shows that they are generally in agreement with the true distributions up to third-order moments, but with systematically heavier tails. In particular, higher-order statistical features (kurtosis, multimodality) seem much harder to acquire. Our findings suggest that humans have only minor constraints on learning lower-order statistical properties of unimodal (including peaked and skewed) distributions of time intervals under the guidance of corrective feedback, and that their behavior is well explained by Bayesian decision theory.
| Human performance in a timing task depends on the context of recently experienced time intervals. In fact, people may use prior experience to improve their timing performance. Given the relevance of time for both sensing and acting in the world, how humans learn and represent temporal information is a fundamental question in neuroscience. Here, we ask subjects to reproduce the duration of time intervals drawn from different distributions (different temporal contexts). We build a set of models of how people might behave in such a timing task, depending on how they are representing the temporal context. Comparison between models and data allows us to establish that in general subjects are integrating task-relevant temporal information with the provided error feedback to enhance their timing performance. Analysis of the subjects' responses allows us to reconstruct their internal representation of the temporal context, and we compare it with the true distribution. We find that with the help of corrective feedback humans can learn good approximations of unimodal distributions of time intervals used in the experiment, even for skewed distributions of durations; on the other hand, under similar conditions, we find that multimodal distributions of timing intervals are much harder to acquire.
| The ability to estimate motor-sensory time intervals in the subsecond range and react accordingly is fundamental in many behaviorally relevant circumstances [1], such as dodging a blow or assessing causality (‘was it me producing that noise?’). Since sensing of time intervals is inherently noisy [2], it is typically advantageous to enhance time estimates with previous knowledge of the temporal context. It has been shown in various timing experiments that humans can take into account some relevant temporal statistics of a task according to Bayesian decision theory, such as in sensorimotor coincidence timing [3], tactile simultaneity judgements [4], planning movement duration [5] and time interval estimation [6]–[8].
Most of these studies [3], [4], [6], [8] exposed the participants to time intervals whose duration followed some simple distribution (e.g. a Gaussian or a uniform distribution), and then assumed that the subjects' internal representation of it corresponded to the experimental distribution. As a more realistic working hypothesis, we can expect the observers to have acquired, after training, an internal representation of the statistics of the temporal intervals which is an approximation of the true, objective experimental distribution. It can be argued that this approximation in most cases would be ‘similar enough’ to the true distribution, so that in practice the distinction between subjective and objective distribution is an unnecessary complication. This is not exact though, first of all because it is unknown whether the similarity assumption would hold for complex temporal distributions, and secondly because the specific form of the approximation can lead to observable differences in behavior even for simple cases (see Figure 1).
We propose that understanding how humans learn and approximate temporal statistics in a given context can help explaining observed temporal biases and illusions [9]. Previous studies have shown that human observers exhibit specific idiosyncrasies in judging simultaneity and temporal order of stimuli after repeated exposure to a specific inter-stimulus lag (temporal recalibration) [4], [10], [11], in encoding certain kinds of temporal distributions in the subsecond range [12] or in estimating durations of very rare stimuli (oddballs) [13], so it is worth asking whether people are able to acquire an internal representation of complex (e.g. very peaked, bimodal) distributions of inter-stimulus intervals in the first place, and what are their limitations.
Bayesian decision theory (BDT) provides a neat and successful framework for representing the internal beliefs of an ideal observer in terms of a (subjective) prior distribution, and it gives a normative account on how the ideal observer should take action [14]. A large number of behavioral studies are consistent with a Bayesian interpretation [15]–[17] and some results suggest that human subjects build internal representations of priors and likelihoods [15], [18], [19] or likelihood and loss functions [20]. We therefore adopted BDT as a framework to infer the subjects' acquired beliefs about the experimental distributions. However, the behavior of a Bayesian ideal observer depends crucially not only on the prior, but also on the likelihoods and the loss function, with an underlying degeneracy, i.e. distinct combinations of distributions can lead to the same empirical behavior [21]. It follows that a proper analysis of the internal representations cannot be separated from an appropriate modelling of the likelihoods and the loss function as well.
With this in mind, we analyzed the timing responses of human observers for progressively more complex temporal distributions of durations in a motor-sensory time interval reproduction task. We provided performance feedback (also known as ‘knowledge of results’, or KR) on a trial-by-trial basis, which constrained the loss function, speeded up learning and allowed the subjects to adjust their behavior, therefore providing an upper bound on human performance [22], [23]. We carried out a full Bayesian model comparison analysis among a discrete set of candidate likelihoods, priors and loss functions in order to find the observer model most supported by the data, characterizing the behavior of each individual subject across multiple conditions. Having inferred the form of the likelihoods and loss functions for each subject, we could then perform a nonparametric reconstruction [24] of what the subjects' prior distributions would look like under the assumptions of our framework and we compared them with the experimental distributions. The inferred priors suggest that people learn smoothed approximations of the experimental distributions which take into account not only mean and variance but also higher-order statistics, although some complex features (kurtosis, bimodality) seem to deviate systematically from those of the experimental distribution.
Subjects took part in a time interval reproduction task with performance feedback (trial structure depicted in Figure 2 top; see Methods for full details). On each trial subjects clicked a mouse button and, after a time interval ( ms) that could vary from trial-to-trial, saw a yellow dot flash on the screen. They were then required to hold down the mouse button to reproduce the perceived interval between the original click and the flash. The duration of this mouse press constituted their response ( ms) for that trial. Subjects received visual feedback on their performance, with an error bar that was displayed either to the left or right of a central zero-error line, depending on whether their response was shorter or longer than the true interval duration. In different experimental blocks we varied both the statistical distribution of the intervals, , and the nature of the performance feedback, i.e. mapping between the interval/response pair and the error display, , relative to the zero-error line. For each experimental block, subjects first performed training sessions until their performance was stable (around 500 to 1500 trials), followed by two test sessions (about 500 trials per session). Testing with a block was completed before starting a new one.
Different groups of subjects took part in five experiments, whose setup details are summarized in Table 1 (see also Methods). In brief, Experiment 1 represented a basic test for the experimental paradigm and modelling framework with simple (Uniform) distributions over different ranges. Experiment 2 compared subjects' responses in a simple condition (Uniform) vs a complex one (Peaked, one interval was over-represented), over the same range of intervals. Experiment 3 verified the effect of feedback on subjects' responses by imposing a different error mapping . Experiment 4 tested subjects in a more extreme version of the Peaked distribution. Experiment 5 verified the limits of subjects' capability of learning with bimodal distributions of intervals.
We first present the results of the first two experiments in a qualitative manner, and then describe a quantitative model. Results of the other three experiments that test specific aspects of the model or more complex distributions are presented thereafter.
In the first experiment the distribution of time intervals consisted of a set of six equally spaced discrete times with equal probability according to either a Short Uniform (450–825 ms) or Long Uniform (750–1125 ms) distribution. The order of these blocks was randomized across subjects. The feedback followed a Skewed error mapping . The ‘artificial’ response-dependent asymmetry in the Skewed mapping was chosen to test whether participants would integrate the provided feedback error into their decision process, as opposed to other possibly more natural forms of error, such as the Standard error or the Fractional error (see later, Bayesian model comparison).
We examined the mean bias in the response (mean reproduction interval minus actual interval, , also termed ‘constant error’ in the psychophysical literature), as a function of the actual interval (Figure 3 top). Subjects' responses showed a regression to the mean consistent with a Bayesian process that integrates the prior with sensory evidence [4], [6], [8], [15]. That is, little bias was seen for intervals that matched the mean of the prior (637.5 ms for Short Uniform, red points, and 937.5 ms for Long Uniform, green points). However, at other intervals a bias was seen towards the mean interval of that experimental block, with subjects reporting intervals longer than the mean as shorter than they really were and conversely intervals shorter than the mean as being longer than they really were. Moreover, this bias increased almost linearly with the difference between the mean interval and the actual interval. Qualitatively, this bias profile is consistent with most reasonable hypotheses for the prior, likelihoods and loss functions of an ideal Bayesian observer (even though details may differ).
The standard deviation of the response (Figure 3 bottom) showed a roughly linear increase with interval duration, in agreement with the ‘scalar property’ of interval timing [25], according to which the variability in a timing task grows in proportion to the interval duration.
These results qualitatively suggest that the temporal context influences subjects' performance in the motor-sensory timing task in a way which may be compatible with a Bayesian interpretation, and in agreement with previous work which considered purely sensory intervals and uniform distributions [6], [8], [26].
As in the first experiment six different equally-spaced intervals were used, with two different distributions. However, in this experiment both blocks had the same range of intervals (Medium: 600–975 ms). In one block (Medium Peaked) one of the intervals (termed the ‘peak’) occurred more frequently than the other 5 intervals, that were equiprobable. That is, the 675 ms interval occurred with with the other 5 intervals occurring each with . In the other block (Medium Uniform) the 6 intervals were equiprobable. The feedback gain for both blocks was again the Skewed error map .
Examination of the responses showed a central tendency as encountered in the previous experiment (Figure 4 top). However, despite the identical range of intervals in both blocks, subjects were sensitive to the relative probability of the intervals [27]. In particular, the responses in the Peaked block (light blue points) appeared to be generally shifted towards shorter durations and this shift was interval dependent (see Figure 5). This behavior is qualitatively consistent with a simple Bayesian inference process, according to which the responses are ‘attracted’ towards the regions of the prior distribution with greatest probability mass. Intuitively, the average (‘global’) shift of responses can be thought of as arising from the shift in the distribution mean, from the Uniform distribution (mean 787.5 ms) to the Peaked distribution (mean 731.3 ms); whereas interval-dependent (‘local’) effects are a superimposed modulation by the probability mass assignments of the distribution. This is only a simplified picture, as the biases depend on a nonlinear inference process, which is also influenced by other details of the Bayesian model (such as the loss function), but the qualitative outcome is likely to be similar in many relevant cases.
The standard deviation of the responses showed a significant decrease in variability around the peak for the Peaked condition (Figure 4 bottom; two-sample F-test ). This effect could be simply due to practice as subjects received feedback more often at peak intervals, however the local modulation of bias previously described (Figure 5) suggests a Bayesian interpretation. In fact, because of the local ‘attraction’ effect, interval durations close to the peak would elicit responses that map even closer to it, therefore compressing the perceptual variability, an example of bias-variance trade-off [6].
The results of the second experiment show that people take into account the different nature of the two experimental distributions, in agreement with previous work that found differential effects in temporal reproduction for skewed vs uniform distributions of temporal intervals on a wider, suprasecond range [27]. The performance of the subjects in the two blocks is consistent with a Bayesian ‘attraction’ in the response towards the intervals with higher prior probability mass. Moreover, although the average negative shift in the response observed in the Peaked condition versus the Uniform one might be compatible with a temporal recalibration effect that shortens the perceived duration between action and effect [11], [28], [29], the interval-dependent bias modulation (Figure 5) and the reduction in variability around the peak (Figure 4 bottom) suggest there may instead be in this case a Bayesian explanation.
In order to address more specific, quantitative questions about our results we set up a formal framework based on a Bayesian observer and actor model.
We modelled the subjects' performance with a family of Bayesian ideal observer (and actor) models which incorporated both the perception (time interval estimation) and action (reproduction) components of the task; see Figure 2 (bottom) for a depiction of the generative model of the data. We assume that on a given trial a time interval is drawn from a probability distribution (the experimental distribution) and the observer makes an internal measurement that is corrupted by sensory noise according to the sensory likelihood , where is a parameter that determines the sensory (estimation) variability. Subjects then reproduce the interval with a motor command of duration . This command is corrupted by motor noise, producing the response duration – the observed reproduction time interval – with conditional probability density (the motor likelihood), with a motor (reproduction) variability parameter. Subjects receive an error specified by a mapping and we assume they try to minimize a (quadratic) loss based on this error.
In our model we assume that subjects develop an internal estimate of both the experimental distribution and error mapping (the feedback associated with a response to stimulus ), which leads to the construction of a (subjective) prior, , and subjective error mapping ; the latter is then squared to obtain the loss function. This allows the prior and subjective error mapping to deviate from their objective counterparts, respectively and .
Following Bayesian decision theory, the ‘optimal’ action is calculated as the action that minimizes the subjectively expected loss:(1)where the integral on the right hand side is proportional to the subjectively expected loss. Combining Eq. 1 with the generative model of Figure 2 (bottom) we computed the distribution of responses of an ideal observer for a target time interval , integrating over the hidden internal measurement which was not directly accessible in our experiment.
Therefore the reproduction time of an ideal observer, given the target interval , is distributed according to:(2)Eqs. 1 and 2 are the key equations that allow us to simulate our task, in particular by computing the mean response bias and standard deviation of the response for each interval (Section 1 in Text S1). Eq. 1 represents the internal model and deterministic decision process adopted by the subject whereas Eq. 2 represents probabilistically the objective generative process of the data. Notice that the experimental distribution and objective error mapping do not appear in any equation: the distribution of responses of ideal observers only depends on their internal representations of prior and loss function.
Eqs. 1 and 2 describe a family of Bayesian observer models, a single Bayesian ideal observer is fully specified by picking (i) a noise model for the sensory estimation process, ; (ii) a noise model for the motor reproduction process ; (iii) the form of the prior ; and (iv) the loss function (Figure 6 and Methods). To limit model complexity, in the majority of our analyses we used the same likelihood functions (, and their parameters , ) for both the generative model (Eq. 2) and the internal model (Eq. 1). Analogously, for computational reasons in our basic model we assumed a quadratic exponent for the loss function (Eq. 1); in a subsequent analysis we relaxed this requirement (Section 2 in Text S1).
To study the nature of the internal model adopted by the participants, we performed a full Bayesian model comparison over the family of Bayesian ideal observer models. For each participant we assumed that the sensory and motor noise, the approximation strategy for the priors, and the loss function were shared across different experimental blocks. The model comparison was performed over a discrete set of model components, that is, possible choices for the priors, loss functions and shape of likelihoods (Figure 6). In particular, priors and loss functions did not have continuous parameters, as a parametric model would likely be ambiguous or hard to interpret, with multimodal posterior distributions over the parameters (as multiple combinations of likelihoods, prior and cost function can make identical predictions). Instead, we considered a finite number of parameter-free models of loss function, prior and shape of likelihoods, leaving only two continuous parameters for characterizing the sensory and motor variability.
Both sensory and motor noise were modelled with Gaussian distributions whose means were centered on the interval and whose standard deviations could either be constant or ‘scalar’, that is, grow linearly with the interval (Figure 6 i and ii). We used two parameters, and , which represent the coefficient of variation of the subject's sensory and motor noise. For the scalar case this simply specifies the coefficient of proportionality of the standard deviation with respect to the mean, whereas in the constant case it specifies the proportion of noise with respect to a fixed interval (787.5 ms).
We considered three different possible subjective error metrics corresponding to the Skewed error (the error map we provided experimentally), the Standard error , and a Fractional error (Figure 6 iv), which were then squared to obtain the loss function (see also Methods). Note that scaling these mappings does not change the optimal actions and hence the model selection process.
We compared different approximation schemes for the priors, such as the true discrete distribution (Figure 6 iii, a) or a single Gaussian whose mean and standard deviation matched those of the true prior (b). We also considered two smoothed versions of the experimental distribution with a weak (c) and strong (d) smoothing parameter, or some other block-dependent approximations, e.g. for the Uniform blocks we considered a uniform distribution over the stimulus range (e); see Methods for a full description. To constrain the model selection process, we assumed that subjects adopted a consistent approximation scheme across blocks.
For each participant we computed the support for each model based on the psychophysical data, that is the posterior probability of the model, Pr(model| data). Assuming an a priori indifference among the models, this corresponds (up to a normalization factor) to the model marginal likelihood Pr(data| model), which was obtained by numerical integration over the two-dimensional parameter space ( and ).
We then calculated the Bayesian model average for the response mean bias and standard deviation, shown by the continuous lines in Figure 3 and 4. Note that the Bayesian model ‘fits’ are obtained by computing the marginal likelihood of the models and integrating the model predictions over the posterior of the parameters (model averaging), with no parameter fitting. The mean biases fits show a good quantitative match with the group averages ( for all blocks); the standard deviations are typically more erratic and we found mainly a qualitative agreement, as observed in previous work [6].
For each participant of Experiments 1 and 2 we computed the most probable (i) sensory and (ii) motor likelihoods, (iii) priors and (iv) loss function (Table S1). The model comparison confirmed that the best noise models were represented by the ‘scalar’ variability, which had relevant support for both the sensory component (7 subjects out of 10) and the motor component (8 subjects out of 10). This result is consistent with previous work in both the sensory and motor domain [5], [6], [25], [30]. The most supported subjective error map was the Skewed error (7 subjects out of 10), which matched the feedback we provided experimentally. The priors most supported by the data were typically smooth, peaked versions of the experimental distributions. In particular, according to the model comparison, almost all subjects (9 out of 10) approximated the discrete uniform distributions in the Uniform blocks with normal distributions (same mean and variance as the true distribution; Figure 6 iii top, b). However, in Experiment 2 most people (5 out of 6) seemed to approximate the experimental distribution in the Peaked block not with a standard Gaussian, but with a skewed variant of a normal distribution (Figure 6 iii bottom, d, f and g), suggesting that their responses were influenced by higher order moments of the true distribution and not just the mean and variance (see Discussion).
For Experiment 2 we also relaxed some constraints on the priors, allowing the model selection to pick a Medium Uniform prior for the Medium Peaked block and vice versa. Nevertheless, the model comparison showed that the most supported models were still the ones in which the priors matched the block distribution, supporting our previous findings that subjects' responses were consistent with the temporal context and changed when switching from one block to another (as visible in Figure 4).
To study in detail the internal representations, we relaxed the constraint on the priors. Rather than choosing from a fixed set of candidate priors (Figure 6 iii), we allowed the prior to vary over a much wider class of smooth, continuous distributions. We assumed that the noise models and loss function emerging from the model comparison were a good description of the subjects' decision making and sensorimotor processing in the task. We therefore fixed these components of the observer's model and inferred nonparametrically, on an individual basis, the shape of the priors most compatible with the measured responses (Figure 7; see Methods for details).
Examination of the recovered priors shows that the subjective distributions were significantly different from zero only over the range corresponding to the experimental distribution, with only occasional tails stretching outside the interval range (e.g. Figure 7 bottom left). This suggests that in general people were able to localize the stimulus range in the blocks. The priors did not typically take a bell-like shape, but rather we observed a more or less pronounced peak at the mean of the true distribution, with the remaining probability mass spread over the rest of the range. Interestingly, the group averages for the Uniform priors over the Short, Medium and Long ranges (Figure 7 top right, both, and bottom right, light brown) exhibit very similar, roughly symmetrical shapes, shifted over the appropriate stimulus range. Conversely, the Peaked prior (Figure 7 bottom right, light blue) had a distinct, skewed shape.
To compare the inferred priors with the true distribution, we calculated their distribution moments (Table 2). We found that the first three moments of the inferred priors (in the table reported as mean, standard deviation and skewness) were statistically indistinguishable from those of the true distributions for all experimental conditions (Hotelling's multivariate one-sample test considering the joint distribution of mean, standard deviation and skewness against the true values; for all blocks). This result confirmed the previously stated hypothesis that participants had developed an internal representation which included higher order moments and not just the mean and variance of the experimental distribution. However, when including the fourth moment (kurtosis) in the analysis, we observed a statistically significant deviation of the recovered priors with respect to the true distributions (Hotelling's test with the joint distribution of the first four moments; for all blocks); in particular, the inferred priors seem to have more pronounced peaks and/or heavier tails. First of all, note that the heightened kurtosis is not an artifact due to the averaging process across subjects or the sampling process within subjects, as we averaged the moments computed for each sampled distribution (see Methods) rather than computing the moments of the average distribution. In other words, all recovered priors are (on average) heavy tailed, it's not just the mean prior that it is ‘accidentally’ heavy tailed as a mixture of light-tailed distributions. So this result could mean that the subjects' internal representations are actually heavy-tailed, for instance to allow for unexpected stimuli. However, there could be a simpler explanation that the presence of outliers arise from occasional trivial mistakes of the participants. We, therefore, considered a straightforward extension of our model which added the possibility of occasional ‘lapses’ with a lapse rate , where the response in a lapse trial is simply modelled as a uniform distribution over a wide range of intervals (Section 3 in Text S1). In terms of marginal likelihood, generally the models with lapse performed better than the original models, but with no qualitative difference in the preferred model components. Crucially, we did not observe a significant change in the kurtosis of the recovered priors, ruling out the possibility that the heightened kurtosis had been caused by trivial outliers.
Our analysis therefore showed that, according to the inferred priors, people generally acquired internal representations that were smooth, heavy-tailed approximations to the experimental distributions of intervals, in agreement up to the first three moments.
In our ideal observer model we compared three candidate loss functions: Skewed, Standard and Fractional (Figure 6 iv). The results of the model comparison in the first two experiments with Skewed feedback showed that there was a good match between experimentally provided feedback and subjective error metric. However, we could not rule out the possibility, albeit unlikely, that participants were ignoring the experimental feedback and following an internal error signal that just happened to be similar in shape to the Skewed error. We therefore performed an additional experiment to verify that subjects behavior is driven by the feedback provided.
We again used a Medium Uniform block but now with Standard error as feedback (see Figure S5 in Text S2). The model comparison for this group showed that the responses of 4 subjects out of 6 were best explained with a Standard loss function. Moreover, no subject appeared to be using the Skewed loss function (Table S1). These results confirm that most people correctly integrate knowledge of results with sensory information in order to minimize the average (squared) error, or an empirically similar metric. Furthermore, all inferred individual priors showed a remarkable agreement with a smoothed approximation of the experimental distribution of intervals (Figure 8 top), suggesting that the Standard error feedback may be easier to use for learning. As in the previous experiments, the average moments of the inferred priors (up to skewness) were statistically indistinguishable from those of the true distribution, with a significant difference in the kurtosis (Table 3 left; Hotelling's test, first three moments: ; first four moments: ).
In the Peaked block we did not observe any significant divergence from the Bayesian prediction. However, the ratio of presentations of ‘peak’ intervals (675 ms) to the others was low (1.4) and possibly not enough to induce other forms of temporal adaptation [29], [31]. To examine whether we might see deviations from Bayesian integration for larger ratios we therefore tested another group of subjects on a more extreme variant of the Peaked distribution in which the peak stimulus had a probability of and therefore a ratio of about 4.0. We provided feedback through the Standard error mapping, as the previous experiment had showed that subjects can follow it at least as well as the Skewed mapping.
Due to the large peak interval presentation frequency we had fewer test data points in the model fitting. Therefore, we constrained the model comparison by only considering the Standard loss in order to prevent the emergence of spurious model components capturing random patterns in the data. We found that the recovered internal priors were in good qualitative agreement with the true distribution, with statistically indistinguishable means (Figure 8 bottom, and Table 3; one sample two-tailed t-test ). When variance and higher moments were included in the analysis, though, the distributions were significantly different (Hotelling's test, mean and variance: ; first three moments: ; first four moments: ) suggesting that the distribution may have been ‘too peaked’ to be learnt exactly; see Discussion. Nevertheless, the observed biases of the responses were well explained by the basic Bayesian models (group mean: ), and the standard deviations were in qualitative agreement with the data (Figure S6 in Text S2).
Our previous experiments show that people are able to learn good approximation of flat or unimodal distributions of intervals relatively quickly (a few sessions), under the guidance of corrective feedback. Previous work in sensorimotor learning [15] and motion perception [32] has shown that people can learn bimodal distributions. Whether the same is attainable for temporal distributions is unclear; a recent study of time interval reproduction [27] obtained less definite results with a bimodal ‘V-shaped’ distribution, although training might have been too short, as subjects were exposed only to 120 trials in total and without performance feedback.
To examine whether subjects could easily learn bimodality of a temporal distribution with the help of feedback we tested two new groups of subjects on bimodal distributions of intervals on a Medium range (600–975 ms, as before) and on a Wide range (450–1125 ms), providing in both cases Standard feedback. In the Medium Bimodal block the intervals at 600 and 975 ms had each probability , whereas the other four middle intervals (675, 750, 825, 900 ms) had each probability . In the Wide Bimodal block the six ‘extremal’ intervals (450, 525, 600 ms and 975, 1050, 1125 ms) had each probability whereas the middle intervals had probability . Note that in both cases extremal intervals were four times as frequent as middle intervals.
In the Medium Bimodal block, subjects' responses exhibited a typical central tendency effect (Figure 9 top left) which suggests that people did not match the bimodality of the underlying distribution. To constrain the model comparison we inferred the subjects' priors under the assumption of scalar sensory and motor noise models and Standard loss function, as found by our previous analyses. As before, we first used a discrete set of priors (see Methods) that we used to compute the model ‘fit’ to the data and then we performed a nonparametric inference. The nonparametrically inferred priors for the Medium Bimodal distribution (Figure 9 top right) suggest that on average subjects developed an internal representation that differed from those seen in previous experiments and, as before, we found a good agreement between moments of the experimental distribution and moments of the inferred priors up to skewness (Table 4 left). However, results of the Bayesian model comparison among a discrete class of flat, unimodal or bimodal priors do not support the hypothesis that subjects actually learnt the bimodality of the experimental distribution (data not shown). Part of the problem may have been that in the Medium Bimodal distribution the two modes were relatively close, and due to sensory and motor uncertainty subjects could not gather enough evidence that the experimental distribution was not unimodal (but see Discussion). We repeated the experiment therefore on a wider range with a different group of subjects.
The pattern of subjects' responses in the Wide Bimodal block shows a characteristic ‘S-shaped’ bias profile (Figure 9 top right) which is compatible with either a flat or a slightly bimodal prior. The nonparametrically inferred priors for the Wide Bimodal distribution (Figure 9 bottom right) again suggest that on average subjects acquired, albeit possibly with less accuracy (Table 4 right), some broad features of the experimental distribution; however individual datasets are quite noisy and again we did not find strong evidence for learning of bimodality.
Our results with bimodal distributions confirm our previous finding that people seem to be able to learn broad features of experimental distributions of intervals (mean, variance, skewness) with relative ease (a few sessions of training with feedback). However, more complex features (kurtosis, bimodality) seem to be much harder to learn (see Discussion).
Our main finding is that humans, with the help of corrective feedback, are able to learn various statistical features of both simple (uniform, symmetric) and complex (peaked, asymmetric or bimodal) distributions of time intervals. In our experiments, the inferred internal representations were smooth, heavy tailed approximations of the experimental distributions, in agreement typically up to third-order moments. Moreover, our results suggest that people take into account the shape of the provided feedback and integrate it with knowledge of the statistics of the task in order to perform their actions.
The statistics of the responses of our subjects in the Uniform blocks were consistent with results from previous work; in particular, we found biases towards the mean of the range of intervals (central tendency) [6], [8], [26], [33] and the variability of the responses grew roughly linearly in the sample interval duration (scalar property) [6], [34]. The responses in the Peaked and High-Peaked blocks showed analogous biases, but they were directed towards the mean of the distribution rather than the mean of the range of intervals (the two means overlapped in the Uniform case) [27]. We also observed a significant reduction in variability at the peak. These results were sufficient to suggest that subjects considered the temporal statistics of the context in their decision making processes. We found a similar regression to the mean for a ‘narrow’ bimodal distribution (Medium Bimodal), in qualitative agreement with previous work that found a simple central tendency with a ‘V-shaped’ temporal distribution [27] (although with very limited training, no feedback and a suprasecond range). However, for a bimodal distribution on a wider range we observed ‘S-shaped’ biases which seem compatible with a nonlinear decision making process [15]. However, more refined conclusions needed the support of a formal framework.
Our modelling approach consisted of building a family of Bayesian observer and actor models, which provided us with a mathematical structure in which to ask specific questions about our subjects [35], going beyond mere statements about Bayesian optimality. In particular, we were interested in (1) whether people would be able to learn nontrivial temporal distributions of intervals and what approximations they might use, and (2) how their responses would be affected by performance feedback. Our observer model resembled the Bayesian Least Squares (BLS) observer described in [6], but it explicitly included an action component as part of the internal model. Moreover, to answer (1) we allowed the prior to differ from the experimental distribution, and to study (2) we considered additional shapes for the loss function in addition to the Standard squared loss .
The Bayesian model comparison gave us specific answers for each of our subjects, and a first validation came from the success of the most supported Bayesian observer and actor models in capturing the statistics of the subjects' responses in the task. However, goodness of fit per se is not necessarily an indicator that the components found by the model comparison reflected true findings about the subjects, rather than ‘overfitting’ arbitrary statistical relationships in the data. This is of particular relevance for Bayesian models, because of the underlying degeneracy among model components [21].
Our approach consisted in considering a large, ‘reasonable’ set of observer models that we could link to objective features of the experiment. This does not solve the degeneracy problem per se but it prevents the model comparison from finding arbitrary solutions. In particular, the set of experiments was designed in order to provide evidence that each element of the model mapped on to an experimentally verifiable counterpart; crucially, we found that a change in a component of the experimental setup (e.g. experimental distribution and feedback) correctly induced a switch in the corresponding inferred component of the model (prior and loss function). We also avoided overfitting by limiting our basic models to only two continuous noise parameters, which were then computed through model averaging and further validated by independent direct measures.
To further validate our methods, we directly measured the subject's noise parameters (sensory and motor noise, and ) in separate tasks and compared them with the model parameters , inferred from the main experiments (see Section 4.1 in Text S1 for full description). The rationale is that, in an idealized situation, we would be able to measure some features of the subjects with an objective, independent procedure and the same features would be predictive of the individual performances in related tasks [16]. The measured parameters were highly predictive of the group behavior, and reasonably predictive at the individual level for the sensory parameter, confirming that the model parameters were overall correctly representing objective ‘noise properties’ of the subjects.
Overall, our modelling techniques were therefore validated by (a) goodness of fit, (b) consistency between inferred model components and experimental manipulations, and (c) consistency between the model parameters and independent measurements of the same quantities.
Given the validation of the results of the model comparison, we performed a nonparametric inference of the priors acquired by participants during the task. Other recent works have inferred the shape of subjective ‘natural’ perceptual priors nonparametrically, such as in visual orientation [24] and speed [36] perception, but studies that focussed on experimentally acquired priors mostly recovered them under parametric models (e.g. Gaussian priors with variable mean and variance) [35], [37]–[39]. The nonparametric method allowed us to study the accuracy of the subjects in learning the experimental distributions, comparing summary statistics such as the moments of the distributions up to fourth order. Note that the significance and reliability of the recovered priors is based on the correctness of our assumptions regarding the observer and actor model; unconstrained priors might capture all sorts of statistical details, one of the typical objections to Bayesian modelling [40]. However, by dividing the model selection stage (and its validation) from the prior reconstruction process we prevented the most pathological forms of overfitting.
The internal representations inferred from the data show a good agreement with the central moments of the true distributions typically up to third order (mean, variance and skewness). Subjects however showed some difficulties in learning variance and skewness when the provided distribution was extremely peaked, with a width less than the subjects' perceptual variability. This discrepancy observed in the High-Peaked block may have arisen because (a) the experimental distribution's standard deviation was equal or lower in magnitude compared to the perceptual variability of the subjects (experimental distribution standard deviation: 80.5 ms; subject's average sensory standard deviation at the mean of the distribution: ms; mean sd across subjects) and (b) due to the shape of the distribution, subjects had much less practice with intervals away from the peak. Another explanation is that subjects' representation of relative frequencies of different time intervals was systematically distorted, with overestimation of small relative frequencies and underestimation of large relative frequencies (see [41] for a critical review), but note that this would arguably produce a change in the mean of the distribution as well, which we did not observe.
Moreover, the recovered priors in all blocks had systematically heavier tails (higher kurtosis) than the true distributions. By exploring an extended model that included lapses we ruled out that this particular result was due to trivial outliers in our datasets. However, our results are compatible with other more sophisticated reasons for the heavy tails we recovered, in particular (a) the objective likelihoods might be non-Gaussian, with heavier tails [42], and (b) the loss functions might follow a less-than-quadratic power law [43], hypothesis for which we found some evidence, although inconclusive, by studying observer models with non-quadratic loss functions (Section 2 in Text S1). Experimentally, both (a) and (b) would imply that in our datasets there would be more outliers than we would expect from a Gaussian noise model with quadratic losses.
Our experiments with bimodal distributions show that, although people's responses were affected by the experimental distribution of intervals in a way which is clearly different from our previous experiments with uniform or peaked distributions, the inferred priors in general fail to capture bimodality and are consistent instead with a broad uniform or multimodal prior (where the peaks however do not necessarily fall at the right places). Note that the average sensory standard deviation for subjects in Experiment 5 was ms (Medium Bimodal; mean sd across subjects) and ms (Wide Bimodal), calculated at the center of the interval range. In other words, in both blocks, the centers of the peaks were well-separated in terms of perceptual discriminability (on average by at least four standard deviations). This suggests that most subjects did not simply fail to learn the bimodality of the distributions because they had problems distinguishing between the two peaks.
Lag adaptation is a robust phenomenon for which the perceived duration between two inter-sensory or motor-sensory events shortens after repeated exposure to a fixed lag between the two [10], [11], [44]; see [45] for a review. It is currently uncertain whether lag adaptation is a ‘global’ temporal recalibration effect (affecting all intervals) [46], ‘local’ (affecting only intervals in a neighborhood of the adapter lag) [47], or both. What is clear is that lag adaptation cannot be interpreted as a Bayesian effect in terms of prior expectations represented by the sample distribution of adaptation and test intervals, since its signature is a ‘repulsion’ from the adapter as opposed to the ‘attraction’ induced by a prior [4], [47], [48].
Our experimental setup for the peaked blocks mimicked the distributions of intervals of typical lag adaptation experiments [11], [29], with the adapter interval set at 675 ms (the ‘peak’). However, we did not detect any noticeable disagreement with the predictions of our Bayesian observer model and, in particular, there was no significant ‘repulsion effect’ from the peak, neither global nor local. Our results suggest that people are not subject to the effects of lag adaptation, or can easily compensate for them, in the presence of corrective feedback.
Sensorimotor lag adaptation seems to belong to a more general class of phenomena of temporal recalibration which induce an adjustment of the produced (or estimated) timing of motor commands to meet the goals of the task at hand. In the case of experimentally induced actuator delays in a time-critical task, such as controlling a spaceship through a minefield in a videogame [49] or driving a car in a simulated environment [50], visual temporal information about delays provides an obvious, compelling reason to recalibrate the timing of actions. However, feedback regarding timing performance need not be provided only in temporal ways. Previous studies have shown that people take into account performance feedback (knowledge of results) when the feedback about the timing of their motor response is provided in various ways, such as verbal or visual report in milliseconds [23], [51] or bars of variable length [52]. Interestingly, people tend to also follow ‘erroneous’ feedback [52]–[54]. However, this can be explained by the fact that people's behavior in a timing task is goal-oriented (e.g. minimizing feedback error), and therefore these experiments suggest that people are able to follow external, rather than erroneous, feedback. In fact, when participants are told that feedback might sometimes be incorrect, which corresponds to setting different expectations regarding the goal of the task, they adjust their timing estimates taking feedback less into account [53]. Ambiguity regarding the goal of a timing task with non-obvious consequences – as opposed to actions that have obvious sensorimotor consequences, such as catching a ball – can be reduced by imposing an explicit gain/loss function [5], [55], and it has been found that people can act according to an externally presented asymmetric cost (even though their timing behavior is not necessarily ‘optimal’ [55]).
Our work extends these previous findings by performing a model comparison with different types of symmetric and asymmetric loss functions and providing additional evidence that most people are able to correctly integrate an arbitrary external feedback in their decision process, while executing a sensorimotor timing task, so to minimize the feedback error.
There is growing evidence that many aspects of human sensorimotor timing can be understood in terms of Bayesian decision theory [3], [5], [6]. The mechanism through which people build time estimates, e.g. an ‘internal clock’, is still unclear (see [56] for a review), but it has been proposed that observers may integrate both internal and external stochastic sources of temporal information in order to estimate the passage of time [7], [57].
Inspired by these results, in our work we assumed that people build an internal representation of the temporal distribution of intervals presented in the experiment. However, for all timing tasks in which more or less explicit knowledge of results is given to the subjects (e.g. ours, [6], [26]), an alternative explanation is that people simply learn a mapping from a duration measurement to a given reproduction time (strategy known as table look-up), with no need of learning of a probability distribution [58]. At the moment we cannot completely discard this possibility, but other timing studies have shown that people perform according to Bayesian integration even in the absence of feedback both for simple [4], [8] and possibly skewed distributions [27], suggesting that people indeed take into account the temporal statistics of the task in a context-dependent way. Moreover, previous work in motor learning in the spatial domain has shown that people do not simply learn a mapping from a stimulus to a response, but adjust their performance according to the reliability of the sensory information [15], a signature of probabilistic inference [59]. Analogous findings have been obtained in multisensory integration [18], [60], [61] and for visual judgements (‘offset’ discrimination task) under different externally imposed loss functions [20], crucially in all cases without knowledge of results. All these findings together support the idea that sensorimotor learning follows Bayesian integration, also in the temporal domain. However, the full extent of probabilistic inference in sensorimotor timing needs further study, possibly involving transfer between different conditions in the absence of knowledge of results [58].
Our results answer some of the questions raised in [6], in particular about the general shape of the distributions internalized by the subjects and the influence of feedback on the responses. An avenue for further work is related to the detailed profile of the likelihoods and possible departures from the scalar property [34], [62] (see also Section 4 in Text S1), especially in the case of complex experimental distributions. It is reasonable to hypothesize that strongly non-uniform samples of intervals might affect the shape of the likelihood itself, if only for the simple reason that people practice more on some given intervals. Cognitive, attentional and adaptation mechanisms might play various roles in the interaction between nonuniform priors and likelihoods, in particular without the mitigating effect of knowledge of results. A relatively less explored but important research direction involves extending the model to a biologically more realistic observer and actor model, examining the connections with network dynamics [12], [63] or population coding [31], bridging the gap between a normative description and mechanistic accounts of time perception. Another extension of the model would consider a non-stationary observer, whose response strategy changes from trial to trial (even after training), possibly in order to account for sequential effects of judgement which may be due to an iterative update of the prior [64]–[66]. Finally, whereas our analysis suggests that subjects found it relatively easy to learn unimodal distributions of intervals, bimodal distributions seemed to represent a much harder challenge. Further work is needed to understand human performance and limitations with multimodal temporal distributions.
The University of Edinburgh School of Informatics ethics committee approved the experimental procedures and all subjects gave informed consent.
Twenty-five subjects (17 male and 8 female; age range 19–34 years) including the first author participated in the study. Except for the first author all participants were naïve to the purpose of the study. All participants were right-handed, with normal or corrected-to-normal vision and reported no neurological disorder. Participants were compensated for their time and an additional monetary prize was awarded to the three best naïve performers (lowest mean squared error).
The first author took part in three of the experiments and was included as he represents a highly trained and motivated participant. Therefore it allowed an informal means to assess whether the author's data was different from those of the naïve participants which could reflect a lack of training or motivation. However, analysis of the author's datasets (response biases and moments of the inferred priors) were statistically indistinguishable from the other participants and therefore his data was included in the analysis.
Participants sat in a dimly lit room, 50 cm in front of a Dell M782p CRT monitor (160 Hz refresh rate, resolution). Participants rested their hand on a high-performance mouse which was fixed to a table and hidden from sight under a cover. The mouse button was sampled at 1 kHz (with a ms latency). Participants wore ear-enclosing headphones (Sennheiser EH2270) playing white noise at a moderate volume, thereby masking any experimental noise. Stimuli were generated by a custom-written program in MATLAB (Mathworks, U.S.A.) using the Psychophysics Toolbox extensions [67], [68]. All timings were calibrated and verified with an oscilloscope.
Each trial started with the appearance of a grey fixation cross at the center of the screen (27 pixels, diameter). Participants were required to then click on the mouse button at a time of their choice and this led to a visual flash being displayed on the screen after a delay of ms which could vary from trial to trial. The flash consisted of a circular yellow dot ( diameter and above the fixation cross) which appeared on the screen for 18.5 ms (3 frames). The ‘target’ interval ms was defined from the start of the button press to the first frame of the flash, and was drawn from a block-dependent distribution . Participants were then required to reproduce the target interval by pressing and holding the mouse button for the same duration. The duration of button press ( ms) was recorded on each trial. Participants were required to wait at least 250 ms after the flash before starting the interval reproduction, otherwise the trial was discarded and re-presented later. After the button release, 450–850 ms later (uniform distribution), feedback of the performance was displayed for 62 ms. This consisted of a rectangular box (height , width ) in the lower part of the screen with a central vertical line representing zero error and a dotted line representing the reproduction error on that trial. The horizontal position of the error line relative to the zero-error line was computed as either (Skewed feedback) or (Standard feedback), depending on the experimental condition, with pixels (). Therefore, for a response that was shorter than the target interval the error line was displayed to the left of the zero-error line, and the converse for a response longer than the target interval. The fixation cross disappeared 500–750 ms after the error feedback, followed by a blank screen for another 500–750 ms and the reappearance of the fixation cross signalled the start of a new trial.
Each session consisted of around 500 trials and was broken up into runs of 84–96 trials. Within each run the number of each interval type was set to reflect the underlying distribution exactly and the order of the presentations was then randomized. However, for the High-Peaked session we ensured that each less likely interval was always preceded by 3–5 ‘peak’ intervals. Subjects could take short breaks between runs.
Each experiment consisted of a number of blocks, each comprising of several sessions. Within each block, the sessions were identical with regard to interval and feedback type. The participants were divided into experimental groups as follows (see also Table 1):
Experiment 1: Short Uniform and Long Uniform blocks with Skewed feedback (4 participants, including the first author). Experiment 2: Medium Uniform and Medium Peaked blocks with Skewed feedback (6 participants, including the first author). Experiment 3: Medium Uniform block with Standard feedback (6 participants, including the first author). Experiment 4: Medium High-Peaked block with Standard feedback (3 participants). Experiment 5: Medium Bimodal with Standard feedback (4 participants) and Wide Bimodal with Standard feedback (4 participants).
The order of the blocks for Experiments 1 and 2 were randomized across subjects. Each block consisted of three to six sessions, terminating when the participant's performance had stabilized (fractional change in mean squared timing error between sessions less than 0.08). For Experiment 5 we required participants to perform a minimum of five sessions.
We examined the last two sessions of each block, when performance had plateaued so as to exclude any learning period of the experiment. We analysed all trials for the uniform distributions and Wide Bimodal block. For the non-uniform distributions, we picked a random subset of the frequently-sampled intervals such that all intervals contributed equally in the model comparison (results were mostly independent of the chosen random subset), with the exception of the Wide Bimodal block for which we would have had too few data points per interval. For each subject we therefore analysed about 1000 trials for the Uniform or Wide Bimodal blocks, 500 for the Peaked or Medium Bimodal block and 200 trials for the High-Peaked block. We discarded trials with timestamp errors (e.g. multiple or non-detected clicks) and trials whose response durations fell outside a block-dependent allowed window of 225–1237 ms (Short), 300–1462 ms (Medium), 375–1687 ms (Long) and 225–1687 ms (Wide), giving 124 discarded trials out of a total of 30000 trials (). Note that 93% of the discarded trials had response intervals less than 150 ms, which we attribute to accidental mouse presses.
|
10.1371/journal.pbio.1001250 | The Chromosomal Passenger Complex Activates Polo Kinase at Centromeres | The coordinated activities at centromeres of two key cell cycle kinases, Polo and Aurora B, are critical for ensuring that the two sister kinetochores of each chromosome are attached to microtubules from opposite spindle poles prior to chromosome segregation at anaphase. Initial attachments of chromosomes to the spindle involve random interactions between kinetochores and dynamic microtubules, and errors occur frequently during early stages of the process. The balance between microtubule binding and error correction (e.g., release of bound microtubules) requires the activities of Polo and Aurora B kinases, with Polo promoting stable attachments and Aurora B promoting detachment. Our study concerns the coordination of the activities of these two kinases in vivo. We show that INCENP, a key scaffolding subunit of the chromosomal passenger complex (CPC), which consists of Aurora B kinase, INCENP, Survivin, and Borealin/Dasra B, also interacts with Polo kinase in Drosophila cells. It was known that Aurora A/Bora activates Polo at centrosomes during late G2. However, the kinase that activates Polo on chromosomes for its critical functions at kinetochores was not known. We show here that Aurora B kinase phosphorylates Polo on its activation loop at the centromere in early mitosis. This phosphorylation requires both INCENP and Aurora B activity (but not Aurora A activity) and is critical for Polo function at kinetochores. Our results demonstrate clearly that Polo kinase is regulated differently at centrosomes and centromeres and suggest that INCENP acts as a platform for kinase crosstalk at the centromere. This crosstalk may enable Polo and Aurora B to achieve a balance wherein microtubule mis-attachments are corrected, but proper attachments are stabilized allowing proper chromosome segregation.
| When cells divide, their chromosomes segregate to the two daughter cells on the mitotic spindle, a dynamic macromolecular scaffold composed of microtubules. Each chromosome consists of two sister chromatids. Microtubules attach to the chromatids at structures called kinetochores, which assemble at the surface of the constricted centromere region where the sister chromatids are most closely paired. To segregate correctly, sister kinetochores must attach to microtubules emanating from opposite spindle poles. Kinetochore attachment to microtubules occurs randomly and mistakes occur frequently. For example, both sister kinetochores may attach to one pole, or one kinetochore may attach to both poles simultaneously. Two protein kinases, Aurora B and Polo, have essential roles in regulating this process: Aurora B triggers the release of incorrect attachments and Polo strengthens the grip that correctly attached kinetochores have on microtubules. In this work, we have investigated the potential functional links between these two crucial enzymes at centromeres in cells of the fruitfly. We found that early in division, Aurora B and Polo both interact with a structural partner protein named INCENP at centromeres. This allows Aurora B to phosphorylate Polo, thereby activating it. We show that coordinating the activities of these two central mitotic kinases is crucial for successful cell division, and that this mechanism is conserved in human cells.
| Executive decisions concerning when cells enter and exit mitosis are made by Cdk1 with cyclins A and B as cofactors. Once cells have entered mitosis, Plk1 and the Aurora kinases direct spindle formation, regulate chromosome attachments to spindle microtubules, ensure the operation of the spindle checkpoint, and enable daughter cells to complete cytokinesis (reviewed in [1]–[4]). Plk1 and Aurora A also function in the regulation of mitotic entry (reviewed in [5]).
In higher eukaryotes, Plk1 and Aurora B have potentially antagonistic activities during the early stages of chromosome attachment and alignment on the mitotic spindle. Plk1 phosphorylation of kinetochore components and microtubule plus-end-associated proteins is required for the establishment of stable kinetochore-microtubule (KT-MT) interactions. Electron micrographs of human cells treated with the Plk1 inhibitor BI2536 show fewer microtubule connections per kinetochore [6]. Tension-sensitive phosphorylation of BubR1 by Plk1 regulates the initial stability of KT-MT interactions [7], as do phosphorylation of CLIP-170 [8] and NudC [9]. Plk1 also phosphorylates components of the Ska and KNL-1/Mis12/Ndc80 (KMN) kinetochore complexes as well as centromere proteins CENP-B, CENP-C, CENP-E, and CENP-F. However, the function of these phosphorylations is not known [10].
The chromosomal passenger complex (CPC), consisting of Aurora B kinase, INCENP, Survivin, and Borealin [11], has a role in the correction of kinetochore-microtubule attachment errors by promoting the release of kinetochore-microtubule attachments [11]–[15]. The localization of the CPC relative to kinetochores is critical for regulation of kinetochore-microtubule attachments [16]. CPC targeting to inner centromeres occurs as a result of Survivin binding to histone H3 phosphorylated on Thr3 by Haspin kinase [17]–[19] and is helped by an Aurora B-dependent positive feedback loop [20].
The temporal and spatial regulation of Plk1 activation is complex. While Polo activity is not required for mitotic entry in unperturbed cell cycles, its activation by T-loop phosphorylation is needed for its functions during mitosis. At centrosomes, Aurora A kinase-Bora phosphorylates Plk1 on Thr210, thereby activating Plk1 at the G2-M transition in human cells [21],[22]. Subsequently, Plk1 triggers the degradation of both Bora [23],[24] and Aurora A [25]. The regulation of Plk1 activity at kinetochores is a critical and largely unstudied question.
Here, we have examined the mechanism of Polo kinase activation at Drosophila and human kinetochores. We show that Aurora B is required for Polo T-loop phosphorylation at the centromere, but not at centrosomes. Our studies identify a new regulatory link between the Aurora B and Polo kinases mediated by INCENP. Furthermore, we demonstrate that this mechanism of regulation of Polo kinase at the kinetochore by the CPC is conserved in human cells. These results support our previous hypothesis that INCENP acts as a platform coordinating the activities of these kinases on chromosomes during early mitosis [26].
As a starting point to examine the relationship between Polo and the CPC in Drosophila, we compared their localization in space and time. INCENP and the CPC are diffuse during early prophase in Drosophila cultured cells (Figure 1A2). In contrast, Polo kinase in these early stages is localized at centromeres (Figure 1, arrows; Figure S1A), at centrosomes, where it colocalizes with Aurora A (Figure 1A–D, asterisks; Figure S2), and also at the nuclear envelope (Figure 1A3, blue arrowhead), where it has been proposed to promote nuclear envelope breakdown (NEB) [8].
Later during prophase and early in prometaphase, INCENP concentrates at specific, brightly stained chromosomal regions that probably correspond to heterochromatin. At this time, Polo kinase is already concentrated at centromeres (Figure 1B3). This is the earliest stage at which we observe partial colocalization between Polo (Figure 1B3,4 white arrows) and the CPC.
In metaphase, when chromosomes are bioriented and under tension, INCENP is concentrated on inner centromeric threads that extend between bioriented sister kinetochores, running parallel to microtubules (Figure 1C2,4; Figure S1D). At this stage, most chromosome-associated Polo is detected in the outer kinetochore (Figure 1C3,4, arrows) and does not colocalize with INCENP. Later, in early anaphase, INCENP and Polo are observed on threads parallel to central spindle microtubules (Figure 1D and unpublished data), although prominent Polo labelling is still detected at the kinetochore.
We conclude that Polo concentrates at centromeres before INCENP does and that both proteins transiently colocalize there during early prometaphase.
To ask if this colocalisation of INCENP and Polo reflects a direct interaction between the two proteins, soluble, bacterially expressed, full-length Drosophila GST-INCENP (Figure 1E right panel) was mixed with in vitro-translated Polo kinase using Aurora B and luciferase as positive and negative controls, respectively. The mixture was then incubated with glutathione beads and bound proteins detected by SDS-PAGE. Robust binding was observed between GST-INCENP and Polo or Aurora B, but not with the luciferase control (Figure 1E). Interestingly, this interaction did not require CDK phosphorylation of INCENP, as previously described for the binding of mammalian INCENP to Plk1 [27].
We confirmed that a physical interaction between INCENP and Polo also occurs in vivo by immunoprecipitation from cell extracts. Cell lines stably expressing either Polo-GFP or Aurora B-GFP (positive control) were lysed and the tagged protein immunoprecipitated with anti-GFP. INCENP was readily detected in both immunoprecipitates by immunoblotting (Figure 1F).
To determine more precisely when and where INCENP and Polo interact during the cell cycle, we used a proximity ligation assay (PLA) to map sites where INCENP, Polo, and Aurora B are in close proximity. PLA is based on conventional double staining using primary antibodies raised in different species. The secondary antibodies used for detection are tagged with short DNA oligonucleotides. If those oligonucleotides are close enough to allow them to be bridged by hybridization with circle-forming oligonucleotides (distance between antigens of 10–30 nm, [28]), the circle can be amplified by rolling circle DNA synthesis and a positive PLA signal is obtained. That signal requires not only the close proximity of the antigens but also their favourable spatial conformation and absence of structural obstacles so that the oligos can interact and subsequent reactions take place. Thus, only a subset of actual interactions between proteins is detected with the PLA technique.
We validated the PLA assay by first confirming the known interaction between endogenous INCENP and Aurora B [29]–[32]. Indeed, we readily observed PLA signals associated with chromosomes in prophase and prometaphase cells (Figure 1G1). This confirmed a recent study that found positive PLA signals between various members of the CPC in all phases of mitosis in human cells [33].
In Drosophila, the positive PLA signals overlapped with Aurora B-GFP, confirming that epitopes on INCENP and Aurora B are in close proximity in the inner centromere (Figure 1G4). A parallel assay using antibodies against INCENP and GFP confirmed this close association of INCENP with exogenous GFP-tagged Aurora B (unpublished data). In two negative controls, we failed to observe PLA signals using antibodies to INCENP and γ-tubulin and between Polo and the centromere histone CENP-A/CID (Figure 1J,K).
We next used PLA to define where in cells interactions occur between INCENP and Polo. Cells co-stained for INCENP and Polo showed PLA signals on chromosomes in early mitosis (Figure 1H1). These signals colocalized with Aurora B-GFP, confirming that the interaction occurs at inner centromeres (Figure 1H4). PLA also detected a close association between INCENP and PoloT182ph, the activated form of Polo kinase, using an anti-phospho-epitope-specific antibody (Figure 1I1, see validation of the antibody below). These positive PLA signals were also present in inner centromeres (Figures 1I4).
We conclude that INCENP and Polo physically interact and are in close proximity at inner centromeres during early mitosis in Drosophila.
The association between INCENP and Polo described above suggested that INCENP and other CPC components might have a role in Polo activation by T-loop phosphorylation. To test this hypothesis, we asked whether reducing INCENP protein levels by RNAi affected the localization and activation of Polo kinase, as detected by monitoring Polo T-loop phosphorylation. Plk1 phosphorylation on the highly conserved T-loop residue Thr-210 (Thr-182 in Drosophila Polo, Figure 2A) is crucial for kinase activation [34]. T182 of Drosophila Polo is a major phosphorylation site detected by mass spectrometry, and a phosphomimetic mutation at that site (PoloT182D) increases Polo kinase activity in vitro (VA, unpublished observations).
To examine Polo activation, we used a phospho-epitope-specific antibody raised against human Plk1T210ph that also recognises Drosophila PoloT182ph (Figure 2B). In asynchronous cultures, PoloT182ph was barely detectable by Western blotting. However, we could readily observe endogenous PoloT182ph using this antibody after treatment with the phosphatase inhibitor okadaic acid (OA; Figure 2B). This signal disappeared following Polo depletion by RNAi, confirming that it comes from the Polo protein (see below). Moreover, Polo-Myc gives a signal at the expected higher molecular weight, while PoloT182A-Myc and PoloT182D-Myc are not recognized by this antibody (Figure 2B). In these experiments, Polo was detected as a doublet by Western blotting. This mobility shift is not caused by T182 phosphorylation and may be caused by an as-yet uncharacterized modification.
We used this antibody to examine the distribution of PoloT182ph by immunofluorescence in cycling DMel-2 cells that had not been treated with okadaic acid. The antibody detected PoloT182ph at centromeres/kinetochores, centrosomes, the cleavage furrow, and midbody. This staining was largely abolished following Polo RNAi-depletion (Figure S3).
Although most Polo accumulates at prometaphase kinetochores (Figure 2C3; Figure S2B,C; Figure S4A–C linescans), a minor fraction of the kinase localizes to inner centromeres (Figure 2C3 arrows; Figure 2C linescan). Indeed, the active kinase (detected with anti-PoloT182ph) is clearly detectable at inner centromeres (Figure 2C4+inset, arrows), where it colocalizes with INCENP, as predicted by the PLA results (Figure 2C4; linescan). We first detect this inner-centromeric pool of active Polo in late prophase cells (Figure S4A–C). PoloT182ph is no longer detected at the inner centromeres of chromosomes aligned at the metaphase plate. Instead it accumulates at kinetochores in metaphase cells (Figure 2D4, arrows; linescan).
Depletion of INCENP by RNAi substantially reduced levels of activated Polo T182ph at kinetochores (Figure 2E4,F4; linescans). In contrast, total Polo localized normally to kinetochores following INCENP knockdown (Figure 2E3,F3; linescans). This is consistent with the observation that Polo localization to this region precedes that of INCENP (Figure 1A). Importantly, we could still readily detect active PoloT182ph at centrosomes of cells following INCENP knockdown (Figure 2E4,F4, asterisks).
These experiments reveal that INCENP is required for T182 phosphorylation and activation of Polo kinase at inner centromeres in early mitosis. Activation of centrosomal Polo does not require INCENP.
In order to investigate the function(s) of Polo T182 phosphorylation in mitosis, we established stable cell lines allowing inducible expression of PoloWT-GFP or PoloT182A-GFP. Endogenous Polo could be depleted in those cells by RNAi against the 3′UTR of the native transcript (Figure 3A). Expression of PoloWT-GFP rescued the viability and proliferation of cells depleted of endogenous Polo. However, expression of PoloT182A-GFP did not, and cells died (unpublished data). Thus, Polo T-loop phosphorylation is essential for viability.
Polo-depleted cells accumulated in mitosis, exhibiting phenotypes similar to those observed for the first polo mutants [35]. Expression of PoloWT-GFP restored mitotic progression in cells depleted of endogenous Polo (Figure 3B), but expression of PoloT182A-GFP did not. Cells expressing only PoloT182A-GFP accumulated in prometaphase/metaphase (Figure 3C), often with unaligned chromosomes (Figure 3D–F). Interestingly, while the loss of Polo led to an increase in monopolar spindles, substitution of endogenous Polo with PoloT182A-GFP did not (Figure 3D–F). This suggests that T-loop phosphorylation of Polo may be dispensable for its role in bipolar spindle assembly.
The observation that INCENP-dependent activation of Polo by phosphorylation at T182 at centromeres/kinetochores is required for chromosome alignment in prometaphase is consistent with the known role of Polo in regulating kinetochore function.
Because the best known role of INCENP is to activate Aurora B kinase in the CPC, we next asked whether Aurora B has a role in Polo T-loop phosphorylation at centromeres. Drosophila Polo T182 (corresponding to human Plk1 T210) is preceded by a conserved stretch of basic residues resembling the consensus site for Aurora kinases (Figure 2A) [31],[36],[37]. Indeed, Drosophila Aurora B complexed with a fragment of INCENP can directly phosphorylate Polo in vitro (Figure 4A). A T182A mutation in the Polo used as a substrate reproducibly reduced its phosphorylation by about one half. Thus, Polo T182 is a major phosphorylation site for Aurora B (Figure 4A). Similar results were obtained using human Aurora B on GST-PoloWT or GST-PoloT182D (unpublished data).
Kinase inhibition studies suggest that Aurora B is responsible for PoloT182 phosphorylation in vivo. Binucleine 2 is the only specific Aurora B kinase inhibitor described to date that is effective in Drosophila cells [38],[39]. When DMel-2 cells were treated with 20 µM Binucleine 2 for 2 h, H3S10ph was undetectable in mitotic cells (unpublished data; [39]) and INCENP and Aurora B were dispersed in clumps on the chromosomes ([38],[39]; Figure S5B,C, compare with Figure S5A). Both of these phenotypes are characteristic of the loss of Aurora B function [40],[41].
Aurora B kinase activity is required for Polo activation at kinetochores, and levels of kinetochore-associated PoloT182ph were greatly reduced in Binucleine 2-treated mitotic cells (Figure 4B4–D4; Figure 4E). In contrast, we observed no obvious difference in the localization of bulk Polo kinase in those cells (Figure 4B3–D3; Figure 4E). Importantly, as in the case of INCENP RNAi, we could still detect activated Polo kinase at centrosomes in the same cells (Figure 4B4–D4 asterisks).
As independent confirmation of the inhibitor studies, RNAi-mediated depletion of Aurora B also led to disappearance of the PoloT182ph signal observed in Western blots after OA treatment of cells, while total Polo levels remained unchanged (Figure 4F). In striking contrast, the PoloT182ph signal actually increased after partial Aurora A depletion (Figure 4Fa), perhaps because cells accumulated in mitosis.
The above results suggested that Aurora B rather than Aurora A plays a major role to promote PoloT182 phosphorylation at centromeres in Drosophila cells. In order to exclude that our failure to detect PoloT182ph by Western blotting following Aurora B depletion was due to a cell cycle block outside mitosis caused by OA treatment, we examined the effects of RNAi depletion of Aurora A, Aurora B, and INCENP on the PoloT182ph signal at centromeres in individual mitotic cells without okadaic acid treatment. Brief (3 d) dsRNA treatments were used to avoid an accumulation of binucleate cells caused by failure in CPC function in cytokinesis.
Depletion of Aurora B or INCENP led to a significant reduction of the PoloT182ph signal at centromeres (Figure 4G). This effect was specific to centromeres, and PoloT182ph levels at centrosomes were unaffected following depletion of Aurora B or INCENP (Figure S6A). In contrast, Aurora A depletion had no effect on levels of PoloT182ph at centromeres, but led to a modest reduction in PoloT182ph levels at centrosomes.
Together, these results confirm that Aurora B and INCENP are required for Polo activation at the centromere/kinetochore in early mitosis and strongly implicate Aurora B as the kinase responsible.
The CPC is required for Polo kinase activation at centromeres in live animals, and not only in aneuploid cultured cells. To demonstrate this, we examined flies homozygous for the hypomorphic female-sterile allele incenpQA26, a point mutation in the highly conserved IN-box domain [42].
We observed a strong signal of PoloT182ph concentrated at kinetochores in wild-type mitotic neuroblasts (Figure 5A3). In third instar larval neuroblasts from the incenpQA26 mutant, 27% of mitoses (n = 290) showed obvious defects in INCENP localization, with the protein spreading onto chromosome arms (Figures 5B2, S5E). This was never observed in wild-type neuroblasts (n = 303; Figures 5A, S5D). The incenpQA26 mitotic phenotype (Figure S5E) resembles the Binucleine 2-induced phenotype, with INCENP dispersed in clumps on the chromosome arms in affected cells (Figure 4C,D; Figure S5B,C). Levels of PoloT182ph at kinetochores were substantially reduced in incenpQA26 mutant mitoses showing this characteristic incenp phenotype (Figure 5B3; Figure 5E). In contrast, overall levels of Polo at kinetochores remained similar to wild type (Figure 5G and I).
To test if PoloT182 phosphorylation requires Aurora B activity in vivo, we dissected whole brains, treated them with Binucleine 2, and processed them for immunostaining as above. After a 2-h incubation in 20 µM Binucleine 2, Histone H3S10ph (a reporter for Aurora B activity) was undetectable in mitotic cells (unpublished data). Drug-treated neuroblasts also showed the characteristic dispersion of INCENP in clumps on chromosome arms (Figure 5C2,D2 arrowheads; Figure 5H).
As predicted, PoloT182ph was virtually undetectable at kinetochores of Binucleine 2-treated neuroblasts, but remained readily observable at centrosomes (Figure 5C3,D3 asterisks). Total Polo levels at kinetochores remained similar to wild type in these cells. Thus, both Aurora B activity and INCENP are required for Polo T182 phosphorylation at kinetochores in mitotic Drosophila neuroblasts.
As independent confirmation that the CPC contributes to Polo regulation in vivo, we conducted a genetic experiment in a sensitized background. Tubulin-Gal4-driven overexpression of Polo mutated in a conserved destruction box (PoloΔdb—Figure S7) is semi-lethal at the pupal stage, suggesting that excessive Polo activity is detrimental to development. Interestingly, a decrease in the levels of functional INCENP rescued the semi-lethality of PoloΔdb-expressing flies. This was observed when one wild-type allele of incenp was replaced with either of the alleles incenp EC3747 or incenp QA26 (Figure S7). A heterozygous deletion removing Aurora B did not cause a significant rescue in the same assay, suggesting that INCENP may be the limiting CPC component for Polo regulation.
Together, these results confirm that the CPC contributes to the regulation of Polo function at kinetochores in vivo.
Importantly, the regulation of Polo T-loop phosphorylation described above for Drosophila is conserved in human cells. We used siRNAs to deplete either Aurora B (Figure 6A) or INCENP (Figure 6C) in HeLa cells and subsequently measured the levels of total Plk1 and Plk1T210Ph at kinetochores. Both INCENP and Aurora B depletion caused a dramatic reduction in Plk1T210Ph levels in early mitosis (Figure 6Ad3,B,Cd3,D). Levels of Plk1 were also slightly reduced, confirming that the CPC is at least partly required for the stable localization of Plk1 to mammalian kinetochores [27]. Comparable results were obtained when we treated cells with the Aurora B kinase inhibitor ZM447439 (Figure S8), confirming that Aurora B activity is indeed required for the presence of Plk1T210Ph at kinetochores in human cells.
Our results thus indicate that the CPC activates Polo kinase by T-loop phosphorylation at centromeres in both flies and humans.
Coordination of Polo and Aurora B activity at kinetochores is critical in early mitosis, as the two kinases play potentially antagonistic but complementary roles in regulating kinetochore-microtubule interactions. Aurora B is essential for the correction of aberrant attachments [13]–[15], and indeed, tethering Aurora B too close to kinetochores interferes with the formation of stable attachments [16]. In contrast, Plk1 activity is required for initial stabilisation of microtubule attachments to kinetochores [7]–[9]. We suggest that interactions with INCENP may provide a mechanism to coordinate the activities of these two essential kinases during early mitosis.
Recent studies suggest that Plk1 is activated at centrosomes when its T-loop (T210) is phosphorylated by Aurora A kinase–Bora, and that this promotes the G2/M transition upstream of Cdk1 [21],[22], although Polo activity is not required for mitotic entry ([3]; this paper–Figure 3). How Plk1 is activated at kinetochores remained an important unsolved question. Our present results show that Aurora B and INCENP, which are concentrated at inner centromeres [43],[44], function there to activate Polo by phosphorylating its T-loop.
Plk1 recruitment to centromeres in late G2 has been variously proposed to be mediated by Bub1 [45], INCENP [27], and BubR1 [7]. Another report implicated the self-primed interaction of Plk1 with PBIP1/CENP-U [46]. This could potentially explain why Plk1 activity is reportedly required for its localisation to kinetochores in human cells [47].
Our RNAi studies confirmed that Plk1 is partially dependent on the CPC for its centromeric localization in human cells. However, this appears not to be the case in Drosophila, where Polo is present at centromeres before NEB, at a time when INCENP is not yet concentrated at inner centromeres and before PoloT182ph, the active form of the kinase, is detected there. Indeed, we observed no significant decrease in kinetochore-associated Polo levels after INCENP RNAi in Drosophila cells.
Although Polo targeting to kinetochores is independent of the CPC in Drosophila, its activation there does require the CPC with active Aurora B. Our data suggest that INCENP binding to Polo facilitates its subsequent activation by Aurora B kinase (Figure 7B,C). Indeed, INCENP and Polo interact physically in vitro and co-immunoprecipitate in mitotic cell extracts. Although most centromeric Polo kinase is concentrated in the outer kinetochore in prophase and prometaphase, active Polo (PoloT182ph) is also found in inner centromeres, where it overlaps with INCENP as confirmed by a proximity ligation assay (PLA).
A range of evidence presented here suggests that Aurora B is the upstream kinase responsible for Polo kinase activation at centromeres. Firstly, Aurora B phosphorylates Polo at Thr182 in vitro. Secondly, RNAi depletion of INCENP or Aurora B, but not Aurora A, reduces levels of active PoloT182ph at kinetochores. Thirdly, tissue culture cells and third larval instar neuroblasts treated with a specific inhibitor of Drosophila Aurora B kinase show decreased levels of PoloT182ph at kinetochores. In all of the preceding experiments, PoloT182ph levels are affected at kinetochores but not at centrosomes, where Polo is presumably activated by Aurora A [21],[22]. Importantly, this involvement of Aurora B in Polo activation at centromeres discovered in Drosophila is conserved for Plk1 in human cells.
Our results suggest a model for interactions between Polo kinase and the CPC at centromeres (Figure 7). In Drosophila cells, Polo targets to centromeres before the CPC is recruited by Survivin binding to histone H3T3ph [17]–[19]. At inner centromeres of chromosomes whose kinetochores are not under tension, Polo now binds to INCENP. This promotes Polo kinase activation, as Aurora B phosphorylates PoloT182. We suggest that interactions with INCENP link the two complementary kinase activities, thereby potentially creating a microtubule attachment/detachment cycle at kinetochores. Such a cycle would not be possible without a balancing phosphatase activity, and PP2A-B56 has recently been shown to oppose both Aurora B and Plk1 activities at kinetochores to promote stable attachments [48].
At metaphase, when chromosomes are bioriented and under tension, the CPC and Polo kinase exhibit only a partial overlap. A weakening of the INCENP/Polo PLA signals in metaphase suggests that Polo may be released from INCENP after its activation—possibly moving to the outer kinetochore (Figure 7D). During metaphase, the CPC localizes in the inner centromere, stretching between sister kinetochores, whereas Polo and PoloT182ph concentrate mainly at the kinetochores. This separation may be necessary to allow Polo-mediated stabilisation of kinetochore-microtubule attachments. The coordinated activities of both kinases at kinetochores and their tension-mediated separation might facilitate a dynamic equilibrium between attached and unattached kinetochores, selectively stabilizing proper chromosome attachments.
In summary, our results reveal that INCENP and Aurora B are responsible for Polo kinase activation at centromeres but not at centrosomes during mitosis. These experiments support the hypothesis that INCENP acts as a scaffold integrating the cross-talk between these two important mitotic kinases [26].
Fly strains were grown at 25°C in standard Drosophila medium. The following stocks were used: Canton-S; incenpQA26/SM6a, incenpEC3747/SM6a, Tubulin Gal-4/TM3. UASp-POLOΔDB-MYC transgenic flies were generated by BestGene Inc. Immunostaining of testes and larval neuroblasts was performed as described previously [49].
Primary antibodies and dilutions for immunofluorescence analysis were as follows: mouse monoclonal B512 anti-αTubulin (SIGMA, 1∶2,000); Rabbit Polyclonals Rb-801, Rb-803 [41], 1∶500); mouse monoclonal anti-PhosphoT210 Plk1 (Abcam ab39068, 1∶100); mouse monoclonal anti-Polo Mab294 (kindly provided by A. Tavares and David Glover,1∶100); rabbit polyclonal anti-Aurora A (1∶100) and anti-Aurora B (1∶500) [50],[51]; and monoclonal anti-Myc 9E10 (Santa Cruz). Secondary antibodies were obtained from Jackson Immunoresearch.
The AC5-Polo-GFP cell line was described previously [52], and the AC5-Aurora B-GFP, AC5-Polo-Myc, AC5-Polo-T182A-Myc, and AC5-Polo-T182D-Myc stable cell lines were generated following the same protocol. Cell lines were grown in Express-Five medium (GIBCO) containing 20 µg/ml blasticidin.
Cells were treated with either DMSO or 20 µM Binucleine-2 for 2 to 4 h before being processed for immunostaining as described previously [41]. For experiments shown in Figures 3, 4F,G, and S6, 1.2×106 D-Mel2 cells were transfected in 6-well plates with 20 µg of dsRNA using Transfast reagent (Promega). Cells were analysed 3 or 4 d later by immunofluorescence and immunoblotting. The control dsRNA was generated against the sequence of the bacterial Kanamycin resistance gene. For experiments shown in Figures 3, 4G, and S6A, cells were seeded on coverslips and treated for 10 s in BRB-80+0.1% NP-40 before a 20 min fixation in BRB-80+4% formaldehyde. Cells were then permeabilized for 10 min in BRB-80+0.1% Triton X-100 and blocked for 1 h in PBS+0.1% Tween20+1% BSA. Primary antibodies were diluted in PBS+0.1% Tween20+1% BSA and incubated overnight at 4°C. Secondary antibodies were incubated 2 h at room temperature. Coverslips were mounted with Vectashield+DAPI. Images were taken using an AxioImager epifluorescence microscope.
Proximity Ligation Assay was performed using Duolink QL (Olink, Uppsala, Sweden) following the manufacturer's protocol. Duolink anti-rabbit plus probe, anti-mouse minus probe, and anti-rat minus probe were used. The following antibody pairs were used for the assay: Rabbit polyclonal anti-Incenp Rb801 [41], 1∶500/mouse monoclonal anti-Polo Mab294 (kind gift of A. Tavares,1∶100); Rabbit polyclonal anti-Incenp Rb801 ([41], 1∶500)/mouse monoclonal anti-PhosphoT210Plk1 (Abcam,1∶100); Rat monoclonal anti-Incenp (Kind gift of Kim McKim, 1∶300)/Rabbit polyclonal anti-Aurora B 963 ([41],1∶500); Rabbit polyclonal anti-Incenp Rb801 ([41], 1∶500)/mouse monoclonal anti-GFP (Roche, 1∶500); Rabbit polyclonal anti-Incenp Rb801 ([41], 1∶500)/mouse monoclonal anti-γTubulin (Sigma 1∶50); and Rabbit polyclonal anti-CID (a gift from S. Henikoff,1∶500)/mouse monoclonal anti-Polo Mab294 (1∶100). In each experiment a negative control using only one antibody of each pair was included.
For each antibody pair, exponentially growing DMel-2 cells were seeded on Con-A treated coverslips and fixed for immunostaining as described previously [41]. After overnight incubation with primary antibody at 4°C, half of the samples were processed following the normal immunostaining protocol [41] and the other half was used for the PLA assay.
Imaging was performed using Olympus IX-71 microscope controlled by Delta Vision SoftWorx (Applied Precision, Issequa, WA, USA). Image stacks were deconvolved, quick-projected, and saved as tiff images to be processed using Adobe Photoshop. Linescans were generated using Image-Pro software.
POLO-T182A, T182D, and POLOΔDB (R309A, L312A) were generated in the pDONR221 (Invitrogen) using QuickChange (Stratagene). The expression vectors pAC5-POLO-MYC, pAC5-POLO-T182A-MYC, pAC5-POLO-T182D-MYC, and pUASp-POLOΔDB-MYC were generated by Gateway recombination (Invitrogen) of pDONR-based entry clones into pDEST-AC5-Cterm-MYC and pDEST-UASp-Cterm-MYC, respectively. POLO-WT and POLO-T182A were cloned into pETDuet for expression as N-terminal fusions with a HIS tag at the MCS1 position. Aurora B was cloned into pDONR221, which was then recombined into pDEST-AC5-Cterm-GFP to generate pAC5-AURORA B-GFP.
GST tagged full-length Drosophila Incenp was expressed in bacteria (BL21) and purified on Glutathione sepharose beads as described previously [42]. Polo, Aurora B, and Luciferase were in vitro translated using a coupled transcription/translation reticulocyte lysate system (Promega's TNT system). Binding buffer—50 mM Tris pH 7.5, 10 mM MgCl2, 1 mM EGTA, 1 mM DTT, 0.1% Triton X-100, 0.5 mM PMSF, and 1 mg/ml CLAP.
Mouse anti-GFP (Roche) and mouse IgG (Abcam)—negative control—antibodies were crosslinked to protein G Dynabeads (Invitrogen) at 0.5 µg of antibody/1 µl of beads. Exponentially growing D-Mel2 cells were lysed on ice in lysis buffer (for Polo-GFP cell line: 40 mM Tris-Cl [pH 7.5], 100 mM NaCl, 1 mM PMSF, 1 mM DTT, 10 mM EGTA, 1% Triton-X-100, and protease inhibitor cocktail (Roche, UK); for GFP-Aurora B cell line: 50 mM Tris-Cl (pH 8.0), 150 mM NaCl, 1 mM EDTA, 1% NP40, 0.5% deoxycholate, and protease inhibitor cocktail -Roche). Cell lysates were separately incubated with either mAb anti-GFP or mouse IgG bound to Dynabeads protein G for 1 h at 4°C. Samples were spun down, then washed first with lysis buffer, and then twice with wash buffer (40 mM Tris-Cl [pH 7.5], 100 mM NaCl, 1 mM PMSF, 10 mM EGTA, 0.1% Triton-X-100, protease inhibitor cocktail, Roche, UK). Finally the beads were boiled in Laemmli sample buffer. All samples were subjected to SDS-PAGE and analyzed by immunoblotting as described before.
HIS-Polo and HIS-Polo-T182A were expressed in BL21 bacteria from pETDuet-based constructions (see above). Protein purification was done using Talon resin (Clontech) and purified proteins were stored on the resin at −80°C.
For the kinase assay, HIS-Polo and HIS-Polo-T182A on the resin were incubated with Drosophila Aurora B in complex with INCENP645–755 [39] for 15 min at 30°C in 20 mM K-HEPES pH 7.5, 2 mM MgCl2, 1 mM DTT, 500 mM ATP, 5 mCi 32P-g-ATP. Reactions were initiated by the combination of the substrate-bound resins to a fixed volume of a master mix containing all other reagents. Reactions were stopped by the addition of Laemmli sample buffer. Samples were separated by SDS-PAGE and transferred to nitrocellulose. Quantitative, sub-saturation measurements of radioactivity and Polo Western blot signals were obtained using a PhosphorImager and a Typhoon luminescence reader, respectively.
HeLa Kyoto were grown in Dulbecco's modified Eagle's medium, supplemented with 10% foetal calf serum, 0.2 mM l-Glutamine, 100 U/ml penicillin, and 100 µg/ml streptomycin.
RNAi experiments were performed using annealed siRNA oligos (Qiagen) diluted in serum free OptiMem and transfected using HiPerFect reagent (Qiagen) according to the manufacturer's protocol. HeLa cells were seeded on coverslips at a density of 1×105 cells/ml and diluted siRNA was added to cells so that the final concentration of siRNA was 40 nM. Coverslips were fixed at 48 h. For control transfections non-silencing random scramble siRNA oligos were used at the same concentration. The full sequences of siRNA oligos used are as follows: for Aurora B siRNA, 5′-AACGCGGCACTTCACAATTGA-3′; for INCENP siRNA 5′-AGATCAACCCAGATAACTA-3′. For drug treatments, ZM447439 (Tocris Bioscience) or DMSO as control were added to the cells at the concentration of 3 µM for 1 h.
All fixation, permeabilisation and immunostaining were performed at room temperature, as previously described [53]. Anti-Aurora B rabbit polyclonal at 1∶100 (Abcam, ab2245), anti-INCENP rabbit polyclonal at 1∶100 (Upstate), anti-Plk1 1∶100 mouse monoclonal (Abcam), anti-P-Plk1 (T210) 1∶100 mouse monoclonal (Abcam), and anti-phospho-Histone H3 (Ser10) rabbit polyclonal (Upstate). All affinity purified donkey secondary antibodies (labelled either with FITC, TRITC, or CY5) were purchased from Jackson Immunoresearch.
Quantification of Plk1 and Plk1T210Ph on approximately 1,000 centromeres per condition was carried out as follows: Deconvolved images were imported into OMERO [54] and segmentation of centromere foci (ACA, Cy5, reference channel) performed using Otsu segmentation implemented in Matlab. Masks stored in OMERO were then used to calculate intensities, and output into comma-separated value file for plotting in Excel.
|
10.1371/journal.pgen.1002803 | Patterns of Evolutionary Conservation of Essential Genes Correlate with Their Compensability | Essential genes code for fundamental cellular functions required for the viability of an organism. For this reason, essential genes are often highly conserved across organisms. However, this is not always the case: orthologues of genes that are essential in one organism are sometimes not essential in other organisms or are absent from their genomes. This suggests that, in the course of evolution, essential genes can be rendered nonessential. How can a gene become non-essential? Here we used genetic manipulation to deplete the products of 26 different essential genes in Escherichia coli. This depletion results in a lethal phenotype, which could often be rescued by the overexpression of a non-homologous, non-essential gene, most likely through replacement of the essential function. We also show that, in a smaller number of cases, the essential genes can be fully deleted from the genome, suggesting that complete functional replacement is possible. Finally, we show that essential genes whose function can be replaced in the laboratory are more likely to be non-essential or not present in other taxa. These results are consistent with the notion that patterns of evolutionary conservation of essential genes are influenced by their compensability—that is, by how easily they can be functionally replaced, for example through increased expression of other genes.
| In any given organism, a fraction of all genes in the genome are required for viability; if they are experimentally deleted, the organism dies. Interestingly, the set of essential genes is usually not identical even for closely related organisms. Genes that are essential in one organism are sometimes nonessential in sister taxa or even missing from their genomes. This suggests that, in the course of evolution, some genes can be rendered non-essential and consequently can be lost. How can genes become non-essential? It is possible that changes in an organism's living conditions render previously essential functions unessential. Alternatively, it is possible that, during evolution, the function of an essential gene can be taken over by another gene, so that the essential gene becomes dispensable. Here, we tested the second hypothesis experimentally in the laboratory. We tried to replace the functions of essential genes in the bacterium Escherichia coli. We find that the genes that can easily be replaced in the laboratory are also more likely to be lost in the course of evolution. This suggests that differences in the evolutionary fate between essential genes can be partially explained by how easily their functions can be taken over by other genes.
| Essential genes code for central cellular processes required for the viability of an organism. Many recent studies have used experimental data to determine gene essentiality in a large number of bacteria [1]–[11]. The central role that essential genes play suggests they should be highly conserved during evolution, and several comparative genomic analyses have confirmed this hypothesis [12]–[15]. A second implication of this pattern of conservation is that essential genes tend to remain essential during evolution: if a gene is essential in one organism, then orthologues of that gene are usually essential in other organisms (Figure 1).
However, there are many genes that do not follow these patterns: some genes that are essential in one organism are non-essential in other organisms; in other cases, genes that are essential in one organism are absent or have been lost from the genomes of other organisms [13], [15], [16]. Instances in which essential genes have become non-essential, or have been lost completely from genomes, suggest that either changes in physiological or environmental conditions have altered the essentiality of a gene, or that the genetic context has changed in a way to allow loss of a previously essential function. In this case, a second gene (either paralogous or unrelated to the original essential gene) may now perform the essential function. This raises the question of whether there is a connection between compensability and conservation level of essential genes.
The possibility of a connection between compensability and gene conservation has been raised on at least one occasion previously. Geissler et al. [17] observed that the Escherichia coli cell division protein ZipA is poorly conserved in other taxa. Under the assumption that other proteins must fulfill this role in these other taxa, they looked for suppressor mutations that would obviate the requirement for ZipA, and found that a single mutation in FtsA suppressed the lethal zipA phenotype.
Here we used a systematic approach to investigate how frequently the functions of essential genes of Escherichia coli can be replaced under laboratory conditions, and whether the frequency of this process correlates with patterns of evolutionary conservation.
To gain insight into this question, we used the following methodology. First, we compromised the function of an essential gene in Escherichia coli by decreasing its expression with a tightly regulated promoter. We then increased the expression level of a large number of other genes to identify genes that are capable of compensating for the function of the impaired essential gene. Repeating this process for a large number of essential genes, we isolated a set of genes that can functionally replace the essential genes when overexpressed. We find that although the majority of these compensating genes are not homologous to the impaired essential genes, they exhibit similar functions. In a few cases, the compensating genes are capable of fully replacing the functions of the essential gene, allowing the deletion of the essential gene from the genome. Finally, we show that those essential genes whose function can be compensated for in the laboratory are more likely to be non-essential or not present in other bacterial genomes, raising the possibility that similar compensatory mechanisms may allow essential gene loss to occur in natural populations.
Many previous studies have shown that gene essentiality is a mutable characteristic and is dependent on both the genetic background of the organism or the environmental conditions [18]–[21] (such genes are termed conditionally essential genes). The results we present here imply that in some cases it may be possible to predict which essential genes are more likely to be conditionally essential.
We constructed a collection of Escherichia coli strains in which essential genes were placed under the control of a conditionally expressed promoter. The essential genes were selected from a variety of functional classes [22], and exhibit a wide range of conservation levels [15]. In addition, some of these genes have been consistently found to be essential across all bacteria that have been examined empirically, while others are essential in only a few (Figure 1). A total of 26 genes were chosen (approximately 10% of the essential gene content of E. coli); six of these genes are in essential tandem operons (nrdAB, dnaTC and glmUS). To control the expression of the essential genes, we replaced their native promoters with the arabinose-inducible araBAD promoter (Para; see Methods). By shifting these mutant strains from medium with L-arabinose to medium without L-arabinose and supplemented with D-glucose, expression of the essential gene was repressed, and in all cases this resulted in growth inhibition or severe growth defects (Figure 2). We also tested if a plasmid encoding the corresponding essential gene rescued the lethal growth phenotype, and this was the case, except for the operons dnaTC and nrdAB (Table S1). This is most likely due to the fact that these are tandem operons, and transformation and maintenance of two separate plasmids that complement the function of each gene is unlikely, as the plasmids share the same replication origins and resistance markers.
Next, we assembled a library of overexpression plasmids using the ASKA(-) strain collection [23]. The ASKA(-) library consists of 4123 clones that each contain a plasmid with one E. coli open reading frame cloned behind an IPTG-inducible Plac promoter and a N-terminal (His)6-tag. Addition of IPTG induces strong expression of the downstream open reading frame. We removed the 26 clones containing plasmids with the essential genes mentioned above from the ASKA library. We then pooled the remaining 4097 clones and extracted plasmid DNA from this mixed pool.
Each conditional lethal mutant was transformed with an aliquot of the purified plasmid pool and plated on restrictive agar plates (where the essential genes were not expressed) with IPTG to induce expression of plasmid-encoded genes. We induced expression of the plasmid encoded genes with 50 µM IPTG, because higher induction levels are known to be deleterious for growth: 51% of E. coli proteins expressed under control of Plac at 1 mM IPTG cause lethality [23]. We also measured transformation efficiency: when transformed and plated under permissive conditions with selection for the plasmid-encoded chloramphenicol resistance, all strains except murA and fldA gave rise to at least 1.2×104 colonies, which is the minimal number of clones required for coverage of 95% of all variants transformed [24] (see Table S1).
We recovered up to 10 transformants from each plate, and restreaked them onto restrictive agar plates with IPTG to confirm growth. Upon successful growth of these clones, plasmids were extracted. To discriminate between possible chromosomal suppressor mutations and high copy suppressors (HCS) encoded on plasmids, each plasmid was retransformed into the ancestral conditional lethal mutant under permissive conditions, and colonies were tested for growth under restrictive conditions. In case of successful growth, the plasmid was sequenced. To control for suppression by multiple plasmids, we purified plasmids from the ASKA library and repeated the retransformation test, yielding the same results as before. For cases in which high copy suppression by the purified plasmid could not be confirmed, a second round of transformation and selection as described above was carried out.
For six strains (Para-aspS, plsC, plsB, ffh, glmUS, and gltX), although colonies were recovered, no HCS plasmids were isolated after two rounds of transformation and selection (Figure 1). In these cases it is possible that chromosomal mutations were responsible for rescuing the conditional lethal phenotypes, possibly by mutation of the Para promoter to mitigate repression. In three cases, no colonies were recovered at all (Para-gyrA, metK, murA). For murA, this might have been due to poor transformation efficiency. Finally, four conditional lethal mutants (Para-adk, dnaTC, proS, yeaZ) were recovered repeatedly with a plasmid coding for the gene ybiV. We presumed that ybiV interfered with the function of the arabinose promoter: all five essential genes are functionally different, and expression of ybiV promoted growth of these mutants under all restrictive conditions. Furthermore, ybiV has been found previously in screens using a Para construct [25]. We thus excluded the HCS ybiV from the subsequent analyses.
This left us with ten essential genes for which we had identified one or more HCS (Figure 1). For each of these ten genes, we subjectively chose one HCS for further analysis – with the exception of the essential gene degS, for which we included both HCS.
Next, we tested whether the recovered HCS plasmids could replace the functions of their respective essential genes, or whether viability might rely on low-level transcription from the repressed Para promoter. We attempted to knock out the corresponding essential genes in strains harboring the HCS plasmid, with expression of the suppressor induced using 50 µM IPTG. We were successful in four cases (these essential gene - HCS pairs were: dapA/nanA, spoT/mutT, pyrH/cmk and fldA/fldB; Figure 3). We were not able to delete these four essential genes from a strain carrying an empty control plasmid.
The other six essential genes could not be deleted from strains containing the HCS plasmids. This suggested that in the presence of the HCS, low-level expression of the essential genes was sufficient to allow growth. Without the HCS, this residual low-level expression did not allow growth (Figure 4). Alternatively, it is possible that the high copy suppressor increased expression from Para, and thus restored normal levels of the essential proteins. Therefore, we tested whether any of these seven HCS (two HCS were included for degS) increased expression from the Para promoter. We used a chromosomally encoded Para-phoA fusion to monitor expression from Para under conditions where expression of the HCS is induced. HCS clones overexpressing yciR, yhbJ, and ftnA exhibited slightly lower levels of PhoA-activity compared to controls, while overexpression of degP, rho and dpiA resulted in slightly elevated levels of PhoA-activity (approximately 1.5-fold increase over the control). However, this activity was more than 50-fold below the activity of Para when induced with 0.1% L-arabinose (Figure S1). This suggested that none of these HCS rescued the conditional lethal phenotype through increasing expression of the essential gene.
To test if the HCS genes caused a general non-specific rescue, we purified the HCS plasmids from the ASKA collection and transformed these plasmids into each conditional lethal mutant. We tested for regrowth in the same way as in the retransformation test described above. With one exception (see below), none of the HCS plasmids restored growth in any other conditional lethal strain except the strain it was recovered from. Therefore we assumed that the observed high copy suppression is due to a specific mechanistic link between the depletion phenotype and the high copy suppressor (Table 1), rather than a consequence of a high copy suppressor-mediated induction of expression from Para.
The single exception to this pattern was the HCS ftnA (coding for ferritin), which rescued ftsK- or nrdAB depletion. ftnA and nrdB exhibit structural homology (Table 2), suggesting that this is a specific functional replacement. In the case of ftsK, the mechanism of suppression is less clear. One possibility is that the FtnA protein alleviates oxidative stress [26] that results from the loss of FtsK, as a consequence of double strand breaks in chromosomal DNA [27]. However, in both cases, the mechanism does not appear to be moderated through FtnA restoring expression from Para (Figure S1).
These data thus show that out of the 23 essential genes or operons that we assessed, the functions of four could be completely replaced by non-orthologous genes. In six additional cases, the functions of the essential genes could be almost completely replaced: over-expression of a second gene enabled cellular viability even when the expression of the essential gene was largely abolished. In contrast, without overexpression of this second gene, no growth occurred.
We also quantified how well the high copy suppressors restored growth in the four strains in which we could knock out the essential gene, by measuring how the growth yield depended on the dosage of HCS expression. The complemented knockout mutants showed qualitatively different responses to increasing expression of the HCS, as measured by the amount of the inducer IPTG added to the growth medium (Figure 5). The dapA knockout exhibits very low levels of growth at all levels of inducer, suggesting that NanA has only a low level of activity toward the DapA substrate, or that very high levels of activity are required to sustain growth. On the other hand, both the spoT and pyrH knockouts exhibit growth even when the suppressor is uninduced, suggesting that either these proteins are much more promiscuous, or that only low levels of activity are required for growth (Figure 5).
To gain insight into the mechanisms of suppression, we compared the amino acid sequences (using Smith-Waterman alignments [28]) and protein structures (using pairwise structural alignments [29]; Table S2) of all complementing HCS – essential gene pairs. Of the four gene pairs for which we could knock out the essential gene, two HCS share homology in amino acid sequence and protein structure with the suppressed essential gene: dapA and nanA share amino acid homology (Table 2), with nanA being the closest homologue of dapA in E. coli. However, phylogenetic analysis shows that they are only distantly related, and most likely diverged before the most recent common ancestor of all bacteria (Figure S2B). The second HCS – essential gene pair showing amino acid homology is fldA and fldB (Table 2), with fldB being the closest homologue of fldA in the E. coli genome. Phylogenetic analysis suggests that these two proteins diverged after the origin of gamma-proteobacteria (Figure S2A).
In the set of six HCS for which we were unable to knock out the corresponding essential gene, one pair exhibits sequence homology: the proteins DegS and DegP [30] (Table 2). Again, degP (with degQ) is the closest homologue of degS in the E. coli genome.
In several cases, the functions of the essential gene and its complementing HCS appear to be related. Besides having amino acid and structural similarities, the three pairs mentioned above (dapA-nanA, fldA-fldB, and degS-degP) have known functional similarities.
DapA and NanA both belong to the N-acetylneuraminate lyase subfamily and catalyze similar biochemical reactions [31], although dapA is essential and nanA is non-essential. It has previously been shown that a single amino acid exchange can turn NanA, an N-acetylneuraminate lyase, into an efficient dihydrodipicolinate synthase, which is the dedicated function of DapA. This amino acid exchange was hypothesized to optimize the turnover rate rather than the specificity of the reaction [32]. Therefore, replacement of DapA by NanA may be an example of how the increased concentration of an enzyme with promiscuous activity can promote an essential biochemical reaction and restore viability in the absence of the gene originally encoding the essential function.
FldA and FldB are both flavodoxins. FldB is non-essential, in contrast to FldA [1], and cannot replace the function of FldA when expressed from its native promoter [33]. However, our results show that that FldB can replace the function of FldA when expressed at a high level.
DegS activates the sigma E stress response via proteolytic degradation of the anti-sigma factor RseA [34] triggered by misfolded outer membrane porins [35], and it seems that DegP can fulfill the same function [30]. Although this implies that DegP overexpression might fully compensate for DegS function, we were unable to delete degS when overexpressing degP.
Two of the essential gene – HCS pairs that exhibit no apparent amino acid or structural similarity do exhibit functional similarity: pyrH/cmk and spoT/mutT. cmk and pyrH both code for a nucleotide kinase: PyrH converts uracil monophosphate to uracil diphosphate, while Cmk converts cytosine monophosphate to cytosine diphosphate. It is also known that Cmk can use both cytidine and uridine (the primary substrate of PyrH) as substrates. Additionally, it has been shown previously that cmk can act as a high-copy suppressor of a temperature-sensitive pyrH allele [36]. Here we have shown that cmk is fully suppressive by deleting the entire pyrH locus.
Both mutT and spoT can recognize phosphorylated guanosines as substrates, and cleave phosphoryl groups. SpoT is a key enzyme of the stringent response and hydrolyzes penta/tetra guanosine phosphate ((p)ppGpp) [37], [38]. Deletion of spoT leads to the accumulation of (p)ppGpp, which in turn activates stringent response and leads to cessation of cell growth. One possible mechanism of suppression is that mutT can cleave phosphoryl groups from (p)ppGpp, converting it into a phosphorylated guanosine that no longer triggers the stringent response, thus allowing cell growth.
Other possible functional similarities between pairs of high copy suppressors and essential genes are listed in Table 1.
We also objectively evaluated the hypothesis that there are specific functional relationships between essential genes and their HCS by testing whether the functional annotations of essential genes and their HCS are more similar than expected by chance. We collected all the GO molecular function annotations [39] for each essential gene and its corresponding HCS, and calculated a functional distance between each pair (see Methods). Using this functional distance measure, we found that essential genes and their complementary HCS genes are much more similar in function than would be expected by chance (p = 0.0024, one-tailed Kolmogorov-Smirnov test). When we exclude the homologous pairs of genes (degS-degP, dapA-nanA, and fldA-fldB, a weak signal of functional similarity remains (p = 0.05, one-tailed Kolmogorov-Smirnov test). Thus, it appears that the HCS genes compensate for the deleted essential genes through specific complementation of the missing function. On the other hand, ybiV, which is likely to be a non-specific HCS, shows no pattern of having greater functional similarity to its paired essential genes than would be expected by chance.
Comparative genomic analyses have shown that genes that are essential in E. coli tend to be conserved in other bacterial taxa [12]–[15]. In addition, recent empirical results have shown that the essential functions of genes tend to be conserved: genes that are essential in one taxon have orthologues that are essential in other taxa (Figure 1). There are, however, exceptions: some essential genes are less well conserved, or have become non-essential in some taxa.
Here, we have shown that under laboratory conditions, the functions of many essential genes can be completely or partially replaced by homologous or unrelated non-essential genes. A simple explanation might connect these two observations: if it is difficult to replace the function of an essential gene, then this gene should be both highly conserved and consistently essential across bacterial taxa. We thus asked whether it is more difficult to find genes providing compensatory functions for genes that are both conserved and consistently essential.
We used data on conservation (see Methods) and empirical assessments of essentiality [1], [3]–[11] to test this hypothesis. Of the 23 essential genes or operon pairs that we investigated, eight are both conserved and essential for all bacteria in which essentiality has been empirically assessed (Figure 1). Within this set, we found an HCS for only a single gene, ygjD. 15 of the essential genes or operons that we considered are either not fully conserved across bacteria or are nonessential in some taxa, or both. Of these, we found HCS for 9 of the 15; if we exclude the operon pairs, as it may generally be more difficult to find compensatory functions for both genes, this fraction increases to 8 out of 12. The probability of finding so few HCS for conserved and consistently essential genes by chance is 0.037 and 0.025, respectively (one-tailed Fisher's exact test). The estimated odds ratios are (with 95% upper limits in parentheses): 0.11 (0.89) and 0.083 (0.75), respectively. These data suggest that suppressors of conserved consistently essential genes are approximately one fifth as likely to be found as suppressors for genes that are less conserved or are non-essential in some taxa.
Thus, genes that are ancient, strongly conserved, and consistently essential across taxa appear to be persistently essential under laboratory conditions.
The data here support the hypothesis that in some cases, simply increasing the expression level of specific non-essential genes can render essential genes non-essential (i.e. high copy suppression). Sequence and structural comparisons showed that some of the HCS genes were homologous to the essential genes whose function they replaced. However, these homologues tended to be distantly related, with divergence times ranging from before the root of all bacteria, to soon after the origin of gamma-proteobacteria. In addition, in the majority of cases, homology was not required for HCS to occur, highlighting the possibility that even when there is no detectable homology, elevated expression can act as a mechanism allowing functional replacement. For example, although the essential gene – HCS pairs spoT-mutT and pyrH-cmk do not share detectable similarity on protein structure or sequence level, their biochemical activity is apparently similar enough to allow complete functional complementation.
In several instances, the HCS that we recovered did not allow the deletion of the corresponding essential gene. We hypothesize that a combination of very low expression levels of the essential gene and expression of the relevant HCS allowed suppression of the lethal phenotype. However, in almost all cases, the suppression of the conditional lethal phenotype seemed to be based on a specific mechanistic link, a hypothesis that was further supported by the finding that the molecular functions of essential genes and their dedicated high-copy suppressors are far more similar than would be expected by chance.
Finally, we have shown that when the function of an essential gene can be replaced in the laboratory, orthologues of that gene are more likely to be non-essential or absent from the genomes of other bacterial taxa. This observation suggests that compensability may influence patterns of evolutionary conservation: the functions of some essential genes are easier to replace than others, and the genes that perform such functions may be lost more often over evolutionary time.
Previous studies have looked for high copy suppressors of lethal phenotypes [25], [40], [41]. The majority of these studies have been performed on a smaller scale or by screening for suppressors of mutations that cause non-lethal phenotypes, with the aim to investigate gene function. Our study is a comparative and systematic attempt to quantify the frequency of suppressors of essential genes, and to test if there is a statistical association between gene conservation and compensability.
The potential for finding redundant, yet non-orthologous genes that can functionally replace essential genes might be a function of genome size. Previous work has shown that bacterial species with large genomes have fewer essential genes than species with small genomes [4], [5]. One explanation for this observation is that in large genomes, there is a greater chance that a second gene encodes a similar function. Thus, the chance to replace essential gene functions with other functions could be greater in species with larger genome sizes and a generalist lifestyle. It would thus be interesting to test how the results of this study compare with additional studies in bacteria having much larger (or smaller) genomes. Indeed, in bacteria with small genomes, almost all essential genes are also highly conserved; thus finding conditionally essential genes may prove far more difficult.
Overall, our work provides a novel explanation for the different patterns of conservation that are observed for essential genes, and emphasizes that gene essentiality is a fluid characteristic, even over short periods of evolutionary time.
All strains were grown in LB media (Sigma) or LB agar plates (1.5% agar, Sigma), and L-arabinose or D-glucose (both Sigma) was supplemented as indicated. E. coli strains MG1655 and DY330 were described previously [42], [43] and grown at 37°C and 32°C, respectively, with vigorous shaking. AB330 is a Lac+ derivative of DY330, and was received from Alex Boehm, University of Wurzburg, Germany. P1 transduction and TSS transformation were done as described elsewhere [44], [45]. Strains harboring a pKD4 derived kanamycin resistance cassette were grown with 50 µg/ml kanamycin sulfate (Sigma), and strains with ASKA(-) plasmids with 15 µg/ml chloramphenicol (Calbiochem). Ampicillin (Fluka) 25 µg/ml was used to select for Para-phoA insertion in attB. Strains transformed with pCP20 [46] were grown at 32° in the presence of 15 µg/ml chloramphenicol (Calbiochem). IPTG (isopropyl thiogalactopyranoside) was from Sigma.
No comprehensive collection of conditional lethal mutants of essential genes is available. To construct a collection, we selected 23 essential genes and operon pairs from E. coli that exhibited varying levels of conservation across other bacterial taxa [15]. Genes were balanced for functional categories, but otherwise random. This group of essential genes covers nearly 10% of the essential gene content of E. coli MG1655.
Before we selected essential genes for our experiment, we discarded genes located in operons coding for other essential genes, because insertion of the Para construct in front or inside operons might have strong polar effects. Three exceptions were made: nrdAB, dnaTC and glmUS are essential tandem operons whose gene products interact physically or are involved in the same cellular processes. We assumed that the construct we use to repress transcription abolishes expression of both genes.
We used the previously described strain TB55 [47] as PCR template for construction of arabinose-inducible conditional lethal mutants (analogous to Roux et al. [48]), with the aim of tightly linking a kanamycin marker to the arabinose-inducible promoter of the araBAD operon. This strain allows the generation of a PCR product that contains an outward facing kanamycin resistance marker on one end, and on the other end an outward facing arabinose-inducible promoter. Insertion of this construct in front of essential genes and fusion of the Para promoter to transcriptional or translational start sites allows control of expression of selected essential genes [48]. We used TB55 to generate PCR products flanked by 40 to 42 base pairs homology to the upstream region of essential genes of interest. The PCR product spanned the kanamycin resistance gene, araC and the full intergenic region between araC and araB.
Next, we constructed TB741, a strain that allowed us to monitor expression of Para from a second, independent arabinose-inducible araBAD promoter. To that end, we combined a phoA knockout acquired from KEIO clone JW0374 [1], and, after removal of the kanamycin resistance marker with pCP20 [46], a Para-phoA construct was inserted into attB (derived from E. coli strain SA22 (a gift from Prof. Winfried Boos, University of Konstanz, Germany) with P1 phage transduction. All strains used in this study can be found in Table S3.
All conditional lethal mutant strains were constructed initially with the same primer design, which included the following: deletion of 40 to 100 base pairs of the upstream region of the gene of interest by insertion of the PCR product generated from TB55, and fusion of the start codon of the gene of interest with the start codon of araB.
We were not able to recover clones with a conditional lethal character for yeaZ and murA following this methodology. Therefore we fused the transcription initiation site of araB to the predicted transcriptional start sites of murA and yeaZ (from www.regulondb.ccg.unam.mx), yielding conditional lethal clones. All oligonucleotide sequences can be found in Tables S4 and S5.
As mentioned above, strain TB55 was used to generate PCR products that contained a kanamycin cassette adjacent to araC, the full Para-region and 42 to 45 base pairs at the 5′ and 3′ -prime ends that were homologous to the upstream and N-terminal region of the essential gene of interest. DY330 cells were grown in LB medium supplemented with 0.2% arabinose and made electro- and recombination competent as described previously [49]. After electroporation, cells were rescued in LB medium containing 0.2% arabinose and incubated at 32° for 1.5 hours prior to plating on arabinose- and kanamycin - containing LB plates. Clones were checked on LB plates supplemented with 0.4% glucose to confirm their conditional lethal character. The constructs were then moved by P1-transduction into TB741, and conditional lethality was assessed again on LB plates with 0.4% glucose. All promoter fusions as well as the adjacent araC gene were verified by sequencing.
We used the ASKA(-) strain collection [50] to construct a plasmid pool that contained all Escherichia coli open reading frames. The ASKA(-) library consists of 4123 clones, each one carrying a plasmid with one open reading frame. We pin-replicated clones into 96-well plates containing LB medium (Sigma) and 15 µg/ml chloramphenicol. Plates were incubated for 48 hours at 37°C. Then, 20 µl of each well were pooled, but clones containing plasmids that coded for essential genes of interest in our experiment were excluded. Plasmids were extracted using a plasmid preparation kit (Promega), following the recommendations of the manufacturer.
Each conditional lethal mutant was grown in LB medium with 0.1% arabinose (Sigma) to an OD600 nm of 0.4 to 0.8. During the preparation of electrocompetent cells, the density of all cultures was adjusted to an OD600 nm of 1 to guarantee an equal number of cells per transformation event. Each clone was electroporated with 1 µl of plasmid pool (DNA concentration approximately 330 ng/µl). Cells were rescued with 1 ml LB medium with 15 µg/ml chloramphenicol, and 100 µl was immediately removed and transferred to 900 µl LB medium with 0.1% arabinose and 15 µg/ml chloramphenicol to estimate transformation efficiency. To select for high copy suppressors, cells were spread on LB agar plates containing 0.4% glucose (to enhance repression of Para), 50 µM IPTG and 15 µg/ml chloramphenicol, and incubated at 37°C until colonies appeared, or maximally 3 days to minimize the formation of colonies that might arise due to chromosomal suppressor mutations.
We recovered up to 10 colonies per transformation event and restreaked them onto plates with glucose, IPTG, and chloramphenicol to verify growth. After successful regrowth, clones were grown in liquid cultured overnight with 0.1% arabinose and plasmid was extracted (Promega). The purified plasmid was retransformed [44] into fresh ancestral conditional lethal mutant strains under permissive conditions, and 4 independent colonies of each transformation event were tested on permissive and restrictive plates for growth. This procedure directly tested for suppression mediated by more than one plasmid: only upon successful regrowth of all 4 clones, were the plasmids sequenced using the primer 5′-GCGGATAACAATTTCACACAGA-3′. Cases in which all four clones did not grow were discarded from further analysis. The outcome of this retransformation test was based on the transformation method that exhibits a comparably low efficiency [44], decreasing the probability of transforming two different plasmids into the same cell, making it unlikely that more than one plasmid was responsible for high copy suppression.
After this verification procedure, we went back to the original ASKA(-) library, purified plasmids that we recovered from the screen (except yciR; this gene was not contained in the clone at the indicated position in the collection), and repeated the procedure. This led to exclusion of two high copy suppressors for aspS, and verified all other suppressive plasmids we found in the screen.
To determine if expression of a high copy suppressor lead to strongly increased expression of the Para promoter, we assayed the activity of a Para-phoA fusion inserted into the lambda attachment site, using a previously described procedure [51]. Briefly, we transformed plasmids coding for high copy suppressors (and as a control the empty plasmid pCA24N) into TB741, and grew clones overnight with 50 µM IPTG, 0.4% glucose and 15 µg/ml chloramphenicol in 96 well plates, replicating each clone independently 16 times. To estimate the maximum expression level of Para-phoA, we induced 16 replicates of TB741 harboring pCA24N with 0.1% arabinose. After overnight growth, cultures were spun down, resuspended in phoA buffer (150 mM TrisHCl adjusted to pH 9), diluted 1∶2 into fresh phoA buffer to a volume of 180 µl, and the OD600 nm was measured. One drop (approximately 10 µl) of a 1% SDS solution (Sigma) and 25 µl of a 10 mg/ml PNPP (4-nitrophenylphosphate, Sigma) solution was added. After incubation at room temperature for 24 hours the OD550 nm and OD420 nm were measured, and PhoA activity determined using the formula described in [51].
To delete spoT, pyrH, fldA and dapA, plasmids encoding mutT, cmk, fldB and nanA were transformed into AB330 and expression was induced with 50 µM of IPTG (or 1 mM for nanA). Knockouts were achieved following previously described methods [43], [49] using a pKD4-derived kanamycin cassette flanked by homologous ends. Successful deletions were moved into MG1655 (harboring ASKA(-) plasmids coding for HCS) with P1 transduction [45] with addition of IPTG and verified by PCR using primers upstream and downstream of the insertion.
To test for specific interactions between the depletion of essential genes and expression of plasmid-based non-complementing high copy suppressors, the HCS plasmids purified from the ASKA(-) library were transformed under permissive conditions into each conditional lethal mutant, and regrowth was checked as described for the initial screening procedure.
Gene deletion mutants were grown overnight at 37°C in 96-well plates with shaking at 400 rpm in 8-fold replication, in LB medium supplemented with 1 mM IPTG and 15 µg/ml chloramphenicol. Cultures were spun down, washed once in LB medium, and diluted 1∶10−4 into fresh medium with IPTG concentrations as indicated. Optical density at 600 nm was measured every 30 minutes, for 7.5 hours in total.
To analyze differential growth of conditional lethal mutant strains (with suppressive plasmids, empty control plasmids, deletion of essential genes or ancestral conditional lethal mutants), we grew the corresponding clones in 96-well plates overnight. As a control, each conditional lethal mutant and the ancestral TB741 strain were transformed with the empty plasmid pCA24N. Conditional lethal mutants were grown with 0.1% L-arabinose and, if required, in presence of 15 µg/ml chloramphenicol to select for ASKA(-) plasmids. Essential gene deletion mutants were cultured with 1 mM IPTG to induce expression of high copy suppressors and to decrease the likelihood of genetic suppressor mutations. After overnight growth, cultures were serially diluted by repeatedly transferring 20 µl of culture into 180 µl of LB medium. Of this dilution series, 5 µl of the indicated dilutions were spotted onto plates supplemented with arabinose, glucose, chloramphenicol and IPTG as indicated and incubated as indicated.
We used assignments based on previously published data [15]. Briefly, we used reciprocal shortest distance [52] to find potential orthologues of the relevant E. coli genes in the respective genomes. Two genes that are reciprocally the most closely related were denoted as orthologues if they aligned over more than 60% of the longer gene. In cases in which no orthologues were found, we used the MicrobesOnline database to search for genes named as putative orthologues. In this way, we found two additional putative orthologues, one for ftsK in S. pneumonia, and a second for plsC in S. aureus.
We used data from ten empirical studies on essentiality [1], [3]–[11] to determine whether or not genes orthologous to those in E. coli were essential in other bacterial taxa.
Orthologous and homologous genes from a range of bacterial taxa were selected and aligned using Muscle v3.8.31 [53] with default parameters. The alignments were cleaned using GBlocks 0.91b [54] with length of non-conserved positions set to 32, the number of flank and conserved positions set to minimum values, minimum block length to 2, and allowed gaps set to all. This alignment was used as input into MrBayes 3.1.2 [55] with a mixed amino acid model and invariant plus gamma distributed rate variation across sites. The chains were run for 200,000 (dapA/nanA), or 1,000,000 (fldA/fldB) generations, and the last 20% of the run was used for construction of a majority rule tree.
We obtained molecular function annotations from the GO database (www.geneontology.org/GO.downloads.annotations.shtml; 5/20/2011) for all annotated E. coli genes. We also obtained the relationships between all GO categories (www.geneontology.org/GO.downloads.ontology.shtml; OBO v1.2). GO annotations are related in a tree-like manner, beginning with broad, non-specific parent categories (e.g. “binding”), each of which have more specific child categories (e.g. “acyl binding”). Thus, we quantified functional distance as the number of parent categories that separate any two genes, normalized by the total number of parent categories for each gene. We calculated this distance between each essential gene and all other genes in the genome, and compared this to the distance between the essential genes and their complimentary HCS. This yields a number between 0 and 1, specifying the fraction of genes in the genome that are less functionally similar than the essential gene and its HCS. If functional similarity does not play a role for the essential gene - HCS pairs, we would expect this number to be 0.5, on average, and distributed uniformly between 0 and 1. Instead, we found that for 12 out of 13 essential gene HCS pairs, this distance was less than 0.5 (i.e. they were more similar than the average pair of genes); for 8 out of 13 pairs, the distance was less than 0.25. We used a Kolmogorov-Smirnov test to compare this distribution to the distribution expected if there were no functional relation between the essential gene and its HCS (the uniform distribution).
All statistical tests were done in R v2.13.1 [56].
|
10.1371/journal.pgen.1008294 | Ddc2ATRIP promotes Mec1ATR activation at RPA-ssDNA tracts | The DNA damage checkpoint response is controlled by the phosphatidylinositol 3-kinase-related kinases (PIKK), including ataxia telangiectasia-mutated (ATM) and ATM and Rad3-related (ATR). ATR forms a complex with its partner ATRIP. In budding yeast, ATR and ATRIP correspond to Mec1 and Ddc2, respectively. ATRIP/Ddc2 interacts with replication protein A-bound single-stranded DNA (RPA-ssDNA) and recruits ATR/Mec1 to sites of DNA damage. Mec1 is stimulated by the canonical activators including Ddc1, Dpb11 and Dna2. We have characterized the ddc2-S4 mutation and shown that Ddc2 not only recruits Mec1 to sites of DNA damage but also stimulates Mec1 kinase activity. However, the underlying mechanism of Ddc2-dependent Mec1 activation remains to be elucidated. Here we show that Ddc2 promotes Mec1 activation independently of Ddc1/Dpb11/Dna2 function in vivo and through ssDNA recognition in vitro. The ddc2-S4 mutation diminishes damage-induced phosphorylation of the checkpoint mediators, Rad9 and Mrc1. Rad9 controls checkpoint throughout the cell-cycle whereas Mrc1 is specifically required for the S-phase checkpoint. Notably, S-phase checkpoint signaling is more defective in ddc2-S4 mutants than in cells where the Mec1 activators (Ddc1/Dpb11 and Dna2) are dysfunctional. To understand a role of Ddc2 in Mec1 activation, we reconstituted an in vitro assay using purified Mec1-Ddc2 complex, RPA and ssDNA. Whereas ssDNA stimulates kinase activity of the Mec1-Ddc2 complex, RPA does not. However, RPA can promote ssDNA-dependent Mec1 activation. Neither ssDNA nor RPA-ssDNA efficiently stimulates the Mec1-Ddc2 complex containing Ddc2-S4 mutant. Together, our data support a model in which Ddc2 promotes Mec1 activation at RPA-ssDNA tracts.
| The ATR-ATRIP complex is recruited to sites of DNA damage by replication protein A-bound single-stranded DNA (RPA-ssDNA), and controls cellular responses to various types of DNA damage. The recruitment of ATR-ATRIP to RPA-ssDNA is not sufficient for the kinase activation. The activation of ATR-ATRIP requires activators including TopBP1 and ETAA1. In budding yeast, Mec1-Ddc2 (the ATR-ATRIP homolog) is activated through a similar mechanism. Activation of Mec1-Ddc2 requires Ddc1 (Rad9 homolog), Dpb11 (TopBP1 homolog) and Dna2. Our studies of the ddc2-S4 mutation have shown that Ddc2 promotes not only Mec1 recruitment but also Mec1 activation. In this study we show that Ddc2 promotes Mec1 activation independently of Ddc1/Dpb11/Dna2 function in vivo. We also show that Ddc2 regulates Mec1 activation through ssDNA recognition in vitro. Although RPA does not directly stimulate Mec1 activity, it can enhance ssDNA-dependent Mec1 activation. Our results suggest that ATR/Mec1, like ATM/Tel1 and DNA-PKcs, is activated upon DNA damage recognition.
| Chromosomes are constantly damaged by exogenous and endogenous threats [1]. The repair of damaged chromosomes is therefore critical for maintaining genome stability. The DNA damage response consists of multiple pathways controlled by the phosphatidylinositol 3-kinase-related kinases (PIKK) [2, 3]. These kinases include DNA-dependent protein kinase catalytic subunit (DNA-PKcs), ataxia telangiectasia-mutated (ATM), and ATM and Rad3-related (ATR). Although all these PIKKs respond to DNA damage, their DNA damage specificities are different. ATM and DNA-PKcs are activated by double-stranded DNA breaks (DSBs), whereas ATR responds to various types of DNA lesions with single-stranded DNA (ssDNA) [4, 5].
The Mre11-Rad50-Nbs1 complex recruits and activates ATM at DNA ends [6] whereas the Ku complex recruits and activates DNA-PKcs at DNA ends [3, 7]. The recruitment of ATM and DNA-PKcs is thus coupled to the kinase activation. Replication protein A (RPA) is the major protein that binds ssDNA with a high affinity [8]. ATR interacts with a partner ATRIP and recognizes RPA-covered ssDNA (RPA-ssDNA) [4, 5]. However, the recruitment of the ATR-ATRIP complex (ATR-ATRIP) to RPA-ssDNA is not sufficient for ATR activation. Indeed, ATR-ATRIP is stimulated by checkpoint regulators including TopBP1 and ETAA1 [4, 5]. TopBP1 is recruited to sites of DNA damage or stalled replication forks although the mechanism for the recruitment is not fully understood [9–12]. TopBP1 appears to engage with the Rad9-Rad1-Hus1 (9-1-1) complex at dsDNA-ssDNA junctions [4, 5]. Subsequently, TopBP1 directly stimulates the ATR-ATRIP kinase [4, 5, 13]. ETAA1 interacts with RPA and acts at stalled replication forks [14–16]. Like TopBP1, ETAA1 directly activates ATR-ATRIP [14, 15]. Thus, ATR-ATRIP is recruited by recognizing RPA-ssDNA and subsequently activated through multiple steps [4, 5].
In budding yeast, the Mec1-Ddc2 complex (Mec1-Ddc2) corresponds to ATR-ATRIP [17, 18]. Mec1-Ddc2 interacts with RPA-ssDNA to accumulate at sites of DNA damage [17]. The Ddc1-Mec3-Rad17 complex (the yeast 9-1-1 complex) recruits Dpb11 (TopBP1 ortholog) to the dsDNA-ssDNA junction [19, 20]. In budding yeast, both Ddc1Rad9 and Dpb11TopBP1 stimulate Mec1-Ddc2 kinase activity [21–25]. These observations have established the model in which the Ddc1-Dpb11 axis governs the checkpoint pathway by directly activating Mec1-Ddc2. In parallel with the Ddc1-Dpb11 axis, the Dna2 nuclease/helicase stimulates Mec1 kinase activity and controls DNA damage and replication checkpoints in S phase [26]. All Ddc1, Dpb11 and Dna2 proteins utilize the unstructured domains with aromatic amino acid residues (Trp or Tyr) to increase the catalytic activity of Mec1 [23, 24, 26, 27]. Thus, Ddc1, Dpb11 and Dna2 appear to activate Mec1 through a similar mechanism.
In budding yeast, Mec1 activates the downstream kinase Rad53 (Chk2 homolog) to enforce DNA damage checkpoint response [28, 29]. Signal transduction from Mec1 to Rad53 requires checkpoint mediators, such as Rad9 and Mrc1; Mec1 phosphorylates Rad9 and Mrc1 to promote their interaction with Rad53 at sites of DNA damage [30–33]. Rad9 controls checkpoint throughout the cell-cycle whereas Mrc1 is specifically required for the S phase DNA damage checkpoint [28, 32, 34–36]. Mrc1 associates with components of the replication fork in S phase [37, 38]. In contrast, recruitment of Rad9 to sites of DNA damage is a highly regulated process that involves three distinct mechanisms [39, 40]. First, the TUDOR domain of Rad9 interacts with K79-methylated histone H3 [41, 42]. Second, the tandem BRCT domain of Rad9 interacts with S129-phosphorylated histone H2A [43]. Finally, Rad9 binds to the Dpb11 scaffold protein [39, 44]. Histone H3 methylation is a constitutive mark on chromatin [45] and phosphorylated histone H2A spreads over around DNA lesions [46, 47]. However, the Dpb11 scaffold protein localizes to DNA lesions; indeed, Mec1 phosphorylates Ddc1 to promote Ddc1-Dpb11 interaction [39, 44, 48]. Thus, the Ddc1-Dpb11 axis not only stimulates Mec1 kinase activity but also promotes Rad9 recruitment to sites of DNA damage.
We have identified a separation-of-function ddc2 mutation (ddc2-S4) that causes defects in Mec1 activation but does not affect Mec1 recruitment [49]. However, it is not known how Ddc2-dependent Mec1 activation triggers checkpoint signaling. Moreover, the underlying mechanism of Ddc2-dependent Mec1 activation remains to be determined. To understand the significance of Ddc2-dependent Mec1 activation, we further characterized the ddc2-S4 mutation by carrying out genetic experiments. We found that the ddc2-S4 mutation impaired Rad9 and Mrc1 phosphorylation after DNA damage, consistent with the previous observation that the ddc2-S4 mutation is defective in checkpoint activation throughout the cell cycle [49]. The Ddc1-Dpb11 axis and Dna2 contribute to Mec1 activation in S phase [26]. We also found that S-phase checkpoint signaling is more significantly defective in ddc2-S4 mutants than in ddc1- and dna2-deficient mutants. Thus, Ddc2 appears to promote Mec1 activation independently of Ddc1/Dpb11 and Dna2 function. We further examined the effect of ddc2-S4 mutation on kinase activity of Mec1-Ddc2 using an in vitro reconstitution system. Whereas ssDNA stimulated Mec1 activity, RPA did not. However, RPA was found to promote ssDNA-dependent Mec1 activation. The Mec1-Ddc2 complex containing Ddc2-S4 mutant (Mec1-Ddc2-S4) exhibited a basal kinase activity in vitro. However, neither ssDNA nor RPA-ssDNA efficiently stimulated Mec1-Ddc2-S4. Our results support a model in which Ddc2 promotes Mec1 activation through ssDNA recognition.
Mec1 phosphorylates the Rad9 checkpoint mediator to promotes Rad9-Rad53 interaction, which is essential for Rad53 activation [30, 31]. The ddc2-S4 mutation confers defects in DNA damage checkpoint activation and damage-induced Rad53 phosphorylation at G2/M [49]. To understand the significance of Ddc2-dependent Mec1 activation, we first examined the effect of ddc2-S4 mutation on Rad9 phosphorylation after DNA damage at G2/M (Fig 1A). Cells expressing HA-tagged Rad9 protein (Rad9-HA) were arrested with nocodazole and exposed to methyl methanesulfonate (MMS). Cells were then subjected to immunoblotting analysis with anti-HA antibodies to monitor Rad9 phosphorylation (Fig 1A). Rad9 underwent phosphorylation in wild-type cells after MMS treatment. Phosphorylation was decreased in ddc2-S4 mutants but less significantly compared with in ddc2Δ mutants. Thus, the ddc2-S4 mutation impairs Rad9 phosphorylation after DNA damage.
One explanation for Rad9 phosphorylation defect in ddc2-S4 mutants is that Rad9 does not efficiently localize to sites of DNA damage. We next examined the effect of ddc2-S4 mutation on Rad9 accumulation at sites of DNA damage by chromatin immunoprecipitation (ChIP) assay (Fig 1B). In budding yeast, HO endonuclease introduces a sequence-specific DSB. We used an experimental system in which cells carry the GAL-HO plasmid and contain a single HO cleavage site at the ADH4 locus [49]. In this system, HO endonuclease, expressed after incubation with galactose, generates a single DSB at the ADH4 locus. Cells expressing HA-tagged Rad9 (Rad9-HA) protein were transformed with the GAL-HO plasmid. Transformants were grown initially in sucrose to repress HO expression, and then transferred to medium containing nocodazole to arrest at G2/M. After arrest, galactose was added to induce HO expression. Cells were then analyzed by the ChIP assay using anti-HA antibodies. Rad9 accumulated at sites of DNA damage less efficiently in ddc2-S4 mutants than in wild-type cells. However, the ddc2-S4 mutation conferred a milder defect in Rad9 accumulation compared with the ddc2Δ mutation (Fig 1B).
Rad9 limits the Sae2- and Sgs1-dependent pathway and interferes with DNA end resection [50]. We next addressed whether the ddc2-S4 mutation affects the kinetics of DNA end resection. To this end we monitored ssDNA generation at two EcoRI restriction sites near the HO cleavage site (at 0.8 kb or 5.8 kb from the site) by a quantitative PCR-based method [51] (Fig 2A). PCR amplifies only resected DNA because the EcoRI restriction enzyme can cleave unprocessed DNA (S1 Fig). The PCR amplification, normalized to the efficiency of HO cleavage, corresponds to the rate of DNA end resection [51]. The ddc2-S4 mutation did not significantly affect DNA end resection (Fig 2B). RPA, consisting of Rfa1, Rfa2 and Rfa3, binds to ssDNA tracts [8]. We also examined whether the ddc2-S4 mutation affects RPA accumulation near the DSB (Fig 2C). No apparent defect in Rfa2 association was observed in ddc2-S4 mutants. These result support the previous finding that the ddc2-S4 mutation does not affect Mec1 localization to sites of DNA damage [49]. Our results are also consistent with the current view that Mec1 positively controls DNA end resection although it promotes Rad9 accumulation at sites of DNA damage [52].
Mec1 phosphorylates Ddc1 to recruit Rad9 near sites of DNA damage through the Dpb11 scaffold [39, 44, 48]. We next examined the effect of ddc2-S4 mutation on Ddc1 phosphorylation after DNA damage (Fig 3A). Cells expressing HA-tagged Ddc1 protein were treated as above and subjected to immunoblotting analysis to monitor Ddc1 phosphorylation. Ddc1 phosphorylation was decreased in ddc2-S4 mutants compared with wild-type cells. Mec1 and Ddc1 are independently recruited to sites of DNA damage [53, 54]. We confirmed that the ddc2-S4 mutation has no impact on Ddc1 accumulation at sites of DNA damage (Fig 3B). Thus, the ddc2-S4 mutation impairs Ddc1 phosphorylation that promotes Ddc1-Dpb11-Rad9 assembly at sites of DNA damage [39, 44, 48].
Mec1 phosphorylates the Mrc1 checkpoint mediator that is essential for Rad53 activation during S phase [32, 35]. Notably, Mrc1-dependent Rad53 activation does not require Rad9 or Ddc1/Dpb11 function [55]. We further examined the effect of ddc2-S4 mutation on Mrc1 phosphorylation after DNA damage (Fig 4A). Wild-type, ddc2-S4 or ddc2Δ cells expressing HA-tagged Mrc1 protein were arrested in G1 with α-factor and released into medium containing MMS. Cells were harvested at the indicated times after release. We confirmed that cells remained within S phase at the time point after release (Fig 4A). Mrc1 phosphorylation was decreased in ddc2-S4 mutants but less significantly than in ddc2Δ mutants. Thus, Ddc2-dependent Mec1 activation also controls the Mrc1-dependent checkpoint pathway during S phase.
The Ddc1-Dpb11 axis and Dna2 control DNA damage checkpoints in S phase [26]. Two substitution mutations at the N-terminus of Dna2 (dna2-W128A, Y130A; hereafter called dna2-AA) abrogate its checkpoint function [26]. We next compared Rad53 phosphorylation in ddc2-S4, ddc2Δ and ddc1Δ dna2-AA mutants in S phase (Fig 4B). Cells expressing HA-tagged Rad53 protein were arrested at G1 and released into MMS or hydroxyurea (HU) as above. Rad53 phosphorylation was decreased in ddc1Δ dna2-AA mutants. However, a more significant defect was observed in ddc2-S4 or ddc2Δ mutants (Fig 4B). Our results agree with the previous observation that ddc1Δ dna2-AA cells are less defective in Rad53 activation than mec1Δ cells [56]. We next compared DNA damage sensitivity of ddc2-S4 and ddc1Δ dna2-AA mutants. While the ddc2-S4 and the ddc1Δ dna2-AA mutation caused similar sensitivities to MMS, ddc2-S4 mutants were more sensitive to HU than ddc1Δ dna2-AA mutants (Fig 4C). We further addressed whether the residual checkpoint activation in ddc1Δ dna2-AA mutants depends on Mec1 function in S phase (Fig 4D). The introduction of a mec1Δ mutation decreased damage-induced Rad53 phosphorylation in ddc1Δ dna2-AA mutants. Together, these results suggest that Ddc2 promotes Mec1 activation through a Ddc1/Dpb11/Dna2-independent mechanism.
Mec1 phosphorylates two subunits of RPA, Rfa1 and Rfa2, in response to DNA damage [57, 58] although the significance of RPA phosphorylation in checkpoint signaling is not fully understood [59]. We examined the effect of ddc2-S4 mutation on Rfa2 phosphorylation after DNA damage (Fig 5A). Wild-type and ddc2-S4 mutants were arrested with nocodazole at G2/M and exposed to MMS. Cells were then analyzed by immunoblotting with anti-Rfa2 antibodies. We found that damage-induced Rfa2 phosphorylation was decreased in ddc2-S4 mutants. The ddc2-S4 mutation does not affect Mec1 or RPA accumulation at sites of DNA damage [49] (Fig 2C). Thus, the ddc2-S4 mutation impairs Rfa2 phosphorylation in vivo.
Mec1/ATR phosphorylates RPA efficiently in the presence of ssDNA in vitro [21, 58, 60]. However, whether RPA or ssDNA modulates ATR/Mec1 activity remains to be determined. To understand a role of Ddc2 in Mec1 activation, we reconstituted an in vitro system using purified Mec1-Ddc2 and RPA proteins. We have purified Mec1-Ddc2 and Mec1-Ddc2-S4 through a two-step affinity chromatography after overexpressing FLAG-tagged Mec1 and His-tagged Ddc2 protein in yeast cells (S2 and S3 Figs). In agreement with the previous studies [21, 58], Mec1-Ddc2 phosphorylated RPA efficiently in the presence of ssDNA (S4 Fig).
The ddc2-S4 mutation contains two substation mutations (K263E, H382Y). K263 is implicated in Mec1-Ddc2 homodimerization [61] whereas H382 is in close proximity to the 177KKRK180 motif which is involved in DNA binding [61, 62] (Fig 5B, S5 Fig). Ddc2/ATRIP itself interacts weakly with ssDNA but RPA stimulates ssDNA binding of Ddc2/ATRIP [17, 62]. We determined the effect of ddc2-S4 mutation on the interaction of Mec1-Ddc2 with ssDNA or RPA-ssDNA by a pull-down assay. Mec1-Ddc2 and Mec1-Ddc2-S4 were found to bind similarly to ssDNA or RPA-ssDNA in vitro (Fig 5C), consistent with the observation that the ddc2-S4 mutation does not affect Mec1 accumulation at sites of DNA damage [49]. We further examined the effect of ddc2-S4 mutation on Ddc2-ssDNA binding. We prepared MBP-fused Ddc2 or Ddc2-S4 from E. coli and further examined whether they interact with oligonucleotides in the presence or absence of RPA (Fig 5D). MBP-Ddc2 and MBP-Ddc2-S4 interacted similarly with ssDNA and RPA-ssDNA. MBP alone did not exhibit oligonucleotide binding (Fig 5D). Thus, the ddc2-S4 mutation does not affect its own ssDNA- or RPA-ssDNA-binding abilities.
We next examined whether Mec1-Ddc2-S4, like Mec1-Ddc2, phosphorylates RPA efficiently in the presence of ssDNA (Fig 5E). Mec1-Ddc2 and Mec1-Ddc2-S4 similarly phosphorylated Rfa2 in the absence of ssDNA. However, in the presence of ssDNA, Mec1-Ddc2-S4 phosphorylated Rfa2 less efficiently compared with Mec1-Ddc2 (Fig 5E). A similar defect in Rfa1 phosphorylation was observed with Mec1-Ddc2-S4 (Fig 5E). Together, these results raise a possibility that Ddc2 upregulates Mec1 kinase activity by interacting with ssDNA or RPA-ssDNA.
We addressed whether RPA or ssDNA regulates Mec1 activity using GST-Rad53 as a substrate (Fig 6). GST-Rad53 lacks the N-terminal kinase domain of Rad53; therefore, no phospho-incorporation into GST-Rad53 was observed without Mec1-Ddc2 [49] (Fig 6A and 6B). We first tested the effect of various concentrations of RPA on Mec1 kinase activity. RPA had no significant impact on Mec1 activity using GST-Rad53 as a substrate (Fig 6A).
We next investigated the effect of ssDNA on Mec1 activity using various lengths (20, 40, 60 or 80 mer) of oligo(dT) (Fig 6B and 6C). Although no apparent stimulation was observed with a 20 mer oligo(dT) (oligo(dT)20), longer oligonucleotides, oligo(dT)40, oligo(dT)60 and oligo(dT)80, were found to increase Mec1 activity more efficiently. Similar activation was observed with oligo(dT)80 and an 80-mer biotinylated oligonucleotide containing all DNA bases (bio-oligo(dN)80) (Fig 6C). We note that biotinylation of oligonucleotide has no impact on Mec1 activation (S6 Fig). High concentrations of 80-mer oligonucleotides (125 nM) were required to reach maximum activation compared with the concentration of Mec1 (5 nM). We further tested whether longer ssDNA stimulates Mec1 more strongly using 5 kb ΦX174 phage ssDNA (Fig 6D). A single-stranded form of ΦX174 phage stimulated Mec1 at much lower concentrations compared with 80-mer oligonucleotides. The maximum activation obtained with ΦX (3 fold) was slightly higher than that of 80-mer oligonucleotides (2.5 fold) (Fig 6C and 6D). These results indicate that ssDNA stimulates Mec1 in a dosage-dependent and length-dependent manner.
We then determined the effect of ddc2-S4 mutation on ssDNA-dependent Mec1 activation using 80-mer oligonucleotides (Fig 6E). Mec1-Ddc2-S4 and Mec1-Ddc2 exhibited similar basal kinase activities. However, Mec1-Ddc2-S4 was not efficiently stimulated by ssDNA. Thus, ssDNA appears to stimulate Mec1 activity through a Ddc2-dependent mechanism.
We next determined the combination effect of RPA and ssDNA on Mec1 kinase activity using GST-Rad53 as a substrate. RPA prompts ssDNA binding of Mec1-Ddc2 or Ddc2 (Fig 5C and 5D) whereas ssDNA stimulates Mec1 kinase activity (Fig 6). We thus expected that RPA promotes ssDNA-dependent Mec1 activation. However, Mec1-Ddc2 was found to interact with RPA independently of ssDNA in vitro, in agreement with the current view that the N-terminus of Rfa1 interacts directly with the N-terminus of Ddc2 [63, 64] (Fig 7A). Therefore, RPA by itself could compete with ssDNA-bound RPA for Mec1-Ddc2 binding (S7 Fig). Moreover, RPA is a good substrate of Mec1 (S4 Fig); that is, RPA could compete with GST-Rad53 as a Mec1 substrate [65]. Hence, high RPA concentrations could have negative impacts on ssDNA-dependent Mec1 activation in vitro.
We first examined the effect of RPA on Mec1 activation with a low concentration of 80-mer oligonucleotides (12.5 nM) (Fig 7B). We note that only weak Mec1 activation was observed at this concentration (Fig 6B and 6C). We incubated oligonucleotides with various concentrations of RPA and subsequently with Mec1-Ddc2 to initiate the kinase reaction. Lower concentrations of RPA enhanced Mec1 kinase activity whereas higher concentrations of RPA attenuated (Fig 7B and 7D). We further tested the effect of RPA on Mec1 activation with a higher concentration of oligonucleotide (125 nM) (Fig 7C). Again, lower concentrations of RPA stimulated Mec1 activity whereas higher concentrations of RPA attenuated. These results are consistent with the hypothesis that RPA promotes ssDNA-dependent Mec1 activation although high RPA concentrations have negative impacts on ssDNA-dependent Mec1 activation in vitro. The stimulatory effect of RPA was less pronounced when Mec1-Ddc2 was incubated with a higher concentration of oligonucleotides (Fig 7C and 7D), consistent with the observation that ssDNA, but not RPA, stimulates Mec1-Ddc2 activity (Fig 6).
As discussed above, the ddc2-S4 mutation causes a defect in ssDNA-dependent Mec1 activation (Fig 6E) although it does not affect RPA-ssDNA binding of Mec1-Ddc2 (Fig 5C and 5D). We then determined the effect of ddc2-S4 mutation on Mec1 activation in the presence of oligonucleotides (12.5 nM) and RPA (0, 5, 10 nM) (Fig 7E). Mec1-Ddc2-S4, unlike Mec1-Ddc2, was not efficiently stimulated by RPA-ssDNA (Fig 7E). Together, our results support a model in which Ddc2 mediates Mec1 activation through ssDNA recognition while RPA prompts ssDNA binding of Mec1-Ddc2 at sites of DNA damage.
Previous studies have established the model in which ATRIP/Ddc2 interacts with RPA-coated ssDNA and recruits ATR/Mec1 to sites of DNA damage [4, 5]. However, Ddc2 appears to stimulate Mec1 kinase activity at sites of DNA damage [49]. In this study we have further characterized the ddc2-S4 mutation by carrying out genetic and biochemical experiments. The ddc2-S4 mutation causes defects in phosphorylation and accumulation of the Rad9 checkpoint mediator at sites of DNA damage. The ddc2-S4 mutation also confers a defect in phosphorylation of the S-phase specific Mrc1 checkpoint mediator. The Ddc1-Dpb11 axis and Dna2 contribute to Mec1 activation in S phase [26]. Notably, the ddc2-S4 mutation causes a more significant defect in S phase checkpoint signaling than the ddc1Δ dna2-AA mutation. Thus, Ddc2 controls Mec1 activation through a Ddc1/Dpb11/Dna2-independent mechanism. We further examined the effect of ddc2-S4 mutation on Mec1 kinase activity using an in vitro reconstitution system. ssDNA, but not RPA, stimulates Mec1-Ddc2 kinase activity. However, RPA can promote ssDNA-dependent Mec1 activation. Neither ssDNA nor RPA-ssDNA stimulates Mec1-Ddc2-S4 effectively. Our data support a model in which Ddc2 increases Mec1 kinase activity upon ssDNA recognition.
The ddc2-S4 mutation confers a defect in Rad9 phosphorylation and accumulation at sites of DNA damage. Mec1 phosphorylates Rad9 to allow Rad9-Rad53 interaction and subsequent Rad53 activation [30, 31, 34]. Rad9 accumulates at sites of DNA damage by interacting with K79-methylated histone H3, S129 phosphorylated histone H2A and the scaffold protein Dpb11 [39, 40]. Mec1 phosphorylates Ddc1 to promote Ddc1-Dpb11-Rad9 interaction at sites of DNA damage [39, 44, 48]. In this study we show that the ddc2-S4 mutation confers a defect in Ddc1 phosphorylation after DNA damage. We have previously shown that histone H2A phosphorylation is decreased in ddc2-S4 mutants [49]. Thus, two different Rad9 recruitment mechanisms are defective in ddc2-S4 mutants. Rad9 recruitment defect may compromise Rad9 phosphorylation in ddc2-S4 mutants because Mec1 accumulates and phosphorylates Rad9 at sites of DNA damage [30, 31, 34]. Mec1 phosphorylates Dpb11 and enhances the stimulating effect of Dbp11 on Mec1 kinase activity in vitro [25, 66]. Similar to Dpb11, Ddc1 directly activates Mec1 kinase in vitro [21, 23]. It is not known whether phosphorylated Ddc1 stimulates Mec1 kinase activity more effectively than non-phosphorylated one.
The ddc2-S4 mutation causes a more significant defect in S phase checkpoint activation than the ddc1Δ dna2-AA mutation. Notably, the residual checkpoint activation in ddc1Δ dna2-AA cells largely depends on Mec1 function. These results suggest that Ddc2 promotes Mec1 activation through a Ddc1/Dbp11/Dna2-independent mechanism. Mec1 phosphorylates the Mrc1 checkpoint mediator and activates Rad53 in S phase [32, 35]. Notably, Mrc1-dependent Rad53 activation does not require Rad9 or Ddc1/Dpb11 function [55]. Recent evidence suggests that the dna2-AA mutation affects DNA replication or repair function rather than checkpoint activation [27]. Thus, Ddc2-dependent Mec1 activation appears to play a key role in the stimulation of the Mrc1 checkpoint pathway during S phase.
The ddc2-S4 mutation causes a defect in ssDNA-dependent Mec1 activation. Then how does ssDNA stimulate Mec1 kinase activity? Mec1-Ddc2 forms a dimer of heterodimers through multiple interfaces including the PIKK regulatory domain (PRD) [67]. Interestingly, the PRD is closely positioned near the catalytic and activation loop at the kinase domain, thereby blocking kinase activity and substrate entry [61, 67]. We propose that ssDNA binding of Ddc2 triggers conformation changes of the Mec1-Ddc2 homodimer, which could open up the catalytic active site (Fig 8A). The ddc2-S4 mutation carries two substation mutations (K263E, H382Y). K263 contributes to Mec1-Ddc2 homodimerization [61] (S5 Fig) whereas H382 is positioned near the putative DNA binding (177KKRK180) region [61] (Fig 5B). Thus, the ddc2-S4 mutation may affect conformation changes of Mec1-Ddc2 homodimer upon ssDNA binding. Mec1 phosphorylates Mec1 activators (Ddc1, Dpb11 or Dna2) after DNA damage [66, 68, 69]. Conformation changes of the kinase domain could therefore enhance binding of Mec1 activators to its own kinase domain. Consistent with the view, Dpb11 has been shown to activate Mec1 more strongly in the presence of RPA and ssDNA [22].
RPA at lower concentrations promotes ssDNA-dependent Mec1 activation in vitro. Ddc2 and RPA recognize ssDNA through different mechanisms; the KKRK motif of Ddc2 is implicated in DNA binding [62] whereas RPA utilizes its own DNA binding domain (DBD) [8] (Fig 8B). However, the N-terminus of Ddc2 interacts with RPA independently of ssDNA although other domains of Ddc2 may be involved in RPA interaction [49, 63, 64] (Fig 8B). Thus, RPA-Ddc2 interaction could stimulate Ddc2-ssDNA binding by providing an additional ssDNA-binding interface, thereby boosting ssDNA-dependent Mec1 activation. However, we cannot fully exclude the possibility that RPA, once coated on ssDNA, acquires the ability to directly stimulate Mec1 activity. RPA at higher concentrations attenuates ssDNA-dependent Mec1 activation in vitro. One explanation is that RPA by itself competes with RPA-ssDNA for Mec1-Ddc2 binding. Alternatively, there would be substrate competition between RPA and in vitro substrates. At the moment, it remains to be determined which property of RPA down-regulates ssDNA-dependent Mec1 activation in vitro. Previous in vitro studies have shown that RPA-ssDNA has no apparent impact on ATR-ATRIP/Mec1-Ddc2 kinase activity; however, the effect of different RPA concentrations has not been extensively investigated [21, 22, 63, 70]. RPA not only stabilizes ssDNA but also stimulates various repair processes [8, 71–75]. Moreover, RPA binding to ssDNA is highly dynamic using different binding modes [76, 77]. Interestingly, RPA depletion modulates ssDNA generation and Mec1 activation differently [74]. Thus, dynamic interactions between ssDNA, RPA and Mec1-Ddc2 might be important for efficient Mec1 activation. A high-resolution structure of Mec1-Ddc2 has been recently reported [61]. However, the structure and dynamics of how Mec1-Ddc2 and RPA assemble on ssDNA remain to be elucidated.
In summary, we have shown that Ddc2 promotes Mec1 activation independently of Ddc1, Dpb11 and Dna2. We have also provided evidence supporting that Ddc2 promotes Mec1 activation through ssDNA recognition. ATR/Mec1 recognizes RPA-ssDNA and controls many cellular activities during DNA replication and repair [4, 5]. Our studies thus provide insight into how RPA-containing DNA structures modulate ATR/Mec1 activation, and suggest that ATR/Mec1, like DNA-PK and ATM/Tel1, is activated upon the recruitment to sites of DNA damage.
pRS424-GAL-FLAG-MEC1 is a high-copy plasmid version of YCp/pRS316-GAL-FLAG-MEC1 [78]. The GAL1-GAL10 promoter region was amplified by PCR with the primer pair 3016 and 3017, fusing a sequence encoding MEHHHHHH to the GAL1 promoter. The PCR product was cleaved with EcoRI and MluI. The DDC2 or ddc2-S4 coding sequence [49] was amplified by PCR with the primer pair KS460 and KSX001, fusing a HA epitope to the N-terminus of Ddc2 or Ddc2-S4, respectively. The PCR product was cleaved with MluI and SalI. The EcoRI-MluI and the MluI-SalI fragments were cloned into YEplac195, generating YEp195-GAL-His-HA-Ddc2 or YEp195-GAL-His-HA-Ddc2-S4, respectively. The YCpT-Rad53-HA plasmid has been described [49]. The dna2-W128A, Y130A (dna2-AA) mutation [26] was integrated into the own locus after PCR fusion [79] using primers KS2943, KS2944, KS2955 and KS2946. The MRC1-HA::TRP1 construct was generated by a PCR-based method [80] using the primer pair KS3649 and KS3650. The strains used in this study are listed in S1 Table.
Quantitative PCR analysis of DNA end resection was performed as described previously [51]. HO cleaves the HO cut site and generate a DSB. The DNA was digested with the EcoRI restriction enzyme that cleaves the amplicons at 0.8 kb and 5.8 kb from the DSB, but not in the SMC2 control region. The ssDNA percentage over total DNA was calculated using the following formula: % ssDNA = [100/[(1+2ΔCt)/2]]/f, in which ΔCt values are the difference in average cycles between digested template and undigested template of a given time point and f is the HO cut efficiency [51]. HO cutting efficiency was determined as described [81]. The oligonucleotides used are listed in S2 Table.
The yeast strain (mec1Δ ddc2Δ sml1Δ; KSC3218) was transformed with pRS424-GAL-FLAG-MEC1 and YEp195-GAL-His-HA-DDC2 or YEp195-GAL-His-HA-DDC2-S4. Transformed cells were grown in sucrose media (2% sucrose 0.05% glucose) to a log-phase and incubated with 2% galactose for 5 hr to induce expression from the GAL promoter. Crude extracts were prepared from 10 gram of cells in 50 ml of buffer A (20 mM Tris-HCl [pH 8], 10% glycerol, 3 mM DTT, 0.1% Triton X-100) containing 1 mM EDTA, 100 mM NaCl and inhibitors (1 mM phenylmethylsulfonyl fluoride (PMSF), 1 μg/ml leupeptin, 1 mM benzamidine, 1 mM Na3VO4) by vortexing with 600 μl of glass beads. After clearing by centrifugation, supernatant was incubated with 2 ml of ANTI-FLAG-M2 affinity agarose (Sigma) for 2 hr. Resin was washed with 20 ml of buffer A containing 400 mM NaCl, 10 ml of buffer A containing 100 mM NaCl, 20 ml of buffer A containing 100 mM NaCl, 5 mM MgCl2 and 1mM ATP, and 20 ml of buffer A containing 100 mM NaCl. FLAG-tagged protein was eluted with 4 ml of buffer A containing 100 mM NaCl, 300 μg /ml of FLAG-peptide (Sigma), 2.5 mM MgCl2, 5U of Benzonase (Millipore). The FLAG-eluate was incubated with 1 ml of Ni-NTA-agarose (Clontech) for one hour, washed with 5 ml of buffer A containing 100 mM NaCl. Bound protein was eluted with 1.5 ml of buffer A containing 150 mM NaCl and 300 mM imidazole and then concentrated using a Vivaspin 500 column (GE Healthcare) with buffer A containing 150 mM NaCl. All the protein purification procedures were performed at 4°C.
The coding sequences for DDC2 and ddc2-S4 were amplified by PCR using YCpT-myc-DDC2 [82] and YCp-myc-DDC2-S4 [49] with the primer pair KS3620 and KSX001, and cloned into the BamHI and SalI sites of pMAL-c2X (New England Biolabs) to generate the plasmid pMAL-Ddc2 and pMAL-Ddc2-S4, respectively. Proteins were expressed in E. coli Rosetta (Novagen) after the incubation with 1 mM IPTG at 30°C for 4 hr. The cell pellet from one liter of culture was suspended in 50 ml of buffer M (25 mM Tris-HCl pH 7.5], 10% glycerol, 0.5 mM EDTA, 1 mM DTT) containing 300 mM NaCl and protease inhibitors (leupeptin and pepstatin A at 5 μg/ml each, 1 mM PMSF). After sonication, crude cell lysates were clarified by centrifugation and then incubated with pre-equilibrated 1 ml of amylose resin (New England Biolabs) for 2 hr. After washing with buffer M containing 1 M NaCl, bound proteins were eluted with 2 ml of buffer M containing 300 mM NaCl and 10 mM maltose. Eluates were pooled and concentrated using Vivaspin 500 columns.
Streptavidin beads (4 μl; Pierce) coated with biotinylated oligonucleotides were incubated with RPA for 30 min and further incubated for 30 min after the addition of Mec1-Ddc2 or MBP-Ddc2 proteins in 500 μl of the binding buffer B (20 mM Tris-HCl [pH 7.5], 100 mM NaCl, 0.01% NP-40, 10% glycerol, 100 μg/ml bovine serum albumin) at 30°C. Beads were recovered and subjected to immunoblotting analysis.
Mec1-Ddc2 and RPA were incubated with or without oligonucleotides in the binding buffer B containing ANTI-FLAG-M2 affinity agarose for 30 min at 30°C. Beads were subjected to immunoblotting analysis.
Kinase reactions were carried out by the addition of Mec1-Ddc2 or Mec1-Ddc2-S4 in 40 μl (final volume) of the kinase buffer (20 mM Hepes-KOH [pH 7.5], 10 mM NaCl, 10 mM MgCl2, 4 mM MnCl2, 50 μM ATP) containing 5 μCi [γ-32P] ATP (3,000 Ci/mmol) and 400 nM GST-Rad53. To detect RPA phosphorylation, GST-Rad53 was omitted in kinase reactions. GST-Rad53 was purified as described [49]. Each reaction contains 5 nM purified Mec1-Ddc2 or Mec1-Ddc2-S4. Before initiating kinase reactions, Mec1-Ddc2 or Mec1-Ddc2-S4 was incubated with RPA, ssDNA/oligonucleotide or RPA-ssDNA complex in 4 μl of 15 mM Tris-HCl [pH7.5], 100 mM NaCl, 0.025 mM EDTA for 15 min at 30°C. RPA-ssDNA complex was prepared by mixing RPA and ssDNA/oligonucleotide for 30 min at 30°C. After 10 min of incubation at 30°C, the kinase reactions were terminated by the addition of 5x SDS-sample buffer. The reaction mixtures were separated on SDS-polyacrylamide gels, and phosphorylation was quantified with a phosphor imager system (Typhoon 8600, GE Healthcare).
Cells were incubated with α-factor (6 μg/ml) or nocodazole (15 μg/ml) for 2 hr to synchronize at G1 or G2/M, respectively. Chromatin immunoprecipitation assay and immunoblotting analysis were carried out as described [49, 78]. DNA flow cytometry was carried out by using FACSCalibur (BD Biosciences) [49]. Budding yeast RPA protein was purified as described [83]. Anti-Rfa1 and anti-Rfa2 antibodies were obtained from Steve Brill (Rutgers, Piscataway). Anti-Rad53 antibody (EL7.E1) was purchased from Abcam. The ribbon diagram of Ddc2 was generated by PyMOL (Palo Alto, CA) using the PDB data base (PDB ID: 5X60).
|
10.1371/journal.pgen.1003558 | H3.3-H4 Tetramer Splitting Events Feature Cell-Type Specific Enhancers | Previously, we reported that little canonical (H3.1–H4)2 tetramers split to form “hybrid” tetramers consisted of old and new H3.1–H4 dimers, but approximately 10% of (H3.3–H4)2 tetramers split during each cell cycle. In this report, we mapped the H3.3 nucleosome occupancy, the H3.3 nucleosome turnover rate and H3.3 nucleosome splitting events at the genome-wide level. Interestingly, H3.3 nucleosome turnover rate at the transcription starting sites (TSS) of genes with different expression levels display a bimodal distribution rather than a linear correlation towards the transcriptional activity, suggesting genes are either active with high H3.3 nucleosome turnover or inactive with low H3.3 nucleosome turnover. H3.3 nucleosome splitting events are enriched at active genes, which are in fact better markers for active transcription than H3.3 nucleosome occupancy itself. Although both H3.3 nucleosome turnover and splitting events are enriched at active genes, these events only display a moderate positive correlation, suggesting H3.3 nucleosome splitting events are not the mere consequence of H3.3 nucleosome turnover. Surprisingly, H3.3 nucleosomes with high splitting index are remarkably enriched at enhancers in a cell-type specific manner. We propose that the H3.3 nucleosomes at enhancers may be split by an active mechanism to regulate cell-type specific transcription.
| In our previous study, we unexpectedly discovered that nucleosomes containing the variant H3.3 histones experience substantial splitting events, resulting hybrid nucleosomes containing both “old” and “new” H3.3–H4 dimers. Here, we mapped the genomic distribution of these splitting events at the genome-wide level and studied the connections among gene transcriptional activity, H3.3 nucleosome occupancy, H3.3 nucleosome turnover and H3.3 nucleosome splitting events. We found that H3.3 nucleosome splitting events are better markers that reflect the transcriptional activity. Moreover, we discovered that H3.3 nucleosome splitting events feature the cell-type specific enhancers, which do not appear to the mere consequence of H3.3 nucleosome turnover. These findings may suggest an active mechanism regulating the H3.3 nucleosome splitting events at the enhancers.
| H3.3 is a variant histone that differs from the canonical H3 histones by four amino acids [1]–[4]. Unlike the canonical histones that are incorporated in the replication-dependent pathway, H3.3 histones can also be deposited in a replication-independent manner [5]. Genome-wide profiling experiments in Drosophila cells demonstrated a general enrichment of H3.3 histones at actively transcribing genes [6] and a localized enrichment at the Polycomb responsive elements (PRE) [7]. In mammals, the HIRA complex mediates the incorporation of H3.3 histones at active genes [8], [9] whereas the ATRX-DAXX complex mediates the deposition of H3.3 histones at telomeric and pericentric heterochromatin [9]–[11].
Histone modifications carry important epigenetic information [12]–[14]. Understanding how the patterns of histone modification are transmitted to daughter cells during mitotic division is a highly interesting topic [15]–[20]. We reported that the lysine methylation of histones does not necessarily proceed in a symmetrical fashion within each nucleosome [21] and that canonical (H3.1–H4)2 tetramers undergo conservative segregation during replication-dependent chromatin assembly [22]. These studies ruled out a model in which the faithful copying of modifications within each nucleosome serves as the general mechanism for the inheritance of histone modification-based epigenetic information [23]. However, the existence of such a mechanism at specific genomic regions remains possible [24], for example, at certain regulatory sites [25].
Unlike the canonical (H3.1–H4)2 tetramers that rarely split, we reported that the (H3.3–H4)2 tetramers experience splitting events at a ratio of approximately 10% in each cell division in HeLa cells [22]. Here, we report the mapping of occupancy, turnover rate and splitting events for H3.3 nucleosomes at the genome-wide level. We found a remarkable enrichment of the (H3.3–H4)2 tetramer splitting events at cell-type specific enhancers, which may suggest a potential connection between the H3.3 nucleosome splitting and the maintenance of the lineage-specific transcription status.
To map the genome-wide distribution pattern of the (H3.3–H4)2 tetramer splitting events, “hybrid” mononucleosomes that contain both newly synthesized and existing old H3.3–H4 dimers must be purified. Accordingly, we established a stable HeLa cell line that contains dual, inducibly expressed H3.3: a Flag-tagged histone H3.3 under the control of a tetracycline-inducible promoter and an HA-tagged H3.3 under the control of a Ponasterone A-inducible promoter (Figure S1). We previously reported that ectopically expressed H3 histones accounted for less than 3% of the total H3 histones [22]. Consistently, the HA- and Flag-tagged H3.3 histones were readily detected by antibodies against these epitope tags; in contrast, the larger version of the ectopically expressed H3.3 histones were barely detectable using antibodies against H3 (Figure 1A).
The Flag-H3.3 and HA-H3.3 histones were then allowed to express at distinct time periods, which designates the HA-H3.3 histones as the “old” H3.3 and the Flag-H3.3 histones as the “new” H3.3 (Figure 1B). Mononucleosomes were prepared from these cells (Figure S2A) and then subjected to chromatin immunoprecipitation (ChIP). The Flag-H3.3 or HA-H3.3 containing mononucleosomes were purified with a single-round ChIP to generate the pools of “new” or “old” H3.3 nucleosomes, respectively, and the split H3.3 nucleosomes were selectively purified by sequential ChIPs with antibodies against the Flag and HA tags (Figure S2B). To ensure that we indeed sequence DNA samples from the split mononucleosomes, the library DNA fragments between 200 and 300 bp were size-fractionated (Figure S2C) prior to single-end sequencing (Figure S2D), because 92 bp of adapter sequences were ligated to the DNA samples during library construction.
To genome-widely map total H3.3 nucleosomes distribution, sequencing results from the two single-round ChIPs, which consisted of both the old (HA-H3.3) and new (Flag-H3.3) nucleosomes were pooled and analyzed using the sliding-windows method (See the Materials and Methods). In total, we identified 732,944 well-positioned H3.3 nucleosomes.
The genome-wide turnover kinetics for nucleosomes has been reported previously [26]. However, the genome-wide H3.3 nucleosome turnover pattern has not been specifically determined. Because we selectively purified the old (HA-tagged) and new (Flag-tagged) H3.3 nucleosomes (Figure 1B), we were able to compare their genomic profiles. Generally, both the new (Flag-tagged) and old (HA-tagged) H3.3 histones were enriched around the transcription start sites (TSS) and depleted at the transcription end sites (TES) (Figure 1C–1D). However, the old H3.3 nucleosomes displayed a broader cleft around the TSS than the new H3.3 nucleosomes (Figure 1C–1D), suggesting that the H3.3 nucleosomes at the TSS experience a higher turnover rate than the H3.3 nucleosomes located elsewhere.
The above experiments provided some hints about the turnover of H3.3 nucleosomes. However, these experiments were not specifically designed for determining the turnover rate of H3.3 nucleosomes. We attempted to develop a mathematic model using the above data set, but too many approximations had to be incorporated into the equations, which may affect the accuracy of the model. In order to directly measure the turnover rate of H3.3 nucleosomes, we performed a second set of experiments, in which we induced the expression of HA-H3.3 histones for 48 h and then switched it off (Figure 2A). Cells were harvested at 0 h, 24 h and 48 h after the termination of induction. Mononucleosomes were prepared from these cells and then subjected to ChIP-Seq with antibodies against HA (Figure S2D).
The ChIP-Seq profiles at 24 h and 48 h time points were compared to the ChIP-Seq profile at 0 h to generate the turnover index (T24 and T48) respectively, for each H3.3 nucleosome (See the Materials and Methods for details). We calculated the Pearson correlation of T24 and T48 and they displayed clear positive correlation (r = 0.72). To directly visualize the above results, we plotted the two-dimensional histogram for T24 and T48 of all H3.3 nucleosomes. Indeed, turnover index at these two time points displayed clear positive correlation, and T48 were generally higher than T24 (Figure 2B). These results collectively reflected the linearity and continuity of H3.3 nucleosome turnover during the tested time window. In all further analysis, we focused on T48 because that was the same time window we used to determine the splitting events and T48 would allow us to perform a direct comparison.
We analyzed the H3.3 nucleosome turnover event along protein coding genes. Indeed, the H3.3 nucleosomes at the TSS regions had higher turnover index and active genes generally displayed higher H3.3 nucleosome turnover (Figure 2C), confirming what we observed earlier (Figure 1C–1D).
We sorted all of the genes by their expression levels, from high to low, as represented by the RPKM (reads per kb of exon per million mapped reads) obtained from the RNA-Seq experiments, and then plotted the turnover index of the H3.3 nucleosomes at the +1 nucleosome of each gene against its RPKM. The H3.3 nucleosome turnover index at this region displayed a moderate decline for active genes within top 60% expression levels, although the expression levels of these genes could differ for more than 300 fold determined by their RPKM (Figure 2D). H3.3 nucleosome turnover at the TSS regions appeared in a bimodal distribution (Figure 2D) suggesting that genes are either active with high H3.3 nucleosome turnover or silenced with low H3.3 nucleosome turnover, rather than exhibiting a linear correlation between H3.3 nucleosome turnover and the transcriptional activity.
We then sorted the H3.3 nucleosomes by their corresponding turnover index (T48) from high to low. H3.3 nucleosomes that scored within the top 5% of the turnover index were defined as “high turnover” nucleosomes, and those scored within the bottom 5% of the turnover index were defined as “low turnover” nucleosomes. The “high turnover” H3.3 nucleosomes were relatively enriched at the promoters, 5′ UTRs and 3′ UTRs (Figure 2E), which is consistent with our observations in Figure 2C.
“Hybrid” mononucleosomes containing both newly synthesized Flag-H3.3 and old HA-H3.3 were purified and mapped according to experimental procedures described in Figure 1A. Then we developed a computational model to score the (H3.3–H4)2 tetramer splitting events and assigned an H3.3 nucleosome splitting index (S) for each H3.3 nucleosome (See the Materials and Methods for details).
Two challenges need to be addressed while developing the mathematic model for scoring the splitting events. Firstly, endogenous H3.3 histones exist in our system and they could form heterotetramers with both tagged versions, which cannot be monitored in our study. To solve this problem, we started with an adequate amount of cells, approximately 1.2×109 cells (Figure S2B). The portion of detectable splitting events at any given genomic loci could be estimated to be [% Flag-tagged H3.3]×[% HA-tagged H3.3]×splitting rate, which should be approximately at the range of 10−5–10−4, because the tagged-H3.3 histones were at the level of approximately 5–10% total H3.3 histones and the global splitting rate was approximately 10% [22]. Therefore, for any given loci, we were able to capture approximately 104∼105 splitting events, which allowed us to study a representative population of the total splitting events and to obtain a relative measurement of the splitting rate.
Secondly, at regions with the highest turnover rate, it is a concern that our approach may fail to capture the split nucleosomes. To clarify this concern, we categorized all H3.3 nucleosomes according to their turnover index (T48) range and compared their splitting index. H3.3 nucleosomes with higher turnover index clearly associated with higher splitting index (Figure 3A). To obtain an amplified view for H3.3 nucleosomes with the highest turnover rate, we further categorized these nucleosomes according to their turnover index (T48) range, and observed a similar trend, with the exception of the last group (T48 range 0.99–1.0). But there were only 13 H3.3 nucleosomes defined in this group, and the results for this group may not be statistically meaningful. Taken together, H3.3 nucleosomes with the highest turnover index (T48) generally displayed the highest splitting index, suggesting that our approach could capture the splitting events at regions with the highest turnover.
The above results suggested that H3.3 nucleosomes with higher turnover tend to associate with higher splitting events. To test the relationship between the H3.3 nucleosome turnover and splitting events further, we calculated the Pearson correlation between the turnover index (T) and splitting index (S) for all of the H3.3 nucleosomes and found a moderate positive correlation (r[S, T] = 0.3).
We then sorted all of the H3.3 nucleosomes by their corresponding splitting index (S), from high to low. The H3.3 nucleosomes scoring within the top 5% of the splitting index were defined as “split” nucleosomes, which possessed higher numbers of reads in the sequential ChIP than in the single-round ChIPs. The H3.3 nucleosomes scoring within the bottom 5% of the splitting index were defined as “non-split” nucleosomes, which possessed high numbers of reads in the single-round ChIPs but no reads in the sequential ChIP. Next, we calculated the frequency at a discrete turnover index range for the total H3.3 nucleosomes, the “split” nucleosomes and the “non-split” nucleosomes, respectively (Figure 4). The “non-split” H3.3 nucleosomes displayed a moderate lower turnover profile than the total H3.3 nucleosomes (Figure 4A, 4C and 4D), while the “split” H3.3 nucleosomes were slightly enriched at the high turnover range (Figure 4B and 4D). Nevertheless, the majority of non-split nucleosomes were within the high turnover range (Figure 4C–4D). These observations suggest that the H3.3 nucleosome splitting events are unlikely to be merely the consequence of the H3.3 nucleosome turnover.
We next examined the relationship between the H3.3 nucleosome splitting events and transcriptional activity. For total H3.3 nucleosomes, split H3.3 nucleosomes (within the splitting index top 5%) and non-split H3.3 nucleosomes (within the splitting index bottom 5%) localized at genes, we individually examined their distribution profiles within different classes of genes that were categorized by their expression levels. Approximately 36% of total H3.3 nucleosomes and 25% of non-split H3.3 nucleosomes were localized at genes within the top 25% for expression levels. In contrast, 41% of the split H3.3 nucleosomes were located at these genes (Figure 5A), which is a significant difference with P value less than 0.0001 analyzed with Chi-square test. On the other hand, we found that 7% of total H3.3 nucleosomes and 5.5% of the split H3.3 nucleosomes were localized at genes within the bottom 25% for expression levels. However, 12% of non-split H3.3 nucleosomes were located at these genes (Figure 5A), which is also a significant enrichment with P value less than 0.0001. These data suggest that the split H3.3 nucleosomes were relatively enriched at active genes and non-split H3.3 nucleosomes were enriched at inactive genes.
For each gene, we scored its normalized split nucleosome occupancy (the number of split nucleosomes normalized against the number of H3.3 nucleosomes) and the non-split nucleosome occupancy (the number of non-split nucleosomes normalized against the number of H3.3 nucleosomes) and then plotted them against the RPKM. The split nucleosomes were enriched at the active genes, even after the normalization against the levels of H3.3 occupancy, while the non-split nucleosomes showed enrichment at the inactive genes (Figure 5B). Therefore, we conclude that the H3.3 nucleosome splitting events are better markers of active genes than H3.3 nucleosome occupancy.
The H3.3 nucleosomes were reported to display cell-type specific enrichment at intergenic regions bound by multiple transcription factors, suggesting an enrichment of the H3.3 nucleosomes at the enhancers [9], which prompted us to interrogate the splitting events at the enhancers. Interestingly, it appears to be quite obvious that split H3.3 nucleosomes are enriched at a number of enhancers that we looked into (Figure 6A and Figure S3).
Genome-wide distribution of enhancers was previously determined in HeLa cells, based on distinct pattern of histone modifications, including enrichment of H3K4me1 and H3K27ac [27]. Because we used the same cells in this study to map the splitting events, we were able to examine the H3.3 nucleosome splitting events at these enhancers (each enhancer was arbitrarily defined as a 2 kb region). We first scored the relative enrichment for the H3.3 nucleosomes, the split H3.3 nucleosomes and the non-split H3.3 nucleosomes at various genomic features. Indeed, the H3.3 nucleosomes were enriched at the enhancers in HeLa cells, with a comparable fold-enrichment in the promoter regions and 5′ UTRs (Figure 6B). Strikingly, the split H3.3 nucleosomes were far more enriched at the enhancers than any of the other genomic features tested (Figure 6B). We found that 10% of the total H3.3 nucleosomes were located at the HeLa enhancers whereas 31% of the split H3.3 nucleosomes (within the splitting index top 5%) were located at the HeLa enhancers; in contrast, only 3% of the non-split H3.3 nucleosomes (within the splitting index bottom 5%) were located at the HeLa enhancers (Figure 6C). In addition, approximately 37% of all of the intergenic split H3.3 nucleosomes were specifically located at the HeLa enhancers, but such enrichment was not observed at the K562 cell-specific enhancers (Figure 6C). These data suggest that H3.3 nucleosome splittings are frequent events that feature cell-type specific enhancers.
In the above analysis, an arbitrary 5% cut off was employed. To obtain the continuity of this analysis, we sorted all H3.3 nucleosomes by their splitting index from high to low. Then we used a sliding window of 5000 H3.3 nucleosomes and analyzed the percentage of those nucleosomes that reside in the enhancers. The percentage of H3.3 nucleosomes reside in the HeLa enhancers declined continuously along with the reduction of their splitting index (Figure 6D). We noticed that H3.3 nucleosomes with high splitting index also displayed a minor enrichment at the K562 enhancers (Figure 6D). Interestingly, such enrichment was diminished when common enhancers between HeLa cells and K562 cells were excluded from the analysis (Figure 6E). This further supports that H3.3 splitting events are enriched at active enhancers in a cell-type specific manner.
To further investigate the relationship between the H3.3 nucleosome splitting events and enhancers, we divided the entire human genome into 10-kb intervals and sorted them by their split H3.3 nucleosome numbers. We then plotted the H3.3 nucleosome split number of these 10-kb intervals against their overlapping percentage with the enhancers. Those genomic intervals with higher numbers of split H3.3 nucleosomes clearly displayed a higher overlap with the HeLa enhancers but not the K562 enhancers (Figure 6F). Moreover, the overlapping percentage with the enhancers dropped to background level when the number of split H3.3 nucleosomes declined to zero. A minor overlap with K562 enhancers was observed for the genomic regions with high numbers of split H3.3 nucleosomes (Figure 6F), which was again diminished when the common enhancers between the two cell lines were excluded (Figure 6G).
We also sorted the 10-kb genomic intervals by their number of H3.3 nucleosomes, and then analyzed their overlap with enhancers. Those regions with high numbers of H3.3 nucleosomes were significantly enriched at the HeLa enhancers, but also at the K562 enhancers, regardless whether the common enhancers were excluded from the analysis or not (Figure S4). These data collectively suggest that the enrichment of H3.3 nucleosome splitting events, but not H3.3 occupancy, feature cell-type specific enhancers.
Considering the high turnover H3.3 nucleosomes (within the turnover index top 5%) were also enriched at enhancers (Figure 6C), and H3.3 nucleosome splitting index displayed modest positive correlation with turnover index (T48) (Figure 3 and Figure 4), it is necessary to examine the correlation between the splitting events and turnover at the enhancers.
We first categorized H3.3 nucleosomes into two groups, the enhancer group (all H3.3 nucleosomes reside in enhancers) and the non-enhancer group (all H3.3 nucleosomes do not reside in enhancers), and then we compared the splitting index of these two groups within the same turnover range. Interestingly, enhancer H3.3 nucleosomes displayed higher split index than non-enhancers H3.3 nucleosomes with similar turnover index (Figure 7A).
We then performed similar comparison for three groups of H3.3 nucleosomes located at the enhancers, promoters and 5′-UTRs, because these regions displayed comparable enrichment of H3.3 nucleosomes (Figure 6B). H3.3 nucleosomes at the enhancers were clearly associated with higher splitting index than H3.3 nucleosomes at the promoters or 5′-UTRs with similar turnover index (Figure 7B).
Furthermore, for H3.3 nucleosomes at comparable turnover ranges, significantly higher percentage of enhancer H3.3 nucleosomes were called as “split” nucleosomes than the non-enhancer H3.3 nucleosomes (Figure 7C). These results support the notion that splitting events are not merely the consequence of nucleosome turnover and there might be active splitting mechanism(s) at the enhancers. We also noticed that the percentage of “split” nucleosomes displayed a difference greater than two-fold between the enhancer group and the non-enhancer group at most turnover range (Figure 7C), except for H3.3 nucleosomes with the highest turnover (T48 between 0.95 and 1), but the difference remained to be statistically significant (Figure 7C). These results are consistent with our observation that high turnover (Figure 3 and Figure 7) and enhancer enrichment (Figure 6 and Figure 7) are both associated with “split” H3.3 nucleosomes.
Finally, to exclude any potential bias of our sequential ChIP experiments, we co-expressed both tagged H3.3 nucleosomes at the same time and then performed single-round and sequential ChIP-Seq experiments (Figure S2D). As expected, profiles of the single-round and sequential ChIP-Seq results were highly similar (Figure S5) and no specific enrichment of dual-tagged H3.3 nucleosomes was observed at the HeLa enhancers (Figure 7D).
Active genes have higher H3.3 nucleosomes occupancy [5], [6], [9] and higher nucleosome turnover [26]. In this report, we found that active genes are also associated with higher H3.3 nucleosome turnover and higher H3.3 nucleosome splitting events. But what are the relationships among all these events? Do they simply reflect one event at the active genes or they might have different roles? Recently, the yeast nucleosomes, which consist the “H3.3-like” H3 histones [3] were shown to display some level of splitting events [28], similar to the human H3.3 nucleosomes [22]. In yeast, active genes also tended to have higher nucleosome splitting signals [28], similar to our observation (Figure 5). Actively transcribing genes have a higher nucleosome turnover and a higher nucleosome splitting; therefore it appears to be logical to think that these events might be directly correlated with each other. However, by comparing these parameters at the genome-wide level, we found that, although these events are indeed correlated with active transcription, there is only a moderate correlation between the H3.3 nucleosome turnover and the H3.3 nucleosome splitting events (Figure 3 and Figure 4). Nucleosomes with the same turnover index exhibited different splitting indexes at enhancer regions and non-enhancer regions (Figure 7), which also suggests these events are not directly correlated. We believe that this is because of the fact that neither of these events is linearly correlated with the transcriptional activity (Figure 3D and Figure 5B).
The surprising observation that H3.3 nucleosomes with high splitting index were remarkably enriched at cell-type specific enhancers (Figure 6 and Figure 7) may suggest a role for nucleosome splitting in regulating cell-type specific transcription, especially when these splitting events are clearly not the mere consequence of H3.3 nucleosome turnover (Figure 7). We propose the existence of active mechanism(s) at cell-type specific enhancers, which may regulate lineage-specific transcription.
This study represents a first attempt to unveil the role of H3.3 nucleosome splitting events. The unexpected observation of the enrichment of these events at enhancers is highly interesting. However, there are more questions than answers at this stage regarding the functional significance and molecular mechanism of this observation. Function-wise, it would be interesting to ask whether this event is related to the transmission of epigenetic modifications, as previously proposed [25] or whether this event maybe relevant to cell fate determination. One interesting experiment is to transplant the current detection system into the stem cell systems, and to ask whether splitting events may localize differently in cells at the self-renewal stage and differentiated stage. It is also highly interesting to understand the molecular mechanism of the splitting events and even to manipulate the splitting events. We speculate that the H3.3 deposition chaperones and the chromatin remodelers may participate in the splitting events. However, it is highly challenging to draw a firm conclusion without a strategy that uncouples the splitting events and the H3.3 nucleosome deposition pathway.
Stable HeLa cells expressing Flag-H3.3 histones under the control of a tetracycline-inducible promoter were established in our previous work [22]. These cells were stably engineered with the pIND system (Invitrogen) to express the HA-H3.3 histones under the control of a Ponasterone A-inducible promoter.
Mononucleosomes were prepared, according to the literature [29], using 1.2×109 cells that were sequentially induced (Figure 1B). Affinity purified mononucleosomes were eluted with Flag or HA peptides according to our previous reports [21], [22]. DNA samples were extracted from affinity-purified mononucleosomes by phenol extraction and ethanol precipitation. Solexa sequencing libraries were constructed with the NEBNext DNA Sample Prep Master Mix Set 1 Kit following the manufacturer's instructions, and then subjected to single-end sequencing on Illumina Genome Analyzer II.
The sequencing reads of Flag-H3.3 (new) and HA-H3.3 (old) were pooled and shifted to the center of fragment size of 150 bp. The middle half size (i.e. 75 bp) of each read was kept for the following analysis. The bigwig profile was generated with the NPS program [30]. A 150 bp-window was used to scan the hg19 genome with a 10 bp step. The window was defined as a nucleosome if the reads profile met the following criterions: the middle point of the window was the highest or the 2nd highest and greater than 9; the middle point was greater than the 30% percentile of the window. For nucleosomes with the distance of 10 bp, the more symmetric one was kept. “Adjacent” nucleosomes called by the program with more than 50% overlap were considered as fuzzy nucleosomes and excluded from further analysis.
The RNA-Seq experiment was performed according to a previous publication [31]. The sequencing reads were mapped to human genome hg19 using Tophat [32], and the RPKM values were quantified using Cufflinks [33].
To measure turnover index at relatively late time points required for the splitting assay, adequate amount of starting cells (approximately 1×108 cells) were used for ChIP-Seq experiments illustrated in Figure 2A. The turnover index was calculated by comparing HA-H3.3 profiles at time point t (Ht) and time point 0 h (H0). For each H3.3 nucleosome, the signal ratio between the two time points (Ht/H0) was defined as R48 or R24. For each time point t, the signal ratio at H3.3 nucleosome with the lowest turnover in the genome was defined as Rt,max; likewise, the signal ratio at H3.3 nucleosome with the highest turnover in the genome was defined as Rt,min. To normalize the final turnover index (Tt) to the range of 0 to 1, the turnover index for each H3.3 nucleosome was calculated as:(1)For a given genomic location, the amount of dual-tagged “hybrid” nucleosomes (D) sequentially purified according to Figure 1B can be determined by the amounts of Flag-H3.3 nucleosome (F), HA-H3.3 (H), total available H3.3 nucleosomes (N) and the splitting index (S) using the following equation.(2)The variables F, H and D were determined according to the ChIP-seq results from the Flag single-round, HA single-round and sequential ChIP-seq experiments. The variable N cannot be directly determined, but the amount of total available H3.3 nucleosomes should be proportional to the HA-H3.3 nucleosomes after a long period (48 h) of induction (H0), which has been determined in 0 h time point according to experiments described in Figure 2A. Therefore, , and:(3)and:(4)
ChIP-Seq data have been deposited in the NCBI Sequence Read Archive under accession No. SRA043915. (http://www.ncbi.nlm.nih.gov/Traces/sra/sra.cgi)
|
10.1371/journal.pntd.0002050 | Mosquito Cellular Factors and Functions in Mediating the Infectious entry of Chikungunya Virus | Chikungunya virus (CHIKV) is an arthropod-borne virus responsible for recent epidemics in the Asia Pacific regions. A customized gene expression microarray of 18,760 transcripts known to target Aedes mosquito genome was used to identify host genes that are differentially regulated during the infectious entry process of CHIKV infection on C6/36 mosquito cells. Several genes such as epsin I (EPN1), epidermal growth factor receptor pathway substrate 15 (EPS15) and Huntingtin interacting protein I (HIP1) were identified to be differentially expressed during CHIKV infection and known to be involved in clathrin-mediated endocytosis (CME). Transmission electron microscopy analyses further revealed the presence of CHIKV particles within invaginations of the plasma membrane, resembling clathrin-coated pits. Characterization of vesicles involved in the endocytic trafficking processes of CHIKV revealed the translocation of the virus particles to the early endosomes and subsequently to the late endosomes and lysosomes. Treatment with receptor-mediated endocytosis inhibitor, monodansylcadaverine and clathrin-associated drug inhibitors, chlorpromazine and dynasore inhibited CHIKV entry, whereas no inhibition was observed with caveolin-related drug inhibitors. Inhibition of CHIKV entry upon treatment with low-endosomal pH inhibitors indicated that low pH is essential for viral entry processes. CHIKV entry by clathrin-mediated endocytosis was validated via overexpression of a dominant-negative mutant of Eps15, in which infectious entry was reduced, while siRNA-based knockdown of genes associated with CME, low endosomal pH and RAB trafficking proteins exhibited significant levels of CHIKV inhibition. This study revealed, for the first time, that the infectious entry of CHIKV into mosquito cells is mediated by the clathrin-dependent endocytic pathway.
| Deciphering the much neglected aspects of cellular factors in contributing to the infectious entry of CHIKV into mosquito cells may enhance our understanding of the conservation or diversity of these host factors amongst mammalian and arthropod for successful CHIKV replication. The study revealed that the infectious entry of chikungunya virus (CHIKV) into mosquito cells is mediated by the clathrin-dependent endocytic pathway. A customized gene expression microarray known to target the Aedes mosquito genome was used to identify host genes that are differentially regulated upon CHIKV infection. A combination of bio-imaging studies and pharmacological inhibitors confirmed the involvement of clathrin-mediated endocytosis as well as the importance of low endosomal pH during CHIKV infectious entry. Furthermore, the clathrin heavy chain, Eps15, RAB5, RAB7 and vacuolar ATPase B are shown to be essential for the infectious entry process of CHIKV. This study aims to underline the importance of cellular factors, particularly those associated with clathrin-dependent endocytosis, in mediating the infectious entry of CHIKV into mosquito cells.
| Chikungunya virus (CHIKV) is an arthropod-borne virus of the genus Alphaviruses, belonging to the family Togaviridae. It is an enveloped, single-stranded, positive-sense RNA virus with a genome size of approximately 12,000 nucleotides. CHIKV virions measure 60–70 nm in diameter and it contains a spherical capsid with icosahedral symmetry. The viral genome encodes for four non-structural (nsP1–P4) and five structural proteins (capsid, E1, E2, 6K and E3) [1], [2]. Embedded in the lipid bilayer surrounding the viral capsids, the E1 and E2 structural proteins enable the virus to be directed to host cells for attachment and fusion with cellular membranes during infectious entry processes [1], [2]. Chikungunya is defined as “bent walker” in Makonde, which refers to the hunched posture observed in patients suffering from persisting arthralgia [3], [4]. Symptoms typically develop from 4–7 days after the bite of an infected mosquito vector. Characterized by high fever, joint pain, headache, vomiting and maculopapular rash, acute CHIKV infection lasts approximately 1–10 days, while chronic CHIKV infection often results in polyarthralgia and myalgia that persist for long periods. Other CHIKV-associated complications reported include lymphopenia, severe skin lesions, lethal hepatitis and encephalitis, with severe neurological symptoms documented during recent outbreaks in Réunion Island [3], [4].
While human transmission of CHIKV occurs via Aedes (Ae.) mosquitoes, particularly Ae. aegypti and Ae. albopictus, other Aedes species such as Ae. furcifer, Ae. taylori, Ae. luteocephalus, Ae. africanus and Ae. Neoafricanus are involved in enzootic cycles [5], [6]. Alphaviruses can be broadly divided into the New World encephalitic viruses and Old World arthritogenic viruses [7], [8]. Along with other widely recognized Old World alphaviruses such as Sindbis (SINV), Semliki Forest (SFV), Ross River (RRV) viruses, CHIKV is responsible for high morbidity rates, accounting for millions of adverse, albeit non-fatal cases [3], [9], [10]. Genomic analysis of previously and recently identified clinical isolates revealed unique molecular features, most prominently a point mutation in the viral envelope E1 glycoprotein (E1-A226V) [9], which was suggested to increase the capability of viral fusion, assembly and tropism that aids in virus transmission [11], thus accounting for the selective advantage of the viral subtype. The presence of the A226V mutation in the CHIKV E1 gene was also reported during a major outbreak of CHIKV infection in the Indian state of Kerala [12]. Based on an SFV model of infection, replacement of the alanine residue at position 226 of the E1 envelope protein to valine was previously observed to affect membrane fusion and is believed to result in differential cholesterol dependence [10], [13].
Viruses can enter host cells through various pathways such as phagocytosis, macropinocytosis, and receptor-mediated endocytosis. Viruses have evolved the ability to penetrate and release the viral genome into the cell cytoplasm, after binding to the cellular receptor(s). Penetration for enveloped RNA viruses includes endocytosis and membrane fusion, the latter of which can either take place in a pH independent manner at the cell surface or within intracellular vesicles (pH-dependent). Majority of viruses require endocytic internalization for productive infection, with the virions being led to appropriate replication sites, thus bypassing many cytoplasmic barriers [14]. In particular, RNA viruses posses the ability to hijack multiple portals of cellular entry. Endocytic pathways such as clathrin-mediated, clathrin-independent, macropinocytosis, caveolar-mediated and caveolar-independent, have been shown to be utilized by numerous viruses [15], [16]. Other less characterized pathways also include lipid raft-mediated endocytosis, in which dynamin participation has been proposed but has not been determined [14].
Microarray studies performed on arboviruses and its mosquito vectors have been limited and aimed at enhancing diagnostics and understanding immune-based antiviral mechanisms [17], [18]. Such studies were previously performed to analyze gene expression profiles of mosquito midguts in response to Sindbis (SINV) infection, and genes associated with vesicle transport and immune cascades were observed to be involved during the infection [19].
Previous studies have been conducted to investigate the different entry pathways of Alphaviruses on various cell lines. SFV and Venezuelan equine encephalitis virus (VEEV) have been shown to enter mammalian cells through pH-dependent endocytic pathway [20]. Additionally, SINV was observed to infect both mammalian and mosquito cells at neutral pH [21], while VEEV was found to enter Ae. albopictus C710 mosquito cells via pH-dependent endocytosis [22]. Analyses of infectious CHIKV entry have been limited to mammalian cells, with several findings noting that CHIKV infection on HEK293T mammalian cells is independent of clathrin heavy chain and dependent of functional Eps15 [3], [4]. However, little is currently known about the infectious CHIKV entry process and pathway into mosquito cells. Deciphering the much neglected aspects of cellular factors in contributing to the infectious entry of CHIKV into mosquito cells may enhance our understanding on the conservation or diversity of these host factors amongst mammalian and arthropod cells for successful CHIKV replication.
This unprecedented study therefore aims to examine the infectious entry processes of CHIKV in mosquito cells. Different strategies targeting cellular endocytosis were used, including customized microarray profiling of mosquito genes involved in endocytic pathways, treatment with specific drug inhibitors, gene knockdown and expression of dominant negative cellular proteins. We demonstrated, for the first time, that CHIKV preferentially uses a clathrin-mediated and Eps15-dependent pathway to enter Ae. albopictus (C6/36) cells. We also showed the importance of endosomal pH acidification in CHIKV entry. Moreover, results from the siRNA-based knockdown of Rab5 and Rab7 genes suggested that CHIKV entry involves the trafficking of virus particles from early to late endosomes. The novelty of deciphering the infectious entry of CHIKV into C6/36 cells potentially allows for better understanding on the pathogenesis of CHIKV infection and the development of potential antiviral strategies.
Ae. albopictus C6/36 cells (American Type Culture Collection) were maintained at 28°C in Leibovitz-15 (L-15) growth medium (Sigma-Aldrich Corp., St Louis, MO, USA) supplemented with 10% fetal bovine serum (FBS) (Hyclone, Cramlinton, UK). Baby hamster kidney (BHK-21) cells (American Type Culture Collection, ATCC CCL-10) were maintained at 37°C in Rosewell Park Memorial Institute (RPMI-) 1640 growth medium (Sigma-Aldrich Corp) supplemented with 10% FBS (Hyclone).The cells were passaged in T75 flasks (Nunc, Denmark) at a 1∶5 dilution every 3–4 days at 70–80% confluency. For experimental infections, C6/36 cells were seeded in T25 flasks to a confluency of 80% that achieved a cell density of ∼3×106 cells/ml. The C6/36 cells were incubated at 37°C for 1.5 hours during virus infection, before being placed at 28°C throughout the remainder of the experiments, in line with the natural temperature for mosquitoes and mosquito cell incubation.
Singapore/07/2008 CHIKV strain was obtained from National Public Health Laboratory, Ministry of Health, Singapore and propagated in C6/36 cells. Low passages of the virus were used throughout this study. CHIKV strains SGEHICHD122508 – (Accession No.: FJ445502.2) and SGEHIDSD67Y2008 – (Accession No.: EU441882.1) were obtained from Environmental Health Institute, National Environmental Agency, Singapore. These virus strains were propagated in C6/36 cells and utilized in low endosomal pH experiments. The virus titers were quantitated using viral infectious plaques assays performed on BHK-21 cells. Growth kinetics were performed on these three different CHIKV strains, with infected and mock-infected samples harvested at various time points of 0, 6, 12, 24, 36, 48, 72, 96 and 120 hours post infection (p.i) on C6/36 cells. A multiplicity of infection (MOI) of 10 was used for most of the experiments throughout the study, to allow for more accurate observations and better detection of CHIKV entry processes into host cells.
Confluent monolayers of C6/36 cells were infected with CHIKV at an MOI of 10. At 24 hours p.i, the supernatant was harvested by centrifugation at 4,500 rcf for 10 mins. CHIKV particles were then concentrated and partially purified by using a centrifugal filter device (Millipore, Billerica, MA, USA) at 1,077 rcf for 2 hours. The partially purified viruses were then purified even further by sucrose gradient centrifugation at 74,766 rcf for 3 hours at 4°C. Finally, the purified virus pellet was resuspended in Tris buffer (50 mM Tris-HCl [pH 7.4]). The titer of the purified virus preparation was determined by viral infectious plaque assay on BHK-21 cells and was found to be 5×1010 PFU/ml. For negative staining of purified CHIKV preparation, 7.5×108 PFU/15 µl of CHIKV was added to freshly glow discharged, carbon-coated grids, and stained with 2% uranyl acetate for 1 min. The grids were then air dried before viewing under the CM120 Biotwin transmission electron microscope (Philips).
C6/36 cells growing on coverslips were incubated with CHIKV at an MOI of 10 for 1 hour at 4°C with gentle rocking. The cells were subsequently washed three times in ice-cold 1× phosphate buffer saline (PBS) to remove unbound viral particles, prior to further incubation for 1 hour at 37°C in growth medium to enable virus penetration. Extracellular virus particles that failed to enter into cells are then inactivated with acid glycine buffer (pH 2.8) (0.1 M potassium hydrogen phthalate and 0.1 M of HCl). Infectious virus entry was traced at different time points upon the addition of CHIKV to C6/36 cells for up to 1 hour post-infection and processed for either ultrastructural analysis via transmission electron microscopy or immunofluorescence assay.
C6/36 cells (1.2×106) were seeded into 24-well plates, and incubated for 24 hours before the drug treatment assays were performed. Pre-treatment drug assays were performed in favour of co- and post-treatment studies, in order to ensure that potential CHIKV inhibition is most likely to occur at the entry step, as opposed to downstream infective phases, such as viral replication. Hence, to determine the effects of the drugs used to inhibit the CHIKV entry, C6/36 cells were pretreated with drugs at different concentrations for 3 hours at 37°C. The pharmacological inhibitors were then removed and cell monolayers were washed twice with 1× PBS, in order to eliminate the possibility of exposure of the virus to the inhibitors. This is to ensure minimal risk of the inhibitors directly influencing the viability of the virus and its subsequent entry into the cells. After 1.5 hours of virus infection at an MOI of 1, the cells were washed thrice with 1× PBS, replaced with fresh L-15 media and incubated for another 24 hours. At 24 hour p.i., supernatants from CHIKV-infected cells were harvested for viral infectious plaque assays. Three independent experiments were carried out for each set of drugs used. Inhibition of receptor- and/or clathrin- mediated endocytosis was performed through the use of chlorpromazine (42, 56, 70 & 84 µM) (Sigma Aldrich) [23], monodansylcadaverine (50, 100, 150 & 200 µM) (Sigma Aldrich) [24] and dynasore (5, 10, 50 & 100 µM) (Sigma Aldrich) [25]. Other inhibitors targeting alternative endocytic pathways included filipin (0.1, 0.5, 1.0, 1.5 & 2.0 µg/ml) (Sigma Aldrich) [26], nystatin (5, 10, 20 & 40 µM) (Sigma Aldrich) [26], methyl-β-cyclodextrin (2.5, 5.0, 7.5 & 10 µM) (Sigma Aldrich) [26] and EIPA (10, 25, 50 & 100 µM) (Sigma Aldrich) [27], [28]. CHIKV infected, 0.1% DMSO treated C6/36 cells acted as solvent control. Endosomal acidification was inhibited by drug treatment of C6/36 cells with concanamycin A (20, 60, 100, 150 & 300 nM –Singapore/07/2008 CHIKV strain) and (80, 100, 150 & 300 nM - CHIKV strains SGEHICHD122508 and SGEHIDSD67Y2008) (Sigma Aldrich) and bafilomycin A (0.1, 1.0, 2.0, 3.0 & 4.0 µM) (Sigma Aldrich) [29], [30]. Other inhibitors performed on C6/36 cells include colchicine (50, 100, 150 & 200 µM) (Sigma Aldrich) [31], nocodazole (1, 5, 10, 15 & 20 µM) (Sigma Aldrich) [32], cytochalasin B (0.1, 1.0, 1.5 & 2.0 µg/ml) (Sigma Aldrich) [23], cytochalasin D (1, 3, 5, 10 & 20 µg/ml) (Sigma Aldrich) [31] and nifedipine (40, 60, 80 & 100 µM) (Sigma Aldrich) [33].
Upon infection, C6/36 cells were harvested at 0 min, 15 mins, 30 mins and 120 mins post infection (pi). At 0 min pi, cells were harvested immediately upon virus inoculation. At each time point, C6/36 cells were washed with 2 ml of the pre-warmed (28°C) maintenance medium. After decanting the maintenance medium, 1 ml of Qiagen Cell Protect solution was added to each flask. Detached cells were transferred into a sterile 2 ml tube and were stored immediately at −80°C until total RNA extraction. Cells were homogenized in 350 µl RLT buffer in QIAshredder spin columns (Qiagen, Hilden, Germany) prior to total RNA extraction with Qiagen RNeasy Protect cell mini kit (Qiagen) according to manufacturer's instructions. Hundred nanograms of total RNA were used for probe synthesis of cy3-labeled cRNA, and hybridizations were carried out on an Aedes mosquito customized gene expression microarray (18760 transcripts from Vector Base Aedes aegypti database with 2 best probes per transcript) in Agilent GE 8×60K array format (Agilent Technologies, California, USA). Hybridization was carried out at 65°C for 17 hours in an Agilent hybridization oven at 10 rpm. After hybridization, microarrays chips were washed in gene expression wash buffer 1 for 1 min at room temperature and 1 min in gene expression wash buffer 2 at 37°C before scanning on the Agilent High Resolution Microarray Scanner (C-model). Raw signal data was extracted from the TIFF image with Agilent Feature Extraction Software (V10.7.1.1). The raw microarray data was processed and analyzed with Partek Genomics Suite (Partek, St Louis, Missouri, USA) to generate values representing fold changes in gene expression. An average of the duplicate values was used to calculate fold change, and each value was then assessed for its statistical significance, using analysis of variance (ANOVA). Host genes demonstrating at least a 1.5-fold change in expression upon CHIKV infection were selected for further investigation. Pathway analysis was subsequently detailed with Ingenuity Pathway Analysis (IPA) 9.0 (Ingenuity Systems 2011, Redwood City, California) and differentially regulated genes involved in the clathrin-mediated endocytic pathway were selected for pathway mapping.
To track the infectious entry process of CHIKV into C6/36 cells at various time points p.i, cells infected with CHIKV at an MOI of 10 were fixed with 2.5% glutaraldehyde (Agar Scientific, Stansted, UK) at 4°C for 20 mins, followed by scraping of the cells and subjecting them to longer fixation at 4°C overnight. The following day, cells were centrifuged and the pellet was washed with PBS and deionized water. The cell pellet was then post-fixed with 1% osmium tetroxide (Ted Pella, Redding, California, USA) and 1% potassium ferro-cyanide for 2 hours, followed by dehydration in an ascending graded series of ethanol and acetone, i.e. 25%, 50%, 75%, 95% and 100% for 10 mins at each concentration. On the following day, cells were infiltrated with resins by passing them through three changes of mixture, comprised of a combination of acetone, ethanol and araldite. The following day, cells were infiltrated with four changes of absolute embedding media with 1 hour incubation at room temperature, 40°C, 45°C and 50°C. After the last spin, cell pellet was resuspended in 100–200 µl of araldite. Mixture was embedded using the BEEN capsule (size 3) and was incubated at 60°C for 24 hours to allow polymerization. The samples were trimmed with an ultramicrotome (Reichert-Jung, New York, USA) and the sections were stained with 2% uranyl acetate and fixed with lead citrate. The stained sections were viewed under the Philip EM 208 transmission electron microscope and images were captured digitally with a dual view digital camera (Gatan, Werrendale, USA).
For immunofluorescence microscopy, C6/36 cell monolayers were first grown on coverslips till 75% confluency. The cells were incubated at 4°C for 30 mins. The cells were allowed to bind to CHIKV at an MOI of 10 for 1 hour at 4°C to allow viral attachment to the cell surface before being shifted to 37°C for 10 mins to promote CHIKV entry into the cell. Cells were fixed in ice-cold methanol at 10 and 15 mins post entry of CHIKV. This is followed by three washes of cold PBS prior to immunofluorescence assay analyses. Rabbit polyclonal antibodies to clathrin (CLTC, Chemicon), early endosomal antigen 1 (EEA1; Novus Biologicals) and CHIKV E2 protein (customized CHIKV13893 B3 rabbit polyclonal, ProSci) were used for immunofluorescence assays. Texas Red (TR)- or FITC-conjugated secondary antibodies were used at a concentration of 1 µg/ml. Lysotracker, a dye for staining live cells were used at a concentration of 75 nM. The specimens were then viewed with Olympus IX81 motorized inverted epifluorescence microscope (Olympus, Tokyo, Japan) with an excitation wavelength of 543 nm for TR and 480 nm for FITC at 63× magnification.
Cell viability upon drug treatments and siRNA transfection was assessed by the Cell Cytoxicity Assay – alamarBlue (Invitrogen, CA, USA) assay according to the manufacturer's recommendations. Briefly, C6/36 cells were seeded in 96-well cell culture plates and subsequently treated with individual siRNAs or drugs for 3 hours, before incubation with alamarBlue reagent solution for 2 hours at 37°C. After which, the plates were subjected to fluorescence detection, at an excitation wavelength of 540 nm–570 nm, and emission wavelength of 580 nm–610 nm (Tecan iControl Reader, Männedorf, Switzerland).
Plasmid constructs of dominant-negative Eps15 (GFP-EΔ95/295) was kindly provided by A. Benmerah, Pasteur Institute, and plasmid constructs backbone EGFP-C2 was purchased from Clontech (CA, USA). Transfections were performed by using Lipofectamine LTX reagents according to manufacturer's recommendation (Invitrogen). Briefly, C6/36 cells were grown on coverslips in 24-well tissue culture plates until 75% confluency. Then, 3.5 µg plasmid constructs were complexed with 4 µl Plus reagent in 25 µl OPTI-MEM medium (Gibco, New York, USA) for 15 mins at room temperature. The mixture was then added to 25 µl OPTI-MEM containing 2 µl Lipofectamine LTX (Invitrogen, USA). After incubation for another 15 mins, the DNA-liposome complexes were added to the cells, prior to further incubation for 3 hours at 37°C. One millilitre of complete growth medium was then added and incubated for another 24 hours before the virus entry assay was carried out.
Different siRNAs targeting various Ae. albopictus genes involved in endocytic processes were selected to perform reverse transfection assays in C6/36 cells, including CLTC (NCBI Accession: XM_001656826), RAB5 (NCBI Accession: XM_001658641), RAB7 (NCBI Accession: EF127648) and vacuolar ATPase B (NCBI Accession: AF092934). The siRNA gene sequences used in this study are, CLTC (CAAUAAAGAUAAUGCCCAU), RAB5 (CGAAUAUUGUGAUUGCGCU), RAB7 (CCUGGAGAAUAGGGCCGUA) and vacuolar ATPase B (GUCAUUCAAGGGAUAAUGU) (Sigma Aldrich). Reverse transfection assays on scrambled siRNA gene sequences were also performed simultaneously to confirm the specificity of the gene targeting siRNAs. The scrambled siRNA gene sequences used in this study are CLTC (ACAGAAUUAAACUACUUGC), RAB5 (ACAGUUUGAGGUACUGUUC), RAB7 (CUCAGAGGGUAACGUCGAG) and vacuolar ATPase B (CUGAAUAUCAGUGGUAUAG). Specific gene targeting siRNAs and scrambled siRNAs were dissolved in DEPC-treated reverse osmosis water to a final stock concentration of 100 µM, and incubated at room temperature for 30 mins with gentle agitation. Different siRNAs were diluted to desired working concentrations of 0.1 nM, 1 nM, 5 nM, 10 nM with serum-free media (Dharmacon, US) and transfection reagent (Dharmafect-1). The specific individual siRNAs that were directed against each of the respective genes were then transfected into C6/36 cells prior to being subjected to CHIKV infection after 48 hours post transfection. The supernatants were then harvested 24 hours p.i for plaque assays.
Validation of gene expression was performed via qRT-PCR. Upon gene silencing, total RNA was extracted from C6/36 cells with RNeasy Extraction Kit (Qiagen). The samples were assayed in a 20 µl reaction mixture containing 10 µl SYBR Green Master Mix (Fermentas, US), 1 µl forward and reverse primer respectively, 1 µl RNA, 1 µl reverse transcriptase and 7 µl nuclease free water. A no-template control was also included. The cycling conditions for one-step SYBR Green-based RT-PCR consisted of a 30-min reverse transcription step at 44°C and 5 mins of Taq polymerase activation at 94°C, followed by 40 cycles of PCR with denaturation occurring at 94°C for 15 s and annealing and extension taking place at 60°C for 30 s. Following amplification, a melting curve analysis was performed to verify the melting temperature of PCR products amplified by the Ae. albopictus gene primer pairs. The primers pairs stated are CLTC (Forward, 5′-CGTTCGGCCAATGCTGC-3′, Reverse, 3′- GGGAAGTCGCTCTGCGCT-5′), RAB5 (Forward, 5′-TCAGCGACAGGCATCGC-3′, Reverse, 3′-CAGCGGTTTTGGCCGAC-5′), RAB7 (Forward, 5′-AACGAAGCGTGCCCAGCAGT-3′, Reverse, 3′-CCGGTTGTTGCGGTCTGCGT-5′), vacuolar ATPase B (Forward, 5′-GCTCGGTCTTCGAGTCGCT-3′, Reverse, 3′-CAGTGTCAGGCGCGAGGTC-5′) and actin controls (Forward, 5′-CCACCATGTACCCAGGAATC-3′, Reverse, 3′-CACCGATCCAGACGGAGTAT-5′).
Where applicable, statistical analyses were performed on repeated measurements using the one-tailed Student's t-test. The significance level was set at p<0.05 (*), p<0.01 (**) or p<0.001 (***). Data shown throughout the study were obtained from three independent experiments.
A customized gene expression microarray chip consisting of 18,760 transcripts targeting the Ae. aegypti mosquitoes was used to profile differentiated regulation levels of host genes necessary for the infectious entry of CHIKV. A total of 579 targeted mosquito genes were found to be differentially regulated – defined as fold change of less than −1.5 or more than 1.5 - upon CHIKV infection. Among these genes – many of which are known to be involved in generalized host immune responses, such as the IFN-associated pathway - are those related to clathrin-mediated endocytosis. Genes associated with other endocytic pathways, such as caveolin-mediated endocytosis and macropinocytosis were not observed to be differentially regulated based on the user-defined criteria. Standard housekeeping genes were also found to exhibit similar expression profiles upon CHIKV infection as mock-infected samples. A brief description of the reported mammalian-based functional roles and the fold changes upon various time points of CHIKV infection for each of the genes is shown in Table 1 and a heat map exhibiting the differential regulation of these genes across all time points of CHIKV infection is shown in Figure 1. These genes, or related genes, have also been mapped onto the clathrin-mediated endocytotic pathway, as shown in Figure S2. Genes known to be associated with clathrin-mediated endocytosis include epsin I (EPN1), epidermal growth factor receptor pathway substrate 15 (EPS15) and Huntingtin interacting protein I (HIP1). EPN1 and EPS15 were found to be upregulated while HIP1 was downregulated upon CHIKV infection. In addition, genes that targeted kinases (MAP2K7, MAP4K4 and MAPK14) were downregulated in the first 15 min of CHIKV infection, although MAP2K7 and MAP4K4 were subsequently found to be upregulated after 30 min and 120 min of infection. Genes involved in vesicle and endosomal transport, such as ATP6V1B2, ATP6V1F, ARFRP1 and RAB34 were also found to be differentially regulated during CHIKV infection. Taken together, analysis of the microarray data suggests the possible involvement of clathrin-mediated endocytosis in the infectious entry of CHIKV.
Based on the microarray findings, we proceeded to employ a combination of bio-imaging techniques including transmission electron microscopy (TEM) and immunofluorescence assays, to further investigate the infectious entry processes of CHIKV. CHIKV was first prepared by a series of concentration and purification procedures. As revealed by negative staining of the virus preparation, a homogeneous population of CHIKV particles with a uniform size of 60–70 nm in diameter (Figure 2a) was obtained. The purified virus particles were subsequently used to map the infectious entry process of the virus into C6/36 cells. In order to visualize synchronized entry of CHIKV into cells, C6/36 cells were first incubated with CHIKV (MOI = 10) at 4°C for 1 hour. Low-temperature treatment allows binding of CHIKV to the cell surface receptors but prevents the internalization of virus particles into the cells. Subsequently, the cells were warmed to 37°C, and the virus-infected cells were processed for embedding and sectioning at appropriate times after warming for transmission electron microscopy. At 5 mins upon warming to 37°C, CHIKV particles (Figure 2b, arrow) were observed to attach on the outer surface of the plasma membrane of C6/36 cells and CHIKV particles (Figure 2b, arrow) were also noted within invaginations of the plasma membrane. These invaginations resembled those of clathrin-coated pits (Figure 2b, arrowheads). Similarly, attachment and localization of CHIKV particles to clathrin molecules were revealed by double-labeled immunofluorescence staining of the cellular clathrin and CHIKV particles by specific antibodies (Figure 2c and Figure 2d).
After 10 mins at 37°C, most of the virus particles were observed within endocytic vesicles. CHIKV virus particles were contained within each of these vesicles (Figure 3a) as revealed at the ultrastructural level by transmission electron microscopy. These virus-containing vesicles were predominantly localized to the perinuclear region in close association with the endoplasmic reticulum (ER). To further characterize the origin of the cellular endocytic vesicles that were involved in the endocytic trafficking process of CHIKV, double-labeled immunofluorescence microscopy assays were performed. Antibodies specific for early endosomes (EEA1) and late endosomes and lysosomes (Lysotracker) were used. At 10 mins after cells were warmed to 37°C, a double-labeled immunofluorescence assay with anti-CHIKV envelope protein and anti-EEA1 antibodies showed colocalization mainly at the cell periphery region, suggesting that the virus particles were trafficked to the endosomes after endocytosis (Figure 3b). By 15 mins after incubation at 37°C, CHIKV particles were found mainly in vesicles (Figure 3c) that were stained with Lysotracker (Molecular Probes), thus indicating the trafficking of the endocytosed CHIKV particles to the late endosomes and lysosomes by this time point. The fluorescent staining was more intense at the perinuclear region. A unique accumulation of a large number of virus-containing late endosomes and lysosomes were observed at the perinuclear region by 15 mins (Figure 3c), and these structures remained predominant until 35 mins p.i. (data not shown).
The results presented above suggested the involvement of a clathrin-mediated endocytic pathway in CHIKV entry into C6/36 cells. In order to further characterize the pathway by which CHIKV enters C6/36 cells, studies of various drugs inhibiting endocytosis and related processes were performed in a dose-dependent manner. C6/36 cells were pretreated with drugs that selectively inhibit receptor-mediated endocytosis [monodansylcadverine [24]], clathrin-dependent endocytosis [chlorpromazine [23] and dynasore [25]] and caveolae-dependent endocytosis [filipin and nystatin [26]]. Involvement of inhibitors associated with other entry pathways such as macropinocytosis [EIPA [27], [28]] and cholesterol-dependent endocytosis [methyl-β-cyclodextrin [26]] was also evaluated. Furthermore, inhibitors targeting actin polymerization [cytochalasin B [23] and cytochalasin D [31]], microtubule polymerization [colchicine [31] and nocodazole [32]] were used to investigate the role of cytoskeleton during CHIKV entry. Treatment of inhibitors associated with the acidification of endosomes [concanamycin A and bafilomycin A [4], [29], [30] as well as the calcium channel flux [(nifedipine [33]] were also performed (Table S1). Minimal cellular cytotoxicity was observed in drug-treated cells throughout the spectra of concentrations used in these experiments.
Viral entry occurs via several endocytic pathways, with the most common being clathrin- and caveolae-mediated endocytosis [34], [35]. Drug treatment assays were carried out to determine whether CHIKV enters C6/36 cells via receptor-mediated endocytosis, and more specifically clathrin- or caveolae-mediated endocytosis. Upon treatment of monodansylcadverine, a well-known pharmacological drug inhibitor that targets receptor-mediated endocytosis [36], dose-dependent inhibition of CHIKV infection was observed, with a 2-log reduction at 150 µM (Figure 4a). Clathrin-mediated endocytic pathways can also be specifically inhibited by drugs such as chlorpromazine and dynasore. Chlorpromazine is a cationic, amphiphilic molecule that disrupts the assembly of clathrin lattices at the cell surface and endosomes [23], [26], whereas dynasore acts as a potent inhibitor of endocytic pathways by disrupting dynamin, thus preventing clathrin coated vesicles formation, [25]. Data revealed dose-dependent inhibition of CHIKV infection, upon treatment with chlorpromazine (Figure 4b) and dynasore (Figure 4c), showing 2-log reductions at 70 µM and 10 µM respectively. This suggests that CHIKV entry into C6/36 cells occurs via clathrin-mediated endocytosis. To eliminate the involvement of other entry pathways during CHIKV infection, drugs known to inhibit caveolae-mediated endocytosis and macropinocytosis were also evaluated. Caveolae-mediated drug inhibitors, filipin and nystatin inhibit virus entry by disrupting the caveolae, thus preventing caveolae formation [26]. Treatment with filipin (Figure. 4d) and nystatin (Figure 4e) did not exhibit inhibitory effects on CHIKV infection at any of the drug concentrations used. These results suggest minimal involvement of caveolae-mediated endocytosis upon CHIKV infection in C6/36 cells.
Early studies on alphaviruses have shown that lipid rafts are crucial players during virus entry, as cholesterol is needed to allow fusion of viruses with the endosomal membrane of host cells [37]. To evaluate the role of membranous cholesterol, treatment with methyl-β-cyclodextrin, a drug inhibitor targeting lipid raft synthesis via the removal of cholesterol by disrupting detergent-insoluble membrane micro-domains (DIMs) was evaluated in CHIKV infection [38], [39]. Results displayed dose-dependent inhibition, showing 2-log reductions at 2.5 mM (Figure 4f) suggesting that CHIKV entry is dependent on lipid raft synthesis targeting on membranous cholesterol. In a previous study, EIPA, an inhibitor of macropinocytosis, successfully inhibited rhinovirus 2 and Coxsackie B3 virus entry into HeLa cells [40]. However, in this study, at low concentrations of 10 and 25 µM, EIPA only displayed minimal inhibitory effects on the entry pathway of CHIKV infection. Instead, CHIKV infection was observed to be enhanced (Figure 4g). Possible reasons could include the activation of reflex mechanisms in cells, thus causing an increase of endocytic uptake through other pathways.
The employment of dominant-negative mutants of Eps15 can be much more specific in targeting the arrestment of clathrin-coated pit formation [41]. GFP-tagged dominant negative mutant of Eps15, (GFP-EΔ95/295), GFP-tagged negative control constructs (GFP-D3Δ2) and internal GFP control were transiently transfected into C6/36 cells [42]. Transfection efficiencies for all constructs were observed to be more than 80% by fluorescence microscopy. Transfected cells were then assayed for their capability to internalize Texas Red- (TR-) conjugated transferrin, a specific marker for clathrin-dependent endocytosis. Indeed, at 48 hours post-transfection, maximal expression of the transfected gene can be observed and the internalization of TR-transferrin was impaired in cells transfected with GFP-EΔ95/295. In contrast, the uptake of TR-transferrin was not affected in cells expressing GFP-D3Δ2 or GFP (data not shown). The dominant negative mutant GFP-EΔ95/295 drastically inhibited CHIKV infection by more than 80% but neither of the control constructs exerted any inhibitory effects on CHIKV infection in C6/36 cells (Figure 5).
Most enveloped viruses require low-endosomal pH to enter host cells via endocytosis, which is maintained by vacuolar proton-ATPases, to trigger fusion of the viral envelope with the endosomal membrane and release the nucleocapsid into the cytosol [31], [42], [43]. Drug treatment assays were performed to examine the low pH-dependence of CHIKV entry using the vacuolar proton-ATPase inhibitors, namely bafilomycin A1 - which inhibits endosomal and lysosomal acidification [29], [30] - and concanamycin A - which inhibits acidification of organelles [44] As shown in Figure 6, bafilomycin A1 and concanamycin A displayed dose-dependent inhibitory levels with at least 2-log reductions at 3 µM (Figure 6a) and 60 nM (Figure 6b) respectively. These results strongly suggest that CHIKV entry process is dependent on low endosomal pH.
In addition, recent studies reported that more sensitive inhibition of E1-226V mutated CHIKV LR-OPY1 strain upon endosomal pH acidification with bafilomycin A1 and chloroquine on Ae. albopictus cells were observed as opposed to CHIKV 37997 African reference strain [45]. Therefore, in our studies, C6/36 cells treated with concanamycin A were tested against local isolates of CHIKV, namely the SGEHIDSD67Y2008 strain, which is similar to the prototypic CHIKV 37997 African reference strain, and the SGEHICHD122508 strain, which closely resembles the E1-226V mutated CHIKV LR-OPY1 strain. Results displayed complete inhibition at 150 nM for the CHIKV SGEHICHD122508 strain (Figure 6c) when compared to the CHIKV SGEHIDSD67Y2008 strain (Figure 6d). These findings matched those observed by Gay et al. (2012), in which mutations in CHIKV strains result in more sensitive inhibitory levels upon endosomal pH acidification.
Involvement of the cellular cytoskeletal network on CHIKV entry was also investigated via treatment with cytoskeleton-disrupting drugs. Actin filaments have been shown to assist the initial uptake of ligands via clathrin-coated pits and the subsequent degradative pathway, whereas microtubules are known to be involved in maintaining endosomal traffic between peripheral early and late endosomes. Cytochalasin B and D are actin-disrupting drugs, which specifically target the actin cytoskeleton by preventing its polymerization into microfilaments and promoting microfilament disassembly [44]. Pretreatment of cells with cytochalasin B and D (Figures 7a and 7b respectively) failed to inhibit CHIKV infection. Similarly, treatment with nocodazole (Figure 7c) and colchicine (Figure 7d), inhibitors resulting in depolymerization of microtubules, showed no inhibition of CHIKV infection, thus indicating that CHIKV entry does not rely on microtubule polymerization [31]. These results suggest minimal involvement of the cytoskeletons in the entry process of CHIKV infection.
Previous studies on herpes simplex viruses have identified the importance of calcium (Ca2+) flux in virus entry for delivering virus capsids to the cytoplasm or nucleus [33]. Therefore, to determine whether Ca2+ flux is important in CHIKV infection, nifedipine, an inhibitor of dihydropyridine L-type voltage sensitive Ca2+ channel flux, was used. However, in this study, nifedipine treatment (Figure 7e) failed to inhibit CHIKV infection, thus indicating that Ca2+ flux is not required for CHIKV infection. From these drug treatment assays, it can thus be concluded that CHIKV entry into C6/36 cells occurs via clathrin-mediated endocytosis. Low endosomal pH is found to play a significant role in CHIKV entry, while the cytoskeleton and Ca2+ flux may not be vital for the endocytic process of CHIKV infection.
Data from the microarray analyses has revealed the differentiated regulation of genes associated with the clathrin-mediated endocytic pathway, and drug treatment assays have validated the involvement of the pathway in the infectious entry of CHIKV into mosquito cells. To investigate the functional roles of genes related to clathrin-mediated endocytosis, siRNAs targeting clathrin-heavy chain (CLTC), Rab proteins (RAB5 and RAB7) and vacuolar ATPases (vacuolar ATPase B) were utilized in further downstream studies. Dose-dependent siRNA-based knockdown of the selected targeted cellular genes was performed in varying siRNA concentrations (0.1, 1, 5, 10 nM) on C6/36 cells, prior to being subjected to CHIKV infection. Scrambled siRNAs were included as controls to ensure the specificity of the siRNAs used in this study. Minimal cellular cytotoxicity was observed in siRNA-treated cells throughout the spectra of concentrations used in these experiments (data not shown). RNA expression levels of the knocked-down genes were analyzed, with the non-infected samples being harvested at 48 hours post transfection. Significant reduction was observed in the levels of gene expression of CLTC, RAB5, RAB7 and vacuolar ATPase B relative to non-transfected cells (TC) (Figure S2a–S2d, solid bars). In contrast, data for scrambled siRNA gene expression showed similar levels of gene expression to TC samples (Figure S2a–S2d, striped bars). These results suggested that the siRNA knockdown of the targeted cellular genes is specific.
Effects of the scrambled siRNAs showed minimal inhibition of CHIKV infection relative to CHIKV-infected non-transfected cells (PTC) (Figure 8a–8d, striped bars). However, cells with specific siRNA knockdown of CLTC gene showed dose-dependent reduction in the infectious viral titre of CHIKV, with a 1-log reduction at 5 nM, relative to the PTC samples (Figure 8a, solid bars). siRNAs targeting the endosomal trafficking pathway (RAB5 and RAB7), which are involved in viral entry via the trafficking of the early and late endosomes, prevented CHIKV infection in a dose-dependent manner, showing a 3-log reduction in infectious virus titre at 5 nM RAB5 siRNA (Figure 8b, solid bars). A 1-log reduction in CHIKV titre at 1 nM RAB7 siRNA (Figure 8c, solid bars) further accounts for the trafficking of internalized CHIKV particles from early endosomes to the late endosomes. In addition, silencing of vacuolar ATPase B, involved in endosomal acidification, also led to a decrease in CHIKV infection in a dose-dependent manner, with a 2-log reduction at 5nM (Figure 8d, solid bars). These results further confirmed our earlier findings that CHIKV entry into Ae. albopictus (C6/36) cells occurs via clathrin-mediated endocytosis and is dependent on low pH endosomal acidification.
Interest on deciphering virus entry into host cells has been steadily gaining momentum over recent years, in the hope to establish potentially powerful anti-viral strategies against these medically important human pathogens. Studies have shown that numerous viruses enter via receptor-mediated and/or clathrin-mediated pathways [34], [35]. The entry process of many enveloped viruses typically begins with the fusion of viral envelope glycoproteins at the plasma membrane allowing internalization of viral nucleocapsids at neutral pH. Virus entry can also occur via endocytosis prior to fusion with the endocytic membrane, whereby hydrophobic virus fusion proteins undergo conformational changes upon exposure to acidic pH resulting in the release of viral nucleocapsids into the cytoplasm. Receptor-mediated endocytosis forms the predominant mode of entry, often mediated by the formation of clathrin-coated pits, prior to subsequent transport of viruses to early endosomes, where the low pH environment triggers fusion [46]. Meanwhile, clathrin-mediated endocytosis primarily entails the binding of extracellular cargo molecules to specific cell-surface receptors. These receptors, along with other membrane proteins entering via endocytosis, are transported by the intracellular adaptor proteins to endocytic sites. Together with clathrin, the adaptor protein forms an enclosed coat at the plasma membrane. The coated membrane then bends to form invaginations resembling clathrin-coated pits that pinch off to form cargo-filled vesicles [47].
Nevertheless, analyses of these entry modes have been predominantly demonstrated in mammalian cells. Indeed, the involvement of endocytic pathways in the entry of alphaviruses has been extensively studies, with SFV and SINV found to penetrate target cells through clathrin-dependent endocytosis [3], [15], [48], [49]. Few studies have however been documented on endocytic entry pathways of arboviruses into mosquito cells. This study shows, for the first time, CHIKV infectious entry into Ae. mosquitoes cells via clathrin-mediated endocytosis. Although a recent study has shown CHIKV entry in mammalian cells via clathrin-independent endocytosis [3], [4], [7], earlier findings indicated the dependence of CHIKV infectious entry in mammalian cells on clathrin [3], [4], [7]. This work thus indicates that the infection mechanism in mosquitoes and mammals may have indeed occurred through a common conserved endocytic pathway.
A variety of experimental approaches was used in this novel study including microarray gene profiling, bioimaging studies (transmission electron microscopy, double-labeled immunofluorescence microscopy), pharmacological inhibitors, overexpression of dominant-negative mutant of Eps15 and siRNA-based knockdown of genes involved in the endocytic pathway.
A customized gene expression microarray was first conducted to identify host genes necessary for the infectious entry of CHIKV into mosquito cells. Several genes that were differentially regulated during CHIKV infection have been known to be involved in clathrin-mediated endocytosis (Table 1), including EPN1, EPS15 and HIP1. EPN1 is an accessory protein that interacts with EPS15 - a clathrin-coat-associated protein that binds the α-adaptin subunit of the clathrin adaptor AP2 (AP2A1) [50] - and clathrin, as well as with other accessory proteins for the endocytosis of clathrin-coated vesicles. It facilitates the rearrangement of the clathrin lattice, resulting in the formation of clathrin-coated invaginations and fission [51]. HIP1 plays a role in clathrin-mediated endocytosis and trafficking by regulating clathrin assembly via binding to a highly conserved region of clathrin light chain [52]. The microarray analysis also revealed the involvement of kinase-targeting genes (MAP2K7, MAP4K4 and MAPK14) - associated with the signal transduction processes of viral entry [53] – during early CHIKV infection. In addition, ATP6V1B2 and ATP6V1F, components of V-ATPases, were also found to be differentially expressed during the initial phases of CHIKV infection. This suggests a significant role for V-ATPases, which have been identified in intracellular compartments such as clathrin-coated vesicles and endosomes and are therefore essential in clathrin-mediated endocytosis [54]. The upregulation of ARFRP1 suggests the importance of vesicle and endosomal transport in early CHIKV infection, while the downregulation of RAB34 - which is a member of the Rab family small GTP-ases that regulates vesicle budding, docking and fusion, and has been predominantly associated with membrane ruffles and macropinosomes and promotes macropinosome formation [55] – eliminates the possible engagement of micropinocytosis for CHIKV infectious entry into mosquito cells. Taken together, analysis of the microarray data suggests that CHIKV entry occurs via clathrin-mediated endocytosis.
Downstream assays were subsequently performed in order to validate the microarray findings. Transmission electron microscopy analyses showed the presence of CHIKV particles within invaginations of the plasma membrane, resembling those of clathrin-coated pits. Furthermore, characterization of the vesicles involved in the endocytic trafficking processes of CHIKV revealed the translocation of the virus particles to early endosomes and subsequently to late endosomes and lysosomes. To this end, double-labeled immunofluorescence assays were performed with the early endosomal marker, EEA1 and late endosomal and lysosomal marker, Lysotracker. Colocalization of virus particles were observed upon double-labeling with anti-CHIKV envelope protein and anti-EEA1 antibodies, thus indicating the trafficking of CHIKV particles to endosomes upon entry into mosquito cells. These endosomes were also observed to be closer to the cell periphery. Subsequent labeling with Lysotracker showed that endocytosed CHIKV particles were trafficked from early to late endosomes and lysosomes (Figure 3c).
Further analyses of CHIKV internalization into C6/36 cells was determined by treating cells with a set of pharmacological inhibitors targeting receptor-, clathrin-, caveolae- mediated endocytosis, cholesterol-dependent endocytosis and macropinocytosis. Significant results from treatment with monodansylcadverine, chlorpromazine and dynasore proved the involvement of receptor- and/or clathrin- mediated endocytosis (Figure 4a–c).
The importance of lipid rafts has been widely acknowledged, with studies showing that DIMS, found in the plasma membrane of cell surface, posses the ability to isolate cholesterol into the hydrophobic pocket, thus aiding in entry of viruses [38], including Simian virus 40 (SV40) [56]. Moreover, studies in RNA viruses, such as HIV-1, have determined virus entry into host cells via lipid rafts, and treatment with methyl-β cyclodextrin resulted in blockade of trans-epithelial transcytosis of HIV-1 and reduction of envelope fusion [57]–[60]. Similarly, we reported in this study that methyl-β cyclodextrin treatment showed inhibition of CHIKV entry via C6/36 cells, thus suggesting that the infectious entry process of CHIKV is dependent on lipid raft synthesis targeting membranous cholesterol. In contrast, treatment with inhibitors such as flilipin, nystatin and EIPA, had minimal effects on inhibiting CHIKV infection, thus eliminating the possibility of CHIKV entry via other pathways.
Earlier studies have shown that the mutant form of Eps15, EΔ95/295 which contains a 200-amino acid deletion, prevented the association with AP2, thus inhibiting the entry of VEEV via clathrin-mediated endocytic pathway [30]. We observed similar observations in this study, with the overexpression of EΔ95/295 found to reduce the infectious entry of CHIKV. It can therefore be concluded that CHIKV entry into C6/36 cells occurs via clathrin-mediated endocytosis.
Earlier studies have shown that E1 constitutes the fusion protein of the alphaviruses [61]–[63]. In the endosomal vesicles containing endocytosed CHIKV particles, the E1–E2 heterodimer undergoes a conformational change upon exposure to low pH. This causes rearrangement to a homotrimeric complex of E1 formation, leading to increased activity for membrane fusion [64], [65]. Membrane fusion processes occur rapidly via the insertion of hydrophobic fusion peptides to form pores in cellular and viral membranes [66], thus releasing the nucleocapsid into the cytoplasm of the cell even before the degradation of the lysosomes [67]. Requiring low pH endosomal exposure, alphaviruses exposed to lysotromphobic weak bases such as bafilomycin A1, chloroquine and concanamycin A, are unable to undergo membrane fusion due to neutralization of pH in the endosomes [67], [68]. For instance, infection of SFV on Ae. albopictus cells was inhibited upon treatment with inhibitors targeting low-endosomal acidification [66].
A recent study on E1-A226V mutated CHIKV LR-OPY1 strain showed that it is more sensitive to inhibition via endosomal pH acidification with bafilomycin A1 and chloroquine on Ae. albopictus cells as opposed to the CHIKV 37997 African reference strain [45]. These two strains possess 85% nucleotide sequence identity, differing only in the E1 protein at position 226 [9]. Furthermore, CHIKV infection of C6/36 cells was found to be sensitive to inhibitors of the v-ATPase and chloroquine, a weak base that accumulates in the acidic parts of the cell and inhibits the acidification of endocytic compartments [45]. Similarly, in our own study, we observed lower levels of inhibition for the CHIKV SGEHIDSD67Y2008 strain - which has features common to those of the CHIKV 37997 African reference strain - than those of the CHIKV SGEHICHD122508 strain, which resembles the E1-A226V mutated CHIKV LR-OPY1 strain. The results revealed that while both the CHIKV SGEHICHD122508 and SGEHIDSD67Y2008 strains require endosomal acidification for optimal infection of Ae. albopictus cells, the former is more sensitive to inhibition as compared to the latter. This could be due to the differential sensitivities of the CHIKV strains to lysomotropic agents and weak bases, as similarly reported in previous studies [3], [4].
Our findings in this study were also further evaluated via siRNA-based dosage dependence analyses of several cellular genes associated with clathrin-mediated endocytosis and endosomal acidification. siRNA targeted against CLTC showed significant inhibition of CHIKV infection, thus further strengthening our earlier findings, in which treatment with clathrin-mediated endocytic associated inhibitors showed similar dose-dependent inhibitory trends of CHIKV infection. Similarly, silencing of vacuolar ATPase B also led to a decrease in CHIKV infection, strongly demonstrating that CHIKV entry requires low endosomal pH. Previous studies have shown that RAB5 and RAB7 proteins are usually associated with the translocation of viruses from the early to late endosomes [69]. In particular, mammalian cells infected with SFV were found to require the integrity of RAB5 proteins for productive infection [4], while RAB5 and RAB7 proteins were identified to play significant roles in the productive infection of vesicular stomatitis Indiana virus (VSV) and SFV in mosquito cells [69], [70]. Similarly, we observed significant inhibition of CHIKV infection upon siRNA-based knockdown of these genes, thus suggesting that CHIKV entry involves the translocation from early endosomes after clathrin-mediated endocytosis to late endosomes.
Based on our unprecedented findings in this novel study, it can thus be concluded that CHIKV infectious entry into Ae. albopictus cells occurs via clathrin-mediated endocytosis and is dependent on low endosomal pH acidification and the presence of membranous cholesterol. Elucidation of the infectious entry of CHIKV into mosquito C6/36 cells will contribute towards better understanding of CHIKV pathogenesis, thus enabling future development of antiviral strategies against the infectious entry process of CHIKV.
|
10.1371/journal.pmed.1002706 | Raltegravir-intensified initial antiretroviral therapy in advanced HIV disease in Africa: A randomised controlled trial | In sub-Saharan Africa, individuals infected with HIV who are severely immunocompromised have high mortality (about 10%) shortly after starting antiretroviral therapy (ART). This group also has the greatest risk of morbidity and mortality associated with immune reconstitution inflammatory syndrome (IRIS), a paradoxical response to successful ART. Integrase inhibitors lead to significantly more rapid declines in HIV viral load (VL) than all other ART classes. We hypothesised that intensifying standard triple-drug ART with the integrase inhibitor, raltegravir, would reduce HIV VL faster and hence reduce early mortality, although this strategy could also risk more IRIS events.
In a 2×2×2 factorial open-label parallel-group trial, treatment-naive adults, adolescents, and children >5 years old infected with HIV, with cluster of differentiation 4 (CD4) <100 cells/mm3, from eight urban/peri-urban HIV clinics at regional hospitals in Kenya, Malawi, Uganda, and Zimbabwe were randomised 1:1 to initiate standard triple-drug ART, with or without 12-week raltegravir intensification, and followed for 48 weeks. The primary outcome was 24-week mortality, analysed by intention to treat. Of 2,356 individuals screened for eligibility, 1,805 were randomised between 18 June 2013 and 10 April 2015. Of the 1,805 participants, 961 (53.2%) were male, 72 (4.0%) were children/adolescents, median age was 36 years, CD4 count was 37 cells/mm3, and plasma viraemia was 249,770 copies/mL. Fifty-six participants (3.1%) were lost to follow-up at 48 weeks. By 24 weeks, 97/902 (10.9%) raltegravir-intensified ART versus 91/903 (10.2%) standard ART participants had died (adjusted hazard ratio [aHR] = 1.10 [95% CI 0.82–1.46], p = 0.53), with no evidence of interaction with other randomisations (pheterogeneity > 0.7) and despite significantly greater VL suppression with raltegravir-intensified ART at 4 weeks (343/836 [41.0%] versus 113/841 [13.4%] with standard ART, p < 0.001) and 12 weeks (567/789 [71.9%] versus 415/803 [51.7%] with standard ART, p < 0.001). Through 48 weeks, there was no evidence of differences in mortality (aHR = 0.98 [95% CI 0.76–1.28], p = 0.91); in serious (aHR = 0.99 [0.81–1.21], p = 0.88), grade-4 (aHR = 0.88 [0.71–1.09], p = 0.29), or ART-modifying (aHR = 0.90 [0.63–1.27], p = 0.54) adverse events (the latter occurring in 59 [6.5%] participants with raltegravir-intensified ART versus 66 [7.3%] with standard ART); in events judged compatible with IRIS (occurring in 89 [9.9%] participants with raltegravir-intensified ART versus 86 [9.5%] with standard ART, p = 0.79) or in hospitalisations (aHR = 0.94 [95% CI 0.76–1.17], p = 0.59). At 12 weeks, one and two raltegravir-intensified participants had predicted intermediate-level and high-level raltegravir resistance, respectively. At 48 weeks, the nucleoside reverse transcriptase inhibitor (NRTI) mutation K219E/Q (p = 0.004) and the non-nucleoside reverse transcriptase inhibitor (NNRTI) mutations K101E/P (p = 0.03) and P225H (p = 0.007) were less common in virus from participants with raltegravir-intensified ART, with weak evidence of less intermediate- or high-level resistance to tenofovir (p = 0.06), abacavir (p = 0.08), and rilpivirine (p = 0.07). Limitations of the study include limited clinical, radiological, and/or microbiological information for some participants, reflecting available services at the centres, and lack of baseline genotypes.
Although 12 weeks of raltegravir intensification was well tolerated and reduced HIV viraemia significantly faster than standard triple-drug ART during the time of greatest risk for early death, this strategy did not reduce mortality or clinical events in this group and is not warranted. There was no excess of IRIS-compatible events, suggesting that integrase inhibitors can be used safely as part of standard triple-drug first-line therapy in severely immunocompromised individuals.
ClinicalTrials.gov NCT01825031.
International Standard Randomised Controlled Trials Number ISRCTN 43622374.
| Individuals in Africa who are HIV positive and initiating treatment with severe immunosuppression (CD4 < 100 cells/mm3) have high risks of dying shortly after starting treatment.
It would be expected that adding an integrase inhibitor (raltegravir) to standard triple-drug antiretroviral therapy for 12 weeks would reduce HIV viral load faster: we wanted to find out whether this would reduce high early mortality.
We randomly allocated 1,805 adults, adolescents, and older children to receive standard antiretroviral therapy with or without 12 weeks of adjunctive raltegravir.
We found that the group receiving 12 weeks of adjunctive raltegravir had significantly faster declines in HIV viral load.
However, we found no significant difference in deaths within 24 weeks of starting treatment between those receiving adjunctive raltegravir (97/902 [10.9%]) or not (91/903 [10.2%]); nor were there differences in clinical disease progression, immune reconstitution, inflammatory syndrome, or adverse events.
In this study, we found that intensifying the potency of initial antiretroviral therapy did not reduce early mortality on treatment, providing no support for its widespread use in individuals initiating treatment with severe immunosuppression.
However, it also did not increase the rates of clinically important immune reconstitution inflammatory syndrome, suggesting that integrase inhibitors could replace other components of standard first-line antiretroviral therapy safely.
| Despite World Health Organisation (WHO) guidelines recommending universal antiretroviral therapy (ART) regardless of cluster of differentiation 4 (CD4) cell count [1], 20%–25% of individuals infected with HIV in sub-Saharan Africa continue to present for care with severe immunosuppression (CD4 count < 100 cells/mm3) [2]. Of these, about 10% die in the first 3 months after ART initiation [3–6]. Causes of early death are multifactorial and are similar for adults and older children [5]. Possible strategies to reduce this excess mortality could aim to accelerate immune restoration by controlling HIV viraemia more rapidly; to control or prevent overt clinical infections; or to provide nutritional support to enhance immune response to infection and reverse metabolic deficiencies.
Standard WHO-recommended ART [1] consists of two nucleos(t)ide reverse transcriptase inhibitors (N(t)RTI), with a non-nucleoside reverse transcriptase inhibitor (NNRTI). Another key drug class is integrase strand-transfer inhibitors (INSTIs), which result in faster decline in HIV viral load (VL) over the first 3 months of combination therapy [7,8]. Several studies have investigated how this might impact chronic inflammation and immune restoration; in some, INSTIs appeared to improve markers of inflammation (including microbial translocation) and/or T-cell activation [9–14], but in others there were no clear differences in inflammatory and other markers (i.e., D-dimer, interleukin 6 [IL-6], and T-cell activation) [15]. The clinical importance of intensified or quadruple ART in patients with VL >100,000 copies/mL or advanced disease remains unclear.
To our knowledge, no randomised trial has been powered to test the hypothesis that the INSTI-associated rapid reduction in HIV viraemia translates into clinical benefit, mortality reductions in particular, or, conversely, is associated with harm through increased risk of immune reconstitution inflammatory syndrome (IRIS). We therefore conducted the Reduction of EArly mortaLITY (REALITY) randomised clinical-endpoint trial (NCT01825031; ISRCTN43622374) to compare three interventions to reduce early mortality in adults and older children initiating ART with CD4 <100 cells/mm3 in four sub-Saharan African countries: adjunctive intensification with raltegravir [16] (the first licenced INSTI), enhanced anti-infective prophylaxis, and food supplementation. Here, we report on the impact of adjunctive raltegravir intensification.
Adults, adolescents, and children aged ≥5 years infected with HIV, diagnosed through national screening programmes, not on ART and reporting no previous ART, and with CD4 <100 cells/mm3 were approached consecutively for screening from inpatient and outpatient facilities at clinics at eight urban/peri-urban regional hospitals in Zimbabwe, Uganda, Malawi, and Kenya. When CD4 counts were not routinely performed at diagnosis, those with new HIV diagnoses were approached consecutively for CD4 testing and study screening. Participants were enrolled if they had screening CD4 count <100 cells/mm3, were ART-naive, were not pregnant/breastfeeding, had not used single-dose nevirapine to prevent mother-to-child transmission, had no contraindications to trial drugs, and provided written informed consent. The trial was approved by Ethics Committees in Zimbabwe, Uganda, Malawi, Kenya, and the United Kingdom. The protocol and CONSORT checklist are provided as S1 Protocol and S1 CONSORT checklist.
Participants were randomised 1:1 to initiate open-label ART with 2NRTI+NNRTI either alone (standard ART) or with 12 weeks’ raltegravir (raltegravir intensification) using standard doses (see S1 Text). Standard ART was tenofovir-disoproxil-fumarate+emtricitabine or zidovudine+lamivudine in adults and adolescents and abacavir+lamivudine or zidovudine+lamivudine in children aged 5–<13 years, with nevirapine or efavirenz, according to physician choice and local standard of care. Participants were also factorially randomised 1:1 to 12 weeks enhanced anti-infection prophylaxis versus standard-of-care co-trimoxazole prophylaxis [17], and 1:1 to 12 weeks ready-to-use supplementary food (RUSF) versus no routine supplementation [18] (see S1 Text for details).
Randomisation was stratified by centre, age (</≥13 years), and the other factorial randomisations. A computer-generated sequential randomisation list using variably sized permuted blocks was prepared by the trial statistician and incorporated securely into the online trial database. The list was concealed until allocation after eligibility was confirmed by local centre staff, who then performed the randomisation.
Nurse visits at weeks 2, 4, 8, 12, 18, 24, 36, and 48 included symptom checklist, self-reported adherence assessment (with standard adherence support), medication dispensing, and body composition assessment using bioimpedance (TANITA BC-420MA). At weeks 4, 12, 24, 36, and 48, history and examination were undertaken by a physician; haematology, biochemistry, and CD4/CD8 assays were performed at local laboratories; and plasma was stored for retrospective VL and genotyping (results not available in real time). Nurses/physicians were unblinded. Laboratory tests were assayed blind to randomisation. Toxicity substitution and/or second-line switch were at physician discretion, following WHO guidelines [19]. Participants exited the trial after 48 weeks. Consent was obtained to ascertain vital status in all participants at trial closure (May 2016, i.e., beyond 48 weeks from randomisation) from the ART programmes where they had transferred to care at trial exit.
The primary outcome was all-cause mortality to 24 weeks. Secondary outcomes to 48 weeks were all-cause mortality; serious adverse events (SAEs) following International Committee on Harmonization definitions, grade-4 adverse events (AEs), and AEs leading to modification of ART/trial drugs; specific mechanisms of each intervention (CD4 count change; incidence of tuberculosis, cryptococcosis, candidiasis, severe bacterial infections); changes in weight/BMI; hospitalisations; and self-reported ART adherence/acceptability. Other prespecified outcomes included VL suppression (originally planned to be assayed only in a random subset; ultimately assayed in all participants), genotypic resistance (assayed in samples from Kenya, Uganda, and Zimbabwe, see S1 Text), and body composition. All AEs were graded following established guidelines [20,21]. HIV integrase genotypes were assayed for raltegravir-intensified participants with VL >1,000 c/mL at 12 weeks, and reverse transcriptase genotypes were assayed for all participants with VL >1,000 c/mL at 48 weeks. An Endpoint Review Committee (ERC) (majority independent members), blind to trial drugs received, adjudicated secondary clinical endpoints and trial-drug relatedness against protocol-defined criteria and assessed compatibility of clinical events with IRIS. Clinical endpoints were predominantly clinically defined, as microbiological and other diagnostic facilities were limited in most centres. Because of this known limitation, IRIS was not a secondary outcome and was considered an exploratory analysis. However, IRIS was part of the standardised ERC assessment and ERC case record form from the start of the trial and was prospectively ascertained on all reviewed events. Events were considered IRIS compatible if there was atypical or exaggerated presentation of an opportunistic infection or tumour soon after ART initiation. VLs were only measured retrospectively on stored plasma and so were not available contemporaneously for assessment of IRIS compatibility; the earliest post-randomisation CD4 count was done at 4 weeks and so was not available for events before this. As access to diagnostic testing was limited, previously published definitions [22,23] were used but in modified form; the ERC adjudication relied heavily on the description of the clinical presentation and any radiology/microbiology/histology provided by site, and the timing of the presentation and its evolution in relation to ART initiation.
A total of 1,800 adults/children provided >80% power to detect 50% relative reductions in 24-week all-cause mortality from 7% to 3.5% (two-sided alpha = 0.05), allowing 5% lost to follow-up. Interim data were reviewed by an independent Data Monitoring Committee (three annual meetings) using the Haybittle-Peto criterion (p < 0.001). Randomised groups were compared following the intention to treat principle using log-rank tests for time-to-event outcomes (censoring those lost to follow-up), Fisher exact tests for binary outcomes, and generalized estimating equations (GEEs) with independent working correlation for global tests of repeated measures (logit/normal distributions for binary/continuous outcomes, respectively). Primary analyses were stratified by randomisation stratification factors (details in S1 Text). Analyses used Stata v15.1.
Between 18 June 2013 and 10 April 2015, 1,805 participants were randomised to standard ART (n = 903) or raltegravir-intensified ART (n = 902) (Fig 1).
Baseline characteristics were well balanced between randomised groups (Table 1). Median age was 36 years; 72 (4.0%) participants were aged 5–17 years. Median CD4 count was 37 cells/mm3 and VL was 249,770 copies/mL; 1,334/1,804 (73.9%) had VL ≥ 100,000 copies/mL. Despite this, 854 (47.3%) participants were WHO stage 1/2. A total of 56 (3.1%) participants were lost to follow-up (no clinic attendance for >91 days). Before death/loss to follow-up, 12,664/12,944 (97.8%) scheduled visits were completed.
A total of 901 (99.9%) participants randomised to raltegravir-intensified ART were prescribed raltegravir. Overall, in the first 12 weeks on ART the raltegravir-intensified ART group spent 98.2% of person-time prescribed raltegravir versus 0.0% in the standard ART group, compared with 0.6% versus 0.1%, respectively, from 12 to 48 weeks (S1 Fig). Self-reported ART adherence was high but was significantly poorer in the raltegravir-intensified ART group during the first 12 weeks only (p < 0.001) (overall p = 0.08, S2A Fig), particularly in those taking other ART once daily (S2B Fig). At last follow-up, 885 (98.0%) standard ART versus 880 (97.6%) raltegravir-intensified ART participants remained on first-line ART (exact p = 0.53), with 67 (7.4%) standard ART versus 52 (5.8%) raltegravir-intensified ART participants having made within-class substitutions (exact p = 0.18).
Despite slightly poorer self-reported adherence, early VL suppression was significantly and substantially faster with raltegravir intensification (Fig 2A), with 41.0% (343/836), 71.9% (567/789), 76.7% (594/774), and 80.7% (661/757) participants, respectively, recording <50 copies/mL 4, 12, 24, and 48 weeks after randomisation versus 13.4% (113/841), 51.7% (415/803), 74.7% (586/784), and 79.2% (591/746) in the standard ART group, respectively (p < 0.001, p < 0.001, p = 0.36, and p = 0.47, respectively). Overall, therefore, 59.4% versus 43.6% of time at risk through 24 weeks was spent with VL <50 copies/mL in raltegravir-intensified versus standard ART, respectively. At higher thresholds, differences at week 4 were even greater and attenuated more rapidly but persisted through 24 weeks (S3 Fig). At 48 weeks, there was no evidence of differential suppression at 50–5,000 copies/mL (p > 0.4). Mean change in log10 VL to week 4 was −3.4 (standard error [SE] 0.03) versus −2.7 (SE 0.03) in raltegravir-intensified ART versus standard ART (p < 0.001), with geometric mean VL 80 c/mL and 480 c/mL, respectively.
By 24 weeks (primary endpoint), 97 (10.9%) raltegravir-intensified ART versus 91 (10.2%) standard ART participants died (adjusted hazard ratio [aHR] = 1.10 [95% CI 0.82–1.46], log-rank p = 0.53, Fig 3A), with no evidence of interaction with other randomisations (pheterogeneity > 0.7). Absolute 24-week mortality difference was +0.6% (−2.2%, +3.5%). By 48 weeks, 110 (12.4%) versus 115 (13.0%) participants, respectively, had died (aHR = 0.98 [0.76–1.28], log-rank p = 0.91; absolute difference −0.6% [−3.7%, +2.5%]) with no evidence of differences by cause (Table 2). There was no difference in longer-term (after week 48) mortality (aHR = 0.95 [0.74–1.21], log-rank p = 0.69, S4 Fig). Estimated mortality rates decreased sharply from day 19 through week 12 (Fig 3B), with 71 (31.6%) of the 225 deaths occurring by 4 weeks, and 147 (65.3%) by 12 weeks. There was no evidence of early differences in mortality rates (S5 Fig) and no evidence of heterogeneity in the impact of raltegravir intensification over time on ART (pheterogeneity = 0.14, comparing 0–24 versus 24–48 versus 48+ weeks on ART). Of nine preplanned and five exploratory subgroup analyses (S6 Fig; details in S1 Text), only one had weak evidence for variation across dual NRTIs (pheterogeneity = 0.04, suggesting harm from raltegravir intensification in combination with tenofovir-emtricitabine; pheterogeneity = 0.06 comparing tenofovir-emtricitabine versus zidovudine-lamivudine [excluding abacavir-lamivudine]). Specifically, there was no evidence that mortality differences depended on pre-ART VL or pre-ART CD4 (pheterogeneity = 0.84/0.49 and 0.22/0.27, respectively, using categorical/continuous interactions). No subgroup analyses suggested heterogeneity in the impact of raltegravir intensification on suppression <50 copies/mL at week 4 (pheterogeneity > 0.05) (S7 Fig).
There was no evidence of an impact of raltegravir intensification on any disease progression clinical outcome (Table 2).
Integrase genotypes were obtained in 33 raltegravir-intensified ART participants with VL >1,000 (median 47,724) copies/mL at 12 weeks (see S2 Text). One participant had predicted intermediate-level (mutations T97A+R263K) and two predicted high-level (N155H, T97A+Y143R/C) raltegravir resistance mutations (0.6% of those randomised, accounting for missing genotypes using probability weights). No patient had predicted intermediate- or high-level dolutegravir resistance.
Reverse-transcriptase genotypes were available for 75 raltegravir-intensified ART versus 87 standard ART participants with VL >1,000 (median 89,815) copies/mL at 48 weeks. The NRTI mutation K219E/Q (p = 0.004) and the NNRTI mutations K101E/P (p = 0.03) and P225H (p = 0.007) were less common in raltegravir-intensified ART, with no evidence of differences for other mutations (p > 0.1, S8 Fig). There was marginal evidence suggesting less intermediate- or high-level resistance with raltegravir-intensified ART to tenofovir (24.0% [18/75] raltegravir-intensified ART versus 37.9% [33/87] standard ART, p = 0.06), abacavir (40.0% [30/75] raltegravir-intensified ART versus 54.0% [47/87] standard ART, p = 0.08), and rilpivirine (38.7% [29/75] raltegravir-intensified ART versus 52.9% [46/87] standard ART, p = 0.07) (S9 Fig) (see S2 Text for details).
Absolute CD4 count increases were similar in both groups through 24 weeks (p = 0.76; Fig 2B); however, there was marginal evidence of a small difference at 48 weeks (+161 [standard deviation ±4.4] cells/mm3 raltegravir-intensified ART versus +148 [±4.4] cells/mm3 standard ART, adjusted difference +11.4 [95% CI −0.4 to +23.1], p = 0.06). There was similar weak evidence of slightly (2%–5%) greater percentages with CD4 ≥ 200 cells/mm3 at later time points (p = 0.049 at week 24, p = 0.07 at week 48, S10 Fig). Changes in weight were also similar through week 24 (p = 0.52; Fig 2C), with small but significant differences appearing from 36–48 weeks. Similar late differences were observed for BMI in adolescents/adults (data not shown), fat mass (S11A and S11B Fig), and muscle mass (S11C and S11D Fig). There was no evidence of heterogeneity in these differences by age (S12 Fig). Absolute CD8 count increases were similar in both groups through 48 weeks (p = 0.82, S13 Fig).
There was no evidence for differences in time to first SAEs (log-rank p = 0.88; S1 Table), grade-4 AEs (log-rank p = 0.29), grade-3/4 AEs (log-rank p = 0.72), grade-4 AEs definitely/probably/possibly related to ART (log-rank p = 0.52), grade-4 AEs definitely/probably related to ART (log-rank p = 0.07), AEs leading to ART modification (log-rank p = 0.51) or new hospitalisations (log-rank p = 0.59) (Table 2). Raltegravir was modified for AEs in 19 (2.1%) participants (S2 Table), including renal (n = 6) events, hepatic (n = 6) events, hypersensitivity reactions (n = 3; two discontinued with efavirenz, which was subsequently restarted; one discontinued with tenofovir+emtricitabine+efavirenz, subsequently started lopinavir+ritonavir+tenofovir+emtricitabine+raltegravir), and Stevens-Johnson syndrome in one participant (discontinued lamivudine+zidovudine+efavirenz+raltegravir at week 7). Only one participant experienced a grade-4 AE adjudicated as definitely/probably related to raltegravir (Table 2). There was no evidence of any difference in total hospitalisation-days (2,367 raltegravir-intensified ART versus 2,685 standard ART, rank-sum p = 0.56) or total hospitalisations (198 versus 233, respectively, Poisson p = 0.09).
Fatal or nonfatal events judged compatible with IRIS occurred in 89 (9.9%) raltegravir-intensified ART versus 86 (9.5%) standard ART participants (log-rank p = 0.79) (Table 2); 36 (4.0%) versus 31 (3.4%), respectively, experienced fatal IRIS events (p = 0.54). Tuberculosis IRIS events occurred in 53 (5.9%) participants with raltegravir-intensified ART versus 54 (6.0%) with standard ART (exact p = 1.00) and cryptococcal IRIS events in 15 (1.7%) versus 16 (1.8%) participants, respectively (exact p = 1.00) (S3 Table). IRIS events occurred a median 3.4 (IQR 2.0–6.3) weeks from randomisation, with rates declining from the third week on ART (S14 Fig). IRIS was more common in participants initiating ART at older ages (p = 0.005), with lower CD4 counts (p < 0.001) or with pre-existing TB (p = 0.007); IRIS was less common in those initiating ART with enhanced prophylaxis against opportunistic infections (p = 0.001) (S4 Table). There was no evidence that raltegravir intensification was associated with IRIS after adjusting for these factors (p = 0.63).
In this large trial in adults and older children with CD4 <100 cells/mm3 in sub-Saharan Africa, we found that 12-week raltegravir intensification of standard ART was well tolerated and reduced plasma HIV VL substantially faster than standard ART alone, but that this had no discernible clinical benefit and no effect on mortality; however, nor was there any evidence of increased rates of IRIS. This is important given the current move to first-line INSTI-based ART; all INSTIs have similarly rapid VL reductions [24], so this result can likely be extrapolated to other INSTIs.
The underpinning hypothesis was that mortality might be reduced by accelerating plasma HIV RNA decline with INSTIs, because initiating ART reduces mortality compared with rates observed pre-ART [25]. One reason this did not occur could be that the differential VL suppression, although significant and substantial, was still too small to reduce mortality. As VL reductions are similar across INSTIs [24], other INSTIs would likely produce similar results. Other explanations for lack of mortality benefit despite significantly faster VL declines may be because there was a lag in, or no, improvement in early functional CD4 recovery or because the disease may simply have been too severe to be reversed despite rapid VL decline. CD4 and CD8 T cells were the only markers of immune restoration measured in real time in all participants, so we cannot investigate further; but there was no differential effect on CD4 (or CD8) counts, at least in the first 24 weeks. This might be expected, as immune restoration does not only depend on HIV viraemia driving inflammation and T-cell turnover [26]. What is intriguing in this context is the small but significant difference in CD4 count at week 48 following only 12 weeks’ raltegravir intensification, which is consistent with previous findings of modestly greater longer-term CD4 increases in HIV seroconverters initiating integrase-inhibitor-containing ART [27] and the STARTMRK trial [28]. We are unable to explore whether CD4 immune restoration continued to diverge after 48 weeks to larger, more clinically relevant differences [29] but plan to measure inflammatory biomarkers [30,31], T-cell subsets, and markers of T-cell turnover and activation in a subset of participants in whom peripheral blood mononuclear cells were stored. However, our results do suggest that quadruple ART is unlikely to provide any clinical benefit over standard triple-drug ART in patients with VL >100,000 copies/mL or advanced disease. Furthermore, the lack of evidence of clinical benefit does not support evaluation of adjunctive INSTIs in less immunocompromised patients, whose risk of clinical events is markedly lower [5].
With increasing moves towards replacing NNRTIs with INSTIs in first-line ART, findings from retrospective observational studies [32,33] have raised major concerns regarding the potential for increased IRIS events with faster VL declines, which may have particularly important consequences in severely immunocompromised patients and where CD4 counts may no longer be available, so that closer clinical monitoring cannot be instigated. Our randomised trial found no evidence of higher rates of clinical IRIS events associated with the significantly faster VL declines with raltegravir-intensified ART. This finding also has relevance for the increasing numbers of treatment-experienced patients returning after previous disengagement from care [34], for whom INSTIs are an attractive option given challenges in accessing resistance testing. Whilst unaffected by raltegravir intensification, IRIS events were significantly reduced by the enhanced-prophylaxis bundle also investigated in this trial [35].
One potential benefit from faster declines in VL is reduced potential for onward transmission, particularly because almost half of the trial participants had minimal symptoms (WHO stage 1 or 2). However, the differential suppression occurred between weeks 4 and 12, with overall only 15.8% less time spent with VL <50 c/mL in the standard ART group through week 24 (albeit predominantly at the time of greatest risk of clinical events). Therefore, benefits at a population level would likely be modest at best. However, such a strategy might have particular value in swiftly reducing VL in women identified as HIV-infected during pregnancy to reduce mother-to-child transmission [36].
Poorer compliance with raltegravir-intensified ART could explain the lack of clinical benefit, although the substantial differences in early VL suppression suggest that adherence to raltegravir-intensified ART was reasonable. However, the additional twice-daily raltegravir pill reduced self-reported adherence to ART in participants receiving other ART once daily by about 4% (S2B Fig). Association with twice-daily dosing, rather than pill burden, is supported by the lack of difference in self-reported ART adherence between participants randomised in this trial to enhanced-prophylaxis versus standard-of-care co-trimoxazole (one additional pill once daily from day 5 to 12 weeks) [17]. Although the induction of UGT1A1 by efavirenz (reducing raltegravir levels by about one third) and/or incomplete adherence in those on twice-daily regimens might have reduced efficacy, there was no evidence of heterogeneity in VL suppression at week 4, or mortality, according to initial NNRTI or according to whether other ART was once daily versus twice daily.
At 48 weeks, 20.0% of participants had a VL ≥50 copies/mL and 12.6% had ≥1,000 copies/mL; this may be unsurprising given advanced disease at ART initiation but limits power to investigate drug resistance. Despite similar rates of virological failure ≥1,000 copies/mL, raltegravir-intensified ART was associated with lower rates of the NRTI mutation K219E/Q and the NNRTI mutations K101E/P and P225H. A limitation is that we did not assay samples at ART initiation; however, randomising large numbers should provide balance between raltegravir-intensified ART and standard ART in transmitted drug resistance. This may suggest that these specific mutations arise early in treatment and, even if then suppressed, ultimately re-emerge at failure. However, differences in predicted NRTI/NNRTI resistance were at most marginal for tenofovir, abacavir, and rilpivirine, and raltegravir intensification did not substantially protect against developing clinically meaningful NRTI/NNRTI resistance. Major integrase mutations potentially compromising raltegravir occurred in very small numbers and did not preclude future use of dolutegravir.
The trial’s major strengths were the mortality primary endpoint in a well-defined high-risk population from eight HIV clinics across four African countries, increasing generalisability. Limitations include the fact that some WHO 3/4 and IRIS events were probably misclassified due to limited clinical, radiological, and/or microbiological information for some participants, reflecting available services at the centres, meaning we were also unable to accurately distinguish between IRIS-associated unmasking versus new active disease. Furthermore, the first post-randomisation CD4 was done at 4 weeks only and VL data were retrospective and not available for review. However, adjudications were made blinded to trial drugs, minimising bias in the overall findings, and the primary endpoint, 24-week mortality, was completely objective. IRIS rates were lower than some previous studies but likely included most clinically important events. Ultimately, the trial was well powered to detect differences in mortality, providing confidence that even if some IRIS events were missed, these were not leading to excess deaths given the relatively tight 95% CI around absolute mortality differences. We did not assess VL response before 4 weeks (first stored sample), when one third of the deaths had already occurred. We were unable to genotype Malawian samples (12.5% of the participants) and obtained integrase genotypes in only 73% of samples with VL >1,000 copies/mL at 12 weeks, possibly because of challenges with primers for non-clade-B viruses. We recruited fewer children than planned, limiting our ability to assess effects in this subgroup.
Overall, REALITY trial findings do not support intensification of ART with the INSTI, raltegravir, for the first 12 weeks in severely immunocompromised adults and children infected with HIV. In contrast, the enhanced prophylaxis bundle also tested in the REALITY trial provided substantial reduction in both mortality [17] and morbidity (including IRIS) [35] and could also be practically implemented at primary care level, provided CD4 counts were available to identify those at risk. Successful future interventions would need to improve both the rapidity and functional quality of immunological recovery and/or reduce inflammation and/or further improve prevention and management of IRIS. Earlier diagnosis and initiation of lifelong ART should continue to be a major focus, particularly in sub-Saharan Africa, as, despite recommendations for universal ART, the proportion of ‘late presenters’ appears to have plateaued in both low- and middle-income countries [37] and high-income countries [38], suggesting that substantial numbers will continue to present late for care in low- and middle-income countries for the foreseeable future. Finally and importantly, this study provides no evidence that moving to INSTIs as part of standard first-line therapy, as currently planned across many low- and middle-income countries, will increase rates of clinically important IRIS in such individuals.
|
10.1371/journal.pgen.1003016 | Mismatch Repair Balances Leading and Lagging Strand DNA Replication Fidelity | The two DNA strands of the nuclear genome are replicated asymmetrically using three DNA polymerases, α, δ, and ε. Current evidence suggests that DNA polymerase ε (Pol ε) is the primary leading strand replicase, whereas Pols α and δ primarily perform lagging strand replication. The fact that these polymerases differ in fidelity and error specificity is interesting in light of the fact that the stability of the nuclear genome depends in part on the ability of mismatch repair (MMR) to correct different mismatches generated in different contexts during replication. Here we provide the first comparison, to our knowledge, of the efficiency of MMR of leading and lagging strand replication errors. We first use the strand-biased ribonucleotide incorporation propensity of a Pol ε mutator variant to confirm that Pol ε is the primary leading strand replicase in Saccharomyces cerevisiae. We then use polymerase-specific error signatures to show that MMR efficiency in vivo strongly depends on the polymerase, the mismatch composition, and the location of the mismatch. An extreme case of variation by location is a T-T mismatch that is refractory to MMR. This mismatch is flanked by an AT-rich triplet repeat sequence that, when interrupted, restores MMR to >95% efficiency. Thus this natural DNA sequence suppresses MMR, placing a nearby base pair at high risk of mutation due to leading strand replication infidelity. We find that, overall, MMR most efficiently corrects the most potentially deleterious errors (indels) and then the most common substitution mismatches. In combination with earlier studies, the results suggest that significant differences exist in the generation and repair of Pol α, δ, and ε replication errors, but in a generally complementary manner that results in high-fidelity replication of both DNA strands of the yeast nuclear genome.
| The stability of complex and highly organized nuclear genomes partly depends on the ability of mismatch repair (MMR) to correct a variety of different mismatches generated as the leading and lagging strand templates are copied by three polymerases, each with different fidelity. Here we provide the first comparison, to our knowledge, of the efficiency of MMR of leading and lagging strand replication errors. We first confirm that Pol ε is the primary leading strand replicase, complementing earlier assignment of Pols α and δ as the primary lagging strand replicases. We then show that MMR efficiency in vivo strongly depends on the polymerase that generates the mismatch and on the composition and location of mismatches. In one extreme case, a flanking triplet repeat sequence eliminates MMR altogether. Overall, MMR is most efficient for mismatches generated at the highest rates and having the most deleterious potential, thereby ultimately achieving high-fidelity replication of both DNA strands.
| Three processes operate to ensure faithful replication of the eukaryotic nuclear genome [1], [2]. The first is the ability of DNA polymerases α, δ and ε to selectively insert correct rather than incorrect nucleotides onto correctly aligned rather than misaligned primer-templates. The second is proofreading, the 3′ exonucleolytic excision of errors from the primer terminus during replication. The third is mismatch repair (MMR) of errors that escape proofreading (reviewed in [3]–[7]). MMR begins when a mismatch is recognized by homologues of the bacterial MutS homodimer, either Msh2-Msh6 (MutSα) or Msh2-Msh3 (MutSβ). This recognition initiates a series of steps that ultimately remove the replication error from the nascent strand and allow new DNA to be synthesized accurately.
The origin and nature of the strand discrimination signal used for MMR in vivo remains uncertain. MMR requires the presence of a discontinuity in the newly synthesized strand. At least in vitro, this discontinuity can be a nick or gap located either 3′ or 5′ to the mismatch, with the protein requirements for MMR differing somewhat depending on the location of the DNA ends relative to the mismatch. This provides an attractive possibility (reviewed in [3]), namely that MMR may be directed to the nascent strand by the 3′ ends of growing chains at the replication fork and/or by the 5′ ends of Okazaki fragments that are transiently present during lagging strand replication. That the latter could provide a higher signal density for MMR of lagging strand replication errors was suggested in an earlier study of MMR of a damaged (8-oxo-G-A) mismatch [8]. This leads to a previously unexplored question addressed by the present study, i.e., is the efficiency of MMR similar or different for mismatches generated during leading and lagging strand replication?
Investigation of this question is complicated by the fact that DNA polymerases α, δ and ε (Pols α, δ and ε, respectively) are all required to efficiently replicate the nuclear genome [9], and these polymerases have different error rates and error specificities [2], [10]. Over the years, multiple models have been considered for the division of labor among these three polymerases during replication (reviewed in [9]–[12]). Among these models, recent evidence [2], [13], [14] suggests that under normal circumstances, the leading strand template is primarily replicated by Pol ε, while the lagging strand template is replicated by Pol α-primase and Pol δ.
Although MMR corrects errors made by all three polymerases [2], [13], [15]–[21], it has only recently become possible to determine the extent to which MMR efficiency, and possibly MMR enzymology, varies depending on the replicase that made the error, the nascent strand containing the error and/or the location of the error within a DNA strand. We are investigating these variables using Saccharomyces cerevisiae strains containing mutant alleles of the POL1 (Pol α), POL2 (Pol ε) and POL3 (Pol δ) genes. These mutant alleles, pol1-L868M [18], [19], pol2-M644G [13] and pol3-L612M ([2] and references therein), encode enzymes with single animo acid replacements at the polymerase active site that reduce the fidelity of DNA synthesis. As a consequence, strains harboring these alleles have elevated spontaneous mutation rates, thereby allowing assignment of responsibility for most in vivo errors to a chosen mutator polymerase, rather than its wild type counterparts [2], [13]. In strains containing these mutator polymerases, URA3 mutation rates and mutational spectra can be determined and used to calculate the rates for specific mutations, e.g., single base substitutions and insertions/deletions (indels) in various sequence contexts. Comparison of these rates in MMR-proficient yeast strains to strains that lack MSH2-dependent MMR yields a calculation of the apparent MSH2-dependent MMR efficiency for a variety of replication errors generated during replication in vivo.
Using this approach, we recently described the efficiency of repairing lagging strand replication errors generated by L868M Pol α and L612M Pol δ [21]. Here we extend the effort using yeast strains encoding M644G Pol ε, allowing the comparison of MMR correction efficiencies for replication errors made by each of the three eukaryotic replicative polymerases. The results indicate that on average, MMR balances the fidelity of leading and lagging strand DNA replication, but with exceptions that place some base pairs at high risk of mutation from replication infidelity even in cells with normal MMR.
The present study presents what to our knowledge is the first direct comparison of MMR efficiency for errors made by all three replicases in vivo, thereby providing insights into the contribution of MMR to leading and lagging strand replication fidelity. This comparison is a continuation of efforts to examine the possibility that MMR may be directed to the nascent strand by the 3′ ends of growing chains at the replication fork [22], and/or by the 5′ ends of Okazaki fragments that are transiently present during lagging strand replication [8].
Our previous inference that Pol ε is a leading strand replicase was based on patterns of rare mutations in one gene (URA3) at one locus (AGP1) [13]. Two recent studies have made it feasible to test Pol ε strand assignment using a different biomarker, ribonucleotide incorporation into nuclear DNA. The first study demonstrated that, in addition to reduced fidelity for single base mismatches, M644G Pol ε also has reduced sugar discrimination, i.e., it incorporates rNTPs into DNA much more readily than does wild-type Pol ε [23]. In that study, rNMPs incorporated into nascent DNA during replication by M644G Pol ε were detected as alkali-sensitive sites in the nuclear genome of a pol2-M644G rnh201Δ strain, which lacks the ability to repair rNMPs in DNA due to deletion of the RNH201 gene encoding the catalytic subunit of RNase H2. A more recent study exploited this fact to probe the genomic DNA of a homologous S. pombe polε-M630F rnh201Δ mutant strain by strand-specific Southern blotting [14]. When strand-specific probes flanking ARS3003/3004 were used, the results revealed that more rNMPs were incorporated into the nascent leading strand than into the nascent lagging strand. This led to the interpretation that, as in budding yeast, fission yeast Pol ε is also the primary leading strand replicase [14]. Using this same strategy, we examined the strand specificity of rNMP incorporation in S. cerevisiae pol2-M644G rnh201Δ strains with the URA3 reporter in one of two possible orientations, using alkali treatment and subsequent probing for either the nascent leading or lagging strand with strand-specific URA3 probes (Figure 1A). One of the two strands from each pol2-M644G rnh201Δ strain was preferentially sensitive to alkaline hydrolysis (Figure 1B). In each case, this corresponded to the nascent leading strand products of replication (probe A in orientation 2 and probe B in orientation 1). These results strongly support the idea that Pol ε preferentially replicates the leading strand template. Note that the distribution of ribonucleotides within the two strands across the whole genome remains to be determined and could differ.
The strategy used here to study strand-specific MMR involves measuring spontaneous mutation rates in yeast strains with the URA3 reporter gene present in either of two orientations, both proximal to ARS306, a well-characterized, early-firing replication origin [24]. In our initial study of the role of Pol ε in replication [13], we compared mutation rates in MMR proficient (MSH2+) strains with wild type Pol ε (encoded by the POL2 gene) to rates in strains with the pol2-M644G mutation. The pol2-M644G strains had elevated mutation rates [13], an observation that is reproduced here (Table 1). The majority of 5-FOA resistant mutants had single-base mutations in the URA3 gene. In orientation 1, these were predominantly A-T to T-A mutations at base pairs 279 and 686. These mutations were rare in orientation 2 (partial spectra in [13], complete spectra in Figure S1A). This strong orientation bias, and the fact that the in vitro error rate for template T-dTMP mismatches by M644G Pol ε is much higher than the error rate for template A-dAMP mismatches, implies that Pol ε participates in leading strand DNA replication [13]. Two later studies [2], [25] indicated that Pol δ primarily acts as a lagging strand polymerase and has a less substantial role in leading strand replication. This further implied that Pol ε not only participates in leading strand DNA replication, but that it is the major leading strand replicase.
The pol2-M644G msh2Δ mutants have strongly elevated mutation rates relative to the MSH2+ strains (Table 1), indicating that the vast majority of the mutations are made by M644G Pol ε. In the absence of mismatch repair, most 5-FOA resistant mutants contained single base changes that were widely scattered throughout the URA3 coding sequence (Figure S1B). As compared to MMR proficient pol2-M644G strains, base pairs 279 and 686 in pol2-M644G msh2Δ strains did not stand out as hotspots for A-T to T-A transversions in orientation 1, even though base substitution and single base deletion hotspots were observed at several other locations (Figure S1B).
The data in Table 1 and Figure S1 were used to calculate rates for single base mutations in the MMR-proficient and msh2Δ strains (Table S2). The ratio of these rates reflects the apparent MMR correction efficiency for each type of error, and the results can be compared (see discussion) to those reported earlier [21] for replication errors made by L868M Pol α and L612M Pol δ. As noted previously [21], [26]–[29], certain correction factors could be higher if some mismatches in the MMR proficient strains are not subject to MMR, either because they are damaged or because they are generated during DNA transactions that occur outside of replication.
Conclusions about the overall balance of repair between strands and polymerases derive from collective consideration of all single base mismatches. In the pol2-M644G strain background, the MMR correction factor for all single base mismatches is 250-fold (Table 2; Figure 2A, blue bar; Table S2), i.e., on average, 249 of 250 single base replication errors generated by M644G Pol ε are corrected by MMR. This correction factor is higher than for L612M Pol δ (Table 2; Figure 2A, green diamond), but lower than for L868M Pol α (Table 2; Figure 2A, red diamond). As a consequence, the mutation rates for all three variant polymerase strains are similar when MMR is operative (top line in Table 2). Average correction factors are high for each of the four classes of single base changes generated by M644G Pol ε (Figure 2A), in the following order: deletions (1,500-fold), insertions (1,100-fold), transitions (440-fold) and transversions (72-fold). Correction factors vary widely between specific positions in the URA3 open reading frame. Figure 2B–2D show eight locations where it is possible to compare MMR of the same mismatch generated by M644G Pol ε during leading strand replication (blue bars) or by Pol α (red diamonds) and δ (green diamonds) during lagging strand replication (expanded from [21]). In order to maintain equivalent template context, leading strand errors found in one URA3 orientation in the pol2-M644G strains are always compared to lagging strand errors found in the other URA3 orientation in the pol1-L868M and pol3-L612M strains. For example, the correction factors in Figure 2B for deleting an A-T pair from the three longest runs of A-T pairs in the URA3 coding sequence (base pairs 174–178, 201–205 and 255–260, Figure S1) are each inferred to involve a single unpaired T. The comparative MMR correction factors and their implications are considered in the Discussion.
In contrast to the efficient repair of most single mismatches, the rate of A-T to T-A transversions at base pair 686 in orientation 1 (Figure S1 and Table S3) is no higher in the pol2-M644G msh2Δ strain than in the MMR-proficient pol2-M644G strain (Table S2). This indicates that T-T mismatches generated at base pair 686 during leading strand replication by M644G Pol ε are not efficiently corrected by MMR (Figure 3, “T-dT, 686,” dark blue bar). This contrasts with an average of 41-fold correction (dark blue bar on left) of the same mismatch inferred at all other A-T base pairs in URA3, i.e., A to T substitutions in orientation 1 and T to A transversions in orientation 2 (Figure S1). Adjacent to base pair 686 is a triplet repeat sequence, 5′-ATT ATT ATT gTT (designated here as ATT3). For several reasons (see Discussion), we speculated that this sequence might suppress MMR at base pair 686. To test this, we constructed strains in which ATT3 was modified to 5′-ATA ATC ATA gTT (designated ATT0, see Figure 3), with the three (underlined) changes interrupting the repeat units without changing the amino acid sequence. We then measured spontaneous mutation rates and generated mutational spectra (Figure S2) to determine if the flanking sequence changes allowed MMR of T-T mismatches at base pair 686. The results (Table S3) indicate that this is indeed the case. The MMR correction factor at base pair 686 increased to 35-fold (Figure 3, p≤0.001), indicating that 97% of T-T mismatches are repaired when base pair 686 is flanked by ATT0.
Single base-base mismatches are repaired by MutSα (Msh2-Msh6) but not by MutSβ (Msh2-Msh3) [3]–[7], implying that the ATT3 sequence is suppressing repair of the T-T mismatch that would normally occur via MutSα. However, given evidence that MutSβ can bind to a non-B-DNA structure that can form in a triplet repeat sequence and promote triplet repeat expansion (reviewed [30]), we examined whether suppression of MMR at base pair 686 might depend on MutSβ. This was done by calculating the A-T to T-A mutation rate at base pair 686 in URA3 orientation 1 in a pol2-M644G msh3Δ rnh201Δ strain [31]. The calculated A-T to T-A rate is 17×10−8, which is no lower than observed here in the Msh3+ strain (6.8×10−8, Table S3). Thus suppression of MMR by ATT3 is independent of MutSβ.
This study provides new insights into relationships between the intrinsic asymmetry of DNA replication and MMR in yeast.
We previously inferred that Pol ε participates in leading strand replication using base substitutions as biomarkers for leading strand replication. These events are rare, occurring approximately once per 10 million incorporations. The present study uses ribonucleotides as an independent and much more abundant biomarker. The preferential presence of ribonucleotides in the nascent leading strand observed here in pol2-M644G rnh201Δ strains (URA3 orientation 1 and orientation 2; Figure 1) strongly supports the inference that Pol ε primarily participates in leading strand replication. This does not preclude occasional Pol ε participation in lagging strand replication. The interpretation that Pol ε primarily participates in leading strand replication lends credibility to the interpretations presented below regarding the efficiency of repairing mismatches made by Pol ε during leading strand replication as compared to mismatches of similar composition made by Pols α and δ during lagging strand replication. An additional notable point here is that the sizes of the nascent leading strand fragments resulting from alkaline hydrolysis of DNA from the pol2-M644G rnh201Δ strains (Figure 1) indicate that approximately one ribonucleotide may be incorporated for every 1,000 deoxyribonucleotides. This density of ribonucleotide incorporation into DNA is about four orders of magnitude higher than for A-T- to T-A transversions. Thus ribonucleotides mapped by deep sequencing techniques could serve as high density, genome-wide biomarkers of Pol ε action in vivo during replication and possibly during repair and recombination.
The average MMR correction factors for errors made by M644G Pol ε are highest for indels, intermediate for transitions and lowest for transversions (Figure 2A). This rank order is common to E. coli [28], [29], [32] and to errors made by yeast Pols α and δ [21], suggesting that MMR has conserved the ability to most efficiently correct the most potentially deleterious errors (indels), and also the base-base mismatches made at the highest rates by both bacterial and eukaryotic replicases. This general principal is qualified by the observation that MMR efficiency varies, even for the same inferred mismatch (e.g., either an extra T, a C-dT or a G-dT mismatch, Figure 2B, 2C and 2D, respectively) made by the same polymerase (M644G Pol ε) during replication of the same (leading) strand. Most sequence-dependent variations in MMR efficiency seen here are in the 2- to 10-fold range (Figure 2) depending on the comparison. That such variations are typically small is perhaps expected, since MMR is needed to preserve the stability of nuclear genomes despite their enormous sequence complexity.
Variations due to mismatch composition and location are consistent with biochemical studies showing differences in MMR in vitro [33] and with mutational studies in vivo in which the identity of the replicase that made the mismatch was unknown. Several explanations for variations in eukaryotic MMR efficiency can be explored in the future. For example, the efficiency with which E. coli repairs transversion mismatches in phage λ increases with increasing G-C content in neighboring nucleotides [32], and recognition of certain mismatches by MutSα is influenced by a 6-nucleotide region surrounding the mismatch [34]. Thus it may be that flanking sequences, such as those shown in Figure 2, influence eukaryotic MMR efficiency in vivo by modulating (i) mismatch binding by MutSα, which contacts several base pairs on either side of the mismatch [35], (ii) base pair stacking, since a MutSα-bound mismatched base stacks with a conserved phenylalanine in Msh6, and/or (iii) DNA flexibility, since MutSα-bound mismatched DNA is kinked, and a transition between bent and unbent DNA may be critical for limiting MMR to processing of mismatched as compared to matched base pairs [36]. Variations in MMR efficiency might also depend on proteins that operate downstream of mismatch binding, such as MutLα or exonucleases, or they may reflect other variables, such as the timing of nucleosome reloading behind the replication fork, nucleosome dynamics and/or chromatin remodeling.
A striking observation here is the apparent absence of MMR of the A-T to T-A transversion at base pair 686 (Figure 3), which is inferred to result from a T-T mismatch made by M644G Pol ε during leading strand replication. This lack of repair contrasts sharply with efficient repair at many other locations. For example, the deletion mismatch at base pairs 255–260, which is predicted to involve a mismatch containing a single unpaired T in the template (Figure 2B), has an approximately 6000-fold higher correction factor than for the T-T mismatch at base pair 686. Lack of repair at base pair 686 is not due to a general inability to correct A-T to T-A transversion mismatches, because the average correction factor for these events elsewhere in URA3 is 41-fold (Figure 3). The absence of correction at position 686 led us to test whether MMR was inhibited by the adjacent 5′-ATTATTATTgTT sequence. There were several reasons to suspect that this could be the case. The sequence is A-T rich and may have unusual helical parameters that could diminish MMR. For example, sequences containing larger numbers of ATT repeats can form a non-hydrogen bonded structure [37], and can be induced into hairpins by the DNA minor groove binding ligand DAPI (4′,6-diamidino-2-phenylindole) [38], [39]. Triplet repeat sequences can form non-B-DNA structures that bind MMR proteins (reviewed in [30]), and they are often associated with genome instability (reviewed in [40]), albeit characterized by indels rather than base substitutions. In addition, recent studies have demonstrated that nucleosomes influence the behavior of MMR proteins and visa versa (e.g., see [41]–[44]), and nucleosome binding to DNA is influenced by DNA sequence, with A-T-rich dinucleotides such as those present in ATT3 having an important role in nucleosome positioning (e.g., see [45], [46] and references therein).
For these reasons, we examined MMR at base pair 686 after changing the flanking sequence to eliminate the triplet repeats and decrease A-T content by one base pair. The results indicate that these changes allowed correction of 97% of the mismatches generated by M644G Pol ε at base pair 686 (88% correction at the lower 95% confidence limit, Figure 3). This suggests that the ATT3 flanking sequence is a natural cis-acting suppressor of the normal MSH2-dependent MMR machinery. Suppression does not decrease upon deletion of MSH3, and thus is MutSβ independent, unlike triplet repeat expansion [30]. Collectively, position 686 and ATT3 are an example of what has been called an “At Risk sequence Motif” [47], i.e., a naturally occurring DNA sequence that results in inefficient operation of a DNA transaction required for genome stability. The fact that one such sequence exists in the 804 base pair open reading frame of URA3 leads one to wonder how many natural suppressors of MMR might be present in nuclear genomes. This issue is currently being investigated using the deep sequencing approach previously used to infer that Pol δ is a lagging strand replicase across the yeast genome [25]. Experiments are also planned to examine which (if any) of the possibilities mentioned in the preceding section may be relevant to inefficient MMR at base pair 686.
We previously suggested that MMR may be directed to the nascent strand by the 3′ ends of growing chains at the replication fork [22], and/or by the 5′ ends of Okazaki fragments that are transiently present during lagging strand replication [8]. The 5′ ends of Okazaki fragments, and perhaps the PCNA required to process these ends, could potentially provide a higher signal density for MMR of lagging strand replication errors as compared to errors generated during leading strand replication, which is thought to be more continuous. If so, then MMR might be more efficient in correcting lagging strand errors. In an initial test of this hypothesis, we found that mutagenesis due to a mismatch formed at one particular G-C base pair during replication of unrepaired 8-oxo-G in ogg1-deficient yeast was lower for lagging as compared to leading strand replication, and importantly, that this bias was largely eliminated in MMR defective strains [8]. Among several possible explanations that we considered for loss of the strand bias, one was that 8-oxo-G-dA mismatches made during lagging strand replication may be more efficiently corrected than are 8-oxo-G-dA mismatches made during leading strand replication. A major goal of the present study was to test this hypothesis for multiple, natural (i.e., undamaged) mismatches generated at different locations during replication of a larger target sequence. The present study accomplishes this, and allows the first direct comparison of MMR efficiency for errors made by all three replicases, to our knowledge, thereby providing insights into the contribution of MMR to leading and lagging strand replication fidelity.
From the results in Figure 2, we conclude that in general, mismatches made by all three replicases are repaired very efficiently. This is logical given the need to preserve genetic information in both DNA strands. This conclusion is independent of various models regarding which DNA polymerase replicates which strand (reviewed in [11], [12]). Other implications derive from the model wherein Pols α and δ are the primary lagging strand replicases and Pol ε is the primary leading strand replicase. In our earlier report [21], we pointed out that correction factors were higher for mismatches made by Pol α than for the same mismatches made by Pol δ, suggesting that the 5′ ends of Okazaki fragments may be strand discrimination signals and that MMR efficiency may be related to the proximity of a mismatch to that signal. This is interesting given that DNA polymerase ε is highly processive, at least as processive as DNA polymerase δ, and that leading strand replication is thought to be largely continuous [48], [49], [50]. It is of course conceivable that leading strand replication may not be as continuous as current models imply. If leading strand replication is indeed largely continuous, then the fact that MMR corrects most Pol ε errors about as efficiently as it corrects errors made by Pols α and δ (Figure 2) implies the existence of MMR signals other than the 5′ ends of Okazaki fragments, and these can very efficiently direct MMR to the nascent leading strand. Possible signals for leading strand replication include the above-mentioned 3′ ends of growing chains at the replication fork [22], [51], [52], nicks introduced into the nascent leading strand by nucleases, and/or asymmetrically bound PCNA [8], [53]. PCNA is a particularly attractive possibility for differentially modulating the efficiency of MMR of errors made by the three replicases, because it is involved in early steps in MMR (see [3]–[7] for review]), it does not influence DNA synthesis by Pol α, and it does stimulate DNA synthesis by both Pol δ and Pol ε, albeit through different PCNA-polymerase interactions (see [9] and references therein).
The results in Figure 2 further suggest that, even for the same mismatch (extra T, G-dT or C-dT) in a common sequence context, MMR efficiency varies depending on which polymerase made the error. In two of three instances involving deletion of a single template T (Figure 2B), the repair of mismatches made by Pol δ is higher than for mismatches made by Pol ε. This correlates with the observation that Pol δ generates this mismatch in vitro at a higher rate than does Pol ε [54]. Similarly, transitions and transversions (Figure 2A) and several site-specific base substitutions (Figure 2C and 2D) generated by Pol α are corrected more efficiently than are mismatches generated by Pol δ and Pol ε. Pol α lacks an intrinsic proofreading exonuclease activity and is less accurate than proofreading-proficient Pols δ and ε (reviewed in [10], [55]). Thus the present study of mismatches generated by Pol ε extends the idea that MMR has evolved to most efficiently correct the most deleterious mismatches (i.e., indel mismatches). Within classes of similar deleterious potential (base-base mismatches), evolution has produced the highest efficiency versus the most frequently generated mismatches. In a model wherein Pol ε is the major leading strand replicase and Pols α and δ conduct about 10% and 90% of lagging strand replication [2], respectively, the results (Table 2; Figure 2A, average repair for single base errors) further suggest that MMR balances the fidelity of replication of the two strands despite the use of replicases with substantially different fidelity and error specificity.
The strains used in this study, the measurements of spontaneous mutation rates and the sequencing of URA3 mutants were as previously described [2], [13], [21], save that MSH2 was deleted from haploid pol2-M644G strains rather than diploid. The ATT3 to ATT0 conversion was made via site-directed mutagenesis and integration pop-out [56] in a strain with wild type polymerases. PCR product containing the ATT0 URA3 allele was then transformed into msh2Δ backgrounds and proper insertion verified via sequencing.
Genomic DNA was isolated from exponentially growing cultures (grown in YPDA at 30°C) using the Epicentre Yeast DNA purification kit. Five µg of DNA was treated with 0.3 M KOH for 2 h at 55°C and subjected to alkaline-agarose electrophoresis as described [23]. Following neutralization, DNA was transferred to a charged nylon membrane (Hybond N+) by capillary action and probed by Southern analysis. Strand-specific radiolabeled probes were prepared from a PCR-amplified fragment of URA3 template, using a previously described procedure and probe design [14].
See Text S1.
|
10.1371/journal.pntd.0005727 | Disease severity in patients with visceral leishmaniasis is not altered by co-infection with intestinal parasites | Visceral leishmaniasis (VL) is a neglected tropical disease that affects the poorest communities and can cause substantial morbidity and mortality. Visceral leishmaniasis is characterized by the presence of Leishmania parasites in the spleen, liver and bone marrow, hepatosplenomegaly, pancytopenia, prolonged fever, systemic inflammation and low body mass index (BMI). The factors impacting on the severity of VL are poorly characterized. Here we performed a cross-sectional study to assess whether co-infection of VL patients with intestinal parasites influences disease severity, assessed with clinical and haematological data, inflammation, cytokine profiles and BMI. Data from VL patients was similar to VL patients co-infected with intestinal parasites, suggesting that co-infection of VL patients with intestinal parasites does not alter disease severity.
| Visceral leishmaniasis (VL), a disease caused by a parasite, Leishmania, belongs to the most neglected tropical diseases: they mainly occur in low-income countries and affect the poorest populations. The parasites are transmitted via the bite of an insect vector and migrate to the internal organs. When VL occurs, the patients will present with enlarged spleen and liver, a disturbed haematological profile with low blood cell counts, systemic inflammation and malnutrition. This stage of the disease is fatal if left untreated. The factors that influence VL disease severity are poorly characterized. Here, we recruited patients with VL in an area of Ethiopia with a high prevalence of intestinal parasites. Our aim was to assess whether the disease was more severe in VL patients co-infected with intestinal parasites. Our results show that clinical and haematological data, inflammation, cytokine profile and nutritional status of VL patients are similar in VL patients and in VL patients co-infected with intestinal parasites. These results suggest that co-infection of VL patients with intestinal parasites does not impact on disease severity.
| The leishmaniases are a group of neglected tropical diseases (NTDs), largely affecting the least developed regions of the world. There is an estimated 350 million people in 98 countries at risk of leishmaniasis [1, 2]. Leishmaniases are a spectrum of diseases: presentations vary from subclinical infection through a range of dermal presentations of varying severity (cutaneous leishmaniasis, CL) to visceral leishmaniasis (VL). VL is caused by Leishmania (L.) donovani or L. infantum and is the most severe form of the disease. An estimated 200,000 to 400,000 new cases of VL with an incidence of 50,000 deaths occur each year, however these numbers are widely acknowledged to be a gross underestimation of the real burden [3, 4]. In global estimates, Sudan, South Sudan, Ethiopia, Kenya and Somalia account for the second largest number of annual VL cases, after South Asia [3]. VL inflicts an immense toll on the developing world and impedes economic development, with an estimated loss of 2.3 million disability-adjusted life years. There is no effective vaccine; currently used chemotherapy is toxic and increasing drug resistance is reported [5]. Infection with Leishmania parasites can be asymptomatic or can manifest as a progressive disease. VL is characterised by hepatosplenomegaly, fever, weight loss, hyperglobulinemia and pancytopenia [6]; if left untreated, it is almost always fatal. In Ethiopia, VL is caused by L. donovani and it is one of the most significant vector-borne diseases; Ethiopia has the second largest number of VL cases in sub-Saharan Africa with an estimated annual burden of 4500 to 5000 new cases. VL is worsened by malnutrition and HIV co-infection, and it has been suggested that intestinal parasitic infections might also impact on disease severity by modulating cell-mediated immunity and by worsening malnutrition [6]. Helminth infections are characterised by a strong Th2 response [7] and it has been suggested that this might suppress a protective Th1 response in VL patients and therefore contribute to the strong immunosuppression characteristic of these patients [8]. In addition, intestinal parasites may also contribute to malnutrition by competing for nutrients in the gut, inducing chronic inflammation and causing malabsorption. The Northwest of Ethiopia, where the current study took place, has a high prevalence of intestinal parasitic infections (both protozoa and helminths) [6] and malnutrition appears to be relatively common [9]. However, precise information about the impact of co-infection with intestinal parasites on the severity of patients with VL is scarce.
In the current study, we measured the impact of intestinal parasite co-infections on the disease status of patients with VL, before the start of anti-leishmanial treatment. Clinical data were collected and haematological data, inflammatory mediators and cytokines were determined. All these parameters were compared between patients presenting with VL and VL patients co-infected with intestinal parasites.
The study was approved by the Institutional Review Board of the University of Gondar (IRB, reference SBMLS/1199/07). For this cross-sectional study, a cohort of 42 male non-endemic healthy controls were recruited amongst the staff of Gondar University Hospital and 60 male VL patients were recruited from the Leishmaniasis Research and Treatment Center of Gondar University Hospital before treatment. All patients were male and migrant workers, and indeed the large majority of VL patients at the Leishmaniasis Research and Treatment Center are migrant workers. No women presented with visceral leishmaniasis during our study.
The exclusion criteria were age (<18 years), and co-infection with tuberculosis, malaria and HIV. The diagnosis of VL was based on positive serology (rK39) and the presence of Leishmania amastigotes in spleen or bone marrow aspirates [10]. Written Informed consent was obtained from each patient and control. Patients were treated with a combination of sodium stibogluconate (20mg/kg body weight/day), and paromomycin (15mg/kg body weight/day) injections, given intramuscularly for 17 days or with Ambisome® (max of 30mg/kg body weight, with 6 injections of 5mg/kg body weight /day) and showed an initial clinical cure rate of 100% after treatment, defined as follows: at the end of successful treatment, patients look improved, afebrile, and usually have a smaller spleen size than on admission and an improved hematological profile.
The diagnosis of intestinal parasites was made in fresh stools, before the start of anti-leishmanial treatment, collected in a clean screw top container. Part of the collected stool was processed using the direct saline wet mount procedure [11, 12]; part was examined using the Kato-katz technique [11, 12] and the rest was processed using the "formol ether concentration" technique [12]. All preparations were examined for the presence of parasites by microscopy, within 30 minutes after collection; each stool sample was examined by two experienced laboratory technicians. Of note, only active intestinal infections were taken into account: Giardia and Entamoeba cysts were observed in 3 and 5 patients, respectively, however, no trophozoites were detected in their stools and therefore were not categorized as causing an active infection. Six ml of blood was collected in EDTA tubes before the start of treatment and was processed within 10 minutes after collection: following density gradient centrifugation on Histopaque-1077 (Sigma), the plasma was isolated from the top layer and frozen immediately. To count the percentages of eosinophils, a drop of whole blood was smeared onto a glass slide, stained with Giemsa and the percentages of eosinophils per 100 white blood cells were counted microscopically.
The BMI was measured as follows: Weight (kg)/ (height (m)) 2. Normal BMI was defined as ≥18.5, moderate malnutrition as BMI = 18.4–16.5 and severe malnutrition as BMI < 16.5.
The haematological profiles were determined by using an automated CELL-DYN1800 Haematology Analyser, USA.
The ELISA for the determination of the acute phase C-reactive protein (CRP) was performed as described in [13] (detection limit: 0.3μg/ml). ELISA kits were used for the determination of IFN-γ, IL-6, IL-8 and IL-10 levels (Ready-SET-Go! ELISA Sets) in plasma, according to the manufacturers’ protocol. The detection limits for these ELISA were 4pg/ml, 2pg/ml, 2pg/ml and 32pg/ml, respectively. Elastase and Myeloperoxidase (BioVendor) were used according to the manufacturers’ protocol. The detection limits for these ELISA were 0.2pg/ml and 0.4ng/ml, respectively.
Cytokine analysis for interleukin (IL)-2, -4, -5, -12 and IL -13 was performed by using the Luminex 200 system (USA, Multiplex Map Kit) and the plate was analyzed using the Luminex 100 system. The detection limits for these tests were 2.5pg/ml, 0.1pg/ml, 3.1 pg/ml, 2.5pg/ml and 3pg/ml, respectively.
The enzymatic activity of arginase in the plasma was measured as previously described [14]. Briefly, urea concentrations were first determined without the activation and hydrolysis steps; these values were subtracted from those obtained by measuring the urea levels. One unit of enzyme activity is defined as the amount of enzyme that catalyzes the formation of 1 μmol of urea per min.
Data were evaluated for statistical differences using two-tailed Mann-Whitney (GraphPad Prism 6) and differences were considered statistically significant at p<0.05. Results are expressed as median± SEM.
Sixty male VL patients were recruited into our study and their clinical data are summarized in Table 1. Their intestinal parasite (IP) status was determined and as shown in Table 2: 48.3% (29 patients = IP+) were positive for intestinal parasites: 18 VL patients were co-infected with hookworms, 8 with Ascaris (A.) lumbricoides, 4 with Schistosoma(S.) mansoni, 2 with Entamoeba (E.) histolytica, 1 with Trichuris(T.) trichiura, and 1 with Strongyloides (S.) stercoralis. Of note, both S. mansoni and S. haematobium are prevalent in Ethiopia, mainly in school-age children [15, 16], but since our aim was to study the impact of intestinal parasites on VL disease severity, we focus on S. mansoni. Out of the 29 VL patients co-infected with intestinal parasites, 5 patients were co-infected with 2 different intestinal parasites (Table 2). When stratified according to their intestinal parasite (IP) status, the two groups were of similar age (IP+: 22.0±1.3 and IP-: 24.0±0.9, p>0.05); had a similar duration of illness (IP+: 8.0±0.8 and IP-: 6.0±1.0 weeks, p>0.05); similar parasite grade (spleen IP+: 3.0±0.2 and IP-: 3.0±0.3, p>0.05); similar spleen size (IP+: 10.0±0.8 and IP-: 10.0±0.7, p>0.05) and liver size (IP+: 12.0±0.5 and IP-: 3.5±0.4, p>0.05); and similar BMI (IP+: 17.4±0.3 and IP-: 17.2±0.3, p>0.05) (Table 1 and Fig 1).
We cannot exclude that we have underestimated strongyloidiasis due to the sensitivity of diagnostic tests based on parasitological examination, however, we did not have access to PCR or serological tests.
30.4% of the VL patients IP+ and 22.7% of VL patients IP- presented with diarrhoea. Importantly, co-infection with intestinal parasites did not affect the treatment of VL patients, as all 29 IP+ VL patients were successfully treated in the same time frame and had a positive initial clinical cure.
Next, we assessed the haematological profile of the 2 cohorts of patients and as shown in Table 3 and Fig 2, VL patients presented with severe pancytopenia, and anaemia, as compared to healthy non-endemic controls (S1 Table). No significant differences (p>0.05) were observed in neutrophil, white blood cell (WBC) and platelet (Plt) counts, haemoglobin (Hgb) and haematocrit (Hct) between VL patients and VL patients co-infected with IP. The percentages of eosinophils in whole blood was low (<3) or undetectable in both groups, therefore no statistical differences could be evaluated.
VL is characterized by a strong systemic inflammation as measured by high levels of inflammatory cytokines in the plasma of these patients [8]. To assess whether co-infection of VL patients with IP had an impact on the systemic inflammation, the levels of CRP, IL-6 and IL-8 were measured in the plasma of the two groups of VL patients. The results confirm the strong inflammatory status, and as shown in Table 4 and Fig 3, no significant differences were observed. We also assessed the levels of arginase, myeloperoxidase and elastase in the plasma of the two groups of VL patients; these enzymes are all found in primary granules of neutrophils, which are the last granules to be released following activation of neutrophils. Results presented in Table 4 and Fig 3 show that the levels of these enzymes were similar in the plasma of the two groups of VL patients.
VL infections in human are characterised by high levels of IL-10 and IFN-γ [17]. To assess whether co-infection of VL patients with IP affects the systemic cytokine profile, we measured by Luminex and ELISA an array of different cytokines. Our results show that the Th1 (Table 5), Th2 (Table 6) and IL-10 (Table 7) cytokine profiles were similar in the plasma of both groups of VL patients.
Our results show that the clinical and haematological data, inflammatory markers, cytokine profiles and BMI are similar in VL patients co-infected with intestinal parasites and VL patients with no intestinal parasites. The logistic regression model was used to test the association of IL-10, IL-8, IFN-γ, Plt, HCT, Hgb, WBC and BMI with IP. The IP variable was defined as positive response for at least one variable presented in Table 2. Due to low frequency of infections for the variables T. trichiura, E. histolytica and S. stercoralis (the last three variables reported in Table 2 with maximum number of infections of 1,2 and 2, respectively), co-infection patterns were investigated for hookworms, S. mansoni and A. lumbricoides using Fisher exact test (for the three possible pairs: hookworms and S. mansoni, hook worms and A. lumbricoides and S. mansoni and A. lumbricoides). In all cases it was found to be not significant.
All VL patients recruited at the LRTC presented with severe disease. Disease severity in VL patients has been associated with hepatosplenomegaly, low BMI, pancytopenia, anaemia and high parasite burden in splenic aspirates. Our results show hepatosplenomegaly in all but one patient (these measurements were not taken in two patients); that the majority of the VL patients were malnourished: 11 had a normal BMI (>18.5), 23 were moderately malnourished (BMI 16.5–18.4) and 26 were severely malnourished (<16.5) and there was no significant difference in BMI between VL patients with and without co-infection with intestinal parasites. Of note, we collected only one stool sample, however we maximised our chances to detect parasites by using three different techniques: direct saline wet mount, Kato-katz and "formol ether concentration" techniques. Furthermore, the neutrophil and white blood cell counts of both cohorts of VL patients were below the normal range. It is likely that the undetectable or very low percentages of eosinophils are due to the severe pancytopenia characteristic of VL patients. Finally, Leishmania parasites were present in the spleen or bone marrow aspirates of both patient cohorts. The results showing that VL patients are malnourished are in agreement with our previous results, showing that the large majority of VL patients are malnourished [18, 19]. Intestinal parasitic infections have long been associated with malnutrition in children [20, 21]: clinical manifestations of intestinal parasitic infection range from acute or persistent diarrhoea to dysentery, resulting in inflammation and nutrient malabsorption [22]. However, studies about intestinal parasitic infections and malnutrition in adults are less common and sometimes conflicting. For example, a study performed in Brazil showed that hookworm infection was associated with low BMI in adults [23]. But in a recent study performed in Ethiopia, the prevalence of undernutrition was similar in individuals infected with or without intestinal parasite infections [24]. In line with these results we show that the BMI of VL patients co-infected with IP were similar to those of VL patients. VL patients in the North West of Ethiopia are usually admitted with progressed, severe visceral leishmaniasis and since undernutrition and therefore a low BMI is characteristic of acute disease in patients with VL, [18, 19], any impact of intestinal parasite-related malnutrition might be hidden by the extent of malnutrition due to VL. Our results are in disagreement with those by Mengesha et al. [25], which show that VL patients were more likely to be severely malnourished when co-infected with IP. There are important differences between the recruitment of patients between our study and that of Mengesha et al., as we excluded other co-infection, such as TB, malaria and most importantly HIV; and indeed, we had previously shown that the BMI of HIV/VL patients was even lower as the BMI of VL patients [18]. In addition, we only considered patients with active intestinal parasitic infections, i.e. for example, patients with cysts of Giardia or Entamoeba, but no trophozoites, were excluded from our study. We did not find any significant differences in the median age, duration of illness, size of the spleen or liver and parasite grade in the spleen. Of note, in contrast to VL patients, none of the VL patients co-infected with IP had a maximum parasite grade of 5+ or 6+; however, the numbers of patients with high parasite burden are too low to make a significant conclusion. All the haematological data measured in the two groups were also similar and characteristic of VL patients [19].
VL is associated with high levels of systemic inflammation [8]. We have recently shown that neutrophils, which are major players in the induction and maintenance of inflammation, are highly activated and have degranulated, as shown by increased levels of arginase, myeloperoxidase and elastase in the plasma of VL patients [26]. These results suggested that the elevated levels of arginase, MPO and elastase are markers of a severe and systemic inflammatory response that is at least in part caused by high neutrophil activation. In our two cohorts of IP+ and IP- VL patients, the levels of arginase activity, myeloperoxidase and elastase were comparable. Furthermore, in addition to their potential role as markers of inflammation, these enzymes could also exacerbate immunopathological processes and therefore disease severity: arginase has been shown to play a role in T cell suppression [19] and in Leishmania parasite replication [14]; and both myeloperoxidase and elastase have the potential to affect parasite survival [27–29]. However, our results suggest that co-infection of VL patients with intestinal parasites does not impact on the levels of these three enzymes in the plasma.
Since parasitic infections are associated with an increased Th2-type response, we anticipated that the Th2 cytokines might be increased in VL patients co-infected with IP and that this might affect inflammation, but as shown by the similar levels of inflammatory mediators, the levels of inflammation are not altered in VL patients co-infected with IP; and indeed the levels of IL-5 were similar and those of IL-4 and IL-13 were below the detection limits of the ELISA we used.
In contrast to experimental models of leishmaniasis [30], the cytokine profile in VL patients is not associated with a Th1 or Th2 type cytokine profile. Indeed, high levels of both IL-10 and IFN-γ in the plasma are a hallmark of VL patients [17], IL-4 has been shown to be increased in VL patients [31], but only minimally [32] or even undetectable [33] in other studies. In line with the parameters we measured in the two groups of patients, the cytokine profiles of VL patients co-infected with IP are similar to that of VL patients.
The lack of apparent differences in the parameters we measured between the two groups of patients is likely to be due to the severity of VL, which might mask any differences that might be driven by the presence of intestinal parasites. Indeed, our previous study has shown that VL is severe in patients admitted to the Leishmania Research and Treatment Center in Gondar, as the duration of illness was considerably longer in the patients from the North West of Ethiopia as compared to those from Bihar, in India [34]; and that all patients presented with hepatosplenomegaly, low BMI, pancytopenia, anaemia and high parasite burden in splenic aspirates.
Further studies would benefit from a better characterization of the intestinal parasites in VL patients by using additional techniques, such as PCR, serological tests, as well as quantification of the load of each parasite and the collection of three consecutive samples.
In summary, our results show that co-infection of VL patients with IP does not affect VL disease severity, since clinical, haematological data and the treatment outcome are not altered by co-infection; in addition, none of the immunological markers we measured were different. However, in the light of the continuing debate about deworming [35], these results should be taken with care; our cohort of VL patients suffer from a highly debilitating disease, that is fatal if left untreated, and it might be therefore difficult to accurately dissect the contribution of a less severe disease, such as soil transmitted helminths. Notably, we did not address the medical conditions of these patients during follow-up, and whereas the relapse rate of VL patients is low [36] in this setting, we cannot exclude that these relapses might be linked to the presence of intestinal parasites. Interestingly, one recent study show that intestinal helminth infections, but not protozoan parasites, had a deleterious impact on the clinical course of cutaneous leishmaniasis (CL) caused by Leishmania braziliensis, as shown by increased frequency of mucosal lesions, poorer and longer response to therapy in co-infected CL patients [37]. Importantly, more work is needed to determine whether infection with an intestinal parasite prior to acquiring Leishmania parasites might lead to increased risks of developing leishmaniasis.
|
10.1371/journal.pntd.0001568 | Dominant Cross-Reactive B Cell Response during Secondary Acute Dengue Virus Infection in Humans | The four serotypes of dengue virus (DENV) cause dengue fever (DF) and dengue hemorrhagic fever/dengue shock syndrome (DHF/DSS). Severe disease has been associated with heterotypic secondary DENV infection, mediated by cross-reactive antibodies (Abs) and/or cross-reactive T cells. The role of cross-reactive immunity in mediating enhanced disease versus cross-protection against secondary heterotypic DENV infection is not well defined. A better understanding of the cross-reactive immune response in natural infections is critical for development of safe and effective tetravalent vaccines. We studied the B cell phenotype of circulating B cells in the blood of pediatric patients suspected of dengue during the 2010–2011 dengue season in Managua, Nicaragua (n = 216), which was dominated by the DENV-3 serotype. We found a markedly larger percentage of plasmablast/plasma cells (PB/PCs) circulating in DENV-positive patients as compared to patients with Other Febrile Illnesses (OFIs). The percentage of DENV-specific PB/PCs against DENV-3 represented 10% of the circulating antibody-producing cells (ASCs) in secondary DENV-3 infections. Importantly, the cross-reactive DENV-specific B cell response was higher against a heterotypic serotype, with 46% of circulating PB/PCs specific to DENV-2 and 10% specific to DENV-3 during acute infection. We also observed a higher cross-reactive DENV-specific IgG serum avidity directed against DENV-2 as compared to DENV-3 during acute infection. The neutralization capacity of the serum was broadly cross-reactive against the four DENV serotypes both during the acute phase and at 3 months post-onset of symptoms. Overall, the cross-reactive B cell immune response dominates during secondary DENV infections in humans. These results reflect our recent findings in a mouse model of DENV cross-protection. In addition, this study enabled the development of increased technical and research capacity of Nicaraguan scientists and the implementation of several new immunological assays in the field.
| Dengue is the most common mosquito-borne viral infection of humans, with half the world's population at risk for infection. Four different dengue virus serotypes (DENV-1 to -4) can cause the disease, which can be either inapparent or present with flu-like symptoms (Dengue Fever), also known as “breakbone fever”. In a number of cases, the disease can be more severe and sometimes fatal, with signs of bleeding and vascular leakage leading to shock (Dengue Hemorrhagic Fever/Dengue Shock Syndrome). Severe disease has been associated with secondary sequential DENV infections, i.e., infection with a second DENV serotype different from the serotype causing the first infection. No specific treatment or vaccine is available. Understanding how the human immune response develops during a natural infection can be beneficial for future vaccine studies and trials. B cells are a subset of cells that produce antibodies and are thus essential in the response to natural infections and vaccines. We show here that during secondary DENV infections in humans, the B cell immune response to a previous infecting DENV serotype is stronger than the response against the current infecting serotype. In addition, this study allowed the development of research capacity and implementation of new immunological assays in Nicaragua.
| Dengue is the most prevalent mosquito-borne viral disease affecting humans worldwide, mainly encountered in tropical and sub-tropical regions in peri-urban and urban areas, with almost half of the world's population at risk for infection. Dengue is caused by four dengue virus serotypes (DENV-1–4), transmitted by Aedes aegypti and Ae. albopictus mosquitoes. DENV infection can be asymptomatic or can cause a spectrum of disease, which spans from classical dengue (DF) to more severe forms termed dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS) [1]. DF is an incapacitating severe flu-like illness that usually resolves spontaneously. The main symptoms include high fever, retro-orbital pain and headache, muscle and joint pain, and rash. DHF/DSS is a potentially fatal form of dengue. DHF is characterized by hemorrhagic manifestations, platelet count ≤100,000 cells/mL; and signs of plasma leakage that may include elevated hematocrit, pleural effusion, ascites, edema, hypoproteinemia and/or hypoalbuminemia. If plasma leakage continues without appropriate fluid resuscitation, DSS can ensue. DSS presents with signs of circulatory failure (narrow pulse pressure or hypotension accompanied by clinical signs of shock) in addition to the signs and symptoms found in DHF. An estimated 500,000 patients require hospitalization each year for DHF/DSS, a large proportion of whom are children [2]. Recently, the WHO developed a new classification of dengue disease that replaces the traditional classification and includes Dengue with or without Warning Signs and Severe Dengue [3]. This new classification has proven to be useful in clinical management of DENV-infected individuals; however, it may be less well-suited for pathogenesis studies [4].
The four DENV serotypes co-circulate in regions like South-East Asia where dengue is hyper-endemic. In contrast, in Nicaragua, one DENV serotype tends to dominate for several years, before being replaced by another serotype, with lower-level co-circulation of other DENV serotypes. DENV-3 has been the dominant serotype circulating in the period 2008 to 2011 in Nicaragua [5]. Prior to this, DENV-2 was the predominant serotype between 1999 and 2002 and again between 2005 and 2007 [5], [6], [7], [8], while DENV-1 predominated between 2002 and 2005 [9]. DENV-4 circulates at a low level in Nicaragua [9].
Although a large proportion of DENV infections remain asymptomatic, epidemiological studies have demonstrated an association between more severe disease and secondary (2°) heterotypic DENV infections with a distinct serotype from the primary (1°) DENV infection [10], [11], [12], [13], [14]. This increase in severity during 2° heterotypic DENV infections has been attributed to antibody (Ab)-dependent enhancement (ADE), where Abs to the 1° infecting serotype bind but do not neutralize the second infecting serotype, instead facilitating an increase in viral uptake by Fcγ-receptor bearing cells [15], [16], [17]. In addition to ADE, cross-reactive T cells, formed during the 1° DENV infection, are over-activated, inducing a “cytokine-storm” syndrome implicated in the pathogenesis of shock syndrome and severe disease [18], [19], [20], [21].
No specific treatment is currently available for dengue, and vaccines trials are in Phase 1 and 2. A better understanding of the immune response developed during natural infections may be beneficial for future vaccine design as well as for defining correlates of protection for the current vaccine trials. Indeed, a balanced and long-lasting T cell, B cell and Ab response against the four serotypes is the goal of an effective tetravalent vaccine. While cross-reactive pre-formed Abs have been implicated in ADE, a cross-reactive B cell and Ab response may be beneficial and protective [22], [23], [24], [25]. In addition, we and others have shown in a mouse model of DENV infection that cross-reactive T cells can be protective [25], [26], [27]. Clearly, in humans, cross-reactive immune responses can be protective, as the majority of 2° DENV infections are asymptomatic or result in mild disease [12].
Different B cell compartments can be identified according to their phenotype, and several B cell subsets circulate in the blood during the acute phase of an infection. Naïve B cells, memory B cells and plasma cells (PCs) are phenotyped by staining with surface markers followed by flow cytometry [28]. During a 1° infection, naïve B cells are stimulated and develop into antigen-specific B cells. These B cells either differentiate into memory B cells, which reside in the secondary lymphoid organs, or into PCs, which secrete antigen-specific Abs. Prior to differentiation into PCs, B cells undergo several cycles of proliferation and differentiate into an intermediate state called plasmablasts (PBs) [28]. Short-lived PCs are active during the acute infection, while long-lived PCs (LLPCs) migrate to the bone marrow and are responsible for long-term humoral immunity [29], [30]. Memory B cells, which retain antigen-specific Abs at their surface, undergo affinity maturation, and only the clones bearing the Abs with the highest affinity survive long-term [31]. This process takes several weeks after the acute infection and continues despite the absence of circulating antigen. Memory B cells are the cells implicated in the antigen recall response and are rapidly activated during a 2° infection [28].
In this study, we analyzed the phenotype of circulating B cells by flow cytometry during the acute phase of infection in patients suspected of dengue presenting to the National Pediatric Reference Hospital, the Hospital Infantil Manuel de Jesús Rivera (HIMJR), in Managua, Nicaragua. The striking increase we observed in the percentage of PB/PCs in DENV-positive patients prompted us to analyze the DENV-specific B cell response by ELISPOT ex vivo (representing the circulating PCs at the time of infection) in acute 2° infections, against the current infecting serotype (DENV-3) and against a heterotypic serotype (DENV-2). In addition, we studied the DENV-specific avidity of serum IgG during acute infection and the neutralization capacity of the serum during the acute phase and at 3 months post-onset of symptoms. We found a higher number of cross-reactive DENV-specific PCs, which was associated with greater cross-reactive DENV-specific serum avidity during the acute phase of the infection, suggesting an important role for cross-reactive memory B cells in 2° DENV infections.
The protocol for this study was reviewed and approved by the Institutional Review Boards (IRB) of the University of California, Berkeley, and of the Nicaraguan Ministry of Health. Parents or legal guardians of all subjects provided written informed consent, and subjects 6 years of age and older provided assent.
This study was performed from August 1, 2010, to January 31, 2011, during the peak of the dengue season in the Nicaraguan National Pediatric Reference Hospital, Hospital Infantil Manuel de Jesús Rivera (HIMJR), located in the capital city of Managua. Inclusion criteria included age between 6 months and 15 years of age, fever or history of fever less than 7 days, and one or more of the following signs and symptoms: headache, arthralgia, myalgia, retro-orbital pain, positive tourniquet test, petechiae, or signs of bleeding. Exclusion criteria included: a) a defined focus other than dengue, b) children weighing less than 8 kg, c) children less than 6 months of age, and d) children 6 years of age and older displaying signs of altered consciousness at the time of recruitment. Patient data such as vital signs, clinical data, and radiographic or ultrasound results were collected on a daily basis by trained medical personnel using a standardized clinical report form until discharge. A blood sample was collected daily for a minimum of three days for Complete Blood Count (CBC) with platelets, blood chemistry, and diagnostic tests for dengue. Between days 14 and 21 after onset of symptoms, a blood sample was collected for convalescent follow-up. In addition, blood samples were collected at 3, 6, 12, and 18 months post-illness onset. At each time-point, plasma and peripheral blood mononuclear cells (PBMCs) were prepared and stored in aliquots at −80°C and liquid nitrogen, respectively.
Daily blood specimens were obtained from patients (average 2.7 samples, range 1–3), along with a convalescent/discharge sample (for 96% of the enrolled patients). Analyzed samples were obtained between 1 and 8 days post-onset of symptoms (mean of 5.6±0.08 days). Five mL of blood were collected in EDTA tubes (Becton-Dickenson, Franklin Lakes, NJ) for children with a body weight greater than 10 kg, and 4 mL were collected for children with a body weight equal or less than 10 kg. The transport temperature (∼28°C), time of sample collection, transport, reception, and processing (total = ∼2.5 hours (h)) were strictly controlled using personal data assistants (PDAs) with barcode scanners. Upon receipt in the National Virology Laboratory, an aliquot of 300 µL was removed for flow cytometry staining (see below), and the remaining 4–5 mL of fresh blood was gently pipetted into a Leucosep tube (Greiner Bio-One) containing 3 mL of Ficoll Histopaque (Sigma), and centrifuged at 500× g for 20 minutes (min) at room temperature. The plasma was removed and frozen in aliquots. The PBMC fraction was collected and transferred to a 15 mL conical tube containing 9 mL of PBS with 2% Fetal Bovine Serum (FBS; Denville Scientific Inc.) and 1% penicillin/streptomycin (Sigma). Cells were washed 3 times in this solution by centrifugation at 500× g for 10 min and resuspended in 10 mL of complete media. Before the third wash, an aliquot of 500 µL was used to obtain a cell count using a hematology analyzer (Sismex XS-1000i). After the third wash, cells were resuspended at a concentration of 107 cells/mL in freezing media consisting of 90% FBS and 10% dimethyl sulfoxide and aliquotted. Average yield was 9.6×106 total cells (3×106 to 17.6×106). Cryovials containing the cell suspension were placed in isopropanol containers (Mr. Frosty, Nalgene) at −80°C overnight and then transferred to liquid nitrogen.
Laboratory confirmation of DENV infection consisted of reverse transcription–polymerase chain reaction (RT-PCR) amplification of viral RNA [32]; isolation of DENV in C6/36 Aedes albopictus cells [7]; seroconversion of DENV-specific IgM antibodies as measured by IgM capture enzyme-linked immunosorbent assay (ELISA) [33] between acute-phase and convalescent-phase serum samples; and/or a four-fold or greater increase in total antibody titer, as measured by Inhibition ELISA [9], [34], between paired acute- and convalescent-phase serum samples. Identification of DENV serotype (1–4) was achieved by RT-PCR directed to the capsid gene [32] and/or nonstructural protein 3 gene [35] performed with RNA extracted from serum and/or supernatant of C6/36 cells obtained during virus isolation [36]. Primary DENV infections were defined by an antibody titer by Inhibition ELISA of <10 in acute-phase samples and/or <2,560 in convalescent-phase samples, and secondary DENV infections were defined by an antibody titer by Inhibition ELISA≥10 in acute-phase samples and/or ≥2,560 in convalescent-phase samples [6]. All serologic and virologic assays were performed in the National Virology Laboratory at the National Diagnosis and Reference Center (CNDR) of the Nicaraguan Ministry of Health. All clinical laboratory tests were performed in the Department of Clinical Chemistry at the CNDR or at the clinical laboratory at the Health Center Sócrates Flores Vivas [36] in Managua.
DENV was propagated in Aedes albopictus C6/36 cells (gift from P. Young, University of Queensland, Australia) in M199 medium (Invitrogen) with 10% FBS at 28°C. Cell supernatants were collected on days 5, 6, 7 and 8 post-infection and either frozen at −80°C directly or after concentration. Concentrated virus was prepared by centrifugation through Amicon filters (50 kDa, 3,250× g for 20 min at 4°C). To prepare antigen for avidity and ELISPOT assays, DENV was cultivated in Vero cells in DMEM medium (Invitrogen) with 10% FBS at 37°C and 5% CO2. Cell supernatants were collected on days 5, 6, 7 and 8 post-infection, clarified and concentrated by ultracentrifugation (26,000× g for 2 h at 4°C) and resuspended in TNE (Tris buffer, NaCl and EDTA) or PBS. DENV-2 (strain N172, passage 2) and DENV-3 (strain N7236, passage 3) are clinical strains from two Nicaraguan patients isolated in the National Virology Laboratory in Managua, Nicaragua, and passaged minimally in our laboratory. Virus titers were obtained by plaque assay on baby hamster kidney cells (BHK21, clone 15) as previously described [37]. Raji-DC-SIGN-R cells (gift from B. Doranz, Integral Molecular, Philadelphia, PA) were grown in RPMI-1640 medium (Invitrogen) with 5% FBS at 37°C in 5% CO2 for use in neutralization assays [38], [39].
On days 1, 2, and 3 of hospitalization, 300 ul of fresh whole blood was collected. Red blood cells were lysed using 1× RBC lysis buffer (eBioscience). Cells were then blocked in 5% Normal Rat serum (Jackson ImmunoResearch Inc.) before staining. Cells were stained with anti-CD138 (MI-15) or anti-HLA-DR FITC (G46-6), anti-CD20 PECy7 (2H7), anti-CD27 PE (O323), and anti-CD38 PECy-5 (HIT2). For the analysis of marginal zone (MZ) B cells, cells were stained with anti-IgD FITC (IA6-2), anti-CD20 PECy7, anti-CD27 PE, and anti-IgM PECy-5 (G20-127). Finally, cells were fixed in 2% paraformaldehyde. Samples were analyzed on a 4-color flow cytometer (Epics XL, Beckman-Coulter). Results were analyzed using FlowJo software, version 7.2.5 (TreeStar Software). All flow cytometric analysis was performed in the National Virology Laboratory at the CNDR in Managua.
To quantify the number of DENV-specific PCs, frozen PBMCs from day 6 post-onset of symptoms were thawed and analyzed by ELISPOT ex vivo [40]. Ninety-six-well filter plates were first coated with 10 µg/well 4G2 monoclonal antibody (MAb) (mouse, pan-DENV) overnight at 4°C and then blocked for 2 h at 37°C with RPMI-1640 medium plus 10% FBS. Viruses DENV-2 N172 or DENV-3 N7236 prepared from infected Vero cells by ultracentrifugation were UV-inactivated for 10 min and then incubated with the plates at a dilution of 1∶25 in PBS to capture the virus. To detect the total number of IgG-secreting cells (including both DENV-specific and non-specific ASCs), wells were coated with donkey anti-human IgG (10 µg/mL, Jackson ImmunoResearch Inc.). Virus-coated and anti-IgG-coated plates were incubated for 5–6 h with PBMCs to allow formation of Ab-antigen complexes (anti-DENV Abs with DENV and total IgG with anti-IgG). Duplicate samples of 1×105 PBMCs per well (for wells containing DENV antigen) and 3×104 per well (for wells containing anti-human IgG) were plated in the first well, and four 2-fold dilutions were distributed in the subsequent wells. After the incubation period, cells were removed, and plates were washed and incubated with biotinylated anti-human IgG Ab overnight (1/1,000, Jackson ImmunoResearch Inc.), followed by Streptavidin-Alkaline Phosphatase (AP, Vector Inc.) and BCIP/NBT substrate (Vector Inc.). Resulting spots, representing DENV-specific Ab-producing B cells or total IgG Ab-producing cells, were counted by visual inspection using an inverted microscope. Control wells were coated with 4G2 MAb and PBS with no virus. For each sample, spots counted in the control wells were subtracted from the spots counted in the test wells coated with DENV-specific antigen. ELISPOT responses were considered to be positive if the number of spots was >200 spots/106 PBMCs for total IgG.
Serum samples from the acute phase (day 6 post-onset of symptoms) and 3 months post-onset of symptoms were heat-inactivated at 56°C for 20 min and then diluted using eight 3-fold dilutions, beginning at 1∶10 and extending to 1∶21,870. Neutralization was assessed by flow cytometry using a reporter (GFP) system with pseudo-infectious DENV reporter virus particles (RVPs) [39]. DENV RVP production (DENV-1, Western Pacific 74; DENV-2, S16803; DENV-3, CH53489; DENV-4, TVP360; gift from B. Doranz, Integral Molecular) was performed in 293TREx cell lines as described [38], [39]. Supernatants containing RVPs were harvested, passed through 0.45-µm filters, aliquotted, and stored at −80°C. For all experiments, DENV RVPs were rapidly thawed from cryopreservation in a 37°C water bath and placed on ice for use in neutralization assays. DENV RVPs in RPMI-1640 complete medium were pre-incubated with an equal volume of serially diluted serum samples for 1 h at room temperature with slow agitation. Raji DC-SIGN-R cells were added to each well at a density of 40,000 cells per well, followed by incubation at 37°C in 5% CO2 for 48 h. Cells were subsequently fixed in 2% paraformaldehyde and analyzed for the percentage of cells expressing GFP by flow cytometry (Becton-Dickinson LSRII). The percent infection for each serum dilution was calculated, and the raw data was expressed as percent infection versus log10 of the reciprocal serum dilution. The data were fitted to a sigmoidal dose-response curve, using Prism (GraphPad Prism 5.0 Software) to determine the titer of antibody that achieved a 50% reduction in infection (50% neutralization titer, NT50). The NT50 titer is expressed as the reciprocal of the serum dilution. Maximum infection was determined in the absence of serum.
Serum avidity was measured using a modified ELISA protocol with urea washes [25], [41], [42]. Supernatant from Vero cells infected with DENV-2 N172 and DENV-3 N7236 was ultracentrifuged (26,000× g for 2 h at 4°C) to prepare concentrated virus. Viruses were UV-inactivated for 10 min, plated in carbonate buffer overnight in a flat-bottom 96-well plate, washed, and then blocked with PBS-T (PBS with 0.1% Tween-20) containing 5% nonfat dry milk. Wells were incubated for 1 h with serum samples from 1° or 2° DENV infections diluted in blocking buffer. Convalescent samples (day 14 to 21 post-onset of symptoms) were used for the analysis of 1° DENV infections, while acute samples (day 6 post-onset of symptoms) were used for the analysis of 1° DENV infections. The plates were washed for 10 min with different concentrations of urea (6 M urea for primary DENV cases and 9 M urea for secondary DENV cases) before adding the secondary biotin-conjugated Ab (donkey anti-human IgG) and streptavidin-AP conjugate. Finally, PnPP substrate was added to the wells, and optical density (OD) values were measured at 405 nm using KC Junior software. Background levels were measured in wells that were treated with normal human serum. For each plate, background was subtracted, and percentage of IgG bound was calculated by dividing the adjusted OD after urea washes by the adjusted OD after PBS.
Non-parametric analyses using the two-sided Wilcoxon Rank Sum test were used for pairwise comparisons, and the Mann-Whitney test was used for non-paired analysis. The Spearman test was used to examine correlations. Calculations were performed in GraphPad Prism 5.0 software.
Between August 1, 2010, and January 31, 2011, 216 patients were enrolled for suspected dengue at the National Pediatric Reference Hospital, HIJMR. Twelve patients were excluded from analysis; one patient dropped out of the study after enrollment and 11 patients had an undetermined dengue diagnostic result. Overall, 204 patients were followed up and their characteristics are shown in Table 1. One hundred and thirty patients (63.7%) were laboratory-confirmed as dengue-positive. Among these, 75 (36.8%) were 1° and 55 (63.2%) were secondary 2° DENV infections (Table 1). Serotype identification was achieved in 86.2% of dengue-positive cases, with 108 of 112 (96.4%) confirmed as DENV-3 infections. Of note, the severity of disease was relatively low in this season, with 32 (26.4%) dengue-positive cases classified as DHF/DSS [1]. Prior to circulation of DENV-3 as the dominant serotype in 2008–2010 [5], DENV-2 was the predominant circulating serotype in Nicaragua between 1999 and 2002 and again between 2005 and 2007 [6], [7], [8], while DENV-1 predominated between 2002 and 2005 [9]. Thus, children with secondary DENV infections were most probably previously infected with DENV-1, DENV-2, or both.
Fresh whole blood collected during the first three days of hospitalization in the HIMJR was stained with MAbs and analyzed by flow cytometry in order to phenotype the B cells circulating at the time of infection. Dengue diagnostic (RT-PCR) results were obtained within 24 h after hospital admission. B cells from all cases were phenotyped on day 1, while B cells from all dengue-positive cases and one out of every five OFI cases were phenotyped on all three days. This staining allowed us to distinguish between naïve B cells (CD20+CD27−), memory B cells (CD20+CD27+) and PB/PCs (CD20lowCD27high) (Figure 1A). In addition, among the memory B cells, the marginal zone (MZ) B cell subset was analyzed (IgD+IgM+) (Figure 1B). As expected, the PB/PCs expressed high levels of CD38, which is a marker of cell activation, and variable levels of CD138, which is a cell surface marker found only on PCs. In addition, this population expressed high levels of HLA-DR, indicating activation of these cells (Figure 1A).
The percentages of different B cell subsets were then analyzed over time. While no increase in percentage of PB/PCs over time was observed in OFI cases, this percentage increased and peaked on day 5 post-onset of symptoms in DENV-positive. On day 5, a significant increase in percentage of PB/PCs was found in DENV-positive patients as compared to OFI cases (mean DENV-positive = 4.72±0.97% vs. mean OFI = 0.96±0.69%, p = 0.022) (Figure 2A). Of note, among DENV-positive patients, no statistical difference in percentage of PB/PCs was found at day 5 post-onset of symptoms between 1° and 2° infections (mean 1° = 4.99±1.35% vs. mean 2° = 4.25±1.38%, p = 0.76) (Figure 2B) or between DF and DHF/DSS cases (mean DF = 4.42±1.23% vs. mean DHF/DSS = 5.50±1.51%, p = 0.48) (data not shown). A lower percentage of memory B cells was found on day 4 post-onset of symptoms in DENV-positive cases (mean DENV-positive = 1.93±0.42% vs. mean OFI = 7.52±2.07%, p = 0.020), but no clear increase over time was seen in either of the two populations (Figure 2C). A slightly higher percentage of naïve B cells was noted on day 3 post-onset of symptoms in DENV-positive cases (mean DENV-positive = 7.16±0.76% vs. mean OFI = 5.14±1.52%, p = 0.032), but again no clear increase over time was seen in either population (Figure 2D). A significantly higher percentage of MZ B cells was found on day 2 post-onset of symptoms in OFI cases (mean DENV-positive = 6.57±2.55% vs. mean OFI = 20.82±3.09%, p = 0.020), but no significant differences were found at later time-points (Figure 2E). These data correlate with data on absolute numbers of B cells calculated based on the number of total lymphocytes (Figure S1). Of note, despite a higher number of total lymphocytes in OFI, the numbers of PB/PCs are greater in DENV-positive patients when compared to OFI between days 4 and 6 post-onset of symptoms.
The characteristics of the patients with 2° DENV infections enrolled during the study are shown in Table 2. Among the 55 cases, only confirmed DENV-3-positive cases were processed by ELISPOT to measure the number of DENV-specific PCs circulating in the peripheral blood during the acute phase (day 6 post-onset of symptoms). Concentrated preparations of virions from clinical isolates of DENV-2 and DENV-3 from Nicaragua, minimally passaged in the laboratory, were used as antigen in order to match as closely as possible the virus to which the patients were exposed. Of 33 cases with detectable ASCs, DENV-3-specific PCs represented 11.5% of the total ASC/106 PBMCs (mean DENV-3-specific ASC = 1,008±295 ASC/106 PBMCs and mean total ASC = 8,783±1,028 ASC/106 PBMCs) (Figure 3A).
The median age of patients experiencing secondary DENV infection was 10.5 years, with a range of 5.5 to 15.8 years. According to epidemiological data regarding the DENV serotypes that have been circulating recently in Nicaragua [6], [7], [8], [9], [43], these children could have been previously infected by DENV-1 and/or DENV-2. As these are pediatric cases, the volume of blood drawn is restricted and thus the availability of PBMCs was limited. Therefore, only a subset of samples was processed using a second DENV serotype, in this case DENV-2, in addition to DENV-3 as antigen (Table 2). DENV-2 was chosen to represent a cross-reactive, heterotypic serotype to which patients in the study were likely to have been exposed. A significantly higher number of DENV-2-specific ASC was found in these 2° DENV infections when compared to the number of DENV-3-specific ASC (mean DENV-2 ASC = 4,402±823 ASC/106 PBMCs vs. mean DENV-3 ASC = 1,129±373 ASC/106 PBMCs; p<0.0001) (Figure 3B). DENV-2-specific ASC represented on average 46±7% of the total ASC circulating at the time of infection, compared to 10±3% DENV-3-specific ASC (p<0.0001) (Figure 3C). Overall, these data show an increase in DENV-specific PCs during acute 2° DENV infections, with a greater increase in cross-reactive PCs that are specific to a previous infecting serotype rather than the current infecting serotype. A positive correlation was found between the titer of total DENV-specific Abs as measured by Inhibition ELISA and the number of DENV-2-specific PCs during acute infection, while no correlation was found with the number of DENV-3-specific PCs (Figure 3D and E). This result suggests that the anti-DENV specific Abs are mostly produced by the cross-reactive PCs during an acute 2° DENV infection.
In order to measure IgG serum avidity, we used a modified ELISA with urea washes [25], [41], [44]. The same clinical viral isolates from Nicaragua that were used in the ELISPOT assays were used in the avidity assay. To validate the assay using samples and virus from Nicaragua, we tested a subset of 42 1° DENV-3 cases from the 2010 hospital study. As the amount of IgG is low during the acute phase of 1° infections, we used serum samples from the convalescent phase (day 14 to 21 post-onset of symptoms). The serum avidity of these samples was measured against both DENV-2 and DENV-3. As expected, higher avidity was found against the infecting DENV serotype, DENV-3, with a low level of cross-reactivity against DENV-2 (mean % IgG bound to DENV-3 = 27.7±1.4% vs. mean % IgG bound to DENV-2 = 9.4±0.9%; p<0.0001) (Figure 4A).
We then measured the DENV-specific serum avidity during the acute phase of 2° DENV-3 infections (day 6 post-onset of symptoms). The same subset of samples that was processed for DENV-2 and DENV-3 ELISPOT was processed by the avidity assay. As shown in Figure 4B, the cross-reactive serum avidity against DENV-2 was significantly higher than the homotypic serum avidity against DENV-3 (mean % IgG bound to DENV-2 = 61.3±3.7% vs. mean % IgG bound to DENV-3 = 50.7±3.6%; p = 0.030). Overall, these data show a greater cross-reactive DENV-specific IgG serum avidity as compared to homotypic DENV-specific IgG serum avidity during the acute phase of 2° DENV infections.
Finally, we measured the DENV-specific neutralization capacity of patient serum against the 4 DENV serotypes using an RVP flow cytometry-based neutralization assay. The same subset of samples that was processed for DENV-2 and DENV-3 ELISPOTs was processed by the neutralization assay. The NT50 titer of 2° DENV-3 infections at 3 months post-onset of symptoms is shown in Figure 5A. The NT50 titer was high not only against DENV-3 (mean 986±276), the current infecting serotype, but also against DENV-2 (mean 2039±371). The NT50 against DENV-1 (mean 404±91) and DENV-4 (mean 390±192) were lower but detectable. Thus, after 2° DENV infections, a broad cross-reactive neutralization response develops against the 4 serotypes, consistent with previous reports.
In addition, we measured the NT50 titer of these same samples during the acute phase of the infection at day 6 post-onset of symptoms. As expected, NT50 titers were higher during the acute phase when compared to the 3-month samples. The NT50 titer was high not only against DENV-3 (mean 4783±1687), the current infecting serotype, but also against DENV-2 (mean 3979±1274), DENV-1 (mean 3244±1049), and DENV-4 (mean 4654±1342). Thus, as at 3 months post-onset of symptoms, we found a broadly cross-reactive response to all 4 serotypes during the acute phase of the infection (Figure 5B). Of note, no statistical significant difference was found between anti-DENV-2 and anti-DENV-3 NT50 titers, either during the acute phase or at the 3-month time-point.
In this study, we used flow cytometry to phenotype the B cell components circulating at the time of DENV infection, using fresh whole blood in Nicaragua. In addition, we measured the number of DENV-specific PCs during acute infection by ELISPOT using Nicaraguan virus preparations as antigen. Finally, we measured both the DENV-specific IgG serum avidity and neutralization capacity of the serum against different serotypes of DENV. Overall, we show that a large number of PB/PCs circulate during DENV infection when compared to OFIs, both during 1° and 2° DENV infections. We find a strikingly higher number of DENV-specific PCs and serum IgG avidity directed to a heterotypic DENV serotype (DENV-2) as opposed to the current infecting serotype (DENV-3). Overall, we show that a cross-reactive B cell response dominates during the acute phase of 2° human DENV infections.
A large percentage of PB/PCs circulate in the blood of DENV-infected children during the acute phase of infection, in both 1° and 2° DENV infections, as compared to children with OFIs. Of note, the amount of PB/PCs does not vary with age [45]. The percentage of PB/PCs circulating in the blood peaked at day 5 post-onset of symptoms. While we would have expected a high percentage of PB/PCs in both DENV-infected and OFI patients, the difference was marked and might point to either a stronger B cell response during DENV infections when compared to OFIs or to a difference between the time-points after infection at which the samples were collected in DENV-positive cases versus OFI cases. The definitive diagnosis of OFI cases is not known; however, possible differential diagnoses include influenza, rickettsiosis, and leptosporosis, among others. In an effort to define the possible viral etiology of OFIs, we analyzed DENV-negative cases using viral microarrays followed by deep sequencing and detected Human Herpesvirus 6 sequence and sequences related to other Herpesviridae and Circaviridae [46]. The course of disease of the OFIs, which may be different from dengue illness, and the fact that PB/PCs circulate in the blood for only a short period of time as compared to other B cell components [47] may explain the differences in percentage of PB/PCs between these two groups. In addition, certain viruses, like influenza and measles, are known to depress the immune system [48]; thus, some OFI patients may experience decreased proliferation of B cells either directly or secondarily due to decreased proliferation of T-helper cells, resulting in reduced numbers of PB/PCs. Of note, no difference in percentage of PB/PCs circulating in blood was noted when comparing 1° and 2° DENV infections.
In contrast to PB/PCs, which circulate in the blood during a narrow time-window, the number of memory B cells circulating in the blood increases later during infection [47]. We observed an increase over time of memory B cells in DENV-infected patients, whereas this subset of cells decreased in OFI patients. Marginal zone (MZ) B cells are IgM+ “memory” B cells that have been implicated in the response against encapsulated bacteria, such as S. pneumoniae [49]. These cells are implicated in T-cell-independent immune responses and despite the presence of IgM at their surface, they present hypermutated immunoglobulin receptors [50], [51], [52]. Recently, highly neutralizing IgM+ MAbs have been generated from individuals infected by influenza [53], and these MAbs have been shown to arise from the MZ B cell population [53]. We did not find a clear difference in the percentage of this population between the two groups. Thus, this subset of cells may not play a role during DENV infections.
In order to further characterize the PB/PCs circulating during acute DENV infection, we measured the number of DENV-specific PCs at day 6 post-onset of symptoms by ex vivo ELISPOT, i.e., without any stimulation of the PBMCs. First, we found that DENV-3-specific PCs constitute a substantial proportion (∼10%) of total ASCs in the blood of patients with a 2° DENV-3 infection. Among the patients experiencing a 2° DENV-3 infection, a subset of samples were processed by ELISPOT against both DENV-2 and DENV-3 viruses. Interestingly, we found a higher number of PCs specific for the non-infecting serotype (DENV-2) when compared to the currently infecting serotype (DENV-3). These DENV-2-specific PCs made up 46% of the total ASCs. These findings were associated with the IgG serum avidity data, where higher serum avidity was detected against DENV-2 as compared to DENV-3. Thus, during an acute 2° DENV infection, cross-reactive PCs and cross-reactive Abs responsible for the higher avidity increase more than homotypic PCs and homotypic Abs directed to the current infecting serotype. In addition, a positive correlation between the total anti-DENV Ab titer was found only with DENV-2 specific PCs but not with DENV-3 specific PCs, consistent with other reports [44]. Thus, the increased number of anti-DENV Abs circulating during a 2° infection may be induced by cross-reactive PCs, and this rise in Ab titer is associated with an increased IgG serum avidity against a heterotypic serotype. These findings support the initial concept of “original antigenic sin” in dengue immunopathogenesis, whereby the humoral immune response in a secondary DENV infection is stronger to the prior infecting serotype [54], [55].
These data are in accordance with our findings in our mouse model of sequential DENV infection, where we observed an increase in PCs, memory B cells, and highly avid Abs against the previous infecting serotype rather than against the current infecting serotype [25]. These data are also in accordance with recently published human data, which show an increase in cross-reactive memory B cells and cross-reactive serum avidity during the acute phase of 2° DENV infection in a population of DENV-infected children in Thailand [44]. These two sets of data are complementary, as we measured the number of DENV-specific PCs ex vivo (plated directly for ELISPOT without prior in vitro stimulation) during acute infection, while Mathew et al. [44] measured the number of memory B cells obtained from PBMCs polyclonally stimulated in vitro. Overall, these two studies suggest that the increase in cross-reactive PCs during an acute 2° DENV infection is mediated by the cross-reactive memory B cells formed during a previous infection with a different serotype.
Neutralization assays during the acute phase and at 3 months post-onset of symptoms show a broadly cross-reactive response against the four serotypes of DENV, as previously described [56]. Thus, there appears to be no association during 2° DENV infection between neutralization capacity of the serum and the number of circulating DENV-specific PCs or increased DENV-specific serum avidity. Direct correlation between neutralization capacity of serum and serum avidity has not been shown thus far during DENV infection. In fact, it was found that no direct correlation exists between neutralization capacity and affinity of anti-DENV MAbs [57](K. Williams and E. Harris, unpublished data). In addition, in our mouse model of sequential DENV infection, we demonstrated an uncoupling of the neutralization and avidity responses during 2° DENV infections, with a higher DENV-specific avidity against the 1° infecting serotype and an increased neutralization capacity of the serum against the 2° infecting serotype [25]. Further analysis of 1° and 2° serum samples, including samples from patients enrolled in our Nicaraguan or other cohort studies for which the 1° infecting serotype is known, are needed to further investigate this question in humans.
While other groups have used recombinant proteins for avidity and ELISPOT assays [44], we used viral particles as antigen, prepared from Nicaraguan clinical viral isolates. Previous data have shown that human anti-DENV and anti-West Nile Virus (WNV) Abs bind to the viral prM/M protein and to sites on the envelope (E) protein or on several E monomers on the virion that are not preserved in the recombinant E formulation [58], [59], [60]. Thus, we preferred to use whole viral particles in our assays to better approximate the viral antigen seen by the immune response in vivo. In addition, the use of clinical viral isolates from Nicaragua represents the most relevant viral strains.
The 2010–2011 dengue season in Nicaragua was characterized by low disease severity, with only 30 (23.1%) cases of DHF and 2 cases (1.5%) of DSS in our study. We did not find any difference in the number of DENV-specific PCs or in serum avidity during the acute infection between DF and DHF/DSS, but differences may exist in more severe cases. Further analyses of the B cell response during subsequent seasons with greater severity are warranted to study such associations. In addition, disease severity can be influenced by the serotype-specific sequence of infections and the time interval between sequential DENV infections [61], [62], [63], issues that are better addressed using samples from prospective cohort studies. A separate study of a prior DENV-2 epidemic in Managua revealed a trend towards decreased serum avidity in more severe DSS cases when compared to DF and DHF cases (M.O. Pohl, S. Zompi and E. Harris, unpublished data) using both a urea-based ELISA and a virus competition ELISA [55]. More refined analysis of the serum avidity by surface plasmon resonance may be more sensitive, and such studies are currently underway.
This study has several strengths. Given our established mouse model of DENV infection and disease, we can study the immune response in parallel in mice and humans. The mouse model allows a more complete mechanistic approach, e.g., allowing the investigation of the role of the different immune components during DENV infections [25], while the human studies extend the relevance of the findings to the clinical situation. For the first time, B cell- and Ab-based assays, including ELISPOTs and urea-based ELISAs, were carried out using viral particles purified from clinical isolates from the field as antigen. Using this type of antigen, prepared by propagating the virus in mammalian Vero cells, enables as close an approximation to the in vivo situation as possible. Finally, the flow cytometry was performed at the NVL/CNDR in Managua, Nicaragua. Although this limited the analysis to several four-color panels due to the cytometer available at the CNDR, it allowed analysis of fresh whole blood from children enrolled in the hospital-based dengue study. Importantly, establishing this assay in Nicaragua increased the research and technical skills of NVL personnel, which is complemented by our program of continuous training of Nicaraguan scientists at UC Berkeley in relevant scientific and technical areas. In-country use of the cytometer also resulted in continuous maintenance of the machine, which is now being used for additional projects, such as flow cytometry-based neutralization assays for serological investigation of DENV infection over time.
One of the main limitations of this study was the low level of severity observed during the 2010–2011 season, which did not allow correlations between the number of DENV-specific PCs circulating during acute infection and disease severity to be performed. The use of samples from future more severe epidemics will be useful in investigating this question. In addition, the previous infecting serotype(s) of the 2° DENV infections hospitalized in this study is unknown. The use of samples from cohort studies, in which patients are followed prospectively over time, will allow an improved analysis of the serotype-cross-reactive response initially observed in this study.
Overall, we have shown that during DENV infection, a high number of PB/PCs circulate in the blood and that during 2° DENV infection, the DENV-specific PCs are mostly cross-reactive and likely arise from memory B cells formed during previous heterotypic infections. This is associated with an increase in cross-reactive DENV-specific IgG serum avidity. The assays used in this study were either performed at the NVL in Managua, Nicaragua, or at UC Berkeley in collaboration with a researcher from Nicaragua who was trained in ELISPOT and avidity ELISA assays, thus increasing research capacity of Nicaraguan scientists. In addition, these assays were performed using clinical viral isolates from Nicaragua, better approximating the in vivo situation in humans. Lastly, these assays should be useful in the characterization of the humoral immune response induced by candidate dengue vaccines.
|
10.1371/journal.pgen.1002707 | Brain Expression Genome-Wide Association Study (eGWAS) Identifies Human Disease-Associated Variants | Genetic variants that modify brain gene expression may also influence risk for human diseases. We measured expression levels of 24,526 transcripts in brain samples from the cerebellum and temporal cortex of autopsied subjects with Alzheimer's disease (AD, cerebellar n = 197, temporal cortex n = 202) and with other brain pathologies (non–AD, cerebellar n = 177, temporal cortex n = 197). We conducted an expression genome-wide association study (eGWAS) using 213,528 cisSNPs within ±100 kb of the tested transcripts. We identified 2,980 cerebellar cisSNP/transcript level associations (2,596 unique cisSNPs) significant in both ADs and non–ADs (q<0.05, p = 7.70×10−5–1.67×10−82). Of these, 2,089 were also significant in the temporal cortex (p = 1.85×10−5–1.70×10−141). The top cerebellar cisSNPs had 2.4-fold enrichment for human disease-associated variants (p<10−6). We identified novel cisSNP/transcript associations for human disease-associated variants, including progressive supranuclear palsy SLCO1A2/rs11568563, Parkinson's disease (PD) MMRN1/rs6532197, Paget's disease OPTN/rs1561570; and we confirmed others, including PD MAPT/rs242557, systemic lupus erythematosus and ulcerative colitis IRF5/rs4728142, and type 1 diabetes mellitus RPS26/rs1701704. In our eGWAS, there was 2.9–3.3 fold enrichment (p<10−6) of significant cisSNPs with suggestive AD–risk association (p<10−3) in the Alzheimer's Disease Genetics Consortium GWAS. These results demonstrate the significant contributions of genetic factors to human brain gene expression, which are reliably detected across different brain regions and pathologies. The significant enrichment of brain cisSNPs among disease-associated variants advocates gene expression changes as a mechanism for many central nervous system (CNS) and non–CNS diseases. Combined assessment of expression and disease GWAS may provide complementary information in discovery of human disease variants with functional implications. Our findings have implications for the design and interpretation of eGWAS in general and the use of brain expression quantitative trait loci in the study of human disease genetics.
| Genetic variants that regulate gene expression levels can also influence human disease risk. Discovery of genomic loci that alter brain gene expression levels (brain expression quantitative trait loci = eQTLs) can be instrumental in the identification of genetic risk underlying both central nervous system (CNS) and non–CNS diseases. To systematically assess the role of brain eQTLs in human disease and to evaluate the influence of brain region and pathology in eQTL mapping, we performed an expression genome-wide association study (eGWAS) in 773 brain samples from the cerebellum and temporal cortex of ∼200 autopsied subjects with Alzheimer's disease (AD) and ∼200 with other brain pathologies (non–AD). We identified ∼3,000 significant associations between cisSNPs near ∼700 genes and their cerebellar transcript levels, which replicate in ADs and non–ADs. More than 2,000 of these associations were reproducible in the temporal cortex. The top cisSNPs are enriched for both CNS and non–CNS disease-associated variants. We identified novel and confirmed previous cisSNP/transcript associations for many disease loci, suggesting gene expression regulation as their mechanism of action. These findings demonstrate the reproducibility of the eQTL approach across different brain regions and pathologies, and advocate the combined use of gene expression and disease GWAS for identification and functional characterization of human disease-associated variants.
| Expression quantitative trait loci (eQTL) are genomic loci that influence levels of gene transcripts and can be mapped by genetic linkage in families or eGWAS in unrelated populations [1]. eQTLs are distinct from other complex trait loci, because they directly identify the target gene, since the transcript trait is a reflection of the mRNA level from a single gene. Furthermore, eQTLs imply regulation of gene expression as the mechanism of action for the underlying variants. Recently, few studies identified an enrichment of eQTLs from lymphocytes [2] and lymphoblasts [3] amongst human complex disease and trait GWAS loci, suggesting that eQTLs may be useful in mapping human disease-associated variants.
Most human eQTL mapping studies to date assessed immortalized lymphoblastoid cell lines [4], [5], [6], [7], [8], [9], [10], [11], [12] and family-based samples from the CEPH [4], [5], [6], [7], [8], [13] (Centre d'Etude du polymorphisme humain) or HapMap [10], [11], [14], [15] repositories. Multiple other small and large scale eQTL studies investigated other tissues and populations including lymphocytes [16], monocytes [17], T-cells [18], fibroblasts [18], skin [19], subcutaneous and omental adipose tissue [20], [21], bone [22], liver [23] and brain [24], [25].
Despite the assumption that brain eQTLs would also influence human diseases and traits, there are no systematic gene mapping studies for human diseases that utilize brain gene expression phenotypes. Furthermore, the brain region most relevant for such studies and the influence of brain pathology on eQTL mapping studies are largely unknown. To address these issues, we performed an eQTL using cerebellar tissue from 197 subjects with Alzheimer's disease (AD) neuropathology and 177 with other pathologies (non–AD). We validated the results in a different brain region using temporal cortex samples from 202 ADs and 197 non–ADs (Supplementary Tables 1 and 2 in Dataset S1), 85% of whom overlapped with the cerebellar group. We evaluated significant cisSNPs from our study for association with human diseases/traits using a GWAS catalog [26]. We also assessed our significant eGWAS cisSNPs for association with two central nervous system (CNS) diseases, progressive supranuclear palsy (PSP) [27] and AD risk [28], using two recent GWAS for these diseases.
Our results demonstrate the power of the brain eQTL approach in the identification and characterization of many human CNS and non–CNS disease-associated variants. This study also highlights the remarkable reproducibility of human eQTLs across different brain regions and pathologies, which has implications for the design of eGWAS in general. Combined assessment of eQTLs and disease risk loci can be instrumental in mapping disease genes with regulatory variants.
Levels of 24,526 transcripts for 18,401 genes were measured in 773 brain samples from the cerebellum and temporal cortex of ∼200 ADs and ∼200 non–ADs, using WG-DASL assays. Nearly 70% of all probes could be detected in >75% of the samples tested. All autopsied subjects were genotyped for 313,330 single nucleotide polymorphisms (SNPs) from Illumina HumanHap300-Duo Genotyping BeadChips, as part of the Mayo AD GWAS [29]. An eGWAS testing association of transcript levels with cisSNPs was performed using multivariable linear regression correcting for APOE ε4 dosage, age at death, gender and multiple technical variables. False discovery rate (FDR)-based q values [30] (q) were used for corrections of multiple testing.
To achieve internal replication, we first analyzed the ADs and non–ADs separately. In our cerebellar eGWAS, at q<0.05, there were 5,271 significant cisSNP/transcript associations (1,156 unique genes) in the AD, 4,450 (1,022 unique genes) in the non–AD and 10,281 (1,875 unique genes) in the combined datasets. Q-Q plots suggested a clear excess of significant results (Figure 1, Figure S1a–S1d). 2,980 cisSNP/transcript associations (2,596 unique cisSNPs, 686 unique genes) were significant at q<0.05 in both ADs and non–ADs (Table 1, Supplementary Table 3 in Dataset S1, Figure S2). The direction and magnitude of associations in both groups demonstrate remarkable similarities (Pearson's correlation coefficient = 0.98, p<0.0001). The box plots depicted for some of these top associations (Figure S3a–S3c) demonstrate this replication in ADs and non–ADs. Most associations have an additive or dominant pattern with respect to the minor allele.
To assess the genetic component contributing to gene expression variability, we estimated intraclass correlation coefficients (ICC) [31] in the 15 samples measured in replicate on 5–6 different plates and 2–3 different days. Between-subject variance accounted for a median of 60% of total probe expression variance (Supplementary Table 4 in Dataset S1; Figure S4). The 746 probes for the top 2,980 cerebellar cisSNP associations had higher between-subject variance (median = 78%).
Using multivariable linear regression, we next estimated the percent variation in cerebellar probe expression levels due to the “best” cisSNP for each transcript after accounting for technical and biological covariates. We found that the “best” cisSNP explained a median of ∼3% of the expression variation. For the top 746 probes, the “best” cisSNPs accounted for a median of 18% of the expression variance (Table 2, Supplementary Table 5 in Dataset S1).
The top 2,980 cerebellar eGWAS associations were followed up in the temporal cortex validation study. We found that 2,685 top cerebellar cisSNP/transcript associations could be tested in the temporal cortex (2,387 unique cisSNPs, 677 unique probes and 625 unique genes) (Figure 1, Table 3, Supplementary Table 6 in Dataset S1). A total of 2,089 of these (1,888 unique cisSNPs, 502 unique probes and 471 unique genes) were significant after study-wide Bonferroni corrections, many of which had effect sizes showing remarkable similarity to those from the cerebellar eGWAS (Pearson's correlation coefficient = 0.94, p<0.0001).
The top cerebellar eGWAS results were also compared to published liver [23] and brain [24], [25] eGWAS and overlap was identified for 4–11% of the top transcripts from these published studies (Text S1) Using HapMap2 genotypes, all transcripts and association threshold p<1.0E-4 in our eGWAS, we determined that 24–32% of the top transcripts from the published eGWAS overlapped with ours.
We used the cerebellar eGWAS as the discovery analysis and the temporal cortex eGWAS as the validation; since our goal is to identify significant cisSNP associations while minimizing any confounding factors due to pathology and given the fact that half of our subjects had pathologic AD, in which cerebellum is relatively unaffected whereas temporal cortex is one of the first affected brain regions. Nonetheless, we have also used temporal cortex as the discovery set and cerebellum as the validation, with remarkably similar results (Text S1, Supplementary Tables 7 and 8 in Dataset S1).
To examine whether the brain eGWAS approach identified variants implicated in human diseases/traits, we linked the 2,596 top cerebellar eGWAS cisSNPs to the “Catalog of Published GWAS” [26], which compiles weekly search results from all published GWAS of ≥100,000 SNPs where associations of p≤1.0E-05 are reported. We identified 47 cisSNPs that were also associated with 36 diseases/traits (Table 4, Supplementary Table 9 in Dataset S1). This represents a 2.4-fold enrichment of significant cerebellar cisSNPs amongst disease/trait associated SNPs, which is significant (p<10−6) based on simulations adjusted for minor allele frequencies [3] (Text S1).
Among the 36 diseases/traits associating with top cerebellar cisSNPs were central nervous system (CNS)-related conditions including Parkinson's disease (PD), Moyamoya disease, cognitive performance and attention-deficit hyperactivity disorder (ADHD). We both identified novel cisSNP/transcript associations and confirmed some previously reported ones. We found novel associations between rs6532197, which confers increased risk of PD [32], and higher brain levels of MMRN1 (cerebellar eGWAS p = pCer = 4.86×10−12; temporal cortex eGWAS p = pTCx = 4.57×10−9). MMRN1 encodes for multimerin and was found to be in a region of duplication/triplication with SNCA (encoding α-synuclein), a well-established risk gene in PD [33]. We found no significant cisSNP/SNCA level associations. These results suggest that MMRN1 may deserve further investigations as an additional PD risk gene.
Another example of a cisSNP which associates with human disease risk is rs8070723, the minor allele of which is associated with reduced risk of PD [32] and reduced brain MAPT levels (pCer = 3.36×10−7–7.02×10−69; pTCx = 9.03×10−4–8.61×10−44). Rs11012 minor allele, which confers increased risk of PD [34], showed association with lower brain LRRC37A4 levels (pCer = 1.69×10−33; pTCx = 3.378E−20). MAPT region variants were previously identified to associate with brain levels of MAPT and LRRC37A4 in neurologically normal subjects [27], [32], in a MAPT haplotype H1/H2-dependent manner [27]. Indeed, rs8070723 is in tight linkage disequilibrium with rs1052553 (r2 = 0.95, D′ = 0.97), the major allele of which marks the MAPT-H1 haplotype and associates with higher brain MAPT levels [24].
Many top cerebellar cisSNPs also associate with non–CNS diseases/traits (Supplementary Table 9 in Dataset S1). IRF5 cisSNP rs4728142 is associated with both cerebellar IRF5 levels and risk of systemic lupus erythematosus (SLE) [35]. Previously, IRF variants were shown to influence IRF splicing and expression as well as SLE risk [36], [37]. Interestingly, rs4728142 is also associated with ulcerative colitis (UC) [38] where both IRF5 and TNPO3 are reported as candidate genes. Given its influence on IRF5, but not TNPO3 expression levels, rs4728142 most likely marks IRF5, but not TNPO3 as the candidate UC risk gene.
Our approach to identify human disease-associated SNPs amongst the 2,596 top cerebellar eGWAS cisSNPs may be overly conservative, given our selection criteria to only include transcripts that are detectable in >75% of the subjects and only those cisSNPs that are significant in both independent cohorts (ADs and non–ADs). Furthermore, given that our eGWAS genotyping platform consisted of ∼300 K SNPs, it is plausible that transcript associations with SNPs from the “Catalog of Published GWAS” [26] may be missed if those SNPs did not exist in our platform. To address these issues, we repeated the cerebellar and temporal cortex eGWAS, without restrictions for transcript detection rates and using genotypes imputed to HapMap2 (>2 million SNPs). Comparison of the eGWAS associations with p<1.0E-4 to the “Catalog of Published GWAS” identified 392 unique cerebellar cisSNPs that also associate with 189 human diseases/traits; and 339 such temporal cortex cisSNPs associating with 167 diseases/traits (Text S1, Supplementary Tables 10 and 11 in Dataset S1). Amongst the associations identified by this less stringent approach were those for brain levels of CLU [39], [40], CR1 [40] and GAB2 [41] which were identified as risk loci in GWAS of Alzheimer's disease.
We also performed comparisons of the eGWAS results from the ADs and non–ADs separately to determine whether there were any results unique to these diagnostic groups (Text S1, Supplementary Tables 12, 13, 14, 15 in Dataset S1). Although 13–25% of the disease/trait associations were with cisSNPs that were unique to ADs or non–ADs, all but a few of these could also be identified in the combined analysis of all subjects. There were only 2–7 human diseases/traits with cisSNP associations that were detectable just in ADs or non–ADs, but not the combined group.
Of these unique cisSNP, those that associate with cerebellar levels of C9orf72 in non–ADs are interesting, as these variants were previously identified in GWAS of amyotrophic lateral sclerosis (ALS), where C9orf72 was one of the candidate genes at the disease locus [42], [43]. This gene was recently identified as the most common cause of familial ALS, with a repeat expansion leading to loss of an alternatively spliced transcript [44], [45]. These results further support the utility of the combined eGWAS and disease GWAS approaches in the potential identification of disease genes with modified transcript levels as the plausible disease mechanism.
In a recent PSP GWAS [27], four loci near MAPT, STX6, EIF2AK3, and MOBP conferred significant risk, in addition to three suggestive loci at 1q41 intergenic locus, BMS1 and SLCO1A2. We assessed these seven strongest PSP risk loci in our eGWAS in the ADs, non–ADs and combined datasets, as well as the PSP subset of non–ADs (Table 5, Supplementary Table 16 in Dataset S1). We found novel, significant rs11568563 minor allele associations with reduced brain SLCO1A2 levels (pCer = 2.33×10−8; pTCx = 4.36×10−2–9.14×10−18), which confers increased PSP risk [27]. SLCO1A2 encodes solute carrier organic anion transporter family member 1a2 and is a drug transporter into the CNS [46]. Fine-mapping of the SLCO1A2 region revealed rs11568563 to be the strongest cisSNP influencing brain levels of this gene (Figure S5). This SNP was also identified as the top PSP-associating variant at this locus [27]. All other cisSNPs that associate with brain SLCO1A2 levels have weaker effects that appear to be due to their LD with s11568563, which is a missense coding mutation within SLCO1A2. Whether rs11568563 is merely tagging the functional variant(s) regulating levels of SLCO1A2 or coding changes also influence expressed transcript levels require further investigations. Additionally, MAPT/rs242557 minor allele increased PSP risk [27] and brain MAPT levels (pCer = 9.78×10−3–8.8×10−13, pTCx = 1.1×10−8). MAPT/rs8070723 minor allele associated with lower brain MAPT levels in our eGWAS, decreased PSP risk [27], similar to a PD GWAS [32]. We also found nominally significant increases in brain MOBP levels (pCer = 2.13×10−2–1.71×10−7; pTCx = 1.55×10−2–1.57×10−6) with rs1768208, which increases PSP risk [27].
The recent PSP GWAS by Hoglinger et al. [27] included eQTL analysis for the significant loci using brain expression levels from 387 subjects without clinical neurologic diseases. In addition to associations between MAPT locus cisSNPs with brain MAPT and LRRC37A4 levels, they also detected signals for the nearby ARL17A and PLEKHM1 genes, neither of which were detectable in our eGWAS. They also identified cisSNP associations with brain MOBP levels but even stronger influence on the nearby SLC25A38 levels. We did not identify significant cisSNP/SLC25A38 brain expression associations. Although some of the significant probes for MOBP and MAPT harbor variants within their probe sequence, which may potentially confound associations with expression levels, these genes had other significant probes without any sequence variants (Text S1).
Most non–AD subjects in our study had pathologic diagnosis of PSP (nCer = 98, nTCx = 107, Supplementary Table 2 in Dataset S1). We assessed the 2,980 top cerebellar cisSNP/transcript associations in this subset, and found that most results were consistent with the ADs (Supplementary Tables 17 and 18 in Dataset S1).
To investigate whether any of the significant brain cisSNPs may influence risk of AD, we compared our eGWAS results to the AD risk associations from the large AD GWAS conducted by ADGC [28]. We obtained results of meta-analyses for the ADGC Stage 1+2 cohort (11,840 LOAD vs. 10,931 controls) [28] and investigated those SNPs with suggestive AD risk association in this dataset (pmeta<10−3). To ensure uniform comparison between our eGWAS and the ADGC GWAS, we assessed results from >2 million SNPs for each study using SNPs genome-wide imputed to HapMap phase 2 (release 22). There were 77,126 cerebellar (63,652 unique SNPs, 2,338 unique genes) and 68,172 temporal cortex (57,922 unique SNPs and 2,201 unique genes) cisSNP/transcript associations significant at q<0.05 representing a clear excess (Figure S6). There were 380 cisSNPs that were significant for the cerebellar transcript associations and also had suggestive AD risk associations (2.9-fold enrichment), 432 such temporal cortex cisSNPs (3.3-fold enrichment) and 356 cisSNPs significant in both the cerebellum and temporal cortex (2.7-fold enrichment, p<10−6 for all three analyses) (Figure 1, Supplementary Tables 19 and 20 in Dataset S1).
MAPT and LRRC37A4 cisSNPs, implicated in PSP [27] and PD [32] GWAS and which significantly influenced brain levels of these genes also had suggestive AD risk associations (pmeta = 8.82×10−4–1.53×10−5). CisSNP alleles associating with lower brain MAPT levels were associated with lower AD risk, similar to PD [32] and PSP [27] GWAS, which may suggest a common mechanism for these neurodegenerative diseases. ABCA7, identified recently as a novel LOAD risk locus [28], [47], had significant cerebellar cisSNPs. Further investigations of the other genes with evidence of brain transcript and AD risk association is warranted to understand their role in AD (Text S1).
To ensure that we did not miss any associations due to the stringent eGWAS criteria that we applied, we repeated the analyses using no restrictions for transcript detection rates and eGWAS p value threshold of p<1.0E-4. We also investigated cisSNPs identified in AD and non–AD brains, both separately, and jointly, given that some cisSNP associations may be unique to one group. We compared these eGWAS results to the ADGC GWAS as described above (Supplementary Tables 21, 22, 23, 24, 25, 26 in Dataset S1). Using cerebellar and temporal cortex eGWAS from all subjects, 561 and 488 unique transcripts with cisSNPs that yield suggestive AD risk associations were identified, respectively. There were 259–312 such transcripts identified in each AD or non–AD eGWAS, with >50% overlap between the two diagnostic groups' results, although many of these results could be identified in the eGWAS of combined samples. About 7–10% of the transcripts could only be identified in just ADs or non–ADs, but not the combined eGWAS. Amongst such unique transcripts were CLU and BIN1, which reside at the LOAD GWAS loci [39], [40], [48] and associate with cisSNPs in the cerebellum of non–ADs. Detailed analyses of the CLU locus cisSNP/transcript associations are in-press [49].
In a large eQTL study on 773 brain samples from ∼400 autopsied subjects, we demonstrate significant contribution of genetic factors to human brain gene expression, reliably detected across different brain regions and pathologies. There is significant enrichment of brain cisSNPs amongst disease-associated variants, advocating gene expression changes as a mechanism for the first time for certain genes implicated in human diseases, including PSP (SLCO1A2), PD (MMRN1), Paget's disease (OPTN) while replicating others (e.g. PD/MAPT, SLE/UC/IRF5). MAPT cisSNPs associating with PSP, PD and AD risk highlight potential common mechanisms for these neurodegenerative diseases.
The reported results have several important implications for the genetics of human brain gene expression: First, despite technical challenges of gene expression measurements in post-mortem brain tissue [50], ∼70% of the transcriptome can be reliably detected in >75% of the subjects across two brain regions and different disease pathologies. Second, although there is significant contribution from technical covariates, genetic factors account for a substantial proportion of the variance in brain gene expression levels. We estimate that genetic factors explain an average 3% (range: 0–85%) of the variance in human cerebellar gene expression overall, and 18% (range: 8–85%) of the variance for the top cis-regulated transcripts. These estimates show remarkable similarity to those from other eQTL studies, such as a large, family-based lymphocyte eQTL, where cis eQTLs had an overall median effect size of 1.8% and significant eQTLs accounted for 24.6% of the variance in expression [16]. Similarly, significant cisSNPs explained 2–90% of expression variance in a liver eGWAS [23].
Third, there is remarkable replication of significant cisSNP associations across different brain regions and underlying tissue pathologies. Indeed, the 2,980 top cerebellar cisSNP/transcript associations represent 58% and 68%of all significant associations in the ADs and non–ADs. Since >50% of the non–ADs were comprised of subjects with PSP, we also conducted a separate analysis of this pathologically distinct group of non–ADs and again determined that many of the top cisSNPs were also significant in the PSPs despite the small sample size (n = 98). Importantly, most of the cisSNPs had highly similar effect sizes in the ADs, non–ADs and PSP subset of non–ADs. Furthermore, 78% of the top cerebellar cis-associations were also significant in the temporal cortex. Cerebellum is a relatively unaffected region in AD, whereas temporal cortex is typically one of the first areas to harbor neuropathology [51]. It is not inherently evident whether the unaffected or affected tissue regions would be most suitable for eQTL studies. Whereas unaffected regions would have the advantage of minimizing confounding on expression measurements from pathology (such as inflammation and cell death), affected regions may be more relevant for disease-associated eQTL mapping. The substantial overlap in significant cisSNP associations between different brain regions and disease types in our study implies that sample size may be the most critical element of successful eQTL mapping. In other words, analysis of expression data collected in different tissue regions and diseases, provided there is careful statistical control, could greatly enhance power to detect eQTLs. Nevertheless, there may be important eQTLs that are specific to brain region and disease.
It is not obvious whether the cisSNP that display similar effects in different brain regions and different disease types would have relevance to human disease. The top 2,596 cerebellar cisSNPs that are significant in both ADs and non–ADs, and many of which are also significant in the temporal cortex are also enriched for variants implicated in human disease, including CNS disease, such as PD and PSP. Thus, the fourth implication of our study is that it may be possible to map disease-associated variants using eQTL studies conducted in unaffected tissue or unaffected subjects. In addition to providing a general characterization of the genetics of brain gene expression, this study successfully replicated many previously published cisSNP associations, such as rs8070723/ MAPT level, rs11012/ LRRC37A4 level associations, both of which were implicated in PD. We found novel brain expression level associations for transcripts implicated in disease, including rs11568563 association with SLCO1A2, recently identified in a PSP GWAS. The disease-associating cisSNP associations identified in this study were not restricted to CNS diseases, but also included non–CNS diseases, such as SLE, where we replicated the previously published rs4728142/ IRF5 level associations.
These findings imply that many disease-associated cisSNPs can influence gene expression idependently of tissue/region/pathology, and be mapped reliably in tissue which is unaffected, not disease-related or from unaffected subjects. Indeed, our findings are consistent with a study of lymphoblastoid cell lines from subjects affected and unaffected with asthma, where Dixon et al. [12] found no differences between asthmatics and non-asthmatics. Furthermore, they detected significant transcript level associations with SNPs that also associate with asthma. Emilsson et al. [20] performed eQTL mapping in both blood and adipose tissue and determined that >50% of significant adipose tissue cisSNPs were also significant in blood. This is similar to the overlap we detected for cerebellum and temporal cortex, though two brain regions are more likely to have similar eQTL profiles than two different tissues.
Although many cisSNP effects can be detected in many different tissue types and disease conditions as shown here and by others [12], [20], there conceivably exist expression variants which exert their effects in a tissue or disease-specific manner. For example in the eQTL comparing blood and adipose tissue, Emilsson et al. [20] also found that more transcripts from adipose tissue had significant correlations with obesity-related traits. In reality, both scenarios may be at play, such that some expression variants have more ubiquitous effects, whereas others may need tissue/cell/region/disease specific factors to exert their influence on gene expression. Indeed, many of the CNS disease related cisSNP associations in our brain eGWAS could not be identified in our comparison to a liver eGWAS [23] or an existing database for a LCL eGWAS [12], suggesting that disease-relevant tissue may be necessary to detect effects of certain cisSNPs, and highlighting the value of this brain eGWAS for CNS traits/conditions.
Despite the enrichment of our samples with tissue from AD subjects and our use of both cerebellar and temporal cortex tissue, we did not identify strong transcript associations for some of the top genes recently implicated in AD risk in large LOAD GWAS studies [28], [39], [40], [47], [48]. This could be because the AD risk variants in these genes exert their effects via mechanisms other than influencing transcript levels, namely changes in protein conformation. If so, even the negative results from an eGWAS could be informative in guiding the future deep-sequencing efforts which should focus on coding rather than non-coding, functional regions. Alternative explanations include technical shortcomings, such as inability to measure all transcript species, measurements of global rather than cell-specific gene expression, not including all tested disease-associated variants in our genotyping platform. We also need to consider that the top genes nearest the strongest variants from the LOAD GWAS may not be actual disease genes. These loci require further investigations to account for this possibility. Additionally, our criteria for selection of the top cisSNPs, requiring significance in both ADs and non–ADs, might be too stringent, thereby leading to some false negative results. Finally, it may be possible to identify additional disease-related expression variants by focusing on those that have differential influence in disease vs. non-disease tissue, although this was not a focus of analysis in this study. Given that our non–AD tissue also consisted of subjects with other neurodegenerative diseases, there may be more similarities with the AD tissue, making it more difficult to detect variants with differential disease-related expression-associations in our current study. Nevertheless, we did find associations with cisSNPs for ABCA7, a novel AD risk locus gene [28], [47] and MAPT [52], [53], [24], [54] implicated in AD.
It is important to emphasize that although the identification of transcript level associations provides another layer of confidence for disease-associating variants and genes, it is entirely possible that a variant in an LD region encompassing multiple genes, could be marking a functional disease variant in one gene and an expression variant in another gene. Thus, although highly useful in conjunction with disease association studies, eGWAS should be seen as a guide rather than ultimate evidence in disease-mapping efforts. Similarly, absence of eGWAS associations for a disease-associated variant should not be seen as contradictory evidence, but rather raise the possibility of alternative functional mechanisms for that variant.
Despite the wealth of information our study provides, we acknowledge several shortcomings. First, our non–ADs were not normal controls but often had other brain pathologies. It will be necessary to seek replication of these findings or novel cisSNP/transcript associations in normal brain tissue, as well. Second, we only focused on single SNP associations. The preliminary observations from our eGWAS findings suggest that multiple independent variants may affect brain expression levels of some genes, whereas others might be under the influence of a single strong variant. Finally, like any association study, it is not clear whether the cisSNPs identified in our eGWAS are themselves the functional SNPs or simply in LD with un-genotyped regulatory variants. Future studies focusing on analysis of haplotypes, SNPxSNP interactions, novel variant discovery and functional in-vitro studies testing effects of multiple variants are required to dissect the genetic variation underlying brain gene expression levels.
In summary, this cerebellar eGWAS study and the temporal cortex validations provide insight about the genetics of brain gene expression, a framework to guide future studies with respect to tissue/region/disease choice in eQTL studies, examples about the utility of this approach in gene mapping, replication of some known transcript associations and evidence for novel transcript associations in human disease. Combined eGWAS-disease GWAS approach may provide complementary information in mapping human disease and enable identification of functional variants that may not be possible by either approach alone.
The complete set of results from the brain eGWAS can be accessed at the National Institute on Aging Genetics of Alzheimer's Disease Data Storage (NIAGADS) website at http://alois.med.upenn.edu/niagads/. Questions about the dataset can be addressed to the corresponding author of this manuscript ([email protected]).
All subjects were participants in the published Mayo LOAD GWAS [29] as part of the autopsy-based series (AUT_60–80). All subjects had neuropathologic evaluation by DWD. All ADs had definite diagnosis according to the NINCDS-ADRDA criteria [55] and had Braak scores of ≥4.0. All non–ADs had Braak scores of ≤2.5, and many had brain pathology unrelated to AD (Supplementary Tables 1 and 2 in Dataset S1). Three-hundred forty subjects had measurements in both cerebellum and temporal cortex. This study was approved by the appropriate institutional review board.
To assess the genetic contribution to the variance in human cerebellar gene expression, we first determined between-subject variance, as a percentage of the total variance in probe expression, using ICC [31] for 15 samples measured in replicate on 5–6 different plates and 2–3 different days.
Using multivariable linear regression models, we then calculated the proportion of variance in cerebellar gene expression levels that were explained by technical effects (PCR plate, RIN, (RIN-RINmean)2), biological covariates (APOE ε4 dosage, age at death, gender) and the “best” cisSNP for each probe. These analyses were carried out on the combined dataset consisting of cerebellar expression measurements from 374 subjects and 15,283 probes with at least one cisSNP (Text S1).
We identified 2,980 cisSNP/transcript associations (2,596 unique SNPs, 746 unique probes and 686 unique genes) that achieved genome-wide significance within both the ADs and non–ADs analyses with q values<0.05. All 2,980 cisSNP/transcript associations achieved genome-wide significance with q<0.05 and pBonf<0.05 in the combined ADs+non–ADs analysis. We sought validation of these hits in the temporal cortex of 399 subjects who had WG-DASL whole transcriptome measurements and whole-genome genotypes. RNA extractions, QC, WG-DASL measurements, transcript level detections and association analyses were performed for these temporal cortex samples, in the same manner as that for the cerebellar samples. After appropriate QC, 2,685 of the 2,980 top cerebellar cisSNP/transcript associations remained detectable among the temporal cortex results (2,387 unique SNPs, 677 unique probes and 625 unique genes).
To determine whether the cerebellar eGWAS captured variants implicated in complex diseases/traits, we compared the top 2,980 cerebellar eGWAS cisSNPs with the top disease/trait associated SNPs in the “Catalog of Published GWAS” [26], curated by the National Human Genome Research Institute (www.genome.gov/gwastudies). This catalog compiles weekly search results from all published GWAS of ≥100,000 SNPs where associations of p≤1.0E-05 are reported. The catalog accessed on 04/23/2011 had 5,272 entries. We restricted our search to those entries where the “SNPs” column had only one SNP with an rs number. Thus, haplotypes and variants without rs numbers were excluded. There were 5,101 entries after this exclusion, comprised of 4,248 unique SNPs and 433 unique diseases. One SNP may associate with >1 disease/trait and each disease/trait may have ≥1 associating SNP. This list was linked to the 2,980 top cerebellar cisSNPs by common rs numbers.
To assess whether the number of observed cisSNPs that have both significant cerebellar eGWAS and disease/trait associations represent a significant enrichment, we performed simulations while adjusting for cisSNP minor allele frequencies, as previously reported [3]. We performed 1 million simulations and adjusted for the minor allele frequencies of all the tested cisSNPs in 10 bins from 0–0.05 to 0.45–0.50. Using the total number of cisSNPs that are both transcript and disease/trait associating for each simulation, we obtained an empirical p value and an estimate of fold-enrichment.
Cerebellar eGWAS results were also compared to other published eGWAS results from a human liver [23] and two human brain [24], [25] studies. The methods and results are depicted in Text S1.
To determine whether any of the cisSNPs significant at q<0.05 influenced risk of AD, we obtained meta-analyses results from the ADGC [28]. The cohorts that are assessed by ADGC, as well as the methodological details of the meta-analyses are described in detail in a recent publication [28]. Briefly, the meta-analyses of the ADGC dataset results reported here (Supplementary Tables 17 and 18 in Dataset S1) are generated from the combined analyses of stage 1 and stage 2 cohorts (Text S1), with detailed descriptions provided elsewhere [28]. Stage 1 cohorts are comprised of 8,309 LOAD cases and 7,366 cognitively normal elder controls. Stage 2 has 3,531 LOAD vs. 3,565 control subjects. Each cohort was tested for AD risk association using a logistic regression approach, assuming an additive model and adjusting for age, sex, APOE ε4 dosage and principal components from EIGENSTRAT [59]. The meta-analyses results were generated using the inverse variance method implemented in the software package METAL [61].
|
10.1371/journal.pgen.1002894 | Exome Sequencing Identifies Rare Deleterious Mutations in DNA Repair Genes FANCC and BLM as Potential Breast Cancer Susceptibility Alleles | Despite intensive efforts using linkage and candidate gene approaches, the genetic etiology for the majority of families with a multi-generational breast cancer predisposition is unknown. In this study, we used whole-exome sequencing of thirty-three individuals from 15 breast cancer families to identify potential predisposing genes. Our analysis identified families with heterozygous, deleterious mutations in the DNA repair genes FANCC and BLM, which are responsible for the autosomal recessive disorders Fanconi Anemia and Bloom syndrome. In total, screening of all exons in these genes in 438 breast cancer families identified three with truncating mutations in FANCC and two with truncating mutations in BLM. Additional screening of FANCC mutation hotspot exons identified one pathogenic mutation among an additional 957 breast cancer families. Importantly, none of the deleterious mutations were identified among 464 healthy controls and are not reported in the 1,000 Genomes data. Given the rarity of Fanconi Anemia and Bloom syndrome disorders among Caucasian populations, the finding of multiple deleterious mutations in these critical DNA repair genes among high-risk breast cancer families is intriguing and suggestive of a predisposing role. Our data demonstrate the utility of intra-family exome-sequencing approaches to uncover cancer predisposition genes, but highlight the major challenge of definitively validating candidates where the incidence of sporadic disease is high, germline mutations are not fully penetrant, and individual predisposition genes may only account for a tiny proportion of breast cancer families.
| Currently, we know that a woman who inherits a fault in one of two genes, BRCA1 or BRCA2, has a high risk of developing both breast and ovarian cancer. However, such faults account for only half of all families with a strong family history of breast cancer. In this study, we planned to identify new genes that may be associated with an increased risk of developing breast cancer by looking for faults in every gene in the blood DNA of multiple women with breast cancer from large families with a strong family history of the condition over multiple generations. We can then track which gene fault is present in all the women with breast cancer in that family and in other families, but is not found in the women who did not develop breast cancer or have no family history. Using this approach, we identified faults in two genes, Fanconi C and Bloom helicase, in six families. Faults in these genes appear to increase the risk of developing breast cancer. Both these genes work in a similar way as BRCA1 and BRCA2, and this highlights the importance of these functions in preventing breast cancer. Further studies need to be done to confirm our results.
| Around one in six women who develop breast cancer has a first degree relative with the condition [1]. In the mid 1990s, a classical linkage approach identified germline mutations in two genes, BRCA1 and BRCA2, which are associated with a high risk of developing both breast and ovarian cancer [2], [3]. Although BRCA1 and BRCA2-specific genetic testing is rapidly evolving in the clinical setting, mutations in these genes are successful at explaining only around half of the dominant multi-case breast cancer only families [4], and their contribution to the heritable risk of breast cancer has been estimated to be no more than around 20% of the total [5], [6]. Importantly, the identification and management of individuals with high-risk breast cancer predisposition gene mutations is now well accepted in clinical practice. Although evidence-based risk management is only possible in a relatively small group of families, as it is limited by the identification of an underlying genetic mutation, the benefits for those individuals are well established [7].
Through a candidate gene approach, mutations in other high and moderate penetrance cancer-susceptibility genes have been identified in a further small proportion of families but the underlying etiology of the increased susceptibility to breast cancer in the majority of multi-case breast cancer families remains unknown. Recent advances in massively parallel sequencing technology have provided an agnostic means by which to efficiently identify germline mutations in individuals with inherited cancer syndromes at the individual family or cancer-specific level [8], [9]. The aim of this study is to identify through a whole exome sequencing approach, the underlying familial predisposition to breast cancer in multiple multi-generational breast cancer families in whom no BRCA1 or BRCA2 mutation was identified (BRCA1/2 negative families), and to assess the candidate genes identified by this means in a cohort of familial BRCA1/2 negative breast and ovarian cancer patients.
We performed intra-family exome sequence analysis of multiple affected relatives from 15 high-risk, trans-generational breast cancer families in whom full BRCA1 and BRCA2 mutation analysis had been performed and was uninformative in at least one breast cancer-affected family member (Table 1). Sequencing was performed on GAIIx or HiSeq instruments (Illumina). The average read depth achieved for target regions was 83.19 and at least 80% (average 89.12%) of the capture target regions were covered by 10 or more sequence reads for all samples (Table S1). Following data filtering, an average of 35 overtly deleterious and 284 non-synonymous mutations were identified per individual (Table S1).
To identify candidate predisposition genes we only considered those with overtly deleterious mutations that were shared by multiple affected relatives and/or were targeted in more than one family and further priority was given to genes with a role in mechanistically well-established breast cancer–associated DNA repair. A list of all overtly deleterious mutations identified in among the 33 individuals sequenced is provided in Table S2. Two of the fifteen families were found to carry independent heterozygous truncating mutations in the Fanconi Anemia (FA) gene, FANCC. Neither family was reported to be of Ashkenazi Jewish ancestry and the mutations are different to those commonly reported among this ethnic group. Family 1 carried a novel nonsense mutation (FANCC c.535C>T, p.Arg179*) that was present in the youngest affected individual (breast cancer at age 37) and in her mother who had ovarian cancer at age 66, but not in her breast cancer-affected sister who was diagnosed at age 46 (Figure 1). Family 2 was found to harbor a known pathogenic FA mutation (FANCC c.553C>T, p.Arg185*) [10] which was present in two sisters who developed breast cancer aged 36, and bilateral breast cancer aged 46 and 53, respectively. A third family analyzed by exome sequencing was found to carry a heterozygous c.1993C>T mutation in the BLM gene which is predicted to truncate the protein at codon 645 (p.Gln645*). This known pathogenic Bloom syndrome mutation [11] co-segregated with cancer in the family (Figure 1), being present in all three sisters diagnosed with breast cancer aged 39, 39 and 41 years respectively and absent in the two unaffected sisters. Although retrospective likelihood segregation analysis of these limited pedigrees did not reach significance (see Text S1), overall, co-segregation of FANCC and BLM mutations in these families appears consistent with that expected for moderately penetrant breast cancer alleles.
Mutation analysis of all coding exons of FANCC and BLM was extended to the index cases from a further 438 BRCA1/2 negative breast cancer families (from kConFab). This approach identified one further family with a heterozygous, known pathogenic FANCC mutation, (c.67delG, p.Asp23Ilefs*23, rs104886459) [12] and one with a heterozygous pathogenic BLM mutation (c.2695C>T, p.Arg899*) [11]. For FANCC, mutation hotspot exons 2, 5, 7, 14 and 15 were screened in the index cases from an additional 957 BRCA1/2 uninformative breast cancer families attending familial cancer services (including 561 obtained from the Peter MacCallum Cancer Centre Familial Cancer Centre and a further 396 from kConFab). One further family with a heterozygous FANCC c.1661T>C (p.Leu554Pro, rs104886458) missense variant, which is a functionally validated pathogenic FA mutation, was identified [13].
The index case in the FANCC c.67delG family developed breast cancer at age 60 but independent clinical testing subsequently identified a deleterious mutation in BRCA2 (c.8297delC, p.Thr2766Asnfs*11) in other breast cancer-affected family members (Figure 1). Genotyping of both mutations within this family suggests that different individuals may carry risk conferred by one or both of these family mutations.
The index case of the FANCC c.1661T>C family developed bilateral breast cancer at age 44 and 55, but DNA from other family members was not available for segregation analysis. All FANCC variants detected in index cases or controls are summarized in Table S3.
The index case of the BLM c.2695C>T family developed breast cancer at age 33 but segregation analysis showed the mutation was inherited from her father rather than her mother whose reported family history of breast cancer had initiated their recruitment into kConFab (Figure 1). Interestingly, breast cancer was diagnosed much earlier in the index case compared to her maternal relatives (33 years versus 58 to 73 years) possibly indicating a different genetic etiology. Unfortunately data regarding family history on the paternal side are limited. Neither the father nor the paternal grandparents were reported to have developed cancer but no further information regarding number or cancer status of other relatives is available. All BLM variants detected in index cases or controls are summarized in Table S4.
No pathogenic BLM mutations were detected in 464 healthy controls and none have been reported in the 1000 Genomes data (20100804 release, n = 1,092) [14] compared to 2/438 breast cancer families with BLM mutations. Likewise, no known pathogenic or overtly deleterious FANCC mutations were identified among the 464 controls or the 1000 Genomes data or among 654 healthy controls examined in an independent study [15]. The Exome Variant Server (EVS), NHLBI Exome Sequencing Project, Seattle, WA, does report deleterious mutations in FANCC and BLM in 3/3,510 and 4/3,510 individuals of European decent, respectively. However, this cohort includes extreme tail sampling of traits relating to heart, lung and blood disorders. The latter group in particular may be expected to show enrichment for mutations in DNA repair machinery including FA genes. Excluding the Exome Variant Server frequency data, a total of 4/1,395 breast cancer families screened for all or at least the mutation hot spot exons carried overtly deleterious FANCC mutations compared to none among the combined control population (n = 2,210). While this is indicative that overtly deleterious mutation in FANCC and BLM are likely to be very rare in the population this must be considered a crude measure as the controls were drawn from diverse populations the majority of which were not matched to the index cases. However, it is possible that more families in our breast cancer family cohort may be explained by FANCC and BLM mutations since, for both genes, private non-synonymous variants were identified that are predicted to be damaging by in silico algorithms. One such variant, for which there was DNA available for segregation analysis, was FANCC p.Arg185Gln. This variant closely segregated with disease in this family, which included four female blood relatives with breast cancers diagnosed at ages 34, 51, 47 and 62 (Figure 1). The p.Arg185Gln variant was identified in 1/1,395 breast cancer families but not in any of 464 controls and has not been reported in the 1000 Genomes project or EVS database.
Homozygous mutations in FANCC and BLM are responsible for FA (complementation group C) and Bloom syndrome, respectively, and individuals diagnosed with these syndromes have a high risk of cancer. Functionally, the FA and Bloom syndrome pathways play important roles in homologous recombination (HR) based repair of double-stranded DNA breaks [16], [17]. Constitutional inactivating mutations in genes integral to error-free HR and responsible for FA have been clearly associated with an increased susceptibility to both breast and ovarian cancer [16], and include the genes BRCA1, BRCA2 (FANCD1), FANCN (PALB2), FANCJ (BRIP1), RAD51C (FANCO) and RAD51D. Thus, in addition to the direct genetic evidence that we have described here, FANCC and BLM are strong candidates for breast cancer susceptibility genes due to their role in the precise regulation of HR and some of its associated functions. Although there is limited data, heterozygous FANCC mutations have previously been linked to an increased incidence of breast and early onset pancreatic cancer [15], [18], [19], however, no excess breast and ovarian cancer was observed among Ashkenazi Jews carrying the FANCC c.711+4A>T mutation [20]. While another previous study failed to identify overtly pathogenic FANCC mutations in breast cancer, the study cohort size was small (n = 88) [21]. In keeping with our data, two recurrent truncating mutations in the BLM gene were shown in a case control study to be associated with increased breast cancer risk in Russia [22]. Gruber et al reported an elevated risk of colorectal cancer in Ashkenazi Jews carrying the common BLMASH mutation and a non-significant excess of breast cancer [23] although a later study failed to confirm these findings [24].
Further to the germline mutations in FANCC and BLM, exome sequencing identified mutations in the breast cancer predisposition genes, PTEN and BRCA2 in an additional three of the original 15 families (Figure S1). The truncating PTEN mutation (c.217G>T, p.Glu73*) was identified in only one branch of the family suggesting another susceptibility gene may explain the extended family history. Prior to this finding, the treating familial cancer centre reported no PTEN-associated clinical features within the family. In family 5, exome sequencing identified a deleterious BRCA2 mutation (c.5722_5723delCT, p.Leu1908Argfs*2, rs80359530) in two of the three family members tested (Figure S1). The mutation is present in a male diagnosed with breast cancer but not in the youngest affected female relative in the family, who had been offered the original clinical BRCA1 and BRCA2 mutation test in the clinic setting. Similarly in family 6, exome sequencing identified a deleterious BRCA2 mutation (c.26delC, p.Pro9Glnfs*16, rs80359343) in a female diagnosed with breast cancer at age 30, but not in her cousin who was diagnosed at age 36 and was the only family member to have undergone full diagnostic BRCA1 and BRCA2 gene sequencing (Figure S1). These families are interesting in a clinical context since they were designated as unresolved on the basis of best clinical practice and demonstrate the need for targeted sequencing of all proven breast and ovarian cancer susceptibility genes to obtain maximum information in the clinical setting (as previously demonstrated [25]). Our data also highlights the major challenge confounding genetic studies of common adult onset familial disease; the presence of ‘phenocopies’ in families with an inherited genetic predisposition and/or the convergence of pedigrees with different genetic causes (e.g. PTEN family 4). Among the remaining nine breast cancer families there were numerous genes that were recurrently targeted that warrant further investigation. It is noteworthy that in one family, one individual harbored a known FA pathogenic truncating mutation in FANCL. Mutation of this gene is responsible for a very small fraction of FA families and only three pathogenic mutations in FANCL are recorded in the Fanconi Anemia Mutation Database.
In conclusion, we describe two potential breast cancer susceptibility genes FANCC and BLM both of which have functional roles in the regulation of HR. The heterozygous mutation carrier rate in Caucasians for these genes is extremely low (for FANCC it is estimated at 1/3,000 [15], whilst the carrier frequency of BLM mutations is unknown since the syndrome is exceedingly rare) and notwithstanding the possibility of the “winners curse” [26], the exome sequencing data is strongly suggestive that FANCC and BLM represent breast cancer predisposing genes. Together with the recently identified association of RAD51 paralogues with cancer predisposition [27], [28], our findings suggest that the number of unidentified moderate to high-risk susceptibility genes is very much larger than previously expected and the number of families explained by each gene is likely to be much less than 1% (cf. RAD51C [27], [29]). Consequently, providing definitive evidence for a causative role for novel breast cancer genes will be challenging and will require validation of rare mutations in thousands rather than hundreds of families. We predict that this will be a generic problem associated with identifying causative mutations in common diseases such as breast cancer and that validation rather than the technical exercise of exome sequencing is where the real challenge lies.
This study was approved by the Peter Mac Ethics Committee (project numbers 09/62 and 11/50). Informed consent was obtained from all participants. Fifteen high-risk breast cancer families with at least four cases of multi-generational breast cancer including at least one additional high-risk feature (such as bilateral, early onset or male breast cancer, or ovarian cancer) and at least two available blood specimens from breast cancer-affected individuals, were selected for whole exome sequencing from among approximately 800 BRCA1 and BRCA2 mutation negative families from the Kathleen Cunningham Foundation Consortium for Research into Familial Breast Cancer (kConFab), which has been collecting biospecimens and clinical and epidemiological information from families recruited through Familial Cancer Centres in Australia and New Zealand since 1997 [30]. DNA from two or three breast cancer-affected individuals were obtained from each family for analysis (as shown in Table 1), at least one of whom had previously been screened for BRCA1 and BRCA2 mutations (by sequencing of all coding exons and Multiplex Ligation-dependent Probe Amplification). Blood DNA from index cases from a further 834 mutation negative kConFab families and 561 mutation negative families obtained from the Peter MacCallum Cancer Centre Familial Cancer Centre were obtained for mutation analysis of candidate genes. Of those index cases obtained through the Familial Cancer Centre, individuals were breast cancer-affected, had a strong family history and been assessed for the probability of harboring a BRCA1 or BRCA2 mutations using BRCAPRO [31] and had been found on the basis of a verified family and personal history of having a 10% or greater probability. The index cases had undergone full diagnostic BRCA1/2 mutation search and no mutation was identified. However, it should be noted that the majority of these families did not fulfill the very stringent family history criteria that was required for recruitment to kConFab, the research cohort from which the families for the initial exome sequencing were taken [30]. Non-cancer control DNA samples were obtained from kConFab (226 age- and ethnicity-matched best friend controls) and from the Princess Anne Hospital, UK (238 Caucasian female volunteers, as described previously [32]). DNA for candidate gene mutation analysis underwent whole genome amplification (WGA) using Repli-G Phi-mediated amplification system (Qiagen) prior to mutation analysis.
2–3 µg of DNA was fragmented to approximately 200 bp by sonication (Covaris) and used to prepare single- or paired-end libraries using the SPRIworks Fragment Library System I for Illumina Genome Analyzer on the SPRI-TE Nucleic Acid Extractor (Beckman Coulter). Exome enrichment was performed using the NimbleGen Sequence Capture 2.1 M Exome Array, EZ Exome Library (Roche NimbleGen) or SureSelect Human All Exon version 2 or 50 Mb libraries (Agilent Technologies) according to the recommended protocols. Sequencing was performed on GAIIx or HiSeq instruments (Illumina). Library preparation and sequencing details for each sample are provided in Table S1. We did not observe any significant differences in performance of the different exome capture platforms.
Paired-end sequence reads were aligned to the human genome (hg19 assembly) using the Burrows–Wheeler Aligner (BWA) program [33]. Local realignment around indels was performed using the Genome Analysis Tool Kit (GATK) software [34]. Subsequently, duplicate reads were removed using Picard and base quality score recalibration performed using GATK software. Single nucleotide variants (SNVs) and indels were identified using the GATK Unified Genotyper and variant quality score recalibration. Variants were annotated with information from Ensembl release 62 using Ensembl Perl Application Program Interface (API) including SNP Effect Predictor [35], [36]. Single-end sequence reads were aligned as above except duplicate reads were flagged prior to base quality score recalibration and included in variant calling.
Variants were first filtered for confident calls originating from bidirectional sequence reads using a quality threshold of ≥30, read depth of ≥10 and allele frequency ≥0.15. Prior to further filtering, variants were assessed for overtly deleterious mutation in known breast cancer associated genes [25]. Then, all variants present in the dbSNP database v132, except those also reported in the public version of the Human Gene Mutation Database (HGMD) [37] were removed, as were all common variants detected in >10 out of 33 exomes. Next, variants with functionally deleterious consequences (nonsense SNVs, frameshift indels, essential splice variants and complex indels) were identified for evaluation [35]. Functionally deleterious variants were evaluated in each individual as well as pairwise between relatives.
Primers flanking the BRCA2, PTEN, FANCC and BLM mutations identified by whole exome sequence analysis were used to amplify germline DNA from affected index cases and all available relatives. The purified products were directly sequenced using BigDye terminator v3.1 chemistry on a 3130 Genetic Analyzer (Applied Biosystems).
High resolution melt (HRM) analysis was performed on duplicate PCR products amplified from 15 ng WGA DNA. Primer sequences and PCR conditions are provided in Table S5. Melt analyses were performed on a LightCycler 480 Instrument using Gene Scanning Software (Roche). Duplicate PCR products exhibiting variant DNA melt curves were Sanger sequenced to identify sequence variations. All novel sequence variants were confirmed by Sanger sequencing an independent PCR amplified from non-WGA DNA. The functional effect of missense variants were evaluated using in silico prediction tools SIFT and PolyPhen-2 [38], [39].
The following GenBank reference sequences were used for variant annotation: FANCC, NM_000136 BLM, NM_000057; PTEN, NM_000314 and BRCA2, NM_000059.
1000 Genomes Browser, http://browser.1000genomes.org/; Ensembl, http://www.ensembl.org/index.html; The Genome Analysis Toolkit, http://www.broadinstitute.org/gsa/wiki/index.php/The_Genome_Analysis_Toolkit; HGMD, http://www.hgmd.org/; Picard, http://picard.sourceforge.net; HGVS nomenclature for the description of sequence variants, http://www.hgvs.org/mutnomen/; NCBI SNP database, http://www.ncbi.nlm.nih.gov/projects/SNP/; The Fanconi Anemia Mutation Database, http://www.rockefeller.edu/fanconi/; BLMbase mutation registry, http://bioinf.uta.fi/BLMbase/; SIFT, http://sift.jcvi.org/; PolyPhen-2, http://genetics.bwh.harvard.edu/pph2/. Exome Variant Server, http://evs.gs.washington.edu/EVS/.
|
10.1371/journal.ppat.1001094 | A Novel Family of Toxoplasma IMC Proteins Displays a Hierarchical Organization and Functions in Coordinating Parasite Division | Apicomplexans employ a peripheral membrane system called the inner membrane complex (IMC) for critical processes such as host cell invasion and daughter cell formation. We have identified a family of proteins that define novel sub-compartments of the Toxoplasma gondii IMC. These IMC Sub-compartment Proteins, ISP1, 2 and 3, are conserved throughout the Apicomplexa, but do not appear to be present outside the phylum. ISP1 localizes to the apical cap portion of the IMC, while ISP2 localizes to a central IMC region and ISP3 localizes to a central plus basal region of the complex. Targeting of all three ISPs is dependent upon N-terminal residues predicted for coordinated myristoylation and palmitoylation. Surprisingly, we show that disruption of ISP1 results in a dramatic relocalization of ISP2 and ISP3 to the apical cap. Although the N-terminal region of ISP1 is necessary and sufficient for apical cap targeting, exclusion of other family members requires the remaining C-terminal region of the protein. This gate-keeping function of ISP1 reveals an unprecedented mechanism of interactive and hierarchical targeting of proteins to establish these unique sub-compartments in the Toxoplasma IMC. Finally, we show that loss of ISP2 results in severe defects in daughter cell formation during endodyogeny, indicating a role for the ISP proteins in coordinating this unique process of Toxoplasma replication.
| Apicomplexans are the cause of important diseases in humans and animals including malaria (Plasmodium falciparum), which claims over a million human lives each year, and toxoplasmosis (Toxoplasma gondii), which causes birth defects and neurological disorders. These parasites possess a unique cortical system of membrane sacs arranged on a cytoskeletal meshwork, together referred to as the inner membrane complex (IMC). The IMC is the anchor point for the gliding motility machinery necessary for host invasion and also a scaffold around which new parasites are constructed during replication. Here we have uncovered new insights into the organization and function of this structure by identifying and characterizing ISP1-3, a family of proteins that define novel sub-compartments within the Toxoplasma IMC. Residues predicted for myristoylation and palmitoylation are critical in the membrane targeting of these proteins, suggesting that multiple palmitoyl acyltransferase activities reside within the IMC and dictate its organization. Surprisingly, ISP1 is required for proper sub-compartment sorting of ISP2 and 3, revealing a novel hierarchical targeting mechanism for the organization of this membrane system. Disruption of ISP2 results in defects during endodyogeny and a dramatic loss in parasite fitness, revealing that the ISP proteins play an important role in coordinating parasite replication.
| The phylum Apicomplexa contains numerous obligate intracellular pathogens that are the cause of serious disease in humans and animals, greatly influencing global health and causing significant economic loss worldwide. The phylum includes Plasmodium falciparum, the causative agent of malaria which claims 1–2 million human lives annually, and Toxoplasma gondii, a pathogen that infects more than thirty percent of the world's population and causes severe neurological disorders and death in immunocompromised individuals [1]. Most of the drugs used to treat apicomplexans target metabolic pathways or the chloroplast-derived apicoplast [2], [3], [4], but these parasites also possess elaborate and unique structures that are required for replication and invasion and thus represent attractive new targets for therapeutic intervention.
Apicomplexans are grouped with dinoflagellates and ciliates in the alveolata infrakingdom [5]. The unifying morphological characteristic of this group is the presence of alveoli: membrane sacs located beneath the plasma membrane. Molecular phylogenetic data supports this grouping, as does the identification of a conserved family of articulin-like membrane skeleton proteins, the alveolins, which associate with alveoli in all three phyla [6], [7]. While the presence of alveoli is conserved, each of these groups has adapted this peripheral membrane structure for different cellular functions to fit their distinct niches. In dinoflagellates, the alveoli sometimes contain cellulose-based plates that function as protective armor [8]. In contrast, ciliate alveoli are calcium storage devices thought to play roles in regulation of cilia, exocytosis from cortical organelles known as extrusomes, and control of cytoskeletal elements [9], [10], [11].
In apicomplexans, the alveoli in conjunction with an underlying filamentous network are termed the inner membrane complex (IMC) [12], [13]. Flattened alveoli underlie the entirety of the plasma membrane except for a small gap at the apex and base of the cell [14]. These cisternae are organized into a patchwork of rectangular plates capped by a single cone-shaped plate at the apex of the cell. Freeze-fracture studies of the IMC plates expose a lattice of intramembranous particles (IMPs), an arrangement that suggests an association with proteins of the underlying filamentous network and subtending cortical microtubules [15], [16], [17]. Together, these features of the IMC are the foundation for a unique form of gliding motility used for host cell invasion and also serve as the scaffold for daughter cell formation during division [18], [19].
Toxoplasma tachyzoites replicate by endodyogeny, a process of internal cell budding that produces two daughters within an intact mother parasite. Following centriole duplication, daughter cell formation begins with the concurrent assembly of an apical and basal complex [20]. Although these two structures consist of cytoskeletal components that will eventually cap opposite ends of the mature parasite, they are initiated in close spatial and temporal proximity. IMC construction then proceeds by the extension of the basal complex away from the daughter apical complex, generating a bud into which replicated organelles are packaged. Parasite division is completed by a number of maturation steps terminating with the adoption of the maternal plasma membrane [21].
The apical, cone-shaped cisterna is unique in form and presumably the earliest membrane component deposited into the nascent IMC [19]. A number of cytoskeletal IMC markers localize to a region at the parasite apex thought to correspond to this apical-most IMC plate. A GFP fusion of the dynein light chain, TgDLC, can be detected in an apical cap region but predominantly localizes to the conoid and is also found in the basal complex, spindle poles and centrioles. TgCentrin2, the most divergent of the three Toxoplasma centrin homologues, labels the preconoidal rings and a peripheral ring of ∼6 annuli located at the lower boundary of the TgDLC cap. It has been suggested that these annuli lie at the juncture between the apical cap plate and the flanking set of IMC plates [20]. Additionally, PhIL1, a cytoskeletal IMC protein of unknown function, is detected throughout the IMC but strongly enriched in the apical cap and basal complex [22]. Only a few proteins are known to directly associate with the IMC membranes. These include a number of proteins associated with gliding motility [23], [24], [25], as well as the heat shock protein Hsp20 [26] and one isoform of the purine salvage enzyme hypoxanthine-xanthine-guanine phosphoribosyltransferase [27]. Thus, despite the central role of this conserved membrane system in apicomplexan biology, little is known of its composition, organization, and construction.
We present here a family of proteins unique to the Apicomplexa that localize to three distinct sub-compartments of the Toxoplasma IMC. ISP1 localizes to a region corresponding to the apical cap, ISP2 occupies a central IMC region, and ISP3 resides in both the central IMC region and a basal IMC compartment. ISP1 and 3 are early markers for bud formation and label previously unobserved daughter IMC structures in the absence of parasite cortical microtubules, indicating that microtubules are not required for initial assembly of IMC membranes. We show that the ISPs are initially targeted to the IMC by conserved residues predicted for coordinated myristoylation and palmitoylation in the extreme N-terminus of each of these proteins. Interestingly, deletion of ISP1 results in the relocalization of ISP2 and 3 to the apical cap, demonstrating an interactive, hierarchical targeting among this family of proteins to these distinct sub-compartments of the IMC. Finally, disruption of ISP2 results in a severe loss of parasite fitness and dramatic defects in daughter cell formation. Although the ISP2 knockout parasites ultimately compensate for these defects, this data shows an important role for these proteins in the coordination of daughter cell assembly.
We previously generated a panel of monoclonal antibodies against a mixed fraction of T. gondii organelles [28]. One of the antibodies, 7E8, stains a cone-shaped structure at the periphery of the apical end of the parasite (Figure 1A). This staining pattern extends from a gap at the extreme apex (Figure 1A, arrow) ∼1.5 µm along the length of the parasite, a localization suggestive of the apical IMC plate observed by electron microscopy [14]. Colocalization with TgCentrin2 shows that 7E8 staining is delimited at its apex and base by this apical cap marker, indicating that 7E8 does indeed detect a protein associated with the anterior-most IMC plate (Figure 1D).
During early endodyogeny, 7E8 staining is visible in daughter parasites as a pair of small rings within each mother parasite (Figure 1B, arrows). As daughter formation proceeds, this structure enlarges and extends to form the apical cap seen in mature tachyzoites (Figure 1C). The association with forming daughter scaffolds together with the extreme apical gap further suggests that 7E8 labels the apical sub-compartment of the IMC. We also frequently observe 7E8 staining a single dot near the basal border of the cone (Figure 1C, arrow) which is distinct from TgCentrin2 annuli (Figure 1D, inset).
Western blot analysis of Toxoplasma lysates with mAb 7E8 revealed a single band at ∼18 kDa (Figure 1E). We used the 7E8 antibody to isolate its target protein by immunoaffinity chromatography. The isolated protein was separated by SDS-PAGE (Figure 1F), digested with trypsin, and seven peptides were identified by mass spectrometry corresponding to the hypothetical T. gondii protein TGGT1_009340 (Figure 1G). EST and cDNA sequencing confirmed that the gene model is correct. Due to its unique localization, we named this protein IMC Sub-compartment Protein 1 (ISP1).
Examination of the 176 amino acid sequence of ISP1 reveals that it contains a high number of charged residues (∼30%). While there are a relatively large number of ESTs encoding ISP1, the protein lacks conserved domains that could suggest its function. The protein contains a glycine at position two, which is predicted to be myristoylated [29] as well as a pair of cysteines at positions seven and eight strongly predicted to be palmitoylated [30]. Since ISP1 lacks a predicted signal peptide or transmembrane domain, these residues suggested a mechanism for IMC membrane association. BLAST analysis of the ISP1 sequence revealed orthologues across the apicomplexan phylum, including Neospora, Theileria, Cryptosporidia, Babesia, and Plasmodium (Figure S1). Orthologues were also found in Eimeria by BLAST against EST libraries (data not shown). ISP1 also showed significant homology in its C-terminal region to CP15/60, a poorly characterized putative surface glycoprotein in Cryptosporidia [31], [32]. No ISP1 orthologues were identified outside of the phylum indicating that this protein is restricted to the Apicomplexa.
BLAST analysis of the T. gondii genome using the ISP1 sequence identified two additional hypothetical proteins with considerable sequence similarity to ISP1, which we named ISP2 (TGGT1_058450) and ISP3 (TGGT1_094350) (Figure 2A). The greatest degree of sequence similarity between these three proteins exists within the C-terminal two-thirds of their sequences. The N-terminal regions of the proteins are more divergent, but each contain a conserved glycine at position two as well as a pair of conserved cysteines predicted to be myristoylated and palmitoylated, respectively (Figure 2A, boxed residues). ISP2 additionally contains a third cysteine at position five predicted to be palmitoylated. Similar to ISP1, these proteins are highly charged and have a relatively large number of corresponding ESTs. OrthoMCL analysis of the ISPs indicates two ortholog groups within Apicomplexa. ISP1 and ISP2 segregate with one group while ISP3 segregates with another (Figure S1). The Toxoplasma genome may encode a fourth ISP family member (TGGT1_063420), although it does not segregate with any OrthoMCL group. This predicted protein lacks the conserved glycine and cysteine residues present in the N-termini of other ISP proteins. Only a single EST is present for TGGT1_063420, indicating that it is poorly expressed relative to the other ISPs, and thus it was not investigated further.
To localize ISP2 and ISP3 in T. gondii, we expressed each gene under the control of its endogenous promoter with a C-terminal HA epitope tag. Intriguingly, ISP2 localizes to a previously unrecognized central sub-compartment of the IMC, which begins at the base of the ISP1 apical cap and extends approximately two-thirds the length of the cell. The apical boundary of this compartment is delineated by the TgCentrin2 annuli (Figure 2B). The posterior boundary has a jagged edge suggesting it corresponds to discrete IMC plates (Figure 2D, arrows). While the ISP2 signal terminates near the end of the subpellicular microtubules, the termini for these two structures are not identical (Figure S3, WT). Antisera raised against recombinant ISP2 confirmed this central IMC sub-compartment localization, ensuring that exclusion of ISP2 from the apical cap and basal IMC is not an artifact of epitope tagging (Figure S2A).
Similar to ISP2, ISP3 stains the central section of the IMC. However, ISP3 staining extends to the posterior end of the complex, identifying a third sub-compartment of the IMC (Figure 2C). A small gap in ISP3 staining is observed in the posterior region similar to that seen for other IMC proteins [23]. Antisera raised against recombinant ISP3 gave a poor signal by IFA, but was sufficient to confirm localization to both the IMC central and basal sub-compartments (Figure S2B). As with ISP1, ISP2 and ISP3 are visible in forming daughter parasites. Whereas the maternal signals of ISP1 and ISP2 appear to remain stable throughout endodyogeny, the maternal ISP3 signal rapidly attenuates with the onset of endodyogeny while it concentrates in daughters (Figure 2E). Attenuation of ISP3 in mothers and enrichment in daughters was also observed with our polyclonal antibody, indicating this is not the result of a C-terminal processing event that removes the HA epitope tag (Figure S2C). Thus, ISP3 provides an excellent marker for bud initiation, growth, and maturation during endodyogeny (Figure 2E and Video S1).
The observations that the ISPs are visible at the periphery of forming daughters prior to adoption of the maternal plasma membrane and that gaps are present at the extreme apex and base suggests an association with the IMC. To confirm IMC association, we treated extracellular parasites with Clostridium septicum alpha-toxin. This vacuolating toxin causes a dramatic separation of the plasma membrane and the underlying IMC, enabling differential localization of these closely apposed membrane systems [33]. In toxin-treated parasites, the ISP proteins segregate with the IMC and not with the plasma membrane, confirming that the ISPs are indeed IMC proteins (Figure 3A–B).
To ascertain if the ISPs are embedded in the IMC protein meshwork that includes the articulin-like protein IMC1, we performed detergent extractions of extracellular parasites in 0.5% NP-40. In these conditions, each ISP was solubilized similar to the control protein ROP1, while IMC1 remained in the insoluble pellet fraction (Figure 3C). This extraction profile demonstrates that the ISPs are not embedded in the detergent resistant protein meshwork that underlies the IMC membranes.
We disrupted microtubules in intracellular parasites to assess whether the underlying microtubules influence ISP localization. Apicomplexan microtubules are selectively susceptible to disruption by dinitroanilines, such as oryzalin [34]. After 40 hours of 2.5 µM oryzalin treatment, all tubulin is unpolymerized and dispersed. Without spindle microtubules (mitosis) and subpellicular microtubules (budding), productive daughter formation repeatedly fails resulting in an undivided, amorphous mother cell with a polyploid DNA content [35] (Figure 4A). Intriguingly, we observe ISP1 labeling numerous small rings that are centrally located within oryzalin-treated parasites (Figure 4A, inset) of approximately the same dimensions as ISP1 early daughter buds in untreated, replicating parasites (compare with Figure 1B, arrows). Since polymerization of subpellicular microtubules is essential to drive bud extension, these rings likely represent failed attempts to build new daughter buds [36]. A larger peripheral patch of ISP1 with a central hole is also observed, likely representing the original parent apical cap (Figure 4A, arrows). While ISP2 was not observable in these early bud rings (Figure 4B), we did detect ISP3 in these structures within oryzalin-treated parasites (Figure 4C, inset arrows), suggesting that both the apical cap and remaining IMC sub-domains are formed independently of microtubules at a very early stage of bud development. While membrane skeleton proteins are likely candidates for providing the foundation for these structures, we were unable to detect the articulin-like protein IMC1 in these early bud rings, even at lower oryzalin concentrations (0.5 µM) that only disrupt cortical microtubules (Figure 4D and Video S2).
The greatest sequence similarity within the ISP family is present in the C-terminal two-thirds of the proteins while the N-terminal region is more divergent (Figure 2A), thus we reasoned that the unique targeting of each ISP family member might be controlled by its N-terminal region. To test if the N-terminal region of ISP1 is necessary for targeting, we eliminated the first 63 residues to create a truncated protein fused to YFP. ISP164–176-YFP does not target to the IMC but is instead distributed throughout the cytoplasm and nucleus, showing that this N-terminal region is necessary for apical cap targeting (Figure 5A). To determine if the ISP1 N-terminal region is sufficient for targeting, we fused the first 65 residues of ISP1 (containing the putative acylation sequence and divergent N-terminal region) to YFP and expressed this construct in Toxoplasma. The ISP11–65-YFP fusion traffics to the apical cap in an identical fashion to endogenous ISP1 (Figure 5B). To further narrow the N-terminal region required for apical cap targeting, we generated an additional fusion of the first 29 residues of ISP1 (containing the putative acylation sequence) to YFP. This fusion also traffics in a manner identical to full length ISP1 (Figure 5C), demonstrating that this N-terminal domain is both necessary and sufficient for apical cap targeting.
To assess targeting of ISP2 and ISP3, we also created fusions of their N-terminal regions (residues 1–41 and 1–36 respectively) to YFP. The ISP31–36-YFP fusion targets to the central and basal sub-compartments of the IMC but is restricted from the apical cap (Figure 5D), showing that this region is sufficient for proper sub-compartment targeting. In contrast, ISP21–41-YFP localized to the entire IMC, overlapping with endogenous ISP1 in the apical cap and extending into the basal IMC sub-compartment (data not shown). To ensure this change in targeting for ISP21–41 was not an artifact of the YFP fusion, we replaced YFP with an HA tag (shown to have no effect on the targeting of full length ISP2, Figure 5E). The ISP21–41-HA protein also localized throughout the IMC (Figure 5F), demonstrating that the N-terminal domain of ISP2 is sufficient for targeting to the IMC, but not for correct sub-compartment localization.
Protein myristoylation occurs co-translationally through the action of an N-myristoyl transferase [37]. This modification is sufficient to promote transient association with membranes for otherwise cytosolic proteins. This weak membrane affinity can then be stabilized by addition of one or more palmitoylations through the action of a palmitoyl acyltransferase (PAT), effectively locking a protein into a target membrane system in a mechanism known as “kinetic trapping”. The ISPs each contain a second position glycine followed by cysteines within the first 10 residues that are predicted to be myristoylated and palmitoylated, respectively (Figure 2, boxed residues). We mutated the glycine and cysteine residues in HA epitope tagged ISP constructs to examine their effect on targeting. As predicted by the kinetic trapping model, mutation of the second position glycine to an alanine abolished IMC targeting in each family member (Figure 6 and Figures S3 and S4, G2A), resulting in proteins distributed throughout the cytoplasm. Mutation of the cysteine residues to serine was performed individually and together. While only minor defects in targeting were observed with individual cysteine mutations, mutation of both cysteines abolished ISP1 and ISP3 targeting (Figure 6 and Figure S4). In the case of ISP2, targeting was only abolished when all three cysteines were coordinately mutated (Figure S3). While coordinated cysteine mutants of the ISPs are distributed in the cytoplasm similar to G2A mutants, we also often observed perinuclear staining that is especially concentrated just apical of the nucleus (arrows, Figure 6 and Figure S3 and S4). Presumably, myristoylation of these proteins still occurs, but without palmitoylation, these mutants are left to transiently sample the different membrane systems within the cell and therefore may appear concentrated as they associate with the ER and Golgi membranes present in this region. These results demonstrate that these residues are essential to ISP sorting and indicate that coordinated acylation of the ISPs is responsible for IMC membrane targeting.
To assess the function of ISP1, we disrupted the ISP1 gene by homologous recombination (Figure 7A). We identified clones which lacked ISP1 expression by IFA and Western blot (Figure 7B–C), indicating successful disruption of the ISP1 locus and demonstrating that ISP1 is not necessary for in vitro propagation of T. gondii. Disruption of ISP1 did not result in any gross defect in parasite growth. However, we were surprised to find that both ISP2 and ISP3 were relocalized in the Δisp1 strain. In the parental strain, ISP2 staining terminates sharply at the ring of TgCentrin2 annuli bordering the base of the apical cap (Figure 7D, arrowheads). However, in Δisp1 parasites, ISP2 staining extends past this border, relocalizing to the apical cap sub-compartment of the IMC (Figure 7D). Apical cap relocalization is also observed for ISP3 in the Δisp1 strain (Figure 7E). To ensure the ISP2 and ISP3 relocalization to the apical cap is truly a result of the absence of ISP1, we reintroduced the ISP1 gene with a C-terminal YFP fusion into the Δisp1 strain. This fusion protein targets correctly to the apical cap and, importantly, reestablishes the wild-type localization of ISP2 (Figure 8A, insets) and ISP3 (data not shown), excluding them from the apical cap. Thus, ISP1 exhibits a gate-keeping effect on ISP2 and 3, preventing access to the apical cap and establishing a hierarchy of protein targeting among these IMC sub-compartments. To determine if ISP1 performs a broader scaffolding function within the apical cap, we evaluated the localization of TgDLC1 using a GFP fusion; however, we observed no change in the localization of this protein in the absence of ISP1 (data not shown).
Given the ability of ISP1 to exclude other family members from the apical cap, we exploited our ISP11–65-YFP construct to determine whether or not the N-terminal region that is sufficient for apical cap targeting also plays a role in exclusion from this compartment. Expression of this construct in Δisp1 parasites does not result in exclusion of ISP2 (Figure 8B) or ISP3 (data not shown) from the apical cap, demonstrating that distal sequences present in the more conserved regions of ISP1 (residues 66–176) are necessary for exclusion. To further assess whether the C-terminal region from another ISP family member could substitute for the ISP1 C-terminal domain and function in exclusion, we constructed a hybrid protein containing the N-terminal 65 amino acids of ISP1 and the C-terminal region of ISP2 (residues 43–160) fused to YFP. Similar to the ISP11–65-YFP construct, the ISP1N/2C-YFP chimera targets to the apical cap but does not exclude ISP2 (Figure 8C) or ISP3 (data not shown). These results demonstrate that the exclusion activity of the C-terminal region of ISP1 is specific to this family member and cannot be replaced by the complementary region from ISP2.
We created an additional chimera consisting of the N-terminal region of ISP2 (residues 1–41) fused to the C-terminal region of ISP1 (residues 67–176). While the N-terminal region of ISP2 alone targets YFP or HA throughout the IMC (Figure 5F), inclusion of the C-terminal region of ISP1 restricts the localization to the apical cap and central regions of the IMC (Figure 8D, see discussion). In parasites expressing this chimera, ISP2 and 3 are mostly relocalized into the base portion of the IMC (Figure 8E–F, brackets). The fact that the ISP1 C-terminal region is able to exhibit exclusion activity against the other ISPs when artificially targeted to other domains of the IMC strengthens the conclusion that the ISP1 C-terminal region constitutes an ISP exclusion domain.
To further investigate the function of the ISP proteins, we disrupted the genes encoding ISP2 and ISP3 by homologous recombination. To accomplish this, we employed a recently developed Δku80 parasite strain that is highly efficient at homologous recombination [38]. We first removed HPT from the Ku80 locus by homologous recombination and negative selection using 6-thioxanthine, creating Δku80Δhpt strain parasites. We then used this strain to disrupt ISP2 or ISP3 and confirmed these deletions by IFA (not shown) and Western blot (Figure 9A and Figure S5).
In contrast to our findings for Δisp1 parasites, localization of other ISP family members was unchanged in both Δisp2 and Δisp3 strains (data not shown). While no gross phenotype was seen in Δisp3 parasites, the Δisp2 strain parasites were obviously defective in growth as the knockout was rapidly lost from transfected populations and its isolation required cloning early following transfection. To assess this loss in fitness, we performed competition growth assays between parent and Δisp2 parasites by mixing these strains in culture and monitoring the culture composition at each passage. The parental strain rapidly out competed the Δisp2 parasites, confirming a severe fitness loss in these parasites (Figure 9B). Further analysis by IFA revealed that Δisp2 parasites display a number of defects in parasite division. Most frequently, we observed the construction of >2 daughters per mother cell in each round of endodyogeny with some parasites assembling as many as 8 daughters (Figure 9C). To quantify this defect, we stained for ISP1, an early marker for bud formation during endodyogeny, and counted vacuoles containing parasites undergoing endodyogeny and assembling >2 buds. As expected, we saw a dramatic increase in the number of parasites producing more than two daughters in the Δisp2 strain (Figure 9D). Neither Δisp1 or Δisp3 parasites showed any aberration in daughter cell assembly compared to wild-type parasites (data not shown).
Assembly of >2 daughters in Δisp2 parasites sometimes occurred around a single polyploid nucleus with karyokinesis accompanying budding (bottom left parasite, Figure 9C) while other parasites assembled the spindle apparatus and underwent karyokinesis without budding, resulting in a mother parasite with two nuclei (Figure 9E). We also observe parasites containing two discrete nuclei in the process of budding >2 daughters (outlined parasites, Figure 9F).
Less frequently, we observed a catastrophic failure of Δisp2 parasites to appropriately segregate nuclei, resulting in anucleate zoids and nuclei extruded in the vacuole (Figure 9G). These vacuoles also show major defects in apicoplast segregation with a few cells receiving both a nucleus and an apicoplast while some received only an apicoplast and others received neither. Finally, some vacuoles with nuclear segregation defects contained many immature buds within the vacuole (Figure 9H). These buds appear to be outside of any intact parasite and it is unclear if they were initiated within a mother cell and then somehow liberated into the vacuolar space or if they were the result of a budding event that was initiated within the vacuolar space itself. In these vacuoles, several elongated apicoplasts are strung throughout the vacuolar space, associated with the extracellular buds and nuclei.
Surprisingly, the Δisp2 parasites recovered from both the fitness and replication defects after approximately two months of culture (data not shown), preventing complementation by genetic rescue. To ensure these phenotypes are specific to the disruption of ISP2 and not the consequence of any off target effects, we generated a second independent Δisp2 line. This line displayed the same loss of fitness and cell division defects, indicating these phenotypes are specifically linked to disruption of the ISP2 locus (data not shown).
Alveoli are the unifying morphological feature among ciliates, dinoflagellates and apicomplexans where these unique membrane stacks have been adapted to suit these divergent organisms in vastly different niches. In apicomplexans, the membrane stacks (the IMC) have been exploited to provide unique and critical roles in parasite replication, motility and invasion. Freeze-fracture studies reveal a highly sophisticated arrangement of IMC plates with dissimilar organization of IMPs in the apical versus lower plates indicating compositional differences between these regions [14]. Identification of the ISPs clearly demonstrates that the protein constitution of the membrane cisternae is not uniform. The ISP compartments have sharp boundaries (Figure 2B–D), suggesting that they correspond to discrete cisterna or groups thereof (Figure 10A).
ISP1 localizes to the apical cap compartment that is delimited by TgCentrin2 and thus represents the first membrane associated protein of this apical-most IMC plate. Previously, the cytoskeleton-associated proteins PhIL1 and TgDLC1 were shown to localize in part to the apical cap region [20], [22]. The C-terminal half of PhIL1 is sufficient for apical cap localization and also for retaining cytoskeletal association. This portion of the protein lacks predicted transmembrane domains or acylation signals, indicating that it links directly to a sub-domain of the cytoskeleton independent of the membrane stacks. Electron micrographs of detergent-extracted parasites show substantial differences in the cytoskeletal filaments in this region (e.g. thicker filaments and a parallel instead of interwoven arrangement), indicating that distinct sub-domains exist in both the IMC membranes and underlying network [13].
Localization of ISP2 and 3 revealed two additional sub-compartments of the IMC that have not been previously observed: a central compartment labeled by ISP2 and a basal compartment labeled by ISP3. The abutment of ISP2 and ISP3 staining against the posterior end of the apical cap likely corresponds to the junction between the apical cap and the rectangular plates constituting the remainder of the IMC. The presence of TgCentrin2 annuli at this border is striking as centrins are calcium-binding contractile proteins known to play a role in the duplication of microtubule organizing centers [39]. While the ISP3 sub-compartment clearly terminates at the posterior end of the IMC, it is unclear what accounts for the basal boundary of the ISP2 sub-compartment which lies approximately two-thirds down the length of the parasite. One possibility is an association with the cortical microtubules that also terminate in this region [40]. However, the microtubules and ISP2 signal do not consistently terminate at the same point. Alternatively, the signal termination may correspond to another junction of IMC plates and the exclusion of ISP2 from the basal region of the IMC may reflect another point of hierarchical targeting, as we discovered for ISP1 in the apical cap.
While ISP1 and 2 are both retained in mother parasites during endodyogeny, ISP3 maternal staining dissipates as daughter parasites form. The strong ISP3 signal in early buds along with the rapid attenuation of ISP3 signal in the mother during endodyogeny provides an unhampered view of the membranes of the daughter buds (Figure 2E and Video S1). Expression of IMC proteins is tightly regulated during the cell cycle including the ISPs, which show an expression profile similar to that of IMC1 (Michael White, personal communication). Thus, the bright ISP3 staining in daughters and concomitant loss of signal in mother cells could be due to synthesis in daughters and degradation in mothers. Alternatively, since palmitoylation is a reversible lipid modification, recycling by de-palmitoylation at the parent IMC and re-palmitoylation at daughter IMCs could account for the ISP3 dynamics observed.
ISP1 and 3 are localized to numerous ring structures in oryzalin-treated parasites, indicating that initiation of bud IMC assembly repetitively occurs under these conditions and is not dependent on microtubules. Microtubule polymerization is essential for cell division and cortical microtubule extension is thought to drive bud growth, explaining why buds in parasites lacking microtubules never elongate [36]. ISP1 and 3 are localized to distinct compartments in forming daughter cells, demonstrating that IMC sub-compartmentalization is established early during endodyogeny (Video S1). The ISP1 and 3 signals are not always perfectly overlapping in oryzalin-treated cells, suggesting that IMC membrane specialization may be established even in these early bud rings, although the rings are too small to clearly visualize distinct sub-domains. The absence of ISP2 from these rings may indicate later recruitment to daughter buds or simply be a consequence of drastic perturbation of the cell under these conditions.
Some nucleating scaffold element must provide a foundation for these early IMC membrane bud rings. The earliest signs of daughter bud formation observed by electron microscopy are a dome-shaped vesicle and associated microtubules [41]. The basal complex protein TgMORN1 is the earliest protein marker of bud generation, forming a pair of rings around the centrioles after their duplication at approximately the same time daughter conoids are assembled [42]. In oryzalin-treated parasites observed during the first few hours following drug addition, initial TgMORN1 ring formation still occurs and can be followed until cells attempt to bud, at which point the inability to polymerize new microtubules results in drastic loss of parasite morphology. After 24 hours of oryzalin treatment, TgMORN1 localizes in patches sparsely associated with peripheral sheets of IMC membrane skeleton marker IMC1 but does not label anything resembling the bud rings observed for ISP1 and 3 [20], [43]. In our study, IMC1 did not localize to ISP1-labeled bud rings in oryzalin-treated parasites, demonstrating that it is not required for bud initiation. Furthermore, TgMORN1 has been disrupted and shown to be non-essential for parasite growth [44]. Future studies with ISP1 and 3 will enable the discovery of the critical nucleating factors that mediate bud initiation.
Protein acylation is a widely employed eukaryotic mechanism to mediate membrane association of proteins that lack a transmembrane domain. Our mutation of conserved N-terminal residues that are predicted to be myristoylated and palmitoylated indicates that these modifications are responsible for IMC membrane targeting. These mutagenesis studies also agree with our deletion analysis demonstrating that the N-terminal regions of ISP1 and 3 are sufficient for correct targeting. Together, these data suggest a kinetic trapping model for ISP localization in which ISP proteins are first co-translationally myristoylated in the cytosol enabling sampling of membranes, then recognized and palmitoylated by a unique PAT (or PAT activity) that is present in each sub-compartment, thus locking the protein into the appropriate membrane sub-compartment (Figure 10B).
For ISP1 and 3, this multiple PAT model agrees with our deletion analysis showing that N-terminal regions of the proteins are sufficient for sub-compartment localization. Recognition of each ISP protein as a substrate would be determined by the context of the sequences immediately surrounding the residues required for myristoylation and palmitoylation. Indeed, additional deletion analysis showed that the first ten residues of ISP1 mostly retain apical cap targeting (data not shown). In contrast, while the N-terminal region of ISP2 is sufficient for general IMC membrane association, deletion of the C-terminal region or its substitution in the ISP2N/1C chimera alters sub-compartment specificity. These structural changes to ISP2 may remove important information for establishing stringent PAT specificity, permitting incorporation into other IMC sub-compartments. A similar effect was recently discovered for the palmitoylated protein Vac8 in Saccharomyces cerevisiae. While palmitoylation of wild-type Vac8 was only catalyzed by one of the five S. cerevisiae PATs tested, truncation of the Vac8 C-terminus resulted in its palmitoylation by all five PATs [45]. Alternatively, it is possible that palmitoylation of the ISP family is facilitated by a single PAT that is localized throughout all three IMC compartments and regulated by additional cofactors. Modulation of PAT activity against certain substrates by additional protein cofactors has been shown in both yeast and mammalian systems [46], [47].
The presence of a Asp-His-His-Cys-cysteine-rich domain (DHHC-CRD) is the hallmark of PAT activity and has allowed for the identification of several PATs in other systems, including 7 in S. cerevisiae and 23 in mammalian genomes [48]. Within the Toxoplasma genome, 18 DHHC-CRD containing proteins are predicted to be encoded, a relatively higher number among protists (e.g. the Giardia lamblia and Trypanosoma brucei genomes are predicted to contain 9 and 12 PATs, respectively [49], [50]), indicating a more extensive PAT network may be present to accommodate protein sorting within the numerous unique membrane systems in apicomplexans. Future work localizing and characterizing the putative Toxoplasma PATs will distinguish between the possible models for ISP sorting suggested by our data.
Relocalization of the other ISP family members into the apical cap may explain the lack of any gross phenotype in Δisp1 parasites. Whereas targeting to the apical cap is mediated by the N-terminal region of ISP1, relocalization of other family members into this sub-compartment is dependent on the C-terminal portion of this protein. Both the ISP1N/2C and ISP2N/1C chimeras support the conclusion that this gate-keeping is specific to ISP1 and directed against ISP2/3. Interestingly, while distal sequences of ISP2 are also required for its exclusion (as shown by ISP21–41-HA), this is not the case for a comparable truncation of ISP3.
Perhaps the simplest explanation for the mechanism of ISP2/3 exclusion from the apical cap is provided by our multiple PAT model (Figure 10B). This model would suggest that in wild-type parasites, the presence of ISP1, either directly or indirectly via other proteins, modulates PAT activity in the apical cap, thus preventing recognition of ISP2 and 3. In the absence of the ISP1 C-terminal domain, ISP2 and 3 are able to be recognized as substrates of the apical cap PAT and also localize to this compartment. This model would also suggest that the exclusion insensitivity of truncated ISP2 (Figure 5F), as compared to truncated ISP3 (Figure 5D), may simply result from a change in the ability of PATs to specifically recognize and act upon this altered molecule (discussed in the previous section). Alternatively, deletion of ISP1 may result in relocalization of a central sub-compartment PAT into the apical cap, thus enabling ISP2 and 3 to localize to this membrane region.
Finally, it is also possible that ISP1 exclusion is the result of a receptor in the apical cap, which the C-terminal domain of ISP1 binds with a higher affinity than ISP2 or 3. The absence of the ISP1 C-terminal domain would then allow binding of the similar regions of ISP2 and 3 to the receptor in the apical cap. However, the variable exclusion observed in C-terminal truncations of ISP2 and 3 argues against this scenario. We have attempted to identify ISP1 binding partners by immunoprecipitation under gentle conditions but have had no success, indicating that if partners do exist, they are not strongly interacting. Regardless of the precise mechanism, the targeting of the ISP family demonstrates that organization of the Toxoplasma IMC is an interactive, complex process. To our knowledge, this hierarchical targeting is a completely unprecedented mechanism for sorting of palmitoylated proteins in any membrane system. It will be interesting to see if similar mechanisms of membrane organization are present in other members of the eukarya.
Disruption of ISP2 results in defects in daughter cell formation, indicating that ISP2 is important for proper coordination of daughter parasite assembly. Our observation that ∼5% of wild-type parental strain vacuoles assemble >2 daughters is in agreement with previous studies [51]. Toxoplasma populations have been reported to undergo flux in the percentage of parasites displaying this trait due to certain stresses [52], however the dramatic (∼60%) effects on daughter parasite assembly in the Δisp2 strain vastly exceed these previous reports. Furthermore, the severe fitness loss in these parasites indicates this failure to properly coordinate cell division has serious consequences for parasite biology. This could be due to abortive replication events, as we do observe ultrastructural and organelle partitioning defects that are likely terminal (e.g. parasites lacking a nucleus or apicoplast and immature daughter buds within the vacuole, Figure 9G–H). However, many of the Δisp2 progeny produced in parasites assembling >2 siblings appear viable as they seem to properly assembly the IMC and cortical cytoskeleton and also receive nuclear DNA, an apicoplast and a mitochondrion (data not shown). In these cases, poor control over the number of daughter cells being assembled may also render a fitness cost on parasites during the normally efficient proliferative tachyzoite life stage.
The increase in the number of daughter parasites per mother cell results in several outcomes. In some parasites, DNA replication and karyokinesis occur prior to bud formation (Figure 9E–F), while in others, multiple rounds of DNA replication appear to occur without karyokinesis, resulting in large nuclei that are segregated in a single step among multiple daughters (Figure 9D). In either case, mother parasites that produce greater than 2 daughters are no longer performing endodyogeny, but instead replicating by one form or another of endopolygeny [19], [51], [53]. The presence of replication abnormalities in Δisp2 parasites reminiscent of division in other Toxoplasma life stages and other apicomplexan species suggests this protein plays a role in coordinating progress along the proper cell division pathway in tachyzoites and that this coordination is needed to maintain parasite fitness.
It is unclear how Δisp2 parasites ultimately recover from these defects and return to normal growth and replication. In both of the independent ISP2 knockouts performed months apart, the defects in growth and daughter formation were stable for at least two months. Recovery may be due to compensation via the other ISP proteins or may instead involve other players. It will be interesting to determine whether double knockouts of the ISP proteins, or even a triple knockout, will yield a more severe and stable phenotype. These functional implications for ISP2 underscore the idea that apicomplexan-specific processes are likely tied to the many hypothetical genes encoded within these parasites, some of which will provide novel therapeutic targets. The conservation of this family throughout the phylum suggests that the unique ISP targeting mechanism is conserved and raises the possibility that these proteins are more broadly involved in coordinating the various pathways of cell division that are critically important to the pathogenesis of apicomplexan parasites.
Antibodies were raised in mice under the guidelines of the Animal Welfare Act and the PHS Policy on Humane Care and Use of Laboratory Animals. Specific details of our protocol were approved by the UCLA Animal Research Committee.
T. gondii RHΔhpt (parental) strain and modified strains were maintained in confluent monolayers of human foreskin fibroblast (HFF) host cells as previously described [54].
Monoclonal antibodies (mAb) were generated against a mixed fraction of organelles from T. gondii [28]. For immunization, ∼100 µg of purified organelles [55] were injected in RIBI adjuvant into a BALB/c mouse. Following four injections, the spleen was isolated, hybridoma lines were prepared, and supernatants from individual clones screened for antibody reactivity.
The following primary antibodies were used in IFA or Western blot: rabbit polyclonal anti-tubulin [35], rabbit polyclonal anti-SAG1 [56], anti-IMC1 mAb 45.15 [33], anti-ROP1 mAb TG49 [57], and anti-ATrx1 mAb 11G8 [28]. Hemagglutinin (HA) epitope was detected with mAb HA.11 (Covance) or rabbit polyclonal anti-HA (Invitrogen).
Fixation and immunofluorescence staining of T. gondii were carried out as previously described [55]. All cells imaged in this study were formaldehyde-fixed except parasites in Figure S2B, which were fixed with methanol. Image stacks were collected at z-increments of 0.2 µm with an AxioCam MRm CCD camera and AxioVision software on an Axio Imager.Z1 microscope (Zeiss) using a 100x oil immersion objective. Deconvolved images were generated using manufacturer specified point-spread functions and displayed as maximum intensity projections.
The protein recognized by monoclonal antibody 7E8 was isolated from 5×109 T. gondii RH tachyzoites lysed in radioimmunoprecipitation assay (RIPA) buffer (50 mM Tris [pH 7.5], 150 mM NaCl, 0.1% sodium dodecyl sulfate [SDS], 0.5% NP-40, 0.5% sodium deoxycholate). Insoluble material was removed from the lysate by centrifugation at 10,000× g for 30 min after which the remaining soluble lysate fraction was incubated with mAb 7E8 cross-linked to protein G-Sepharose beads (Amersham) using dimethylpimelimidate as previously described [58]. After washing in RIPA buffer, the bound protein was eluted using high pH (100 mM triethylamine, pH 11.5) and the eluate was separated by SDS-polyacrylamide gel electrophoresis (PAGE). Coomassie staining identified a single 18-kDa band, which was excised and trypsin digested before analysis by mass spectrometry at the Vincent Coates Foundation Mass Spectrometry Laboratory, Stanford University Mass Spectrometry (http://mass-spec.stanford.edu).
YFP-αTubulin and mRFP-TgCentrin2 were expressed in parasites using previously described plasmids [20], [59]. HA epitope-tagged lines and YFP fusions pISP1/2/3-HA/YFP were generated by cloning the genomic loci of ISP1 (primers P1/P2), ISP2 (primers P3/P4) or ISP3 (primers P5/P6) into the expression plasmids pNotI-HA-HPT or pNotI-YFP-HPT using the restriction sites HindIII/NotI. These vectors contain a C-terminal HA tag or YFP fusion and selectable marker HPT driven by the DHFR promoter [60]. The ISP11–65 truncation was generated by cloning YFP (primers P7/P8) at the restriction sites EcoRV/PacI in pISP1-YFP. The ISP21–41 truncation was generated by cloning YFP (primers P9/P8) at the restriction sites RsrII/NotI in pISP2-YFP. The ISP31–36 truncation was generated by cloning the ISP3 promoter and residues 1–36 (primers P10/P11) at the restriction sites PmeI/AvrII in the previously described vector ptubYFP-YFP/sagCAT [61]. The ISP164–176 truncation was generated by cloning the ISP1 promoter and start codon (primers P1/P12) at the restriction sites HindIII/EcoRV in pISP1-YFP. The ISP1N/2C chimera was generated by cloning ISP243–160 (primers P13/P4) at the restriction sites EcoRV/NotI in pISP1-YFP. The ISP2N/1C chimera was generated by cloning ISP167–176-YFP (primers P14/P8) at the restriction sites RsrII/PacI in pISP2-HA. For expression, 1.6×107 parasites were transfected with 30 µg of plasmid and then analyzed by IFA as specified in figure legends.
Separation of the parasite IMC and plasma membrane was achieved by treatment with C. septicum alpha-toxin as previously described [33]. Briefly, freshly lysed, extracellular parasites were washed and incubated 4 hrs in serum free media with or without 20 nM activated alpha-toxin. Following treatment, cells were fixed in 3.5% formaldehyde, allowed to settle on glass slides and analyzed by IFA.
Tachyzoites were allowed to infect HFF monolayers on coverslips in media containing 0.5 or 2.5 µM oryzalin (Sigma). Parasites were allowed to grow 30–40 hrs post-infection and then fixed and examined by IFA.
The coding sequences for ISP2 (primers P15/P16) and ISP3 (primers P17/P18) were PCR amplified from T. gondii cDNA and cloned into pET101/D-TOPO (Invitrogen). Constructs were transformed into E. coli BL21DE3 cells, grown to A600 of 0.6–0.8 and induced with 1 mM isopropyl 1-thio-β-D-galactopyranoside (Sigma) for 5 hrs at 37°C. Recombinant ISP2 and ISP3 were purified over Qiagen Ni-NTA agarose under denaturing conditions and eluted with a low-pH buffer as per the manufacturer's instructions. Eluted proteins were dialyzed against PBS and ∼75 µg was injected per immunization into BALB/c mice (Charles River) on a 21 day immunization schedule. Polyclonal antiserum was collected from mice after the second boost and screened by IFA and Western blot analysis.
For detergent extraction experiments, 3×107 freshly lysed parasites were washed in PBS, pelleted and lysed in 1 mL TBS (50 mM Tris-HCl [pH 7.4], 150 mM NaCl) containing 0.5% NP-40 and complete protease inhibitors (Roche) for 15 min at 4°C and then centrifuged for 15 min at 14,000× g. Equivalent amounts of total, supernatant and pellet fractions were separated on a 15% gel, transferred to nitrocellulose and blotted using anti-IMC1, anti-ROP1, mAb 7E8, polyclonal anti-ISP2, and polyclonal anti-ISP3.
Mutations were generated by Quick Change Mutagenesis (Strategene) using HA-tagged, wild-type ISP1, 2 or 3 with mutagenesis primers as follows (forward primer given, reverse compliment was also used): ISP1: G2A (P19), C7S (P20), C8S (P21), C7,8S (P22). ISP2: G2A (P23), C5S (P24), C8S (P25), C9S (P26), C8,9S (P27), C5,8,9S (P28). ISP3: G2A (P29), C6S (P30), C7S (P31), C6,7S (P32). PCR amplified products were treated with DpnI to digest wild-type template and transformed into E. coli. Recovered clones were sequenced to confirm mutations.
The deletion of the ISP1 gene was accomplished by double homologous recombination using a construct derived from the pMini-GFP.ht knockout vector [62] which contains the selectable marker hypoxanthine-xanthine-guanine phosphoribosyltransferase (HPT) and also contains the green fluorescent protein (GFP) as a downstream marker to distinguish homologous and heterologous recombinants. The 5′ flank (3,147 bp) and 3′ flank (3,042 bp) of ISP1 were amplified from strain RH genomic DNA using primer pairs P33/P34 and P35/P36, respectively. These genomic flanks were then cloned into pMini-GFP.ht upstream and downstream of HPT, resulting in the vector pISP1-KO-HPT.
After linearization with NheI, 30 µg of pISP1-KO-HPT was transfected into RHΔhpt parasites and selection for HPT was applied 12 hours post-transfection using 50 µg/ml mycophenolic acid and 50 µg/ml xanthine. Surviving parasites were cloned by limiting dilution eight days post-transfection and screened for GFP by fluorescence microscopy. GFP-negative clones were assessed for absence of mAb 7E8 staining by IFA. Western blot analysis was carried out on whole-cell lysates of Δisp1 clones and parental strains using mAb 7E8 and anti-ROP1 antibody as previously described [55]. The HPT gene was removed from RHΔisp1 + HPT by a second round of double homologous recombination. The pISP1-KO-HPT vector was digested by EcoRV/NheI to remove the HPT gene and then blunted using Klenow enzyme and re-circularized by ligation. The resulting vector was linearized by EcoRI and transformed into RHΔisp1 + HPT, followed by selection for the absence of HPT on 200 µg/ml 6-thioxanthine (Sigma). After 3 weeks of selection, parasites were cloned and screened for the absence of GFP expression. Clones that were GFP-negative were then assessed for the inability to grow in mycophenolic acid and xanthine, indicating loss of HPT. One such clone was chosen and deletion of the ISP1 locus was confirmed by PCR. This clone was designated Δisp1.
The HPT selectable marker was removed from the Ku80 locus of the previously described Δku80 strain [38]. Briefly, 10 µg of a PCR fusion construct containing a 5′ Ku80 flank (primers P37/P38) fused to a 3′ Ku80 flank (primers P39/P40) was transfected into RHΔku80-HPT parasites. Selection against HPT with 6-thioxanthine and confirmation of marker loss were carried out as described above.
For disruption of ISP2, a knockout vector was generated by cloning ∼3 kb 5′ (primers P41/P42) and 3′ (primers P43/P44) genomic flanks into a modified version of pMiniGFP.ht in which HPT was replaced by the selectable marker DHFR-TSc3, yielding the vector pISP2KO-DHFR-TSc3. After linearization by NotI, 30 µg of this vector was transfected into Δku80Δhpt parasites and selection was applied 12 hours post-transfection using 1 µM pyrimethamine. Parasites were cloned and confirmed to lack ISP2 as described above. For disruption of ISP3, the vector pISP3-KO-HPT was generated by cloning ∼3 kb 5′ (primers P45/P46) and 3′ (primers P47/P48) genomic flanks into pMiniGFP.ht. After linearization by KpnI and transfection into the Δku80Δhpt strain, parasites were selected for HPT, cloned and confirmed to lack ISP3 as described above.
Freshly lysed parental and Δisp2 parasites were counted and mixed in desired ratios before infection of 3.3×106 parasites into a T25 flask of confluent HFFs. Parasites were allowed to disrupt the monolayer before passing into a fresh T25. At initial infection and at each passage, samples of the mixed culture were infected into coverslips and allowed to grow 32 hours before fixation and staining with polyclonal anti-ISP2 and rabbit polyclonal anti-tubulin as a co-marker to monitor mixed culture composition. At least 500 vacuoles were counted from each of 4 coverslips per passage. Values represent mean 3 standard deviations for a representative experiment.
Parental line and Δisp2 parasites were infected onto coverslips and allowed to grow 18–24 hours before fixation and staining with mAb 7E8 as a marker for daughter buds and rabbit polyclonal anti-tubulin as a co-marker. Fifty vacuoles containing parasites undergoing bud formation were counted from each of 3 coverslips per sample. Vacuoles containing one or more parasites assembling >2 daughters were scored as aberrant. Values represent the mean ± SD from a representative experiment.
|
10.1371/journal.pgen.1006150 | A Cascade of Wnt, Eda, and Shh Signaling Is Essential for Touch Dome Merkel Cell Development | The Sonic hedgehog (Shh) signaling pathway regulates developmental, homeostatic, and repair processes throughout the body. In the skin, touch domes develop in tandem with primary hair follicles and contain sensory Merkel cells. The developmental signaling requirements for touch dome specification are largely unknown. We found dermal Wnt signaling and subsequent epidermal Eda/Edar signaling promoted Merkel cell morphogenesis by inducing Shh expression in early follicles. Lineage-specific gene deletions revealed intraepithelial Shh signaling was necessary for Merkel cell specification. Additionally, a Shh signaling agonist was sufficient to rescue Merkel cell differentiation in Edar-deficient skin. Moreover, Merkel cells formed in Fgf20 mutant skin where primary hair formation was defective but Shh production was preserved. Although developmentally associated with hair follicles, fate mapping demonstrated Merkel cells primarily originated outside the hair follicle lineage. These findings suggest that touch dome development requires Wnt-dependent mesenchymal signals to establish reciprocal signaling within the developing ectoderm, including Eda signaling to primary hair placodes and ultimately Shh signaling from primary follicles to extrafollicular Merkel cell progenitors. Shh signaling often demonstrates pleiotropic effects within a structure over time. In postnatal skin, Shh is known to regulate the self-renewal, but not the differentiation, of touch dome stem cells. Our findings relate the varied effects of Shh in the touch dome to the ligand source, with locally produced Shh acting as a morphogen essential for lineage specification during development and neural Shh regulating postnatal touch dome stem cell maintenance.
| Sonic hedgehog (Shh) is one of a limited set of signaling molecules that cells use to drive organ formation during development and tissue regeneration after birth. How Shh signaling achieves different biological effects in the same tissue is incompletely understood. Touch domes are unique sensory structures in the skin that contain innervated Merkel cells. Using mouse genetics, we show that touch domes develop in tandem with, but distinct from, primary hair follicles. Moreover, touch dome specification requires a cascade of cell-cell signaling that ends with Shh signaling from an adjacent primary hair follicle. It was previously shown that Shh signaling from sensory nerves regulates the maintenance of touch dome stem cells after birth. Thus, the critical role for Shh signaling in embryonic touch dome specification is dependent on locally produced Shh, whereas the renewal of touch dome stem cells requires Shh transported to the skin by sensory neurons. These observations suggest that the distinct functions of Shh in touch dome development and maintenance correspond to changes in the source of the Shh signal required for the varied effects.
| The Hedgehog (Hh) pathway is conserved across the Metazoa subkingdom, and is one of a small number of intercellular signaling pathways that regulate the differentiation and pattering of morphologically diverse structures during development [1,2]. Postnatally, Hh ligands regulate tissue specific stem cell, homeostasis, and wound healing [3]. The basic molecular mechanisms of Hh signaling are still being investigated, and even less is known about how activation the pathway can result in such pleiotropic functions. Timing and distribution of Hh ligand delivery, ligand concentration, and duration of exposure can all influence signaling outcomes [4]. Multiple additional mechanisms have been proposed to alter Hh signaling, including ligand modification and sequestration, regulation of the primary cilium, modulation of Smo function by kinases, redundancy and cross regulation of Gli transcription factors, altering target gene expression by transcriptional co-regulators, and cross regulation by other signaling pathways [1–5]. In the present study, we discovered that Shh is the final critical element in a signaling cascade that specifies the touch dome lineage in developing mouse skin. Contrasting these findings with the role of Shh in regulating postnatal touch dome stem cells [6], we found the changing function of Shh was accompanied by a change in the source of the ligand, suggesting an additional contextual mechanism that influences the results of Shh signaling.
The functional diversity of vertebrate skin depends greatly on the variety of ectodermal appendages it produces. The development of ectodermal appendages including hair follicles, teeth, sweat glands, and mammary glands is precisely regulated by networks of signaling pathways including Wnt/β-catenin, Eda/Edar, Shh, and BMP [7]. Hair follicle development is particularly well studied as a model of ectodermal appendage development [8]. Identifying and comparing the developmental networks that control the specification and differentiation of ectodermal lineages can provide insights into developmental disorders and genetic diseases.
The touch dome (TD) is a specialized epidermal sensory structure composed of K8+ Merkel cells (MCs) arrayed among columnar basal keratinocytes that express the hair follicle keratin K17. Cells of the TD are morphologically and molecularly distinct from the adjacent interfollicular epidermis. The TD arises from the K14+ ectoderm [9,10] and is maintained as a distinct epidermal lineage by resident stem cells [6,11–13]. In postnatal skin, self-renewal of TD stem cells is regulated by Shh from sensory neurons that innervate the MCs [6]. MC development requires the transcription factor Atoh1 [14] and is regulated by levels of Sox2 expression [15,16]. TD MCs can be identified in the epidermis based on their expression of Atoh1, Sox2, or K8 [17].
TD MC development in the embryonic ectoderm is spatially and temporally associated with that of primary hair follicles. In embryonic day 14 (E14) mice, a primary wave of hair follicle placode induction takes place. These primary follicles produce guard hairs that ultimately comprise ~2% of the adult mouse coat. Secondary (E16) and tertiary (E18) waves of hair follicle induction are responsible for forming the three remaining types of hair follicles [18]. Developing MCs are first detected at ~E15 in association with nascent primary hair germs. Just before birth, TD MCs surround the infundibulum at the top of guard hair follicles. Shortly after birth, the touch dome has migrated into a crescent-shaped domain just caudal to the guard hair follicle (Fig 1A). Planar cell polarity signals regulate the reorganization of the forming TD [19]. Loss of Eda function results in abnormal guard hair formation and an absence of TD MCs, however the mechanism by which Eda signaling impacts MC specification is unclear [20]. Other signals involved in TD MC development are largely unknown.
To elucidate the developmental requirements for TD MC formation, we used genetically modified mice to disrupt signaling pathways known to be important in the development of other ectodermal appendages. We found that embryonic deletion of dermal β-catenin prevented TD MC formation. Similarly, Edar mutant skin failed to generate TD MCs. The mechanism of MC loss in these mice was due to failure of Shh expression by primary hair follicles, as evidenced by an absence of TD MC formation in Shh-null skin. Moreover, independent deletion of Shh or Smo in the embryonic epidermis reveled that intraepithelial Shh signaling from primary hair germs was necessary for TD MC specification. Notably, the loss of MC specification in Shh-deficient skin was observed before any disruption in hair follicle development was apparent. The importance of Shh signaling in TD MC formation was further demonstrated by using a Smo agonist to rescue MC specification in ex vivo-cultured Edar mutant skin. In contrast, Fgf20 was dispensable in TD MC development, demonstrating that the signaling cascades required for guard hair follicle development and TD formation diverge at the level of Fgf20 signaling. Distinction from the forming guard hair follicle was further demonstrated when fate mapping of the follicle lineage showed that TD MC progenitors predominantly arise in the epidermis outside the hair placode. Thus, like other ectodermal appendages, the touch dome is a distinct epidermal lineage whose specification and development requires Wnt-dependent mesenchymal-epithelial interactions and reciprocal signaling within the developing ectoderm, including Eda signaling to primary hair placodes and subsequent Shh production by primary hair germs. The critical role for Shh signaling in embryonic TD specification is dependent on locally produced ligand, whereas the regulation of postnatal TD stem cells requires Shh transported to the skin by sensory neurons. These observations suggest that ligand source influences the differential effects of Shh signaling in the TD.
Mesenchymal-epithelial interaction is critical in hair follicle morphogenesis [21]. In mice with dermal β-catenin conditional knockout, hair follicle development is arrested at a very early stage [22]. We investigated the effect of dermal β-catenin deletion on TD MC development using En1Cre/+; β-cateninflox/Δ mouse embryos where Cre recombination occurs in the dorsal trunk mesenchyme that forms the dermis [22]. As expected, dorsal trunk skin of E17.5 affected mice had no evidence of developing hair follicles on histological sections (Fig 1C). Using whole mount K8 immunostaining to detect MCs, we found K8+ TD MC adjacent to developing guard hair follicles in dorsal trunk skin of control (En1Cre/+; β-cateninflox/+, n = 5) E17.5 embryos (Fig 1B). In contrast, we found no K8 staining in dorsal trunk epidermis of E17.5 En1Cre/+; β-cateninflox/Δ embryos (n = 3, Fig 1B), indicating that β-catenin function in the developing dermis is necessary for the formation of TD MCs. En1Cre is not expressed in proximal limb mesenchyme [23]. As an internal control, we examined limb skin of E17.5 En1Cre/+; β-cateninflox/Δ embryos and observed normal collections of K8+ MCs in TDs adjacent to hair follicles, further implicating dermal Wnt/β-catenin signaling in TD MC development.
Dermal Wnt signaling is necessary for the earliest steps of epidermal patterning, including the upregulation of Edar expression in nascent hair placodes [22]. Edar is the membrane receptor for the TNF family protein Eda. MCs do not form in Eda mutant Tabby mice that have defective guard hair development [20]. We hypothesized that loss of Edar function would also disrupt MC development. To test this, we used point mutant Edardl-J/dl-J (downless) mice that lack guard hairs as their abortive primary hair germs fail to form follicles (n = 3, Fig 2B) [24]. No MCs were detected in postnatal day 0 (P0) dorsal trunk skin by whole mount K8 immunostaining (Fig 2A). We also failed to detect TD MCs in adult Edardl-J/dl-J; Atoh1LacZ/+ mouse skin stained with X-gal (n = 3, Fig 2C) [25], demonstrating that TD MC formation is truly abrogated in the absence of Edar function, and not simply delayed. To examine the early molecular specification of TD MC, we examined gene expression in E15.5 Edardl-J/dl-J dorsal trunk skin (n = 3) using quantitative reverse transcription–PCR (RT-PCR). Relative to control skin, Edar mutant skin shows significantly reduced mRNA levels of the MC differentiation markers Sox2 and Atoh1 (Fig 2D), suggesting that MC specification itself is disrupted. A difference in K8 expression was not detected, likely due to the broad expression of K8 in the periderm of embryonic skin [26]. As Sox2 is also expressed in the forming dermal papilla of E15.5 primary hair germs [16], these results further suggest that dermal papilla differentiation is disrupted in Edardl-J/dl-J mouse skin. Together, these results strongly suggest that Edar signaling in the primary hair placode critically regulates TD MC formation and that Edar loss alone can explain the absence of TD MC development in the dermal β-catenin knockout skin.
There are a number of signaling molecules expressed downstream of Edar in nascent primary hair follicle that could potentially influence MC specification and development. Shh is a morphogen expressed by forming hair placodes [27], and Edardl-J/dl-J mice fail to express Shh in their abortive primary hair germs (Fig 2D) [24]. As nerve-derived Shh was recently shown to regulate TD stem cell renewal in adult mouse skin [6], we hypothesized that Shh production was the mechanism by which primary hair follicles regulate TD MC development. Embryos deficient for Shh survive to birth but die postnatally due to holoprosencephaly [28]. Shh-null mice have defective hair follicle development where hair germs form but do not elongate into follicles [29,30]. We generated ShhGFPcre/GFPcre mouse embryos (n = 3) to assess TD MC formation in Shh null skin [31]. As expected, E18.5 ShhGFPcre/GFPcre hair follicles were arrested as hair germs (Fig 3C). In E18.5 control epidermis (n = 3), K8+ TD MCs were observed in annular clusters around primary hair follicles (Fig 3A and 3B). In contrast, no TD structures were detected in Shh-null mouse embryos, and the epidermis contained very few scattered K8+ cells (Fig 3A and 3B), demonstrating that Shh signaling is essential for TD MC development.
Using the Gli1LacZ reporter allele to visualize cells with active hedgehog signaling [32], we observed diffuse X-gal staining in E17.5 and P0 basal epidermis, dermis, and hair follicles (S1A and S1B Fig), demonstrating that there are hedgehog-responding cells throughout the epidermis of developing mouse skin. This finding is distinct from adult epidermis where Gli1 expression is restricted to the mature touch dome [12,33]. No LacZ reporter expression was detected in P0 ShhGFPcre/GFPcre; Gli1LacZ/+ trunk skin (S1C Fig), suggesting that Shh is the primary Hh ligand regulating skin development.
To test if hedgehog-responding cells can serve as Merkel cell precursors, E15.5 Gli1CreER/+; R26YFP/+ embryos (n = 3) were treated with low-level tamoxifen (2mg to the gravid dam) to genetically label a fraction on Gli1+ cells at the time of MC specification [34]. Scattered labeled cells were visible in the primary hair follicles, dermis, and epidermis of P0 sectioned dorsal trunk skin (S2 Fig). GFP staining was found in 27.6% of K8+ Merkel cells (n = 333). This was comparable to GFP labeling in 24.3% of primary hair follicle epithelial cells (n = 1847), a structure derived from precursors that highly express hedgehog response genes at E15.5 [27]. These fate mapping results demonstrate that hedgehog-responding precursors give rise to TD MCs.
The requirement for Shh in TD MC formation suggests that Shh loss accounts for the MC defect in Edar mutant skin. To test whether Shh signaling was indeed the critical mechanism regulating MC formation downstream of Edar, we used the Smo agonist Hh-Ag1.5 to restore hedgehog signaling in embryonic Edar mutant skin during the window of MC specification. We cultured E13.5 Edardl-J/dl-J skin (n = 3), and after two days (E15.5), used RT-PCR to assess gene expression. In untreated cultures (n = 3), we observed reductions in Sox2, Atoh1, Shh, and Gli1 relative to control skin (n = 3), similar to those seen in uncultured E15.5 Edar mutant skin (Figs 2D and 3D). Smo agonist treatment resulted in significant increases of Gli1 transcription in both control and Edar mutant skin, confirming activation of hedgehog signaling. Smo agonist also resulted in a reduction of Shh expression in control skin, suggesting that some form of negative feedback downregulates Shh transcription. Similarly, Smo agonist further reduced Shh levels in Edar mutant skin. Interestingly, Smo agonist reduced Atoh1 mRNA levels in control skin. This result is consistent with the observation that Shh signaling in neurons prevents Atoh1 degradation [35] and suggests that elevated Atoh1 protein levels in the setting of a Smo agonist can feedback to reduce Atoh1 transcripts. Most importantly, Smo agonist rescued expression of the MC differentiation markers Sox2 and Atoh1 to normal levels in Edar mutant skin (Fig 3D). Together, these results suggest that Shh is both necessary and sufficient for MC formation in embryonic trunk skin and is the critical factor lost in Edar mutant skin.
In embryonic skin, Shh is initially expressed in hair follicle placodes and continues to be expressed in developing hair follicle bulbs [36]. However, Shh is also delivered to the skin by sensory nerves [33]. To test the requirement for follicle-produced Shh, we deleted Shh from the embryonic epidermis using K14-Cre; Shhflox/flox mice [37,38]. Unlike control skin (n = 3), no TD MCs were detected in P0 mutant (n = 3) epidermis by immunostaining (Fig 4A and 4B), indicating that epidermal Shh is necessary for TD MC production. Similar to the Shh null mice, hair follicles failed to develop in K14-Cre; Shhflox/flox skin. Abortive K17+ germ-like structures formed at sites of follicle induction but failed to elongate (Fig 4B, 4C and 4E). At the time of TD MC specification in E15.5 control skin (n = 2), we observed Sox2+ and K8+ cells in the epidermis above and adjacent to primary hair germs. In E15.5 K14-Cre; Shhflox/flox skin (n = 2), neither Sox2 nor K8 staining was observed, suggesting that hair placode/germ-derived Shh is necessary for MC specification and that MC loss occurs prior to, and independent of, the follicle downgrowth defect in Shh mutant skin (Fig 4D and 4E). Sox2+ dermal papillae were observed under the abortive primary hair germs in E15.5 K14-Cre; Shhflox/flox skin (Fig 4E), suggesting epidermal Shh is not necessary for dermal papilla specification. Next, we used K5-tTA; TRE-Cre; Smoflox/flox mice in the absence of doxycycline to delete the obligate Shh signaling mediator Smo in the developing epidermis [39,40]. At E18.5, these mice (n = 2) completely lacked TD MCs that were readily detected by K8 immunostaining in control epidermis (n = 9, Fig 4F), demonstrating that the cells that require Shh signaling for TD MC development reside in the epidermis. Together, these results strongly suggest that intraepithelial Shh signaling from primary hair placodes/early hair germs to epidermal target cells is required for the specification and development of TD MCs.
Although K8+ cells seem to arise within forming primary hair follicles [20], our observations that Sox2+ and K8+ cells first appear in the epidermis above and adjacent to primary hair germs made us question whether TD MC progenitors develop within the hair placode or simply in close proximity to the hair follicle lineage. We used ShhGFPcre/+; R26YFP/+ mice to fate map the hair follicle lineage originating from the Shh-expressing hair placode [41]. To determine whether TD MCs arise from within the hair follicle lineage, we used immunostaining to assess co-labeling of Sox2+ and K8+ MCs with GFP staining in ShhGFPcre/+; R26YFP/+ skin at E14.5 (n = 3), E16.5 (n = 3), and E18.5/E19.5 (n = 3, Figs 5A, 5B, S3 and S4). In total (n = 189 MC), over 89% of MCs failed to stain with GFP, indicating that TD MCs primarily originate from extrafollicular cells. A breakdown of staining characteristics for MCs associated with hair follicles at different developmental stages [42] is shown in Fig 5C. A similar lack of GFP+ K8+ TD MCs was seen in adult ShhGFPcre/+; R26YFP/+ skin (Fig 5D), indicating that the stem cells maintaining TD MCs [13] also principally originate outside of the hair follicle lineage. Together, these results indicate that TD MCs predominantly arise from a developmental linage outside of primary hair placodes, although Shh+ ectodermal cells can give rise to a minor fraction of TD MCs.
The presence of Sox2+ K8- MCs that are more abundant in E14.5 skin versus later stages of development, and the complete absence of Sox2- K8+ cells (Figs 5C and S5A), suggests that Sox2 expression is an early event in MC specification and that K8 upregulation occurs later in MC differentiation. This finding is consistent with prior observations [16] and the fact that adult TD MC progenitors express very low levels of K8 and upregulate K8 expression upon differentiation to mature MCs [6,13].
In E14.5–16.5 skin, TD MCs were scattered in the epidermis adjacent to primary hair germs. In late embryonic and early postnatal skin (E18.5-P4), TD MCs were organized around the infundibulum of primary hair follicles, but were also found clustered on the upper caudal side of primary hair follicles, in a region known to express NCAM (Figs 5A, 5B, S3, S4, S5B, S5C) [20]. Interestingly, the MCs and NCAM+ keratinocytes found in ShhGFPcre/+; R26YFP/+ upper follicles infrequently expressed YFP (Figs 5A and S3). This observation suggests that the caudal side of the upper developing primary follicle does not form by downgrowth of the hair follicle placode but primarily by expansion of extraplacodal cells into the forming follicle. MCs do not persist in the upper region of primary follicles, as K8+ MCs are not detected within P13 guard hair follicles [6]. The purpose of the transient population of follicular MCs during late embryonic and early postnatal development is unclear.
We used mice with dermal β-catenin deletion, Edar mutation, Shh mutation, epidermal deletion of Shh, and epidermal deletion of Smo to show that paracrine Shh signaling within the epidermis, downstream of Wnt and Eda signaling, is essential for TD MC development. However, in all of these mice, there is either a global defect in hair follicle patterning and development, a developmental defect in all hair follicles, or a defect in primary hair follicles. To separate Shh signaling from normal guard hair development, we used Fgf20-null (Fgf20LacZ/LacZ) mice [43]. Like Shh, Fgf20 is expressed by hair placodes and is regulated by Wnt/β-catenin and Eda/Edar signaling. In Fgf20 mutant skin, there is a defect in the formation of hair follicle-associated dermal papillae and a failure in the downgrowth of primary hair germs [43]. Although the abortive primary hair germ phenotype of Fgf20 mutant skin resembles that of Eda and Edar mutant skin, Shh expression is preserved in the defective primary hair follicles [43]. We used RT-PCR to confirm that Shh is expressed in E15.5 Fgf20LacZ/LacZ skin (n = 3) and found Shh levels elevated compared with those in control skin (n = 3, Fig 6A). A small, nonsignificant increase in Gli1 mRNA levels was also observed. In P0 Fgf20 mutant skin (n = 3), we occasionally observed larger follicles that were comparable to guard follicles in control skin (n = 3, Fig 6B), suggesting that after an initial arrest, primary follicle downgrowth can occur. Nonetheless, the primary follicles were not normal, as guard hairs were absent in the coats of juvenile and adult Fgf20LacZ/LacZ mice [43]. In E15.5 Fgf20LacZ/LacZ skin, there were decreased levels of Edar, Sox2, and Atoh1 expression, although only the Atoh1 reduction reached statistical significance (Fig 6A), and a Sox2 reduction was expected based on the dermal papilla defect in this mouse. Despite the reduction in MC factors at the time of MC specification, the pattern and distribution of K8-immunostained TD MCs was normal in P0 Fgf20LacZ/LacZ skin (n = 4 mice, 752 TD, 11,393 MC, Fig 6B and 6D). We did observe a 17% reduction in the mean number of MC per TD relative to control epidermis (n = 2 mice, 348 TD, 6,338 MC, Fig 6B and 6D). Normal-appearing TDs persisted into adulthood with a normal TD density and a continued reduction (14%) in MC/TD observed in adult (P50-P103) Fgf20LacZ/LacZ mice (n = 3 mice, 131 TD, 1685 MC) relative to control (n = 3 mice, 136 TD, 1979 MC, S6 Fig). These results demonstrate that Fgf20 is required for normal guard hair follicle development but is not necessary for TD MC specification; our results also illustrate that when Shh expression is preserved, even abnormal primary hair germs are capable of supporting TD MC development.
Vertebrate skin provides barrier, mechanical, defensive, communicative, sensory, metabolic, and homeostatic functions. This diversity of function is achieved, in part, by specialization of ectodermal appendages that form through a series of mesenchymal-epithelial interactions in the embryo. By regulating common inductive events with similar yet distinct sets of morphogens, great diversity of structure and function is achieved [7]. Although some types of ectodermal appendages are specific to mammals (hair follicles, mammary, and sweat glands), structures that enhance sensation of the outside world are common adaptations across all classes of vertebrates. We have used multiple genetically modified mouse models to elucidate for the first time the developmental signaling cascade required for the formation of sensory TDs in embryonic skin. Only recently was the TD discovered to be a distinct skin lineage, maintained by its own resident stem cells. Here, we find that the mechanism for establishing the TD lineage within the developing ectoderm requires many of the morphogens that drive formation of ectodermal appendages. We discovered that TD MC specification requires Wnt-dependent mesenchymal signals to establish reciprocal signaling within the developing ectoderm, including Eda signaling to primary hair placodes, and subsequent Shh signaling from primary follicles to extrafollicular MC progenitors (Fig 7). Our identification of primary follicle Shh as a critical regulator of MC specification is consistent with TD development being spatially and temporally associated with the first wave of hair follicle induction in embryonic trunk skin, whereas our fate mapping results confirm that the TD lineage is separate from adjacent hair follicles. Together, our findings identify the TD as a distinct ectodermal touch receptor whose development is critically regulated by Wnt, Eda/Edar, and Shh signaling.
Our results support a model of TD MC development in which MC specification is tied to primary hair follicle patterning and morphogenesis up to the point of Shh production by the early follicle and can diverge from follicle development with signals such as Fgf20 that act in parallel or downstream to Shh in the hair follicle. Similarities in the developmental signaling pathways between the TD and primary hair follicles include the importance of mesenchymal Wnt signaling and ectodermal Eda/Edar signaling. Accordingly, there is no Shh expression by primary hair placodes in En1Cre/+; β-cateninflox/Δ skin, Edardl-J/dl-J skin, or Eda mutant Tabby skin [44], and TD MCs fail to form in all these mice. Although Shh signaling is critical in the development of both hair follicles and TDs, the role of Shh appears different for each structure. In the hair follicle, Shh signaling is dispensable for placode induction but is needed for the subsequent growth of the follicle. In the TD, loss of Shh signaling by either deleting Shh production or removing epidermal Smo results in complete loss of MC specification and development—a phenotype that is evident even before changes are seen in the primary hair germ. Another contrasting feature is that a hedgehog signaling agonist was sufficient to rescue MC differentiation in Edar mutant skin, whereas transgenic expression of Shh in Eda mutant Tabby mice failed to restore primary hair follicle development [45]. Downstream of Eda/Edar signaling, Fgf20 expression in the developing hair placode is needed for proper dermal papilla and guard hair formation, while Shh is expressed and TD MCs are able to form. Thus, although TDs and hair follicles share early developmental requirements for Wnt and Eda signaling, they arise from distinct ectodermal compartments, are maintained as distinct lineages, and differ in their specific requirements for Shh and Fgf20.
The coats of Fgf20 mutant mice lack guard hairs, and yet typical-appearing TDs were found at normal density and only a slightly reduced average number of MCs per TD. Interestingly, TD MCs are maintained in adult hairless mice, where hair follicles develop normally but undergo cystic degeneration after the first month of life [6], demonstrating that once established, the TD lineage can persist without an affiliated guard follicle. Thus, the primary hair follicle is a necessary source of Shh for TD MC formation in embryonic skin but is dispensable in postnatal MC maintenance. In postnatal mouse skin, maintenance of TD stem cells requires Shh signaling from sensory neurons that innervate MCs. However, TDs are specified at a time prior to epidermal innervation [46], and embryonic deletion of Shh from sensory neurons had no impact on TD MC formation [6]. It is noteworthy that Shh critically regulates different TD functions during development and in postnatal skin. In the embryo, hair follicle Shh is required to establish the MC lineage. Postnatally, loss of neural Shh blocks the maintenance of TD stem cells, but MC differentiation continues until the progenitor pool is exhausted [8]. This is analogous to the developing telencephalon, where Shh from the prechordal mesoderm and ventral forebrain is a critical morphogen for neural patterning and development [47], however postnatal neural stem cells are maintained by Shh from local niche neurons [48,49]. Shh signaling often shows pleiotropic effects within a given organ system with roles in patterning, specification, and proliferation during development and later functions in stem cell regulation, tissue regeneration, and cancer formation. Our findings and the observations from the central nervous system suggest that altering the source of ligand is an important contextual component influencing the function of Shh signaling within a tissue.
The formation of TDs adjacent to primary hair follicles requires Shh signaling from the nascent follicles. Secondary and tertiary hair follicles also express Shh during their formation, and yet TDs do not form in association with induction of those hair follicles. Moreover, based on the broad expression of Gli1 within the developing epidermis, many embryonic epidermal cells receive Shh signaling. Thus, even though a hedgehog signaling agonist was sufficient to rescue MC differentiation in Edar mutant skin, Shh signaling alone is not sufficient to induce TD MC specification in any and all developing ectoderm. Further experimentation will be necessary to identify the factors that establish the temporal and spatial competency of developing epidermis to respond to Shh signaling with MC specification. However, Polycomb repressive complex 2 (PRC2) activity appears to be important in preventing MC specification around secondary and tertiary hair follicles (personal communication, E. Ezhkova). Notch signaling may also play a role in limiting MC specification, as ectopic expression of Atoh1 is able to induce MC production in some epithelial compartments of the skin, and impeding Notch signaling facilitates this process [50].
Fate mapping showed that MCs arise predominantly from ectoderm outside the forming hair follicle. This is consistent with our observations that the first MCs appeared as Sox2+ cells in the E14.5/15.5 epidermis above and adjacent to forming primary hair germs. We confirmed that late embryonic and early postnatal guard hair follicles contain a discrete population of MCs, however these cells also tend to arise from outside the hair follicle lineage. Although the purpose of the MCs in the upper regions of primary hair follicles during development is unclear, it has been proposed that the NCAM expressed on keratinocytes around these MCs may facilitate MC innervation [20]. Coincidently, the early postnatal window corresponds to a period when TD MC innervation undergoes pruning and maturation [46][51], and the onset of perineural influence on MC maintenance [6]. Currently, it is not known whether the MCs in the follicle during this period eventually die, change their fate, or migrate to TDs in the epidermis.
We have determined that intraepithelial Shh signaling from the developing hair follicle to MC precursors is a critical factor in TD MC production. However, the precise location of the target cells for Shh remain undefined. Because the Hh response gene Gli1 is broadly expressed in the developing epidermis, it cannot be used to identify the specific K5+ Shh-responding cells required for TD MC development. Just as adult TD stem cells are Gli1+ [6], our Gli1 fate mapping experiments in the embryo show that TD MC progenitors are a direct target of Shh signaling. Because MC development within the TD anlage is dependent on Shh signaling from developing hair follicles, it is reasonable to assume the TD anlage forms in proximity to primary hair placodes. Moreover, the appearance of early MCs in the epidermis around primary hair germs and the observation that a minor portion of TD cells originate from the hair placode lineage suggest that the TD anlage is likely the ectoderm immediately adjacent to, and slightly overlapping, the primary hair placode. It is uncertain whether the TD forms from its own placode and associated mesenchymal condensate; however, this is unlikely, as no such structures have been observed. It is more likely that the TD is induced in the adjacent ectoderm by inchoate primary hair follicles or by the same signals that induce the follicles. Nonetheless, as an independent epidermal lineage requiring mesenchymal induction during development, and being absent in mouse models of ectodermal dysplasia (Eda and Edar mutant mice), TDs can be considered ectodermal appendages, or at least accessory structures to ectodermal appendages.
This work elucidates the developmental signaling requirements for the sensory TD and illustrates the commonalities and contrasts that exist between TD development and development of the closely associated guard hair follicle. These results further our functional understanding of skin patterning and development, the repertoire of ectodermal appendage formation, and how specialized sensory structures form prior to interfacing with the sensory nervous system. Intriguingly, along with prior observations, these findings indicate that the distinct functions of Shh signaling in TD development and maintenance correspond to changes in the source of the Shh ligand required for the varied effects.
Mice were housed and bred on an outcrossed Swiss Webster background in a pathogen-free facility at the National Cancer Institute (NCI), Bethesda, MD. Genotyping of mice was performed by allele-specific PCR on DNA extracted from tail tissue. All experiments were performed in accordance with institutional guidelines according to IACUC-approved protocols. En1Cre/+; β-cateninflox/Δ and control samples were provided by Dr. Radhika P. Atit. Some ShhGFPcre/+; R26YFP/+ samples were provided by Dr. Sunny Y. Wong. Edardl-J/+, Atoh1LacZ/+, ShhGFPcre/+, Gli1LacZ/+, Gli1CreER/+, K14-Cre, Shhflox/+, K5-tTA, TRE-cre, Smoflox/+, R26YFP/+, R26LacZ/+, and Fgf20LacZ/+ mice were described previously as cited in the text. For all embryonic and neonatal observations, littermate animals with a wildtype copy of the targeted allele and/or lacking cre recombinase activity were used as controls.
K5-tTA; TRE-cre; Smoflox/flox mice were bred and maintained on a standard rodent diet without doxycycline during embryonic development. Tamoxifen (Sigma) was dissolved in corn oil (20mg/ml) and administered (2mg intraperitoneal injection per gravid mouse) to induce CreER.
Skin was fixed in 4% paraformaldehyde for 15 minutes (for X-gal staining) or overnight (for immunostaining). Tissue was whole mount-stained or cryoprotected overnight in 30% sucrose, embedded in OCT, and frozen; 12-μm sections were obtained.
Standard and whole mount immunostaining procedures were performed. Tissue sections on glass slides were fixed in 4% paraformaldehyde for 15 minutes before incubation in 10% serum in 0.1% PBT (0.1% Triton X-100 in PBS) for 1 hour and then in primary antibody (in 5% serum/0.1% PBT) overnight at 4°C. The primary antibodies used were: rat anti-K8 (1:50, University of Iowa), rabbit anti-K17 (1:200, Epitomics), chicken anti-GFP (1:1000, Abcam), rabbit anti-GFP (1:500, Abcam), rabbit anti-Sox2 (1:500, Stemgent), and rabbit anti-NCAM (1:500, Millipore). Alexa Fluor-conjugated secondary antibodies (1:2000, Invitrogen) were used to detect the signals. Whole mount immunostaining followed the online protocol as described [52]. Concomitant staining of littermate control tissue and control staining where the primary antibody was omitted were used to confirm the specificity of experimental staining. Confocal images were acquired with the Zeiss LSM 710 Confocal system (Carl Zeiss Inc, Thornwood, NY).
The Troma-1/K8 antibody developed by Dr. Philippe Brulet and Dr. Rolf Kemler was obtained from the Developmental Studies Hybridoma Bank, developed under the auspices of the National Institute of Child Health and Human Development and maintained by The University of Iowa, Department of Biology, Iowa City, IA 52242.
Merkel cells per touch dome and touch dome per mm2 numbers were assessed by direct visualization of immunofluorescently stained K8+ Merkel cells in whole mount dorsal trunk skin from experimental and control animals. Counting was performed by a blinded observer. In tissue sections, Merkel cells were visualized by K8 and/or Sox2 staining and were scored for co-staining with GFP. Reporter recombination in primary hair follicles was quantified by counting total GFP staining cells and dividing by total number of DAPI stained nuclei within a preselected region of each follicle.
Total RNA from embryonic trunk skin was purified using RNeasy Micro Kit (Qiagen) and reverse-transcribed into cDNA following the manufacturer’s manual (Invitrogen #11752). Then quantitative RT-PCR, with actin as control, using SYBR Green was performed to detect RNA expression. Data are presented as means ± SD. The sequences of PCR primers are shown in S1 Table.
Embryonic skin culture was performed as described [53]. E13.5 embryos were collected, each embryonic mouse was cut through the sagittal midline and eviscerated, half was cultured in media (DMEM +10% FBS +1% penicillin/streptomycin) as a control group, and half was cultured in media with 25 nM Smo agonist Hh-Ag1.5 (Xcess Biosciences Inc.) as the treated group. Media was changed the next day, and skin was harvested 2 days after culture, followed immediately by RNA isolation.
Population data sets are shown as the mean values, and error bars represent SD. For comparisons between sets, a two-tailed t-test was applied.
|
10.1371/journal.pntd.0001167 | Gene Expression Profiling and Molecular Characterization of Antimony Resistance in Leishmania amazonensis | Drug resistance is a major problem in leishmaniasis chemotherapy. RNA expression profiling using DNA microarrays is a suitable approach to study simultaneous events leading to a drug-resistance phenotype. Genomic analysis has been performed primarily with Old World Leishmania species and here we investigate molecular alterations in antimony resistance in the New World species L. amazonensis.
We selected populations of L. amazonensis promastigotes for resistance to antimony by step-wise drug pressure. Gene expression of highly resistant mutants was studied using DNA microarrays. RNA expression profiling of antimony-resistant L. amazonensis revealed the overexpression of genes involved in drug resistance including the ABC transporter MRPA and several genes related to thiol metabolism. The MRPA overexpression was validated by quantitative real-time RT-PCR and further analysis revealed that this increased expression was correlated to gene amplification as part of extrachromosomal linear amplicons in some mutants and as part of supernumerary chromosomes in other mutants. The expression of several other genes encoding hypothetical proteins but also nucleobase and glucose transporter encoding genes were found to be modulated.
Mechanisms classically found in Old World antimony resistant Leishmania were also highlighted in New World antimony-resistant L. amazonensis. These studies were useful to the identification of resistance molecular markers.
| Leishmania are unicellular microorganisms that can be transmitted to humans by the bite of sandflies. They cause a spectrum of diseases called leishmaniasis, which are classified as neglected tropical diseases by the World Health Organization. The treatment of leishmaniasis is based on the administration of antimony-containing drugs. These drugs have been used since 1947 and still constitute the mainstay for leishmaniasis treatment in several countries. One of the problems with these compounds is the emergence of resistance. Our work seeks to understand how these parasites become resistant to the drug. We studied antimony-resistant Leishmania amazonensis mutants. We analyzed gene expression at the whole genome level in antimony-resistant parasites and identified mechanisms used by Leishmania for resistance. This work could help us in developing new strategies for treatment in endemic countries where people are unresponsive to antimony-based chemotherapy. The identification of common mechanisms among different species of resistant parasites may also contribute to the development of diagnostic kits to identify and monitor the spread of resistance.
| Leishmaniasis refers to a spectrum of parasitic diseases caused by protozoan parasites belonging to the genus Leishmania. The diseases are classified as neglected tropical diseases according to the World Health Organization (WHO) and constitute a public health problem in many developing countries of East Africa, the Indian subcontinent and Latin America. Human leishmaniasis has a prevalence of 12 million cases, with an estimated population of 350 million at risk and an incidence of 2 million new cases annually. Depending on Leishmania species, the host immune response, and environmental factors, leishmaniasis exhibits a broad spectrum of clinical manisfestations [1]. For example, in the New World, Leishmania (Leishmania) amazonensis, Leishmania (Viannia) guyanensis and Leishmania (Viannia) braziliensis are the causative agents of cutaneous and mucocutaneous leishmaniasis while Leishmania (L.) infantum chagasi is the aetiological agent of American visceral leishmaniasis [1], [2].
Pentavalent antimonials (SbV), such as sodium stibogluconate (Pentostam®) and meglumine antimoniate (Glucantime®) have been the first-line drugs in the treatment of all forms of leishmaniasis in South America, North Africa, Turkey, Bangladesh and Nepal. One major drawback of the SbV treatment is the emergence of resistance. For example, more than 60% of patients with visceral leishmaniasis in Bihar State in India are unresponsive to treatment with SbV antimonials [3]. The emergence of antimony resistance is related to inappropriate drug exposure resulting in a build-up of subtherapeutic blood levels and increasing tolerance of parasites to SbV [4]. Other drugs have been introduced as alternative chemotherapeutic agents including pentamidine, paromomycin, liposomal amphotericin B and miltefosine. However, either side effects, lower effectiveness or high cost have limited their use [5].
The mechanisms involved in antimony resistance in Leishmania are partially understood. Antimonial drugs are administered as SbV, a prodrug that is reduced to SbIII, the trivalent and biologically active Sb form [6], [7]. However, the site of this reduction (macrophages and/or parasites) remains unclear. Two genes that encode proteins involved in Sb reduction have been described recently, the arsenate reductase LmACR2 and TDR1 thiol-dependent reductase [8], [9]. Nevertheless, the role of these reductases in antimony resistance is not clear. Non enzymatic Sb reduction is also possible and probably mediated by the reducing agents glutathione (GSH) and trypanothione (T(SH)2) [5], [10], [11]. Once reduced in the macrophages, SbIII uptake is mediated by the aquaglyceroporin1 (AQP1) [12] and downregulation of AQP1 gene expression is correlated to resistance [13].
Increases of T(SH)2 levels have been observed in parasites selected for resistance to SbIII or arsenite [14]. This enhancement is usually related to the increased levels of rate-limiting enzymes involved in the synthesis of GSH (gamma glutamylcysteine synthetase- γ-GCS) and polyamines (ornithine decarboxylase – ODC) [15], [16]. The use of specific inhibitors of γ-GCS or ODC can revert the resistance phenotype in mutants [16]. The ATP-binding cassette (ABC) protein MRPA has been classically related with drug resistance in Leishmania and plays a major role in metal resistance in these parasites [17]. MRPA is a member of the multidrug-resistance protein (MRP) family and its localization in intracellular vesicle membranes strongly suggests that it sequesters Sb-thiol complexes into these vesicles [18]. The MRPA gene has been found frequently amplified in laboratory-selected antimony- or arsenite-resistant Leishmania mutants as well as in field isolates [19], [20], [21].
Improved knowledge of the mechanisms involved in drug resistance using laboratory-selected mutants or field isolates are mostly derived from Old World Leishmania species such as L. tarentolae [22], L. major [23], [24], L. tropica [25], L. donovani [26], and L. infantum [27]. On the other hand, the mechanism of drug resistance in New World Leishmania species remains poorly explored. Nevertheless, phenotypic and molecular characterizations of drug resistance have been recently published for human pathogenic neotropical Leishmania species [28], [29], [30]. Resistance to antimony in L. amazonensis has not been well studied as yet. Understanding the mechanisms responsible for drug resistance in Leishmania could support the design of new strategies for the successful treatment of leishmaniasis as well as the identification of molecular markers for resistance.
Considering the multiplicity of mechanisms leading to antimony resistance, the simultaneous analysis of gene expression could provide useful information about the antimony-resistance mechanisms in Leishmania and help the identification of new pathways involved in resistance. Recent studies have demonstrated the usefulness of whole-genome DNA microarrays for studying drug resistance in Leishmania [31], [32]. In this study, populations of L. amazonensis resistant to SbIII were selected in vitro in order to study global gene expression modulation associated with antimony resistance.
Leishmania amazonensis (MHOM/BR/1989/Ba199) promastigotes were maintained in minimum essential culture medium (α-MEM) (Gibco, Invitrogen, NY, USA), supplemented with 10% (v/v) heat-inactivated fetal calf serum (Multicell, Wisent Inc. Québec, CA), 100 µg/ml kanamycin, 50 µg/ml ampicillin, 2 mM L-glutamine, 5 µg/ml hemin, 5 µM biopterin, (Sigma-Aldrich, St Louis, USA), pH 7.0 and incubated at 25°C in B.O.D incubators (Johns Scientific-VWR, Toronto, CA). The parasites were kindly provided by Dr. Aldina Barral, Gonçalo Muniz Research Center, Oswaldo Cruz Foundation, Brazil [33]. Populations of Leishmania amazonensis promastigotes were selected for SbIII resistance as previously described [19]. The four independent mutants of L. amazonensis Ba199SbIII2700.1 to Ba199SbIII2700.4 were individually selected in 25 cm2 flasks containing 5 ml of α-MEM medium in the presence of SbIII concentrations up to 2700 µM.
L. amazonensis Ba199Sb mutants selected for SbIII resistance were grown in the absence of antimony pressure for 20 passages to test for the resistance stability phenotype [34].
The full genome arrays were described previously [31], [32], [35]. GeneDB version 3.0 of L. infantum genome and L. major genome version 5.2 were used for the probe selection. The microarray chip includes a total of 9173 Leishmania specific probes and control probes and made by Agilent Technologies (Mississauga, ON, CA). These arrays have been used successfully with several species [31], [35].
Total RNA was extracted from 108 promastigotes during the mid-log growth phase using RNeasy Plus mini kit (Qiagen Sciences, Maryland, USA) as described by the manufacturer. The quality (based on the appearance of the spectra) and quantity of RNA were assessed using RNA 6000 Nano Assay chips on Bioanalyzer 2100 (Agilent Technologies Santa Clara, CA, USA). For each probe, 7 µg of RNA were converted to aminoallyl-dUTP incorporated cDNA using random hexamers (Roche, Basel, Switzerland) in presence of Superscript III RNase H reverse transcriptase (Invitrogen, Carlsbad, CA, USA). Aminoallyl-dUTP incorporated cDNA were thereafter coupled to Alexa Fluor 555 or Alexa Fluor 647 (Invitrogen, Carlsbad, CA, USA) according to manufacturer recommendations. Fluorescent cDNA were then purified using the probe purification kit ArrayIt (TeleChem International, Sunnyvale, CA, USA) and quantified spectrophotometrically.
The labeled and purified cDNA from L. amazonensis was mixed with 200 µg/ml sonicated salmon sperm DNA (Agilent Technologies, Santa Clara, CA, USA); 200 µg/ml yeast tRNA (Sigma-Aldrich Ltd, ON, CA); 1 x blocking agent buffer (Agilent Technologies, Santa Clara, CA, USA) and 1 x hybridization buffer (Agilent Technologies, Santa Clara, CA, USA), then mixed, denaturated 3 min at 95°C and incubated 30 min at 37°C. Mixed labeled cDNAs were applied in the hybridization chamber (Agilent Technologies, Santa Clara, CA USA) and the hybridization was performed for 24 h at 65°C into a hybridization oven (GeneChip®, Stovall Life Sciences, Greensboro, NC, USA). Slides were washed 5 min at room temperature in 0,5X SSC, 5% Triton-X102 with gentle agitation and subsequently washed 5 min in pre-warmed 0,1X SSC, 0,005% Triton-X102 at room temperature with occasional stirring.
Detection of Alexa Fluor 555 and Alexa Fluor 647 signals were performed on a G2565CA microarray scanner (Agilent Technologies, Santa Clara, CA, USA) at 5 µm resolution as previously described [32]. The signal intensity data were extracted from the primary scanned images using GenePix Pro 6.0 software (Axon Instruments, Union City, CA, USA). Five different cDNA preparations of each Ba199Sb mutant and their respective Ba199 wild-type were analyzed including dye-swaps. Normalization and statistical analyses were performed in R 2.2.1 software using the LIMMA (Linear Models for Microarray Data) 2.7.3 package [36], [37], [38]. Background correction was performed using the “edwards” method; within-array normalization was done by loess and between array normalization by the Aquantile method. Multiple testing corrections were done using the false discovery rate method with a threshold p value of 0.05. Only genes statistically significant with an absolute ratio greater than 1.5 were considered. Custom R programs were used for the generation of the chromosome expression maps. Data are available with the GEO accession number GSE26159.
Three independent RNA preparations were used for each real-time RTPCR experiment. First-strand cDNA was synthesized from 2.5 µg of RNA using Oligo dT12–18 and SuperScript II RNase H-Reverse Transcriptase (Invitrogen, Carlsbad, CA, USA) according to the manufacturer protocol. Equal amounts of cDNA were run in triplicate and amplified in 20 µl reactions containing 1 x SYBR® Green Supermix (Bio-Rad, Hercules, CA, USA), 100 nM forward and reverse primers and 1 µl cDNA target. Reactions were carried out using a rotator thermocycler Rotor Gene (RG 3000, Corbett Research, San Francisco, USA). Initially, mixtures were incubated at 95°C for 5 min and then cycled 30 times at 95, 60 and 72°C for 15 sec. No-template controls were used as recommended. Three technical and biological replicates were established for each reaction. The relative amount of PCR products generated from each primer set was determined based on the threshold cycle (Ct) value and the amplification efficiencies. Gene expression levels were normalized to constitutively expressed mRNA encoding glyceraldehyde-3-phosphate dehydrogenase (GAPDH, LmjF30.2970). Primers for targeted genes MRPA (LmjF23.0250), NT3 (LmjF13.1210) and LmjF26.2680 were designed using Primer QuestSM (www.idtdna.com/Scitools/Applications/Primerquest). The sequences of the primers for MRPA are forward 5′-TGAGACACGCCGCATCAAGAGTAT-3′ and reverse 5′-TCAATGCTTCCTGCAGTACGAGGT-3′; for NT3 are forward 5′-AAGTTCATCTGGCCTCTCATGGCT-3′ and reverse 5′-GATGGTTGCAAACCACTTGTCCGT-3′; for LmjF26.2680 are forward 5′-ACCCAGTCATTCGTCATGCACTCT-3′ and reverse 5′- ATCTGGTTGACAGCGTCGCAAATG-3′ ; and for the GAPDH control forward 5′-GAAGTACACGGTGGAGGCTG-3′ and reverse 5′-CGCTGATCACGACCTTCTTC-3′.
Genomic DNA was isolated from L. amazonensis Ba199 WT and Ba199 antimony-resistant mutants using DNazol (Invitrogen, Carlsbad, CA, USA) following the manufacturer's instructions. Southern-blots and pulse field gel electrophoresis (PFGE) conditions were done following standard protocols [20]. Genomic DNAs were digested with PvuI and electrophoresed in 1% agarose gel. The fragments were transferred to HybondTM-N+ membrane (Amersham Pharmacia Biotech, Sunnyvale, CA, USA) and submitted to Southern-blot analysis. Chromosomes of L. amazonensis Ba199 WT and antimony-resistant mutants were separated by PFGE in which low melting agarose blocks, containing embedded cells (108 log phase cells/ml) were electrophoresed in a contour clamped homogenous electric field apparatus (CHEF Mapper, Bio-Rad, Hercules, CA, USA) in 0,5 x Tris-Borate-EDTA, with buffer circulation at a constant temperature of 14°C and run time of 30 h. Saccharomyces cerevisiae chromosomes were used as size markers. DNA was transferred to nylon membranes, cross-linked to the membrane with UV light. The blots were hybridized with [α-32P]dCTP labeled DNA probes. The probes used in the present study included a 450 bp MRPA fragment and a α-tubulin probe used to control the DNA loading.
Intracellular thiols were analyzed by derivatizing with mono-bromobimane and separating by high-performance liquid chromatography as described previously [14], [39] using a chromatograph Shimadzu SCL 10A. Thiols were identified from bimane fluorescence with excitation and emission at 360 and 450 nm, respectively using a coupled fluorescence detector (Shimadzu RF-10Axl).
The IC50 values were calculated by linear regression using the software GraphPad Prism 5.0 and Sigma Plot 10.0 for windows. Differences in the level of intracellular thiols were analyzed by one-way ANOVA followed by Dunnett's multiple comparison test post-test using GraphPad Prism 5.0. The level of significance acceptable was 95% (p<0.05).
Four independent mutants of L. amazonensis were selected step by step for antimony (SbIII) resistance. The IC50 value of the sensitive Ba199 strain was 83 µM, whereas the antimony resistant-mutants Ba199SbIII2700.1, 2700.2, 2700.3 and 2700.4 had IC50 values greater than 2700 µM (Table 1), the highest achievable soluble SbIII concentration in α-MEM medium at pH 7. The stability of the resistance phenotype was tested by growing the cells in the absence of SbIII. After 20 passages without drug pressure, only the resistance in mutant Ba199SbIII2700.2 was found to be stable, while the other three mutants showed decreased resistance levels (Table 1). However, reversion was only partial since the Ba199SbIII2700.1, 2700.3 and 2700.4 mutants were not as sensitive as wild-type cells to SbIII (Table 1). The susceptibility to miltefosine in the SbIII-resistant L. amazonensis mutants was also tested. None were cross-resistant but 3 out of the 4 lines were surprisingly hypersensitive to it (Table 1). Intriguingly, we have also observed hypersensitivity to miltefosine in L. infantum SbIII-resistant mutants (W. Moreira and M. Ouellette, unpublished observations).
The Ba199SbIII2700.2 and Ba199SbIII2700.3 lines were selected for gene expression studies using full genome DNA microarrays. We plotted the log2-transformed gene expression ratios of Ba199SbIII2700.2 (red line) and Ba199SbIII2700.3 (blue line) compared to Ba199WT parental strain, as a function of the microarray probes (Fig. 1). Most genes were equally expressed but about 10% of genes showed a statistical significant variation (summarized in Table S1 and detailed in Tables S2 and S3) with approximately 2-fold differential expression but some reached log2-transformed ratio values up to 4 and −4 (Fig. 1). The differential hybridization data were also represented on a chromosome by chromosome basis (Figs. 2 and 3). Upregulated and downregulated genes are indicated by red and green lines, respectively, while equally expressed genes were shown as gray regions. Some obvious changes in gene expression were noticed. A specific region at one telomeric end of chromosome 23 was upregulated in Ba199SbIII2700.2 (Fig. 2), while most genes of chromosome 23 seemed upregulated in Ba199SbIII2700.3 (Fig. 3). Chromosome aneuploidy has been described previously in Old World drug resistant Leishmania [31], [32] and the chromosome maps of Figs. 2 and 3 suggest that this phenomenon also takes place in New World Leishmania species with chromosomes 1, 10, 16, 27 and 31 becoming polyploids in Ba199SbIII2700.2 (Fig. 2) while in addition to chromosome 23, chromosomes 5, 27 and 32 are polyploids and chromosome 4 is haploid in Ba199SbIII2700.3 (Fig. 3). A region of chromosome 35, 250 kb from one telomeric end, corresponds to loci where the expression of genes was down regulated in both Ba199SbIII2700.2 and Ba199SbIII2700.3 mutants (Figs. 2 and 3). The expression of genes part of a region on chromosome 33, 1.5 Mb from one telomere end was also down regulated in both mutants (Figs. 2 and 3).
The array results led to several candidate genes putatively correlated to resistance. Candidate genes could either be highly differentially regulated or part of large regions differentially regulated, as highlighted in Figs. 2 and 3. The genes common to both mutants most differentially down regulated included the hypothetical protein gene LmjF26.2680 and a putative lmgt2 glucose transporter gene LmjF36.6290 (Fig. 1 and Supplementary Tables S2 and S3). On the other hand, the gene common to both mutants most upregulated was corresponding to the nucleobase transporter NT3 LmjF13.1210. The overexpression of NT3 was confirmed by qRT-RTPCR which yielded similar results as found with microarrays with higher expression of NT3 in Ba199SbIII2700.2 compared to 2700.3 (Fig. 1, Fig. 4). We also tested the two other L. amazonensis mutants available and found that NT3 was also overexpressed in Ba199SbIII2700.1 and 2700.4 (Fig. 4). Similarly, we confirmed the down regulation of LmjF26.2680 by qRT-RTPCR not only in Ba199SbIII2700.2 and 2700.3 but also in two other L. amazonensis resistant mutants (Fig. 4). None of the genes described above were previously linked to antimony resistance in Leishmania. For specific larger regions that were presumed to be up or down regulated as determined from the chromosome maps of Figs. 2 and 3, we found that the region of chromosome 23 upregulated in Ba199SbIII2700.2 (Fig. 2) contained several genes (Table S2) including the ABC protein gene MRPA LmjF23.0250, a well established marker of antimony resistance [40], [41]. The MRPA gene was also upregulated in Ba199SbIII2700.3 (Fig. 3) as determined by microarrays (Table S3). As discussed above, two regions of chromosome 35 and 33 appeared to be down regulated in both mutants. The region of chromosome 35 encodes for several hypothetical proteins, but also three proteophosphoglycan (PPG) genes PPG1, PPG3 and PPG5 (Tables S2 and S3). Similarly, the region of chromosome 33 corresponds mostly to hypothetical proteins (Tables S2 and S3). With the exception of MRPA, none of the genes highlighted in this study were previously linked to antimony resistance. We searched for genes that were previously linked to resistance with significant changes in gene expression and found several genes that were upregulated in the Ba199SbIII mutants and that were involved in redox and thiol metabolism such as peroxidoxin (LmjF23.0040), glutaredoxin (LmjF05.0310), trypanothione synthetase (LmjF23.0460; LmjF27.1870), trypanothione reductase (LmjF05.0350), and spermidine synthase (LmjF04.0580) (Tables S2 and S3). The overexpression of several trypanothione biosynthetic genes (e.g. spermidine synthase, trypanothione synthetase) prompted us to quantify the level of intracellular reduced thiols, since resistance to SbIII is often correlated to increased glutathione and trypanothione levels in Old World Leishmania [42]. The antimony-resistant L. amazonensis mutants, with the exception of Ba199SbIII2700.1 (for glutathione), had significant higher levels of cysteine, glutathione and trypanothione (Fig. 5).
We also tested the role of genes previously not associated with resistance, concentrating on some of the genes most differentially expressed. These genes correspond to the hypothetical gene LmjF26.2680 which was down regulated by more than 20-fold in all mutants (Fig. 4) and NT3 that was overexpressed in all mutants as determined by real-time RT-PCR (Fig. 4). Transfection of LmjF26.2680 in wild-type L amazonensis or in its resistant mutants did not change their susceptibilities to SbIII (results not shown). Similarly, transformation and overexpression of NT3, did not lead to higher resistance to SbIII in wild type cells (result not shown).
The MRPA gene was overexpressed in both mutants (Table S2 and S3) and this upregulation was indeed confirmed by qRT-RTPCR in Ba199SbIII2700.2 and 2700.3 but MRPA was also found overexpressed in Ba199SbIII2700.1 and 2700.4 (Fig. 4). The fold increased expression by qRT-RTPCR was higher than what microarray would have suggested. Often, but not always, gene overexpression is correlated to gene amplification in Leishmania [13], [40], [41]. Southern blot analysis and careful densitometric quantification has indeed indicated that MRPA gene copy number is increased in the mutants compared to wild-type cells (Fig. 6A). Increased gene copy number is usually due to the formation of extrachromosomal circular or linear elements [43], [44] although changes in copy number of whole chromosomes have also been reported [31], [32]. Search for extrachromosomal circles failed by standard alkaline lysis extractions and we thus relied on CHEF gels to separate the Leishmania chromosomes and investigated for the presence of short linear amplicons. Hybridization to a MRPA probe showed the presence of linear amplicons in Ba199SbIII2700.1 and 2700.2 while the whole chromosome 23 was increased in copy number in Ba199SbIII2700.3 and 2700.4 (Fig. 6B). These results are consistent with the microarray data (Figs. 2 and 3).
Resistance to antimony in Leishmania has been studied mostly in Old World species and mostly in strains in which resistance was induced under laboratory conditions (reviewed in [5], [42]). However, with a better understanding of in vitro resistance mechanisms, more work has recently been done with clinical isolates and some of the markers highlighted in in vitro studies were shown to correlate with drug resistance in clinical isolates [20], [45]. In general, there is a reasonable agreement between in vitro susceptibility testing and clinical response with Old World Leishmania when assays are carried out with intracellular parasites [45], [46], [47]. However, there are conflicting results in linking in vitro susceptibility testing and clinical responses with New World leishmaniasis [48], [49]. There have been few studies on mechanisms of resistance to antimony in New World parasites and we have thus used here the proven approach of in vitro selected resistant cells. Four independent L. amazonensis clones were selected for resistance to SbIII. Resistance was in general unstable when cells were grown in absence of the drug (Table 1), a result also recently observed with New World Leishmania selected for antimony resistance [28].
To find possible markers of resistance in these L. amazonensis strains, we carried out RNA expression profiling on full genomic DNA microarrays, a technique proven useful to study resistance mechanisms in Leishmania [31], [32], [50]. We found several gene candidates (Table S2 and S3), some for which the expression was highly modulated in comparison to sensitive isolates. Two of these genes (the hypothetical LmjF26.2680 and NT3) were new and were experimentally tested by gene transfection. However, we could not directly link them to resistance. NT3 and LmjF26.2680 were respectively overexpressed and down-regulated in four independent mutants (Fig. 4), and this recurrence would argue for some role in resistance. If it is not directly involved in resistance as the transfection work would suggest, it could either require another product to confer resistance or it may have another more indirect role such as in increased fitness or compensating for other mutations. We noticed that one glucose transporter in Ba199SbIII2700.3 was down regulated (Fig. 1). Decrease glucose uptake, for example by minimizing reactive oxygen species, was suggested as a general mechanism associated with drug resistance in L. amazonensis [51]. Future work will be required to test this. It is also worthnoting that while the expression of NT3 is increased, this is not due to gene amplification. Indeed, the NT3 copy number remains similar to wild-type (result not shown). While changes in expression in resistant isolates are often due to changes in gene copy number, there has been several other reports of increased expression by other means which will likely involve post-transcriptional regulation mechanisms. Indeed, the expression of genes in Leishmania is not controlled at the level of transcription initiation [52], [53].
The microarray work allowed detecting alterations of expression of large regions of genomic DNA and even of whole chromosomes (Figs. 2 and 3). In Leishmania these alterations are usually linked to changes in copy number [31], [32]. One region that attracted our attention was part of chromosome 23. Mutant Ba199SbIII2700.2 had a specific region that was overexpressed while the whole chromosome 23 seemed overexpressed in Ba199SbIII2700.3. The gene MRPA, one marker highly correlated to SbIII resistance in Old World Leishmania, is encoded by chromosome 23. We tested whether this increased expression was due to changes in copy number and Southern blot analysis indeed confirmed that MRPA is amplified (Fig.6A). New World Leishmania is divided in two subgenus: Leishmania and Viannia. Gene amplification is rare in the Viannia subgenus [30] and this may be due to an active RNA interference (RNAi) mechanism in this subgenus but absent in the Leishmania subgenus [54]. It is thus surprising that there is one report of a circular extrachromosomal amplification of MRPA in L. V. guyanensis selected for antimony resistance [55]. There is, however, ample report of gene amplification in the New World Leishmania subgenus whether it is L. amazonensis [56], [57] or L. mexicana [58]. No MRPA amplification has been observed in one L. amazonensis strain selected for SbIII resistance [59] but a circular amplification was observed in L. mexicana selected for resistance to the related metal arsenite [58]. The MRPA containing amplicon in L. mexicana or L. V. guyanensis corresponded to an extrachromosomal circle. In Ba199SbIII2700.2 the amplification was a linear amplicon and extended from the telomeric region to gene LmjF23.0540 (a region of ∼230 kb). All linear amplifications so far described, indeed extended to the telomeric region and are usually forming large inverted duplications [31]. This duplication of the region amplified fits with the size of this linear amplicon (Fig. 6B). Interestingly, we also found an MRPA containing linear amplicon in Ba199SbIII2700.1 (Fig. 6B). The amplicon is smaller, suggesting that a different rearrangement point, usually at the level of inverted repeats [31], [60] has been used. The microarray data indicate that the whole chromosome 23 was increased in Ba199SbIII2700.3 and this was corroborated by Southern blot analysis (Fig. 6). Indeed, the CHEF showed clearly that chromosome 23 had a higher hybridization intensity compared to Ba199SbII2700.2 (Fig. 6B). Interestingly, polyploidy of chromosome 23 was also observed in Ba199SbII2700.4. Intriguingly, this relatively modest increase in copy number was nonetheless correlated to a high MRPA expression at the RNA level (Fig. 4). Thus an increase in MRPA expression in L. amazonensis is correlated to either the formation of extrachromosomal linear amplicons or the increased ploidy of the chromosome.
This study has shown that mechanisms of resistance to antimony found in Old World Leishmania can also be detected in New World species. This includes higher thiol levels (Fig. 5) and increased expression of the ABC MRPA, whose gene product sequesters thiol-metal conjugates into an intracellular organelle [18]. Overexpression of several genes was found to correlate with increased thiols [15], [16], [50] and overexpression of spermidine synthase (leading to polyamines, one constituent of trypanothione) and trypanothione synthase (supplementary Tables S2 and S3) could contribute to the observed increased thiols. Also we noticed that trypanothione reductase was overexpressed and this would maintain thiols into a reduced form and this gene was found overexpressed in field isolates [61]. Many other genes were found to be differentially regulated although analysis of two candidates did not allow finding a role in resistance. Nonetheless with all microarray experiments done with several different species it should now be possible to perform meta-analysis which could direct at further candidates for a better understanding of antimony resistance mechanisms in the protozoan parasite Leishmania.
The study presented here should serve as a useful basis for analyzing antimony resistance in clinical isolates of new world leishmaniasis. Indeed, in vitro work mostly with the promastigote stage of old world leishmaniasis has led to a number of drug resistant markers [5], [42]. These markers were shown to confer resistance in the amastigote or intracellular stage of the parasite [40] and even more importantly in L. donovani field isolates [20], [61], [62], [63]. Since several markers were highlighted here with in vitro resistance in L. amazonensis, it would now be possible to test whether similar resistance mechanisms take place with drug resistant clinical isolates of New World leishmaniasis.
|
10.1371/journal.pntd.0006083 | Longitudinal assessment of anti-PGL-I serology in contacts of leprosy patients in Bangladesh | Despite elimination efforts, the number of Mycobacterium leprae (M. leprae) infected individuals who develop leprosy, is still substantial. Solid evidence exists that individuals living in close proximity to patients are at increased risk to develop leprosy. Early diagnosis of leprosy in endemic areas requires field-friendly tests that identify individuals at risk of developing the disease before clinical manifestation. Such assays will simultaneously contribute to reduction of current diagnostic delay as well as transmission. Antibody (Ab) levels directed against the M.leprae-specific phenolic glycolipid I (PGL-I) represents a surrogate marker for bacterial load. However, it is insufficiently defined whether anti-PGL-I antibodies can be utilized as prognostic biomarkers for disease in contacts. Particularly, in Bangladesh, where paucibacillary (PB) patients form the majority of leprosy cases, anti-PGL-I serology is an inadequate method for leprosy screening in contacts as a directive for prophylactic treatment.
Between 2002 and 2009, fingerstick blood from leprosy patients’ contacts without clinical signs of disease from a field-trial in Bangladesh was collected on filter paper at three time points covering six years of follow-up per person. Analysis of anti-PGL-I Ab levels for 25 contacts who developed leprosy during follow-up and 199 contacts who were not diagnosed with leprosy, was performed by ELISA after elution of bloodspots from filter paper.
Anti-PGL-I Ab levels at intake did not significantly differ between contacts who developed leprosy during the study and those who remained free of disease. Moreover, anti-PGL-I serology was not prognostic in this population as no significant correlation was identified between anti-PGL-I Ab levels at intake and the onset of leprosy.
In this highly endemic population in Bangladesh, no association was observed between anti-PGL-I Ab levels and onset of disease, urging the need for an extended, more specific biomarker signature for early detection of leprosy in this area.
ClinicalTrials.gov ISRCTN61223447
| Leprosy is an infectious disease caused by the bacterium Mycobacterium leprae, which causes skin and nerve damage. Despite worldwide efforts to eliminate leprosy, the number of infected individuals who develop leprosy, is still substantial. Household contacts of new leprosy patients are especially at risk. Early diagnosis of leprosy is key in preventing lifelong handicaps as well as transmission. This requires field-friendly tests that identify individuals at risk of developing the disease before they develop clinical symptoms so that they can receive (prophylactic) treatment. Measuring antibody levels directed against the M.leprae-specific phenolic glycolipid I (PGL-I) provides an indication of the bacterial load. To identify whether anti-PGL-I Ab levels correlate with the development of leprosy in contacts of newly diagnosed leprosy cases, we analyzed these levels in 25 contacts who developed leprosy during 6 years of follow-up and 199 contacts who were not diagnosed with leprosy at 3 time points in 6 years. This study showed that anti-PGL-I Ab levels at intake did not significantly differ between contacts who developed leprosy during the study and those who remained free of disease. Therefore, anti-PGL-I Ab levels alone do not represent a practical tool for prediction of which household contacts will develop leprosy in an endemic area such as Bangladesh, with high levels of patients with paucibacillary leprosy. This stresses the need for a diagnostic test composed of a biomarker signature consisting of multiple biomarkers.
| Leprosy is an infectious disease caused by Mycobacterium leprae (M. leprae), which causes damage to the skin and peripheral nerves[1]. The highest numbers of new leprosy cases are detected in India (127,326 in 2015), Brazil (26,395 in 2015) and Indonesia (17,202 in 2015)[2]. Bangladesh also has highly endemic areas, with a number of new cases of above 3,000 per year[2]. Although leprosy prevalence has decreased tremendously along with the widespread availability of multidrug therapy (MDT) in endemic areas, detection of new cases worldwide has shown only a modest decline in the last five years, and has stabilized in some countries[3]. In Bangladesh, the number of new cases was 3,976 in 2015, compared to 3,848 new cases in 2010[2]. Indirect evidence indicates that worldwide millions of unreported cases linger undetected as a gradual result of a decline in leprosy control activities after the disease was declared eliminated[1]. The continued transmission is probably largely due to M. leprae infected individuals, carrying substantial numbers of bacteria but (yet) lacking clinical symptoms. Thus, early detection and subsequent (prophylactic) treatment of asymptomatically infected individuals as well as subclinical disease is essential to reduce transmission.
Diagnosis of leprosy is still largely dependent on clinical signs and symptoms and detection of acid fast bacteria. However, user friendly lateral flow assays provide new possibilities for rapid diagnosis of leprosy patients in early stages of the disease or of M. leprae infected individuals without any symptoms[4, 5]. Such assays are likely to contribute to reduction of current diagnostic delay in endemic areas and also aid classification of leprosy disease, allowing appropriate treatment. Currently, there is no specific and sensitive test available that can detect asymptomatic M. leprae infection or predict progression to clinical disease[6]. In view of the long incubation time of leprosy (typically 3–5 years) as well as its low incidence, identification of predictive biomarkers requires longitudinal monitoring of M. leprae-specific immunity in those at risk of developing disease. Therefore, investments in large-scale longitudinal follow-up studies, allowing intra-individual comparison of immune profiles in leprosy patients’ contacts, is essential to evaluate which markers correlate with progression to disease and may be used as predictive biomarkers.
M. leprae phenolic glycolipid I (PGL-I) is an extensively studied antigen on the outer surface of the mycobacterium[7]. The existence of high levels of IgM antibodies to PGL-I[5–7], has allowed the development of several tests that were investigated broadly for diagnostic purposes[7–10]. Although useful for identifying multibacillary (MB) leprosy patients, anti-PGL-I antibody (Ab) titers have little value in detecting PB leprosy patients, since the latter develop cellular rather than humoral immunity and therefore often lack antibodies to PGL-I5.
In a previously conducted cluster randomized controlled trial, designated the COLEP study, the effect of single dose rifampicin versus placebo in preventing leprosy in close contacts of newly diagnosed leprosy patients was studied between 2002 and 2009 in a leprosy endemic area in the Northwest of Bangladesh[11, 12]. To investigate whether anti-PGL-I Ab seropositivity can be used as a predictive biomarker for progression to leprosy in contacts, the current study compared anti-PGL-I Ab levels of the prospective cohort at intake and at three time points covering six years of follow-up per contact.
Contacts of leprosy patients were voluntarily recruited as part of the COLEP study (a cluster randomized controlled trial) in 2002 and 2003 in the districts Rangpur and Nilphamari in the northwest of Bangladesh, which is a leprosy endemic area[11, 12]. Eligible participants (patients and contacts) were informed verbally about the study and invited to participate. Written consent was obtained from all participants at recruitment or from the parent or guardian of under 18s. Contacts were followed prospectively from 2002/2003 to 2008/2009 for the development of leprosy. Blood samples were collected by spotting on Whatman filter paper (Sigma) and subsequently stored at -80°C. Blood samples were collected at 4 time points: recruitment into the study, follow-up 1 (FU1; two years after intake), follow-up 2 (FU2; four years after intake) and follow-up 3 (FU3; six years after intake)[12]. Leprosy was diagnosed when at least one of the following signs was present: one or more skin lesions with sensory loss, thickened peripheral nerves, or a positive skin smear result for acid-fast bacilli. Patients with negative smear results and no more than five skin lesions were classified as PB leprosy, and those with a positive smear or more than five skin lesions as MB leprosy[12]. Clinical and demographic data was collected in the COLEP study database[11].
A random sample was taken from 28,092 contacts of leprosy patients recruited within the COLEP study[11]. A total of 239 contacts developed leprosy within the six years of follow-up. 25 contacts were included into this sub-study who were diagnosed with leprosy at either FU1, FU2 or FU3 and for whom filter papers of at least three different time points were available. Out of the contacts who did not develop leprosy, 199 were randomly included using the RAND formula (Excel 2010), aiming for an equal ratio of three age groups (0–14, 15–29, and 30+ years).
The COLEP study represents a unprecedented field trial for leprosy, because it includes valuable longitudinal analysis of contacts and thus is uniquely suited to identify the predictive value of biomarkers. However, the COLEP study did not collect blood samples from contacts as the only samples collected was blood on filter paper. Therefore, this limited biomarker analysis to anti-PGL-I Ab only.
In this part of the country, the new case detection rate of leprosy was 3.21 per 10,000 in 2002 (DBLM Annual Report 2002). In these cases leprosy was diagnosed by active and passive case detection. In 2002 and 2003 random samples from the general population were taken to calculate the prevalence of previously undiagnosed leprosy (PPUL). In the contact group of the COLEP study, the PPUL rate was 73/10,000, compared to 15.1/10,000 in the samples taken from the general population. These cases were found by active door-to-door screening[13].
Disaccharide epitope (3,6-di-O-methyl-β-D-glucopyranosyl(1→4)2,3-di-O-methylrhamnopyranoside) of M. leprae specific native PGL-I glycolipid was synthesized and coupled to human serum albumin (synthetic PGL-I; designated ND-O-HSA). This was generated with support from the NIH/NIAID Leprosy Contract N01-AI-25469 and obtained through the Biodefense and Emerging Infections Research Resources Repository (http://www.beiresources.org/TBVTRMResearchMaterials/tabid/1431/Default.aspx).
Antibodies (IgM, IgG, IgA) against M. leprae PGL-I were detected as described previously[5, 14, 15]. ND-O-HSA was coated onto high-affinity polystyrene Immulon 4HBX 96-well Nunc ELISA plates (Thermo Scientific, Rochester, NY) using 500 ng per well in 50 μl of 0.1M sodium carbonate/bicarbonate pH 9.6 (i.e. coating buffer) at 4°C overnight. Unbound antigen was removed by washing six times with PBS (phosphate buffered saline) with 0,05% Tween 20 (wash buffer). The wells were blocked with PBS containing 1% BSA (bovine serum albumin) (Roche Diagnostics, Germany) for 1 hour at room temperature (RT). Bloodspots were punched from filter papers. Three punches (2 mm each) per individual were added to 100 μl PBST (PBS/0,1% Tween20) and incubated at 4°C in 24 wells plates. After overnight incubation, 50 μl PBST/NRS (PBST + 10% normal rabbit serum) was added to each well and the plates were shaken gently for 1 hour at RT. The eluate was added to the ELISA plates (50 μl/ well) and incubated for 2 hours at RT. After incubating with the primary antibody, the wells were washed six times with PBS with 0.05% Tween 20 (wash buffer), followed by the addition of 50 μl of a 1:8,000 dilution of the secondary antibody anti-human (Dako P0212) for two hours. Following washing the wells with the wash buffer six times, 50 μl of p-nitrophenylphosphate substrate (Kirkegaard and Perry Labs, Gaithersburg, MD) was added. Antibodies (IgM, IgG, IgA) against M. leprae PGL-I were detected as previously described[14]. Absorbance was determined at a wavelength of 450 nm. Samples with an optical density at 450 nm (OD450), after correction for background OD above 0.150, were considered positive. This threshold was determined by a threefold multiplication of an average EC value.
As quality control, anti-PGL-I IgM levels were determined for 10 Dutch leprosy patients by ELISA using serum as well as blood spots on filter paper: Although IgM levels were higher for 9 individuals in sera, all seropositive individuals were also positive using blood spots and OD450 values correlated well (R2 = 0,80).
Multivariable logistic regression was used to calculate adjusted odds ratios for the level of ant-PGL-I Ab levels at intake, and corrected for age and sex. A p-value ≤ 0.05 was used as a cut-off for statistical significance. To investigate the association of changes in anti-PLG-I Ab from baseline to the time of development of leprosy, generalized linear mixed models were used. The dependent variable was the development of leprosy at a time point and the differences in anti-PLG-I Ab levels from baseline were included as independent variables. To adjust for the correlation between intra-individual measurements we included a random intercept for each subject. The difference between the anti-PGL-I Ab levels between contacts of MB or PB index patients was calculated using a t-test comparing averages. All analyses were performed in R version 3.2.0 (R, Vienna, Austria; https://www.R-project.org).
From the 28,092 contacts of leprosy patients recruited within the COLEP study[11], 239 contacts developed leprosy within the six years of follow-up. For 25 contacts who were diagnosed with leprosy during follow-up and 199 contacts who remained free of leprosy, good quality filter paper was available for at least three different time points during follow-up.
Characteristics of the study populations are shown in Table 1 and Table 2. Of the 25 contacts who developed leprosy, 10 contacts developed leprosy at 2 years after intake (FU1), 7 contacts at 4 years after intake (FU2) and 8 contacts at 6 years after intake (FU3). Four contacts (16%) developed MB leprosy and 21 (84%) developed PB leprosy. This is the same proportion of MB versus PB as in the total group of new leprosy cases diagnosed within the COLEP study[12] 4 years after intake (24 MB contacts versus 126 PB contacts; 16% versus 84%). The group was evenly distributed for sex (M/F = 1.17:1) and age categories. For 10 contacts the index patient had MB leprosy, whereas for 15 contacts this was PB leprosy.
The anti-PGL-I Ab levels at intake were compared between the two groups of contacts (Fig 1). In the group of contacts who developed leprosy, the average anti-PGL-I Ab titer at intake was 0.11, and varied between zero and 0.424. 6 of these 25 (24%) contacts who developed leprosy had a positive anti-PGL-I Ab level of >0.150 at intake. In the group who did not develop leprosy, the average anti-PGL-I Ab titer was 0.10 and varied between zero and 1.275. 35 out of 199 (17.6%) contacts who did not develop leprosy had a positive anti-PGL-I Ab level of >0.15 at intake. No significant association was observed for the anti-PGL-I Ab levels at baseline (OR: 1.01 (0.78, 1.31), 95% CI p = 0.94) between the two groups.
To further analyze the longitudinal pattern of PGL-I serology in contacts, the anti-PGL-I Ab levels are depicted at different follow-up times, comparing the titers of contacts developing leprosy (Fig 2A) to the titers of contacts without leprosy (Fig 2B). The difference between anti-PGL-I Ab level at diagnosis was compared to the anti-PGL-I Ab level at intake. This difference was minus 0.047, indicating that the level of anti-PGL-I Ab titer was lower at time of diagnosis compared to time of intake. Next we calculated the difference between anti-PGL-I Ab titer at various time points of follow-up to the anti-PGL-I Ab level at intake of the contacts who did not develop leprosy using a generalized linear mixed model analysis. Thus, for all contacts who did not develop leprosy, we compared the anti-PGL-I Ab level at FU1 to intake, the level of FU2 to intake and the level at FU3 to intake. If a contact developed leprosy at FU2 (or FU3), we also included the difference between anti-PGL-I Ab titer at FU1 (and FU2) and intake into the group of contacts who did not develop leprosy. Differences in anti-PGL-I Ab levels had no significant association with the development of leprosy at either of the three follow-up time points (OR: 0.62 (0.15, 2.62), p = 0.52). Thus changes in anti-PLG-I Ab levels are not predictive of disease progression in contacts of new leprosy patients in Bangladesh. Since MB patients harbor a higher quantity of bacteria than PB, we separately considered the longitudinal pattern of the anti-PGL-I Ab levels in the four contacts who developed MB leprosy (Fig 2C). The mean OD450 at the time of diagnosis for both MB/BT and PB patients was below threshold for positive (< 0.15). Also, no increase in anti-PGL-I Ab levels was observed at the moment of leprosy diagnosis; actually, anti-PGL-I Ab levels were often even lower at diagnosis time compared to intake. The findings indicate that not only for newly diagnosed PB, but also for MB patients, anti-PGL-I Ab levels do not represent a practical tool for prediction of leprosy.
Although several studies described that positive anti-PGL-I Ab titers in household contacts of leprosy patients were related to a higher risk of developing leprosy[16–19], reports also indicated that more than half of the individuals with antibodies against PGL-I will never develop leprosy[16, 17]. Besides, diagnosis based only on seropositivity for anti-PGL-I Abs would leave more than half of the new leprosy cases undetected[16, 18, 19]. To study the value of anti PGL-I Ab as a predictor of leprosy in those at risk of developing leprosy in a highly endemic area, we here analyzed the anti PGL-I Ab levels in the blood of 224 contacts of leprosy patients in the Northwest part of Bangladesh. However, no association was found between anti-PGL-I Ab levels and onset of disease in this population.
As part of a variety of studies investigating the use of serology for prediction of leprosy in those at risk of developing disease, a study in the state of Minas Gerais, Brazil[18] suggested that anti-PGL-I serology in household contacts of leprosy patients can be used to identify leprosy at a preclinical stage. This study identified more contacts with suspected leprosy in the group with positive anti-PGL-I levels (9.62%) than in the test-negative group (1.76%). However, out of the 52 contacts with positive anti-PGL-I serology, only 5 had leprosy. The anti-PGL-I seropositivity was higher in those contacts exposed to patients with MB leprosy than PB leprosy, which is probably due to the higher bacterial load in MB patients and therefore higher exposure rates of their contacts.
In another Brazilian study[17], performed in Rio de Janeiro, leprosy diagnosis had a strong association with anti-PGL-I seropositivity at intake. A significantly higher proportion of healthy contacts with anti-PGL-I Abs (5.6%) developed leprosy during the follow-up period compared with those without (2.3%). Anti-PGL-I seropositive contacts had a 3.2-fold higher risk of developing leprosy compared with seronegative contacts.
A third study performed in Cebu (the Philippines)[16] showed that household contacts of MB leprosy patients with anti-PGL-I Abs have a 7.65-fold-higher risk of developing leprosy in the six years of active surveillance than seronegative contacts. It is noteworthy that out of the 27 contacts developing leprosy, 13 remained seronegative, indicating that half of the new leprosy cases would not be detected when solely anti-PGL-I serology would be used as a predictive diagnostic tool. This particularly applies to PB cases, as all of the 10 newly diagnosed MB patients were or became seropositive. On the other hand, 85 out of the 99 anti-PGL-I Ab positive contacts never developed leprosy, implying a false positivity rate of 86% when using anti-PGL-I serology as a predictive marker for leprosy.
Barretto et al.[20] showed that the odds of seropositive versus seronegative school children developing leprosy within two years is 2.7 times higher in an hyperendemic region in the Amazones of Brazil (State of Pará). Thus, this would indicate a > 90% probability of detecting at least one new case among 10 seropositive individuals in 2 years. On the other hand, 5 of 11 new cases found amongst school children in these high-risk areas in Brazil tested negative for anti-PGL-I Abs. Furthermore, no significant difference between the median anti-PGL-I Ab titer of new cases and of healthy school children was observed. Of note is that a significant increase in the anti-PGL-I IgM titers was found at the time of diagnosis compared to intake. The group that did not develop leprosy also demonstrated an increase in their average antibody titers, although the most significant increase was observed in the group that developed disease. These findings in Brazil stand in contrast to our current study in Bangladesh, in which hardly any difference or even a slight decrease in the anti-PGL-I Ab levels was observed in the contacts who developed leprosy.
A recent meta-analysis among household contacts of new leprosy patients in French Polynesia, Zaire, Papua New Guinean, Venezuela, Brazil, India and Philippines[19] shows that the risk of developing leprosy is about three times higher in those who are positive for anti-PGL-I Abs compared to the seronegative group, with the odds ratio varying from 2.72 to 3.53. However, the sensitivity of anti-PGL-I Ab tests as predictor of the development of clinical leprosy was found to be lower than 50% in all studies. Thus, selecting contacts with anti-PGL-I antibodies for prophylaxis, although possibly beneficial for reduction of transmission, would only prevent less than half of the leprosy cases among contacts. Our findings in contacts in Bangladesh are in line with those of the meta-analysis by Penna et al.[19] as well as the other studies discussed above, since development of leprosy was not associated with the level of anti-PGL-I seropositivity at intake, clearly indicating that also in Bangladesh anti-PGL-I Ab tests lack the ability to early diagnose leprosy amongst leprosy contacts[21–23], if used as a stand alone tool.
Most of the leprosy patients’ contacts in our study developed PB leprosy (21 out of 25), which offers an explanation for the lack of increase of anti-PGL-I titers at leprosy diagnosis. Importantly, in Bangladesh, the percentage of PB cases amongst new leprosy cases is generally higher than in other countries in Asia, especially southeast Asia where predominantly MB patients are found[2]. This phenomenon is probably due to a combination of genetic factors as well as early case detection. Bangladesh is a high endemic area with a high rate of active case-finding, which leads to a lot of PB cases being found. In contrast, low endemic areas with little active case-finding have higher numbers of MB cases, since PB is often self-healing. In our study, only four household contacts developed MB leprosy (out of the 25 total number of new leprosy patients). PB leprosy in general is characterized by low levels or absence of antibodies against M. leprae antigens[16], which is in line with our finding that there was no significant difference in anti-PGL-I Ab level at intake compared to leprosy diagnosis. Schuring et al[24] found that anti-PGL-I seropositivity was associated with bacterial index (BI). However, most contacts in our study had PB and therefore an undetectable BI. Separate evaluation of the four MB patients did not show any differential increase in anti-PGL-I Ab level in this group either. This is in line with the findings of van Hooij et al[4, 25], showing low levels of anti-PGL-I Ab in all patients, including MB. Moreover, anti-PGL-I IgM levels could not be used to discriminate PB patients or household contacts from endemic controls. In leprosy endemic countries other than Bangladesh, where MB leprosy is more prevalent, the longitudinal pattern of anti-PGL-I Ab levels could hold more diagnostic value. Besides this, anti-PGL-I antibodies can represent a useful tool for monitoring effectiveness of treatment of leprosy (reactions), since effective treatment is associated with decrease in antibody levels[26].
As a part of the COLEP trial, half of the new leprosy contacts received placebo and the other half single dose rifampicin. It can be expected that single dose rifampicin could lower the anti-PGL-I antibody level in subjects with a relatively high bacterial load. However, it is unknown how soon and to which extent the antibody titre is suppressed. Furthermore, there is also the possibility that subjects become re-infected with M. leprae due to continued exposure to an unknown source. Also, in the absence of complete ‘sterilisation’ of M. leprae in these subjects, the bacterium may start to multiply again after the effect of rifampicin has waned. So although the antibody titre may certainly have decreased due to single dose rifampicin, it is unknown whether this effect would be apparent after 2 years, at the moment of first blood sampling.
Furthermore, it is worthy to note that leprosy is a complicated disease with different immunological processes playing a role in disease progression, which in turn are affected by factors such as genetics[27], co-infections[28] as well as food-shortage[29]. The combination of these factors with the long incubation time that elapses before leprosy becomes clinically manifest, makes predicting which M. leprae exposed individuals will progress to disease complicated. For example, certain helminth-derived proteins can bias the host immune response towards an anti-inflammatory Th2 response, which may facilitate M. leprae growth or progression to MB leprosy[28]. Furthermore, a period of food shortage can reduce cell mediated immunity of individuals incubating M. leprae, causing the development of clinical disease[29].
Recent advancements in leprosy biomarker research[15] have shown that IFN-γ responses measured after stimulation with leprosy-unique antigens can be used as a measure for M. leprae exposure. In particular, the combination of humoral and cellular biomarkers increased efficiency to distinguish M. leprae infected from non-infected individuals, patients from contacts, or lepromatous from tuberculoid patients compared to serology alone[4, 15]. In view of the findings in this study as well as our previous studies on cellular biomarkers[4, 15, 26, 30], field-friendly tests using a biomarker signature would improve identification of contacts who are at risk of developing leprosy as well as asymptomatic, infected individuals who can transmit bacteria. In current longitudinal studies on biomarker identification, a new lateral flow test format is used (UCP-LFA)[4, 25], that not only allows field-use but also provides a permanent record as the luminescent signal on the LF strips does not fade. Such tests would represent a useful contribution to current pilot studies on the effectiveness of SDR as leprosy post-exposure prophylaxis (LPEP)[31], allowing more selective targeting for prophylaxis as well as preventing overtreatment.
In view of the dichotomy of the leprosy spectrum in terms of immunity against M. leprae, current research is focused on identification of predictive biomarker profiles associated with early stage leprosy, consisting of multiple cellular and humoral (disease-specific) biomarkers. Early diagnosis of leprosy and subsequent appropriate multidrug therapy (MDT) will not only decrease severe nerve damage and subsequent lifelong handicaps, but also significantly contribute to further decrease of M. leprae transmission. This study shows that measurement of anti-PGL-I Abs alone is not sufficient to predict the development of clinical leprosy amongst household contacts of newly diagnosed leprosy cases in (highly) endemic area such as Bangladesh. Because of the high number of PB patients in Bangladesh, using anti-PGL-I titers as a screening test to discriminate which contacts to treat, may lead us to miss a lot of potential new cases.
Ethical clearance was obtained from the Ethical Review Committee of the Bangladesh Medical Research Council in Dhaka (ref. no. BMRC/ERC/2001-2004/799). All subjects were informed verbally in their own language (Bengali) about the study when they were invited to participate. Written consent was received from each adult, while a parent or guardian had to sign the consent form for children who participated in the study.
|
10.1371/journal.pgen.1003452 | Sensory Neuron-Derived Eph Regulates Glomerular Arbors and Modulatory Function of a Central Serotonergic Neuron | Olfactory sensory neurons connect to the antennal lobe of the fly to create the primary units for processing odor cues, the glomeruli. Unique amongst antennal-lobe neurons is an identified wide-field serotonergic neuron, the contralaterally-projecting, serotonin-immunoreactive deutocerebral neuron (CSDn). The CSDn spreads its termini all over the contralateral antennal lobe, suggesting a diffuse neuromodulatory role. A closer examination, however, reveals a restricted pattern of the CSDn arborization in some glomeruli. We show that sensory neuron-derived Eph interacts with Ephrin in the CSDn, to regulate these arborizations. Behavioural analysis of animals with altered Eph-ephrin signaling and with consequent arborization defects suggests that neuromodulation requires local glomerular-specific patterning of the CSDn termini. Our results show the importance of developmental regulation of terminal arborization of even the diffuse modulatory neurons to allow them to route sensory-inputs according to the behavioural contexts.
| Serotonin, a major neuromodulatory transmitter, regulates diverse behaviours. Serotonergic dysfunction is implicated in various neuropsychological disorders, such as anxiety and depression, as well as in neurodegenerative disorders. In the central nervous systems, across taxa, serotonergic neurons are often small in number but connect to and act upon multiple brain circuits through their wide-field arborization pattern. We set out to decipher mechanisms by which wide-field serotonergic neurons differentially innervate their target-field to modulate behavior in a context-dependent manner. We took advantage of the sophisticated antennal lobe circuitry, the primary olfactory centre in the adult fruitfly Drosophila melanogaster. Olfactory sensory neurons and projection neurons connect in a partner-specific manner to create glomerular units in the antennal lobe for processing the sense of smell. Our analysis at a single-cell resolution reveals that a wide-field serotonergic neuron connects to all the glomeruli in the antennal lobe but exhibits the glomerular-specific differences in its innervation pattern. Our key finding is that Eph from sensory neurons regulates the glomerular-specific innervation pattern of the central serotonergic neuron, which in turn is essential for modulation of odor-guided behaviours in an odor-specific manner.
| Serotonin, 5-hydroxytryptamine (5-HT), an evolutionarily ancient monoamine, plays diverse roles in the brain [1], [2], [3]. In the mammalian brain, serotonin is implicated in the regulation of behavioural arousal and control of motor output [4], [5] with a proposed phylogenetically ancient function in modulating a drive to withdraw from dangerous and aversive environments and seek contentment [6]. In the fruitfly, Drosophila melanogaster, serotonin regulates diverse aspects of behaviour such as aggression, sleep, circadian rhythm, learning and memory [7], [8], [9], [10], [11]. It is estimated that there is one serotonergic neuron per million in the mammalian central nervous system, yet, when axon terminals are examined in the rat cortex, as many as 1/500 are serotonergic [2], suggesting that a small set of neurons may act through their broad arborization pattern to play roles in modulating many brain circuits. Understanding how serotonin and other neuromodulators function to modify intrinsic dynamic properties of neuronal circuits and thereby alter animal behaviour, is a daunting task. An iconic preparation in which this has been carried out is the circuit that drives pyloric rhythm in the crab/lobster stomatogastric system [12], [13]. Such studies have led to the view that understanding the function of brain circuits not only requires a characterization of intrinsic dynamic properties of constituent neurons and their connectivity but also an understanding of how specific neurotransmitters and neuromodulators impinge on the circuit [14].
Functional imaging and electrophysiology suggests that serotonergic modulation of olfactory information is an important conserved feature [15], [16], [17]. In the Drosophila antennal lobe (AL), innervated by ∼2500 olfactory sensory neurons (OSNs), ∼150 projection neurons (PNs), and ∼200 local interneurons (LNs), the CSDn is the sole serotonergic neuron [18], [19], [20]. This and its accessibility to genetic manipulation [18], [21] allow the development of the capacity for serotonergic modulation to be studied in the context of the well-characterized olfactory glomerular system.
While the CSDn's axonal terminals spread over multiple glomeruli in the adult AL [18], it also exhibits glomerular-specific differences in innervation pattern (this study). Such wide-field arborizations, with variations in specific glomeruli, are seen in multi-glomerular olfactory LNs [22], [23], but the underlying mechanisms that regulate these arborizations have not been studied. This is in contrast with the many elegant studies that have led to significant understanding of mechanisms underlying targeting of the uni-glomerular OSNs and PNs [24], [25], [26]. The glomerular-specific pattern of wide-field interneurons is also likely to be important for their function as context- specific modulators of olfactory information, a hypothesis that has not been tested. Serotonergic neurons have been suggested to act in a paracrine manner: serotonin-containing varicosities release serotonin that can diffuse away and act on extra-synaptically located receptors [27]. While the arbors of such diffuse neuromodulatory neurons are suggested to be distributed to optimize efficient coverage of brain regions, the heterogeneous distribution of the terminal arbors of the CSDn in the AL suggests the possibility that arborization in a specific glomeruli is an important functional feature and could be behaviourally relevant, a view which we test and show to be valid.
In searching for the mechanistic underpinning of the CSDn's terminal aroborization pattern we homed in on Eph-ephrin signaling as a likely candidate. Eph receptors (Eph) form the largest family of receptor tyrosine kinases (RTKs) and mediate contact-dependent bidirectional communication between cells through short-range interactions [28], . Such short-range interactions between axonal arbors and their target cells could be relevant for emergence of regional differences in the arborization pattern of neurons in the CNS. We find that an Eph/ephrin signaling-mediated repulsion plays a key role in glomerular-specific positioning of axonal terminals of the CSDn. Sensory neurons differentially express Eph, which interacts with Ephrin on the CSDn to establish glomerular-specific innervation pattern of the CSDn axonal terminals. Further, we show that this glomerular-specific innervation pattern of the CSDn allows it to modulate olfactory behaviour in an odor-specific manner.
We have determined the function of the CSDn in modulating odor-guided behaviour and shown that its glomerular-specific modulatory properties are dependent on the developmental regulation of its terminal arborization. Since the CSDn is the only serotonergic neuron in the AL, our study behaviourally dissects out the role of this important neuromodulator in the olfactory system and shows, for the first time, how its function is developmentally put in place. Our results also point to how sensory neurons, which are targeted to specific glomeruli, could locally regulate terminal arbors of other wide- field neurons. Finally, we examine Eph-ephrin signaling at the resolution of a single neuron, for the first time, to show how short-range signaling can sculpt local pattern, and thereby, function.
We had earlier characterized the development the CSDn in Drosophila [18], [21]. In these studies, the CSDn [18] is labeled using a combination of cis-FRT/FLP and Gal4/UAS method [31], [32]. This method can result in activation of CD8::GFP reporter protein expression in the CSDn in one antennal lobe, while the neuron on the contralateral side remains unlabeled, thereby allowing the examination of its arbors without the pattern being obscured by its homolog in the other hemisegment. Although the CSDn's terminal arbors in the contralateral AL innervate all glomeruli [18], a closer examination showed clear glomerular-specific differences in the innervation pattern (Figure 1A, 1E). We focused on glomeruli whose function in olfactory perception is well established in behavioural assays allowing us to correlate connectivity of the CSDn with its function in modulating behaviour. We therefore analyzed the VA1d, DA1, VA1l/m, DL3, which respond to fly- derived odors [33]. Of these, sensory neurons innervating DA1 and DL3 respond to the pheromone cis-vaccenyl acetate - cVA [33], [34], [35]. We also examined the V glomerulus, which responds to Carbon dioxide (CO2) [36], [37]. Quantification of axonal branch tip number of the CSDn in these glomeruli demonstrated prominent glomerular-specific differences in its innervation pattern: VA1d and V were innervated by many arbors while DA1, VA1l/m and DL3 received fewer inputs from the CSDn (Figure 1A, 1E; Figure S1 and Table S1). In order to understand the cellular and molecular mechanism(s) underlying such differences in innervation pattern of the wide-field neuron we analyzed the possible role of signaling molecules and observed a clear disruption of this pattern in Ephrin hypomorphs (Figure 1B, 1F; Figure S1 and Table S1). Axonal branch tip number increased dramatically in DA1, VA1l/m and DL3 glomeruli of Ephrin hypomorphs while innervations to glomeruli VA1d and V is comparable to controls (Figure 1B, 1F; Figure S1 and Table S1): The glomeruli that normally had fewer arbors of the CSDn (DA1, VA1l/m and DL3) were densely innervated in Ephrin hypomorphs, whereas arbors in densely innervated glomeruli (VA1d and V) remained unchanged in this mutant. Further, CSDn-specific expression of Ephrin rescued glomerular-specific innervation pattern defects observed in Ephrin hypomorphs (Figure 1C, 1D, 1G; Figure S1 and Table S1) suggesting that Ephrin is required autonomously in the CSDn although it is widely expressed in the developing AL (Figure 1H–1L). Overexpression of Ephrin in the CSDn did not change overall pattern of axonal branch tip distribution although a small decrease in final branch tip number was observed (Figure 1C, 1G; Figure S1 and Table S1). This reduction in the overall branch tip number could either be due to increased Eph-mediated repulsion or due to other as yet unknown molecular interactions within the AL.
While Ephrin was required in the CSDn for positioning its terminal arbors in a glomerular-specific manner (Figure 1A–1D and 1F–1G), expression analysis showed that it is uniformly distributed in the developing AL (Figure 1H–1L) and thus may not provide the positional information for glomerular-specific branching. We therefore examined the expression of Eph, the receptor for Ephrin, in the developing AL. Interestingly, Eph expression, as revealed by an Ephrin-Fc probe [38], was detected in a small subset of glomeruli within the developing AL from 50 h after puparium formation (50 hAPF; Figure 2A–2D). Most prominent Eph expression was detected in DA1, VA1l/m and DL3 glomeruli. These are the same glomeruli that receive fewer arbors of the CSDn in control animals and show substantial increase in innervation by the CSDn in Ephrin hypomorphs. The observation of commissural expression of Eph (arrow in Figure 2C and 2E) along with the above glomerular specific pattern suggests that the OSNs are the source of Eph. Consistent with this interpretation, targeted expression of EphRNAi in sensory neurons (pebbled-Gal4/+; UAS EphRNAi/+) abolished Eph expression in the AL (Figure 2E–2F; Figure S2). Targeted misexpression of Eph in sensory neurons (pebbled-Gal4/+; UAS Eph/+) lead to Ephrin-Fc labeling in the whole AL, further validating the specificity of the Ephrin-Fc probe (Figure S2). Targeted expression of the EphRNAi in the projection neurons or in the local interneurons did not affect glomerular-specific Eph expression (data not shown). Furthermore, in amos mutant animals, of the genotype amos1/Df(2L)M36F-S6 [39], which lack most OSNs , the AL expression of Eph is also substantially reduced (Figure 2G–2H). Taken together, we conclude that Eph is expressed by a small set of sensory neurons and enriched in cognate glomeruli that received reduced arbors of the CSDn compared to other glomeruli where Eph levels are low.
Ephrin expressed by the CSDn may initiate repulsive interactions upon encountering high levels of Eph on sensory neurons. This hypothesis predicts that high levels of Ephrin ectopically expressed in other interneurons in these glomeruli would result in their arbors being repelled by high Eph expression. To test this hypothesis, we overexpressed Ephrin in PNs and focused our analysis on their arbors in the high Eph-expressing VA1l/m glomerulus, visualized using the Or47b::rCD2 strain (Gal4-GH146,UASmCD8::GFP; Or47b::rCD2; UASEphrin). Indeed, targeted overexpression of Ephrin in PNs resulted in a drastic reduction of PN innervations in the VA1l/m glomerulus (Figure 2I–2J), consistent with the view that Eph-ephrin signaling mediates a repulsive interaction within the developing AL. Similar effect of Ephrin misexpression on PN arborization was observed in other high-Eph expressing glomeruli, DL3 and DA1 (Figure S3). This suggests that under normal circumstances, CSDn-derived Ephrin could interact with sensory neuron-derived Eph to appropriately position terminals of the CSDn in a glomerular-specific manner.
In order to directly assess the role of sensory neuron-derived Eph, we used a combination of Gal4/UAS and LexA/lexAOp dual expression system. We generated RN2flp, tub>stop>LexA::VP16; lexAOpCD2GFP line which labels the CSDn (Figure 3A) and showed a clear glomerular-specific arborization pattern similar to that seen in the GAL4 reporter (Figure 3B, 3D). OSN-specific knockdown of Eph, achieved by targeted expression of EphRNAi in OSNs driven by the pebbled-Gal4, leads to increased innervation of CSDn in DA1, VA1l/m glomerulus (Figure 3C, 3D; Figure S1) similar to the phenotype that we observed in Ephrin hypomorphs (Figure 1). Such a change was also seen for DL3 glomerulus (Figure 3D, Figure S1). These results implicate OSNs in a previously unknown role in the development of a central neuron through their regulated expression of Eph. OSN terminals enter the lobe at 22 h APF and are key components of glomerular development [40]. OSN expression of Eph in the developing antennal lobe becomes prominent after 50 hAPF (Figure 2A–2D). To further validate the role of OSNs in CSDn patterning, we examined the CSDn arborization pattern in animals developing without antennae [41] and thus without the antennal OSNs (Figure 3F) or in animals in which antennae are transformed to legs (Figure 3H). In both the cases, the innervation pattern of CSDn in the antennal lobe was uniform (Figure 3F, 3H), unlike control animals where axonal terminals exhibited glomerular-specific differences in the innervation pattern (Figure 3E, 3G). Taken together, these data substantiate a role for OSNs in providing positional cues necessary for glomerular-specific arborization patterning of an identified central serotonergic neuron.
Eph-ephrin interactions can lead to diverse outcomes in terms of attraction, repulsion and cell adhesion in a context-dependent manner. High affinity Eph/ephrin signaling is known to initiate contact-dependent repulsion while low level signaling can lead to attraction and directed neuronal branch extension [42], [43], [44], [45]. We further investigated how Eph/ephrin signaling levels could control the final arborization pattern of the CSDn. To achieve a complete loss of Eph-ephrin signaling we utilized an allele EphX652 in where Eph expression is completely abolished [38]; Figure S4. Since Eph is expressed in the OSNs, we first tested the role of Eph during the development of OSNs and projection neurons (PNs), the primary synaptic partners of the OSNs. Terminals of OSNs (Figure 4A–4F) and uniglomerular PNs (Figure 4I–4J) develop normally in Eph null animals suggesting that Eph is not necessary for development of these components of the AL circuit, which have uniglomerular projections. Next, we asked if misexpression of Eph in the majority of the OSNs during a time window when Eph is expressed in very few glomeruli would affect OSN patterning in the AL. To this end, we used Or83bGal4 [46], which drives Gal4 expression in ∼80% of the OSNs starting from mid-metamorphosis. Misexpression of Eph using Or83bGal4 did not affect OSN patterning in the AL (Figure 4G, 4H). Overall, these observations allow us to argue that Eph signaling does not play any obvious role in OSN/uniglomerular PN patterning within the AL.
Surprisingly, terminal innervations of CSDn were reduced in animals homozygous for EphX652 to all the glomeruli examined (Figure 5B, 5H and Table S1). This was in marked contrast to the situation where Eph-ephrin signaling was not completely abolished but only reduced in the Ephrin hypomorphs (Figure 1B) or where Eph was knocked down specifically in the OSNs (Figure 3C). The CSDn innervation pattern was differentially affected in the latter cases and glomeruli with normally less innervations showed a substantial increase, leaving the densely innervated glomeruli unaffected. These differences in phenotypes indicate a requirement of Eph signaling at multiple stages of the CSDn development. Complete loss of Eph throughout development might influence overall branching and hence we observed reduced arborization of the CSDn in Eph null. On the other hand, OSN-derived Eph controls glomerular-specifc innervation of the CSDn during pupal stages. In any event, our observations suggest a key role for Eph/ephrin pathway in patterning axonal terminals of the CSDn. To further test this, we ectopically expressed Eph in the CSDn. Targeted ectopic expression of Eph in the CSDn resulted in striking reversal of axonal branch tip distribution in the glomeruli (Figure 5C, 5I and Table S1). Axonal terminals of Eph-expressing CSDn preferentially innervated glomeruli with high Eph and completely avoided VA1d glomerulus, which expresses low Eph (Figure 5C, 5I). This exquisite mistargeting further strengthens the suggestion that levels of Eph/ephrin signaling control glomerular-specific innervation of this serotonergic neuron. One possibility is that preferential targeting to high Eph-expressing glomeruli could be due to attractive homotypic interactions between Eph expressing neurons. Eph-mediated homotypic interactions have been shown to promote cell adhesion between Eph-expressing cells during rhombomere-boundary formation in zebrafish [47]. Another possibility, not excluding the first, is that Eph-ephrin interaction within CSDn could result in ‘cis inhibition’ [28], [48] of the signaling pathway due to simultaneous presence of Eph and ephrin in the same cell, which in turn could reduce repulsive interaction and increase the attractive one.
We next examined if the developmental timing of the CSDn arborization is consistent with OSN derived Eph playing a role in the process. Glomerular-specific innervation of the CSDn involves directed growth of terminals to the target glomeruli. At 50 h after puparium formation (APF), very few arbors of CSDn were seen in the regions of the antennal lobe where VA1l/m, VA1d, DA1 and DL3 glomeruli were developing (Figure 5D). An adult-like pattern was seen by 70 h APF without an intermediate stage where excess arbors were seen (Figure 5E). Terminals of the CSDn failed to innervate these glomeruli in Eph null animals (Figure 5F–5G). The time course of the development of glomerular-specific arborization of the CSDn coincided with the expression profile of Eph, described above and is consistent with a role for Eph/ephrin pathway as regulators of this process. These observations demonstrate that the final arborization of the CSDn is not an outcome of excess growth in every glomerulus, followed by pruning but is an outcome of the repulsive signaling operating in high-Eph expressing glomeruli, which restrict the growth of CSDn terminals during development.
We next examined if the extent of glomerular-specific arborizations of the CSDn has functional implications in behaving animals. To address this, an understanding of the role of the CSDn in odor-guided behaviours in Drosophila is first required. The CSDn is the only identified source of serotonin in the Drosophila AL [18], [49] suggesting an important role for this neuron in modulating olfactory perception. Although functional imaging studies have demonstrated that serotonin can change response properties of neurons in the AL [16], a direct demonstration of behavioural requirement of this neuron is lacking. We used the R60F02Gal4 strain [50] which consistently labels the CSDn bilaterally in the adult brain (Figure 6A), providing an advantage over the cis-FRT/FLP method, for behavioural analysis. R60F02Gal4's restricted expression in the central brain, with prominent expression in the CSDn and only a few arborizations in the suboesophageal ganglion provides an excellent reagent for behavioural experiments (Figure 6A). We validated that R60F02Gal4 indeed labels the CSDn in two ways. Firstly, the anatomy of its projections (Figure 6A, 6Ai and 6Aii) was similar to the described characteristic anatomy of the CSDn [18]. Furthermore by examining serotonin immunoreactivity in a genetic background where R60F02Gal4 expresses GFP, it was found that the only serotonin positive neuron in the AL co-localized with the GFP (Figure 6Aiii–vi) confirming that the Gal4 indeed specifically labels the CSDn.
For behavioural analysis, we selected two odorants; CO2 (perceived by the low Eph-expressing V glomerulus) and cVA (perceived by the high Eph-expressing DA1 and DL3 glomeruli) as innervations of the CSDn in the cognate glomeruli have been characterized by us. The behavioural response of wild-type adult Drosophila towards these odorants and the underlying neural circuitry is understood in good detail [34], [36]. CO2 is a repulsive stress pheromone in flies and is sensed by the V glomerulus [36]. Blocking evoked neurotransmitter release from the CSDn by targeted expression of tetanus neurotoxin light chain [TNTG; 51] rendered animals behaviourally more sensitive towards CO2 and these animals exhibited increased repulsion to CO2 compared to controls (Figure 6B, p = 0.017). Further, suppressing excitability of the CSDn by ectopic expression of an inward rectifying human K+ channel, Kir2.1 [52] in the neuron resulted in an increased CO2 avoidance behaviour (Figure 6C, p<0.01). Perturbation of neuronal activity during development has known consequences on the dendritic pattern of the CSDn [18], [21] and could be argued that this affects the behaviour. In order to circumvent the behavioural effects deriving from a developmental requirement of neural activity we manipulated the CSDn activity only during adulthood by using the temperature-sensitive Gal80 repressor of Gal4 (Gal80ts) [53]. Adult-specific suppression of the CSDn excitability by overexpression of the Kir2.1 in adult flies lead to increased CO2 sensitivity (Figure 6D) suggesting that the CSDn function in modulating olfactory behaviour is required during adulthood. In order to further validate the view that behavioural defects are indeed through serotonin signaling, we analyzed the expression pattern and function of serotonin receptors in the AL and then manipulated them. A Gal4 reporter line for serotonin receptor 5-HT1BDro [9] labels a small set of local interneurons in the adult AL (Figure 6E) suggesting that these neurons could be possible downstream target of serotonin released by the CSDn. RNAi-mediated knock down of 5-HT1BDro [9] in 5-HT1BDro expression domain lead to an increase in CO2 avoidance behaviour (Figure 6F). However, 5-HT1BDro is also expressed in the mushroom body neurons [9], which are a crucial component of the olfactory circuit underlying olfactory learning and memory [54], [55]. In order to define better, the domain of 5HT1BDro expression relevant in mediating CO2 avoidance behaviour, 5HT1BDro levels were ‘knocked-down’ using an RNAi construct [9] driven by the 5-HT1BDro-Gal4 driver in a context where Gal80 repressor of Gal4 is expressed under a mushroom-body promoter [56]. These animals will have normal 5HT1BDro in the mushroom body neurons, due to Gal80 repressing GAL4 expression in this tissue, but lowered expression in the olfactory local interneurons due to RNAi. Behavioural experiments show that these animals exhibit an increased CO2 avoidance behaviour. Taken together, these observations suggest that the CSDn releases serotonin as a neuromodulatory transmitter and serotonergic receptor-expressing local interneurons play an important role in CO2 sensitivity.
Next, we tested the role of the CSDn in cVA-dependent courtship behaviour. cVA, a male pheromone, is transferred to females during mating and renders them less attractive to other males in subsequent encounters. Virgin males therefore, show reduced courtship towards cVA-treated females [35]. The males sense the presence of cVA through OSNs that target to DA1 and DL3 glomeruli [34], [35]. Blocking neurotransmitter release from the CSDn by targeted expression of tetanus neurotoxin light chain in the CSDn resulted in reduced behavioural sensitivity towards cVA and these males exhibited increased courtship towards cVA-treated females compared to controls (Figure 7A, p = 0.028). Taken together, these experiments demonstrate a role for the CSDn in modulating olfactory perception of behaving animals in an odor-dependent manner.
Having established a role for the CSDn function in odor-response modulation, we examined the basis for this modulation. Modulation could be achieved in a variety of ways, such as the differential expression of serotonin receptors in the AL or/and by the differential arborization (as observed in the present study, Figure 1A), which in turn may result in differential levels of local serotonin release by the CSDn. Suppressing the function of the CSDn causes reduced behavioural sensitivity towards cVA, indicating that serotonin release is important for enhanced sensitivity towards cVA (Figure 7A). The level of serotonin release in the cVA-specific glomeruli (DA1 and DL3) is likely to be more in cases where there is an increase in the innervations of the CSDn to these glomeruli. Innervations in these glomeruli increase heavily in Ephrin hypomorphs compared to control (Figure 1) predicting that Ephrin hypomorphs should be much more sensitive to cVA. This was indeed the case; Ephrin hypomorphs showed a remarkable behavioural sensitivity to cVA and thus showed highly reduced courtship towards cVA-treated females (Figure 7B, p<0.001). If increased behavioural sensitivity in Ephrin hypomorphs is indeed due to increased DA1/DL3 innervations by the CSDn then rescuing the CSDn branching pattern to control levels should show a rescue of the behavioural phenotype. Targeted expression of Ephrin in the CSDn in Ephrin hypomorphs leads to a partial rescue of the behavioural sensitivity of Ephrin hypomorphs towards cVA (Figure 7C, p = 0.004 compared to Ephrin hypomorphs). This suggests that the increased sensitivity to cVA in the Ephrin hypomorphs is indeed due to the increased innervations of the CSDn. However, the absence of a complete rescue of the behavioural phenotype suggests the possibility that the terminals of other interneurons are defective in the relevant glomeruli in Ephrin hypomorphs. As mentioned earlier, the widespread expression of Ephrin in the lobe indicates that other interneurons may also require the molecule. Nevertheless, a partial behavioural rescue by Ephrin expression in the CSDn in Ephrin hypomorphs suggests that Eph/ephrin signaling has a role in development of the pheromone modulatory circuit and regulates correct positioning of neuronal arbors in a manner relevant for behaviour. Normal courtship in Ephrin hypomrphs (male courtship index = 0.72±0.032; n = 33) is comparable to controls (male courtship index = 0.76±0.037; n = 36) suggesting that these animals don't display a defect in courtship behaviour. A similar analysis could not be performed for Eph null animals as these showed severely reduced normal courtship (data not shown). We next checked whether this is true for the other odor we have examined, CO2. The CSDn innervations in the V glomerulus of Ephrin hypomorphs are comparable to controls (Figure 1) and their response towards CO2 is also comparable to control animals (Figure 7D, p = 0.98). However, EphX652 null animals, which have reduced innervations of the CSDn in the V glomerulus show an increased repulsion to CO2 when compared to controls (Figure 7E, p = 0.003). This phenotype is comparable to what we observed upon silencing or blocking neurotransmitter release from the CSDn (Figure 6). Thus, the olfactory sensitivity towards CO2 changes only in the contexts where the CSDn branching has been affected in V glomerulus. Taken together, our data suggests that the serotonergic CSDn has a modulatory effect in olfactory behavioural sensitivity and glomerular positioning of its terminals during development is essential for its function in the adult.
Our study demonstrates the Eph-ephrin dependent control of terminal-arborization pattern of an identified serotonergic neuron and shows that this regulation is put in place by sensory neuron derived Eph. Further, we link this Eph-ephrin mediated regulation of arborization to neuromodulation dependent behaviour. Eph-ephrin signaling in neural development has been studied [57], [58] but not at the level of an identified neuron and not in a serotonergic neuron. We suggest that such signaling, which combines repulsive as well as attractive responses could play a broad role in sculpting the target-domain of many wide-field neurons. Consistent with this view, we observe arborization defects in multi-glomerular interneurons when Eph-ephrin signal is compromised but not in typical PNs. PNs however show arborization defects upon misexpression of ephrin, demonstrating that Eph-ephrin repulsive signals can operate ectopically. Local targeting of neuronal arbors is a prerequisite to partner-specific connectivity and is thought to be achieved by differential expression of cell-surface molecules on pre- and post-synaptic cells [reviewed in 59], [60]. Eph/ephrin signaling could be part of a code of cell-surface molecules, which are thought to regulate local targeting and partner-specific neuronal connectivity in the AL [61].
Several studies on the OSN and PN targeting in the Drosophila antennal lobe have led to a view that PNs organize the coarse map of the antennal lobe and thus provide spatial information necessary for appropriate fine-targeting of other lobe neurons [26], [62]. Glomerular organization of the antennal lobe is complete by ∼48 hAPF and synaptogenesis starts between 48 and 72 hAPF [63]. Our work shows that this developmental time window is not only relevant for synaptogenesis in the antennal lobe but also for OSN-driven patterning of wide-field interneurons. A small set of OSNs start to express Eph at the onset of the synaptogenic time window and provide spatial information to growing axonal terminals of the CSDn. Eph expressed by OSNs may not influence gross targeting of PNs as PN targeting occurs much earlier in the AL. However, OSN-derived Eph may regulate patterning of axonal terminals of other interneurons, which elaborate their branches during late metamorphosis. It will be interesting to see if selective Eph expression in the OSNs during the phase of synaptogenesis requires olfactory co-receptor expression or neuronal activity.
That the CSDn is a modified larval neuron [18] and that the glomerular-specific terminal pattern is set-up during pupal development both raise the possibility that serotonin release from this neuron has a role in antennal lobe development and plasticity. This possibility emerges from the very elegant set of studies from the Beltz laboratory [64], [65], which demonstrate a role of serotonin through its receptors, in adult neurogenesis in decapod crustaceans. One possibility is that the CSDn acts in the Drosophila larva to influence neurogenesis in the adult, during larval or pupal life, by regulating the specific LN and PN stem-cell linages and their neuronal morphogenesis in the antennal lobe [22], [23], 66. Another possibility, not excluding the first, is that serotonin is relevant to experience dependent changes in the glomeruli, such as observed in Drosophila [67]. We see no obvious alteration in the size of the antennal lobe in contexts where the CSDn function is blocked (data not shown) and, a detailed developmental role for serotonin is outside the scope of the current study. Nevertheless, the CSDn's singular presence in the antennal lobe make studying the developmental role of serotonin an attractive direction and an area that will surely be embarked on soon.
In most brain regions closely studied, each neuromodulatory transmitter is usually released by more than one neuron and co-expression with other neurotransmitters is not uncommon [68]. The Drosophila antennal lobe is likely to be similar and a recent study using mass- spectrometry and genetic tools suggests presence of a large number of neuromodulators in the AL [69]. This makes the linking of the development of identified neurons with their role in behaviour difficult to tease apart. The CSD neuron is special in that it is the only serotonergic neuron that innervates the AL and does not appear to have a co-transmitter. The CSD neuron preparation is thus valuable in that it allows the examination of neuromodulation from development of its anatomy to the role of this anatomy in behaviour. While there may well be a matrix of neuromodulators which together function in the behavioural paradigms we have tested, our results on ablating the function of the CSDn suggest that this neuron is likely to be a key player. Serotonergic neurons usually have a diffuse neuromodulatory role in the CNS. In such contexts, serotonin is able to diffuse from the release site in order to act upon extra-synaptic receptors and serotonergic neurons often branch in a manner to have complete coverage of the neuropil [2], [14], [27], [70]. The context-dependent response to serotonin is mediated by multiple serotonin receptors, which initiate diverse intracellular signal transduction pathways and also differ in their expression pattern in the central nervous system [71], [72]. Our analysis at the resolution of an identified neuron suggests that contextual specificity is also regulated at the level of innervation pattern and connectivity of serotonergic neurons. Our data points to a general mechanism underlying the emergence of contextual specificity in neuromodulation: peripheral neurons developmentally regulate the extent of innervation by modulatory neurons, which in turn, regulate the extent of neuromodulation of specific sensory pathways and behavioural output in the adult.
Our preparation allows the study of neuromodulatory regulation at every scale—from developmental anatomy to behaviour—and does so at the level of a single, identified neuron. A key gap in our study, which we recognize and are addressing as a longer-term direction, is the absence of a neurophysiological response. Published evidence demonstrates the physiological consequences of ectopic serotonin on the antennal lobe. Dacks et al used a genetically encoded Calcium indicator G-CaMP [16] to examine the responses of PNs to ectopic administration of serotonin. They argue that for some odors, serotonin could function by increasing projection neuron sensitivity. Importantly, they show that for odors that activate a wide-range of glomeruli, serotonin enhances PN responses in only some of these glomeruli. The natural suggestion from our study is that this differential alteration of PN responses in a glomerular-specific manner could be, at least in part, due to the specific arborization pattern of the CSDn termini. The response to serotonin is complex and not restricted to PNs alone. Dacks et al also demonstrate that serotonin enhances the responses of inhibitory LNs too. They argue that the effect of serotonin observed on PNs could be an indirect consequence of GABA from inhibitory LNs pre-synaptically acting on OSNs whose modulated function alters PN response. Experiments to test this or related models of serotonin dependent neuromodulation are technically challenging, require substantial time: Such experiments will require, for example, the measurement of OSN, LN and PN physiology when 5-HT receptors are blocked or absent in LNs. For now however, our findings that the CSD neuron's arborization affects behaviour in a manner similar to that seen when its activity is blocked or when 5-HT1BDro receptor levels are down-regulated combined with the studies from Dacks et al [16] strongly suggest that the developmental regulation of local serotonin activity on neurons of the antennal lobe is an important component in the fly's olfactory response.
Why might flies differentially modulate two different olfactory responses? While CO2 is a stress odor for the fruit fly, cVA detection provides information about its mate and thus, each eliciting very different kind of behavioural responses. The mechanism of olfactory processing of CO2 is distinct from that of most other odorants: olfactory perception of CO2 requires co-expression Gr21a and Gr63a which belong to the Gustatory Receptor (GR) family rather than the Olfactory Receptor (DOR) family [73], [74]. Further, CO2 and cVA-sensing OSNs exhibit differences in GABABR expression and consequently employ heterogeneous GABA-mediated presynaptic gain control [75]. In a natural context in the wild, presence of multiple odorants is expected. Differential modulation of functionally distinct odor-processing pathways could be used to advantage for an animal in the wild allowing it to adapt and fine-tune innate behavioral responses according to its immediate environment. Our data points to an element in the complex set of parts which puts such a system in place during development. Another level of sophistication might be added to the olfactory circuit by differential expression of serotonin receptors.
RN2flp, tub>CD2>Gal4, UASmCD8::GFP [18] and w; If/CyO, wg-Z; tub84B-FRT-stop-FRT-LexA::VP16, RN2Flp (this study; referred as RN2flp, tub>stop>LexA::VP16 in the manuscript) flies were used for labeling the CSDn. Gal4-GH146 was a gift from RF Stocker [19]; Gal4-LN1 and Gal4-LN2 were provided by Kei Ito [23]; Or83b-Gal4 was kindly provided by LB Vosshall [46]. 5-HT1BDro-Gal4 and UAS 5-HT1BDro-RNAi lines were kindly provided by Amita Sehgal [9]. EphX652/CiD [38], UAS Eph and UAS Ephrin lines were kindly provided by JB Thomas [38], [76]. lexAop-rCD2::GFP was provided by Tzumin Lee [77]. Pebbled-Gal4 [78] was provided by Rachel Wilson. R60F02-Gal4 was a gift from Gerald Rubin and it was generated as described in [50]; the 892 bp enhancer fragment in R60F02 derives from the acj6 gene and is delineated by the PCR primers caccagtgtcctgccggcgggcgaaaaga and aggtgccgcaatggaagtccttttt. UAS TNTG and UAS TNTVIF were kindly provided by Sean Sweeney [51]; UAS Kir2.1 was a gift from Richard Baines [52]. amos1/Df(2L)M36F-S6 was provided by Andrew German [39]. wglacZ was a gift from JK Roy [79]. MB-Gal80 [56] was a gift from Andre Fiala. Antp, EphrinKG09118 [38] and Or47b-CD2 were obtained from Bloomington Drosophila stock center, Indiana University, USA. wg1-16 was obtained from the Drosophila genetic resource, Kyoto, Japan. UAS EphRNAi (v4771) was obtained from the Vienna Drosophila RNAi Center [80]. All flies were maintained under standard conditions at 25°C unless otherwise indicated. For pupal timing, white prepupae (0 h after puparium formation – 0 hAPF) were collected and placed on a moist filter paper in humid conditions. This stage lasts for about an hour, thus setting the accuracy of staging; the pupal stage lasts 100 hours under conditions in our laboratory.
The tub>stop>LexA::VP16 construct was created by replacing the Gal80 coding region of a pCasper-tub-Gal80 construct with >stop>LexA::VP16. Therefore the LexA::VP16 was amplified via Polymerase Chain Reaction using the pBluescript-LexA::VP16 vector (Lai and Lee, 2006) as a template with the following primers: forward primer-GGG CTA GAG CGG CCG CGG CTA GCG CTC GCG ATA AGC TT and reverse primer- CAA AGA TCC TCT AGA GCC CCC TAC CCA CCG TAC TC. The resulting NotI-NheI-LexA::VP16-Xba PCR fragment was ligated via NotI and XbaI in an open pCasper-tub-NotI-XbaI vector (all enzymes from NEB). A successful ligation was verified via sequencing. The minimal >stop> cassette was inserted via ligation using the NheI side in front of the LexA::VP16 coding region. The orientation of the cassette was verified via digestion and sequencing. DNA for injection was purified using a Qiagen Midi Kit and transgenic lines were generated by BestGene Inc., (Chino Hills, CA, USA).
Brains from 2–4 days old adults were dissected and stained as described in [81]. Primary antibodies used were mouse anti-Bruchpilot/mAbnc82 (1∶20; DSHB), rabbit anti-GFP (1∶10,000; Molecular Probes, Invitrogen, Delhi, India), rabbit anti-Dephrin [1∶1000; Kind gift from Andrea Brand, 82] and rabbit anti-Serotonin (1∶500; Sigma). Secondary antibodies used were Alexa 488, Alexa-568 and Alexa-647 coupled antibodies generated in goat (Molecular Probes; 1∶400). Samples were mounted between coverslips with a spacer in 70% glycerol. Optical sections of 1 µm step size were analyzed using Olympus Fluoview version 1.4a, ImageJ (http://rsb.info.nih.gov/ij/; Wayne Rasband, NIH, USA) and Adobe Photoshop 7.0 (Adobe Systems, San Jose, CA, USA).
Ephrin-Fc fusion probe [kind gift from Alan Nighorn, 38] was used to visualize Eph receptor expression pattern in the developing antennal lobe. Protocol from Kaneko and Nighorn [83] was used for Drosophila pupal brains. Ephrin-Fc specifically recognizes Drosophila Eph [38] and Ephrin-Fc immunoreactivity is completely abolished in EphX652 mutant pupal brains (Figure S1).
Glomeruli were identified using the standard 3-D map of the Drosophila antennal lobe [84]. Quantification of axonal terminal branch tip number was carried out in ImageJ (http://rsb.info.nih.gov/ij/; Wayne Rasband, NIH, USA) using the particle analysis and cell counter plugin. Branch tips of the CSDn in the individual glomerulus were marked manually using the plugin. Data was analyzed and represented as histogram using Microsoft Excel and graphpad instat. Statistical significance was determined using Mann-Whitney test and one-way repeated measure ANOVA test using Sigmaplot software.
The assay was performed as previously described [34]. We tested courtship response of individually reared 5–6 day old virgin males of desired genotypes when introduced to age matched virgin CSBz females (reared in vials with ∼10 females), which were applied 0.2 µl of cVA (Pherobank, Netherlands) (diluted in Acetone) or only Acetone (as control). Concentration of cVA used was 1∶100 unless mentioned in particular experiment. The courtship response was recorded by videotaping (Sony Handycam DCR DVD910E & Sony DSC H9) in a chamber (Diameter = 1.5 cm; Height = 5 mm) for 10 mins, from which the courtship index was calculated manually as described previously [34].
CO2 response index of 4–5 day old flies were measured using an upright Y-Maze apparatus as described elsewhere [67]. CO2was drawn through one arm of the maze, and control air was drawn through the other arm. Flies starved overnight were allowed into the entry tube, and their preference for the arm with the CO2 (O) vs. the control arm with air (C) was quantified as a response index [RI; the difference in the number of flies in the CO2 and control arms as a fraction of the total flies RI = (O−C)/(O+C)]. Behavioural analysis was done in a double-blind manner.
|
10.1371/journal.pgen.0030061 | The Association of a SNP Upstream of INSIG2 with Body Mass Index is Reproduced in Several but Not All Cohorts | A SNP upstream of the INSIG2 gene, rs7566605, was recently found to be associated with obesity as measured by body mass index (BMI) by Herbert and colleagues. The association between increased BMI and homozygosity for the minor allele was first observed in data from a genome-wide association scan of 86,604 SNPs in 923 related individuals from the Framingham Heart Study offspring cohort. The association was reproduced in four additional cohorts, but was not seen in a fifth cohort. To further assess the general reproducibility of this association, we genotyped rs7566605 in nine large cohorts from eight populations across multiple ethnicities (total n = 16,969). We tested this variant for association with BMI in each sample under a recessive model using family-based, population-based, and case-control designs. We observed a significant (p < 0.05) association in five cohorts but saw no association in three other cohorts. There was variability in the strength of association evidence across examination cycles in longitudinal data from unrelated individuals in the Framingham Heart Study Offspring cohort. A combined analysis revealed significant independent validation of this association in both unrelated (p = 0.046) and family-based (p = 0.004) samples. The estimated risk conferred by this allele is small, and could easily be masked by small sample size, population stratification, or other confounders. These validation studies suggest that the original association is less likely to be spurious, but the failure to observe an association in every data set suggests that the effect of SNP rs7566605 on BMI may be heterogeneous across population samples.
| Obesity is an epidemic in the United States of America and developing world, portending an epidemic of related diseases such as diabetes and heart disease. While diet and lifestyle contribute to obesity, half of the population variation in body mass index, a common measure of obesity, is determined by inherited factors. Many studies have reported that common sequence variants in genes are associated with an increased risk for obesity, yet most of these are not reproducible in other study cohorts, suggesting that some are false. Recently, Herbert et al. reported a slightly increased risk of obesity for people carrying two copies of the minor allele at a common variant near INSIG2. We present our attempts to further evaluate this potential association with obesity in additional populations. We find evidence of increased risk of obesity for people carrying two copies of the minor allele in five out of nine cohorts tested, using both family- and population-based testing. We indicate possible reasons for the varied results, with the hope of encouraging a combined analysis across study cohorts to more precisely define the effect of this INSIG2 gene variant.
| Body mass index (BMI) is a heritable measure of obesity that is routinely obtained in large cohorts, is correlated with other measures of obesity, and predicts morbidity and mortality from obesity-related diseases [1–4]. Thus, BMI is a readily accessible trait that can be used to screen for genetic variants that increase an individual's risk for obesity and its complications. There have been more than one hundred publications reporting association between common genetic variants and BMI, but few of the associations have been reproducible in multiple populations [5]. Genotyping of variants has increased exponentially in scale over the past few years, and much more comprehensive screens of common genetic variation for association with obesity are now possible. The poor rate of reproducible findings in association studies in general and obesity in particular are likely due to a combination of false-positive results, underpowered attempts to reproduce associations with modest effects, systematic bias due to technical artifacts or population stratification, and perhaps true heterogeneity in effect across populations due to differences in genetic or environmental modifiers [6,7]. Thus, new reports of association require rapid, well-powered studies to validate true associations or identify false positives that could otherwise trigger unwarranted investigation of spurious findings.
Recently, Herbert and colleagues, including several of the authors of this study, reported a novel association between homozygosity for the minor allele of a single nucleotide polymorphism (SNP), rs7566605, and increased BMI [8]. The SNP has no known function, and the closest gene codes for the insulin signaling protein type 2 (INSIG2), a hijacking protein in the endoplasmic reticulum that, in response to changes in lipid levels, impedes the movement of sterol regulatory element binding proteins to the Golgi apparatus for processing and ultimately its release to act as a nuclear transcription factor and regulator of lipid biosynthesis [9–11]. Animal data suggests a role for INSIG2 in increasing triglyceride level in rats [12], as well as linkage to obesity phenotypes in mice [13].
The association of SNP rs7566605 with obesity was initially found in a set of related individuals from the Framingham Heart Study (FHS) offspring cohort [8]. The SNP was genotyped in five additional cohorts, and the association was observed again in four of these, including population-based studies, case-control samples, and family-based cohorts. However, no significant association was found in a fifth cohort (the Nurses Health Study [NHS]), where a slight trend in the opposite direction was seen. Approximately 10% of individuals were homozygous for the minor allele (C/C), and in a meta-analysis of the case-control samples (including the NHS cohort and excluding the FHS discovery cohort), these individuals had a 22% increased risk of obesity (defined as BMI ≥ 30 kg/m2). In the NHS cohort alone, the 95% confidence interval (CI) for the odds ratio (OR) for obesity was 0.58–1.13. Subsequently, two further groups reported no evidence of association in large cohorts, and a third found association only for people on the overweight end of their population [14–17].
We considered several possible explanations for observing an association in four cohorts but not in the fifth. The failure to observe association in the NHS sample could be due to more modest effects in this cohort and therefore inadequate sample size, population stratification, ascertainment bias, other unmeasured confounders, or any combination of these. It is also possible that evidence in the four cohorts was falsely positive, for any of a combination of reasons that could include hidden population substructure, technical artifacts, or statistical fluctuations causing false positives. However, because of the consistency across multiple cohorts, including studies with family-based design, we felt that these explanations were less likely. Finally, it is also possible that the association is heterogeneous across populations, either due to differences in ascertainment, or differences in genetic or environmental modifiers. Of these possibilities, it is most critical to assess first whether the original associations were spurious, so as to avoid further efforts expended on a false finding. Our primary objective was therefore to test additional large populations to evaluate further the validity and generalizability of this association. By studying these additional populations, including a sample with longitudinal data, we hoped to better assess the strength and consistency of the association between increased BMI and the risk genotype at rs7566605, and perhaps generate hypotheses about any inconsistencies in this association.
Descriptions of the cohorts used in this study are presented in Table 1, Table S1, and in the Methods. These nine cohorts are drawn from eight different populations and include a total of almost 17,000 individuals. The cohorts were not ascertained for BMI, except for the Essen study cohort, which was selected from the upper (BMI ≥ 30 kg/m2) and lower (BMI < 20 kg/m2) ends of the BMI distribution of their population and a portion of the African-American sample that was enriched for obese individuals. We tested for association with obese (BMI ≥ 30 kg/m2) versus non-obese (BMI <30 kg/m2) and also with BMI as a continuous trait, to mimic the association tests performed in the initial publication. All analyses were performed under a recessive model, with the prior hypothesis that C/C homozygotes would have a higher BMI than individuals in the other two genotype classes.
The frequency of C/C homozygotes was increased in obese individuals compared to non-obese control individuals in several cohorts (Table 2). Nominally significant (two-tailed p < 0.05) associations between obesity (BMI ≥ 30 kg/m2) and the C/C were present in three samples: the Iceland cohort (OR = 1.29, 95% CI = 1.06–1.57, p = 0.0064), the Essen cohort (OR = 1.75, 95% CI = 1.15–2.68, p = 0.008), and in one of six exam cycles within the longitudinal data from the FHS cohorts (Table 2). In the Iceland cohort, the homozygote C/C genotype was associated with a 0.69 kg/m2 increment in BMI, which is in good agreement with the effect observed by Herbert et al. [1].
The KORA S3, Maywood, and Scandinavian cohorts, and five of six exam cycles in the FHS cohort, did not show nominally significant associations under a recessive model. The Scandinavian, FHS, and Maywood samples may have been too small to achieve statistical significance with an association of the magnitude estimated by Herbert et al. (OR = 1.22). The Scandinavian cohort had an estimated OR (1.25) similar to the original report, but a p value of 0.46 and a wide 95% CI around the estimated OR (0.69–2.24). In particular, this cohort had only 120 people with BMI > 30 kg/m2, and the power to achieve nominal significance for an OR of 1.22 (as estimated in the original report) is only 15%. The estimated OR in the Maywood cohort was 0.88 but the CIs were also wide (p = 0.68, 95% CI = 0.49–1.59), which suggests that the sample was also underpowered to find this modest association and/or that the effect in this sample is smaller than in the original report.
The KORA S3 sample was much larger (851 obese and 3,233 non-obese), but had an OR of 0.90, with a 95% CI of 0.71–1.16, suggesting that the association is either more modest or absent in this cohort, limited to a particular subgroup of this population (see Discussion), and/or that when several samples are tested, some statistical fluctuation either away from or toward the null is expected. Association tests in the FHS cohort between the C/C genotype and obesity showed some apparent variability, achieving significance in some but not all of the six exams, with p values ranging from 0.003–0.51(Table 2); correcting the best p value for having tested six exams suggests that the totality of these findings are consistent with a replication (corrected p value = 0.018). There was no formal evidence of heterogeneity across the six exams (p = 0.47), and the 95% CIs for all exams include an OR of 1.22 (Table 2).
We also analyzed the five population-based samples—Maywood, Iceland, KORA S3, Scandinavia, and FHS (see Methods for details)—for association with BMI as a continuous trait, again under a recessive model controlling for age and gender. We saw similar results to those observed for the dichotomous analysis, with nominally significant associations between C/C homozygotes and increased BMI observed in the Iceland and FHS cohorts but not in KORA S3, Maywood, or Scandinavia (Table 2). When we analyzed association with BMI at each exam cycle from FHS separately, there was no significant evidence of association in a recessive model. The effect estimates trended in the same direction (exam 3, two-tailed p value = 0.096) (Table 2) as did estimates in the analysis using z-scores for BMI (see Methods) and mean z-score over six exams (unpublished data).
Finally, we tested SNP rs7566605 for association with increased BMI in three family-based samples, using PBAT [18]. Two of the three cohorts showed an association between SNP rs7566605 and BMI as a continuous trait under a recessive model (Table 3). (A dichotomous analysis was not done in these cohorts, because the definition of obesity we used for the remainder of the samples [BMI > 30 kg/m2] was not applicable to the children that made up a substantial part of each cohort.) The family-based portion of the Scandinavian cohort was composed of adults, but the incidence of obesity was only 13% (n = 66), limiting the power of a dichotomous analysis. Because BMI changes rapidly during childhood, we compared the results for the pediatric cohorts using three different measured outcomes: BMI, BMI adjusted for age and gender, and BMI-for-age percentile (Centers for Disease Control and Prevention 2000 National Center of Health Statistics); the p values for the corresponding FBAT statistics were essentially identical in each cohort (unpublished data).
To estimate the overall significance and effect size in the samples we studied, we performed a pooled analysis for both the unrelated and family-based cohorts. These combined analyses, which included both cohorts that showed association and those that did not, yielded independent, statistically significant associations for both the unrelated samples (Table 4) and the family-based samples (Table 3). Combining the p values of the family-based studies using Fisher's method provided evidence of replication (Fisher's combined p = 0.004; Table 3). For the unrelated samples (Table 4), we compared obese and non-obese people, and performed a combined analysis using each exam cycle of the FHS cohort in turn. Since the Essen cohort was ascertained as a severe obesity cohort with non-age matched controls, we tested for heterogeneity between studies using a modified Breslow-Day test [19,20]. There was evidence for heterogeneity when including the Essen cohort (p values for homogeneity = 0.007–0.08) so this cohort was excluded from the combined analyses. Mantel-Haenszel two-tailed p values ranged from 0.011 using FHS exam 3 to 0.054 using FHS exam 6 (Table 4). In these combined analyses, the estimated OR for obesity (BMI > 30 kg/m2) associated with the C/C homozygous genotype ranged from 1.13 to 1.18, somewhat lower than the effect size estimated by the original report [8]. There was also modest evidence of heterogeneity; p values for homogeneity ranged from 0.03 to 0.20, depending on which exam from FHS was included in the combined analysis (Table 4), suggesting that there might be some real variability in effect size across the samples in this study.
Association testing in these nine cohorts shows further evidence that individuals homozygous for the C/C genotype at SNP rs7566605 have a higher BMI and a higher risk of obesity. The association is detectable in diverse cohorts, in children as well as in adults, and in both family-based and population-based samples. The association is not likely to be due to stratification because it was seen in family-based samples such as Costa Rica and CAMP, which are immune to stratification, and because the original publication also described associations in family-based testing [8].
The effect of ascertainment on these analyses could potentially provide confounding of the association in four of these studies. Because index children in family-based studies in CAMP and Costa Rica were ascertained on the basis of asthma, a spurious association between SNP rs7566605 and BMI could be found if the SNP of interest was directly associated with asthma. However, none of the other cohorts were ascertained in this manner, lessening concerns about this source of bias as a potential cause of false-positive associations. In addition, the Scandinavian sample was ascertained as control subjects for a diabetes case control study (similar to the NHS in the original report). A further bias could potentially have been introduced by the selection of non-obese people in the Essen cohort who have a younger mean age than the obese people from this cohort (Table 1). The lean controls (mean BMI = 18.2 kg/m2) are less likely to be obese later in life, but a small portion of them could be misclassified as non-obese, which would tend to bias the estimate toward the null. Of note, the combined analysis remains significant even if we include this study (unpublished data).
The longitudinal nature of the FHS data may provide a clue to a possible cause for inconsistency in the association between SNP rs7566605 and obesity. In this cohort, a stronger effect on BMI was seen in the data from the first three exams than in the last three exams. The individuals at each exam are largely overlapping, making confounders less likely to explain a positive association in the early exam data and a lack of association in later data. Assuming that the association in this cohort is not a false positive due to statistical fluctuation, then the passage of time is the most likely explanation for the diminution of the association in this cohort. The decreasing evidence of association in theory could be due to an interaction with age, namely decreasing effect size with increasing age. Alternatively, a change in the environment could have diminished the strength of the association over time; this would be, in theory, consistent with a well documented “secular trend” of increased obesity over the relevant time period [21,22]. A preliminary and post hoc examination of the FHS data suggests that age may play an important role in modifying the strength of the association (unpublished data). This hypothesis would also be consistent with stronger effects in controls matched for early-onset disease (such as asthma) than in controls matched for later-onset diseases (such as diabetes). Finally, an additional post hoc analysis of the KORA S3 data suggests a stronger association in the most severely obese individuals (OR for BMI ≥ 38 kg/m2 was 1.78, 95% CI, = 0.99–3.21, p = 0.054), who perhaps became obese at an earlier age. Although these hypotheses are speculative at this time, they and other possibilities could and should be tested by a formal meta-analysis of our data, recent studies showing no association [14–16], and additional data that are likely to emerge. We (I.M.H. and colleagues) are in the process of organizing a meta-analysis to reexamine the INSIG2 association in light of these hypotheses to better understand the relationship of this gene to obesity in the population.
In summary, the association of SNP rs7566605 with higher BMI is found in diverse populations. The number of studies in which a nominal association has been observed (five out of the nine cohorts reported here) appears more frequently than expected by chance. However, a more precise assessment of this apparent excess of associations will depend on the availability of a complete set of studies of this polymorphism. Large sample sizes were required to observe the association, but even some large samples have not demonstrated an association with this allele, possibly due to modification by age or other issues related to ascertainment. A combined analysis of both positive and negative studies presented here suggests that the association is valid but also suggests the possibility of heterogeneity across populations. Additional data, both positive and negative, ideally from large samples with good information regarding potential confounders and in a format suitable for meta-analysis, would be required to confirm the existence of heterogeneity and to further refine the estimate of the effect of this SNP on BMI in different populations. However, the evidence to date suggests that this variant has a detectable influence on BMI in a diverse range of populations.
DNA samples were obtained from a large group of 5,187 Icelanders. The study group was composed of individuals who participated in studies of the genetic etiology of cardiovascular and metabolic diseases and the majority of these subjects were recruited as unaffected relatives of probands or as controls and did not have any history of metabolic or cardiovascular diseases. All participants in the study signed informed consent. All personal identifiers associated with tissue samples, clinical information, and genealogy were encrypted by the Icelandic Data Protection Authority, using a third-party encryption system in which the Data Protection Authority maintains the code [23]. Association testing was done according to that of the KORA S4 study design described in Herbert et al [1]. OR of genotype G1 (C/C) compared to genotype G0 (G/C + G/G) was calculated by [n(G1)/m(G1)]/[n(G0)/m(G0)], where n and m denote genotype counts in obese and non-obese individuals, respectively. The genotyping procedure has been previously described [24]. Genotype call rate was 97.3%. p value and CI were adjusted for relatedness of the individuals using simulations as previously described [25]. In each simulation, genotypes for the SNP are simulated through the Icelandic genealogy and the association test repeated treating those genotypes as real genotypes. By repeating this procedure 50,000 times we get the standard deviation of log(OR) under the null hypothesis of no association, which is used to calculate both the p value and the CI. We regressed the log transformed values for BMI on C/C carrier status by adjusting for age and sex in the multiple regressions as shown in Table 2.
In the Southern German region of Augsburg, which includes the city of Augsburg and the two surrounding counties, population-based surveys of the 25–74-y-old population were implemented in 1984 as part of the World Health Organization's Multinational Monitoring of Trends and Determinants in Cardiovascular Disease [MONICA]) project and continued since 1996 within the German Kooperative Gesundheitsforschung in the Region Augsburg (KORA) platform. The third survey, KORA S3, which was the study used in our analysis, was conducted in 1994–1995. Subjects (4,856) were recruited via registry according to the same protocol as the fourth survey (KORA S4) performed in 1999–2001, which was part of the initial replication samples in Herbert et al. The KORA surveys were described previously [22,26]. Genotyping was performed using a MALDI-TOF mass spectometry system (MassEXTEND; Sequenom, http://www.sequenom.com) and the call rate was 99.3%.
DNA samples were obtained from 1,515 unrelated people from the offspring generation of the FHS [27]. We considered the possibility of overlap between the “unrelated plate” of the offspring cohort used here and with the family-based panel, approximately half of which was used in the analysis in the Herbert et al. report. There were 283 people who overlap between the “unrelated plate” and the full family-based panel, so these 283 people were excluded from the analyses reported here. The samples were genotyped using allele-specific primer extension of amplified products with detection by MALDI-TOF mass spectroscopy using a Sequenom platform as previously described [28–30]. Genotype call rate was 99.1% with no discordancies among replicate samples. Association testing was done with linear regression using BMI log transformed and adjusted for age and gender at all six exams.
DNA samples were obtained from 874 unrelated people, self-described as African-Americans, from the same cohort as was described in the original association report [8]. Unrelated people were selected from this population for genotyping. In 270 families, the most obese sibling was chosen to enrich the sample for obese people in the case-control comparison. These were not included in the quantitative trait analysis as described below in Statistical Analysis. Samples were genotyped as previously described [8,29]. Genotype call rate was 97.9% with no discordancies among replicate samples. Association testing was done with linear regression modeling of using log BMI corrected for age and gender with genotype in a recessive and additive model.
DNA samples were obtained from 1,381 adults from Marburg, of which 990 were obese cases (BMI ≥ 30 kg/m2; mean BMI 36.02 ± 5.38 kg/m2) and 391 were lean controls (BMI ≤ 20 kg/m2, mean BMI 18.17 ± 1.00 kg/m2 [31]. Genotyping was carried out by PCR-RFLP with Bsp143I (digests the C-allele) (primers: 5′-TGAAGTTGATCTAATGTTCTCTCTCC-3′ and 5′-AAACCAAGGGAATCGAGAGC-3′). Association analysis under the recessive model, by χ2 testing.
Nuclear families (415) of children with asthma in the Central Valley of Costa-Rica, a relative genetic isolate of predominantly Spanish and Amerindian ancestry [32,33]. Children and their families were enrolled as described previously [34] and anthropometric measurements of all probands included weight and height. However, this population was not ascertained based on morphometric phenotypes. Genotyping was performed using the Illumina BeadStation 500G system (http://www.illumina.com). Genotyping completion rate was >99.8% with no discordances among replicate genotypes. Of the 415 families with genotypic data, 408 had complete phenotypic data and were included in the analysis.
The Childhood Asthma Management Program (CAMP) is a multicentered North American clinical trial designed to investigate the long-term effects of inhaled antiinflammatory medications in children with mild to moderate asthma [35]. Children ages 5 through 12 were eligible for inclusion in the study if they had a diagnosis of asthma and no other clinically significant conditions. Height and weight measurements were collected on these children during the prerandomization period. Of the 1,041 children originally enrolled, 968 children and 1,518 parents contributed DNA samples for genetic studies. Complete nuclear families (408) of self-described non-Hispanic white race with baseline BMI measurements are included here. Genotyping was performed using the Sequenom genotyping platform.
The unrelated sample consisted of individuals from the Botnia Study chosen as control subjects from two cohorts to study diabetes. The first group were controls from a Scandinavian sample of 471 case-control pairs individually matched for gender, age, BMI, and geographic region in Sweden and Finland. The second group were from a Swedish sample of 514 case-control pairs who were individually matched for gender, age and BMI. Subjects were characterized as unaffected for diabetes by glucose tolerance testing as previously described [29]. The family cohort was comprised of 512 unaffected siblings from a Scandinavian sample of 1,189 siblings with and without diabetes, as previously described [36,37]. The samples were genotyped using by an allele-specific primer extension of amplified products with detection by MALDI-TOF mass spectroscopy using a Sequenom platform as previously described [28,29]. Genotype call rate was 96.5% with one Mendel error in one family and no discordancies among replicate samples.
The genotype data in each population was tested for deviation from Hardy-Weinberg and found to be consistent (p value > 0.01). Tests for association of rs7566605 with obesity were performed for the five population-based cohorts under a recessive model, classifying non-obese people as BMI < 30 kg/m2 and obese as BMI ≥ 30 kg/m2. Significance was assessed using a χ2 test with one degree of freedom and two-tailed p values were reported. The Mantel-Haenszel method was used for the combined analysis, and testing for heterogeneity was performed using the Breslow-Day test, as described previously [7,19,20].
For the four samples that had population-based components, an association analysis was performed using BMI as a continuous trait, adjusting for age and gender. A second analysis of the FHS cohort was done to make use of longitudinal data collected across six exams, approximately 4 y apart spanning 26 years from 1971–1997. For each exam, Z scores were calculated by the following process: within each decade of life and gender, log BMI was regressed against age. A Z score was then calculated for these age-adjusted BMIs based on the mean and standard deviation within each decade and gender for each exam. These were then analyzed using standard regression methods (implemented in SAS) for each exam individually, and also for the mean of all available Z scores across the six exams. For the KORA S3, Maywood, and Scandinavian cohort analyses we used standard linear regression with log transformed BMI and adjusted for age and gender. The linear regression analysis in the Maywood cohort excluded 270 people, who had been selected as the most obese person in their family, to avoid possible bias. The Iceland analysis was done with log transformed BMI as a continuous trait under a recessive model, adjusting for age and sex in the multiple regression (sex + age + sex × age).
Association testing of rs7566605 in the family-based cohorts was performed using the FBAT-approach as implemented in PBAT [18,38], with BMI as a quantitative (continuous) trait adjusted for age and gender by Z score under a recessive model. For the Costa Rica and CAMP populations, tests were also done for the outcome BMI adjusted for age and gender, and BMI-for-age percentile (Centers for Disease Control and Prevention 2000 National Center of Health Statistics). Because these studies were similarly sized, a combined analysis was performed using Fisher's method for combining p values, in which twice the negative sum of the natural log of k one-tailed p values is distributed as a χ2 distribution with 2k degrees of freedom [39]. In this method, a one-tailed p value for an effect in the opposite direction is first corrected by subtracting the p value from one; as all the effects in our studies were in the same direction, this correction was not necessary.
The National Center for Biotechnology Information (NCBI) (http://www.ncbi.nlm.nih.gov) accession numbers for the gene and gene product discussed in this paper are INSIG2 (NM_016133) and INSIG2 (NP_057217). |
10.1371/journal.pntd.0003669 | Cessation of Mass Drug Administration for Lymphatic Filariasis in Zanzibar in 2006: Was Transmission Interrupted? | Lymphatic filariasis (LF) is targeted for elimination through annual mass drug administration (MDA) for 4–6 years. In 2006, Zanzibar stopped MDA against LF after five rounds of MDA revealed no microfilaraemic individuals during surveys at selected sentinel sites. We asked the question if LF transmission was truly interrupted in 2006 when MDA was stopped.
In line with ongoing efforts to shrink the LF map, we performed the WHO recommended transmission assessment surveys (TAS) in January 2012 to verify the absence of LF transmission on the main Zanzibar islands of Unguja and Pemba. Altogether, 3275 children were tested on both islands and 89 were found to be CFA positive; 70 in Pemba and 19 in Unguja. The distribution of schools with positive children was heterogeneous with pronounced spatial variation on both islands. Based on the calculated TAS cut-offs of 18 and 20 CFA positive children for Pemba and Unguja respectively, we demonstrated that transmission was still ongoing in Pemba where the cut-off was exceeded.
Our findings indicated ongoing transmission of LF on Pemba in 2012. Moreover, we presented evidence from previous studies that LF transmission was also active on Unguja shortly after stopping MDA in 2006. Based on these observations the government of Zanzibar decided to resume MDA against LF on both islands in 2013.
| Lymphatic filariasis was highly endemic in Zanzibar when MDA commenced in 2001 to eliminate the disease. In 2006, Zanzibar, in the United Republic of Tanzania, was the first territory in Africa to complete five rounds of annual treatment using a combination of albendazole and ivermectin at 100% geographic coverage and achieving effective treatment coverage of over 65% during each round. MDA was stopped in 2006 after sentinel site surveys revealed parasite infection rates of zero in both humans and mosquito populations. In 2012, when new tools became available to verify the absence of transmission, we asked the question if transmission was truly interrupted when MDA was stopped in 2006. In January 2012, we performed the WHO recommended transmission assessment surveys (TAS) on the main islands of Unguja and Pemba to verify the absence of LF transmission in line with ongoing efforts to shrink the LF risk map. Altogether, 3275 children were tested on both islands and 89 were found to be CFA positive; 70 in Pemba and 19 in Unguja. The distribution of schools with positive children was heterogeneous with pronounced spatial variation on both islands. Based on the calculated TAS cut-offs of 18 and 20 CFA positive children for Pemba and Unguja respectively, we demonstrated that transmission was still ongoing in Pemba where the cut-off value was exceeded. We also presented evidence from previous entomological studies that LF transmission was active on Unguja shortly after stopping MDA in 2006. Based on these findings we concluded that LF transmission was still active in Zanzibar, and one million people at risk of acquiring LF, and recommended the resumption of MDA on both islands to eliminate the disease. In 2013, the government of Zanzibar decided to resume MDA with ivermectin plus albendazole on both islands.
| Lymphatic filariasis (LF) is a major cause of acute and chronic morbidity and a significant impediment to socioeconomic development in 73 countries in Africa, Southeast Asia, the Americas, and the Pacific region. The World Health Organization (WHO) estimates that in 2012 more than 1.4 billion people living in these countries were at risk of acquiring the infection [1]. LF infection occurs through intense and long-term exposure to mosquito bites from several genera of anopheline and culicine mosquitoes that are carriers of the three parasites that cause human filariasis (Wuchereria bancrofti, Brugia malayi and B. timori [1]. The Global Programme to Eliminate Lymphatic Filariasis (GPELF) recommends annual mass drug administration (MDA) using albendazole in combination with either diethylcarbamazine (DEC) or ivermectin for 4–6 years as the main strategy to interrupt transmission of the disease [2]. From 2000 to 2012, more than 4.4 billion doses were used to treat over 500 million people in 56 countries, making the GPELF one of the most rapidly expanding global health programs in the history of public health [1]. Many countries implementing MDA have completed more than 5 consecutive annual treatment rounds but only China and South Korea have been verified by WHO for achieving elimination of LF [3]. In 2006, Zanzibar, in the United Republic of Tanzania, which started MDA in October 2001 [4], was the first country in Africa to complete five rounds of treatment using a combination of albendazole and ivermectin at 100% geographic coverage and achieving effective treatment coverage rate of over 65% during each round [3, 5]. A detailed description of the first round of MDA in 2001 and the subsequent four treatment rounds, including treatment coverage for each round, has been published previously [4, 5].
Parasite infection rates and intensities in the human and mosquito populations decrease after several rounds of MDA, but individuals may remain microfilaria and antigenemia-positive even after transmission has been interrupted [6]. A standard methodology called Transmission Assessment Survey (TAS) has been described by WHO to assess whether the prevalence of infection has been lowered to a level where recrudescence is unlikely to occur, even when MDA interventions have been stopped [6, 7]. After five rounds of MDA, an implementation unit (IU) is considered eligible for TAS if treatment coverage exceeds 65% on each round and the prevalence of microfilaraemia is below 1% on sentinel and spot check sites.
Treatment was stopped in Zanzibar in 2006, after five rounds of MDA and 2 sentinel and 12 spot-check site surveys in high risk urban and rural areas revealed parasite infection rates of zero in both humans and mosquitoes [3]. In line with ongoing efforts to shrink the LF map, we performed TAS in January 2012 to determine whether the successive rounds of MDA carried out between 2001 and 2006 achieved the interruption of disease transmission in the two main islands. Demonstrating the absence of active transmission of LF in Zanzibar is the first step in the verification process that could result in its reclassification as non-endemic and consequently, shrinking the LF map by yet another country.
Zanzibar is part of the United Republic of Tanzania located 35 km off the mainland Tanzania coast. It comprises two main islands, Pemba and Unguja, and a number of sparsely populated islets; the land areas of Unguja and Pemba are 1,654 km2 and 984 km2 respectively. About 1 million people live in Zanzibar with 65% of the inhabitants residing on Unguja. Unguja is the largest and most populated island of Zanzibar, with more than 40% of the population residing in Zanzibar town, the administrative and commercial centre of the two islands. Pemba, on the other hand, has three towns forming concentrated urban centres. Although Zanzibar is non-endemic for onchocerciasis, onchocerciasis transmission occurs in mainland Tanzania. Considering that there is free movement and settlement of the population in both the island and mainland communities, it was agreed to use ivermectin in combination with albendazole as the treatment regimen for LF [5]. In Zanzibar, the causative agent of LF is Wuchereria bancrofti which is transmitted by the urban mosquito, Culex quinquefasciatus [8, 9].
Prior to the first MDA round in 2001, parasitological and entomological surveys were carried out in two sentinel sites, Kizimkazi and Kwahani, in Unguja to collect baseline data for impact monitoring. The two sites represented the areas of highest risk of exposure to LF in rural and urban communities respectively in Zanzibar. Kizimkazi, which had a population of 3,037, is located in the dry arid stony area along the coast of the rural southern district. Pit latrines, cesspits and soakage pits, which serve as good breeding grounds for Culex mosquitoes, were common in Kizimkazi. Kwahani, located in the urban district of Unguja, had a population of 4,550 and featured many open drains where Culex mosquitoes breed. Before MDA started in 2001, the MF prevalence rates in Kizimkazi and Kwahani were 17.1% and 7.5% respectively, based on examining 500 people from each site [5]. The baseline surveys took place in September 2001 and were annually repeated before each MDA round. Finger prick blood samples (100μl) were obtained at night and examined using the counting chamber method as described in the WHO guidelines for MF surveys [7]. Treatment coverage rates over the 5 years varied from 76% (in 2001) and 83% during the subsequent 4 years with the last MDA round completed in November 2005. The prevalence of microfilaremia in the two sentinel sites dropped from 17.8% and 7.2% before treatment to 1.0% and 0.0% respectively after the fifth round [10].
Transmission assessment surveys (TAS) were performed in January 2012, six years after stopping MDA, using standard operating procedures (SOP) and WHO guidelines, which have been described in detail elsewhere [6, 7]. Zanzibar met the TAS eligibility requirements of having completed at least five effective rounds of MDA in all IUs with coverage >65% over the total population and MF rates <1% in each of the sentinel sites after five MDAs. It was decided to conduct the survey in two evaluation units (EU); one per island, as this would provide a clear picture of the current LF transmission status on each island while still having less than 2 million people on each EU. The TAS design was based on the net primary school enrolment rate, the target population size and number of schools on each island, and according to the rules established for Culex-W. bancrofti transmission areas [7]. A Microsoft Excel computer tool, the Survey Sample Builder (SSB) [7], was used to generate random number lists and inform TAS design calculations, including sample size and sampling intervals for the two TAS evaluation units. School surveys were conducted on both islands where net primary enrolment rates exceeded 75%. Children 6–7 years old were targeted for the TAS because antigenaemia in young children would reflect recent and active transmission, while antigenaemia in older children and adults may be related to infections that occurred before MDA. All children in grades 1 and 2 were eligible including a small proportion of those outside this age range and the oldest was eight years old. The sample sizes for the two evaluation units generated by the SSB were 30 school clusters on each island, which encompassed 1556 children in Pemba and 1684 children in Unguja.
The TAS critical cut-off value represents the threshold of infected individuals below which transmission is expected to be no longer sustainable, even in the absence of MDA.
If the total number of positive cases is at or below the critical cut-off value, the EU ‘passes’ the survey and MDA is not consider to be required [7]. TAS sample sizes and critical cut-off values are powered so that the EU has at least a 75% chance of passing if the true antigen prevalence is half the threshold level (2% for Culex, Anopheles, and Mansonia vector areas, and 1% for Aedes vector areas). In addition, there is no more than a 5% chance of passing if the true prevalence is greater than or equal to the threshold level. The critical cut-offs generated for Pemba and Unguja were 18 and 20 respectively.
The Binax NOW Filariasis immunochromatographic card test (ICT) (Alere Inc., Scarborough, ME) was used to detect circulating filarial antigen (CFA) as described in the WHO guidelines [7]. Briefly, 100 μl of finger-prick blood was collected from each individual and then transferred to an ICT card test using a calibrated capillary tube. The test was read 10 minutes after closing the card, as instructed by manufacturers. A positive control filarial antigen was used to confirm the quality of the ICT cards. All positive controls turned out positive. The antigen contains the epitope present in circulating Wuchereria bancrofti antigen that is detected by the Binax Filariasis Now test. This was necessary to instil confidence in the large number of negative results expected. There was however a very minimal risk of false positives in Zanzibar which is not endemic for Loa loa [11]. ID number, class and test result of each child tested was recorded.
Data analysis was conducted using SPSS and the results were mapped using ArcGIS 10.1 (ESRI, Redlands, CA). The test results from each cluster in the different EUs were collected by the MOH survey team and the Public Health Laboratory Ivo de Carneri staff on Pemba. Data was entered using a Microsoft Access data base specifically developed by the Centre for Neglected Tropical Diseases (CNTD) in the Liverpool School of Tropical Medicine (LSTM) to support TAS. Two independent office workers entered data using a double entry system that automatically compare the two entries and detect any possible errors. Discrepancies were checked at CNTD for accuracy.
Ethical clearance for the study was granted by the LSTM Research Ethics Committee of the Liverpool school of Tropical medicine which approved the use of oral consent (Research Protocol 11.89RS). After approval from the Ministry of Education and school authorities, the MoH team met with the head master of each school to obtain permission to conduct the survey and to schedule a date for meeting with parents. The survey team explained the purpose of the surveys and received oral consent from teachers and parents. Written consent was not required for surveys conducted by the Ministry of Health as part of the disease control activities. Non consenting parents or non-assenting children could drop out from the study at that time or any time during the study.
Altogether, 3 275 children were tested on both islands and 89 were found to be CFA positive. The CFA point prevalence for Pemba and Unguja were 5.4% (70/1298) and 0.9% (19/1977) respectively. Even though the sample size required for Pemba was 1556, we decided to discontinue the tests when the number of positive children largely exceeded the critical cut-off value of 18 for the EU. Some parents did not consent to including their children in the survey but no mop up was required since the critical cut-off was largely reached. The CFA prevalence for boys (0.91%) and girls (1.0%) was almost identical on Unguja, where the whole sample size was surveyed. The distribution of schools with antigen positive children was very heterogeneous on both islands with pronounced spatial variation between and within districts as shown in the Fig. 1.
A total of 30 schools, 9 from each of the four districts on Pemba Island were surveyed and CFA positive children were detected in every school with the exception of Cheke ckeke in south Pemba, where all 296 children were negative. Among the schools with positive children the CFA prevalence rates varied from 1.17% in Mkoani in South Pemba to 14.64% in Micheweni in North Pemba. The overall CFA prevalence for Pemba was 5.4% but the district specific CFA rates for Mkoani, Wete and Micheweni were 1.0%, 7.8% and 10.3% respectively.
In Unguja, where a total of 1684 children were tested, the total found to be CFA positive (19) was just below the critical cut off value of 20. The distribution of positive schools on the island was very heterogeneous with three (Central, North B and Urban) of the 6 districts revealing no positive children. Among the 3 districts with positive school, the district specific CFA rates were 4.7% (North A), 3.2% (South) and 0.4% (West).
Between 2001 and 2006, the Zanzibar programme for the elimination of LF carried out effective annual MDA campaigns to interrupt the transmission of the disease [5]. To effectively coordinate MDA in this predominantly Muslim country, the last Saturday of October was designated as the annual Filaria day (F-day) with a ‘mop-up day' on Sunday. Ivermectin in combination with albendazole were administered by highly motivated community drug distributors known as Filarial Prevention Assistants (FPAs) who ensured a high treatment coverage ranging from 70 to 80% for all five rounds. MDA was stopped in 2006 after sentinel site surveys revealed parasite infection rates of zero in both humans and mosquito populations as reported by Mohammed in his 2009 PhD thesis (Lymphatic Filariasis in Zanzibar: Epidemiology, Elimination and impact) with the University of Liverpool, United Kingdom [12]. The infection rates in humans and mosquitoes were determined by night blood survey and dissections respectively. It had been demonstrated in Egypt, where W. bancrofti was also transmitted by Culex mosquitoes, that five rounds of MDA using albendazole plus DEC can interrupt the transmission of LF in a population of 2.5 million [13].
Lymphatic filariasis was endemic on both islands before MDA commenced in 2001 as described in detailed investigations carried out many years prior the initiation of MDA [5, 9, 14]. Cross-sectional clinical, parasitological and entomological surveys for LF, conducted in urban and semi urban communities on Pemba in 1990 revealed that LF endemicity and vector species composition had not changed significantly for 15 years [9]. MF prevalence rates on Pemba during a survey conducted in 1975 ranged between 11.8% and 16.2% for people aged above 1 year [9]. Clinical manifestations in the form of hydroceole and lymphoedema were also common on the island, with prevalence of 22.4% and 1.4% respectively for adults above the age of 15 years [9]. Similarly, surveys conducted on Unguja in 1975 showed that the overall prevalence of clinical signs among men aged 15 years and older was 29.6% for hydrocele and 7.9% for elephantiasis, while the MF rates varied from 7.0% to 39.0% [12].
Our TAS results showed that five rounds of MDA in Zanzibar either failed to interrupt the transmission of LF on Pemba, where the TAS cut-off of 18 was surpassed by a huge margin early in the survey, or resurgence occurred after MDA was stopped in 2006. Unfortunately, the sentinel sites selected for monitoring the impact of MDA in Zanzibar did not include communities on Pemba and therefore the intensity of LF transmission on the island in 2006 could not be verified. Studies elsewhere have demonstrated that the vector of LF in Zanzibar, the highly efficient Culex quinquefasciatus, can sustain transmission in areas of low density microfilaraemia, even when MF is undetectable using traditional diagnostic methods based on around 60 μl [15]. In addition, recent studies on mainland Tanzania has also demonstrated that transmission of LF can persist after seven rounds of MDA in urban areas where Culex quinquefasciatus are the main vectors [16]. Infective mosquitoes were found in communities in India where MF rates dropped to zero after six rounds of treatment with DEC or ivermectin suggesting that transmission can occur in the absence of detectable MF if Culex mosquitoes are the vectors [15, 17].
Understanding the transmission dynamics of LF by different species of mosquitoes is essential for the rational planning of control measures and impact assessment. An important determinant of transmission efficiency is the genera of vector species involved [18]. For filariasis transmission to be interrupted, vector density or microfilaria intensity needs to be lowered below a threshold that ensures no new infections occur. This threshold for parasite density has been shown by quantitative models to be higher for anopheline which interact with W. bancrofti in a density dependent vector-parasite relationship known as facilitation. The relationships associated with culicine mosquitoes are known as limitation and proportionality and together with facilitation they describe the quantitative relation between microfilarial uptake and yield of infective L3 larvae in the mosquito vectors [18]. Based on these vector-parasite relationships, the TAS cut-off value for Aedes species (1%) is lower than that for Anopheles and Culex species (2%) [7]. The basis of grouping Culex with Anopheles species in determining these TAS cut-off values is unclear but analysis of eliminations threshold for Anopheles and Culex species suggest a higher threshold for the former [18]. The persistence of transmission in Culex transmission zones has led to growing concerns about the effectiveness of using MDA alone to eliminate LF without the inclusion of vector control [14, 19, 20]. On the other hand, once Anopheles-transmitted LF was eliminated in Solomon Island in the 1970s [20, 21], resurgence was never detected and it was declared non endemic by WHO in 2009 [22],
Vector control is effective against LF [20] but active vector control intervention did not resume in Zanzibar until after MDA for LF was stopped in 2006, when the Zanzibar Malaria Control Programme (ZMCP) started the distribution of free long lasting insecticidal nets (LLINs) targeting mainly pregnant mothers and children under the age of five years [23]. Bednet usage was initially lower in Pemba in comparison to Unguja but by 2008 every household in Zanzibar received two LLINs and, since 2006, six rounds of indoor residual spraying (IRS) have been conducted with synthetic pyrethroid lambda-cyhalothrin (ICON) resulting in over 90% coverage of all dwellings. IRS and LLINs target both endophilic (indoor resting) and endophagic (indoor feeding mosquitoes) mosquitoes including the vectors of LF on Zanzibar [20]. The combination of IRS and LLINs with other interventions resulted in a dramatic reduction of malaria prevalence in Zanzibar from 40% in 2005 to between 0.2 and 0.5% in 2011/2012 [23]. The low prevalence of LF infection in children in Unguja may be partly explained by the impact of the vector control measures as previous efforts to control LF in Unguja by vector control resulted in 65% reduction in mosquito density in houses [23]. The use of LLINs alone have resulted in the interruption of LF transmission in communities in Nigeria [24, 25] and Papua New Guinea [26]. Based on quantitative analysis of elimination thresholds for LF, the probability that the parasite will be eliminated following six rounds of MDA increases as the vector biting rates decrease [18].
The distribution of schools with antigen positive children was very heterogeneous on both islands with pronounced spatial variation between and within districts. Although Unguja barely passed the transmission interruption verification test by revealing fewer (19) CFA positive children than TAS cut-off of 20, four of the six positive schools had CFA positive rates higher than 5% and could enable transmission. Dissection of 6568 Cx. quinquefasciatus mosquitoes in 2006 found none to be infected with W. bancrofti but PCR assays on 5184 specimens collected between November 2007 and February 2008 showed a maximum likelihood infection rates of 1.13% (0.82% − 1.52%) that suggested ongoing transmission. It is therefore very likely that transmission was ongoing on the Island in 2006 when MDA was stopped [12].
The decision to stop MDA after several effective rounds of MDA should be based on statistically robust methodology. A recent multicenter evaluation to define endpoints for MDA in 11 countries concluded that TAS was superior to previous WHO guidelines used to determine when to stop MDA [6]. It was shown to be a practical and effective evaluation tool for stopping MDA although its validity for longer-term post-MDA surveillance will require the use of more sensitive tools for detection infection in humans and mosquitoes.
In conclusion, our findings in 2012 suggested that LF transmission was still active on Pemba. We also presented evidence from previous entomological studies that LF transmission was active on Unguja shortly after stopping MDA in 2006. Based on these findings including the heterogeneous distribution of CFA positive children in Unguja, and the high number of positives found compared (19) to the cut-off value (20) the government of Zanzibar decided to resume MDA with ivermectin plus albendazole on both islands in 2013. TAS will be repeated in 2015 after two rounds of treatment. However, the interpretation of the results from the TAS survey in 2015 may be confounded by the lack of treatment in Zanzibar for 7 years and pre-TAS sentinel site surveys in other evaluation units, including outside Zanzibar, may be required to determine if the criteria for TAS still holds.
|
10.1371/journal.pgen.1006765 | Suspected Lynch syndrome associated MSH6 variants: A functional assay to determine their pathogenicity | Lynch syndrome (LS) is a hereditary cancer predisposition caused by inactivating mutations in DNA mismatch repair (MMR) genes. Mutations in the MSH6 DNA MMR gene account for approximately 18% of LS cases. Many LS-associated sequence variants are nonsense and frameshift mutations that clearly abrogate MMR activity. However, missense mutations whose functional implications are unclear are also frequently seen in suspected-LS patients. To conclusively diagnose LS and enroll patients in appropriate surveillance programs to reduce morbidity as well as mortality, the functional consequences of these variants of uncertain clinical significance (VUS) must be defined. We present an oligonucleotide-directed mutagenesis screen for the identification of pathogenic MSH6 VUS. In the screen, the MSH6 variant of interest is introduced into mouse embryonic stem cells by site-directed mutagenesis. Subsequent selection for MMR-deficient cells using the DNA damaging agent 6-thioguanine (6TG) allows the identification of MMR abrogating VUS because solely MMR-deficient cells survive 6TG exposure. We demonstrate the efficacy of the genetic screen, investigate the phenotype of 26 MSH6 VUS and compare our screening results to clinical data from suspected-LS patients carrying these variant alleles.
| The colorectal and endometrial cancer predisposition Lynch syndrome (LS) is caused by an inherited heterozygous defect in one of four DNA mismatch repair (MMR) genes. Deleterious mutations (e.g., protein-deleting or -truncating) in DNA MMR genes unambiguously allow for the clinical diagnosis LS and hence enable appropriate surveillance measures to be taken to reduce cancer risk and ensure early detection of tumors. However, currently about one-third of detected MMR gene variants are subtle with less clear functional consequences: missense mutations affecting a single amino acid may be innocuous, hence not causing LS, or partially or fully destroy protein function. As long as uncertainty exists about their pathogenicity, such mutations are labeled ‘variants of uncertain (clinical) significance’ (VUS). VUS hamper genetic counseling and therefore the need for functional testing of VUS is widely recognized. To functionally annotate MMR gene VUS, we have developed a high content cellular assay in which the VUS is introduced in a cell culture by oligonucleotide-directed gene modification. Should the VUS be deleterious for MMR, the modified cells survive exposure to the guanine analog 6-thioguanine (6TG) and 6TG-resistant colonies appear. Should the mutation not affect MMR, no colonies appear. Here we present the adaptation and application of this protocol to the functional annotation of variants of the MMR gene MSH6. Implementation of our assay in clinical genetics laboratories will provide clinicians with information for proper counseling of mutation carriers and treatment of their of tumors.
| Lynch syndrome (LS) is an autosomal-dominantly inherited predisposition to a variety of malignancies at a young age, mainly colorectal cancer (CRC) and endometrial cancer (EC) [1]. It is caused by inactivating germ-line mutations in the DNA mismatch repair (MMR) genes MLH1, MSH2, MSH6 or PMS2, or a deletion in the 3’ region of the EPCAM gene that affects MSH2 expression [2–6].
The DNA MMR system is essential for the fidelity of DNA replication. Its primary function is the correction of base-base mismatches and insertion-deletion loops that may arise during DNA replication. Base-base mismatches are recognized by the MSH2-MSH6 heterodimer while MSH2-MSH3 detects loops of unpaired bases. Following mismatch binding, the MSH heterodimers recruit another heterodimer, MLH1-PMS2, to coordinate removal and resynthesis of the error-containing strand [7–9]. A second function of the DNA MMR system is to mediate the toxicity of certain DNA damaging agents such as methylating agents and thiopurines. These DNA damaging agents create adducts in the genome that give rise to mismatches when replicated. The DNA MMR system recognizes the mismatches but will remove the incorporated nucleotide rather than the lesion itself, creating a repetitive cycle of nucleotide incorporation and deletion that ultimately leads to DNA breakage and cell death [10,11]. In the absence of MMR, cells tolerate methylation damage, but consequently show high levels of DNA damage-induced mutagenesis on top of a strongly elevated level of spontaneous mutagenesis [12].
LS patients inherit a functional and a mutant copy of one of the DNA MMR genes. For cells to become MMR-deficient and develop a mutator phenotype that accelerates carcinogenesis, somatic loss of the wild-type allele is required [13]. Microsatellite instability (MSI), i.e., length alterations of repetitive sequences like (CA)n or (A)n, and loss of immunohistochemical staining (IHC) for MMR proteins are considered hallmarks of LS tumors. Analysis of MSI and IHC on tumor tissue can identify patients who may suffer from LS. For a definitive LS diagnosis, however, sequence analyses must reveal a pathogenic germline mutation in one of the DNA MMR genes or the 3’ region of EPCAM [14,15]. Many LS-associated sequence variants are nonsense and frameshift mutations that clearly truncate the protein and unambiguously abrogate MMR activity. Missense mutations that only alter a single amino acid are also frequently identified in suspected-LS patients. The functional implications of these variants are less clear. Consequently, the diagnosis of suspected-LS patients carrying missense variants is difficult in the absence of clear segregation and functional data. As long as the phenotype of these variants of uncertain significance (VUS) is unclear, non-carriers cannot safely be discharged from burdensome surveillance programs [16]. Surveillance programs have proven to significantly reduce morbidity and mortality in LS patients [1,17,18], but pose unnecessary psychological and physical stress on carriers of innocent VUS as well as pressure on preventive healthcare. Therefore, techniques that characterize MMR gene VUS and enable the identification of individuals at risk are urgently needed.
While in the past primarily MSH2 and MLH1 were sequenced to identify LS-causing mutations, in recent years MSH6 has been gained fame for causing LS due to the advancement of DNA sequencing. However, MSH6 mutation carriers can be difficult to diagnose because they may not entirely fulfill the criteria for LS diagnosis: their age at cancer onset is often later than for MLH1 and MSH2 mutation carriers, and their tumors occasionally stain for MSH6 and have no or low MSI [19–21]. We therefore extended the applicability of the oligonucleotide-directed mutagenesis screen we recently described for the identification of pathogenic MSH2 variants to MSH6 variants [22]. The genetic screen uses oligonucleotide-directed gene modification (oligo targeting) [23] to introduce variant codons into the endogenous Msh2 gene of mouse embryonic stem cells (mESCs) and subsequently identifies pathogenic variants by selecting for cells that are resistant to the thiopurine 6-thioguanine (6TG). Here we present the applicability of this screen for the characterization of MSH6 VUS.
The oligonucleotide-directed mutagenesis screen takes a four step approach to the identification of pathogenic MSH6 mutations (Fig 1): 1) site-directed mutagenesis to introduce the variant of interest into a subset of Msh6+/- mESCs, 2) selection for cells that consequently lost MMR capacity, 3) PCR analysis to exclude cells that lost MMR capacity due to loss of the Msh6+ allele (loss of heterozygosity events), 4) sequence analysis to confirm the presence of the planned mutation in the MMR-deficient cells.
mESCs provide a good study model because the human and mouse MSH6 amino acid sequences share over >86% identity (S1 Fig) and mouse models can be made from these cells if VUS need to be studied in vivo. Msh6+/- mESCs only contain one wild type Msh6 allele (Msh6+); the other allele was disrupted by a puromycin-resistance gene and therefore inactivated (Msh6-) [24]. Hence introduction of a specific mutation into the one active Msh6 allele will lead to expression of solely the variant protein and allow immediate investigation of its phenotype. To achieve this, Msh6 was site-specifically mutated by oligo targeting, a gene modification technique that uses short single-stranded locked-nucleic-acid-modified DNA oligonucleotides (LMOs) (with either sense or antisense orientation) to substitute a single base pair at a desired location. LMO-directed base-pair substitution can be achieved at an efficiency of 10−3; thus, about 1 in every 1000 LMO-exposed Msh6+/- mESCs will contain the desired mutation [23]. To determine whether the substitution abrogated Msh6 activity and this subset of cells consequently lost MMR activity, LMO-exposed mESCs were treated with 6TG. The thiopurine DNA damaging agent 6TG is highly toxic to MMR-proficient but only moderately toxic to MMR-deficient cells [11]. Therefore, the appearance of colonies that survived mild 6TG selection is indicative for loss of MMR capacity. Loss of MMR capacity may arise due to the introduced mutation or due to loss of heterozygosity events that caused loss of the functional Msh6 allele. To exclude the latter from further investigation, a PCR that detected the presence of both the disrupted and non-disrupted Msh6 alleles was performed [24]. 6TG-resistant colonies that maintained both Msh6 alleles were sequenced to confirm the presence of the planned mutation.
To demonstrate the ability of the oligonucleotide-directed mutagenesis screen to distinguish pathogenic MSH6 mutations from polymorphisms, a proof of principle study was performed with MSH6 variants G1139S and L1087R that were previously proven to be pathogenic and not pathogenic, respectively [25], as well as all classified pathogenic and not pathogenic missense variants described in the International Society for Gastrointestinal Hereditary Tumours (InSiGHT) colon cancer variant database (http://insight-group.org/). This database uses available clinical, in vitro and in silico data to categorize DNA MMR gene sequence variants according to a five-tiered classification scheme as: class 5, Pathogenic; 4, Likely pathogenic; 3, Uncertain; 2, Likely not pathogenic; and 1, Not pathogenic [26]. Msh6+/- mESCs were first exposed to antisense oriented LMOs encoding the desired base-pair substitution. If subsequent 6TG selection did not reveal resistant colonies encoding the planned mutation, the screen was repeated with sense oriented LMOs.
LMO-mediated introduction of both, pathogenic and not pathogenic variants led to the appearance of 6TG-resistant colonies. For each LMO, we picked and analyzed 18 colonies. The vast majority of 6TG-resistant colonies obtained with LMOs encoding polymorphisms had lost the wild-type Msh6 allele by loss of heterozygosity (LOH) events, as inferred from allele-specific PCR analysis. Sequencing of the few 6TG-resistant colonies that had retained both Msh6 alleles (±6%), did not detect any mutation (Fig 2A). These background colonies apparently arose from cells that for unknown reasons survived 6TG exposure. Of the 6TG-resistant colonies that emerged following LMO-mediated introduction of pathogenic mutations, ±40% still contained both Msh6 alleles. Sequence analysis detected pathogenic mutations in all but one of these 6TG-resistant colonies (Fig 2B; S2A Fig). Thus, the oligonucleotide-directed mutagenesis screen detected all 4 pathogenic mutations and not one of the 5 non-pathogenic variants, indicating it is capable of distinguishing pathogenic MSH6 mutations from polymorphisms.
We used the oligonucleotide-directed mutagenesis screen to investigate the phenotype of 18 MSH6 VUS described in literature and the InSiGHT database as well as 8 MSH6 VUS detected in suspected-LS patients from the Erasmus Medical Center Rotterdam and the Radboud University Medical Center Nijmegen (see S1 and S2 Tables for clinical data [27–38]; see S3 Fig for location of variants in MSH6 [39,40]). Of the 26 variants, 18 were not present in 6TG-resistant colonies and hence do not appear to affect MMR activity. Mutations R510G, A586P, G683D, F703S, L1060R, E1191K, T1217D and T1217I were identified in 6TG-resistant colonies by sequence analysis (Fig 3A and 3B; S2B Fig). Of note, variants R510G and F703S were detected in only two colonies out of five and four, respectively, that had not resulted from LOH (Fig 3B). Given the low frequency of LMO-mediated base-pair substitution, we consider the presence of a variant allele in two independent colonies indicative for pathogenicity. The MMR abrogating effect of all Msh6 variants conferring 6TG-resistance was further characterized by Western blot analyses, MSI assays and methylation-damage-induced mutagenesis assays.
The effect of the identified MMR abrogating mutations on MSH6 and MSH2 protein levels was evaluated by Western blot analyses (Fig 4). MSH6 and MSH2 form a heterodimer; consequently, a drop in MSH6 levels is often associated with a slight decrease in MSH2 protein stability. Protein levels were quantified with respect to Msh6+/- mESCs, which maintain a functional MMR system with about two-third of the MSH6 level observed in Msh6+/+ mESCs [25]. Known pathogenic mutations V397E, L448P, G1137S and R1332Q reduced MSH6 levels to 7–33% of that seen in Msh6+/- mESCs. The R1332Q mutation is located in the splice donor site of exon 9 which may explain the appearance of a shorter protein. The drop in MSH6 levels seen for the known pathogenic mutations was mirrored by variants A586P, G683D, F703S and L1060R that reduced protein levels to 7–24%. Variants R510G, E1191K, T1217D and T1217I maintained relatively high MSH6 levels of 59–79%.
MSI in MSH6 mutation carriers is largely restricted to mononucleotide markers [41]. To investigate the effect of the detected Msh6 variants on MSI we used a (G)10-neo slippage reporter. The neomycin resistance gene (neo) in this reporter is rendered out of frame by a preceding (G)10 repeat. When DNA polymerase slippage errors at the (G)10 repeat such as the deletion of one G or insertion of two Gs remain unnoticed, the neo becomes in frame and generates Geneticin-resistant cells. Hence the number of Geneticin-resistant colonies is indicative of the frequency of neo-restoring slippage events and the MMR capacity of the cells [42]. The slippage rates, i.e., the chance of a slippage event occurring during one cell division, in 6TG-resisant Msh6 VUS expressing mESCs ranged from 5.3x10-5 to 5.1x10-4; which is around the average rate of 1.9x10-4 observed for the known pathogenic mutations and 140 to 1340-fold higher than the slippage rate of 3.8x10-7 seen for Msh6+/- MMR-proficient mESCs (Fig 5).
In addition to increased spontaneous mutagenesis events, MMR-deficient cells also experience increased methylation-damage-induced mutagenesis [43]. To study the influence of the detected MMR attenuating Msh6 variants on methylation-damage-induced mutagenesis, mESCs were exposed to the methylating DNA damaging agent N-methyl-N’-nitro-N-nitrosoguanidine (MNNG) and the number of cells that consequently attained mutations was quantified. In MMR-proficient cells, DNA replication across MNNG-induced O6-methylguanine lesions is impaired by futile cycles of MMR, ultimately leading to cell death and suppression of methylation-damage-induced mutagenesis. Under MMR-deficient conditions, however, the MNNG-induced mismatches are not recognized and remain in the genome leading to the accumulation of mutations. To provide a quick read out for the frequency of mutation accumulation, we measured the number of MNNG-exposed cells that became resistant to a high dose of 6TG for an extended period. Solely cells that carry an inactivating mutation in Hprt survive stringent 6TG treatment because HPRT is required for 6TG to behave as a DNA damaging agent. All detected Msh6 variant cell lines showed an elevated MNNG-induced mutator phenotype when compared to the MMR-proficient Msh6+/- mESCs (Fig 6).
According to literature MSH6-G566R may be pathogenic [29,44], yet our screen did not identify this variant in 6TG-resistant colonies. Hence we investigated whether the MMR abrogating effect of Msh6-G565R could have been missed by the screen due to technical difficulties. Rather than applying 6TG selection after oligonucleotide-directed mutagenesis, we purified Msh6G565R/- mESCs using a Q-PCR-based protocol [25] (S2C Fig) and subsequently examined their MMR capacity. Exposure of Msh6G565R/- cells to increasing doses of 6TG revealed that they were equally sensitive to 6TG as Msh6+/- cells (Fig 7A). In the MSI assay, Msh6G565R/- mESCs did not experience significantly more slippage events than the MMR-proficient control (Fig 7B). Thus, Msh6-G565R did not attenuate MMR consistent with the oligonucleotide-directed mutagenesis screening result.
The results of our study demonstrate the oligonucleotide-directed mutagenesis screen we previously described for the characterization of MSH2 VUS [22] can be extended to MSH6 VUS. Combining oligo targeting in Msh6+/- mESCs with 6TG selection and sequence analysis allows pathogenic MSH6 variants to be distinguished from polymorphisms. The efficacy of the genetic screen was established in a proof of principle study with 4 known pathogenic MSH6 mutations and 5 polymorphisms. This number was low because of the paucity of MSH6 variants that were classified with 100% certainty. Not one of the 5 non-pathogenic variants was identified as MMR abrogating. Also, among the 26 MSH6 VUS we subsequently analyzed, not one of the 4 variants classified as likely not pathogenic was identified as pathogenic by our screen. Finally, functional assays established that one of the VUS that was not detected as pathogenic by the screen indeed did not influence MMR activity (G565R). Hence the false positive rate of the screen, i.e., the chance the screen identified a VUS as MMR abrogating while it was a priori or a posteriori identified as (likely) non-pathogenic was <1/10, giving a specificity >90.0%. The sensitivity of the genetic screen is a measure of the false negative rate; it is the likelihood that a pathogenic mutation is not detected. All 6 InSiGHT classified pathogenic and likely pathogenic variants as well as the previously proven pathogenic G1139S mutation were recognized as MMR abrogating by the screen, translating to a sensitivity of >85.7%.
We used the oligonucleotide-directed mutagenesis screen to investigate the MMR capacity of 26 MSH6 VUS. Eight of these were found in suspected-LS patients from two medical centers in the Netherlands. From this clinical cohort, the mouse equivalents of mutations R511G, A587P and F706S were detected by our screen and shown to abrogate MMR. However, R510G and F703S were detected in only 2/5 and 2/4 6TG-resistant colonies, respectively, that had retained two Msh6 alleles, while the other pathogenic variants were present in virtually all colonies diploid for Msh6 (Figs 2B, 3A and 3B). The poorer recovery of R510G and F7103S mutants may have been due to a lower success rate of LMO-mediated base-pair substitution. The pathogenic phenotype observed for these three variants is in line with clinical data: all three variants were detected in patients with MSI-H LS-related tumors and with a family history of LS-related tumors. In the case of VUS A587P and F706S, relatives with LS-related tumors carried the same mutation. IHC also demonstrated MSH6 was absent in the patients encoding MSH6-A587P and MSH6-F706S; the IHC data for MSH6-R511G were inconclusive.
The other 5 variants in the clinical cohort, A25S, E221D, G670R, R922Q and c.3438+6T>C, were not identified as MMR abrogating. VUS E221D, G670R and R922Q were found in patients who also harbored a second, known pathogenic mutation in one of the DNA MMR genes that was likely causative for the LS phenotype. E221D was also detected in a second patient who was 83 years old and did not have a family history suspicious for LS. MSH6-A25S was found in a typical LS tumor, i.e., a colon tumor showing MSI, loss of heterozygosity of MSH6, and loss of MSH6 protein expression. The patient however only had one relative with a colorectal tumor and this tumor was not MSI-high and stained positive for all MMR proteins. A previous in vitro study also suggested MSH6-A25S is not pathogenic [45]; it could be that the tumor arose due to a missed somatic mutation. VUS c.3438+6T>C was found in a patient with a family history suspicious of LS. We however do not know if the relatives with LS-associated cancers also carried this specific MSH6 sequence variant. IHC failed in the index patient carrying the c.3438+6T>C variant, therefore we cannot exclude that a somatic mutation or MLH1 hypermethylation caused the MSI in the tumor. Tumor tissue of one family member was tested and showed no MSI and normal IHC. It is also possible that the genetic screen was unable to identify c.3438+6T>C as pathogenic due to differences between the human and mouse MSH6 sequences. While the MSH6 coding sequence is highly conserved, intron sequences are more variable between species (S4 Fig shows human and mouse sequence around c.3438+6). Hence there is a chance that variant c.3438+6T>C affects splicing in man but not in mice. According to several splice site prediction programs (NNSPLICE, GeneSplicer, Human Splicing Finder), however, c.3438+6T>C does not affect splicing.
The other 18 MSH6 VUS we studied were attained from literature and the InSiGHT database. The genetic screen found 5 of these variants abrogate MMR: G686D, L1063R, E1193K, T1219D and T1219I. The detection of G686D and L1063R is in line with their InSiGHT classification, which describes the mutations as likely pathogenic. Variant E1193K has previously been suggested to cause LS in studies that identified the mutation in patients with ECs that were MSI and did not stain for MSH6 [27,28]. Not much clinical data is available for VUS T1219D but Msh6T1217D mice were demonstrated to have increased cancer susceptibility [46]. VUS T1219I has been described in a CRC patient who had a family history of CRC and a MSI tumor that stained positive for MSH6, the latter being consistent with the high levels of this variant protein we observed in mESCs. Both clinical and in vitro data indicate MSH6-T1219I abrogates MMR activity [37,45].
MSH6 VUS R128L, R468H, V509A, Y556F, P623A, S666P, E983Q, R1095C, T1255M and R1304K were not identified as pathogenic in our screen. These sequence variants were classified as likely not pathogenic by InSiGHT, identified in patients with MLH1 promoter methylation or with MSS and MSH6 positive tumors, or observed in patients for whom little clinical data was available. VUS S285I, G566R and T1142M were also not detected as MMR attenuating by our screen, yet they seem suspicious for pathogenicity based on available data. MSH6-T1142M was previously suggested to be probably pathogenic based on clinical data describing the variant in a 27 year old patient with polyps who met the Bethesda guidelines, had a 61 year old mother with polyps, and did not carry pathogenic mutations in any other MMR gene nor showed MLH1 promoter methylation in the tumor [36]. VUS S285I and G566R were detected in CRC patients with MSI (low and high, respectively) tumors that had loss of heterozygosity of MSH6 [29]. Cyr and Heinen [44] investigated the effect of these two mutations on mismatch binding and processing: variant S285I was not found to have a specific MMR attenuating effect but variant G566R was suggested to abrogate MMR by interfering with the ATP-dependent conformational change that must take place to activate downstream repair pathways upon mismatch binding. We therefore purified Msh6G565R/- mESCs and assessed their MMR capacity. The Msh6G565R/- cells behaved like MMR-proficient Msh6+/- mESCs, confirming the result of the oligonucleotide-directed mutagenesis screen. Despite the good performance of our screen and the high amino acid conservation of MSH6, we cannot exclude Msh6-G565R was not identified as pathogenic due to differences between mice and men. To fully dissuade this argument we will need to develop the oligonucleotide-directed mutagenesis screen in human cells.
The oligonucleotide-directed mutagenesis screen presented here is a relatively simple tool that can be used to investigate the pathogenic phenotype of many MSH6 VUS in parallel. While the evolutionary conservation of MMR justifies the use of mouse cells for the majority of VUS, testing of splice-site and intronic mutations necessitates adaptation to human cells. Also, as long as uncertainty exists about its specificity and sensitivity, functional testing needs to be combined with clinical data and in silico estimations to arrive at a reliable classification of VUS. Conforming the updated American College of Medical Genetics and Genomics (ACMG) standards and guidelines for sequence variant interpretation, we are currently transferring our functional tests to certified Clinical Genetics laboratories and creating an infrastructure where test results are compared and interpreted taking into account all available data. In this way, LS mutation carriers can be identified with the highest certainty and enrolled in tailored surveillance programs while relatives without the mutation can be excluded from surveillance.
The genetic screen was developed in Msh6+/- mESCs, which contain one active Msh6 allele (Msh6+) and one Msh6 allele that was disrupted by the insertion of a puromycin resistance marker (Msh6-) [24]. The MSH6 variants under investigation were introduced into the Msh6+/- mESCs by oligo targeting using LMOs [23]. 7x105 Msh6+/- mESCs were seeded in BRL-conditioned medium on gelatin-coated 6 wells and exposed to a mixture of 7.5 μl TransIT-siQuest transfection agent (Mirus), 3 μg LMOs and 250 μl serum-free medium the following day. After 3 days, 1.5x106 LMO-exposed cells were transferred to gelatin-coated 10 cm plates and subjected to 6TG (250 nM) (Sigma-Aldrich) selection. After 10 days the 18 largest 6TG-resistant colonies were picked. Cells that became 6TG-resistant due to loss of heterozygosity events were excluded from further analyses using a PCR specialized to detect the presence of both the disrupted and non-disrupted Msh6 alleles [24]. 6TG-resistant mESCs that maintained both Msh6 alleles were sequenced to confirm the presence of the planned mutation.
Western blot analyses were performed as described in Wielders et al. [25]. Rabbit polyclonal antibodies against mMSH2 (1:500) [47] and mMSH6 (1:500) [24] as well as mouse polyclonal antibody against γ-Tubulin (1:1000; GTU-88 Sigma-Aldrich) were used as primary antibodies. Protein bands were visualized using IRDye 800CW goat anti-rabbit IgG and IRDye 800CW goat anti-mouse IgG secondary antibodies (Li-cor) and the Odyssey scan. The infrared fluorescent signals measured by the Odyssey scan are directly proportional to the amount of antigen on the Western blots, allowing quantification of the protein bands.
mESCs were electroporated with the (G)10-neo Rosa26 targeting vector as described in Dekker et al. [48]. The (G)10-neo Rosa26 targeting vector is composed of a promoterless histidinol resistance gene as well as a neomycin resistance gene (neo) that is rendered out of frame by a preceding (G)10-repeat [42]. Once electroporated, 106 cells were seeded on gelatin-coated 10 cm plates in BRL-conditioned medium and exposed to Histidinol (3mM) (Sigma-Aldrich). Successful integration of the vector into the Rosa26 locus of the Histidinol-resistant colonies routinely occurs at a frequency of ±95% and was confirmed by Southern blot analyses. The individual successfully targeted colonies were subsequently expanded to 107 cells and transferred to gelatin-coated 10 cm plates at a density of 105 cells per plate for Geneticin selection (600 μg/ml) (Life Technologies). After 10 days, the number of Geneticin-resistant colonies was counted and the slippage rate of the variant mESCs calculated using the formula: 0.6 x Geneticintotal = N x p x log (N x p), where Geneticintotal is the number of Geneticin-resistant colonies, N the number of cells to which the culture was expanded, and p the number of mutations per cell division. Experiments were performed in quadruplicate and statistical differences calculated using a one-tailed, unpaired t-test with Welch’s correction.
The MNNG-induced mutagenesis assay was performed as described in Claij and te Riele [43]. 2.5x106 variant mESCs were seeded on an irradiated mouse embryonic fibroblasts feeder layer in 10 cm plates and exposed to 0 or 4μM MNNG (Sigma-Aldrich) for 1h the following day. 40 μM O6-benzylguanine was present in the medium from 1h prior to the MNNG treatment until 6 days after, at which point 1.5x106 cells were transferred to gelatin-coated 160 cm2 plates for 6TG selection (10 μg/ml). After two weeks of 6TG selection, the number of resistant colonies and hence the frequency of MNNG-induced Hprt mutants could be determined. Experiments were performed in duplo and the statistical difference between MNNG-treated Msh6+/- mESCs and MNNG-treated variant cell lines calculated using a one-tailed, unpaired t-test with Welch’s correction.
Msh6G565R/- mESCs were made as described by Wielders et al. [25]. Variant G565R was introduced into Msh6+/- mESCs by oligo targeting and a pure Msh6G565R/- mESC clone was obtained by consecutive rounds of seeding and mutation specific PCR: oligonucleotide-exposed cells were expanded and subsequently seeded on a 96-well plate at a density of 5000 cells per well. A mutation-specific quantitative PCR was used to identify wells that contained Msh6G565R/- mESCs. Positive wells were reseeded at lower density and positive wells again identified by Q-PCR. A pure clone was finally obtained by seeding single cells per well. Sequence analysis confirmed the creation of Msh6G565R/- mESCs.
The 6TG sensitivity of Msh6G565R/- mESCs was investigated by exposing the variant cell line to increasing doses of 6TG, as described in Wielders et al. [49]. MMR-deficient Msh6-/- and MMR-proficient Msh6+/- and Msh6+/+ mESCs were taken along for comparison.
We investigated the pathogenic phenotype of MSH6 VUS that were found in suspected-LS patients at the Clinical Genetics departments of the Erasmus Medical Center Rotterdam and Radboud University Medical Center Nijmegen. We collected tumor characteristics, age at diagnosis, results of molecular diagnostics and germline mutation analysis, and family history from medical records. MSI analysis was performed with the Bethesda panel [50] or with the Promega pentaplex MSI analysis [51]. IHC for MLH1, MSH2, MSH6 and PMS2 protein was performed as described previously [52]. Germline mutation analysis of MSH6 was performed by sequencing and multiplex ligation dependent probe amplification. The in silico prediction model PolyPhen [53] was used to estimate the chance of a variant being deleterious.
|
10.1371/journal.pntd.0003868 | PKC/ROS-Mediated NLRP3 Inflammasome Activation Is Attenuated by Leishmania Zinc-Metalloprotease during Infection | Parasites of the Leishmania genus infect and survive within macrophages by inhibiting several microbicidal molecules, such as nitric oxide and pro-inflammatory cytokines. In this context, various species of Leishmania have been reported to inhibit or reduce the production of IL-1β both in vitro and in vivo. However, the mechanism whereby Leishmania parasites are able to affect IL-1β production and secretion by macrophages is still not fully understood. Dependent on the stimulus at hand, the maturation of IL-1β is facilitated by different inflammasome complexes. The NLRP3 inflammasome has been shown to be of pivotal importance in the detection of danger molecules such as inorganic crystals like asbestos, silica and malarial hemozoin, (HZ) as well as infectious agents. In the present work, we investigated whether Leishmania parasites modulate NLRP3 inflammasome activation. Using PMA-differentiated THP-1 cells, we demonstrate that Leishmania infection effectively inhibits macrophage IL-1β production upon stimulation. In this context, the expression and activity of the metalloprotease GP63 - a critical virulence factor expressed by all infectious Leishmania species - is a prerequisite for a Leishmania-mediated reduction of IL-1β secretion. Accordingly, L. mexicana, purified GP63 and GP63-containing exosomes, caused the inhibition of macrophage IL-1β production. Leishmania-dependent suppression of IL-1β secretion is accompanied by an inhibition of reactive oxygen species (ROS) production that has previously been shown to be associated with NLRP3 inflammasome activation. The observed loss of ROS production was due to an impaired PKC-mediated protein phosphorylation. Furthermore, ROS-independent inflammasome activation was inhibited, possibly due to an observed GP63-dependent cleavage of inflammasome and inflammasome-related proteins. Collectively for the first time, we herein provide evidence that the protozoan parasite Leishmania, through its surface metalloprotease GP63, can significantly inhibit NLRP3 inflammasome function and IL-1β production.
| Leishmania parasites are the causative agent of leishmaniasis, a wide spread disease in tropical and subtropical areas. The microorganisms have been shown to be well-adapted to their hosts and are able to enter their target cells where they replicate themselves. To ensure these processes, Leishmania disrupts a multitude of cellular signals and protective mechanisms, which overall attenuates immune responses against the parasites. A key factor for inflammatory processes, also during infections, is IL-1β. As previous studies suggested a dysregulation of IL-1β levels after infection with Leishmania parasites, we herein investigated the underlying mechanisms. Our work reveals that Leishmania suppressing IL-1β production through its virulence factor GP63. Furthermore, our data suggests that the parasites can dampen the maturation of IL-1β after different stimuli. In this regard we established a role for the suppression of the kinase PKC and the generation of reactive oxygen species, as well as the cleavage of cellular proteins that are important for IL-1β-generation. Thus, we here present a novel aspect for how Leishmania parasites can counteract host protective mechanisms.
| Leishmania parasites, which are the causative agent of leishmaniasis, are able to both survive and proliferate within macrophages. The protozoan parasites evolved strategies to avoid phagocyte activation during infection by seizing control of key signaling pathways [1,2]. Studies previously implicated the metalloprotease GP63—a major virulence factor of Leishmania parasites—in a variety of parasite survival mechanisms. In this context, GP63 has been suggested to affect amongst others Leishmania binding to macrophages, phagocytosis of parasites, evasion of complement-mediated lysis and protozoan migration through the extracellular matrix [1,3]. Furthermore, GP63 has been identified as a key Leishmania virulence factor that modulates cellular signalling through the subversion of host protein tyrosine phosphatase (PTP) function [4,5,6]. In this context GP63-mediated PTP-cleavage, results in the activation of the respective phosphatases. This mechanism was identified for the SH2 domains-containing protein tyrosine phosphatase (SHP-1) and protein-tyrosine phosphatase 1B (PTP-1B) [6]. Besides phosphatases, GP63 has been shown to cleave other targets within the cells including kinases like TAB 1 and transcription factors, including AP-1 and NF-κB [7,8]. The importance of the host PTP-modulation and the subsequent inhibition of signaling pathways is emphasized by the observation that key pro-inflammatory mediators such as nitric oxide (NO), IL-6 and TNFα were subsequently downregulated by Leishmania [4,5,9].
Another factor of pivotal importance for inflammatory processes that has been studied in the past in the context of Leishmania infections is IL-1β. In this regard Leishmania infections have been reported in a variety of studies to alter IL-1β production dependent on the parasite species used [4,10,11,12,13]. This included the deregulation of the IL-1β release due to parasite infections upon usage of known IL-1β inducers like LPS, IFN-γ or nigericin. However, the means used by the parasites to interfere with inflammasome activation remain unclear to date.
IL-1β is translated as an inactive precursor—pro-IL-1β (31 kDa), which is processed into active IL-1β by the multi-protein inflammasome-complex upon stimulation of the cells. Integral components of the inflammasome complexes are caspase-1, responsible for proteolytic cleavage of the IL-1β precursor, [14] a member of the NOD-like receptor (NLR) family, which acts as the sensor component of the inflammasome and ASC, a CARD/PYD protein that serves as a docking and activation platform for caspase-1 and the respective NLR [14]. Dependent on the NLR-protein within the complex, inflammasomes have been shown to respond to a variety of stimuli including bacterial and viral pathogen associated molecular patterns (PAMPs) like microbial nucleic acids or proteins and danger associated molecular patterns (DAMPs) [15,16]. In context of the latter, the NLRP3-containing inflammasome was found to be critical for the recognition of inorganic crystals such as malarial hemozoin (HZ), silica and asbestos, as well as other DAMPs like cardiolipin, ATP and uric acid (MSU) [17,18,19,20]. Canonical NLRP3-inflammasome activation requires two signals. The first signal results in an increased expression of pro-IL-1β and the NLR protein as basal levels are insufficient to facilitate inflammasome activation. Typical initial signals are relayed through pattern recognition receptors such as TLRs or receptors for cytokines (e.g. type I IFNs) [15]. Furthermore, inducers of PKC and MAPK-dependent signaling such as phorbol myristate acetate (PMA) have been used as first signals for inflammasome activation in vitro [21,22]. The second signal induces the oligomerisation and complex formation of the inflammasome that allows the processing of pro-IL-1β. Complex formation and activation can be triggered in different ways including ROS generation, potassium efflux, lysosomal damage and mitochondrial destabilization or damage [23]. Following this rationale of NLRP3 inflammasome organization and activation, modulation—for instance after Leishmania infections—may take place at the level of priming, complex assembly or complex activation as has been shown for a variety of other pathogens that interfere with inflammasomes.
In this study we report that Leishmania parasites, through its virulence factor GP63 inhibit IL-1β production and secretion induced by different NLRP3 inflammasome activators. Importantly, this was observable in both a murine and human model system for Leishmania infection. Dysregulation is achieved due to GP63-dependent interference with signaling pathways upstream of the inflammasome, which affect ROS generation. In this context our data suggests PKC signaling and its downregulation is pivotal for the Leishmania-mediated downregulation of inflammasome activation. In addition, Leishmania GP63 also seems to specifically target components of the inflammasome for proteolytic cleavage that is most likely the basis for the suppression of IL-1β production after ROS-independent activation of the NLRP3 inflammasome.
L929 and THP-1 cell lines were obtained from ATCC (Manassas, VA, USA). Cells were cultured in RPMI-1640 medium supplemented with Penicillin-Streptomycin-Glutamine (PSG; both reagents from Wisent, Saint-Jean Baptiste, QC, Canada) and fetal bovine serum (FBS; Gibco Burlington, ON, Canada). In some experiment the culture medium was exchanged to MEM Alpha medium (Gibco, Burlington, ON, Canada). Reagents used included linezolid, ATP, silica and hemin (> 99% of purity; Sigma-Aldrich, Oakville, ON, Canada); GÖ6983 (Biomol, Farmingdale, NY, USA); asbestos (Structure Probe, West Chester, PA, USA), MSU (Alexis Biochemical, Farmingdale, NY, USA); protease inhibitor cocktail (Roche, Mississauga, ON, Canada) and PVDF membrane (BIO-RAD, Mississauga, ON, Canada). All others unlisted or not indicated reagents were purchased from Sigma-Aldrich (Oakville, ON, Canada) Antibodies used in experiments included anti-human NLRP3 and ASC (Alexis Biochemical, Farmingdale, NY, USA), anti-human pro-IL-1β, anti-human and murine caspase-1 (Santa Cruz; Dallas, TX, USA), anti-phospho-tyrosine/HRP (eBiosciences; San Diego, CA, USA), anti-human mature IL-1β (Cell signaling Technology, Danvers, MA, USA; Rockland-Immunochemicals, Limerick, PA, USA), phospho(Ser)-PKC-substrate antibody (Cell signaling Technology; Danvers, MA, USA), anti-GP63 (obtained from Dr. McMaster, University of British Columbia, Vancouver, Canada) and anti-murine IL-1β (R&D systems, Minneapolis, MN).
Synthetic HZ was generated as previously described [24,25]. Briefly, 0.8 mmol crystalline hemin (>99% of purity) was dissolved in degassed NaOH (0.1M) for 30 minutes with stirring. After, the pH was adjusted with propionic acid to 4 and the material was allowed to anneal at 70°C for 18 hrs. The supernatant was removed and the crystals were incubated three times with NaHCO3 (0.1M) for three hours. In between incubations, samples were briefly washed with milliQ H2O. Thereafter, the crystals were washed three times with methanol and milliQ H2O in an alternating fashion. Subsequently, the samples were dried in a vacuum oven overnight over phosphorous pentoxide (Sigma-Aldrich, Oakville, ON, Canada). Synthetic HZ samples were analyzed by X-ray powder diffraction, scanning electron microscopy (SEM), and infra-red spectroscopy to characterize the crystalline state of HZ.
Leishmania major (L. major), L. mexicana, L. major GP63 knock out (GP63 KO), and L. major GP63 rescue (the GP63 gene was inserted into L. major GP63 KO parasites [26]) were used in different experimental setups. All Leishmania parasites were maintained at 25°C in SDM-79 culture medium supplemented with 10% FBS by bi-weekly passage and used for different applications after 6–7 days of culture (stationary phase). Stationary phase parasites were either used to infect macrophages (at a ratio of 20:1), to recover the culture supernatant for GP63 (L. mexicana) purification or to generate parasite secretome or exosome preparations. To generate Leishmania-conditioned medium (LCM), all species of Leishmania were adapted to grow in DMEM medium (Wisent, Saint-Jean Baptiste, QC, Canada) supplemented with 10% of FBS and 1% of PSG. After 7 days in culture, LCM was collected by centrifugation of parasite cultures (1,000 x g, 5 min) and subsequent filtration with 0.22 μm filters.
Leishmania GP63 was purified using an immunoaffinity column. The antibodies used to purify Leishmania GP63 were specific to L. mexicana GP63 [27]. The antibody was cross-linked using the Affi-Gel HZ Immunoaffinity kit (BIO-RAD, Mississauga, ON, Canada). GP63 was purified from the supernatant of stationary L. mexicana cultures and concentrated by centrifugation using Amicon Ultra centrifugational filters (EMD Millipore, Etobicoke, ON, Canada) at 4,000 rpm for 10 minutes and stored at -80°C.
Exoproteome was prepared as described previously [28]. Briefly, stationary L. mexicana promastigotes were washed 3 times with PBS, incubated in phenol red-free and serum free DMEM (Wisent, Saint-Jean Baptiste, QC, Canada) for 4 hrs and culture supernatants were centrifuged twice at 4,000 rpm for 10 min. Subsequently, the material was either concentrated using 10 kDa cut off Amicon Ultra centrifugational filters (EMD Millipore, Etobicoke, ON, Canada) and used as secretome preparations or centrifuged (100,000 x g, 60 min at 4°C) to isolate exosomes. Protein concentration was determined using Bradford reagent (BIO-RAD, Mississauga, ON, Canada).
THP-1 cells were cultured with RPMI-1640 medium supplemented with 10% FBS, 1% PSG, 50 μM of 2-β-mercaptoethanol, 4.5 g/L of Glucose and 1 mM sodium pyruvate. For THP-1 differentiation 1.5 x106 cells/mL were incubated with 0.5 μM of PMA. After three hrs cells were washed, plated (0.75 x 106 cells/mL in 6 wells plates) and incubated for 20–24 hrs. As a consequence the phagocytic properties of the cells were increased and expression of inflammasome proteins and pro-IL-1β was induced. In some experiments THP-1 cells were incubated with 50 ng/ml PMA for 24 hrs. Cells were infected with indicated Leishmania spp. at a ratio of 1:20 (macrophages:parasites) or incubated with purified GP63, secretome preparations, exosomes preparations or Leishmania culture medium (LCM). Cells were washed after times of infection dependent on the experimental setup and the medium was replaced with MEM Alpha medium without FBS. Cells were subsequently stimulated with indicated concentrations of HZ, silica, asbestos, MSU or ATP for 6 hrs. Linezolid was incubated for 18 hrs.
Bone marrow cells were obtained by flushing out the femurs and tibias from 6 weeks old C57Bl/6 mice. Subsequently, erythrocytes were lysed using NH4Cl (155 mM) in Tris/HCl (10 mM), pH 7.2. Bone marrow derived cells were counted, seeded and incubated in RPMI-1640 medium supplemented with 1% of PSG, 10% FBS and 30% (v/v) L929 cell culture supernatant. Cells were cultured for 7 days, exchanging the culture media every second day. For assays, BMDM were harvested and seeded (0.75 x 106/mL) in RPMI medium supplemented with 5% FBS and 1% of PSG. The following day, cells were primed with LPS (100 ng/ml, 3 hrs) and infected with Leishmania spp. at a ratio of 1:20 (macrophages:parasites). After infection for variable time periods dependent on the experimental setup, cells were washed, medium was replaced with MEM-Alpha without FBS and cells were stimulated with indicated concentrations of HZ or linezolid for 6 hrs or 18 hrs respectively.
Supernatants were collected at indicated time points and proteins were precipitated with trichloroacetic acid at a final concentration of 10%. Precipitated proteins were dissolved in Tris/HCl 0.1 mM pH 8.0 and laemmli sample buffer [29]. Cell extracts were obtained by lysing cells with either Igepal (Sigma-Aldrich, Oakville, ON, Canada) containing lysis buffer (1% Igepal in PBS, 20% Glycerol, protease inhibitor cocktail, 2 mM Na3VO4 and 1 mM NaF) or for caspase-1 detection Triton-X-100 (Fisher Scientific, Walham, MA, USA) containing lysis buffer (1% Triton-X-100 in 10 mM Tris/HCl pH 7.5, 150 mM NaCl, 5 mM EDTA and protease inhibitor cocktail). Supernatant and cell lysate samples were subjected to SDS-PAGE and immunoblot analysis. SDS-PAGE/Immunoblot: SDS-PAGE and Immunoblot were performed following protocols previously published [30]. For the detection of caspase-1 p10, 4–12% NuPAGE gels (Invitrogen) were used. After protein transfer onto PVDF membranes, detection of target proteins was achieved through specific primary antibodies and matched secondary HRP-conjugated antibodies.
NUNC maxisorb 96 well plates (Nalge NUNC, Richester NY, USA) were coated with 100 μl/well of capture antibody (SET TO GO kit, eBiosciences, San Diego, CA, USA) overnight and blocked with 200 μl/well assay diluent solution 1 hr at RT. After blocking, 100 μl of standard proteins or samples were added to each well and incubated for 2 hrs. After 5 washes, 100 μl/well of detection antibody were added and incubated 1 hr at RT. For cytokine detection, 100 μl/well of Avidin-HRP were added and incubated for 30 min. Afterwards, 100 μl/well of substrate solution were added for 15 min. 50 μl of stopping solution were added and plates were read at 450 nm in an ELISA reader (Elmer EnSpire Multimode Plate Reader, Perkin Elmer, Waltham, MA, USA) and concentrations were calculated according to a standard curve.
PMA-differentiated THP-1 cells (0.1x106 cell/100 μl) were seeded in opaque 96 well plates. Cells were infected as indicated with Leishmania parasites for 2 hrs. Cells were washed with PBS and incubated with phenol red free RPMI (Wisent, Saint-Jean Baptiste, QC, Canada) containing 20 mM of 2,7-dichlorofluorescein diacetate—DCFH-DA (Sigma-Aldrich, Oakville, ON, Canada) for 10 min at 37°C. Subsequently, cells were stimulated as indicated adding inflammasome activators. The rate of DCFH-DA oxidation to DCF was observed with a SpectraMax M3 (Molecular Devices, Sunnyvale, CA, USA) fluorescent plate reader at a 488 nm excitation wavelength and a 525 nm emission wavelength.
Unpaired Student’s t-test was used when comparing two groups. The differences were considered significant for p < 0.05. Statistical analysis was performed using Prism 5.00 software (GraphPad, San Diego, CA).
C57BL/6 mice were purchased from Charles River Laboratories and Jackson Laboratories, and were kept in pathogen-free housing. All research involving mice was carried out according to the regulations of the Canadian Council of Animal Care and was approved by the McGill University Animal Care Committee under ethics protocol number 4859. Mice were euthanized using CO2 asphyxiation followed by cervical dislocation.
Upon activation macrophages can produce a large array of pro-inflammatory molecules including IL-1β, which is produced by inflammasome complexes. The NLRP3 inflammasome acts as an intracellular signaling platform which is able to sense a variety of exogenous signals like asbestos, silica [17] as well as the malarial pigment HZ [20,31] and DAMPs such as ATP [18] and MSU [21]. In initial dose-response experiments using synthetic HZ we confirmed the ability of HZ to induce IL-1β maturation in PMA-differentiated THP-1 cells (S1 Fig). The observed, dose-dependent IL-1β secretion was comparable to the results obtained in the case of silica or MSU treatment of cells. Infection of THP-1 cells with Leishmania did not induce IL-1β release as shown for L. mexicana (S2 Fig). Leishmania parasites are well known for their ability to block and inhibit various microbicidal functions of macrophages [32], therefore we sought to elucidate whether infections with Leishmania would inhibit IL-1β production by macrophages, which has been indicated to possibly serve a host protective role in murine models of infection [33,34]. As previously introduced the NLRP3 inflammasome activator HZ causes the IL-1β maturation and secretion in PMA-differentiated macrophages. Pre-infection with L. mexicana and L. major, revealed a parasite-dependent block of IL-1β maturation and release (Fig 1A). The impaired production of IL-1β was not restricted to a system of human cell culture but was also observable when parasite infection preceded inflammasome activation in murine BMDMs (Fig 1C). In those experiments due to the availability of antibodies the processing of caspase-1 into its active fragments p10 and p20 could be observed. Notably processing of caspase-1 was absent after infection of cells possibly due to the lack of inflammasome activation or complex formation. Identical experimental setups for infections as in Fig 1A using L. major GP63 KO and L. major GP63 rescue parasites support the hypothesis that Leishmania’s capability to inhibit IL-1β maturation and release was GP63-dependent (Fig 1A–1E). Thus, IL-1β secretion was not impaired in the absence of the protease (Fig 1A). This finding was further supported by experiments using pretreatment of THP-1 cells with Leishmania culture supernatant (LCM) instead of parasites (Fig 1B). The attenuated IL-1β secretion coincided with the presence of GP63 in the LCM. We were able to show that LCM of L. mexicana and L. major GP63 wild type parasite cultures contained GP63 (Fig 1D). To further evaluate the impact of secreted leishmanial factors like GP63 on IL-1β production the supernatant of L. mexicana cultures was concentrated and used in titration experiment on PMA-differentiated THP-1 cells (Fig 1E). IL-1β maturation and release were stimulated by a variety of known NLRP3 inflammasome inducers, namely HZ, silica and asbestos. In all cases we observed an inhibition of IL-1β secretion in a dose-dependent manner by the L. mexicana culture supernatant. Leishmania GP63 can be found either intracellular in the protozoan endoplasmic reticulum, membrane bound via a GPI anchor or secreted without the GPI anchor. During infection GPI anchored, membrane bound GP63 (GPI-GP63) can also be cleaved and released into the supernatant [2,35]. Therefore, the similarities of our results using either Leishmania infections or Leishmania culture supernatant are most likely to be attributed to the presence and activity of GP63.
In recent years, several studies have described the observation that proteins are secreted as exosecretome by Leishmania parasites upon 37°C temperature shock [28] or as exosomes during the culture of parasites [36,37]. Both Leishmania secretome and exosomes have been shown to contain GP63. Thus, we evaluated whether leishmanial secretome or exosome preparations would inhibit IL-1β maturation and secretion. As expected the results were in accordance with the data acquired using parasite infections and culture supernatant treatment of cells, with both secretome and exosomes inhibiting IL-1β production induced by either HZ or MSU (Fig 2C). The importance of the metalloprotease GP63 was clearly demonstrated through experiments using purified GPI-GP63 (Fig 2A and 2B) from L. mexicana stationary phase cultures to pretreat THP-1 cells. This resulted in an attenuation of the HZ-induced IL-1β maturation and its release (Fig 2B) identical to infection experiments previously shown. Collectively, these results provide convincing evidence that Leishmania GP63 is the causative factor for an impaired IL-1β production by the NLRP3 inflammasome complex, which was observed after infections with Leishmania parasites prior to cell stimulation.
Activation of the NLRP3 inflammasome due to DAMPs is often associated with ROS production and ROS-induced or -dependent signaling [38,39]. In this context the molecular basis of ROS generation has been under debate for some time and recent hypothesis include damage to mitochondria as a possible ROS-source and propose thioredoxin-interacting protein TXNIP may act as a ROS-sensor [40]. Leishmania has been shown to interfere with the generation of ROS and other microbicidal molecules [15] and has been described to be involved in the inflammasome activation [15]. Using known danger molecules like HZ and silica we determined, that both crystalline agents readily induce the generation of ROS in THP1 cells (Fig 3A). Therefore, we hypothesized that a Leishmania-dependent decrease of ROS-species or an impaired ROS production could be the basis of the diminished IL-1β maturation/release previously observed. Consequently, infection of THP-1 cells with L. mexicana led to an abrogated ROS production even after HZ or silica stimulation, supporting our hypothesis (Fig 3B). As our previous results suggested the possibility of a GP63-mediated inflammasome suppression, we included purified L. mexicana GP63 (pGP63) in our experimental setup. Pretreatment of THP cells with pGP63 from L. mexicana supernatant was also sufficient to reduce ROS levels to a similar extend as Leishmania infections (Fig 3B).
We previously presented evidence, that HZ-induced NLRP3 inflammasome activation is dependent on Syk activation and signaling [20]. Furthermore, in a variety of studies it has been suggested that Syk activation in turn can be coupled to PKC signaling [41]. PKC activation has previously been associated with ROS production [42,43]. Therefore, we sought to analyze, whether HZ affects PKC activation as well as PKC-mediated phosphorylation and if PKC-dependent signaling may be of importance for the HZ-driven ROS production and inflammasome activation. The analysis of PKC-dependent protein phosphorylation in PMA-differentiated THP-1 cells and BMDMs revealed, that HZ indeed led to an increased phosphorylation of PKC substrates (Fig 4A and S3 Fig). Specific inhibition of PKC using the PKC-inhibitor GÖ6850 [44] was able to counteract the augmented PKC-substrate phosphorylation levels after application of HZ (S4 Fig). The examination of ROS production after the loss of PKC-dependent phosphorylation and PKC-signaling revealed a significant decrease of intracellular ROS-generation in both THP-1 cells (Fig 4B, upper panel) and LPS-primed BMDM (Fig 4B, lower panel). Additionally, to analyze whether PKC-mediated ROS generation was connected to the attenuated IL-1β maturation and/or release we investigated IL-1β levels after PKC-inhibition. In accordance with the data shown previously, IL-1β maturation is abrogated after suppression of PKC-dependent signaling through the application of the PKC inhibitor GÖ6850 (Fig 4C). Interestingly, we already established in the past that PKC activation can be negatively modulated by Leishmania infections [45]. Consequently, experiments with L. mexicana preceding HZ stimulation showed that PKC-dependent phosphorylation in this context is clearly altered by Leishmania parasites presenting a possible explanation for our previous observations (Fig 4D). Through the use of L. major GP63 wild type (Lmj WT) and L. major GP63-/- (Lmj KO) parasites we were able to further substantiate the dependency of PKC-dependent ROS-reduction on GP63 activity in both THP-1 and BMDM cells (Fig 4E) Taken together our results suggest that PKC activation is a signaling event upstream of IL-1β production after HZ stimulation, which is disrupted by Leishmania. In conclusion the analysis of ROS-production and PKC-signaling after infection, the use of pGP63, GP63-/- parasites and the chemical inhibition of PKC, suggest that PKC-dysregulation most likely through GP63 impairs IL-1β release.
Our previous data showed that Leishmania infections of macrophages prevent the maturation of pro-IL-1β to mature IL-1β upon stimulation possibly intervening with a host protective effect. Thus far, the impairment of IL-1β after infection was attributed to a suppression of PKC-dependent signaling and the loss of ROS production. As the enclosed data supports that GP63 is closely associated with these events, we wanted to examine if Leishmania parasites and GP63 may also interfere with inflammasome activation through proteolytic cleavage of inflammasome components. We and others demonstrated that GP63 can cleave targets containing the following amino acid-motives: polar/hydrophobic/basic/basic amino acids (P1- P’1-P’2-P’3) [43,46]. A first indication for GP63-dependent cleavage of inflammasome components was obtained by experiments using BMDMs. There, we observed that after infection with Leishmania we were able to observe cleavage of pro-IL-1β (Fig 1C). An in silico sequence analysis for putative GP63 cleavage sites revealed the possibility of additional GP63 cleavage sites in the sequences of inflammasome complex and associated proteins. Thus, the sequences of human and murine NLRP3; pro-IL1β and TXNIP—a protein that has been suggested to possibly be involved in ROS-mediated inflammasome activation [40]–contain putative cleavage sites for GP63 (Fig 5A and 5C). As GP63 facilitated cleavage is not necessarily restricted to the proposed cleavage motif and to confirm our in silico findings we performed Western blotting analysis of infected THP1 and LPS-primed BMDM cells. In accordance with the in silico data, GP63 seemed to be able to directly interfere with the inflammasome complex. Thus, we observed cleavage of NLRP3 after infection with Leishmania (Fig 5). This process was GP63-dependent as illustrated by the results for L. major wt, GP63KO and GP63 rescue parasites. Although, the in silico analysis did not suggest a GP63-mediated cleavage of ASC or Caspase-1 we choose to analyze both as they are an integral part of the inflammasome complex. Neither pro-caspase-1 nor ASC showed any cleavage after infection. In addition, as anticipated cleavage and/or cleavage fragments were detected for pro-IL-1β and TXNIP in lysates of cells infected with L. mexicana, L. major wt or L. major GP63 rescue expressing GP63, but not in cells infected with L. major GP63 KO. Taken together our data suggests that Leishmania is able to impair inflammasome activation through different GP63-dependent alterations of proteins and signaling pathways.
A controversial question of inflammasome activation is the dependency of ROS for the activation and assembly of the complex. Recent data associates mitochondrial damage with the activation of the NLRP3 inflammasome. In this context cardiolipin seems to work as a DAMP, able to induce inflammasome complex formation and ultimately IL-1β. The previous data presented different ways how leishmanial GP63 can suppress inflammasome activity. This included the cleavage of NLRP3. Therefore, we wanted to know whether these processes might also have a direct effect on IL-1β production. Thus, we observed the IL-1β generation in a ROS-independent experimental setup using the antibiotic linezolid [19]. Linezolid stimulation of either THP-1 or LPS-primed BMDM cells resulted in the maturation of IL-1β as previously published (Fig 6A and 6B). When cells were infected prior to linezolid stimulation, IL-1β release appeared reduced for both murine and human cells. In accordance to previously shown data Leishmania infection alone did only lead to a minimal production of mature IL-1β (Fig 6A). Taken together this finding may indicate, that the infection of cells with Leishmania can abrogate both ROS-dependent and -independent inflammasome activation, possibly through different mechanisms as both ROS-inducing signaling events are blocked and inflammasome components are cleaved due to the leishmanial protease GP63.
Leishmania parasites have evolved many mechanisms to hijack macrophage microbicidal functions in order to survive and proliferate within the phagocytes. In the present work we addressed how Leishmania parasites can attenuate IL-1β production through the leishmanial virulence factor GP63 during infection. In the past, the activity of inflammasomes and the associated production of especially IL-1β has been correlated with the host protection against parasitic infections, for instance in the case of T. cruzi or T. gondii [47,48,49]. In the case of Leishmania parasites the importance and the role of inflammasomes and IL-1β is very controversially discussed, mainly due to the use of different Leishmania species, different leishmanial developmental stages and different infection models. Previous work indicated the possibility of a species-dependent dysregulation of inflammasomes and inflammasome-related pathways and implicated different leishmanial virulence factors. Reports showed that L. donovani and L. tropica do not induce IL-1β production, and negatively modulate the capacity of IFN-primed human or LPS-primed murine peritoneal macrophages to produce IL-1β upon activation [10,50,51]. The focus of several studies was a parasite-mediated dysregulation of IL-1β on a transcriptional level after infection of human [52,53] or murine phagocytes [54]. In this context, Hatzigeorgiou and collaborators [55] implicated a LPG-dependent interference with IL-1β mRNA that translated into both decreased stability and production of IL-1β mRNA and consequently reduced transcription of the IL-1β gene [52,55]. However, data of Cillari et al. [54] and Gurung et al. [56] indicated that L. major infections might increase inflammasome activity and cytokine production during long term infections. Some reports indicate that the role the inflammasome is dependent on the model for infection with a possible delay in the resolution of cutaneous lesions in the absence of IL-1β [33,34]. Thus indications exist for a possible inflammasome-mediated host protective mechanism in some murine models of infection. In contradictory reports working with the new world Leishmania species L. amazoniensis, parasites have been described to be able to both suppress and induce IL-1β secretion by infected cells. On the one hand a study by Ji et al. provided data that the infection of C57Bl/6 mice with L. amazoniensis led to a delay in the secretion of chemokines and cytokines including IL-1β in vivo [11]. On the other hand a recent report of Lima-Junior et al. presented data that inflammasomes and IL-1β are involved in the control of L. amazoniensis infections of C57Bl/6 mice as shown by in vitro and in vivo studies using mice and BMDMs of deficient in IL-1β production (including caspase-1 and NLRP3 KO mice) [13].
Our infection experiments with L. major and L. mexicana revealed an attenuated capability of macrophages to produce and secrete IL-1β when stimulated with the specific and well characterized NLRP3 inflammasome agonist HZ [20]. This we observed in C57Bl/6-derived BMDMs as well as after infection of human cells, which may indicate a role in the circumvention of a host protective mechanism by Leishmania. In PMA-differentiated THP-1 cells as well as in LPS-primed macrophages, L. major and L. mexicana inhibited NLRP3 inflammasome activation as indicated by reduced levels of secreted IL-1β. A possible explanation to the divergent result to previous reports, like Lima-Junior et al. [13], could be the difference in Leishmania spp. used. As introduced before, especially for L. amazoniensis previous data has been controversial. Furthermore, it is to be noted, that L. amazoniensis exhibits a rather unique pathogenesis and a very peculiar intracellular compartmentalization after host infection, which is not observable with the species used in our report [57]. Another crucial difference is potentially the experimental setup, specifically the time of incubation used to detect IL-1β production. It is in fact conceivable that longer periods of infection as examined in the work of Lima-Junior et al. may affect the secretion of IL-1β through cell death related events [58,59]. On this note, data published by Gomes et al. is noteworthy [60]. In their experiments using L. braziliensis IL-1β production was dependent on the developmental form of the parasite used. Infections with amastigotes led to IL-1β maturation while promastigotes did not. Thus the transition from promastigotes to amastigotes during infection may be of the essence for a Leishmania-mediated effect on IL-1β maturation. In addition, we want to point out the fact that infection in our experiments preceded inflammasome activation while previous work predominantly analyzed IL-1β levels over time after infection in vivo or after stimulation of cells with an inflammasome inducing agent like LPS and subsequent infection in vitro [11,13]. Collectively, our results may indicate that an initial block of IL-1 maturation may prevent a host protective effect, thus facilitating parasite survival.
Importantly, none of the previous reports analyzed the possibility of a GP63-mediated effect on IL-1β maturation. In our study, we were able to observe an attenuated IL-1β production using not only parasites but Leishmania culture supernatants, leishmanial secretome and exosome preparations as well, all of which have been shown to contain the metalloprotease GP63 [2,28]. Moreover, purified GP63 exhibited similar effects when used on human THP-1 cells prior to inflammasome activation with HZ. We previously showed that the malaria pigment HZ elevates ROS-levels leading to inflammasome activation [20]. ROS have been shown to contribute to parasite clearance and are inhibited by Leishmania parasites [1]—an effect that can be mediated through the activity of the metalloprotease GP63. Although at this point we cannot rule out the involvement of other leishmanial factors in the inhibition of ROS generation, our data and the usage of purified recombinant GP63 strongly suggests an important role of the protease in this context. Our work identified PKC-signaling as the mechanism upstream of the observed ROS induction after treatment of THP-1 cells with HZ. In the past, different studies presented evidence for a role of PKC both upstream and downstream of ROS generation [42,43,61]. Our results show that PKC-signaling and especially PKC-dependent ROS-generation can mediate inflammasome activation. Some studies previously indicated similar implications for PKC in the activation of the NLRC4 inflammasome [62]. In this context PKCδ-mediated NLRC4-phosphorylation was suggested as the basis of the observed effects [62]. Thus, we here clarify how PKC signaling may also affect NLRP3 inflammasome activation in response to DAMPs like HZ.
In agreement with previous data we were also able to establish that Leishmania is capable to alter the previously introduced HZ-mediated PKC activation after Leishmania infection. As a consequence stimulated cells exhibited a loss of IL-1β production. Initial reports using L. donovani suggested that LPG was involved in the alteration of PKC-signaling as purified LPG prevented PKC activation in macrophages after stimulation with either LPS or PKC-activators [63]. Nevertheless, it has been described in recent years that the inhibition of the oxidative burst in macrophages as well as the associated signaling events, including PKC, were in part mediated by the parasites surface molecules LPG and GP63 [45,64] after infection. Our findings are also corroborated by previous reports that revealed a GP63-mediated interference with PKC-signaling and PKC targets. In this regard, Corradin et al. presented evidence that the PKC substrates myristoylated alanine-rich C kinase substrate (MARCKS)-related protein (MacMARCKS) and myristoylated alanine-rich C kinase substrate (MARCKS), the latter which is of importance for cell motility, adhesion, endo-, exo- and phagocytosis as well as for the interplay of calmodulin and PKC signaling, [65] are targeted by GP63 for cleavage [66,67].
The interference of pathogens with inflammasomes has been shown in a number of bacterial or viral infections. This includes the expression of decoy proteins that bind NLRs or ASC, factors that block caspase-1 activity or scavenger receptors for IL-1β [68]. Interestingly, this also includes Zmp1 a Zn2+-metalloprotease expressed by Mycobacteria spp. that interferes with caspase-1 activation [69]. The leishmanial Zn2+-metalloprotease GP63 has been shown to facilitate cleavage of a multitude of cellular substrates, most notably cellular phosphatases. In silico data using the GP63-cleavage motif [46] indicated a possible GP63-mediated processing of inflammasome or inflammasome-associated proteins, including NLRP3 and pro-IL-1β, which we were able to confirm in infection experiments of both murine BMDMs and human THP-1 macrophages. Interestingly, we observed that one of the GP63-cleaved proteins was TXNIP, which has been shown to facilitate ROS-dependent inflammasome activation [40]. Thus, our data indicates that Leishmania may employ different GP63-linked strategies to impair secretion and maturation of IL-1β during infection, the downregulation of ROS on the one hand and the cleavage of inflammasome and inflammasome-related proteins on the other hand. The relevance of the latter mode of inflammasome inhibition is illustrated by the diminished release of IL-1β after stimulation of cells with linezolid, a ROS-independent inflammasome inducer [19].
IL-1β has been associated with the control of parasitic infections possibly including various Leishmania species. Collectively, we here provide evidence that Leishmania major and mexicana parasites are able to dampen IL-1β secretion during initial stages of infection, rendering cells non-responsive towards stimulation of the NLRP3 inflammasome. This may substantiate a host protective mechanism that has been suggested previously [33]. Moreover, we here show that the observed reduction of IL-1β maturation after infection takes place in both a murine and a human infection model. Our finding that the parasites can impair cytokine secretion through both the downregulation of ROS and possibly the proteolytic cleavage of inflammasome and inflammasome-related proteins strongly supports an important role of this mechanism in the formation of infection. Thus, our data presents a novel way whereby Leishmania ensures the infection of their target cells emphasizing the parasites ability to overcome host protective functions during infection.
|
10.1371/journal.pcbi.1002455 | The Emergence and Early Evolution of Biological Carbon-Fixation | The fixation of into living matter sustains all life on Earth, and embeds the biosphere within geochemistry. The six known chemical pathways used by extant organisms for this function are recognized to have overlaps, but their evolution is incompletely understood. Here we reconstruct the complete early evolutionary history of biological carbon-fixation, relating all modern pathways to a single ancestral form. We find that innovations in carbon-fixation were the foundation for most major early divergences in the tree of life. These findings are based on a novel method that fully integrates metabolic and phylogenetic constraints. Comparing gene-profiles across the metabolic cores of deep-branching organisms and requiring that they are capable of synthesizing all their biomass components leads to the surprising conclusion that the most common form for deep-branching autotrophic carbon-fixation combines two disconnected sub-networks, each supplying carbon to distinct biomass components. One of these is a linear folate-based pathway of reduction previously only recognized as a fixation route in the complete Wood-Ljungdahl pathway, but which more generally may exclude the final step of synthesizing acetyl-CoA. Using metabolic constraints we then reconstruct a “phylometabolic” tree with a high degree of parsimony that traces the evolution of complete carbon-fixation pathways, and has a clear structure down to the root. This tree requires few instances of lateral gene transfer or convergence, and instead suggests a simple evolutionary dynamic in which all divergences have primary environmental causes. Energy optimization and oxygen toxicity are the two strongest forces of selection. The root of this tree combines the reductive citric acid cycle and the Wood-Ljungdahl pathway into a single connected network. This linked network lacks the selective optimization of modern fixation pathways but its redundancy leads to a more robust topology, making it more plausible than any modern pathway as a primitive universal ancestral form.
| The existence of the biosphere today depends on its capacity to fix inorganic into living matter. A wide range of evidence also suggests that the earliest life forms on Earth likewise derived their carbon from . From these two observations one can assume that the global biological carbon cycle has always been based on , and we show here that this assumption can be used as a powerful constraint to help organize and explain the deep evolution of life on Earth. Using a novel method that fully integrates aspects of metabolic and phylogenetic analysis, we are able to reconstruct the complete early evolutionary history of biological carbon-fixation, relating all ways in which life today performs this function to a single ancestral form. The diversification in carbon-fixation appears to underpin most of the deepest branches in the tree of life, and this early metabolic diversification – reaching back to the first cells – appears to have been driven not by the contingencies of history, but by direct links to the physical-chemical environment. The ancestral carbon-fixation pathway that we identify is different from any modern form, but better suited to the capabilities of the earliest primitive cells.
| Six different autotrophic carbon-fixation pathways have been identified across the tree of life [1]–[3]. It has been recognized that some of these pathways share reaction sequences, but a comprehensive framework does not yet exist to interpret the relatedness among all these extant phenotypes, or to judge which if any is the best candidate for a preserved ancestral form [3]–[6]. Phenotypes exclusive to both the bacterial and archaeal domains have been found, but a full explanation for the patterns of exclusivity has not yet been given [1], [2]. Discussions of the ancestral mode of carbon fixation have focussed primarily on the Wood-Ljungdahl (WL) pathway [7] – as part of the reductive acetyl-CoA pathways – and on the reductive citric acid cycle (rTCA) [8], but diverse observations of metabolic universality and simplicity, network topology, and phylogenetic distribution have not yet been given a single compatible interpretation [3]–[6]. As will be elaborated upon below, throughout this work we use the concepts of carbon-fixation ‘pathways’ and ‘phenotypes’ interchangeably as we focus only on the initial biochemical sequences of how enters living cells.
While consensus remains elusive on which pathway (if any) represents the ancestral form of carbon fixation, it has become increasingly accepted over the last 30 years that the first forms of life were likely chemoautotrophic, deriving all biomass components from , and all energy from inorganic redox couples in the environment [3]–[6], [9]. (Photoautotrophs, in contrast, derive energy from sunlight, while heterotrophs derive energy and biomass from organic sources of carbon.) Most discussions of autotrophy in the origin of life are complicated because they combine chemical requirements for carbon and energy uptake with considerations of whether organisms or syntrophic ecosystems were required to complete the required pathways, and the ways the resulting population processes would have related to genomics on one hand and to metabolic regulations on the other. The disentangling of these issues is addressed in [10], and we will not revisit them here. Instead we focus on the purely biochemical requirements for autotrophy, and adopt the working hypothesis that at the ecosystem level the biosphere has been autotrophic since its inception. Autotrophy provides a simple yet powerful constraint from which we derive a coherent framework for the evolution of all extant fixation pathways.
The modern biosphere may be described, most fundamentally, as implementing a biological carbon cycle based on , in which carbon fixation is the metabolic anchor embedding life within geochemistry. If the earliest ecosystems were also autotrophic, then a carbon cycle based on must have existed continuously to have supported biosynthesis. Any local ecosystems that could no longer fix either would have become appendages to neighboring autotrophic ecosystems, or would have dwindled and been recolonized by such ecosystems. Under either scenario it must be possible, relative to appropriately defined ecosystem boundaries, to trace all extant carbon-fixation pathways through fully autotrophic sequences to the earliest forms. The question for historical reconstruction is then whether these sequences are sufficiently simple that their innovations could have occurred independently, or whether complex, cross-species and interdependent sequences of innovations sustained continuous carbon-fixation. In the latter case ecosystems become the relevant units of phylogenetic comparison, and clear lineages in carbon-fixation may be difficult, if not impossible, to reconstruct. A history of independent innovations, in contrast, allows us to be indifferent to the distinction of ecosystems and species, and should make clear carbon-fixation lineages distinguishable.
In this paper we reconstruct the evolutionary sequences that relate modern carbon-fixation pathways to each other and to a single ancestral form, showing that the history of carbon-fixation can indeed be described as simple sequences of independent innovations in autotrophy. To define the constraint of autotrophy we will use metabolic-flux balance analysis, and because we do not use it to model cellular-level mechanisms of either regulation or heredity, it does not distinguish among strictly defined autotrophic species, populations of diverse cells (or pre-cells) tightly linked by transfer of genes and metabolites [11], [12], or syntrophic ecosystems. Therefore the ‘carbon-fixation phenotypes’ that we analyze may, but need not, have corresponded to phenotypes integrated within single species. The intensity of gene transfer and the integrity of species lineages thus become moot points, unless they lead to signatures of complexity and non-independence in the sequence of carbon-fixation phenotypes. Moreover, we focus here on the integrated pathways of carbon fixation alone, requiring only that in the bottom-up construction of biomass all initial metabolic branching points be directly accessible from , and do not extend to the more complicated reconstruction of full intermediary metabolism. The remarkble modularity and redundancy of carbon fixation pathways, and the small number of metabolites through which they are connected to the rest of metabolism, make this separation feasible for autotrophy. We return to the complexities that arise upon consideration of heterotrophic organisms and larger components of intermediary metabolism in a later section. We thus describe a diversification of the of input channels of carbon into the biosphere, from which downstream anabolic pathways may diverge in different organisms. (Anabolism is the process by which life constructs larger organic metabolites from smaller ones, while catabolism is the converse breakdown of larger into smaller molecules.) Distinctions caused by the participation of heterotrophs in complex ecosystems, as well as those among organisms that share carbon-fixation pathways but use these to feed diversified intermediary metabolism, all arise outside the networks we model here. The ancestral pathways we reconstruct are therefore best understood as divergences in the ecosystem-level metabolic foundation of an emerging biosphere. Our evolutionary reconstructions are based on a novel integrated metabolomic/phylogenomics approach, whose basic principles we outline next.
The two main statistical tools that exist to probe genetic information and study the early metabolic evolution of life are phylogenetics [13]–[15] and metabolic flux-balance analysis (FBA) [16], [17]. Whole-genome FBA models have become a widely used and successful method to study the metabolism of individual organisms [16], [18]. One can use this approach and target deep-branching organisms in an attempt to study conserved metabolic features near the base of the tree of life. However, due to the inherent ambiguities and errors in using gene sequence comparisons to determine the presence of enzymes [19], [20] and therefore reactions, heuristics are needed to fill the “gaps” in the initial network that is derived from the genome to produce a viable organism model [16], [18]. These techniques work remarkably well in predicting overall growth rate dynamics under various conditions, especially for well-characterized model organisms [16], [18]. It is less clear, however, what confidence to assign to such heuristic rules when targeting individual pathways of evolutionary interest, especially for deep-branching organisms that lack extensive laboratory characterization and that have significant gene divergences from well-characterized model organisms. Similarly, phylogenetics based on gene presence/absence and sequence similarity, without other prior constraints, has given significant insights into the relatedness and historical divergences among organisms [13], [14]. However, branching relationships become more ambiguous at greater phylogenetic depths, as tree-like descriptions fail for whole organisms due to extensive lateral gene transfer (LGT) near the root [12], [14], [21]. This can make the phylogenetic position (and evolutionary divergence) of metabolic pathways uncertain, as has been the case for carbon-fixation pathways [1], [2].
Phylogenetics and flux-balance analyses, if used together, have complementary strengths that may ameliorate some of these problems. We refer to the joint use of the two methods as “phylometabolic” analysis, and illustrate its main features in Fig. 1. Instead of using heuristics to fill gaps in order to complete an initial network derived from genome annotation (as for example in Fig. 1A), we compare the gene profile for a metabolic pathway in a focal organism to those in related organisms both within and across neighboring clades as shown in Fig. 1B. By focusing at the pathway level, the comparison may reveal variations in multi-genic functional units, providing context for the completion of individual organism networks, while also restricting the plausible sequences for divergence in the evolution of metabolic structure. Especially in cases where individual reactions or growth conditions, rather than complete pathways, have been chemically characterized in different organisms, comparison of functional units can pool evidence that would not be restrictive in isolation. Conversely, as shown in Fig. 1C, placing hypothesized constraints on functional units, such as requiring the continuous production of essential metabolites, can lead to specific conclusions about uncertain clade-level branches in phylogenetic trees. This may be understood as adding a semantic dimension for the contribution of genes to phenotypes, which provides a different form of disambiguation, when phylogenies based on gene presence/absence alone yield poorly defined deep branching relationships as summary statistics for reticulated networks of single-gene histories [14], [21]. The reconstructed “phylometabolic tree”, shown in Fig. 1D, by including multiple complete pathways to common essential metabolites, then suggests which evolutionary substitutions are allowed (at either organism or ecosystem levels) among these pathways by the constraint that essential metabolites are continuously produced. As we will show for the evolution of carbon-fixation, complete pathway reconstructions, combined with characterization of reaction mechanisms and enzymes, often suggests the causes for reconstructed divergences. In an analysis of large or highly ambiguous networks, statistical methods already employed for gene phylogenies, or those used to suggest enzyme functions in automated FBA modeling, could be combined into joint-maximum-likelihood or Bayesian Markov Chain Monte Carlo (MCMC) reconstruction algorithms. The small size and high parsimony of the network we will consider permitted manual reconstruction.
Future work will extend these methods to the reconstruction of full models of intermediary metabolism – in a separate manuscript we will describe the metabolic reconstruction of the deep-branching hyperthermophilic chemoautotroph Aquifex aeolicus. For the carbon-fixation networks considered here, important interactions of phylogenetic and flux-balance constraints occur at two distinct points in the network. The first, of the kind shown in Fig. 1B, determines our reconstruction of the ancestral synthesis route to glycine and serine, which from the perspective of input form a unique set among the monomer precursors to biomass (e.g. amino acids, nucleotides). It has been observed [22] that all anabolic pathways originate from five universal precursors: acetyl-CoA, pyruvate, oxaloacetate, succinyl-CoA and -ketoglutarate, and that all of these are intermediates in the citric acid (TCA) cycle. Even in organisms whose carbon-fixation pathways do not pass through this cycle, TCA arcs are used as supplementary pathways to connect primary carbon inputs to these standard precursors. The sole exceptions are glycine and serine (and a few higher order derivatives of these), which, as shown in Fig. 2, can in some organisms be produced directly from , bypassing both the TCA intermediates and all of the recognized complete carbon-fixation pathways. Operation in the fully reductive direction of the folate-based chemistry that forms the core of this pathway has only been observed within the context of the complete Wood-Ljungdahl pathway, in which acetyl-CoA is synthesized as the final step. However, all steps in this sequence to glycine and serine are fully reversible [23], [24], and so there is no reason a priori to exclude the possibility of much wider use of this pathway in the context of carbon-fixation.
In the next section we will present evidence strongly suggesting that this pathway is in fact widely present across the tree of life without the final step to acetyl-CoA in organisms that lack alternate routes to glycine and serine, and that it forms the most likely ancestral pathway to these amino acids. This is surprising, not only because carbon fixation on folates had previously only been recognized in anaerobic organisms using it as the sole route to fix , but also because our results suggests that organisms commonly use two disjoint parallel pathways to fix carbon. The presence of two parallel carbon-fixation pathways in an autotroph was only recently noted in a single, late branching -Proteobacterium [25] (begging the question how common this might be [2]) but even in that case the two pathways are connected through metabolic intermediates. We will show evidence that direct reductive synthesis of glycine and serine combines with all other carbon-fixation pathways, and that in many of these cases the two pathways supply distinct, disconnected components of biomass.
Perhaps the most important consequence of these findings is that it significantly increases the similarity among all carbon-fixation phenotypes. In particular, the most similar carbon-fixation phenotypes are now the deep branching form of rTCA, which combines a complete rTCA cycle with an incomplete WL pathway to synthesize glycine and serine, and Wood-Ljungdahl, which combines a complete direct reductive sequence from to acetyl-CoA with partial TCA cycle sequences to complete the set of universal anabolic precursors. Moreover, among all carbon-fixation machinery, the folate-based direct reduction of and partial TCA sequences appear to have the most universal distributions across the tree of life, supporting an early appearance for both. Together these various observations suggest general principles underlying the evolutionary diversification of carbon-fixation, as well as an avenue to the reconciliation of phylogenetic and metabolic observations that previously appeared incompatible.
The elaboration of these results below will lead to us to a second junction at which we will invoke phylometabolic constraints, this time of the type in Fig. 1 C/D, in reconstructing the complete evolutionary sequences that connect all carbon-fixation phenotypes to a single ancestral form. As we go through these analyses we also make observations on the nature of the associated reactions and enzymes of key steps in the different forms of carbon-fixation, from which we identify plausible ecological and evolutionary explanations for the divergences at different stages. In particular we include a section immediately following the analysis of glycine/serine pathways that describes in detail how one of the major driving forces - energy optimization - is inferred from a wide range of evidence. The discussion of this driving force, and its interaction with others, will then set the stage for interpreting the full evolutionary history of carbon-fixation. We begin, however, by describing direct reduction of one-carbon () units in more detail, which is not only an essential source of core metabolites in all WL organisms, but occurs ubiquitously as a module for carbon fixation in interaction with other autotrophic pathways.
Direct reduction follows the ‘central superhighway’ of tetrahydrofolate (THF) metabolism (some archaea use tetrahydromethanopterin () and related compounds, all analogues of THF [24]), which links the synthesis of purines, thymidilate, formyl-tRNA, serine, and methyl-group chemistry mediated by S-adenosyl methionine (SAM) [24]. The core reactions of this pathway, summarized in Fig. 2, are widespread in both oxidative and reductive form throughout the tree of life [24], [26]. In their reductive form these reactions, followed by the acetyl-CoA synthase reaction – catalyzed by one of the most oxygen-sensitive enzymes in our biosphere [27] – make up the WL pathway, which is the principal carbon fixation route in a variety of bacteria and archaea, including acetogens and methanogens, sulfate reducers, and possibly anaerobic ammonium oxidizers [2], [28]. In autotrophs, the WL pathway couples in a variety of ways to an incomplete rTCA cycle forming what are collectively known as reductive acetyl-CoA pathways. We will present evidence from genome comparisons that direct reduction is not only a carbon fixation route in WL organisms, but that it is ubiquitous, and was actually the ancestral route to glycine and serine, which took on diversified roles independent from the complete WL pathway very early in the rise of oxygen. The extant glycine and serine synthetic pathways provide the key constraints in reconstructing ancestral carbon fixation, so we describe these next. From their phylogenetic distribution, and their energetics in the context of fully autotrophic networks, we then reconstruct a sequence for the major innovations in carbon fixation.
Three main synthetic pathways to glycine and serine are found in modern organisms and provide evidence about ancestral forms. In eukaryotes and most late-branching bacteria, serine is derived from 3-phosphoglycerate (3-PG) as a branch from either glycolysis or gluconeogenesis [26], [29], through reactions of Fig. 2. Serine is subsequently cleaved to glycine, donating a methylene group to THF, and glycine can be further broken down in what is known as the glycine cleavage system to and , donating a second methylene group to THF [23]. We will refer to this route as the “oxidative” pathway because the first step from 3-PG involves the oxidation of the alpha-hydroxyl to a carbonyl group, and parts of the THF pathway operate in the oxidative direction. In the alternative, direct synthesis of glycine from units (reactions in Fig. 2), the THF-mediated reactions proceed exclusively in the reductive direction, and we refer to this route as the “reductive” pathway. The third major route to glycine is via reductive transamination of glyoxylate [29] (reaction 12 in Fig. 2), which is important in cyanobacteria and plants undergoing photorespiration [30]. Following its synthesis from glyoxylate, one molecule of glycine may be cleaved to and , donating a methylene group to THF, which is then combined with a second glycine to produce serine. In photosynthetic organisms, glyoxylate arises from the Calvin-Benson-Bassham (CBB) cycle upon exposure to oxygen [30], but in other organisms it can arise from other sources such as cleavage of isocitrate in the glyoxylate shunt [29]. We track the distribution of glyoxylate transaminase in our gene profile comparisons as the key reaction in all these pathways, and refer to them collectively as the “glyoxylate” pathways. The synthesis of glycine through cleavage of Threonine has been shown to play a role in some organisms (notably Saccharomyces cerevisiae [31], [32]), but its physiological importance is generally not well understood [33], [34]. We interpret this route as a salvage pathway and do not consider it further here.
To understand the distribution and reconstruct the history of glycine and serine synthesis, we acquired gene profiles for all three pathways from the UNIPROT database [35], for all strains in a wide range of deep-branching organisms (see Methods for details). UNIPROT derives from the manually reviewed SWISSPROT database, which has one of the lowest rates of mis-annotation among public databases [20]. We find the complete gene complement for the reductive pathway widely distributed among both bacteria and archaea, as shown in Fig. 3, including many non-WL organisms that lack the genes for alternative routes. This latter group includes members of clades that use the rTCA cycle (the Nitrospirae) [36] or the 3-Hydroxypropionate (3-HP) bicycle (the Chloroflexi) [37], anaerobic and aerobic heterotrophs (the Thermotogae and Isosphaera pallida of the Planctomycetes) [38], [39], and also several archaea (all listed in Table S1 in Text S1). The very wide distribution of direct reduction on THF suggests that hybrid carbon fixation is much more common than has been recognized, and that the reductive pathway is not limited to WL organisms in which it is the sole pathway. Indeed, this pathway appears to be more widely distributed than any single primary fixation pathway.
A detailed distribution of gene profiles for the three pathways is shown in Table 1. Although the reductive pathway requires 7 or 8 steps (heterotrophic from formate or autotrophic from , respectively) to reach glycine, versus 3 and 1 in the alternative pathways, its frequency in the sample is nearly double the combined frequencies of the oxidative and glyoxylate pathways.
The reductive pathway from appears in two common forms. The full pathway (“Red” in the tables) comprises 8 reactions. In the alternative form (denoted “Red(-2)”), reaction 2, attaching formate to THF, does not appear in genome searches based on sequence similarity, but all 7 other reactions are present. We will argue from detailed analysis of gene frequencies and growth conditions, that the 7-reaction form is active and is in fact a carbon fixation pathway, suggesting that formate incorporation is catalyzed by an unrecognized protein or an unidentified function among the known THF-interconversion enzymes. We suspect that an alternate route involving incorporation through rather than of THF may be active in these cases (see Text S1 for further details). In Table 1 pathways from (autotrophic form) and formate (heterotrophic form) are combined under columns Red and Red(-2). For full 8-step autotrophic gene profile see e.g. organisms MTA, CAG and NDE in Table S1 in Text S1. For full alternate 7-reaction autotrophic gene profiles see e.g. organisms AAE, CCH or NPU in Table S1 in Text S1. Full organism names are also found in Text S1. The most informative distributions come from three clades that are consistently placed among the deepest-branching bacteria: the Firmicutes, Thermotogales and Aquificales [40]–[42]. Thermotogales, Aquificales, and several groups of Firmicutes are among the most thermophilic bacteria and are generally restricted to hydrothermal vents [7], [38], [42], [43]. These environments are among the least changed from early-Earth conditions, and clades apparently restricted to them throughout history may well be the most conservative of metabolic features from the base of the tree [44].
Among the Firmicutes, a remarkably diversified clade, the reductive pathway is the common form, the oxidative pathway is less common, and glyoxylate pathways are very rarely found. The only systematic exception to the common pattern among Firmicutes is found in the Lactobacillales, where the ‘glycine cycle’ (reactions in Fig. 2) is completely absent, apparently having been replaced by the oxidative pathway. Indeed, this group shows the most complete such replacement found among deep-branching bacterial clades. The Lactobacillales, however, are mesophiles, highly adapted to environments rich in organic carbon, and are known to have a high degree of associated gene loss and acquisition through LGT [45]. Therefore we conclude that the reductive pathway is the ancestral route to glycine and serine in the Firmicutes.
The Thermotogales and Aquificales are much less diverse than the Firmicutes, comprising almost exclusively hydrothermal vent/spring organisms. Metagenomic evidence suggests the existence of specialized mesophilic Thermotogales [46], [47], but the amino acid composition of reconstructed ancestral states of a large number of gene families supports a highly thermophilic origin for this clade as a whole [42]. Mesophilic Thermotogales have not yet been cultured and our genomic sample includes thermophiles only. As can be seen, all Thermotogales and Aquificales lack two, or even all three, of the genes to synthesize serine from 3-PG, and none has the gene for glyoxylate transaminase. As the alternative, all Thermotogales have a complete reductive pathway, while several Aquificales show the 7-reaction sequence missing only the ATP-dependent formyl-THF synthase (reaction 2). Evidence that formate is, nevertheless, taken up in the reductive pathway by Aquificales comes from the presence of a formate dehydrogenase (reaction 1) that, in obligate autotrophic members of this clade, has been shown not to operate in the oxidative direction [48], [49]. Experiments in [48] followed the protocol of [49] that test only for the oxidative direction and found no activity in obligate autotrophs. Hence, this enzyme likely functions as a reductase rather than a dehydrogenase, despite the gene nomenclature.
In the remaining deep bacterial lineages, oxidative and reductive pathways are co-present, although the reductive pathways remain more common. The glyoxylate pathway is common only in cyanobacteria (for reasons explained below), and otherwise only in the Halobacteria, a late-branching mesophilic archaeal clade restricted to hypersaline environments. Therefore all distributional evidence suggests that not only is the (either latent or active) reductive pathway common across the bacterial domain, but also that it represents the ancestral pathway to glycine and serine. This leads to the general conclusion that the ancestral function of THF-based chemistry was completely reductive starting from , until alternative routes to glycine and serine became available and parts of this chemistry could reverse direction.
The pattern of pathways for glycine synthesis in archaea is more complex than in bacteria, because pterin diversity is greater (see next section and Fig. 4), and their functions and associated enzymes have been characterized in much less detail [24], [50], [51]. However, pathway distributions, in particular the widespread presence of the glycine cycle, continue to suggest the reductive pathway as the archaeal ancestral form. In bacteria this cycle appears to be a good indicator for the presence of the reductive pathway, with nearly all (213 out 217) strains that have a complete glycine cycle showing either the complete (Red) or alternate (Red(-2)) forms of the direct reductive pathway. Among non-methanogenic archaea a majority has a complete or nearly complete cycle. Of these, some show a complete reductive pathway, while most lack only the genes specific to bonding at of THF (reactions in Fig. 2). As the syntheses of archaeal (non-THF) pterins and of THF use different enzymes starting from the first commited step from GTP [50], [51], suggesting a deep evolutionary divergence of these molecules, it does not seem completely surprising that enzymes performing the equivalent pterin- interconversions would not show up in homology searches against the THF- interconversion enzymes, even if this pathway is in reality present and active. This underscores the importance of further characterization of archaeal pterin- chemistry (see also Text S1).
Among archaea a complete oxidative pathway occurs only in the absence of a glycine cycle (a stricter version of a similar alternation in bacteria), and the majority of these cases are found among methanogens whose -based pathways play fewer biosynthetic roles than their THF homologues [24]. In the following section we combine the distribution signature with an energy analysis of the different pathways, from which it can be seen that the loss of the glycine cycle in methanogens was probably a result of pterin optimization within this derived subgroup of Euryarcheota.
Genes for the -methanofuran system that interconverts oxidation states of one-carbon units have been found in both archaeal and bacterial clades, and this observation has been the basis for some hypotheses that they were present in the LUCA [52]–[54]. Several lines of evidence, however, argue against this conclusion. The most direct counterevidence comes from the synthesis and structural variation within the folate family, to which both THF and belong. Both structures are derived from GTP, but their synthetic pathways diverge as shown in the left panel of Fig. 4 [55]. Whereas in the synthesis of THF the final steps are simply the addition of an aminobenzoate and one or more glutamates, followed by the reduction of a double bond, in the case of the aminobenzoate is first combined with a phosphoribosyl pyrophosphate (PRPP) before being incorporated at the homologous step [56]. This is then followed by the addition of a second PRPP, an -ketoglutarate [56], two methyl groups derived from SAM [57] and finally the reduction of the same double bond, leading to the same chirality [58], as in THF. The synthesis of is thus an elaborated version of the synthesis of THF, which has led and its structural variants (shown in the right panel of Fig. 4) [59]–[64] to be termed ‘modified folates’ [65].
The variation within this group is based around two central structural features that likely affect chemistry. Combining aminobenzoate with PRPP eliminates the electron-withdrawing carbonyl group of aminobenzoate (indicated in red in Fig. 4), raising the electron density at in the final pterin structure and lowering the free energies of the bound units in the direct reduction of [24]. The addition of structural methyl groups (indicated in purple in Fig. 4) results in steric hindrance and a more rigid pterin structure for than for THF, in turn lowering the conformational entropies of pterins containing single-bound units [24]. The order of these modification in the synthesis of and the nature of the observed variation within the family of modified folates – elimination of the carbonyl group is universal outside of THF; the number of methyl group varies from zero to two in the same order in which they are added in – suggests a step-wise exploration of folate modification. Further variation outside of these two aspects mostly occurs within a molecular appendage which is far removed from the active site and is thus unlikely to affect local chemistry.
Within these general outlines, a wide range of structural variants is observed within this class of molecules (see Fig. 4). Of these, only THF and are observed within the bacteria (the latter only in a few clades), while all are observed within the archaea. The general premise that diversity remains highest in the domain of origin, and the fact that is an apparent end-point in the step-wise modification within this class of molecules suggests already that the exploration in folate modifications occurred within the archaea, and that genes for were subsequently laterally transfered to bacteria. This hypothesis is further strengthened if we consider the ecological roles of the metabolic machinery, which are very different in the two domains. Methanogenic archaea use this machinery in both the oxidative and reductive direction, with autotrophs among this group using it exclusively in the fully reductive direction starting from , which as we previously showed was most likely the ancestral function of folate chemistry. In bacteria, by contrast, the methanopterin system is used exclusively in the oxidative direction, either as part of a methylotrophic metabolism or possibly to serve in formaldehyde detoxification. Even methylotrophic bacteria that have been classified as autotrophs first oxidize reduced forms of carbon before feeding it into traditionally recognized -fixing pathways such as the Calvin-Benson Cycle [66]. To summarize: not only do we find a much greater structural diversity within the modified folates (including a range of apparent structural intermediates between THF and ) present in the archaeal domain, we also find the likely ancestral function of this whole class of molecules preserved in exclusively within this domain. Variations in the synthesis, structures and ecology of the folate family are all thus explained consistently by an emergence within the archaea and subsequent transfer to the bacteria, as they would not be by the reverse transfer from bacteria to archaea, or by the presence of in the LUCA.
Finally we briefly discuss phylogenetic studies of the distribution of genes, some of which have been used to argue against the transfer of genes from archaea to bacteria. While several phylogenetic studies [53], [67] led to the assumption that the most probable scenario was a transfer from archaea to bacteria [53], [67]–[69], one study reached a different conclusion [54]. In this work, unrooted trees were built for genes found within the Planctomycetes, proteobacteria, and methanogenic archaea. It was then argued that because all three clades separate in most such trees, transfer from methanogens to bacterial clades (including sequential transfers) could be eliminated, leading to the conclusion that these genes were most likely present in the LUCA [54]. We first note that this study does not address variations in the structure and synthesis of the modified folates or their ecological roles as just discussed. In addition, in any tree with just three clades, the topology automatically renders any two clades monophyletic relative to the third, which means that transfer from methanogens to the ancestor of proteobacteria and Planctomycetes cannot in fact be eliminated on these topologies alone. There are several trees in [54] that show non-methanogenic archaea branching between methanogens and bacterial clades, but these refer to genes for the biosynthesis of , which because modified folates are common across archaea are less useful in distinguishing these scenarios. Finally, transfer from archaea to bacteria explains the pattern of absence of genes in most bacterial clades, which in the case of a presence in the LUCA would require an unexplained mechanism for massive differential loss within the bacteria [68]. Thus we conclude from a wide range of evidence that incorporates phylogenetics and the synthesis, structural diversity, and ecology of the folate family, that emerged within the archaea, and that this particular modified folate (but not others) was subsequently transferred to bacteria, where in a few clades it was adopted in a later derived functional role.
If the folate family diversified within the archaea as we have proposed, through step-wise modification to the synthesis of THF, what then drove these modifications? As alluded to in the description of the individual key structural changes, we argue here that it was an energetic optimization of the fixation of through direct reduction. We calculated the biosynthetic cost of glycine and serine in units of ATP and reductant ( equivalents) for both the reductive and oxidative pathways, shown in Table 2. Since the cost of the oxidative pathway rises with general cost of fixing carbon for different primary fixation pathways, while the reductive pathway has a fixed cost, Table 2 only shows cost for the oxidative pathway as part of the two most energy efficient autotrophic strategies, the rTCA cycle and the reductive acetyl-CoA pathways [1], [70]. It can be seen that the only autotrophic context in which the oxidative pathway is more efficient than the reductive pathway is methanogenesis. The combined effect of higher electron density on of attained by fusing aminobenzoate with PRPP before incorporation, the lowering of the entropy of single-bound -folate structures through the addition of structural methyl groups – both of which may have on average conveyed an energetic advantage in isolation – and the usage of the methanofuran thus resulted in a reduction of the number of ATP's required for the uptake of through this pathway. The higher electron density at of that facilitates the uptake of without ATP hydrolysis increased the stability of the C- bond, but thereby sacrificed the capacity to donate the methylene group and synthesize glycine directly, explaining the absence of the glycine cycle in methanogens. (We have noted that, in contrast, most non-methanogens show all genes for this cycle.) In addition, the lowering of the entropies of single-bound - molecules may contribute to the smaller free energy differences between these and closed ring (methenyl- and methylene-) - structures, allowing easier transitions between oxidation states and thus facilitating reversal of pterin chemistry from the reductive to the oxidative direction [24]. This robust change in functionality would also explain why only among all the modified folates was transfered from archaea to bacteria. Finally, as we shall see below, energy optimization through elimination of redundant ATP-consuming reactions was likely a more general selective force that also explains the initial emergence of WL phenotypes.
We interpret the pathway distributions as showing, generally, that energy optimization is a secondary selection force to oxygen toxicity, when the latter is present. For example, once acetyl-CoA synthase was eliminated in the direct reduction of upon oxygen exposure in deep-branching autotrophs, modification within the folate family as just described would no longer have been advantageous as the oxidative synthesis of glycine and serine could no longer be connected to folate-mediated direct reduction of , eliminating this route to lowering ATP cost. Another example is the cyanobacteria, which use the glyoxylate pathways to synthesize glycine and serine even though energy accounting suggests that a hybrid strategy involving the reductive pathway would be more economical than using glyoxylate emerging out of the CBB cycle, which has one of the highest ATP costs of carbon-fixation [1], [70]. However, it is known that 2-phosphoglycolate, the precursor to glyoxylate in these organisms, is formed when replaces in the CBB Rubisco reaction, and subsequently inhibits the cycle [30]. In this case the adoption of the glyoxylate pathway thus furnishes a mechanism to remove -induced growth inhibition and to recycle 2-phosphoglycolate through anabolism. We predict that the reductive pathway retains a role in cyanobacteria living under anoxic [71] or high- [30] conditions, where the CBB cycle does not produce (significant) 2-phosphoglycolate, or in mutants where the glyoxylate pathways have been deactivated [30].
Having established that direct reduction of is the most likely ancestral metabolic pathway to glycine and serine, as well as the ancestral function of THF- chemistry in general, we next consider a full evolutionary reconstruction of carbon-fixation. We first note, as briefly mentioned in the introduction, that with these new results the deep-branching rTCA and Wood-Ljungdahl carbon-fixation phenotypes show a high degree of similarity. For deep-branching rTCA phenotypes we find parallel use of an incomplete WL pathway that lacks only the final synthesis of acetyl-CoA, which as noted previously is catalyzed by one of the most oxygen sensitive enzymes in the biosphere. Anaerobic WL phenotypes, in contrast, do possess this complete sequence to acetyl-CoA, and they then complete the set of universal anabolic precursors through a variety of incomplete rTCA cycles. Closer inspection of these incomplete cycles shows that while connection to the universal anabolic precursors is always maintained, in all cases the usage of one of the redundant ATP-dependent steps involved in thioester bond formation is eliminated: the synthesis of citryl-CoA from citrate (which is subsequently cleaved to acetyl-CoA and oxaloacetate), or the synthesis of succinyl-CoA from succinate [72], [73]. The incomplete WL pathway as part of the hybridized deep-branching rTCA cycle is thus associated with oxygen sensitivity, while the incomplete rTCA cycles as part of deep-branching WL are associated with ATP economy. The template that underlies both is a fully connected rTCA-WL network.
The modular role and ancestral status of direct reduction thus anchors the most fundamental division in carbon fixation strategies, between the autocatalytic rTCA loop and the non-autocatalytic reductive acetyl-CoA pathways, and suggests that the linked rTCA-WL network preceded both. From the linked rTCA-WL phenotype, oxygen toxicity to the acetyl-CoA synthase causes divergence of Aquifex-like rTCA phenotypes, while energy optimization through elimination of redundant ATP-dependent citryl-CoA or succinyl-CoA synthetases leads to the divergence of WL phenotypes. Other fixation pathways may be derived from these by loss or addition of only a few key reactions, linked again to ATP, oxygen sensitivity, or in some cases alkalinity or redox states. Within our basic assumption that the biosphere has always been autotrophic from , we may therefore ask: 1) whether all carbon-fixation phenotypes may be connected while maintaining uninterrupted access to all initial anabolic branching points from ; 2) whether these connections can be made with no repeated innovations; 3) which networks that are unobserved in extant biology must be posited to connect all networks that are observed. These questions are answered by arranging the known phenotypes, and the new hybrid forms revealed here, as nodes on a tree according to parsimony, as outlined in Fig. 1D, where links represent evolutionary innovations.
A maximum-parsimony tree connecting all known autotrophic pathways, in which all nodes represent viable carbon-fixation phenotypes, is shown in Fig. 5. We note first the position of the linked rTCA-WL network between rTCA and WL phenotypes. Fully connecting the deep tree of life while maintaining autotrophy requires connecting rTCA and WL phenotypes. No single change can connect them while maintaining uninterrupted access to all essential branching points from , and the only sequence of two changes that does maintain autotrophy passes through the linked rTCA-WL network. The ancestral state required to connect the network is thus consistent with the alternative selective filters from oxygen and ATP use discussed above. From this inserted node, we may then connect all other observed carbon-fixation phenotypes, with no further insertions needed, through sequences of single changes for which we can invoke plausible ecological causes.
If we assume that once the complex, ATP-dependent, citryl-CoA or succinyl-CoA synthetase enzymes were lost they could not easily be recovered, we can then explain the absence of the rTCA cycle in the archaea, together with the curious, chimeric Dicarboxylic/4-Hydroxybutyrate (DC-4HB) and 3-Hydroxypropionate/4-Hydroxybutyrate (3HP-4HB) cycles as the only autocatalytic loop pathways in this domain [1], [2]. Once autocatalytic rTCA cycling has been lost (as in the emergence of WL phenotypes), organisms can only survive subsequent loss of the acetyl-CoA synthase due to oxygen exposure if they either can survive as heterotrophs, or else possess a latent cycle that can be activated. Indeed, in the Euryarcheota all non-methanogens are heterotrophs lacking acetyl-CoA synthase. In the Crenarcheota, in turn, the DC-4HB cycle shares the first arc of the rTCA cycle (acetyl-CoAsuccinyl-CoA) but not the second (succinyl-CoAacetyl-CoA), and it has a significantly higher ATP cost per carbon fixed than does rTCA [1], [70]. Therefore, an ancestral status for DC-4HB cannot be motivated as requiring fewer chemical innovations, nor could this pathway have competed with rTCA energetically. If, however, a subgroup of WL organisms that possessed all the steps of the second 4HB arc for other purposes were exposed to oxygen, the activation of the cycle and the transition to this carbon-fixation phenotype would be enforced to maintain autotrophy from . The second arc of DC-4HB originates from acetate along the isoprene synthesis pathway essential to archaeal lipids, and from succinate along a fermentative pathway using the key 4-hydroxybutyryl-CoA dehydratase [74]–[76]. Both of these sequences have been found in the Clostridial clade of the Firmicutes [75], [76], which also contains many WL organisms, supporting this scenario. The most elaborate case is that of Clostridium kluyveri, which contains pathway segments to reach all DC-4HB intermediates, though these are not used as part of an autocatalytic cycle, instead supporting a fermentative heterotrophic metabolism [76].
In a similar fashion, the 3HP-4HB phenotype then branches from the DC-4HB phenotype through replacement of the first rTCA arc (acetyl-CoAsuccinyl-CoA), which was still preserved in DC-4HB. The central difference, from the perspective of carbon-fixation, between the rTCA arc and the 3HP arc that replaces it is that the latter requires only uptake of bicarbonate whereas the former takes in as well. The transition to 3HP-4HB thus appears to be driven by changes in the alkalinity in the environment, with an equilibrium shift from dissolved toward bicarbonate favoring the emergence of the 3HP pathway. The subsitution of the second rTCA arc by the 4HB arc within the crenarcheota had already removed all reactions involving in the second half of the pathway, thus imposing no further barriers to the transition from DC-4HB to 3HP-4HB in response to alkalinity. The archaeal case contrasts with the emergence of the 3HP pathway in bacteria, as part of the 3-HP bi-cycle, which similarly uses only bicarbonate uptake and has thus also been recognized as an adaptation to alkaline conditions. In the latter case, both (rTCA) arcs in the initial phenotype utilized uptake reaction, and adaptations to avoid these resulted in a more complex pathway structure. In addition to the appearance of the 3HP arc this involves the reversal of part of the first rTCA arc and the appearance of an alternate completion of the 3HP arc through combining with a glyoxylate that is one of the products of the reversed (first) rTCA arc. We note, however, that these complex pathways result from the relatively simple and uniform chemistry of aldol reactions, and that part of the 3HP-bicycle overlaps with the glyoxylate shunt, which is a similar bicycle.
The emergence of the 3HP pathway in both archaea and bacteria as just discussed is one of the two main parsimony violations in Fig. 5. The 3HP pathway uses two key biotin-dependent carboxylation reactions (to malonyl-CoA and methylmalonyl-CoA), which suggests a bacterial origin as this enzyme class features prominently in the biosynthesis of the fatty acids that make up bacterial membranes, in contrast to archaeal lipids, which are based on isoprenoid backbones [77]. However, a comparison of enzymes for thioesterification of propionate to propionyl-CoA has been interpreted as implying convergent evolution for these essential steps in the pathway [78]. Since the appearance of the 3HP pathway is recognized as an adaptation to alkaline conditions in both bacteria and archaea, either convergence or LGT is plausible due to restricted common environments. The other main parsimony violation in Fig. 5 is the parallel emergence of the oxidative route to serine. Oxidative synthesis of serine involves three of the most ubiquitous reaction types/enzymes within metabolism, the dehydrogenation of an alcohol to a carbonyl (reaction 11), a transamination of that carbonyl to an amine group (reaction 10), and a dephosphorylation (reaction 9). The recurrent emergence of this pathway through either duplication or promiscuity of common enzymes is therefore not likely to be a low-probability event, and thus not a significant violation of the assumptions behind a maximum-parsimony reconstruction.
The tree of Fig. 5 separates the Firmicutes together with all archaea as the branch from the linked rTCA-WL network originally driven by energy optimization in the absence of oxygen. All other bacterial lineages separate by an early loss of acetyl-CoA synthase. This basic division – with Firmicutes separate from all other bacterial clades – is supported by phylogenetic studies that focus on directed insertion-deletions in paralogous informational and operational genes [79], and on universal orthologous genes [41]. The fact that all archaea separate as a monophyletic branch from a pool of ancestral bacteria may be associated with the more tree-like archaeal phylogeny, compared to the more web-like phylogeny of bacteria, which suggests higher rates of LGT in the latter [21]. The co-evolution of the archaeal DNA-replication system together with less permeable isoprenoid membranes [77], [80], resulting in lower LGT in this domain, are possible mechanisms for isolation.
The carbon-fixation phenotypes at all internal nodes of our reconstructed phylometabolic tree are found preserved in extant organisms, except for the linked rTCA-WL network at the root node. While many of these derived nodes incorporate parallel fixation pathways, in all such cases the parallel pathways are disjoint and supply carbon to distinct portions of biomass. The linked rTCA-WL network is qualitatively different, having two separate input channels that supply acetyl-CoA, resulting in a kind of metabolic redundancy. We must ask, does our choice to insert an unobserved node, in order to connect the tree while preserving autotrophy, justify the reconstruction of an ancestor with a form of redundancy not found in any of its descendants, and if so, what does this reconstruction tell us about the evolution of earliest life? We address both questions by considering the relation of network topology and redundancy to self-amplification.
The capacity of life for exponential growth, resulting from proportional self-amplification by metabolic and other networks with “autocatalytic” topology [81], [82], is essential to self-repair and robustness in the face of perturbations. At the small-molecule substrate level the rTCA cycle is network-autocatalytic, and thus capable of exponential growth above a threshold rate of production of acetyl-CoA, but also fragile against collapse if production falls below this threshold [83]. The threshold fragility, which may have been a more serious problem in an era of primitive catalysts or regulation, is removed while preserving autocatalysis if acetyl-CoA is independently supplied by WL. Conversely, WL in modern organisms may be considered network autocatalytic at the level of whole-cell physiology including enzymes and cofactors, but it relies on the integrity of synthetic pathways for these molecules which are more complex even than the small-molecule substrate. rTCA, which provides an independent channel for carbon fixation and synthesis of precursors, would thus reciprocally support the robustness of WL in the earliest era of organic (versus proposed earlier mineral [69], [84]) cofactors.
The difference in network topology between our reconstructed root, and its derived descendants, would then reflect a shift in the character of natural selection acting on the earliest versus later cells. In early cells (or pre-cells) with imprecise or unreliable enzyme function, consequent leaky pathways and fluctuating cofactor concentrations, or unreliable regulation of anabolism, robustness inherent in the topology of the substrate network would have carried a selective advantage. (Note that anabolism is a form of “parasitic side-reaction” from rTCA cycling, and that an inadequately regulated anabolism can as readily carry carbon fixation below the threshold for self-maintenance, as external factors can). As each of these cellular-level mechanisms was refined, redundant self-amplification of small-molecule substrates would have become unnecessary, and conservation of ATP or adaptation to oxidizing environments would have become more advantageous.
These observations also motivate our reference to the linked rTCA-WL network as a “root” phenotype in a maximum-parsimony tree that, by its construction method, is otherwise unrooted. All other phenotypes in the tree may be explained as evolutionary diversifications away from the linked rTCA-WL network, which is both a template for these divergences, and because of its redundancy, a more plausible candidate for a primitive ancestral form than any modern phenotype. The good overlap of our carbon-fixation tree with the later branchings of bacteria and archaea indicates the fundamental role of autotrophy in shaping the deep evolution of the biosphere, and suggests that the later nodes describe not only cellular life, but emerging well-resolved clades. The less-clear separation of bacteria and archaea near the root in Fig. 5, the correspondence of these branches to the reticulated domain of gene phylogenies [12], [14], [19], the need and character of the inserted root node, and the flexible interpretation of our carbon-fixation phenotypes as species or consortia, leave open the possibility that the earliest branches were stages of chemical evolution that preceded modern life [3]–[6].
Our phylometabolic reconstruction, and the surprising tree that it yields, has focused on the particular function of carbon fixation. However, the same methods could be extended to a fuller description of core metabolism, and from the phenotypes we have already shown, we may anticipate certain specific complications that will be introduced with a wider reconstruction. These reflect the changing nature of ecological interactions with increasing oxygenation, and they give added insight into the interpretation of the high degree of parsimony we have shown for innovations in carbon-fixation, most of which took place in anaerobic or micro-aerobic conditions.
For any phylogenetic reconstruction, it is important to remember that the nodes and links on a tree are summary statistics for relatedness of samples taken from a population process that may have been very complex. A high degree of parsimony in a tree does not indicate the absence of complex structure in populations, constraints on innovation, or ecological interactions; at most it indicates a lack of specific evidence that innovation required anything more than rare variations and environmental selection in vertically transmitted phenotypes. The cases in which violations of parsimony are inescapable provide evidence that multiple levels of organization were causally essential to the course of innovation, whether these were latent constraints causing some innovations to recur (leading to evolutionary convergence), or ecological interactions leading to gene or pathway transfer. The structure of parsimony violation then indicates what forms of multilevel interaction must be deduced to explain evolutionary causation.
Two illustrative cases of parsimony violation that we have elaborated are the transfer (or combined transfer/convergence) of the 3HP pathway, and the gradual elaboration of pterin cofactors followed (as we argue) by the late transfer specifically of from archaea to bacteria. The parallel innovation of the 3HP pathway in Chloroflexi and in the Chrenarchaeota entails duplication of an entire (and rather elaborate) pathway segment, and not merely a single key gene. It is favored in specialized environments which we would expect to create long-term association between inhabiting species, these environments contain a stressor (alkalinity) which we expect to induce gene transfer, and relative to the very ancient divergences in our tree, the innovations of 3HP occur late, at an era when we expect organism lineages to have evolved refractoriness to many forms of gene transfer [85]–[87], along with more integrated control of chromosomes. Similar long-term associations in anaerobic environments such as coastal muds are believed to have led to the (otherwise uncommon) aggregate transfer of a large complement of operational genes from -proteobacteria to Aquificales [88].
The transfer of reflects an even more fundamental link between oxygen and ecosystem structure. As it occurs in methanotrophs and methylotrophs, is used for oxidation of methane and other reduced species, the most extensive form of heterotrophy of reduced carbon, which relies on environmental oxidants to link methane producers and consumers. This function, driven by trophic interactions, is layered over the foundation of reduction on folates which we reconstruct as ancestral in the clades harboring methanotrophs and methylotrophs. The complexity of methylotrophy [89] anticipates the enormous diversification of catabolic pathways that becomes available with oxidation of reduced biotic carbon [17], but which depends on details of ecological provision and accessibility of carbon sources.
Finally, it is interesting that many of these complexities would fall cladistically near the tips of the tree in Fig. 5, suggesting that innovation in carbon fixation ceased, to be replaced by innovations in carbon exchange through ecosystems, on a horizon coinciding with the rise of oxygen. This pattern brings into sharper relief the striking lack of multilevel dynamics that would distinguish organism and ecosystem roles in the reconstruction we have shown.
In this paper we have demonstrated a novel method that integrates constraints from FBA and from phylogenetics. Individually, metabolic and phylogenetic reconstructions are both subject to ambiguities, especially for deep-branching lineages with significant gene divergence from well-characterized model organisms, and extensive LGT near the root [12], [14], [19]. Integrating the two can resolve ambiguities inherent in each, as well as providing new ways to tailor questions to specific features of early evolution.
Here, as a proof of principle, we have presented a coarse-grained reconstruction of the input channels of carbon into the biosphere only up to the initial anabolic branching points. We can relate all modern forms of carbon-fixation to a single ancestral form, and we find that innovations in carbon-fixation were the foundation for most major early divergences in the tree of life. We have also proposed specific causes for the major divergences, and argued for a very small role for lateral gene transfer or convergent evolution. The absence of any phylogenetic signature of important ecological co-evolution for this function, combined with selective forces that originate in energetics or inorganic chemistry, offers additional specific links between genome evolution and the geological record [15]. The consistency of the reconstruction demonstrates, with specific examples, how the fine structure of organic chemistry and geochemistry can enter as detailed constraints on long-range evolutionary dynamics. The hypotheses required by the reconstruction further imply several sets of specific experimental predictions, which are outlined in the Text S1.
By limiting our reconstruction to the networks of core carbon fixation – and by virtue of the modular interface of this function with later stages of biosynthesis – we have selected a problem for which many distinctions between organism and ecosystem (any that do not leave phylogenetic signatures) may be passed over, which falls prior to most ambiguities from reticulated gene phylogenies, and which leads to a high-parsimony tree that can be identified manually. The extension of these methods to larger networks with more ambiguity, and to historical reconstructions for which trees provide less adequate representations of complex population processes, will require formal probability models, which may be implemented in joint-maximum-likelihood or Bayesian MCMC phylometabolic algorithms. The current work suggests ways to increase the number of dimensions of “meaning” that can be used to define such probabilistic models, by placing genes in both physiological and ecological context. In this way we may reconstruct trees of life that reflect more of the multi-level character of evolution and development than is suggested by gene counts, and that capture constraints which may have acted continuously since its emergence.
For a description of the basic principles of phylometabolic analysis, see the introduction and Fig. 1. This method rests in part on metabolic flux-balance analysis (FBA), which has been described in detail elsewhere [90], [91]. Briefly, the core of FBA consists of the following three equations:(1)(2)(3)where is the concentration of metabolite n, is the stoichiometry of metabolite n in reaction m, and is the flux of that reaction. The nm matrix S and the m-dimensional vector are the the stoichiometries and fluxes of the total metabolic network. Under steady state growth, the principle of mass balancing can then be expressed as equation (2). Finally, Z is an objective function that is selected for optimization, and consists of a linear combination of individual individual fluxes weighted by proportionality constants . Z is often chosen to be the biomass composition of an organism, and maximizing its output thus maximizes growth. The full metabolic network of an organism is reconstructed from its annotated genome. While optimization and matching to laboratory growth data of specific organisms requires detailed analysis of the constraints of individual reactions and careful computational modeling, the initial reconstruction of the network requires only that the network represents a viable metabolism capable of supporting growth (Z0).
Our analysis is restricted to the access (from ) of the universal anabolic precursors that represent the initial branching points in anabolism, requiring as in the initial reconstruction of an organism only that they are produced (Z0). These anabolic branching points are acetyl-CoA, pyruvate, oxaloacetate, succinyl-CoA and -ketoglutarate, which is a very small number of intermediates to consider, and because studies of carbon-fixation phenotypes have already shown that they are in all cases reached through reuse of partial TCA sequences in fact no computational analysis is required. As explained in the main text, the only intermediates accessible through pathways that circumvent all the universal anabolic precursors are glycine and serine, so our analysis of metabolic genes focusses on the pathways to these intermediates. Three main pathways to glycine and serine are known (see Fig. 2 and the Results and Discussions section), and we analyze the gene profiles for a large number of species for the presence of the necessary enzymes for each of these three pathways.
The gene profiles for all strains used in this study were obtained from the Uniprot database [35]. For all complete annotated genomes within each clade, gene profiles in the three pathways were obtained by searching for enzyme classes (through EC numbers) and names. The searches were done in a redundant manner to ensure that all naming variations in the annotation of a given gene/enzyme were included in the final result. While in general we used the annotations as given in the database, in a few cases a (Uniprot built-in) BLASTp search was done to confirm the absence or presence of an enzyme if the profile of a particular strain seemed to contradict the pattern of the clade overall. These are highlighted in Table 1 and Table S1 in Text S1. If within a particular strain all enzymes of a given pathway showed up in our database search the pathway (either active or latent) was counted as present in that strain.
The EC numbers and one common variation of the full names of the numbered reactions as shown in the various Figures, and throughout the text are as follows: 1 = formate dehydrogenase (EC: 1.2.1.2); 2 = -formyl-THF synthase (EC: 6.3.4.3); 2A = -formyl-THF cycloligase (EC: 6.3.3.2); 3 = methenyl-THF cyclohydrolase (EC: 3.5.4.9); 4 = methylene-THF dehydrogenase (EC: 1.5.1.5); 5 = dihydrolipoamide dehydrogenase (EC: 1.8.1.4); 6 = aminomethyltransferase (EC: 2.1.2.10); 7 = glycine dehydrogenase (EC:1.4.4.2); 8 = serine hydroxymethyltransferase (EC: 2.1.2.1); 9 = phosposerine phosphotase (EC: 3.1.3.3); 10 = phosphoserine aminotransferase (EC: 2.6.1.52); 11 = 3-phospho-glycerate dehydrogenase (EC: 1.1.1.95); 12 = alanine-glyoxylate transaminase (EC: 2.6.1.44).
|
10.1371/journal.pntd.0005777 | Pathological manifestations in lymphatic filariasis correlate with lack of inhibitory properties of IgG4 antibodies on IgE-activated granulocytes | Helminth parasites are known to be efficient modulators of their host’s immune system. To guarantee their own survival, they induce alongside the classical Th2 a strong regulatory response with high levels of anti-inflammatory cytokines and elevated plasma levels of IgG4. This particular antibody was shown in different models to exhibit immunosuppressive properties. How IgG4 affects the etiopathology of lymphatic filariasis (LF) is however not well characterized. Here we investigate the impact of plasma and affinity-purified IgG/IgG4 fractions from endemic normals (EN) and LF infected pathology patients (CP), asymptomatic microfilaraemic (Mf+) and amicrofilaraemic (Mf-) individuals on IgE/IL3 activated granulocytes. The activation and degranulation states were investigated by monitoring the expression of CD63/HLADR and the release of granule contents (neutrophil elastase (NE), eosinophil cationic protein (ECP) and histamine) respectively by flow cytometry and ELISA. We could show that the activation of granulocytes was inhibited in the presence of plasma from EN and Mf+ individuals whereas those of Mf- and CP presented no effect. This inhibitory capacity was impaired upon depletion of IgG in Mf+ individuals but persisted in IgG-depleted plasma from EN, where it strongly correlated with the expression of IgA. In addition, IgA-depleted fractions failed to suppress granulocyte activation. Strikingly, affinity-purified IgG4 antibodies from EN, Mf+ and Mf- individuals bound granulocytes and inhibited activation and the release of ECP, NE and histamine. In contrast, IgG4 from CP could not bind granulocytes and presented no suppressive capacity. Reduction of both the affinity to, and the suppressive properties of anti-inflammatory IgG4 on granulocytes was reached only when FcγRI and II were blocked simultaneously. These data indicate that IgG4 antibodies from Mf+, Mf- and EN, in contrast to those of CP, natively exhibit FcγRI/II-dependent suppressive properties on granulocytes. Our findings suggest that quantitative and qualitative alterations in IgG4 molecules are associated with the different clinical phenotypes in LF endemic regions.
| Lymphatic Filariasis, also known as elephantiasis, infects an estimated 39 million people in 73 tropical and sub-tropical countries. The most severe clinical manifestations of the disease include swelling of the scrotal area and lower limbs (hydrocele and lymphedema). It is well admitted that host immune reactivity plays a critical role in the pathogenesis of the disease. Previous investigations have linked the non-cytolytic antibody IgG4 to the hyporesponsive states in filarial infections. However, few data exist on how this antibody is involved in the pathogenesis of human filariasis. Here we investigated the role of this antibody in disease pathogenesis by comparing the effect of plasma, IgG and IgG4 fractions from the four clinical categories of individuals; chronic pathology individuals (CP), asymptomatic microfilaria positive (Mf+) and negative (Mf-) and uninfected endemic normal individuals (EN) on activated granulocytes. We could show that granulocyte activation was significantly inhibited in the presence of plasma from EN and Mf+ and that, affinity-purified IgG4 antibodies from EN, Mf+ and Mf- individuals inhibited granulocyte activation in a dose-dependent manner via the immune receptors FcγRI and FcγRII. Our data also reveal significant functional differences between IgG4 molecules from EN, Mf+, Mf- and CP.
| Lymphatic filariasis (LF) also known as elephantiasis is a potentially disabling and disfiguring disease caused in human by vector-borne nematodes Wuchereria bancrofti, Brugia malayi and Brugia timori [1]. The disease has significant social and economic consequences for affected individuals as well as for their families and communities [2]. The current strategy to control the infection is based on mass administration (MDA) of diethylcarbamazine or ivermectin combined to albendazole [2]. After 13 years of the MDA programme, recent estimations in 2015 indicate that 38.47 million LF cases remain [3]. In endemic regions, exposure to the infection leads to different clinical phenotypes. The first group includes putatively immune individuals or endemic normals (EN) who remain infection and disease-free despite continuous exposition to mosquito-transmitted infective larvae (L3) [4]. The second group is defined by a hyper-reactive phenotype and is characterized by chronic lymphatic pathologies (CP) such as lymphedema or elephantiasis. The patients elicit a strong T helper (Th)1 and Th17 immune response that eliminate the microfilarial stage [5] while inducing the production of angiogenic factors like VEGF known to be associated with the development of filarial lymphedema [6]. This severe clinical profile is characterized by high antigen-specific immunoglobulin (Ig)E and low IgG4 [5] [7,8]. The third group includes asymptomatic individuals with latent infection who are free of microfilaria (Mf-) but are positive for circulating filarial antigens (CFA) [9]. The last clinical phenotype includes the majority of infected individuals and is associated with an hyporesponsive immune profile. In this group, individuals present few visible clinical manifestations despite large numbers of circulating microfilariae (Mf+) [10–12]. Subjects from this group commonly present a modified Th2 immune profile with a strong parasite-specific immunoregulatory arm allowing both the presence of adult worms and microfilariae. This modified Th2 response is associated with increased numbers of regulatory cells (Tregs) and alternatively activated macrophages as well as with the secretion of anti-inflammatory cytokines such as IL-10 and TGF-β. This predominantly immunosuppressed environment is associated with elevated levels of antigen specific IgG4 and is directly linked with parasite survival [4,10,13].
These clinical phenotypes strongly correlate with specific antibody isotypes produced in response to the infection. IgG4 antibodies, for example, correlate with the hyporesponsive state observed in Mf+ individuals whereas IgE, IgG1 and IgG3 correlate with CP [7,10,14–16]. IgG4 is structurally and functionally different from its co-class members [17–19]. While IgG1, IgG2 and IgG3 can fix and activate complement, IgG4 has no affinity for the complement and cannot induce antibody-dependent cell mediated cytotoxicity (ADCC). In addition, IgG4 was shown to inhibit antibody dependent complement activation [20] and to compete with IgE for fixation sites on mast cells and eosinophils [21,22]. In contrast, IgE induces mast cell and basophil degranulation [23–25]. These anti-inflammatory properties of IgG4 antibodies were associated with its unique ability to undergo Fab-arm exchange (FAE); resulting in the creation of bispecific, functionally monovalent antibodies [26,27]. However, the role played by filaria-induced antibodies in disease manifestations in LF is still not well understood. No data currently exist on how IgG4 antibodies participate in the modulation of the pathophysiology of filarial infections.
Granulocytes (eosinophils, neutrophils, and basophils) are key effector cells at the frontline against infections with filarial worms [28,29]. During helminth infection, granulocytes are rapidly activated and recruited to sites of infection where they are key producers of Th2 cytokines such as IL-4 and IL-13 [28,30,31]. They also produce “alarmins” which are constitutively available endogenous molecules that are released upon activation and act as chemo-attractants while providing maturation signals to antigen-presenting cells such as dendritic cells (DCs) and macrophages [32–35]. Granulocytes can also attack helminth infections through antibody-dependent cell mediated cytotoxicity (ADCC), which implies the killing of antibody-coated parasites via the release of cytotoxic granules (degranulation). The degranulation is triggered by Fc-receptors (FcRs) recognizing antibody-bound antigen complexes on the cell surface and several cytokines mainly IL-3 and IL-5 [36,37]. Human granulocytes express FcγRI, FcγRIIa/b, FcγRIII, FcɛRI/II and FcαR and can be activated by IgGs, IgE and IgA. The activation of granulocytes can be measured in vitro by monitoring the expression of several activation markers including mainly CD63, a member of the tetraspan membrane glycoprotein family [38,39]. CD63 is an activation marker specific for neutrophils and basophils and, among other markers, for eosinophils [38,40,41] and it is responsible for the retention and sorting of pro-neutrophil elastase in the primary granules of neutrophils [38,40]. Upon degranulation, complete granule contents are released by fusion with the cellular membrane and cytolysis [42]. Granulocytes are characterized by six major granule proteins: major basic protein (MBP), eosinophil peroxidase (EPO), eosinophil cationic protein (ECP), eosinophil-derived neurotoxin (EDN), neutrophil elastase (NE) and histamine. MBP, EPO and ECP are potent helminth toxins [43,44]. These granule proteins have been shown to be involved in the killing of microfilariae of Brugia spp. and are associated in both allergy and helminth models with the development of immunopathology [45–49].
Since granulocytes play a central role in the elimination of large parasites, we hypothesized that antibodies present in plasma of EN, Mf+, Mf- and CP might differently impact granulocyte activation and functions. After comparing the suppressive properties of plasma and purified IgG antibodies from EN, Mf+, Mf- and CP, we found that granulocyte activation was significantly inhibited by plasma from EN and Mf+ individuals whereas plasma of Mf- and CP have no effect. This suppression was dependent on IgG4 in plasma of Mf+ whereas IgG-independent factors seem to be involved in EN. We have also demonstrated that IgG4 actively suppressed granulocyte activation and release of granule contents via FcγRI and FcγRII.
Patients and endemic controls’ samples were collected between 2008 and 2010 in villages of the Ahanta West and Nzema East Districts in the western region of Ghana endemic for LF (S1 Table). No other human filarial species were endemic in the region. The donors were recruited as part of a diagnostic and a clinical trial in LF (Clinical Trials Registration: ISRCTN15216778 and ISRCTN14757) [9,50,51]. Written informed consent was obtained from all participants. Persons eligible for participation were male adults in good health, 18–60 years of age, with a minimum body weight of more than 40 kg and without any clinical condition requiring chronic medication. Exclusion criteria included abnormal hepatic and renal enzyme levels (γ-glutamyltransferase > 28 U/L, glutamyl pyruvic transaminase > 30 U/L, creatinine > 1.2 mg/100 mL) assessed by dipstick chemistry, alcohol, drug abuse or antifilarial therapy in the past 10 months. Study participants were examined by a clinician using physical methods and a portable ultrasound machine (180 Plus; SonoSite, Bothell, WA) as described previously [51]. In addition, the presence of infections with other helminths (Ascaris lumbricoides, Trichuris trichiura, Schistosoma spp.) and protozoa (Plasmodium) was investigated using respectively Kato-Katz and finger prick tests. All samples included in the present study were free of such infections as previously described [9]. Ethical clearance was given by the Committee on Human Research Publication and Ethics at the University of Science and Technology in Kumasi, and the Ethics Committee at the University Hospital Bonn. Microfilarial load was determined by microscopic examination of fingerprick night blood samples as published [51]. Subsequently, 10 mL of venous blood was collected from each eligible volunteer and plasma was taken, aliquoted, and frozen at −80°C until used.
Samples included EN, residing in the endemic region but free of infection (CFA-, Mf-, n = 14), clinically asymptomatic microfilaraemic (CFA+, Mf+, n = 14) and amicrofilaraemic (CFA+, Mf-, n = 14) subjects, positive for circulating filarial antigen and a group of chronic pathological individuals with lymphedema and/or elephantiasis termed “CP” (n = 14), negative for filarial antigen. Also, plasma from European non-endemic blood donors (NEC, n = 14) was used as controls. Serum samples were obtained in an anonymized and de-identified form. All samples and controls used in the present study were randomly picked from the initial batches using a computer-based simple random algorithm as previously described [52]. Each sample in the initial batch was assigned a unique number and the samples corresponding to computer generated list were picked in each group and used in the study.
Brugia malayi worms recovered from the peritoneal cavity of jirds (Meriones unguiculatus) were obtained from NAID Filariasis Research Reagent Resource, FR3 (University of Georgia, Athens, GA). To prepare B. malayi antigen extract (BmAg), 100–300 frozen adult worms were thawed and transferred to a Petri dish pre-filled with sterile PBS (PAA, Pasching, Austria). Following several washes in PBS, worms were placed inside a glass mortar (VWR, Langenfeld, Germany). 3–5 ml of medium (RPMI without supplements) were added and worms were crushed until the solution was homogenous. The extract was then centrifuged for 10 minutes at 300 x g at 4°C to remove insoluble material. The supernatant was carefully transferred to a new tube. Protein concentration was measured using Bradford Assay. Aliquots were stored at -80°C until use. The extract was titrated to determine the optimal concentration for cell stimulation and the level of endotoxin was defined using the Pierce Limulus amoebocyte lysate (LAL) Chromogenic quantification kit (Thermo Fisher Scientific, Schwerte, Germany). The endotoxin level was below the detection limit of 0.1 EU/ml.
Granulocytes used in this study were purified from buffy coats of healthy European donors provided by the Institute for Experimental Haematology and Transfusion Medicine, University Clinic Bonn, Germany. Ethical clearance was given by the Ethics Committee of the University of Bonn (“Ethikkommission der Medizinischen Fakultät der Rheinischen Friedrich-Wilhelms-Universität Bonn”). Granulocytes were isolated using Ficoll-Hypaque (Pancoll, PAN Biotech, Aidenbach Germany) method. The density gradient was performed according to the manufacturer's instructions. Briefly, 15 mL heparinized venous blood samples were diluted with an equal volume of cold phosphate-buffered saline (PBS) in a 50 mL conical centrifuge tube, layered over Ficoll and centrifuged at 900 x g for 30 min at 4°C in a swinging bucket centrifuge (Thermo Scientific, Germany) with brake off. The opaque layer below the Ficoll/plasma interface containing granulocytes was transferred to another tube. After that, red cells were lysed by 10 min incubation at room temperature in 1x red blood cell lysis solution (Miltenyi Biotech, Bergisch Gladbach, Germany). Granulocytes were then centrifuged at 200 x g for 8 min at 4°C to remove contaminating red blood cells. Cell pellets were washed twice at 200 x g for 8 min in RPMI 1640 (Life Technologies, NY, USA). Supernatants were discarded and the purity of isolated granulocytes was assessed by flow cytometry. The purity was routinely ≥ 96%.
Total IgG was isolated from the plasma of EN, Mf+, Mf- and CP using prepacked HiTrap Protein G columns (GE Healthcare, Freiburg, Germany) according to the manufacturer’s instructions. Briefly, 100 μl of plasma samples were diluted with 1400 μl PBS and passed through a pre-equilibrated protein G-Sepharose column (GE Healthcare, Freiburg, Germany). Since Protein G binds to all human IgG subclasses, non-IgG plasma components were washed out from the column. Bound IgG was eluted in 1 ml fractions using IgG Elution Buffer (0.2 M Glycine/HCl, pH 3.0) and neutralized with saturated Tris-HCl (pH 9.0). The antibody concentration was then assessed at 280 nm using a NanoDrop 1000 spectrophotometer (Thermo Fischer Scientific, Wilmington, USA).
IgG4 antibodies were purified from IgG-enriched fractions using the CaptureSelect Human IgG4 affinity matrix (Life Technologies, Paisley, UK) according to the manufacturer’s instructions. Briefly, CaptureSelect affinity matrix was gently loaded and equilibrated in 10 ml affinity chromatography column with 1x PBS (pH 7.3). Diluted IgG-enriched fractions (1:1 volume PBS) were loaded onto the column and the linear flow rate was set at 15 cm/hour. After washing with 1x PBS, the column was eluted with 0.1 M Glycine (pH 3.0) and the fractions were immediately neutralized with Tris-HCl (pH 9.0). IgG4 fractions were collected and the purity of fractions assessed by determining the level of IgG subclasses, IgA, IgE and IgM antibodies by Luminex assay. In addition, the total protein concentration in plasma, IgG and IgG4 fractions was determined using a Bradford Protein Assay Kit (Thermo Fischer Scientific, Wilmington, USA).
IgA purification was done using immobilized Peptide M/Agarose (InvivoGen, San Diego, USA) according to the manufacturer’s instructions. 1 ml of peptide M/Agarose gel was loaded into an appropriate microcentrifuge spin column (Thermo Scientific, Rockford, USA). 1 ml of IgG-depleted plasma from EN were then added to the column and incubated at room temperature for 30 min. After incubation, the column was washed and the flow-through was collected and labeled as IgA negative fractions (IgA-). The bound antibodies were then eluted with IgG elution buffer as described above. The resulting IgA positive (IgA+) fractions were neutralized immediately with neutralization buffer (1 M Tris–HCl, pH 9.0) and store at 4°C until use. Purity was assessed by Luminex analysis as described above. The purity was routinely > 90%
To determine the protein concentration of plasma and purified IgA, IgG and IgG4 fractions, a Bradford protein assay kit (Thermo Scientific, Rockford, USA) was used according to the manufacturer’s instructions. In brief, serial dilutions of bovine serum albumin (BSA) was performed and used as standards against the samples. Another serial dilution of the samples was done in PBS. 300 μl per well of Coomassie blue G-250 (Cytoscelecton, Denver, USA) reagent was distributed in duplicate in a 96 well plate (Greiner Bio-One, Frickenhausen, Germany) and 3 μl of diluted samples and standard were added. The protein concentration in plasma, IgA, IgG and IgG4 fractions was then measured at 595 nm using a SpectraMAX 190 microplate reader (Molecular Devices, California, USA).
To analyze the isotype composition in IgA, IgG/IgG4 positive and negative fractions and in the plasma of EN and LF patients, ProcartaPlex Human Antibody Isotyping Panels (eBioscience, Vienna, Austria) were used according to manufacturer’s instructions. Briefly, antibody coated magnetic bead mixtures were incubated with 25 μl of assay buffer, kit standards or diluted plasma samples in a ProcartaPlex 96-wells plates at room temperature for 1 hour. 25 μl of detection antibodies mixture was then added and the plates were incubated on an orbital shaker at 500 rpm for 30 min. After that, each well was incubated with 50 μl of diluted Streptavidin-Phycoerythrin for 30 min. Plates were then washed using a hand-held magnetic plate washer. All incubations were performed at room temperature in the dark. Afterwards, samples were suspended in 120 μl reading buffer. Data were acquired using a MAGPIX Luminex system (Luminex Cooperation) and analyzed with ProcartaPlex Analyst software 1.0.
The purity of eluted IgG fractions was analyzed by western blot. Samples were treated with 50 mM 2-mercaptoethanol for 5 min, and equal quantities (2.5 μg) of the purified proteins and controls were loaded onto separate lanes of a polyacrylamide gel (10–12%) and resolved by SDS-PAGE (100 v, 45–60 min). The resolved proteins were transferred onto nitrocellulose membranes (GE Healthcare, Freiburg, Germany) using a Bio-Rad Trans-Blot Turbo Transfer system (Bio-Rad, Germany). The membranes were then blocked with gelatin blocking buffer (3% gelatin in Tris Buffered Saline (TBS)) (Bio-Rad, Germany) for 1 hour prior incubation with the primary antibody (polyclonal mouse anti-human IgG (H+L)) (Thermo Scientific, Rockford, USA) for 1.5 hours at room temperature. The nitrocellulose membranes were then washed with TBS/0.05% Tween 20 before incubation for 1 hour with alkaline phosphatase-conjugated goat anti-mouse IgG (Bio-Rad Laboratories, USA). Immune complexes were finally detected with NBT (nitro blue tetrazolium) and BCIP (5-bromo-4-chloro-3-indolyl-phosphate, Bio-Rad Laboratories, USA). Experiments were repeated at least three times.
After establishing the optimal concentrations of IgE, anti-IgE and rIL-3 to induce granulocyte activation and degranulation (S1 Fig), the cells were purified as described above and 2 x 105 cells/well were plated and pre-incubated with 40 ng/ml of natural human IgE antibody (Abcam, Cambridge, UK) for 30 min at 37°C/5% CO2 as previously described [53,54]. The cells were first preactivated at 37°C/5% CO2 for 10 min in the presence of 2 ng/ml rIL-3 (Miltenyi Biotech, Bergisch Gladbach, Germany), before being stimulated with 25 ng/ml anti-IgE mAb (Clone BE5) (Abnova, Taipei, Taiwan) and 10 μg/ml Brugia malayi Ag. Thereafter, the granulocytes were further incubated for 18 hours at 37°C/5% CO2 either alone (with culture medium), or in the presence of appropriately diluted plasma samples (5% v:v, containing 5 μg/ml of total proteins), IgG and corresponding IgG depleted fractions (5 μg/ml of total proteins) or 2.5 μg/ml IgG4 antibodies purified from the IgG positive fractions from the different clinical groups.
Granulocyte culture supernatants were collected after 30 min and 18 hours to assess the level of histamine, and after 18 hours to investigate the release of ECP and NE using ELISA-Kits respectively from Abnova (Taipei, Taiwan), Abbexa (Cambridge, UK) and eBioscience (Vienna, Austria) according to the manufacturer’s recommendations. For histamine detection, samples and standards were first acylated by reacting 50 μl of samples, 25 μl of standards or control with 25 μl of acylation reagent and 25 μl of acylation buffer supplied in the test kit for 45 minutes. 25 μl aliquots of acylated standards, controls and samples were pipetted into wells of the antibody-coated microplate provided with the kit. Then the wells received 100 μl of histamine antiserum and the mixture was allowed to incubate for 3 hours at room temperature. The plates were then washed with the provided washing buffer to remove unbound materials. After that, the bound antibodies were detected using 100 μl of anti-rabbit IgG-peroxidase conjugate using TMB as a substrate. The color was allowed to develop for 20 minutes at room temperature in the dark. The reaction was stopped and the resulting OD values were measured at 450 nm. The histamine concentration, inversely proportional to the OD, was calculated using the SoftMax Pro Data Acquisition and Analysis Software.
For the assessment of ECP and NE, pre-coated ELISA plates were incubated with supernatants and standards for 1 hour at room temperature. The plates were then washed and incubated for 1 hour with HRP-conjugated anti-ECP and anti-NE polyclonal antibodies. After a final wash, the plates were developed using the provided TMB substrate and analyzed at 450 nm.
To assess granulocyte activation and degranulation, the cells were harvested and washed with FACS buffer (PBS/2% FCS) at 1300 rpm for 8 min. 1x105 cells were then resuspended in 100 μl of FACS buffer and blocked with 1 μl of FC- block (Affymetrix eBioscience, San Diego, CA, USA) for 15 min. 5μg/1x105 cells of either anti-human CD66b-FITC (clone: G10F5) or CD63-PE (clone: H5C6) and HLADR-FITC (clone: LN3) (all from Affymetrix eBioscience) were then added and the cell suspension was incubated for 30 min at 4°C. Cells were then washed two times with FACS buffer and fixed in 200 μl PFA (4%). To correct spectral overlaps, fluorescence compensation was done using UltraComp ebeads (Affymetrix eBioscience). Data were acquired and analyzed using a FACS Canto flow cytometer and the BD-FACS-DIVA analysis software (BD Biosciences). Before the analysis, granulocyte viability was assessed by FACS using propidium iodide (PI-PE) and annexin-V (Annexin-FITC) (all from BD Biosciences, Heidelberg, Germany). All samples presented less than 1 and 10% necrotic cells respectively before and after 18 hours incubation.
Granulocytes were cultured as previously described and 2 x 105 cells were harvested and washed with PBS. Then 100 μl of diluted cells were aliquoted into cytospin funnels and spun at 500 x g for 5 min onto glass slides (Engelbrecht, Edermünde, Germany) in a Hettich Cytospin centrifuge (Hettich, Tuttlingen, Germany) and immediately fixed in 4% PFA for 15 min. The slides were then blocked with PBS/1% BSA for 30 min and incubated with the primary antibody (mouse anti-human monoclonal IgG4) (Thermo Fischer Scientific, Rockford, USA) for 1 hour. After washing 3 times, the slides were incubated with the Alexa Fluor 488 coupled secondary antibody (goat anti-mouse polyclonal IgG antibody) (Thermo Fischer Scientific, Rockford, USA) for 1 hour at room temperature in a humidifying chamber. For the investigation of the Fc-receptors associated with IgG4-mediated granulocyte suppression, blocking antibodies against human FcγRI (2 μg/ml) (Biolegend, San Diego, CA, USA, clone:10.1), FcγRII (1 μg/ml) (Biolegend, San Diego, CA, USA, clone: FUN-2) and FcγRIII (4 μg/ml) (Biolegend, San Diego, CA, USA, clone: 3G8) were added before the incubation with purified IgG4 antibodies. Nuclear DNA was labeled with 0.25 μg/ml DAPI (Thermo Fischer Scientific, Rockford, USA) in PBS for 5 min. Cells were then mounted in VECTASHIELD-Antifade mounting medium (Vector Laboratories, CA, USA) and the slides were analyzed using a Zeiss LM-Set Axiocam MRm microscope (Carl Zeiss, Thornwood, NY, USA).
To evaluate the capacity of purified IgG4 from EN, Mf+, Mf- and CP individuals to interact with Brugia antigen, 50 μl Brugia antigen (10 μg/ml) were coated on high binding ELISA plates (Greiner Bio-One, Frickenhausen, Germany) overnight at 4°C. The plates were then washed 5 times with PBS/0.05% Tween 20 and blocked with PBS/1% BSA for 1 hour at room temperature. The wash step was repeated and 50 μl/well of purified IgG4 from EN, Mf+, Mf- and CP (2.5 μg/ml) were added and the plates were incubated again at 4°C overnight. The wells were washed again as described above and diluted biotin-conjugated mouse anti-human IgG4 (clone JDC-14) (1:1000) (from BD Biosciences, Heidelberg, Germany) was added, followed by incubation at room temperature for 2 hours. After an additional washing step, the plates were incubated with 50 μl/well of Streptavidin-HRP for 45 min in the dark. After a final washing step, 50 μl/well TMB substrate solution were added and the reaction was stopped with 25 μl/well 2N H2SO4 (Merck KGAA, Darmstadt, Germany). Optical density was measured at 450 nm using the SpectraMAX ELISA reader and the results were expressed as arbitrary units (AU) using as a standard a plasma sample arbitrarily set at 5 AU.
All statistical analyzes were performed using Prism 5.03 software (GraphPad Software, Inc., La Jolla, USA). Comparative analyzes among groups were conducted using the Kruskal-Wallis test with a Dunn’s nonparametric post-hoc test (> 2 groups). Significance was accepted when p < 0.05. Correlation between the levels of antibodies in IgG negative fractions and inhibition capacity was analyzed using Spearman’s rank correlation.
To define the initial antibody profile of EN, Mf+, Mf- and CP, we compared the plasma levels of IgG1, IgG2, IgG3, IgG4, IgE, IgM and IgA in different groups using a Luminex-based immunoassay. We found that the IgG1 expressions observed in EN and CP were similarly high. In contrast, Mf+ and Mf- individuals presented reduced IgG1 levels (Fig 1A). However, while the highest levels of IgG2 were detected in the plasma of CP individuals, plasmatic IgG2 in EN and Mf- were significantly lower compared to Mf+ and CP (Fig 1B). The differences observed in the expression of IgG3 between the four groups were not statistically significant (Fig 1C). Interestingly, the expression of IgG4 was relatively low in EN, Mf- and CP but significantly elevated in Mf+ (Fig 1D). This contrasts with lower levels of IgE in those patients in comparison to Mf- and patients with chronic pathological manifestations (Fig 1E). In addition, plasma of EN expressed higher IgA levels compared to Mf+, Mf- and CP (Fig 1F), whereas no significant differences were seen in the expression of IgM (Fig 1G).
Because plasma samples from Mf+ individuals presented high levels of IgG4 antibody and since IgG4 antibodies are known to exhibit anti-inflammatory properties, we hypothesized that plasma from Mf carriers, and specifically IgG4 molecules, would preferentially down-modulate granulocyte activation and degranulation. We then next investigated how crude plasma of NEC, EN, Mf+, Mf- or CP modulates the function of IL-3/anti-IgE/BmAg activated granulocytes by monitoring the expression levels of CD63/HLADR and analyzing the release of granule components (histamine, ECP and NE). While plasma from NEC, CP and Mf- had no effect on granulocytes (Fig 2 and S2 Fig), those from EN and Mf+ significantly inhibited activation of granulocytes as indicated by the lower percentages of CD63+HLADR- cells (Fig 2A). Interestingly, the plasma of EN presented a higher inhibitory potential on granulocyte activation when compared to those of Mf+. In line with the activation data, plasma from both EN and Mf+ significantly suppressed the release of histamine (Fig 2B and S2 Fig) and NE (Fig 2C). In addition, histamine levels were higher after 30 min and were lower, but detectable, after 18 hours. However, while the plasma of Mf+ individuals significantly inhibited the release of ECP in granulocyte cultures, those of EN failed to suppress the release of ECP (Fig 2D). These results indicate that, in lymphatic filariasis, active factors in EN and Mf+ infected patients’ plasma environment but not present in Mf- and CP patients impaired granulocyte activation.
To define the role of IgGs in the suppression of granulocytes by plasma of EN and Mf, we depleted IgG antibodies per affinity chromatography. The purity was analyzed (S3 Fig), and the ability of IgG positive and negative fractions to modulate granulocyte activation was tested. Interestingly, while IgG negative (IgG-) fractions of EN significantly suppressed granulocyte activation, IgG positive (IgG+) fractions showed no effect (Fig 3A). Both IgG+ and IgG- fractions from Mf+ significantly inhibited granulocyte activation (Fig 3B) whereas neither IgG+ nor IgG- fractions from NEC, Mf- and CP affected granulocyte activation (Fig 3C–3E). Moreover, in Mf+, the IgG-related inhibition was significantly higher than that observed with negative fractions. These trends were also reflected in the release of histamine (Fig 3F) and NE (Fig 3G). Surprisingly both fractions from EN did not impair ECP release (Fig 3H) when compared with histamine and NE. These data suggest that whereas total IgG from Mf+ individuals inhibited granulocyte activation, IgG-independent factors, seem to be involved in the suppression by plasma of both EN and Mf+ individuals.
To define the IgG-independent factors responsible for granulocyte suppression in EN, we correlated the expression of the remaining antibodies in the IgG-negative fractions (IgA, IgE and IgM) with the ability of these plasma to inhibit granulocyte functions. Interestingly, while IgE and IgM presented no correlation with the inhibitory capacity of IgG-negative fractions of EN (Fig 4A and 4B), IgA expression significantly correlated with the inhibition capacity on activated granulocytes (Fig 4C). In addition, while IgA-depleted fractions [(EN) IgA-] lose their ability to suppress granulocyte activation as shown by similar expression of CD63+ cells when compared to the control, peptide M purified IgA+ fractions significantly reduced granulocyte activation (Fig 4D). These data strongly suggest that IgA expression in EN is associated with the inhibition effect observed when their plasma were incubated with activated granulocytes.
We next investigated whether the modulation of granulocyte activation and degranulation by Mf+ IgG fractions is associated with the presence of the anti-inflammatory isotype IgG4. Highly pure fractions of IgG4 antibodies were prepared and the purity validated (S4 Fig). Thereafter the purified fractions were tested on activated granulocytes. Strikingly, while IgG4 antibodies from EN, Mf+ and Mf- significantly suppressed granulocyte activation (Fig 5A–5C and S4 Fig), those from CP have failed to suppress granulocyte activation (Fig 5D) compared to the control. In addition, the suppressive effect was completely abrogated after IgG4 removal from IgG fractions (Fig 5A–5C), suggesting that the suppressive effect of IgG fractions stems from IgG4 molecules and not from other IgG antibodies. Furthermore, we investigated whether these effects were dose-dependent. Whereas increasing concentrations of IgG4 from EN, Mf+ and Mf- proportionally reduced the percentage of activated cells (CD63+/HLADR-) in a dose-dependent manner, no dose effect was seen when IgG4 from CP patients were used (Fig 5E). Interestingly, the suppressive effect of IgG4 affected mostly neutrophils and basophils but not eosinophils (S5 Fig). Consistent with the granulocyte activation data, we detected lower levels of histamine and NE in supernatants of granulocyte cultures treated with IgG4 antibodies from EN, Mf+ and Mf- compared to those treated with CP-IgG4 (Fig 5F and 5G). However, no significant reduction in the release of ECP was observed after incubation with IgG4 from EN or Mf- (Fig 5H). Table 1 summarizes the effects of plasma, IgG and IgG4 fractions on granulocyte activation.
To further explore the mechanisms by which IgG4 interfere with granulocyte activities, we examined the ability of purified IgG4 antibodies from each group to bind to granulocytes and Brugia antigen. While IgG4 molecules from EN, Mf+ and Mf- were able to interact with effector cells, those from CP had no affinity to granulocytes (Fig 6). However, IgG4 from Mf+ presented a much higher affinity for the cells in comparison to those from Mf- and EN (Fig 6A–6C and 6E). Since differences in the affinity of IgG4 from EN, Mf+, Mf- and CP to form complex with Brugia antigen can also affect granulocyte modulation, we analyzed the capacity of IgG4 from the different groups to interact with Brugia antigen. We detected no significant differences in the capacity of IgG4 from EN, Mf+, Mf- and CP to form a complex with Brugia antigen (Fig 6F).
Because IgG4 from Mf+, Mf- and EN bound to granulocytes, we next investigated which FcγRs are involved in their fixation by using blocking antibodies against FCγRI, FCγRII, and FCγRIII. We observed that the blockade of FcγRI (Fig 7B and 7H) and FcγRII (Fig 7C and 7H) but not FcγRIII (Fig 7D and 7H) significantly reduced IgG4 binding to granulocytes. Interestingly, the capacity of IgG4 to bind to granulocytes was completely abrogated when FcγRI and FcγRII were simultaneously blocked (Fig 7E and 7H). Corresponding results were also observed when the activation of granulocytes in the presence of IgG4 and anti-FCγRs was measured (Fig 7I). These findings suggest that IgG4-mediated granulocyte suppression in Mf+ patients involves FcγRI and FcγRII but not FcγRIII.
The pathology of lymphatic filariasis results from the complex interplay between the pathogenic potential of the parasite, the host’s immune response and collateral bacterial and/or fungal infections [55]. Even though the role of IgG4 in immunosuppression in filariasis is well known [56], only few data exist on their impact on effector cells. Recent investigations suggest that different granulocyte subsets may be important in the immune response to helminth infections [28,29]. Here we used a sequential depletion/purification approach to define the immune components that are responsible for granulocyte suppression in the plasma of individuals from the different LF clinical groups. The advantage of this approach is that, in addition to analysing quantitative differences among these groups, it can be investigated whether a given antibody retains functional capacity when used at normal concentrations. We tested both IgG positive and negative fractions and could show that while IgG+ fractions from Mf+ suppressed granulocytes, those of EN, Mf- and CP had no effect. A further purification step on the IgG+ fractions using anti-IgG4 in the purification matrix indicated that IgG4+ fractions from Mf+, Mf- and EN displayed comparable suppressive capacities. This apparent contradiction with the data obtained using plasma and total IgG is due to the level of purification of IgG4 used here at the same concentration and confirms that quantitative differences in the expression of IgG4 play a significant role in the suppressive properties observed in the plasma of the different clinical groups. These data suggest that IgG4 antibodies from EN, Mf+ and Mf- might have the same overall suppressive properties and the alterations observed when using total IgG or crude plasma are due to differences in the ratios IgG4/total IgG as previously postulated [10,57]. These observations are in line with findings of Mohapatra et al., indicating that plasma of asymptomatic individuals (Mf+) in contrast to those of CP mediated suppression of mitogen-induced proliferation of human PBMCs [58]. Bennuru et al. further demonstrated that sera from CP patients promoted the proliferation of lymphatic endothelial cells whereas those of EN suppressed this proliferation [59]. Plasma of Mf- and CP contain higher levels of pro-inflammatory IgG1-3 and IgE antibodies, known to be relevant for parasite clearance as demonstrated in different animal models of filariasis [60–66] but are also associated with pathology development in CP-patients [4,9,67].
The data obtained using the plasma of CP patients were initially similar to those for EN and Mf-, since they displayed low levels of IgG4 and presented no inhibition effect on granulocytes. However purified IgG4 from this clinical group fundamentally lack suppressive properties even when higher concentrations were used, suggesting the existence of two distinct mechanisms of inhibition of the immunoregulatory antibody IgG4 in CP patients, first by down-modulating its level and second by modifying the properties of the remaining IgG4 molecules. This is likely associated with the proinflammatory environment including Th17 and Th22 characteristic of the CP profile [68]. A recent study on patients undergoing cardiac surgery indicated a strong connection between glycosylation features related to fucosylation, sialylation and bisecting GlcNAc and severity of inflammatory response [69]. Since glycosylation features can affect the function of antibodies [70], this is a possible explanation for the lack of suppressive properties of IgG4 antibodies purified from CP patients. Since no differences were detectable in the purity of the IgG4 positive fractions, and because IgG4 is known to present no allotypic variations [71], post-translational alterations could, as mentioned above, explain why functional differences are observed between IgG4 molecules in our settings. Indeed, different investigations have shown that all IgGs contain a conserved glycosylation site at N297 in CH2 domain that is important for the structural conformation of the Fc region necessary for binding to FcRs and complement factors [71–73]. Differences in the glycosylation states may ultimately influence the effector pathways elicited by the Fc domain [73]. Fucosylation and sialylation for example are two extensively investigated glycan modifications of Fc that significantly modulate the affinity of Fc regions to FcRs [70,73]. In several health and disease settings, a shift toward certain Fab- and Fc-glycoforms of antibodies has been reported [70]. It is very likely that the degree of glycosylation differs in the IgG4 molecules from EN, Mf+, Mf- and CP and subsequently modulates their affinity to FcγRs.
In addition to post-translational alterations, other factors such as the previously described differential recognition of filarial antigens by antibodies in the different clinical groups [74], could also have an impact on the ability of IgG4 antibodies to bind granulocytes. Indeed, differential recognition of filarial antigens could differently affect the formation of antigen-antibody complexes and thus modulate the inhibition capacity of IgG4 on activated granulocytes. Other parameters that potentially can explain the different affinity and inhibition capacity of IgG4 antibodies from the four LF clinical groups are FcR cross-linking and steric hindrance that can respectively reduce availability and access to FcγRs on granulocyte surface [75]. Differences in the inhibition effect could also be modulated by differences in the levels of autoantibodies in the different clinical groups as suggested by Mishra et al. [76].
The most unexpected finding in the present work is the suppression of granulocyte activation by the plasma of EN since this clinical group is usually associated with putative immunity and strong pro-inflammatory responses [77]. Our data indicated an elevated expression of IgA in the plasma of EN and revealed a significant correlation between IgA expression and the suppressive properties in the IgG-negative fractions of EN. Sahu et al. observed similar trends when comparing the expression of filarial-specific IgA in LF endemic populations [78]. In addition, recent investigations indicated that IgA is a multifaceted molecule that can display both pro- and anti-inflammatory properties depending on the environment and can interact with FcαRI on the surface of eosinophils and neutrophils [79,80]. Our data also indicate that plasma from EN failed to significantly inhibit the release of ECP but suppressed histamine and NE suggesting that IgA in the plasma of EN selectively modulate neutrophil and basophil but have less effect on eosinophils.
Our data also indicate that except CP, IgG4 from all clinical groups suppress granulocytes after interaction with both FcγRI and FcγRII confirming results of previous studies indicating that IgG4 binds to FcγRI, FcγRIIa, FcγRIIb and FcγRIIc [81–83]. Activatory FcγRs typically signal through an immunoreceptor tyrosine-based activation motif (ITAM) whereas the inhibitory FcγRIIb triggers signals via immunoreceptor tyrosine-based inhibitory motif (ITIM) [84,85]. Stimulation through ITAM pathway leads to pro-inflammatory activity resulting in destruction and clearance of antigens by phagocytosis, ADCC and promotion of antigen presentation [84]. Bruhns et al. further suggested that IgG4 antibodies display a higher affinity for the inhibitory receptor FcγRIIb [86]. This suggests, in our settings, that IgG4 antibodies may exert their suppressive properties via two distinct but complementary pathways. Suppressive IgG4 antibodies very likely bind to the inhibitory FcγRIIb and deliver a direct anti-inflammatory signal while impeaching pro-inflammatory antibodies (IgG-1-3) to interact with FcγRI.
While investigating the role of immunoglobulins in the modulation of granulocyte activation, we used affinity-purified total IgG, IgA and IgG4. The use of non-antigen specific antibodies was due to technical limitations associated with the amount of patient’s material available. However, the incubation of granulocytes with anti-IgE and IL-3 allows non-antigen specific stimulation of granulocyte subpopulations. Also, due to the well-known technical difficulties associated to the cryopreservation of granulocytes [87,88], the present study used a heterologous system where sera and purified antibody fractions from Ghanaian patients and controls were tested on heterologous granulocytes from healthy European blood donors. Even though a certain level of alloreaction cannot be excluded, our data are validated by the use of the same background for all tested samples. Indeed, all plasma or purified antibody fractions were tested on granulocytes of the same group of donors (n = 9). Also, the use of both heterologous and autologous settings for non-endemic controls showed no impact on the granulocyte activation and histamine release (S6 Fig).
Even though the sample size is relatively small due to the difficulty to recruit patients that have received no anti-filarial treatment after extensive mass drug administration programs and the significant reduction of the disease burden in this region [89,90], the current study extends previous findings suggesting that expression of IgG4 in asymptomatic Mf+ individuals is associated with inhibition of granulocyte functions [10,91,92] and suggests that prominent IgA expression in EN also affects granulocytes’ functions. Our data also provide new insights on a possible role of functional modulation of IgG4 antibodies in CP-patients, providing possible novel clarifications of the mechanisms through which tolerance or pathology is induced in LF, and suggest that IgA and IgG4 may represent meaningful candidates for targeted therapy against LF. Modulating IgG4 and IgA expression and functional properties may for example contribute in the future to the reduction of inflammatory damages in patients with chronic filarial infections.
|
10.1371/journal.pcbi.1003914 | Mouse Hair Cycle Expression Dynamics Modeled as Coupled Mesenchymal and Epithelial Oscillators | The hair cycle is a dynamic process where follicles repeatedly move through phases of growth, retraction, and relative quiescence. This process is an example of temporal and spatial biological complexity. Understanding of the hair cycle and its regulation would shed light on many other complex systems relevant to biological and medical research. Currently, a systematic characterization of gene expression and summarization within the context of a mathematical model is not yet available. Given the cyclic nature of the hair cycle, we felt it was important to consider a subset of genes with periodic expression. To this end, we combined several mathematical approaches with high-throughput, whole mouse skin, mRNA expression data to characterize aspects of the dynamics and the possible cell populations corresponding to potentially periodic patterns. In particular two gene clusters, demonstrating properties of out-of-phase synchronized expression, were identified. A mean field, phase coupled oscillator model was shown to quantitatively recapitulate the synchronization observed in the data. Furthermore, we found only one configuration of positive-negative coupling to be dynamically stable, which provided insight on general features of the regulation. Subsequent bifurcation analysis was able to identify and describe alternate states based on perturbation of system parameters. A 2-population mixture model and cell type enrichment was used to associate the two gene clusters to features of background mesenchymal populations and rapidly expanding follicular epithelial cells. Distinct timing and localization of expression was also shown by RNA and protein imaging for representative genes. Taken together, the evidence suggests that synchronization between expanding epithelial and background mesenchymal cells may be maintained, in part, by inhibitory regulation, and potential mediators of this regulation were identified. Furthermore, the model suggests that impairing this negative regulation will drive a bifurcation which may represent transition into a pathological state such as hair miniaturization.
| The hair cycle represents a complex process of particular interest in the study of regulated proliferation, apoptosis and differentiation. While various modeling strategies are presented in the literature, none attempt to link extensive molecular details, provided by high-throughput experiments, with high-level, system properties. Thus, we re-analyzed a previously published mRNA expression time course study and found that we could readily identify a sizeable subset of genes that was expressed in synchrony with the hair cycle itself. The data is summarized in a dynamic, mathematical model of coupled oscillators. We demonstrate that a particular coupling scheme is sufficient to explain the observed synchronization. Further analysis associated specific expression patterns to general yet distinct cell populations, background mesenchymal and rapidly expanding follicular epithelial cells. Experimental imaging results are presented to show the localization of candidate genes from each population. Taken together, the results describe a possible mechanism for regulation between epithelial and mesenchymal populations. We also described an alternate state similar to hair miniaturization, which is predicted by the oscillator model. This study exemplifies the strengths of combining systems-level analysis with high-throughput experimental data to obtain a novel view of a complex system such as the hair cycle.
| The miniorgan of the hair follicle represents a complex biological system that undergoes repeated phases of death and regeneration over its lifetime [1]–[3]. Understanding of the hair cycle and its regulation would shed light on many other complex systems relevant to biological and medical research including morphogenesis, stem cell biology, response to environmental perturbations and general spatiotemporal patterning [4]. The stages of the hair cycle have been well documented, at least from a morphological standpoint, in mouse models [5]. The period of hair growth, known as anagen, involves rapid proliferation of follicular epithelial cells, such as MatriX (MX) cells in the hair bulb, which surround a key group of mesenchymal cells that form the dermal papilla (DP). Matrix cells differentiate to eventually compose various epithelial populations of the hair shaft. Anagen is followed by catagen, which is characterized by high levels of apoptosis. Finally, telogen is typically described as a quiescent period between growth phases.
The molecular mechanisms underlying this cyclical pattern of death and renewal in hair follicles are not well understood; however, some general concepts, as well as specific molecular regulators, have been identified. One key aspect is the communication between epithelial and mesenchymal cells. Numerous studies have identified physical interactions between these cell populations, as well as several possible signaling molecules [6]. One well studied signaling molecule of the hair cycle is Tgfβ2, which is synthesized and secreted by DP cells. The evidence suggests that, in general, Tgfβ2 suppresses proliferation and induces catagen-like changes in the follicle, including apoptosis of MX cells [7]. However, recent studies have identified a Tgfβ2 mediated pathway which activates epithelial stem cells to promote hair follicle regeneration [8]. This underscores the complexity of the signaling pathways involved.
Mathematical models of general features of hair cycling have also been studied. In a recent study by Murray et al., the authors model follicle growth and coupling as an excitable medium [9]. Their model incorporates general aspects of hair cycle regulation, and shows qualitative agreement to experimental observations. Also recently, Al-Nuaimi et al. developed a general model for hair cycling based on observations in the literature [10]. These authors derived a mathematical, kinetic model which proposed that negative feedback between dynamic MX keratinocytes and static DP cells could reproduce the cyclical growth patterns of the hair follicle. Although these models are significant, they do not attempt to incorporate any specific molecular details in a data-driven approach by formally analyzing large scale experimental data sets. In the study by Lin et al., mRNA microarrays were compiled over the first three rounds of hair growth: morphogenesis, the second naturally synchronized cycle and a depletion-induced cycle [11]. The results demonstrated recurrent gene expression corresponding to hair growth, and the authors specifically focused on genes related to circadian rhythms. However, the study does not address the many other genes observed to have similar patterns. Currently, an unmet need is the development of data-driven approaches that can couple the existing transcriptome-wide data to systems-level properties of hair cycling using formal dynamic models.
Nonlinear dynamical models have provided valuable insights into many oscillating biological systems [12]–[14], and have even been used to suggest general design principles of oscillating metabolic and signaling networks [15]. In general, simplified oscillator models have been developed to describe properties resulting from oscillator interactions or coupling. One such model is the well-studied Kuramoto model [16]. Here all oscillators are interconnected with the same coupling strength, and studied as a single mean field. Although a major simplification, mean field models have successfully described high level properties of many large, complex systems including statistical mechanics (for a review see [17]), economics [18], [19] and even social networks [20]. Importantly, the Kuramoto model is capable of capturing a critical phase transition from an incoherent state to one in which all oscillators converge to a single, coherent cluster. This behavior is referred to as synchronization and is quantified by complex order parameters [21]. Modified and extended versions of the Kuramoto model have been used in many complex systems [22] including synthetic genetic networks [23]; cyclical gene expression and cellular networks [24]; neural networks for memory and brain activity [25], [26]; chemical oscillators [16]; and laser arrays [27], [28]. We would not expect such models to be capable of incorporating or identifying mechanistic molecular interactions or specific details; however, given the above literature evidence, they can be quite successful at describing systems level properties, such as synchronization, and the sufficient, underlying conditions that can produce them.
Our aim here was to investigate a subset of genes whose expression changes as a function of time in a potentially periodic manner, similar to the cyclical nature of hair growth. Previous modeling studies, which have focused on general aspects of hair growth, represent important initial steps in applying mathematical strategies to understanding the hair cycle [9], [10]; however, these models are not driven by molecular-level data. In contrast, other studies use high-throughput molecular-level data to identify important targets, and they apply additional experiments to delineate specific molecular mechanisms [11]; however, these studies are limited to investigation of a small number of genes, and they do not attempt to place the observations into a quantitative modeling context. In this study, we focused on two complementary mathematical modeling strategies that look at high-level features, such as average dynamic behavior, that is based on the individual patterns of thousands of genes. Thus, we are attempting to bridge the gap between the two strategies described above. Using whole skin, transcriptome-wide expression data, we demonstrate the existence of two subsets of genes that have synchronized, out-of-phase expression profiles. Motivated by this observation, we applied a coupled oscillator modeling framework to identify a specific coupling configuration that spontaneously, and stably reproduced the observed synchronization. We then applied a 2-population mixture model to associate the corresponding gene clusters to two computationally determined populations, a rapidly expanding population and a relatively static background population. The estimated population dynamics indicated an association between computationally derived background/expanding populations and the mesenchymal/follicular epithelial cells, respectively. Cell type specific enrichment analysis and experimental imaging with in situ hybridization and immunofluorescence all demonstrated similar associations. The results describe a coupling scheme, between these two cell populations, which would be sufficient to maintain the observed synchronization. Specific signaling molecules were also identified as being priority follow-up targets for drivers of synchronization. To our knowledge this is the first attempt at integrating high-throughput molecular data with a mathematical model to predict systems level properties, such as synchronization and population dynamics.
Given the proposed cyclical nature of hair growth, we investigated the possibility of periodically expressed mRNA in the microarray data collected by Lin et al. [11]. We assumed that such expression patterns may relate to hair cycle regulation. We applied a periodic identification scheme for non-uniformly sampled data [29]. This method estimates a discrete Fourier Series Decomposition (FSD) for each expression signal by robust regression. We identified 4627 probesets (mapping to 3567 unique genes) as significantly periodic signals with a false discovery rate of 10%. Using the semantic measure Normalized Google Distance (NGD), we found that 315 of the corresponding periodic genes had a notable proximity to the hair cycle described in a survey of PubMed abstracts. This translates to an enrichment p-value of 2.5E-5. For example, periodic genes with the lowest NGD are discussed in the literature as being related to hair pigmentation (Mc1r, Tyrp1, Stx17), growth and cycle regulation (Liph, Foxn1), disorders and malformations (Lpar6, Zdhhc13, Krt85) and general associations to hair (Krt28). For a full list of genes and related NGDs to hair see Supplementary File S1.
To further investigate the periodic expression, we assigned a specific frequency and a phase shift to each signal. This was done using the Principal Periodic Component (PPC) as an approximation to the FSD. Both the PPC and FSD reasonably recapitulated the time course trajectories, primarily the low frequency expression signals (Supplementary Figure S1 top). Furthermore, the majority of the periodic signals, 3988 probesets, were associated to this low frequency, which corresponds to a period of 31 days (Supplementary Figure S1 bottom). The 31 day period of expression was on the same time scale as the hair cycle, and further suggested a relationship between the corresponding genes and hair growth and regulation. Although repeating cycles were not directly observed to demonstrate cyclical behavior, we note that the data was a composite of both the second natural and depilation-induced hair growth cycles and, therefore, the expression patterns were common to, at least, these two cycles.
For visual examination, we sorted the periodically expressed probesets by frequency and phase shift (Figure 1A). We compared this data to the PPC trajectories (Figure 1B) to underscore the similarities. We noted a distinct clustering of the expression signals, including two clusters within the low frequency probesets. The lower and upper clusters demonstrate maximal and minimal expression near the end of anagen, respectively. This reciprocal periodic behavior is referred to as out-of-phase periodicity. We calculated the phase shift based on the time to the next maximum value. The phase shift (Figure 1C) clearly identifies these two tightly clustered groups. We associated 1452 probesets to cluster one and 2536 to cluster two. Cluster one and two correspond to probesets that are predicted to reach maximum expression at approximately 33 (first anagen phase after morphogenesis) and 48 (following telogen phase) days postnatal, respectively. The separation of the phase shift further indicates that the two groups are almost exactly out-of-phase. The mean of the two groups is separated by 15.4 days which in polar coordinates corresponds to approximately 180°. Although fast cycling genes were identified, we chose to overlook this group due to a relatively poor fit to the PPC, median coefficient of determination was less than 0.5 (Supplementary Figure S1 Top). However, given that the data was derived from full-thickness mouse skin, it is possible that cyclic gene expression in Keratinocytes could be contributing to the short period signal. This possibility was strengthened with the recent report that human epidermal stem cell functions are regulated by circadian oscillations [30] with a period of 24 hours in vitro. Additional experiments, with higher time-resolution sampling, may better describe fast cycling genes and may provide a link between Keratinocytes, hair-cycling and circadian rhythms.
The clustering of periodic signals was quantified using complex order parameters [21], [31], [32]. Considering only the periodic component of the low frequency expression signals corresponding to a 31 day period, we can visualize the expression as points moving around the unit circle in the complex plane as they travel through the cycles. How tightly grouped these points are can be quantified by a set of order parameters, :(1)where is the number of oscillators and is the instantaneous phase or the position of oscillator on the unit circle at time . See Methods for more details on the formulation of EQ 1. When studying systems of coupled oscillators, the magnitude of , denoted , is used to quantify the coherence or synchronization of the system (Supplementary Figure S2 shows typical values for specific configurations of points). In such systems, a high level of synchronization is typically the result of coupling between oscillators. The low frequency expression signals in the hair cycle demonstrated high out-of-phase synchronization as measured by , as well as a notable asymmetry, due to uneven sized clusters, measured by (Figure 2A). As a negative control, we randomized the oscillator phases to demonstrate and near zero for a similar, but un-clustered system (Figure 2A, black lines). Synchronization was further exemplified by considering order parameters calculated for the specific clusters, denoted and (Figure 2B, green and blue lines corresponding to clusters shown in Figure 1C). This level of synchronization appears to be dynamically stable throughout the time course. If we can expect similar molecular behavior underlying subsequent cycles of hair growth, we would anticipate these periodic expression signals to repeat. In an ideal case the low frequency gene expression could then be viewed much like a system of oscillators, and the observed dynamic stability could be investigated in that context.
We believed that strong synchronization over multiple expression signals was indicative of regulation between the corresponding genes. As mentioned above, systems of multiple agents with periodic behavior are often described using coupled oscillator models. Furthermore, and most significantly, we found a striking similarity between the observed expression in the mouse hair cycle and a simple system of coupled oscillators formulated by Hong and Strogatz [33]. In this model one possible attractor (or long time behavior) was the synchronization of two, asymmetric, out-of-phase, oscillator clusters. Spontaneous, stable synchronization was observed when one cluster was positively coupled to the system's macroscopic rhythm, embodied by complex order parameter (see EQ 1), and the other was negatively coupled. In this case positive or negative coupling indicates oscillators that move towards or away from , respectively. Although such models are a significant simplification from the true biology, the qualitative agreement was encouraging, and we wished to investigate if reasonable insights could be drawn from such an abstraction. In the following we consider a system of Low Frequency Oscillators (LFO) corresponding to the 31 day period expression signals identified above.
We first considered if our system could be quantitatively described by the above model. At the level of individual oscillators this model can be written as(2)where is the phase of the th oscillator, is the natural or intrinsic frequency for , is the coupling constant, is the number of positively coupled oscillators, is the total number of oscillators, and from EQ 1 where the subscript is dropped for simplicity. The dot denotes change with respect to time. Here, we have assumed two groups representing positive and negative coupling denoted by superscripts + and −, respectively. The coupling constants are related by where , , . This model can be simplified to two dimensions when describing only the dynamics of the first order parameter for the two clusters (recall Figure 1C, green and blue clusters), which was what we focused on here as a high level characterization of the system(3)where and are the first order parameters (similar to EQ 1 with ) for the two oscillator groups related to positive and negative coupling, respectively; is the proportion of positively coupled oscillators and . The bar denotes complex conjugate. Given the first order parameters of our observed clusters ( and , recall Figure 2B green and blue lines) and the number of oscillators for the two clusters (which provides ), we solved exactly for the relative coupling strength, , and the intrinsic frequency distribution, . We refer the reader to Methods for a detailed description of this process including additional assumptions. Interestingly, the only stable configuration was negative coupling of cluster 1, the smaller cluster (). Figure 3A shows that this configuration resulted in spontaneous, stable synchronization (config 1, red solid line, a corresponding movie of the individual oscillators is available in Supplementary File S2), a steady-state magnitude of the first order parameter equivalent to the observed magnitude (recall Figure 2A), and clusters that remain 180° out-of-phase (upper panel of Figure 3A). Furthermore, if cluster 1 is positively coupled (), the model system is unstable and no synchronization is observed (Figure 3A purple dashed line, config 2). The results suggest that a positive coupling of cluster 1 to the system is physically unlikely.
This result provides us with our first biological insight, specifically, the genes corresponding to cluster 1 were negatively coupled to the system's macroscopic rhythm, and repelled by this average behavior of the system. If these genes can be associated with specific cell populations, then inhibition or repression of this population by the system would explain the simultaneous negative coupling of a large number of genes. Inhibition at the cellular level could be achieved by regulated apoptosis, which is consistent with a previous model of the hair cycle from the literature [10]. The authors described that negative feedback in the form of regulated apoptosis or inhibition of regulated proliferation could produce observed cyclical hair growth patterns.
Using this coupled oscillator model, we investigated what other states could be possible if specific variables were changed. We constructed a bifurcation diagram that shows the steady-state behavior of the system given different values of (see EQ 3) and assuming that other properties remain constant. Figure 3B shows three qualitatively different states for varying values of . Low values of relate to an incoherent state which has no synchronization and, therefore, no clustering. For large values of , we find the -state in which two clusters are stable, out-of-phase and oscillate with the same long period throughout a range of values, with increasing asymmetry to one cluster. Finally, we observed a traveling wave state for intermediate values of , here the period of the system is greatly reduced and even variable with respect to . These states were described by Hong and Strogatz [33]. Interestingly, the hair system lies at the edge of the -state, near a bifurcation into the traveling wave state. A reduction in corresponds to a decrease in the relative size of cluster 2, and would reduce the effective negative coupling on oscillators in cluster 1, which is what drives the system into a different state. Therefore the model anticipates that on average, removal of oscillators in cluster 2 would result in a loss of regulation which is specifically associated with varying, high frequency oscillations. Furthermore, we noted that the reduced period of the traveling wave state is similar to the fast cycling of short hairs in an existing model of hair cycle [10], which was shown to result from changing parameters associated with negative feedback. This alternative state may be biologically related to the pathological state of hair miniaturization and androgenic alopecia.
Although our model recapitulated the observed synchronization, we emphasize here the various abstractions introduced. First, we did not attempt to model the physical, molecular interactions involved, as we felt such an approach would be too error prone given limited data and a priori knowledge [34]. Instead, we chose to implement the simplest phenomenological model we could conceive to describe observed behaviors. Here oscillators represented probesets with periodic expression patterns (see Methods EQ 7). Modeled oscillators changed due to two terms: the intrinsic term, , and the coupling terms, (both from EQ 2). In a physical model the intrinsic term would represent some constant, external force that independently drives oscillations, in this case the intrinsic term represents all the unknown interactions that were the basis for periodic expression. As a result, insight and prediction related to these interactions was beyond this model's reach. In contrast the coupling terms were of most interest, and represented the average force felt by oscillators of a certain group, due to other oscillators. This coupling term is what drove the observed out-of-phase synchronization. The model was further abstracted as we specifically considered the average dynamics of the two clusters (see EQ 3), which were coupled via the average coupling of the underlying oscillators. In a sense the model treated interactions as being ‘smeared-out over all oscillators. Although exact physical interpretation is difficult, this average effect on the many corresponding genes may be the result of an extensive transcriptional program or a cellular event such as regulated apoptosis or proliferation, and suggests that these clusters may be related to specific cell populations.
Given previous results, we hypothesized that the two observed clusters of LFO gene expression may relate to specific cell types or general cell populations. We noted that observed relative expression changes can be attributed to changes in relative cell population size as opposed to intracellular changes. Furthermore, opposite expression patterns can result when one population is constant in size and the other is changing (See Supplementary Figure S3, and Methods for details). To investigate further, we applied in silico microdissection [35] to estimate properties of two distinct cell populations from the heterogeneous hair cycle data. Here, the 2-population mixture model assumes static intracellular gene expression. Furthermore, we also assume one population is expanding, while the second is a static background population. We emphasize here that we make no assumption as to the specific cell types contributing to these populations nor the relative expression levels of genes within these populations. We then fit this model using all of 45k+ probesets from each micro array chip.
From the heterogeneous hair cycle samples, we were able to approximate the dynamics of dominant expanding cell populations (Figure 4). The trajectory was consistent with cells associated with rapidly proliferating follicle epithelial cells. In particular the model estimated (without any a priori knowledge) a sharp and complete depletion of the expanding population within the catagen time frame. The model also identified differences between samples from the natural and induced cycle. Specifically, the model estimated a slower anagen onset in depletion induced mice, this was also observed by comparing morphologies of tissue sections in Lin et al. [11]. Furthermore, we demonstrated that these features were not estimated in a negative control (see Supplementary Figure S4), using a permutation strategy to simulate data with no hair cycle relationship. This suggested that our results were not due to an artifact or bias introduced in the analysis. These observations suggested that the in silico microdissection procedure was able to identify expanding cell populations compared to a static background populations.
The model was also able to estimate static intracellular expression levels for each population. Combining this with the estimated population size, we were able to compare estimated expression levels to those observed in the data. Overall, we found that the majority of expression signals were not well described by such a simple model (Figure 5A, blue histogram); however, the model was able to describe the majority of the variation in the expression of the LFOs (Figure 5A, pink histogram, 50% of probesets demonstrated a COD>0.5). As above we did not identify such improved statistics of the LFOs in the permuted negative control. This suggested the significance of such a model in describing expression of the LFO subgroup. Using the estimated expression and standard error estimates, we identified probesets with statistically significant differential expression between the two estimated populations (expanding and background) in the form of a t-statistic. The t-statistic shows the expression difference relative to the estimated error. Again more statistically significant differential expression was found in the LFO subgroup (Figure 5B). Probesets from the LFO subgroup were assigned to the population in which they demonstrated a statistically significant increase in expression, 99.7% of the LFO probesets met the statistical requirements for assignment. Figure 5C shows the remarkable similarity between clusters 1 and 2 (recall Figure 1C) and the expanding and background population respectively. In fact, dividing the LFO probesets in this manner yields the same division as achieved when considering oscillator phase (Supplementary Figure S6). Ultimately, this procedure allowed us to associate probesets and corresponding genes to two dynamically distinct populations.
We next investigated the possibility that the estimated populations were associated to specific biological cell types involved in the hair cycle. We made use of two existing studies, Rendl et al. [36] and Greco et al. [37], which dissected hair follicles into specific, predefined cell types. For each cell type, signature genes, genes expressed predominantly in that cell type, were identified. We then calculated the enrichment of these signature genes in the two estimated populations (Supplementary Table S1). In particular, we found a significant enrichment for epithelial MX and, to a much lower degree, ORS cells in the expanding population, and respectively, but not for the background population. Alternatively, we found a significant enrichment for mesenchymal DP cells in the background population, , but not for the expanding population. Other cell types, Melanocytes and Bulge Cells, were found to be significantly enriched in both populations, but to a larger degree in the background population (see Supplementary Figure S7A for relative signature list size and overlap with estimated populations). We also observed overlap between the experimentally determined gene signatures, for example the overlap between Bulge [37] and DP [36] signature genes (see Supplementary Figure S7B) corresponds to an enrichment p-value. The results show that the computationally estimated populations do not uniquely represent a specific, predefined cell type; however, they do have distinct associations. Furthermore the association of epithelial MX and ORS cells to the rapidly expanding population and DP cells to the static background population was consistent with the known relative population dynamics of those cell types. We emphasize that DP population size does change throughout the hair cycle [38]. However, the DP population is not observed to undergo the enormous expansion and apoptosis characterizing hair epithelial cells [39]–[41]. Thus, conceptually the DP may be considered considerably less dynamic than the MX derived population [2], which is consistent with our findings here. For a full list of significant probesets with the t-statistic indicating population association, and other metrics, see Supplementary File S3.
These results provide us with a second biological insight: the genes in LFO cluster 1 were associated with expanding cell populations of the follicle that were enriched for follicle epithelial cells as defined Rendl et al. [36], and cluster 2 was associated with background populations that were enriched for mesenchymal DP cells. This bolsters our previous hypothesis that an inhibitory, possibly apoptotic mechanism, is acting on the expanding, epithelial cell population, and that this mechanism is involved in synchronized gene expression of the hair cycle.
To verify specific gene expression patterns and to investigate the localization of gene products within the hair follicle, we applied qRT-PCR, In Situ Hybridization (ISH), and protein antigen staining. To generate tissue samples, we aligned the hair cycle in 10-week old mice with the shave/depilatory induction protocol. Two animals were sacrificed for each of the five time points considered; however, qRT-PCR was done using four technical replicates for both samples and imaging shows results typical of multiple follicles observed over the two biological replicates. Phenotypic changes were quantified by melanogenesis scoring, which is known to correspond to active hair growth. Both biological replicates were observed to have entered anagen within 16 days of induction and to have completely finished the first round of post induction hair growth by 29 days, as determined by scores increasing from and then returning to zero (Supplementary Figure S8A). Dorsal skin was sampled and prepared for RNA analysis or antigen staining at multiple time points throughout the cycle.
An exhaustive investigation was not within the scope of this work; therefore, we identified a subset of candidate genes for experimental follow-up. Candidates were chosen by considering the significance of periodic expression and the t-statistic derived in the 2-population model. Additionally, we considered genes that had plausible, but not well defined connections to hair growth and regulation as determined by the literature. We chose Signal Transducer And Activator Of Transcription 5A (Stat5a), Fermitin Family Member 2 (Fermt2) and Vimentin (Vim) for candidates associated to the background population, as well as Ovo-Like 1 (Ovol1) and SMAD Family Member 6 (Smad6) for the expanding population. In the follicle dissection study of Rendl et al. [36], Stat5a and Ovol1 were also identified as signature genes for DP and MX cells, respectively; however, the other candidates were not linked to a specific cell type in the same study. The metrics relating to these candidate genes as well as others identified in this study are presented in Supplementary File S2. For candidate and control genes, we confirmed the expected relative expression profiles by qRT-PCR (Supplementary Figure S8B,C). Specifically, we confirmed that the relative expression of background candidates decreased during anagen onset and then increased after completion of anagen with maximums observed in or near the telogen phase, while the expanding population demonstrated a reciprocal profile (recall Figure 1, cluster 2 and 1 respectively).
We investigated the localization of candidate gene products by various imaging techniques in samples corresponding to telogen taken before induction (day 0) and anagen taken 16 days after induction (day 16). Here, we recall that background candidates were determined to be markers for hair follicle cell populations that remain relatively stable throughout the hair cycle; these were also enriched with DP signature genes. Using ISH and fibroblast growth factor 7 (Fgf7) as a control marker [42], we confirmed that the signature gene Stat5a, also identified by the 2-population model, was expressed in DP cells (Figure 6A R1). We note that technical negative and positive controls for ISH can be seen in Supplementary Figure S9. Interestingly, ISH imaging produced similar expression results for background candidates in both telogen (day 0) and anagen phases (day 16) while qRT-PCR and microarray suggested notable differences in relative expression between these phases. These observations were explained by the 2-population model above, where increases in the expanding population decreases the relative size of the background population, resulting in notable expression changes for mixed cell population samples, such as the whole skin samples used in qRT-PCR. Although consistent with our expectations as described, for completeness we add that insensitivity in ISH imaging could provide another explanation for the observations. Furthermore, significant changes of Fgf7 expression have been detected in DP isolates [37], although under different conditions from the data considered here. We observed the same localization pattern in the novel candidate marker, Fermt2. Roles for Fermt2 have been proposed in regulation of the extra-cellular matrix, actin organization as well as cell-ECM focal adhesions [43]. It is also associated with β-catenin/TCF4 complex, and knockdown of Fermt2 leads to loss of β-catenin mediated transcription [44], ultimately affecting myogenic development. Other effectors of the β-catenin/TCF4 complex, such as Wnt, are known to be required for the hair inducing property of DP cells [45]. Background population proteins were also localized by immunofluorescence. The morphology, as determined by brightfield and DAPI staining, was used to identify DP localization. We observed localization of Stat5a protein to the DP in anagen samples; however, localization was much more difficult to assess in telogen samples (Figure 6B R1). Technical issues prevented antibody staining for Fermt2; therefore, we considered an alternate candidate, Vim, which was also associated to the background population. Immunofluorescence demonstrated Vim protein expression localized to the cytoplasm of DP cells in both telogen and anagen phases. We also observed Vim staining in the dermal sheath that surrounds the anagen bulb (Figure 6B R2). While Vim expression in the follicle has been reported [36], [46], [47], it is expressed in other dermal cells and it is thus not specific to the hair follicle. Its expression in both DP cells and in cells adjacent to the hair follicle, or macro-environment, emphasizes the importance of the use of whole skin in our study. For example, Plikus et al. [48], [49] demonstrated that the macro-environment can be the source of paracrine signals that influence the hair cycle.
Imaging results also confirmed candidate markers for the expanding population. Here we recall that the expanding population candidates were determined to be markers for cells whose relative population size increases during anagen, followed by a sharp decline in catagen (Figure 4); these were also enriched for MX cell genes. Using Forkhead box protein N1 (Foxn1) as a positive control for MX cells [50], note: Foxn1 was also identified in the current study as a candidate marker for the expanding population), we observed mRNA expression of the signature gene Ovol1 in the proliferating cell populations of the hair shaft in anagen samples, as described in the literature [36], [50]–[52], Figure 7 R1). We also observed the same expression pattern and localization to the hair shaft for the novel candidate marker Smad6 (Figure 7A R2, day 16). Further evidence of an association between these candidate markers and the expanding population was demonstrated by a general lack of staining in telogen samples. The 2-population model attributes this lack of expression to the absence of the expanding cells in the telogen phase (Figure 4). Additionally, immunofluorescence confirmed the localization of Ovol1 and Smad6 protein near the MX cell marker Foxn1 (Figure 7B). Matching the pattern of Foxn1 expression at day 16, ovol1 and Smad6 stained the epithelial, matrix-derived inner root sheath cells and the upper part of the matrix cells that surround the hair bulb while Smad6 was also detected in the epithelial outer root sheath. Although their general expression was observed, Vimentin and Smad6 were not identified as signature genes for DP and matrix cells respectively by Rendl et al [36]. The results here (Figures 6 and 7) demonstrate their expression in distinct compartments relating to follicle epithelial (Smad6) and background mesenchymal (Vim) cells. The staining results and the 2-population model support the hypothesis that the computationally identified expanding population was associated to follicle epithelial cells. Currently there is no direct evidence of Smad6 in hair cycle regulation; however, Smad6 is a well-known negative regulator of the Tgfβ signaling pathway [53], [54]. Given its role in regulating Tgfβ signaling, as well as its proximity to MX cell markers, Smad6 may be an important candidate for future study in hair cycle regulation. We also noted the periodic expression of bone morphogenetic protein (BMP) genes, which have been documented as important regulators of skin and hair development. Four BMP genes showed periodic expression as LFOs: BMP 1 was found in cluster 1 (dermal papilla-associated) while BMPs 2K, 8a, and 7 were expressed in cluster 2 (matrix-associated). Other BMPs such as 2 and 4 were present in the original data, but the expression data contained too much variability to survive the FDR cutoff. The complete lists of LFO genes that matched the background and expanding population clusters, along with statistical metrics, is in Supplementary File S3.
The coupled oscillator model suggests that out-of-phase clustering is maintained by positive and negative coupling. The two population model indicates that these clusters are associated to specific cell populations. Taken together this is similar to negative feedback, for example the expanding population may drive the background population to produce an inhibitory signal, such as apoptosis, that in turn depletes the expanding population. However, if the background population is static, how is it contributing to such a control loop? For example, when the expanding population is relatively high, one would expect an increase in the expression of the inhibitory genes from the background to drive down the expanding population. One reasonable possibility is that expression changes were occurring within the background population. On average we found that the assumption of static intracellular expression was reasonable enough to estimate population dynamics (recall Figure 4); however, many individual expression profiles were poorly described by this assumption (recall Figure 5A). It is possible that these poorly described genes were undergoing intracellular changes, and could be responsible for the physical communication of inhibitory signals from the static background to the expanding population.
We investigated the possibility that inhibitory signaling genes may be in the DP enriched group identified as LFO cluster 2, but not well described by the static intracellular expression model. We expect such signaling genes to display an increased expression 14 to 16 days after morphogenesis, near the on-set of catagen and before the sharp decline in the expanding population (Figure 4). Using this criterion, we identified 88 expression signals (relating to 74 unique genes; see Supplementary File S4). We observed that these expression signals, on average, are consistent with population driven changes until near catagen on-set, where they begin to increase more than what was explained by static intracellular expression assumed in the 2-population model (Supplementary Figure S10). Of these genes, 50 were annotated as extracellular genes which yields and enrichment p-value of 7.46E-18, improved enrichment over cluster 2 with a p-value of 3.63E-8. For a full list of significant enrichment categories see Supplementary File S5. Interestingly, this relatively short list includes Tgfβ2, which is currently thought to be one of the signaling molecules produced in DP cells to initiate apoptosis in hair epithelial cells at catagen on-set [7].
Given the observed expression signal, membership in DP enriched cluster 2, high enrichment for extracellular genes and inclusion of Tgfβ2, this list may contain potential targets for molecules that communicate an inhibitory signal from the DP to proliferating hair epithelial cells, closing a negative feedback loop. Obviously further experiments will be required to test this hypothesis; however, it does provide a starting point for future validation of the conclusions drawn above and, perhaps, even those identified in the model of Al-Nuaimi et al. [10].
Although our approach provides novel insights and genes associated to the hair follicle, we also recognize that there are several limitations to this study. We studied microarray-derived RNA expression data from developing mouse skin that included non-periodic as well as periodic gene expression patterns. Due to the cyclic nature of the hair cycle, we chose to focus our study on the latter. We emphasize that our study would overlook important regulators of the hair cycle if they were not periodically expressed. Next, we only considered a single time course, experimental study (Lin et al, 2009), which obviously limits the data and conditions available to us (sparse sampling, limited technical replicate measurements and inclusion of only early cycles), and could lead to some important genes and cycle dynamics being excluded from further analysis. Furthermore, biological and technical variation, along with typical tradeoffs in sensitivity versus specificity associated with parameter selection, such as p-value thresholds, will further limit statistical detection of important mRNAs or expression patterns. Due to concerns of batch effects, we did not choose to combine additional datasets from other experimental studies to offset these issues. Instead, we chose to limit the scope of our investigation to describing a specific, but prevalent, dynamic pattern observed in the data. Again, by limiting the scope in this manner, it is likely that some important hair cycle regulators were overlooked. For example, BMP 2 and 4 have been shown to influence anagen initiation [48], [49], but due to sample variability these patterns fell outside the range of detection for this study. However, our investigation did encompass over 3,000 unique genes, where follow-up dynamic, enrichment and experimental analyses all suggested a possible role in the hair follicle and cycle dynamics.
Study design also limited us to time course over a single cycle of follicle synchronized hair growth. We were not able to test if the identified expression patterns, specifically synchronized out-of-phase gene expression, continued for additional cycles. This is a typical experimental limitation due to the loss of follicle synchronization as animals mature, at latter stages of hair growth. This is a different concept from the synchronization describing gene expression patterns. One might still expect that similar gene expression patterns within individual follicles, and the surrounding microenvironment, continue with additional cycles; however, without single follicle tracking, we cannot confirm this. Furthermore, our dynamic coupled oscillator model would never predict such follicle-level de-synchronization, as we did not include any mechanisms for cycle variability nor did we include the concept of individual follicles. Accounting for stochastic variation and spatially modeling individual follicles that are themselves coupled, represents an additional level of complexity that may more accurately model the rich dynamics of the hair system, but was not considered in this study.
Finally, we modeled gene expression from whole skin, since isolation of hair follicles prior to gene expression profiling is resource intensive and was beyond the scope of our work. In doing so, we relied on the 2-population model, cell type specific enrichment (based on experimentally purified cell populations [36], [37]) and experimental imaging to make associations between computationally derived gene groupings and distinct biological populations. While these results were very encouraging, we would like to emphasizes that the computationally derived populations do not represent a specific, predefined cell type. Supplementary Figure S7A shows that the majority of the signature genes were not identified, and the populations contained signature genes from multiple cell types. However, we do note that even experimentally derived gene signatures also demonstrate overlap (see Supplementary Figure S7B). Furthermore, it likely that cell types not investigated by enrichment also contributed to the estimated populations, such as endothelial cells involved in capillary network remodeling, adipocytes or immune cells that may have active roles in hair growth. Ultimately, we identified many associations between the computationally derived gene groupings and distinct, hair cycle relevant cell populations, but we cannot exclude that gene expression unrelated to the hair follicle or hair growth may have contributed to both false positives and negatives.
In this study, we focused on potentially periodic gene expression patterns in whole skin that changed on the same time-scale as cyclical hair growth. We identified two distinct clusters consistent with synchronized, out-of-phase gene expression (Figure 1 and 2). Through nonlinear-dynamic analysis, we proved that a simple, coupled oscillator model was mathematically sufficient to recapitulate this observed synchronization, and that the coupling scheme involves both positive and negative coupling (Figure 3A). We go on to show that these clusters can be associated with either static or expanding cell populations (Figures 4 and 5), and that the size of the expanding population, determined by gene expression data, was consistent with the population dynamics of follicle epithelial cells (Figure 4). Follow-up experimental and enrichment analyses indicated that the corresponding genes (provided as Supplementary Information File S3) were strongly associated with biologically distinct cell-types, such as MX or DP cells (Figures 6 and 7, Supplementary Table S1). Taken together, these results were consistent with regulatory mechanisms involving negative feedback from background mesenchymal cells to the expanding epithelial cells (see summary Fig 8). Finally, we identified a subset of genes that could potentially communicate the inhibitory signal to the follicle (provided as Supplementary Information File S5). Other aspects of the study provided interesting, but speculative, insights on possible alternative hair cycle states that are similar to those of miniaturized hair follicles (Figure 3B).
The model of hair cycle presented here suggests some role for proliferating follicle epithelial cells to be regulated by a systems-level inhibitory response, likely to emanate from the DP. Conceptually, regulated apoptosis could be one way in which a large number of genes from the same general population of cells are inhibited by a second population. This mechanism has also been explored in a kinetic model of hair cycle which shows that negative feedback via DP regulated apoptosis is sufficient to account for the cyclical nature of hair growth [10]. Because mRNA expression data underlies the model presented here, we were able to advance this idea a step further and identify candidate signaling proteins based on the dynamics of the gene expression. Although experimental confirmation of these candidates was beyond the scope of the investigation, we note that the list was highly enriched for extracellular proteins and with only 74 genes we were still able to identify Tgfβ2 as a possible candidate. For clarity, we emphasize here that negative coupling is not the only aspect of our model, which also includes intrinsic oscillations and positive coupling. In fact, our model predicts that reduction of the DP associated oscillators actually results in a shortened cycle (see Figure 3B). Furthermore, the true physical mechanisms underlying the hair cycle are likely far more complex, and studies show that anagen length and hair size actually decrease upon depletion of DP cells [55]. Both our computational and existing experimental results suggest that inhibitory regulation of MX derived cells cannot describe all aspects of the hair cycle; however, it is likely to play an important role, with one possibility being regulated apoptosis [7], [56], and the genes we have identified here could help guide follow-up experiments.
Finally, the most intriguing aspect of this study was the predicted proximity of the hair system to a critical phase transition (Figure 3B). In the observed dynamics the hair system was in a stable -state, in which the proportion of positive and negative oscillators may vary, or at least increase, without affecting the overall period. However, a decrease in this proportion, for example a reduction in positively coupled oscillators, would throw the system in to a travelling wave state, where the two clusters drive each other into higher frequency oscillations. If this behaviour can be validated, it would have important biological implications. Biologically the model suggests that there is a systems-level regulation within and between genes related to follicle epithelial and background mesenchymal cell populations, which represents a balance of negative and positive driving forces. Loss of regulation of the DP population, for example, would mean a decrease in negative forces acting on the epithelial population and would throw the system into a fundamentally different state. This new state would be characterized by a drastically reduced period of expression, similar to hair miniaturization resulting from androgenetic alopecia. Unfortunately, testing of this hypothesis could prove difficult. Removal of several genes via genetic knockout would have consequences not accounted for here; however, inhibition of the physical signaling between the DP and MX derived populations may be more tractable. Excitingly, a recent study by Rompolas et al. [57] introduced new methodology to explore such interactions of the hair follicle in live mice using laser ablation of DP cells. The approach not only allowed for the elimination of a specific cell population, but also removed technical complications associated with follicle synchronization, as individual follicles could be monitored over time. It would be of future interest to see if this methodology could be modified to properly test the predictions presented herein.
To build on the insights of the present framework, we can offer several additional directions for future work. For example, there are several possible advancements to the coupled oscillator model, including: analytically solving the existing coupling scheme for excitable elements (similar to [58]) opposed to oscillators would better capture the pulsing behavior of gene expression and hair growth; integrating noise, known to have a major impact of synchronization [59], could help capture both expression and cycle variability; and coupling together multiple coupled systems could capture associations and variation between follicles. Experimentally, new time course data could identify new behaviors. Performing a single extensive time course from morphology to end of second round of hair growth would capture anticipated transients and determine a proper time scale for synchronization. Increasing the sampling frequency could identify high frequency oscillators and perhaps provide a means to couple circadian rhythm to the current system. Given that we systematically identified two clusters from whole skin data, a beneficial advancement would be to directly collect data from follicle specific cell types, such as MX and DP cells. Producing a time course using purification techniques similar to Rendl et al. would be the most direct way to prevent confounding expression signals from non-relevant cell types and provide a specific interpretation of modeled populations. In our experience, any additional time courses would need to have, at a minimum, a sampling rate 3 times that of the desired frequency for identification. However, we would strongly suggest doubling the number of points in the time course and including 3 replicates at each point for statistical and modeling considerations. Our hope is, that by incremental advancement, the framework provided here can be used to bridge the gap between high-throughput measurement data and systems-level properties of hair cycling.
This study was performed in strict accordance US Animal Welfare Regulations at an AAALAC accredited site. The research protocol was approved by the Institutional Animal Care and Use Committee of Procter & Gamble. Every effort was made to minimize suffering of all animal subjects.
We employed data originally generated in Lin et al. [11]. The authors profiled both second, naturally synchronized and depletion-induced hair cycles. Samples were collected from the upper-mid region of C57Bl/6 mice and analyzed using Affymetrix Mouse Genome 430 2.0. For additional experimental details please refer to the original article.
The raw intensity data was collected from the NCBI Gene Expression Omnibus as accession number GSE11186. The data was uploaded in CEL file format and preprocessed for quality control. Sample GSM281802 was removed based on suspected RNA degradation, a mean correlation coefficient less than 0.95, multiple outliers determined by the MA plot, and high background error and variation determined by RMA. The remaining samples were summarized and normalized using the RMA function from the Bioconductor ‘affy’ library in R, applying quantile normalization and RMA background correction from affy version 1.1. A log base 2 transform was applied to the expression data for all subsequent analysis except for the 2-population mixture model. An R script containing the general QC and RMA methods used can be found in the Supplementary File S6.
The two experimental conditions corresponding to the natural and induced hair cycles, were combined into a single time course as suggested by the original authors. The five sampled time points for the induced cycle, {3,5,8,12,17} days, were mapped to time points in the natural cycle, {24,25,27,29,37} days, based on the morphology of skin sections. Hair morphogenesis during synchronous growth was established by histologic criteria [11]. Combining samples with similar morphologies, and therefore in similar hair cycle phases, we expect to limit the variability of gene expression that is associated to the hair cycle phenotype.
We visualized the expression data using standard heat maps. Multiple values at a given time point were averaged. For visualizing expression values of different scales in a single image it was necessary to normalize the data. Two different normalization schemes were used. Figure 1 focuses on relative levels of periodically expressed genes, here we used a 0–1 normalization: . The associated time scale is ‘days postnatal’ and corresponds to the natural second cycle; the induced cycle was matched to time points as discussed above. In Figure 5 and the 2-population mixture model, we applied a fraction of max normalization, , to capture information of relative fold change in expression. The associated time scale was days from initiation to emphasize differences in the natural and induced cycle, which is assumed to begin after morphogenesis, which is ≈23 days, and depletion, respectively. The normalization schemes described here were for visualization purposes, and were not used in any statistical analysis. For all time series, we present a color bar to roughly indicate the corresponding phase of the hair cycle, the timing for the color bar was taken from [11].
We employed a mathematical description of coupled oscillators to study general features of the synchronization observed in the hair cycle data. We considered a modified version of the simple Kuramoto model [16] suggested by Hong and Strogatz [33]. Hong and Strogatz show that a two group model, one positive and one negative, is sufficient to spontaneously produce two out-of-phase clusters. Oscillators which are positively or negatively coupled will be drawn towards or pushed away from other oscillators on the unit circle, respectively. After some simplification and incorporation of EQ 9 the governing equations reduce to(10)where is the phase of the th oscillator, calculated by EQ 7, which is assigned to group , is the natural or intrinsic frequency for , is the coupling constant for group , and from EQ 9 where the subscript is dropped for simplicity. The dot denotes change with respect to time, and can be calculated using EQ 8. Recall that oscillators represent probesets with expression patterns identified as periodic. Here we assume two groups where and , , . Introducing these two groups to EQ 10 we have(11)where is the number of positively coupled oscillators. We note that EQ 11 is identical to EQ 2 presented in the Results and Discussion, and was included here only to maintain the continuity of the method descriptions. We also reemphasize several simplifications in this formulation. First, the model is a mean field approximation in which each individual oscillator is connected to all other oscillators through the order parameter, . This is derived from an assumed all to all connectivity, which is obviously not expected in a gene network; however, the mean field approximation works well if the effective coupling on the oscillators (or genes) is well described by an average of the individual couplings. Such models have successfully described high level properties of many large, complex systems including statistical mechanics (overview of several models [17]), economics [18], [19] and even social networks [20]. Second, we note that the variables here are considered independent, for example the model assumes that the proportion of oscillators can be varied without affecting other independent variables, such as the ‘intrinsic’ oscillations, . However, in reality removal of genes from the system will have an impact not captured in the model, such as an alteration or even cessation of the assumed ‘intrinsic’ oscillations. Finally, we emphasize that the coupling describes oscillator interactions and not necessarily the underlying driving force for oscillation, which is typically attributed to . With this level of abstraction and simplification it was not possible to describe most of the details of the hair cycle including mechanistic molecular connectivity; however, we were able to describe more general aspects of the system such as a stable, synchronized state.
To solve the system, we follow the original paper [33], and summarize the process here for the reader's convenience. We can reduce the model to a low dimensional system in terms of the first order parameters for each group. First, we consider a system where , we validate the use of this assumption later. Second, we assume the were randomly distributed via a Lorentzian probability distribution . Importantly, we note that use of a single frequency, opposed to a distribution, will not recapitulate the distributed phases observed in Figure 1C [67]. Here, we have moved to a rotating frame such that the mean of is zero; in our system this was done by subtracting the principal periodic component, radians per day. Finally, we can apply the ansatz of Ott and Antonsen [68] which yields(12)where and are the first order parameters (similar to EQ 9 with ) for the two oscillator groups related to positive and negative coupling, respectively; is the proportion of positively coupled oscillators and . The bar denotes complex conjugate. We note that EQ 12 is identical to EQ 3 presented in the Results and Discussion, and was included here only to maintain the continuity of the method descriptions. Using EQ 12 we can solve for the critical value of such that only the incoherent state is stable when and is, therefore, the lower bound for observing synchronization. We can also estimate other bifurcation points of the system, and . For more details see [33].
We solved for various properties of the hair cycle system using EQ 12 and the oscillator state variables solved for above. We assumed the observed period of the hair cycle system was at a quasi-steady-state, where the magnitude of the first order parameter is constant and the rate of change of the phase is also constant. This was demonstrated by observed in Figure 2A. Given the quasi-stead-state, we solved EQ 12 in a rotating frame described above, allowing us to set the left hand side to zero. We considered two possible configurations of assigning clusters (from Figure 1C) to positive or negative coupling groups. After coupling assignment, we calculated , , and from the data and solved EQ 12 for the unknown parameters and . Then and were used to solve for . We note that EQ 12 was solved by letting and lie on the real axis so they were equivalent to and , respectively. This assignment can be done, without loss of generality, for a quasi-steady-state, out-of-phase system. For the configuration with (clust1) = (+) and (clust2) = (−), we found that the system was unstable (Figure 3A purple dashed line). We calculated and which is not physically realizable. However, we found the opposite configuration, with (clust1) = (−) and (clust2) = (+), to be a stable solution with radians per day and (Figure 3A red solid line, recall actual data in Figure 2A red, see Supplementary File S2 for a simulation describing individual oscillators with these properties).
The bifurcation diagram was solved numerically (Figure 2B). We found the long time solutions to the system of ordinary differential equation (EQ 12) for various values of while holding all other variables constant. We found to be sufficient. We verified the assumption of by simulating the low dimensional system, EQ 12, and comparing that to simulations of the high dimensional, EQ 10, with (Supplementary Figure S13). Given noise, due to initial configurations, associated with the finite, high dimensional system, 100 iterations were calculated and the mean and standard deviation were reported. The two representations are sufficiently similar and have nearly identical steady-states. All numerical simulations were performed in Matlab [69] using ‘ode45’. All Matlab scripts necessary to solve for model variables and reproduce simulations found in figures and movies are available in the folder ‘meanField’ in Supplementary File S6.
Observations of two distinct gene expression clusters motivated us to explore possible relationships to different cell populations within the hair follicle. We considered the scenario in which observed expression changes are due to changes in relative cell population size as opposed to intracellular changes. In Supplementary Figure S3, we showed a simple example in which and had a high and low concentration in pop 1, respectively, and reciprocal concentrations in pop 2. While varying the size of pop 1, holding the pop 2 size constant and holding all internal concentration level constant, we can see that the observed concentration of and change, and do so in an out-of-phase manner. To explore this in our system we wished to reverse the process and estimate the relative sizes and intracellular expressions given the observed mixed expression. To achieve this we applied in silico microdissection [35].
Briefly, in silico microdissection works by applying a simple linear model of mixed samples(13)where is the observed expression of gene in mixed sample ; indicates the intracellular expression in populations and ; and is the cell fraction of in mixed sample . Given the cell fraction, , for each sample we can solve EQ 13 for the internal concentration in the two populations. In this situation each gene is an independent problem, each solved via simple linear regression over all mixed samples . This is an overdetermined system if the number of populations considered is less then the number of samples. We can also recast the problem to solve for given the internal concentration for each population. In this situation each mixed sample is now an independent problem, each a constrained linear problem over all genes . Here is constrained between 0 and 1, the problem is convex and can be solved efficiently. An iterative process, similar to expectation-maximization, can be used to solve for both and the y's simultaneously.
We consider a model of an expanding cell population mixed with a constant background population. We treated the hair cycle expression chips as independent mixed samples each with possibly different cell fractions. No information of cycle type or time was needed, nor any strategy for combining samples as in the previous periodic identification. For later comparisons of the induced and natural cycle, we set the time relative to cycle initiation, which we assumed to be after morphogenesis (postnatal day 23) or after depletion. This time scale was only used for graphical representation, and was not used in any calculations. A linearly increasing function from 0 to 1 was used as the initial conditions for , the cell fraction of the expanding population. We also included the expression of all 45k+ probesets without the log2 transform, as suggested in [35], all other preprocessing was the same. Given this, over determined, model we implemented the above iteration strategy to solve for and the relative size (Figure 4) and the internal expression (Figure 5C). We note here that it is not reasonable to expect all intracellular expression to remain constant over the whole hair cycle; however, if the relative population change is large, as seen here, compared to the intracellular expression change for many genes then it is a reasonable assumption.
While calculating the internal expression for the two populations, we also estimated the corresponding standard error using common methods associated with linear regression. The standard error was used to produce a t-statistic and p-value for each probeset, which indicated the extent to which a gene was differently expressed between the two populations (Figure 5B). The probesets for the LFO subgroup with a t-statistic above a 0.10 false discovery rate were assigned to the population in which they were predicted to have higher expression. We note that nearly all of the LFOs met the statistical requirements, 3975 out of 3988. This was equivalent to separating the probesets into two groups based on the estimated t-statistic, and was found to be equivalent to separation of LFOs by phase (Supplementary Figure S6).
We also considered a computational negative control. In the above analysis, we inherently assumed that expression is related to time, after morphogenesis or after depletion. Our population analysis allowed us to then associate expression to relative population size, and therefore, plot relative population size as a function of time. Here we considered a negative control, that expression is random with respect to time, and not related to hair cycle. To achieve this, we randomly permuted (or shuffled) the time courses for each probeset. For a proper comparison the depletion and naturally induced time courses were not intermixed, and kept separate. After permutation, we employed the exact same analysis and plotting procedure as above (used to produce Fig. 4 and 5). The results are shown in Supplementary Figures S4 and S5. In our negative control, we could see that there was no indication of expansion and depletion corresponding with anagen and catagen, respectively (Supplementary Figure S4). Furthermore, we did not observe improved statistics, as in coefficient of determination or the t-statistic (Supplementary Figure S5). We considered this sufficient evidence that the negative control produced only random population changes with respect to time, as expected.
All code for estimating the two populations was written in Matlab and implemented by the scheme discussed in [35]. The constraint linear problem was solved using Matlab Optimization Toolbox function ‘lsqlin’ with default options. Standard errors were estimated using the Matlab Statistics Toolbox function ‘regstats’. Matlab scripts for running all analysis described above and for generating the data in associated figures are available in the ‘2pop’ folder in Supplementary File S6.
We used the online tool DAVID 6.7 to perform basic enrichment analysis [70]. We used the full mouse genome as the background gene set. For biological process enrichment, we used the Gene Ontology annotations under ‘GOTERM_BP_FAT’ and for pathway enrichment, we used KEGG annotations under ‘KEGG_PATHWAY’. The enrichment was done using different target sets indicated in the main text.
We used the Normalized Google Distance (NGD) to estimate enrichment of genes related to hair. The NGD is a semantic similarity measure [71], which for two terms and is defined as(14)where and are the numbers of pages the terms and are found in, respectively, is the number of co-occurrences and is the total number of pages considered. We note that the more frequent and co-occur the lower NGD will be, and that if then NGD = inf. Here we applied the search to all abstracts in the NCBI PubMed database. We calculated the NGD between the term ‘hair’ and all mouse gene symbols. All genes with an NGD to ‘hair’ of less than 1.0 were used as the final set of Hair related genes. We applied a standard enrichment test using a hypergeometric distribution. The target set was the list of all periodic genes and the background set was the full mouse genome.
The threshold of 1.0 was chosen as it is the NGD of the expected value for independent or unrelated terms. Briefly, given a set number of occurrences for a term then the probability of finding term in pages is . Assuming that two terms, and are independent, we have . Plugging these values into EQ 14 we find that NGD = 1.
A cell type enrichment analysis was used to link model populations to specific cell types. Two existing studies, Rendl et al. [36] and Greco et al. [37], dissected hair follicles into specific, predefined cell types relating the the hair follicle: dermal papilla, melanocytes, matrix cells and outer root sheath cells from [36] and Bulge cells from [37]. Using mRNA microarray data the studies defined gene signatures for each population as sets of probesets and corresponding genes with expression nearly exclusive to a particular cell type. Using these signatures to annotate probesets with a particular cell type, we applied standard enrichment tests using a hypergeometric distribution. The target set was the list of genes determined to be in the expanding or background population (seen in Figure 5C) and the background set was the full mouse genome.
Male mice, C57Bl/6 (Charles River Laboratories, Portage, MI) at 62-66 days of age, in the telogen phase of the hair cycle [5] are shaved in the dorsal area (area of 1.5 inches×2 inches) followed by treatment with Nair (Church & Dwight Co.) to the same area for 1 hour before washing off to initiate the hair cycle. Nair depletion induces a similar response as wax by damaging the hair shaft to start a homogenous re-entry into anagen [72]. Mice were collected at various timepoints after induction treatment.
ISH was performed using QuantiGene ViewRNA protocols (Affymetrix, Santa Clara, CA). Five µm formalin fixed paraffin embedded (FFPE) sections were cut, fixed in 10% formaldehyde overnight at room temperature (RT) and digested with proteinase K (Affymetrix, Santa Clara, CA). Sections were hybridized for 3 hours at 40°C with custom designed QuantiGene ViewRNA probes against specific target genes and the positive control genes used were Fgf7 for dermal papilla cells and Foxn1 for matrix cells (Affymetrix, Santa Clara, CA).
Bound probes were then amplified per protocol from Affymetrix using PreAmp and Amp molecules. Multiple Label Probe oligonucleotides conjugated to alkaline phosphatase (LP-AP Type 1) were then added and Fast Red Substrate was used to produce signal (red dots, Cy3 fluorescence). For two color assays, an LP-AP type 6 probe was used with Fast Blue substrate (blue dots, Cy5 fluorescence) followed by LP-AP type 1 probe with Fast Red Substrate (red dots, Cy3 fluorescence) to produce a dual colorimetric and fluorescent signal. The probes sets used for ISH are described in Table S2. Slides were then counterstained with hematoxylin. Serial sections were also subjected to hematoxylin and eosin staining per standard methods to confirm the identity of cells in the region of ISH signals. Images were collected using a Deltavision microscope (Applied Precision), and the fluorescent images were created using softWoRx 5.0 (Applied Precision).
The in situ hybridization assay in this study utilizes branched DNA (bDNA) technology, which offers near single copy mRNA sensitivity in individual cells. The bDNA assay uses a sandwich-based hybridization method which relies on bDNA molecules to amplify the signal from target mRNA molecules. Each probe set hybridizing to a single target contains 20 oligonucleotides pairs. This was followed by sequential hybridization with the final conjugation of a fluorescent dye. Thus, each fully assembled signal amplification ‘tree’ has 400 binding sites for each labeled probe. Finally, when all target specific oligonucleotides in the probe set have bound to the target mRNA transcript, the resulting amplification of signal approaches 8000-fold (20 oligonucleotides times 400 binding sites = 8000 fold).
Immunofluorescence staining was performed on fresh frozen cryosections (10 µM thickness) or FFPE sections (5 µM thickness) of mouse skin to visualize the hair follicles present at Day 0 and Day 16. Cryosections were stored at −80°C until use. Cryosections were dried for 30 min at room temperature and fixed by immersion in ice-cold acetone for 10 mins. Cryosections were then air-dried for 5 mins and washed three times with PBS. For FFPE sections, deparaffinzation was performed using xylene and series of alcohol changes. Antigen retrieval for performed using 0.05% trypsin at 37°C for 20 mins. Both cryosections and FFPE sections underwent the same treatment after this step. The sections were blocked for 1 hour using normal donkey serum (NDS, dilution 1∶10; Sigma-Adhrich) in PBS. Sections were then incubated with specific primary antibodies (as described in Table S3) in 1∶5 dilution of blocking solution for overnight at 4°C and then washed three times with PBS. Next, sections were incubated with Alexafluor-488-conjugated donkey anti-rabbit and Alexafluor-594-conjugated donkey anti-goat antibodies (Vector Laboratories, 1∶500) for 1h at 37°C, washed three times with PBS. Final wash was performed with DAPI and the sections were mounted using Flurosav (Calbiochem). Images were collected using a Deltavision microscope (Applied Precision), and the fluorescent images were created using softWoRx 5.0 (Applied Precision). Foxn1 was used as a positive control for matrix cells based on previous literature [50]. Morphology, as determined by brightfield and DAPI staining, was used to identify DP localization. We considered Fgf7 [42] as a positive control for DP localization; however, all antibodies tested showed significant non-specific staining.
Total RNA was extracted from mouse skin samples at days 6, 16, 23, 29, 38, 44 and 59 using Agilent's Total RNA isolation mini kit (Agilent Technologies). Reverse transcription reaction was performed with 500 ng of total RNA using the Superscript VILO cDNA synthesis kit (Life technologies). A 1∶25 dilution of cDNA was used in the QRT PCR reaction. QRT-PCR was carried out in a 10 µl reaction mixture with gene-specific primers and β-Actin using RT2 SYBR Green ROX qPCR Mastermix (Qiagen). The PCR conditions were 95°C for 10 min, and 40 cycles of 95°C for 15 s, 59°C for 30 s, 72°C for 30 s on the ABI HT 7600 PCR instrument. All samples were assayed in quadruplicate. The differences in expression of specific gene product were evaluated using a relative quantification method where the expression of specific gene was normalized to the level of β-Actin. Primer sequences available in Supplementary Table S4.
|
10.1371/journal.pgen.1006816 | Lack of the PGA exopolysaccharide in Salmonella as an adaptive trait for survival in the host | Many bacteria build biofilm matrices using a conserved exopolysaccharide named PGA or PNAG (poly-β-1,6-N-acetyl-D-glucosamine). Interestingly, while E. coli and other members of the family Enterobacteriaceae encode the pgaABCD operon responsible for PGA synthesis, Salmonella lacks it. The evolutionary force driving this difference remains to be determined. Here, we report that Salmonella lost the pgaABCD operon after the divergence of Salmonella and Citrobacter clades, and previous to the diversification of the currently sequenced Salmonella strains. Reconstitution of the PGA machinery endows Salmonella with the capacity to produce PGA in a cyclic dimeric GMP (c-di-GMP) dependent manner. Outside the host, the PGA polysaccharide does not seem to provide any significant benefit to Salmonella: resistance against chlorine treatment, ultraviolet light irradiation, heavy metal stress and phage infection remained the same as in a strain producing cellulose, the main biofilm exopolysaccharide naturally produced by Salmonella. In contrast, PGA production proved to be deleterious to Salmonella survival inside the host, since it increased susceptibility to bile salts and oxidative stress, and hindered the capacity of S. Enteritidis to survive inside macrophages and to colonize extraintestinal organs, including the gallbladder. Altogether, our observations indicate that PGA is an antivirulence factor whose loss may have been a necessary event during Salmonella speciation to permit survival inside the host.
| During bacterial evolution, specific traits that optimize the organism’s fitness are selected. The production of exopolysaccharides is widespread among bacteria in which they play a protective shielding role as main constituents of biofilms. In contrast to closely related siblings, Salmonella has lost the capacity to produce the exopolysaccharide PGA. Our study reveals that Salmonella lost pga genes, and that the driving force for such a loss may have been the detrimental impact that PGA has during Salmonella invasion of internal organs where it augments the susceptibility to bile salts and oxygen radicals, reducing bacterial survival inside macrophages and rendering Salmonella avirulent. These results suggest that gene-loss has played an important role during Salmonella evolution.
| Escherichia coli and Salmonella enterica are the two core species of the family Enterobacteriaceae, that constitutes a diverse group of bacteria that generally inhabit the gastrointestinal tract of animals. Although these two species are closely related, E. coli comprises commensal bacteria that do not normally cause disease, with the exception of certain pathogenic strains, whereas all members of S. enterica are considered pathogenic. Hence, an intriguing issue regarding bacterial evolution is the identification of determinants that make Salmonella able to establish parasitic interactions but enable E. coli to establish beneficial interactions with the human host. In this regard, it is believed that a combination of different genetic factors accounts for such a difference in virulence: first, Salmonella harbor virulence genes that are not present in E. coli; second, Salmonella may have lost genes from the ancestral core genome that if present, would diminish its pathogenic potential; third, E. coli may carry a virulence suppressor gene(s) that interferes with the synthesis and/or stability of a virulence protein(s); and fourth, Salmonella and E. coli may differ in the regulation of cellular factors important for survival in the host [1–3].
An intriguing difference between Salmonella and E. coli that might account for their distinctive lifestyles as regards the human host is the exopolysaccharide that each species uses to build the biofilm matrix. Bacteria spend most of their lives inside a biofilm surrounded by a highly hydrated layer that provides protection against desiccation, diffusion of antibiotics, toxic metal ions and other compounds, predation by protozoans and the host immune system, amongst others [4,5]. Diversity in biofilm exopolysaccharides composition is high, with some bacterial species being able to produce different types depending on the environmental conditions [6,7]. In parallel to such high diversity and for reasons that remain unknown, a wide range of phylogenetically distant bacteria make use of the same exopolysaccharide to embed themselves inside a biofilm. One example of a “universal” exopolysaccharide is cellulose, composed of β(1–4)-linked D-glucose units, used by a wide variety of bacteria, including both E. coli and Salmonella [8–10], as a significant biofilm matrix component. Another example corresponds to a homopolysaccharide composed of N-acetylglucosamine with β(1–6) glycosidic linkage [11]. Production of this exopolysaccharide was firstly described in Staphylococcus epidermidis and S. aureus where it was referred to as PIA/PNAG [12–14]. Later on, the synthesis of a similar exopolysaccharide was also reported in E. coli where it was named as PGA [15], and also in Acinetobacter baumannii [16], Klebsiella pneumoniae [17], Bordetella bronchiseptica and B. pertussis [18,19], Actinobacillus pleuropneumoniae [20], Yersinia pestis [21], Burkholderia species [22] and Bacillus subtilis [23]. In these bacteria, several functions have been ascribed to PGA such as surface attachment, intercellular adhesion, biofilm formation, epithelial cell attachment, and resistance to antibiotics, antimicrobial peptides and human PMNs [16,19,22,24–29]. In E. coli, the production, modification, and export of PGA requires the machinery encoded by the pgaABCD operon [15]. PgaA and PgaB are needed for poly-GlcNAc export and PgaC and PgaD are necessary for poly-GlcNAc synthesis [30–33]. As it generally occurs for bacterial exopolysaccharides, PGA synthesis is allosterically activated by the second messenger bis-(3’-5’)-cyclic dimeric GMP (c-di-GMP) [30,34,35]. Strikingly, Salmonella lacks the pgaABCD operon and any identifiable genetic loci similar to pga required for PGA synthesis.
Here, we pursue the reasons that explain why E. coli and Salmonella differ in their capacity to produce the PGA exopolysaccharide. We provide evidence that production of PGA reduces Salmonella resistance against bile salts and its capacity to survive inside macrophages, completely impairing the infection cycle and rendering Salmonella avirulent. Together, these observations highlight the relevance of gene loss in the adaptation to novel pathogenic niches and define the loss of the PGA exopolysaccharide as a landmark event during Salmonella speciation.
To investigate whether the presence of the pgaABCD operon in Escherichia and its absence in Salmonella is due to a lineage-specific acquisition in Escherichia or to a loss in Salmonella, we performed different comparative and phylogenetic analyses (see Materials and Methods). Analysis of the genomic context of E. coli PgaA protein in the STRING database [36] correctly identified the presence of four genes in the pgaABCD operon as significantly associated using exclusively gene neighborhood and gene co-occurrence information. The gene cluster, often only presenting the three upstream genes, is widespread among Enterobacteriaceae, being present in 22 species of the 83 available in the database. Besides Escherichia, the genera with the cluster include Klebsiella, Pectobacterium, Yersinia, Citrobacter, and Enterobacter, among others. Analyses of the presence/absence of the genes revealed a similar pattern, confirming the absence of the genes in Salmonella species and other genera. Importantly, both analyses revealed a patchy presence/absence pattern, including many recent apparent losses within some genera such as Citrobacter or Escherichia. We then reconstructed individual phylogenies in each of the genes in the cluster by aligning the top 500 hits of a blastP search in NCBI nr database, after setting a filter to exclude sequences assigned to E. coli. All the top hits belonged to related species of Enterobacteriaceae excluding the possibility of recent, independent transfers of the cluster from a non-Enterobacteriaceae species. Maximum likelihood phylogenies of the four genes produced roughly similar topological arrangements of the included taxa (schematically depicted in Fig 1). We performed a similar analysis with phoH, the gene located in the vicinity of the cluster, encoding a protein with a nucleoside triphosphate hydrolase domain. This gene has a broader distribution, present in 68 of the 83 taxa, a pattern suggesting a vertical inheritance with few independent losses. Importantly, however, for the shared species, the phylogenies of phoH and that of the four genes in the pgaABCD cluster showed an overall similarity (Fig 1, S1 Fig and S1 Dataset). This indicates that the five genes followed a similar evolutionary history, with the exception of differential loss of genes in alternative lineages. This topology was largely congruent with the species tree for Enterobacteriaceae provided in the PATRIC database, which is based on the analysis of several shared genes [37,38], with the notable exception of the position of Yersinia or Serratia strains. Previous studies have shown that losses are more frequent than lateral transfer in the evolution of prokaryotic genomes [39], and lateral transfer would generate discordance between gene trees [40]. Hence, our results point to an overall dominance of vertical inheritance and differential gene loss in the evolution of this gene cluster within Enterobacteriaceae. Considering this scenario, the pgaABCD cluster was lost somewhere after the divergence of Salmonella and Citrobacter clades, and previous to the diversification of the currently sequenced Salmonella strains.
If Salmonella inability to synthesize PGA is exclusively due to loss of the pgaABCD operon, complementation with pgaABCD should be sufficient to restore PGA production. To test this hypothesis, we transformed a S. Enteritidis wild type strain with plasmid pJET::pga carrying the pgaABCD operon of E. coli MG1655 under the control of its own promoter, and analyzed PGA synthesis upon growth under Salmonella biofilm forming conditions (incubation in LB broth, at room temperature, without shaking) using a dot blot assay and an anti-PIA/PNAG antiserum. As expected, PGA was not detected in cell extracts of the wild type strain whereas WT pJET::pga produced PGA and accumulated it throughout the incubation time (Fig 2A and S2A Fig).
We next examined if, as it happens in E. coli [30], PGA production in Salmonella is also dependent on c-di-GMP. To do so, we firstly complemented S. Enteritidis ΔXII with the pgaABCD operon. S. Enteritidis ΔXII is a multiple mutant, derivative of the wild type strain, carrying mutations in all twelve genes encoding GGDEF domain proteins (putative c-di-GMP synthases) and thus incapable of synthesizing c-di-GMP [41,42]. The dot-blot assay showed that ΔXII pJET::pga was unable to produce PGA, confirming that c-di-GMP is indeed essential for PGA production in Salmonella (Fig 2A). Secondly, we constructed a strain in which the adrA gene of Salmonella, which encodes a c-di-GMP synthase, is under the control of a constitutive promoter. This strain (WT PcL::adrA) constitutively produces high levels of c-di-GMP. Upon transformation with pJET::pga, this strain produced higher PGA levels than the wild type strain (Fig 2A and S2A Fig), showing that heterologous PGA synthesis in Salmonella is commensurate to cellular c-di-GMP levels. Finally, and in order to identify the source of c-di-GMP in WT pJET::pga that triggers PGA production, we used a collection of twelve strains, derivatives of ΔXII, each of which contained the chromosomal copy of a single gene encoding a GGDEF domain protein in the original wild type genomic location [41,42]. The analysis of cell extracts of each strain complemented with pJET::pga showed that five GGDEF domain proteins, namely AdrA, YedQ, YegE, YfiN and SEN4316, when individually present in the chromosome of the cell, were able to elicit c-di-GMP dependent PGA synthesis (S2B Fig). Overall, these results showed that heterologous pgaABCD expression is sufficient to restore Salmonella capacity to synthesize PGA and that this synthesis is dependent on c-di-GMP levels that are provided as a pool by different Salmonella c-di-GMP synthases.
In staphylococcal cells, production of PGA can be visualized as a ring of cells adhered to the glass wall at the air–liquid interface, when bacteria are incubated in a glass tube under shaking conditions [43]. To investigate whether Salmonella is likewise able to build a PGA mediated biofilm, we analyzed biofilm formation by WT pJET::pga and WT PcL::adrA pJET::pga after incubation in LB broth, at 28°C for 16 hours under shaking conditions. Only the second strain, which produces constitutive and high levels of c-di-GMP, produced a visible ring of bacteria adhered to the glass wall (Fig 2B). Structure of this PGA based biofilm was then compared with the natural cellulose based biofilm formed by Salmonella using scanning electron microscopy (Fig 2C). To do so, we used a cellulose overexpressing strain (WT PcL::adrA), a PGA positive and cellulose minus strain (ΔbcsA PcL::adrA pJET::pga) and a control strain that produces neither cellulose nor PGA (ΔbcsA PcL::adrA). In the case of the PGA dependent biofilm, cells were tangled up in an abundant extracellular matrix mesh that interconnected the bacteria. Furthermore, spherical, knob-like structures were evident on the bacterial cell surface. These knob-like structures have already been described in PGA (PIA/PNAG) related biofilms of E. coli, Yersinia pestis and Staphylococcus epidermidis [44–46]. On the other hand, bacteria inside a cellulose based biofilm were covered by a sheet-like material [47] that totally encased bacteria and that appeared more compact and structured than the PGA biofilm. To further investigate the differences between both types of biofilms, macrocolony biofilms were grown on LB agar plates (S3A Fig) and a water-droplet analysis of colony hydrophobicity was performed [48]. Results showed that a cellulose mediated biofilm is highly hydrophobic, whereas a PGA based biofilm exhibits intermediate hydrophobicity compared with the non-biofilm producing strain, ΔbcsA PcL::adrA (S3B Fig). Collectively, these findings showed that heterologous PGA expression alongside high c-di-GMP levels enable Salmonella to build a PGA mediated biofilm that greatly differs at the structural level from the natural cellulose based biofilm.
Biofilm exopolysaccharides provide protection from the external environment. Thus, a consequence of PGA loss might be a reduction in Salmonella resistance to environmental threats, unless another compound assumed such a function. To test this hypothesis, we compared the resistance provided by PGA and cellulose to several environmental stresses. Since it has already been described that cellulose mediates chlorine survival of Salmonella and other bacteria [10,49,50], we first analyzed the susceptibility of macrocolony biofilms formed by the cellulose-positive strain (WT PcL::adrA) and the PGA-positive cellulose-negative strain (ΔbcsA PcL::adrA pJET::pga) to chlorine. The non-biofilm producing strain, ΔbcsA PcL::adrA, was used as a control. A 40 min exposure to sodium hypochlorite (200 p.p.m.) caused a decrease of ~5.5 logs in the number of control bacteria, compared to samples treated with only PBS (Fig 3A). Conversely, the same sodium hypochlorite treatment caused a reduction of ~1 log in the number of bacteria inside a cellulose or a PGA based biofilm (Fig 3A). These results determined that the protection against chlorine conferred by PGA is equivalent to that provided by cellulose.
Next, we tested the resistance that PGA and cellulose confer to five minutes of UV light irradiation. Although both exopolysaccharide overproducing strains survived better than the control strain that produces neither polysaccharide, the cellulose-positive cells showed a significantly higher survival rate than the PGA-positive cellulose-negative strain (Fig 3B). Thus, under our experimental conditions, cellulose provides better protection against ultraviolet radiation than PGA.
Microbial biofilm formation and production of extracellular polymeric substances are generally associated with metal resistance and tolerance [51]. To evaluate the protection conferred by cellulose and PGA on Salmonella against heavy metal stress, we treated macrocolony biofilms with 0.5 mM cadmium chloride (CdCl2). Results indicated that cellulose and PGA confer equal resistance to metal toxicity (Fig 3C).
Phages are found in abundance in environmental settings and bacteria have developed sophisticated mechanisms, including biofilm formation, to limit phage reproduction. To address the impact of cellulose and PGA biofilm extracellular matrices on phage infection, we infected bacteria that had been grown on membrane filters under biofilm forming conditions with a P22 phage lysate and analyzed the transduction frequency of a streptomycin resistance cassette. Results showed that, under our experimental conditions, neither exopolysaccharide protected Salmonella from phage infection (Fig 3D).
Overall, these findings suggested that PGA provides, at the most, similar benefits to those conferred by cellulose against environmental threats, at least under the conditions tested. Since both polysaccharides seem to have redundant roles in environmental survival, our results support the idea that during speciation the PGA pathway was lost without affecting survival outside the host during the Salmonella cyclic lifestyle.
During infection, the ability of Salmonella to survive and replicate in the vacuole within host phagocytic cells is essential for systemic disease [52]. To investigate the consequences of PGA production in Salmonella intramacrophage replication, we tested the ability of a PGA producing strain to replicate in RAW264.7 murine macrophages and compared it with that of a cellulose producing strain. To guarantee the synthesis of PGA or cellulose inside macrophages, we created Salmonella strains displaying high c-di-GMP levels inside these cells through the use of the macrophage activated phoP promoter fused to the adrA gene [53]. We firstly constructed WT PphoP::adrA and confirmed that it produced a cellulose based biofilm in response to the low Mg2+ signal activating the phoP promoter (S4 Fig). Then, we engineered ΔbcsA PphoP::adrA PcL::pga, a cellulose mutant that constitutively expresses the PGA synthesis machinery from the chromosome but that synthesizes PGA in a phoP dependent fashion (S4 Fig). As a control, we constructed WT ΔbcsA PphoP::adrA producing neither cellulose nor PGA. The three strains were phagocytosed at similar rates and as it has already been described, the cellulose overproducing strain was defective for replication inside macrophages [53], showing an ~50% intramacrophage survival relative to the control strain ΔbcsA PphoP::adrA (Fig 4A). Remarkably, the PGA producing strain was significantly more attenuated than the cellulose overproducing strain, showing a 7% intramacrophage survival relative to the control strain (Fig 4A).
Salmonella contained within the phagosomal environment encounter a diversity of antimicrobial factors including cationic antimicrobial peptides (CAMP) and reactive oxygen species (ROS) [54]. To investigate the cause(s) behind the low intramacrophage survival phenotype related to PGA production, we firstly performed one-hour polymyxin susceptibility assays [55] of bacterial cells previously grown under low Mg2+ levels, a condition that promotes polymyxin resistance through activation of the PhoP regulon [55,56]. The presence of either polysaccharide, cellulose or PGA, did not have an effect on Salmonella polymyxin resistance (Fig 4B). Then, we investigated whether reduced intracellular replication was linked to increased sensitivity to ROS production by assessing the ability to grow in the presence of 1mM H2O2 (Fig 4C). When wild type Salmonella were inoculated into 1 mM peroxide-containing medium at 107 CFU/ml, there was no increase in cell numbers for the first 3 h of incubation, followed by fast recovery [57]. Growth of the cellulose overproducing and control strains were indistinguishable from that of the wild type, whilst the PGA overproducing strain showed a significant viability loss throughout the incubation time (Fig 4C). Taken together, these results indicated that PGA production has a detrimental effect on Salmonella intramacrophage survival and that such survival decrease may be partially explained by the fact that PGA makes Salmonella more sensitive to oxidative stress.
Since heterologous expression of PGA makes Salmonella less capable to survive inside macrophages, we hypothesized that PGA production might result in virulence attenuation upon infection by the natural oral route of BALB/c mice, which are susceptible to systemic infection with Salmonella. Taking into account that c-di-GMP is involved in modulating the innate immune response [58,59], we constructed a Salmonella strain that constitutively produced PGA from the chromosome, without altering natural c-di-GMP levels. As expected, levels of PGA production by this strain, WT PcL::pga, were lower than those produced by WT pJET::pga (S5 Fig). Additionally, the bcsA gene was mutated in this strain, resulting in ΔbcsA PcL::pga, which produced PGA but not cellulose. Thus, virulence assays were carried out by comparing the pathogenic behavior of the control strain, ΔbcsA, which produces neither cellulose nor PGA, with that of either the PGA producing strain ΔbcsA PcL::pga or the wild type strain, which produces natural levels of cellulose during infection. These two strains did not show any discernable fitness cost compared to ΔbcsA when grown in LB broth at 37°C (S6 Fig). Firstly, the impact of PGA and cellulose synthesis on the capacity of Salmonella to adhere and invade the intestinal epithelium was analyzed by carrying out a competitive index analysis in an ileal loop coinfection experiment (Fig 5A). Both the wild type and ΔbcsA PcL::pga strains showed reduced capacity to adhere and invade the intestinal epithelium compared with the control strain, ΔbcsA. Secondly, we assessed the level of organ colonization following oral co-inoculation of the control strain, ΔbcsA, and either the wild type or ΔbcsA PcL::pga strain. In the case of mice co-infected with the wild type and ΔbcsA strains, the bacterial burden of the wild type was slightly higher than that of ΔbcsA in all organs analyzed (livers, spleens and gallbladders). Conversely, the PGA producing strain showed to be extremely attenuated, since no ΔbcsA PcL::pga bacteria were recovered from the organs examined after co-infection with the control strain (Fig 5B). To exclude the possibility that the control strain outcompetes the PGA producing strain when coinfection experiments are performed, we next compared the virulence of ΔbcsA and ΔbcsA PcL::pga strains by carrying out single infection experiments. Results confirmed that the PGA producing strain was highly attenuated, since mice inoculated with ΔbcsA PcL::pga did not show any disease symptom and most of them presented bacterial counts under the detection limit in livers, spleens and gallbladders (Fig 5C). It is important to note that in the case of gallbladders, the entire organ was plated and that six out of seven gallbladders from mice inoculated with ΔbcsA PcL::pga were free from infection. Thus, these findings reflected the PGA impact on Salmonella intramacrophage survival and supported the view that heterologous PGA production impairs Salmonella survival in orally infected mice.
Bile resistance is indispensable for Salmonella to colonize the hepatobiliary tract during systemic infection and persist in the gall bladder during chronic infection [60,61] and again, this characteristic represents a major difference between Enterobacteriaceae species. Thus, to further examine the consequences of PGA production in the Salmonella infection process, we analyzed the ability of Salmonella PGA producing cells to cope with the presence of bile. Dilutions from cultures of the wild type, the PGA producing strain, ΔbcsA PcL::pga, and their corresponding exopolysaccharide minus strain, ΔbcsA, were spread on LB plates supplemented with 24% bile bovine. Exposure to bile caused a decrease of ~3 logs in the number of cfu of both the wild type and ΔbcsA strain, whereas it provoked a reduction of ~5 logs in the case of the PGA producing strain (Fig 6A). Remarkably, PGA production in E. coli was also very detrimental for bile survival, since an E. coli strain producing PGA showed a ~2.5 logs reduction in bile sensitivity compared either with the wild type or with a pgaC mutant (Fig 6B and S7 Fig). To determine if the observed bile sensitivity mediated by PGA was common to other membrane active agents, we tested the sensitivity of Salmonella and E. coli PGA producing strains to the anionic detergent SDS. The minimal inhibitory concentration (MIC) of SDS for the Salmonella wild type and ΔbcsA strains was 17%, whilst it decreased to 15% in the case of ΔbcsA PcL::pga. On the other hand, the MIC for E. coli MG1655 and ΔpgaC strains was found to be 15%, compared with 7% for the PGA producing strain MG1655 PcL::pga. Altogether, these results indicate that PGA causes a significant reduction in bile resistance both in Salmonella and E. coli and suggest that this negative effect on resistance might be generalizable to other detergents.
Acquisition of new genes is considered to be a mechanism to enhance an organism’s ability to colonize a new environment, resist a specific antimicrobial or evade the immune system [40,62]. However, genomic data reveal that gene loss is also a widespread strategy to enhance bacterial fitness [63–68]. There are at least two reasons why bacteria may loss genes during evolution. A gene product or pathway may become superfluous in the new environment. In the absence of purifying selection, the gene accumulates neutral mutations, generating pseudogenes that may be finally removed from the bacterial genome. Alternatively, the product of the gene may be detrimental, triggering selection to optimize bacterial fitness in the new environment. It is well established that Salmonella evolution towards virulence has, at least, involved the acquisition by horizontal gene transfer (HGT) of a virulence plasmid and several pathogenicity islands that contain the genes necessary for invasion of intestinal epithelial cells and the systemic phase of infection [69,70]. However, the possibility that adaptation of Salmonella to the intracellular environment has occurred through gene loss has rarely been considered [65,71]. McClelland et al. proposed that gene deletion has contributed to genome degradation in S. Paratyphi A and Typhi serovars as they specialized to be human restricted variants. Nevertheless, these authors pointed out that the contribution of gene deletion to this evolution is less obvious than that of point mutations (pseudogenes) since the existence of a deletion is sometimes hard to determine [72]. Our work provides evidence that acquisition by Salmonella of an arsenal of virulence factors might have been useless in a strain producing PGA, supporting the idea that gene transfer and gene loss are inter-related processes, and that both contribute to the ongoing evolution of pathogenicity [73].
All bacterial species adapted to the mammalian intestine are resistant to the antibacterial activity of bile salts. However, the resistance of Salmonella enterica is especially remarkable. During systemic infection, Salmonella is able to transit from the liver into the gallbladder, where it can either induce inflammation and acute infection or persist chronically, creating a carrier state [74–76]. Several cell components and mechanisms have been related with Salmonella resistance to bile [77,78]. On one hand, different efflux pumps transport bile salts outside the cell decreasing their intracellular concentration [79,80]. On the other, diverse strategies that involve membrane reorganization and provide barriers to reduce bile salts uptake have been described, such as remodeling the lipopolysaccharide (both lipid A and O-antigen), changing the length of the enterobacterial common antigen and reducing the content of the Braun lipoprotein bound to the peptidoglycan, the levels of muropeptides cross-linked by 3–3 peptide bridges and the amount of porins sensitive to bile [81–86]. Our finding that constitutive production of PGA causes bile sensitivity in S. Enteritidis suggests an alternative strategy: the removal of compounds (PGA) that render the bacteria susceptible to bile. How PGA causes this effect is presently unclear. PGA represents an unusual bacterial exopolysaccharide, as some GlcNAc residues become deacetylated by the PgaB protein during secretion, providing a positive net charge to the polymer [15,32]. Thus, the presence of PGA may favor the accumulation of anionic bile salts on the bacterial surface.
We showed that constitutive expression of PGA also causes bile sensitivity in E. coli. These results raise the broader question of why E. coli, which displays a fair level of bile resistance necessary to grow in the small intestine, still produces PGA. Bile salts are maintained at high concentrations in the duodenum, jejunum, and proximal ileum. In the distal ileum, bile salts are absorbed into the blood-stream, and the majority of bile is recycled back into the small intestine and does not enter the colon [78]. E. coli resides in the microbiota found in the cecum and colon of humans. Thus, the presence of PGA might be compatible with the bile concentration in the small intestine and not with the concentration in the gallbladder. Alternatively, it is also possible that E. coli has developed regulatory systems to prevent PGA expression in the small intestine.
The second step of the infection process that is negatively affected by the presence of PGA is the survival and replication in the vacuole within host phagocytic cells. During systemic infection, Salmonella survives and replicates in vacuoles within host phagocytic cells where it must overcome the reactive oxygen species produced by macrophages [87]. It has been reported that Salmonella needs to repress cellulose production inside the vacuole through the activation of MgtC, which prevents a rise in c-di-GMP [53]. Increased levels of cellulose interfere with replication inside the vacuole and impair virulence in mice. The mechanisms underlying the antivirulence trait of cellulose has not been determined. We have now found that PGA production also hinders Salmonella division inside macrophages. Regarding this phenotype, we showed that production of PGA increases the susceptibility to H2O2 treatment, thus providing a potential mechanism for this attenuation. The notion that PGA is detrimental during infection of mammal cells is supported by studies with Y. pestis [28]. Y. pestis forms PGA mediated biofilms below 30°C in the blood-feeding fleas favoring the transmission and invasiveness of the bacteria from fleas to mammals [88]. However, PGA production has to be inhibited in the mammal host over 30°C to allow the development of a lethal infection. This temperature dependent regulation of PGA depends on the tight regulation of the c-di-GMP secondary messenger.
Salmonella is an ubiquitous bacterium with a dual intracellular/extracellular lifestyle. Its extracellular life involves survival in the environment, a scenario in which exopolysaccharide-mediated biofilms play an important role, protecting bacteria against environmental threats. Our results indicate that PGA loss provides a fitness advantage when Salmonella colonizes the liver, gallbladder or resides inside the macrophages. However, loss of PGA might have negative consequences for survival in the environment unless another compound off the cell wall was able to compensate for PGA absence. Comparative phenotypic analysis between the protection conferred by PGA and cellulose against environmental threats revealed that PGA confers at the most similar benefits than cellulose, indicating that cellulose is sufficient to provide Salmonella with protection against environmental stresses and compensate for the loss of PGA function.
Our findings provide a plausible explanation for PGA loss from the Salmonella genome during evolution. They also enhance our understanding of the benefits and burdens of a widely used exopolysaccharide to form the bacterial biofilm matrix, highlighting the necessity of additional studies to depict the exact role of PGA at each step of the life cycle. Finally, our study may also encourage microbiologists to turn more attention towards gene loss research as an approach to obtain information about how pathogenic bacteria have adapted to the host.
Protein sequences from E. coli PgaABCD and PhoH were used in a Blastp search against the NCBI non-redundant database accessed in July 2016, using an e-value threshold of 10−5 and excluding from the results hits taxonomically assigned to E. coli. The sequences from the top 500 hits were retrieved for each search and aligned using MUSCLE v 3.8 [89] and then trimmed using trimAl v1.4 [90] (gap-score cut-off 0.9). A Maximum Likelihood phylogenetic reconstruction was performed using phyML v3.0 [91] with the JTT model, setting the number of rate categories to four, and inferring the number of invariant positions and the parameters of the gamma distribution from the data. Branch support was computed using an aLRT (approximate likelihood ratio test) based on a chi-square distribution.
Animal studies were performed in accordance with the European Community guiding in the care and use of animals (Directive 2010/63/EU). Protocols were approved by the ethics committee of the Public University of Navarra (Comité de Ética, Experimentación Animal y Bioseguridad of the Universidad Pública de Navarra) (approved protocol PI-004/11). Work was carried out in the animal facility of the Instituto de Agrobiotecnología, Universidad Pública de Navarra. Animals were housed under controlled environmental conditions with food and water ad libitum. Mice were euthanized by CO2 inhalation followed by cervical dislocation and all efforts were made to minimize suffering.
The strains and plasmids used in this work are described in S1 Table. Escherichia coli and S. enterica subsp. enterica serovar Enteritidis (S. Enteritidis) cells were grown in LB broth and on LB agar (Pronadisa) with appropriate antibiotics at the following concentrations: kanamycin (Km), 50 μg ml-1; ampicillin (Am), 100 μg ml-1; carbenicillin (Cb), 50 μg ml-1; chloramphenicol (Cm), 20 μg ml-1; and streptomycin (Sm) 500 μg ml-1.
Routine DNA manipulations were performed using standard procedures unless otherwise indicated. Plasmid DNA from E. coli was purified using a Quantum Prep plasmid kit (BioRad). Plasmids were transformed into E. coli and S. Enteritidis by electroporation. Transformants carrying Red helper plasmids were made electro-competent as described [10,92]. Restriction enzymes were purchased from ThermoFisher Scientific and used according to the manufacturer’s instructions. Oligonucleotides were synthesized by StabVida (Caparica—Portugal) and are listed in S2 Table. Phage P22 HT105/1 int-201 [93] was used to carry out transductions between strains according to recommended protocols [94].
S. Enteritidis 3934 ΔXII is a multiple mutant carrying mutations in all genes encoding GGDEF domain proteins [42]. Derivatives of ΔXII containing the following single GGDEF protein encoding gene, namely adrA, yeaJ, sen1023, yciR, yegE, yfiN, yhdA, sen3222, and yhjK were constructed as described [41]. In the case of ΔXII+sen2484, ΔXII+yfeA and ΔXII+sen4316 strains, DNA fragments corresponding to the coding sequences of sen2484, yfeA and sen4316 genes were amplified with primer pairs A and D and chromosomal DNA from S. Enteritidis 3934 as a template. Amplified fragments were sequenced and cloned into the pKO3blue plasmid that was electroporated into ΔXII. Integration and excision of the plasmid was performed as described [41] in order to obtain the corresponding restored strains.
To express adrA under the PcL constitutive promoter in S. Enteritidis 3934, a PCR generated linear DNA fragment was used as described [95] with some modifications. The Red helper plasmid pKD46 was transformed into S. Enteritidis 3934, and transformants were selected on LB agar Am after incubation at 30°C for 24 h. One transformant carrying pKD46 was made electrocompetent as described [10]. A DNA fragment containing a kanamycin resistance gene, the PcL promoter and the RBS sequence of the PcL cassette was generated by PCR using primers adrA Km PcL rbs Fw and adra Km PcL rbs Rv and chromosomal DNA from strain MG1655 Km PcL-λATT-GFP as template [96]. Electroporation (25 mF, 200 W, 2.5kV) was carried out according to the manufacturer’s instructions (Bio-Rad) using 50 μl of cells and 1 to 5 μg of purified and dialysed (0.025 μm nitrocellulose filters; Millipore) PCR product. Shocked cells were added to 1 ml of LB broth, incubated for 1 h at 28°C and then spread on LB Km agar to select KmR transformants after incubation at 37°C for 24 h. Transformants were then grown on LB Km broth at 44°C for 24 h and incubated overnight on LB Am agar at 28°C to test for loss of the helper plasmid.
To place the adrA gene under the control of the phoP promoter, a protocol described previously was carried out with some modifications [92]. In a first step, primers Km SceI PphoP adrA Fw and Km SceI PphoP adrA Rv, with 60-bp homology extensions, were used to amplify a kanamycin resistance cassette and an I-SceI recognition site from plasmid pWRG717. This DNA was integrated upstream the adrA gene via λ Red-mediated recombination using plasmid pWRG730, a temperature-sensitive plasmid for independent inducible expression of the λ Red recombinase and I-SceI endonuclease. After confirming proper insertion of the resistance cassette by colony PCR with primers 01-E and Km SceI PphoP adrA Rv, a DNA fragment generated by PCR and derived from oligonucleotides PphoP adrA Fw and PphoP adrA Rv and S. Enteritidis 3934 chromosomal DNA as template, was electroporated into the mutant strain still containing the pWRG730 plasmid. This DNA fragment included the phoP promoter and homology regions used for its upstream adrA integration. After 1 h of incubation at 28°C, 100 μl of a 10−2 dilution was plated on LB agar plates containing 500 ng ml-1 anhydrotetracycline, which induced expression of I-SceI endonuclease. After overnight incubation at 28°C, single colonies were purified and successful recombination was checked by monitoring absence of antibiotic resistance, colony PCR with oligonucleotides 01-E and PphoP adrA Rv, and sequencing of the resulting fragment. Finally, pWRG730 was cured by incubating selected colonies at 44°C.
To insert the pgaABCD genes from E. coli K-12 MG1655 into the S. Enteritidis 3934 chromosome, the T64B prophage site was chosen [97]. Two DNA fragments, sb13 AB and sb13 CD, of ∼500 bp length of the S. Enteritidis sb13 gene, were amplified with primer pairs SmaI sb13 AB Fw/SphI sb13 AB Rv and SphI sb13 CD Fw/SalI sb13 CD Rv, respectively. The PCR products were cloned into the pJET 1.2 vector (ThermoFisher Scientific) and resulting plasmids were digested with SmaI and SphI enzymes in the case of the AB fragment and SphI and SalI enzymes in the case of the CD fragment. AB and CD fragments were ligated in the same ligation mixture with the pKO3 vector [98] digested with SmaI and SalI enzymes, resulting in plasmid pKO3::sb13AD. The pJET::pga plasmid constructed in this study was digested with SphI to obtain a DNA fragment containing the pga promoter and pgaABCD genes. Ppga::pgaABCD was ligated with pKO3::sb13AD digested with SphI, resulting in pKO3::sb13AD-Ppga::pgaABCD plasmid. Integration and excision of the plasmid was used as described [98] to obtain WT Ppga::pgaABCD strain. Insertion of Ppga::pgaABCD into the sb13 gene was confirmed by PCR using primers sb13 OK Fw and pgaA comp Rv. The ability of this strain to produce PGA was not detectable by Dot Blot, probably because heterologous chromosomal expression of the pgaABCD operon under its own promoter was not sufficient to produce evident PGA levels. Thus, a second Salmonella strain was generated in order to express pgaABCD under the PcL constitutive promoter and in the chromosome. To do so, a 427 bp DNA fragment, namely sb13 AB2, of the S. Enteritidis sb13 gene was amplified with primers BglII sb13 AB Fw and BamHI sb13 AB Rv, using S. Enteritidis 3934 chromosomal DNA as template, and cloned into the pJET 1.2 vector (ThermoFisher Scientific). A second DNA fragment containing the PcLrbs promoter [96] and the first 543 bp of the pgaA gene coding sequence was constructed by overlapping PCR, using two separate PCR products. Primers BamHI PcLrbs Fw and sb13 PcLpga Rv were used to amplify the PcLrbs promoter, using E. coli MG1655 Km PcL-λATT-GFP chromosomal DNA as template [96]. Primers PcL pgaA Fw and PstI PcL pgaA Rv were used to amplify 543 bp of the pgaA gene, using E. coli MG1655 chromosomal DNA as template. These two purified PCR products were mixed, and a second PCR using BamHI PcL rbs Fw and PstI PcL pgaA Rv primers was performed to obtain a single DNA fragment, PcLrbs::pgaA, that was cloned into the pJET 1.2 vector (ThermoFisher Scientific). Plasmids pJET::sb13AB2 and pJET:: PcLrbs::pgaA were digested with BglII /BamHI and BamHI/PstI enzymes, respectively, and digestion products were ligated in the same ligation mixture with the pKO3Blue vector [41] digested with BglII and PstI enzymes, resulting in plasmid pKO3Blue::sb13AB2-PcLrbs::pgaA that was electroporated in WT Ppga::pgaABCD strain. Integration and excision of the plasmid was used as described [41] to generate Wt PcLrbs::pgaABCD. Insertion of PcLrbs::pgaABCD into the sb13 gene was confirmed by PCR using primers sb13 OK Fw and sb13 PcL pgaA Rv. Finally, a bcsA mutation was transduced from ΔbcsA strain to generate ΔbcsA::CmR PcLrbs::pgaABCD, which is hereafter abbreviated as ΔbcsA PcL::pga.
To express the pgaABCD operon under the PcL promoter in E. coli MG1655, a PCR generated linear DNA fragment and the Red helper plasmid pKD46 were used as described above. Primers used to generate the DNA fragment containing a kanamycin resistance gene, the PcL promoter and the RBS sequence of the PcL cassette were Km PcL rbs pga Fw and Km Pcl rbs pga Rv.
To delete a 500 bp fragment of the pgaC gene in E. coli MG1655, and as a consequence suppress PGA production in E.coli [15], a protocol described previously was carried out with some modifications [92]. First, primers pgaC Km SceI Fw and pgaC Km SceI Rv, with 60-bp homology extensions, were used to amplify a kanamycin resistance cassette and an I-SceI recognition site from plasmid pWRG717. This DNA was integrated in the pgaC gene using plasmid pWRG730 plasmid and integration was confirmed by colony PCR with primers pgaC Km SceI Fw and pgaD Rv. Phosphorylated 80-mer double-stranded DNA derived from oligonucleotides ΔpgaC Fw and ΔpgaC Rv was electroporated into the mutant strain still containing the pWRG730 plasmid. After 1 h of incubation at 28°C, 100 μl of a 10−2 dilution was plated on LB agar plates containing 500 ng ml-1 anhydrotetracycline, which induced expression of I-SceI endonuclease. After overnight incubation at 28°C, single colonies were purified, and successful recombination was checked by monitoring absence of antibiotic resistance and colony PCR with oligonucleotides ΔpgaC Fw and pgaD Rv. Finally, pWRG730 was cured by incubating selected colonies at 44°C.
PGA exopolysaccharide levels were quantified as previously described [14] with minor modifications. Briefly, cultures in 5 ml LB or LB Cb broth of the strains tested were adjusted to the same number of cells and centrifuged at 18,000 x g for 5 min. Pellets were resuspended in 50 μl of 0.5 M EDTA (pH 8.0) and suspensions were incubated for 5 min at 100°C and centrifuged at 18,000 x g for 5 min. Each supernatant (40 μl) was incubated with 10 μl of proteinase K (20 mg ml-1) (Sigma) for 30 min at 37°C. After the addition of 10 μl of Tris-buffered saline (20 mM Tris-HCl, 150 mM NaCl [pH 7.4]) containing 0.01% bromophenol blue, 5 μl were spotted on a nitrocellulose membrane using a Bio-Dot microfiltration apparatus (Bio-Rad). The membrane was blocked overnight with 5% skimmed milk in phosphate-buffered saline (PBS) with 0.1% Tween 20, and incubated for 2 h with specific anti-PNAG antibodies diluted 1:10,000 [25]. Bound antibodies were detected with peroxidase-conjugated goat anti-rabbit immunoglobulin G antibodies (Jackson ImmunoResearch Laboratories, Inc., West- grove, PA) diluted 1:10,000 and developed using the SuperSignal West Pico Chemiluminescent Substrate (ThermoFisher Scientific). All extracts assayed in a particular experiment were analyzed on the same membrane. Images obtained in a GBox Chemi HR16 system (Syngene) were cut and put together to assemble horizontal figures showing PGA quantification.
The cellulose mediated biofilm formed in glass tubes on standing rich cultures was examined visually after growth in 5 ml of LB broth at room temperature for 72 h [10]. The PGA mediated biofilm was visualized after growth in LB broth at 28°C in an orbital shaker (250 r.p.m) for 16 h [43]. Macrocolony biofilms on the surface of LB agar plates were formed after spotting 50 μl drops of overnight liquid cultures and incubating at 28°C for 48 hours [99].
For scanning electron microscopy bacterial strains were grown under biofilm forming conditions. Growth medium was removed and bacterial cells were fixed by adding a fixation solution (1.3% glutaraldehyde, 0.07M cacodylate buffer and 0.05% rhutenium red). Samples were then washed in and post-fixed by incubation with 2% osmium tetroxide for 1 h. Bacteria were then fully dehydrated in a graded series of ethanol solutions and dried in hexamethyldisilazane (HMDS, Sigma). Finally, samples were coated with 40 Å platinum, using a GATAN PECS 682 apparatus (Pleasanton, CA), before observation under a Zeiss Ultra plus FEG-SEM scanning electron microscope (Oberkochen, Germany) (Laboratoire de Biologie Cellulaire et Microscopie Electronique, UFR Médecine (Tours, France)).
Macrocolony biofilms were formed on the surface of LB or LB Cb agar plates as described above, and a 10 μl water droplet stained with red food colouring was placed on the biofilm to show the hydrophobicity exhibited by the structure [48].
To perform sodium hypochlorite survival analyses, a protocol described previously was carried out with some modifications [10]. Macrocolony biofilms were formed on sterile polymer membrane filters (diameter 47 mm; Millipore) resting on LB agar or LB agar Cb media for 48h at 28°C. Filters were then transferred to an empty petri dish and macrocolonies were treated with 10 ml PBS containing 200 p.p.m. sodium hypochlorite for 40 min at 37°C. Control samples were incubated with 10 ml of PBS. Macrocolonies were harvested with a bent tip and bacteria were washed in PBS three times and suspended in 5 ml of PBS. After vortexing and sonicating (30 sec; potency 3; Branson sonifier 250; microtip), bacteria were enumerated by viable plate counts.
Bacterial strains were grown in LB Cb broth at 28°C in an orbital shaker (200 r.p.m) for 16 h. After sonication (30 sec; potency 3; Branson sonifier 250; microtip), the OD600nm was adjusted to 1 and serial dilutions were plated on four plates of N media agar supplemented with Cb [53]. After 24h of growth at 28°C, two plates were irradiated with UV light for 5 min. All plates were then incubated at 28°C for 48h and the numbers of surviving bacteria were counted. Results are shown as % survival relative to non-irradiated samples. Experiments were conducted in triplicate.
In order to incubate all strains on the same plates, strains ΔbcsA PcL::adrA and WT PcL::adrA were transformed with a pJET empty plasmid.
Macrocolony biofilms on sterile polymer membrane filters (diameter 47 mm; Millipore) resting on LB agar or LB agar Cb media were formed after spotting 5 μl drops of overnight liquid cultures and incubating at 28°C for 48 hours. Filters were then transferred to an empty petri dish and macrocolonies were treated with 10 ml of 0.5 mM CdCl2 for 3 h at 28°C. Control samples were incubated with 10 ml of PBS. Macrocolonies were harvested with a bent tip and bacteria were washed in water three times and suspended in 5 ml of PBS. After vortexing and sonicating (30 sec; potency 3; Branson sonifier 250; microtip), bacteria were enumerated by viable plate counts.
Overnight cultures in LB or LB Cb broth were sonicated (30 sec; potency of 3; Branson sonifier 250; microtip) and the OD600nm was adjusted to 1. Sterile polymer membrane filters (diameter 47 mm; Millipore) were placed on LB or LB Cb agar plates and seeded with a 50 μl drop of each bacterial suspension. Plates were inverted and incubated at 28°C for 48 h to allow macrocolony biofilm formation on top of the filters, that were then transferred to an empty petri dish and treated with a P22 phage lysate generated from the streptomycin resistant strain S. Typhimurium SL1344. After 1h of incubation at 37°C, the entire content of the plates was collected, washed in PBS and plated on LB Sm agar. The number of streptomycin resistant cfu were indicative of transduction efficiency. Experiments were conducted in triplicate.
Macrophage survival assay was conducted essentially as described [53] with some modifications. The murine macrophage cell line RAW 264.7 was propagated in Dulbecco’s modified Eagle’s medium (DMEM) (Gibco) supplemented with 10% fetal bovine serum (Invitrogen) and 1% Penicillin/Streptomycin/Glutamine (Gibco). Macrophages were seeded at a density of 2 x 105 cells per well in 24-well plates 24 h prior to infection. Salmonella overnight cultures grown at 37°C in LB broth were sonicated (30 sec; potency 3; Branson sonifier 250; microtip) and diluted 1:100 in LB broth. Cultures were incubated at 37°C in an orbital shaker (200 r.p.m) to an OD600nm of 1 and the suspension was sonicated again and washed twice with DMEM deprived of serum. Macrophages were then infected with Salmonella strains at a multiplicity of infection of approximately 10:1 and plates were centrifuged at 1000 r.p.m. for 10 minutes at room temperature. After 20 min of phagocytosis, monolayers were washed twice with PBS and treated with gentamicin (100 μg ml-1) for 1 h. To estimate phagocytosed bacteria, samples were then washed three times with sterile PBS and macrophages were lyzed with 1% (vol/vol) Triton X-100-PBS to release intracellular bacteria that were counted by plating 25 μl of serial dilutions onto LB plates. To assess bacterial survival, medium was replaced by DMEM supplemented with 10% FBS and 12 μg ml-1 gentamycin and the cells were incubated at 37°C. After 18 h of infection, wells were washed twice with PBS and were treated with Triton X-100 as indicated above. The percentage survival was obtained by dividing the number of bacteria recovered after 18 h by the number of phagocytosed bacteria and multiplying by 100. At each stage when infected cells were lysed, the number of viable cells in duplicate monolayers infected with each strain was assessed by 0.4% trypan blue exclusion and counting viable cells. No difference in viability was noted between cells infected with the different strains. Experiments were done in triplicate on three independent occasions.
One-hour polymyxin susceptibility assays were performed as described [55]. Polymyxin B Sulfate (Sigma) was used at a final concentration of 2.5 μg ml-1. Data are presented as survival percentage relative to samples incubated in LB without polymyxin. Experiments were conducted in triplicate.
Sensitivity to hydrogen peroxide was tested as previously described [57] with minor modifications. Briefly, overnight cultures were subcultured at 1/100 in 5 ml LB containing either no or 1 mM H2O2 (Merk). Replica cultures were used for each time point. Cultures were grown at 37°C with aeration and collected hourly. LB broth contains ∼30–40 μM Mg2+ [101], which activates the phoP promoter, thus, leading to adrA expression. After sonication (30 sec; potency 3; Branson sonifier 250; microtip) the number of surviving bacteria were counted by plating serial dilutions onto LB plates. Experiments were performed on three separate occasions.
In order to differentiate strains in all mice competitive infections performed, the wild type and ΔbcsA PcL::pga strains were made streptomycin (Sm) resistant through P22 phage transduction of the aadA gene from the natural streptomycin resistant strain S. Typhimurium SL1344 [102].
To compare the in vivo interaction of Salmonella strains with murine intestinal epithelial cells, the ligated ileal loop co-infection model was used as described previously [10,100]. Strains were incubated on LB agar for 48 hours at room temperature, suspended in PBS and sonicated (30 sec; potency 3; Branson sonifier 250; microtip) prior to infection. Competitive index (CI) was defined as the log10 of the ratio of the exopolysaccharide producing strain to control strain recovered (Output) divided by the ratio of the exopolysaccharide producing strain to control strain present in the inoculum (Input). A CI > 0 indicates the exopolysaccharide producing strain with a colonization advantage compared to the control and a CI < 0 indicates the exopolysaccharide producing strain with a colonization disadvantage over the control.
Colonization experiments were carried out with 8-week-old female BALB/c mice (Charles River Laboratories). Mice were acclimated for 7 days after arrival before the experiments were started in the animal facility of the Instituto de Agrobiotecnología, Universidad Pública de Navarra. Food and water were removed, twelve and two hours respectively, before the administration of bacterial suspension. Mice were prefed with 20 μl of 10% sodium bicarbonate 30 min before bacterial inoculation. Water and food were again supplied right after inoculation.
Strains were incubated on LB agar for 48 hours at room temperature, suspended in PBS and sonicated (30 sec; potency 3; Branson sonifier 250; microtip) prior to infection. Mice were inoculated intragastrically with 100 μl of bacterial suspensions. In the case of coinfection experiments, the total bacteria inoculum was 2 x 108 cfu of combined polysaccharide producing strain and ΔbcsA strain at a ratio of 1:1. In the case of individual infections, inoculum was 1 x 108 cfu of the strain analysed. The cfu of each strain in the inoculum (input) were quantified by plating dilution series on LB agar supplemented with chloramphenicol and LB agar supplemented with streptomycin to distinguish between strains. Over the course of infection, mice were examined twice per day and a final disease score was given to each mouse according to clinical signs observed as follows. No clinical signs (0); mild clinical signs: ruffled fur (1); moderate clinical signs: ruffled fur plus, lethargy, hunched posture and decreased activity (2); severe clinical signs: paresis, paralysis, tremor, shivers, ataxia, rigidity (3). When evident signs of disease (score 2 to 3) were observed, mice were euthanized by CO2 inhalation followed by cervical dislocation. Then, dilution series of liver, spleen and gallbladder lysates were plated on LB agar for enumeration of cfu (output), using antibiotic resistance to differentiate strains. Values for CI were calculated as described above.
Bile bovine sensitivity assay was performed as described [82] with minor modifications. Bacterial strains were grown in LB broth at 28°C for 48 h in an orbital shaker (200 r.p.m). Two microliter portions of serial dilutions were incubated for 24 h at 37°C in LB agar plates containing either no or bile bovine (Sigma). The bile concentration used was 24% or 13% when assessing Salmonella and E. coli bile sensitivity, respectively.
To carry out the SDS MIC analysis, bacterial strains were grown in LB broth at 28°C for 48 h in an orbital shaker (200 r.p.m) and diluted such that samples of 2x103 CFU/ml were subjected to various concentrations of SDS in polypropylene microtiter plates (ThermoFisher Scientific). The plates were incubated overnight at 37°C under nonaerated conditions and the wells of the plate were visually analyzed to determine the MICs.
All statistical analyses were performed in GraphPad Prism 5.01. Sodium hypochlorite survival, UV light irradiation data, heavy metals resistance, susceptibility of biofilms to phage infection, macrophage survival and Polymyxin B resistance analyses were analysed by the Mann-Whitney U test. A two-way analysis of variance combined with the Bonferroni test was used to analyse statistical significance in hydrogen peroxide sensitivity assays. A nonparametric Mann-Whitney U test and an unpaired Student’s t test were used to assess significant differences in individual colonization or coinfection experiments, respectively.
|
10.1371/journal.pcbi.1006866 | Competing evolutionary paths in growing populations with applications to multidrug resistance | Investigating the emergence of a particular cell type is a recurring theme in models of growing cellular populations. The evolution of resistance to therapy is a classic example. Common questions are: when does the cell type first occur, and via which sequence of steps is it most likely to emerge? For growing populations, these questions can be formulated in a general framework of branching processes spreading through a graph from a root to a target vertex. Cells have a particular fitness value on each vertex and can transition along edges at specific rates. Vertices represent cell states, say genotypes or physical locations, while possible transitions are acquiring a mutation or cell migration. We focus on the setting where cells at the root vertex have the highest fitness and transition rates are small. Simple formulas are derived for the time to reach the target vertex and for the probability that it is reached along a given path in the graph. We demonstrate our results on several scenarios relevant to the emergence of drug resistance, including: the orderings of resistance-conferring mutations in bacteria and the impact of imperfect drug penetration in cancer.
| How long does it take for a treatment naive, growing bacterial colony to be able to survive exposure to a cocktail of antibiotics? En route to multidrug resistance, what order did the drugs become impotent in? Questions such as these that pertain to the emergence of a significant cell type in a growing population arise frequently. They are often investigated via mathematical modelling but biologically insightful results are challenging to obtain. Here we outline a general framework of a stochastically growing population spreading through a graph to study such questions and provide simple formulas as answers. The significant cell type appears upon the population reaching a target vertex. Due to their simplicity, the derived formulas are widely accessible and can be used to guide and develop intuition on a range of biological scenarios. We demonstrate this on several settings including: how a region where drugs cannot penetrate affects the emergence of resistance, and, the ordering of mutations that leads to drugs being ineffective.
| The timing and manner in which a particular phenotype arises in a population is a central question of theoretical biology [1–13]. A typical scenario is to consider an initially monomorphic, wild type population, composed of cells that can acquire mutations, for example single site substitutions on the genome. The phenotype of interest comes to exist after a cell has accrued a specific set of mutations. The interpretations of this event are application dependent, but examples are the genesis of cancer instigated by mutations in a pair of tumour suppressor genes, or the emergence of multidrug resistance via alterations to the genes coding for the target proteins. Regardless of context, the questions of when, and how, the phenotype emerges are of significant interest.
It is commonly assumed that the population under consideration is of fixed size. However, with the aim of characterising disease progression, an increasing body of research has been developed to examine the evolutionary dynamics of a growing population. These studies have provided insights on a range of applications, including; cancer genetics [14–17], metastasis formation [18–22], drug resistance [23–26], phylogenetics [27], and the impact of poor drug penetration [28–31].
Here we continue in the same vein by considering a stochastically growing cellular population, where cells can transition in such a fashion so as to alter their, and their offsprings, reproductive capabilities. Such a transition might be due to the acquisition of a (epi)genetic alteration or migration into a new environment. As before, suppose we have a cellular state of interest, for example; a given genotype, a spatial location, or a combination of both. Will this state ever be reached? If it is, when is it reached? And by which sequence of intermediate states?
To make the discussion clear, let us consider an example application: the emergence of multidrug resistant bacteria. Suppose an infection begins with a single pathogenic bacterium which is sensitive to two antibiotic therapies, drug A and drug B. In the absence of either drug, the initial bacterium initiates a growing colony. Within the colony, when a cell replicates one of the daughter cells can acquire a mutation yielding resistance to either of the drugs. Here our questions are: how long does it take for multidrug resistance to emerge? En route, is resistance to drug A or drug B more likely to arise first? An ability to answer such questions is key to understanding pre-existing resistance, a common cause of therapy failure in various settings [32]. This scenario is illustrated in Fig 1a. There each vertex represents a cellular type, in this case its resistance profile. The edges represent cell transitions via mutation upon replication.
In this article we focus on the setting where the intermediate states have reduced reproductive ability (fitness) relative to the initial cells. This is primarily motivated by the commonly observed cost of resistance [33–35], whereby cellular populations that are resistant to a given drug grow more slowly than their sensitive counterparts. A second scenario where we expect a reduction in fitness is drug sensitive cells becoming exposed to toxins, which increases the cell’s rate of dying (cytotoxic) or decreases the rate of replication (cytostatic). As in many biological applications the relevant transition rates are small (some representative examples are: measurements for the point mutation rate per cell division of 10−9 in cancer [36] and 10−10 for bacteria [37], while the dissemination rate of pancreatic cancer cells from the primary tumour was estimated as 10−7 per cell division in [19]), we concentrate our efforts on the regime of low transition rates. Our main contribution is to provide simple, intuitive formulas that answer the questions posed in the small transition rate limit. These formulas show explicitly the contribution of the model parameters, e.g. transition rates and fitness reductions, which allow them to be probed for biological insight. This provides relationships which would be difficult to deduce from simulation alone.
We now move to detail the general framework we study, which will be seen to encompass models introduced in previous works as special cases [25, 26, 30, 38]. Our main results concerning when and how a particular cell type emerges are then presented, and we follow this by demonstrating our method on several applications.
Our interest will always be in the emergence of a particular cell type in a growing population which is modelled as a specific form of a multitype branching process [39]. It is convenient to picture the population on a finite, simple, directed graph G = (V, E), see Fig 1b, where each vertex of the graph represents a cell type or state. The number of types in the system is denoted N and so we let V = {1, 2, …, N}. Thus E is a subset of the set of ordered pairs {(i, j): i, j ∈ V, i ≠ j}. For any given cell type there is an associated vertex in V. Henceforth we will refer to cells of type x as cells residing at vertex x. We will be concerned with the timing, and fashion, that vertex N, which we refer to as the target, is populated, when we initiate with cells at vertex 1, which we denote the root. In the example of multidrug resistance, illustrated in Fig 1a, the cells at the root are sensitive bacteria while the target population is resistant to both drugs.
Let (x) represent a cell at vertex x and ∅ symbolise a dead cell. Then, with all cells behaving independently, our cell level dynamics can be graphically represented as:
( x ) → { ( x ) , ( x ) at rate α ( x ) ∅ at rate β ( x ) ( x ) , ( y ) at rate ν ( x , y ) if ( x , y ) ∈ E .
That is: cells divide at rate α(x), die at rate β(x), and transition to a cell at vertex y at rate ν(x, y) if the edge (x, y) exists. We will denote the fitness (growth rate) of cells at vertex x by λ(x) = α(x) − β(x), the difference of division and death rates of vertex x cells. The parameters associated with the vertex 1 population feature prominently and so for convenience we let α = α(1), β = β(1) and λ = λ(1). As mentioned above, in this article we focus on the setting where the vertex 1 population has the largest fitness, which is also positive. That is, we assume that λ > 0 and for 2 ≤ x ≤ N − 1, λ(x) < λ. We do not specify the fitness of the target population (cells at vertex N).
A common variant when modeling transitions (or mutations) is to specify that cells of type x divide at rate α′(x), and then with probability ν′(x, y) a transition occurs to vertex y. For a fixed value of parameters, this formulation of transitions is equivalent to that given above upon letting ν(x, y) = α′(x)ν′(x, y), α(x) = α′(x)(1 − ∑y:(x,y)∈E ν′(x, y)). Also in the model as currently stated, transitioned cells have deterministically fixed new division and death rates. However including a finite distribution of fitness effects, when a transitioned cell is assigned random division and death rates, is actually covered by our model as given thus far. We discuss this point further in the Extensions section.
At a population level, the number of cells at vertex x at time t will be denoted Zx(t). We shall always assume that the system is initiated with only vertex 1 cells, at a quantity z, that is Zx(0) = zδx,1 with the Kronecker delta function δx,y.
We have two primary questions. The first is, having initiated the system with z vertex 1 cells, how long does it take for the population at the target vertex to arise? That is we concern ourself with the distribution of the target hitting time, defined as
T = inf { t ≥ 0 : Z N ( t ) > 0 } . (1)
Now assuming that the target vertex is populated by a founding cell, we ask from which path through the graph G did this founding cell come? This gives rise to a distribution over the set of paths (or walks) from the root vertex to the target vertex, which we aim to characterise. This second question is more precisely formulated in the Path distribution section. Throughout we will assume that we wait for the first cell at vertex N to arise, however it may be that the population founded by that initial cell goes extinct. If instead one wishes to wait for the first ‘successful’ cell, that is the first cell at vertex N to exist whose progeny survives, then all the results presented below hold so long as the mapping ν(x, N) ↦ ν(x, N)λ(N)/α(N) for all target adjacent vertices x, is applied.
Our results are most clearly understood when G is acyclic, which amounts to excluding reverse transitions. Therefore in the following presentation this will be assumed. The setting when this assumption is relaxed will be discussed in the Extensions section.
Consider any path from the root to the target vertex. Intuitively, if the transition rates encountered along this path are larger, the target population will be seeded from this path quicker. Conversely, the smaller the growth rates along this path are, the slower the target will be reached. It will transpire that this intuition is indeed correct. The key quantities that show how the time to the target, and which path is taken, depend on these competing factors will be seen to be the path weights, which we now introduce.
Let the set of paths between the root and the target be denoted P 1 , N. Any p ∈ P 1 , N, will be a sequence p = (p1, p2, …, pl, pl+1), where each 1 ≤ pi ≤ N, l is the path length (the number of edges in p), all pi are distinct (as presently G is acyclic), and p1 = 1, pl+1 = N. Along the path p, let us call the difference between the growth rate at the root and the vertices encountered the fitness costs. Then we define the weight of path p as
w ( p ) = ν ( p 1 , p 2 ) ∏ i = 2 l ν ( p i , p i + 1 ) λ - λ ( p i ) , (2)
that is the product of the transition rates along the path divided by the fitness costs along the path. Throughout the empty product is set to 1. Further, we let the total weight of the target be
ϕ N = ∑ p ∈ P 1 , N w ( p ) . (3)
We will take the case of a path graph as a running example. A path graph, as illustrated in Fig 2, is the scenario in which the only edges are (i, i + 1), for 1 ≤ i ≤ N − 1. There is only one path p between the root and the target and so in this case, ϕ N = w ( p ) = ν ( 1 , 2 ) ∏ i = 2 N - 1 ν ( i , i + 1 ) λ - λ ( i ).
How long do we wait until the target vertex is populated? We answer this question by characterising the distribution of T, defined in (1), when the target seeding transition rates (the transition rates associated with edges into the target vertex) are small. The key tool required is the long-time population numbers at the initial and intermediate vertices, which is discussed further in the Materials and Methods section. Using this we are able to prove the following theorem regarding the target hitting time, whose proof can be found in S1 Appendix (see Sections S3 and S4).
Theorem 1. As the target seeding transition rates tend to 0, we have
P ( T - μ > t ) → ( λ / α 1 + e λ t + β / α ) z (4)
where
μ = 1 λ log λ 2 α ϕ N .
Heuristically, the time until the target is populated is approximately μ plus some noise, where the distribution of the noise is given by the right hand side of (4). Note that this distribution depends only on parameters associated with vertex 1 cells, while all other parameters of the system are bundled into μ. For practical purposes the theorem yields the following approximation
P ( T > t ) ≈ ( λ / α 1 + e λ t ϕ N α / λ 2 + β / α ) z (5)
This approximation will be valid if the target seeding transition rates are small. We do not provide estimates on the approximation error, however simulations (for example see Fig 3b) demonstrate this approximation holds even for only moderately small transition rates.
Notice that it is possible that the target is never populated (T = ∞). In S1 Appendix (Proposition S1) we show that if the vertex 1 population survives forever then the target hitting time is finite with probability one. Furthermore in many relevant cases, namely a large initial population, low death rate, or small transition rates leaving vertex 1, that the target is eventually populated is essentially equivalent to the vertex 1 population surviving. The probability that the vertex 1 population survives is obtained S1 Appendix (Eq. S7), and this quantity thus gives an approximation for the probability that the target population will ever arise, which is
P ( T < ∞ ) ≈ 1 - ( β / α ) z . (6)
This also explains the β/α term in (5), which arises due to lineages stemming from the initial vertex 1 cells going extinct. The corresponding approximate distribution for the target hitting time, when we condition that the vertex 1 population survives is given in S1 Appendix (Eq. S78). Differentiating the conditional hitting time distribution yields the density which is compared with simulations in Fig 3c.
Let us now discuss the time centring term μ, as given in Theorem 1. In the case of β = 0, z = 1, it can be easily seen (recall λ = α − β) that the median of our approximate distribution (5), is exactly μ. More generally, if we let t1/2 be the median time to hit the target, then if the vertex 1 population survives, we have
t 1 / 2 ∼ μ - h ( z ) . (7)
as the target seeding transition rates tend to 0. The shift h(z), whose precise form can be found in S1 Appendix (Corollary S2), has the behaviour
h ( 1 ) = 0 , h ( z )∼ 1 λ log z λ α , z → ∞ . (8)
The shift exists as initiating the system with a larger number of cells at vertex 1, leads to the target being reached faster. In terms of notation f ∼ g means that f/g → 1, when the limit under consideration is taken.
We now return to the running example of the path graph setting which was introduced at the end of the General framework section. With z = 1, (7) yields that the median time for the target population to appear is
t 1 / 2 ≈ μ = 1 λ log λ 2 α ν ( 1 , 2 ) + 1 λ ∑ i = 2 N - 1 log λ - λ ( i ) ν ( i , i + 1 ) , (9)
for small ν(N − 1, N). The first summand on the right hand side comes from waiting for the first transition from the vertex 1 population, while the second is due transitions between the remaining vertices. The first summand is distinct as the vertex 1 growth is the main cause of stochasticity, as discussed in the Materials and Methods section. The population growth and target hitting time is illustrated for the path graph case in Fig 3.
We now move to our second question: which path leads to the target population arising? Our naive expectation is that paths with larger weights will be more likely. This simple conjecture turns out to be true. To show this we first introduce some notation.
To any existing cell, say at vertex x, we may define the cell’s vertex lineage. This tracks the vertices of cells in the ancestral lineage of the cell under consideration and is a sequence of the form (1, …, x). For example a cell with vertex lineage (1, 3, 4) is at vertex 4, and descended from a vertex 3 cell, who in turn descended from a vertex 1 cell. Any vertex lineage is a path in G and we denote the number of cells with vertex lineage q at time t as X(q, t). Note that while there may be multiple ancestors at the same vertex in a given cell’s ancestral lineage, for example several generations of vertex 3 cells before the first vertex 4 cell, we do not record these in the vertex lineage (more precisely qi ≠ qi+1).
Now for any root to target path p ∈ P 1 , N, that is a path of the form p = (1, p2, p3, …, N), the first time that p is traversed can be defined as
T ( p ) = inf { t ≥ 0 : X ( p , t ) > 0 } .
Observe that path p populating the target first is equivalent to T(p) = T. The question of which path initiated the target only makes sense if the target is eventually populated. To ensure that this occurs we condition on a surviving vertex 1 population, and hence let
P 1 ( · ) = P ( · | vertex 1 population survives ) .
Recall that the vertex 1 population surviving is used as a proxy for T being finite (the target is eventually populated), as discussed in the paragraph following (5). We can now state the answer to our second question, which is a special case of Theorem 2 presented below. The probability that the target is populated via path p is simply the path weight of p, suitably normalised, that is
P 1 ( T ( p ) = T ) ≈ w ( p ) ∑ q ∈ P 1 , N w ( q ) = w ( p ) ϕ N (10)
for small target seeding transition rates. One may ask not only if a particular path populated the target, but whether the target was initiated from a given set of paths, for an example see the Imperfect drug penetration: combination therapy section. In this case the probability that the target is populated via a particular set of paths is given by summing the normalised path weights over each path in the set.
If the vertex 1 population avoids extinction then T(p) will be finite for each root to target path p. Therefore instead of only asking which path populated the target first, one might be interested in whether a particular path is traversed significantly faster. For example, does multidrug resistance obtained via one of the paths in Fig 1a occur t days before resistance from the other path. In order to also consider this case, for a root to target path p, we define
T ( ¬ p ) = min { T ( q ) : q ∈ P 1 , N \ { p } } ,
that is the first time the target is reached along any path except p.
We can now quantify the probability of reaching the target via path p more than t time units before any other path. To avoid discussing a technical assumption, which excludes pathological scenarios, the theorem is stated in approximate form. The more precise version can be found in S1 Appendix (Theorem S3 in Section S5).
Theorem 2. For small target seeding transition rates, and t ∈ R,
P 1 ( T ( ¬ p ) - T ( p ) > t ) ≈ w ( p ) w ( p ) + e λ t ∑ q ∈ P 1 , N \ { p } w ( q ) .
Note that letting t = 0 gives (10). In the running example of a path graph we cannot ask which path seeded the target vertex. However we briefly illustrate the usefulness of the above theorem by considering the question of whether the target vertex is populated via a fitness valley (where by fitness valley we mean a path such that cells at the internal vertices have smaller growth rates than the root population).
Let us consider the case depicted in Fig 4a where two paths exist to the target, a direct path and an indirect path. Here N = 3, with paths p(1) = (1, 2, 3), p(2) = (1, 3) and ν = ν(1, 2) = ν(2, 3), ν(1, 3) = ν2. Naively it may be expected that as the product of transition rates along both these paths are equal, and the indirect path (p(1)) contains a vertex with a fitness cost, the direct path (p(2)) is more likely. However, for small ν, Theorem 2 informs us that the target is populated via the indirect path t time units before the direct path, with probability
P 1 ( T ( 2 ) - T ( 1 ) > t ) ≈ [ 1 + e λ t ( λ - λ ( 2 ) ) ] - 1 (11)
where T(i) = T(p(i)). Thus the indirect path is more probable, that is P 1 ( T ( 1 ) < T ( 2 ) ) > P 1 ( T ( 2 ) < T ( 1 ) ), if λ − 1 < λ(2).
Further applications of the results presented thus far will be given in the Applications section. Before this, we discuss the case when G is cyclic, and some extensions to the initial model for which our results still hold.
Before discussing related work and summarising this study, we demonstrate how to apply the results of our Results section on some applications. The approximate formulas given in the Results section will be our key tools, and we now briefly remark on these.
The results of the Results section hold when the final transition rates in each path tends to 0. Therefore in our applications all transition rates, in particular the mutation and migration rates discussed below, will be taken to be small and statements are to be interpreted as approximations that hold true in this limiting regime. We also note that in our model, when transitions represent migrations, they also occur at cell divisions, (x) → (x), (y), instead of at any time, as (x) → (y). This might appear strange from a biological point of view, but the former formulation for transitions has been chosen, as it simplifies the mathematical treatment. More importantly, we expect that for small transition rates these formulations lead to very similar target hitting times. Indeed, simulations support this claim, as presented in Fig B of S1 Appendix. In all applications considered we shall neglect the role of back transitions. This is for simplicity and is in keeping with the previous works that we compare with. The effect of (finitely many) back transitions could be included by using our results on cyclic graphs.
First we consider the impact of imperfect drug penetration on the emergence of resistance. Two recent publications have explored how resistance spreads in this setting, and have shown that poor penetration can accelerate resistance [30, 31]. We are able to recover and extend some of their findings in a rigorous fashion. Next, the ordering by which resistance-causing mutations accrue is investigated, firstly in the setting of cancer and then in bacterial infections. In the case of bacterial infections we examine how the risk of multidrug resistance depends on mutation rates, recovering simulation results in [26]. A particular aim of this section is to illustrate how to apply the results of our Results section, and so, for the readers’ convenience, we have collected the key formulas in Table 1.
The main ingredient in proving Theorem 1, and consequently Theorem 2, is the long-time behaviour of the population of the initial and intermediate vertices. In this section we discuss this asymptotic behaviour.
Firstly we extend the definition of the total path weight given in the General framework section. Let the set of paths between the root and vertex x be denoted P 1 , x. Then using the definition of path weight (2), we let the total weight for vertex x be ϕ x = ∑ p ∈ P 1 , x w ( p ) for 2 ≤ x ≤ N. Further, let us denote the ratio of the total weight to fitness cost for vertex x as
Φ x = ϕ x λ - λ ( x )
for 2 ≤ x ≤ N − 1. To allow us to conveniently describe the growth at vertex 1 also, we set Φ1 = 1.
The target population can only be founded by transitions from cells residing at the neighbouring vertices (vertices connected to the target by an edge). Therefore, understanding the population growth of cells at these vertices is needed to discuss the timing and manner in which cells at the target arise. At large times, this understanding is provided by the following theorem, concerning the population at the initial and intermediate vertices (it is useful to recall that Zx(t) is the number of cells at vertex x at time t and that we initiate with z cells at vertex 1). It follows from a more general result due to Janson [41], which we have tailored to our particular setting.
Theorem 3. With probability one
lim t → ∞ e - λ t ( Z x ( t ) ) x = 1 N - 1 = W ( Φ x ) x = 1 N - 1 . Here W is distributed as the sum of K independent exponential random variables with parameter λ/α, where K is binomial with z trials and success probability λ/α.
A short proof, demonstrating how Theorem 3 can be deduced from [41] is given in Section S2 of the S1 Appendix (see Theorem S1). The distribution of W may be called a binomial-Erlang mixture, being an Erlang distribution with a binomially distributed shape parameter. For the running example of a path graph on N vertices, we see that Φ x = ∏ i = 2 x ν ( i - 1 , i ) λ - λ ( i ) for 1 ≤ x ≤ N − 1. This is demonstrated in Fig 3a. We note that in the case of N = 3, Theorem 3 was previously given in [16] (Theorem 3.2).
Using Theorem 3 and that the immigration rate into the target vertex from any target neighbouring vertex x is Zx(t)ν(x, N) we are able to prove the results of our Results section. Full proofs are provided in S1 Appendix.
|
10.1371/journal.pntd.0006466 | Dissecting the phyloepidemiology of Trypanosoma cruzi I (TcI) in Brazil by the use of high resolution genetic markers | Trypanosoma cruzi, the causal agent of Chagas disease, is monophyletic but genetically heterogeneous. It is currently represented by six genetic lineages (Discrete Typing Units, DTUs) designated TcI-TcVI. TcI is the most geographically widespread and genetically heterogeneous lineage, this as is evidenced by a wide range of genetic markers applied to isolates spanning a vast geographic range in Latin America.
In total, 78 TcI isolated from hosts and vectors distributed in 5 different biomes of Brazil, were analyzed using 6 nuclear housekeeping genes, 25 microsatellite loci and one mitochondrial marker. Nuclear markers reveal substantial genetic diversity, significant gene flow between biomes, incongruence in phylogenies, and haplotypic analysis indicative of intra-DTU genetic exchange. Phylogenetic reconstructions based on mitochondrial and nuclear loci were incongruent, and consistent with introgression. Structure analysis of microsatellite data reveals that, amongst biomes, the Amazon is the most genetically diverse and experiences the lowest level of gene flow. Investigation of population structure based on the host species/genus, indicated that Didelphis marsupialis might play a role as the main disperser of TcI.
The present work considers a large TcI sample from different hosts and vectors spanning multiple ecologically diverse biomes in Brazil. Importantly, we combine fast and slow evolving markers to contribute to the epizootiological understanding of TcI in five distinct Brazilian biomes. This constitutes the first instance in which MLST analysis was combined with the use of MLMT and maxicircle markers to evaluate the genetic diversity of TcI isolates in Brazil. Our results demonstrate the existence of substantial genetic diversity and the occurrence of introgression events. We provide evidence of genetic exchange in TcI isolates from Brazil and of the relative isolation of TcI in the Amazon biome. We observe the absence of strict associations with TcI genotypes to geographic areas and/or host species.
| T. cruzi is a zoonotic protozoan parasite infecting mammals and widely dispersed throughout endemic Latin America. It is known to possess considerable genetic diversity, comprising six discrete genetic lineages designated Discrete Typing Units (DTUs) TcI-TcVI. TcI is the most genetically diverse DTU and the most frequently sampled lineage in Brazil. We use a combination of high resolution molecular techniques to analyze the genetic diversity of Brazilian TcI isolates obtained from a wide geographical area encompassing five distinct biomes isolated from different mammal hosts and insect vectors. The results reveal significant genetic diversity and no clear association of genotypes with areas or host/vector species. Evidence from incongruent phylogenetic topologies based on nuclear and mitochondrial markers are indicative of genetic exchange and/or introgression events. The relevance of these findings in the context of population structure, ecology and epizootiology is discussed.
| Trypanosoma cruzi, a protozoan parasite (Kinetoplastidea: Trypanosomatidae), is known to possess a complex epidemiology and is widely distributed from the southern states of the United States of America to the Argentinian Patagonia. T. cruzi is a pervasive zoonosis capable of affecting more than 150 domestic and wild mammal species, distributed across 8 orders. T. cruzi infection in humans, may result in Chagas disease [1–3].Transmission to humans is mainly vectorial in endemic areas and over 100 species of hematophagous triatomine insects can harbor the parasites [4, 5]. Moreover, migration of individuals from highly endemic regions to the United States and Europe has resulted in significant public health concerns in recipient countries [6]. Domestic transmission of Chagas disease (CD) in Brazil by Triatoma infestans has been successfully interrupted [7]. However, human infection by T. cruzi is re-emerging as a food-borne disease in previously non-endemic areas [8–10]. Annual outbreaks have occurred, particulary in the northern Brazilian Amazon region during the past decade. Here, some local products derived from fruit juice have been contaminated with infected feces of triatomine bugs of different genera [9, 11, 12].
T. cruzi is characterized by a remarkable genetic heterogeneity [13, 14] and is currently comprised of six lineages or discrete typing units (DTUs), designated TcI to TcVI [15, 16]. In addition, recent evidence also supports the existence of a seventh lineage (TcBat) associated to bats [17]. The most genetically distant DTUs are TcI and TcII [18]. The evolutionary origins of TcIII and TcIV was initially proposed to be the result of an old hybridization between TcI and TcII [19], however more recent evidence shows that TcIII and TcIV have no hybrid origin, but rather are a monophyletic group with TcI that diverged from TcII [20, 21]. TcV and TcVI are known hybrid lineages which share haplotypes with TcII and TcIII [22, 23]. Whether given subpopulations of the parasite are associated with particular vector or host species or with distinct human disease characteristics is still unresolved.
TcI is the most frequently isolated DTU in the sylvatic environment, infecting diverse host and vector species across the Americas with an ancestral parental origin estimated at ~0.5–0.9 MYA [24, 23]. In Brazil, it is also the most widely distributed DTU, in terms of geography and diversity of host and vector species. Furthermore, it is, by far, the most genetically diverse DTU [25–30]. Llewellyn et al. [31] applied Multilocus Microsatellite Typing (MLMT) to the study of TcI population substructure in samples that originated from eight countries, isolated from 18 host and vector species, across 48 tandem repeats [32]. Results revealed extensive intra DTU diversity and spatial structuring of specific genotypes associated with acute oral outbreaks or vectorial infections in Venezuela. In addition, remarkable genetic diversity, through multiclonality, was observed when a single Didelphis reservoir host of TcI was studied [33].
Attempts to subdivide TcI strains into epidemiologically relevant groups are ongoing [34]. Herrera et al. [28] and Cura et al. [35] described five haplotypes associated with transmission cycles in Colombia, Chile and Bolivia. Ramirez et al. [36], used MLST to identify TcI genotypes specifically associated with human infection (TcIDOM) and others associated with peridomestic/sylvatic areas. MLST exploits nucleotide diversity present in four to ten single-copy housekeeping genes and has previously been applied to the study of T. cruzi using different marker combinations for lineage assignment and intraspecific characterization [18, 37]. Evidence for genetic exchange in TcI has been reported, for example, in strains isolated from Didelphis marsupialis and Rhodnius prolixus in the Amazon Basin [38] and in a domestic/peridomestic TcI population in Ecuador [39]. Experimental generation of intra-lineage hybrids suggest that TcI also displays a potential for genetic exchange [40].
Mitochondrial DNA in T. cruzi has a unique structure and function consisting of approximately 20–50 maxicircles (~20kb) and thousands of smaller minicircles (~1.4kb) [38]. Maxicircle DNA is uniparentally inherited and represents a useful taxonomic marker as it is highly mutable in comparison to nuclear DNA. Messenger et al. [41] developed a high resolution maxicircle multilocus sequence typing (mtMLST) scheme to describe intra-DTU diversity in TcI, revealing multiple mitochondrial introgression events and heteroplasmy within South American TcI. Introgression had already been detected in North America [21, 22] and in Brazil [42] and also Bolivia [43].
Together these studies illustrate several remarkable characteristics of TcI, namely the immense geographic distribution, diversity of host and vector species, extensive genetic diversity, and the capacity for genetic exchange. However, little is known about TcI in Brazil and extraordinarily there is only one relevant Brazil centric publication specifically addressing diversity of TcI [42]. Unlike Colombia and Venezuela, in Brazil there is no evidence of population substructure in the context of geographical distribution of intra DTU genotypes, distribution of host/vector species, or genotypes associated with acute outbreaks of CD in Brazil. In the present work, we comprehensively analyze a large cohort of Brazilian TcI isolates from five ecologically disparate biomes. Through the use of high resolution nuclear markers (MLST and MLMT) and a maxicircle region (COII), we investigate the phyloepizootiology of TcI from different Brazilian biomes. The study described herein was conducted with the following major hypothesis: DTU I of T. cruzi in Brazilian isolates displays extensive heterogeneity with no particular association of subpopulations to geographic areas, or host/vector species.
A total of 78 TcI isolates were supplied by Coleção de Trypanosoma sp de Mamíferos Silvestres, Domésticos e Vetores COLTRYP/FIOCRUZ. deposited by several researchers and maintained in liquid nitrogen. DNA was extracted immediately following initial isolation in NNN medium and one round of expansion in LIT. The isolates had previously been confirmed as TcI using Mini-Exon PCR [44].
In this work, TcI isolates were characterized using three high resolution methods comprising MLST, MLMT and maxicircle sequencing (COII) using appropriate reference isolates. Full isolate details are shown in S1 Table and include characterization methods applied to each sample, isolate localities and collection dates. To increase the robustness of the results, microsatellite information from 50 additional isolates, published by Lima et al. [42], was included in our MLMT analyses.
Isolates originated from vector and mammalian reservoir hosts across five Brazilian biomes; namely Atlantic Forest, Amazon, Caatinga, Cerrado, and Pantanal (Fig 1 and S1 Table).
The Cerrado biome is primarily open scrubland (savannah) covering approximately 2 million km2 of Central Brazil, comprising 23% of the total land surface area [45]. Scrubland is interspersed with gallery forests and is seasonally dry but with permanent swamplands dominated by Mauritia flexuosa palm trees [46].
The Pantanal biome is a large seasonal floodplain covering approximately 140000 km2 at the core of South America [47]. It is a biodiversity hotspot and freshwater ecosystem of global significance containing diverse mammal species and habitats. Climate instability results in periodic floods and droughts, affecting the population number and behavior of some species [48].
The Amazon biome is in the largest hydrographic basin of the world, comprising 44% of the South American subcontinent. The biome is a complex mosaic of very diverse ecosystems, dominated by tropical rain forests, with semi-arid regions, and a variety of man-made landscapes. The Amazon biome contains the greatest biological diversity (in absolute terms) on the planet [49].
The Atlantic Forest biome extends from the south of Pernambuco to the south of Rio Grande do Sul, and it is characterized by humid tropical forest. This biome is extensively impacted by human activities. It originally encompassed 12 percent of the national territory but only 1 to 5 percent (less than 100,000 km2) is intact today [50]. Containing more than 8,000 endemic species, the Atlantic Forest is recognized as one of the world’s most significant biodiversity hotspots [51].
The Caatinga biome in northeast Brazil, comprises a semi-arid ecological landscape with only 1% of its territory currently conserved, it is threatened by agriculture and cattle ranching [52]. This biome is characterized by clay and sandy soils with open plains supporting flora that is typical of semi-arid regions [53].
The TcI COII locus was amplified and sequenced according to Messenger et al. [41]. Nucleotide sequences per gene fragment are available at GenBank under accession numbers: MF781085-MF781124. Phylogenies were constructed implementing the substitution model based on the Akaike Information Criterion (AIC) in MEGA 6 [68]. To compare nuclear and mitochondrial topologies, Maximum-Likelihood (ML) phylogenies were constructed (T92+I model, Tamura 3-parameter) which assumes that a fraction of sites is evolutionarily invariable [68]. TcIII (CM17) and TcIV (Saimiri3 cl1, X10/610 cl5, ERA cl2 and 10R26) strains were included as outgroups (accession numbers: JQ581330.1, JQ581331.1, JQ581329.1, JQ581328.1 and JQ581327.1, respectively) [41].
Twenty-five microsatellite loci were amplified as previously described by Llewellyn et al. [31] with some modifications (S3 Table). Markers were distributed across 11 chromosomes, including six groups of physically linked loci [69]. The following reaction conditions were implemented across all loci: a denaturation step of 4 mins at 95°C, 30 amplification cycles 95°C (20 s), 57°C (20 s), 72°C (20 s) with a final 20 mins elongation step at 72°C. Reactions were performed in a final volume of 10 μL containing, 1X ThermoPol Reaction Buffer (New England Biolabs (NEB, UK), 4 mM MgCl2, 34 μM dNTPs; 0.75 mM of each primer, 1 unit of Taq polymerase (NEB, UK) and 1 ng of genomic DNA. Five fluorescent dyes were used to label forward primers, 6-FAM & TET (Proligo, Germany), NED, PET & VIC (Applied Biosystems, UK). Allele sizes were determined using an automated capillary sequencer (ABI 3730, Applied Biosystems, UK), in conjunction with a fluorescently tagged size standard (GeneScan– 500 LIZ, Applied Biosystems, UK), and manually checked for errors in GeneMapper software v3.7 (Applied Biosystems, UK).
Microsatellite data were assessed in accordance with Lewellyn et al. [31]. Individual-level clustering defined by Neighbour-Joining (NJ) phylogenies (DAS: 1 –proportion of shared alleles at all loci/n) between microsatellite genotypes was calculated in MICROSAT v. 1.5 [70] under the infinite-alleles model (IAM). To accommodate multi-allelic genotypes (≥3 alleles per locus), a script was developed in Microsoft Visual Basic to generate random multiple diploid re-samplings of each Multilocus profile. A final pair-wise distance matrix was derived from the mean across multiple re-sampled datasets and used to construct a NJ phylogenetic tree in PHYLIP v3.67 [71]. Majority rule consensus analysis of 10,000 bootstrap trees was performed in PHYLIP v 3.6 by combining 100 bootstraps generated in MICROSAT v. 1.5 [70], each drawn from 100 randomly re-sampled datasets.
Population assignment with a prior assumption of subdivision by collection sites was estimated with the Bayesian clustering program Structure v. 2.3 [72]. We assumed the admixture model due to the lack of information regarding ancestry, with correlated allele frequencies (i.e. frequencies in different populations are similar as a consequence of migration or shared ancestry) [72]. Simulations were set at 106 Markov Chain Monte Carlo (MCMC) interactions, with 2.5 x 105 iterations as burn-in. Ten independent runs were performed for each value of K (that correspond to the number of groups, 2–10), as suggested by Pritchard et al. [72]. The most likely K value was estimated with the ΔK method [73].
An alternative approach to summarize genetic polymorphism was performed using a non-parametric approach, free from Hardy-Weinberg assumptions. Briefly, a K-means clustering algorithm, executed in ADEGENET [74] was used to identify the optimal number of ‘true’ populations, with reference to the BIC, which reaches a minimum when approaching the best support for assignment of isolates to the appropriate number of clusters. The relationship between clusters and the strains contained within them was evaluated using a discriminant analysis of principal components (DAPC), as described in Jombert et al. [74].
A single randomly sampled diploid dataset, generated using a custom Microsoft Visual Basic script to re-sample random multiple diploid combinations of each Multilocus profile, was used for all subsequent analyses, as described in Jombert et al [75]. Population genetic statistics were calculated considering strains assigned to their DAPC-derived population clusters. DTU-level genetic diversity was evaluated using sample size corrected allelic richness (Ar) in FSTAT v 2.9 [76]. Intra-population sub-clustering was calculated as mean pairwise DAS values and associated standard deviations in MICROSAT v1.5 [70]. Sample size corrected private (population-specific) allele frequency per locus (PA/L) was calculated in HP-Rare [77]. Mean FIS, a measure of the distribution of heterozygosity within and between individuals, was calculated per population in FSTAT 2.9. FIS varies between -1 (all loci are heterozygous for the same alleles) and +1 (all loci are homozygous for different alleles). DTU-level heterozygosity indices were calculated in ARLEQUIN v3.11 [78] and associated significance levels for p-values derived after performing a sequential Bonferroni correction to minimize the likelihood of Type 1 errors [79]. Population subdivision was estimated using pairwise FST, linearized with Slatkin’s correction, in ARLEQUIN v 3.11. Statistical significance was assessed via 10.000 random permutations of alleles between populations. Three different strategies were performed to group the samples and calculate pairwise FST values: i) using isolate collection locations to investigate local diversity, ii) to assess levels of gene flow between the five ecologically disparate biomes and, iii) investigate the role of host/vector specificity in the context of host movement and the distribution of TcI genotypes. Within-population subdivision was evaluated in ARLEQUIN v 3.11 [74] using a hierarchal analysis of molecular variance (AMOVA). A Mantel test for the effect of isolation by distance within populations (pairwise genetic vs. geographic distance) was implemented in GENAIEX 6.5 using 10,000 random permutations [80]. The association between host/vectors and genotypic clusters based on DAPC were calculated using contingency tables along with a Chi-squared test.
Nucleotide sequences per gene fragment are available from GenBank under the accession numbers: MF615620-MF615679; MF615680-MF615739; MF615740-MF615799; MF615800-MF615859; MF615860-MF615919; MF615920-MF615979.
In our cohort of 78 isolates, 60 isolates were successfully characterized using all six MLST markers, 62 isolates by maxcicircle gene sequencing and 42 using MLMT markers. In particular, only twenty two isolates could be analyzed using all three methods. Furthermore, 50 more isolates were reassessed with MLMT, totaling 92 isolates considered for microsatellite analysis. S1 Table provides details of the typing methodologies applied to each particular isolate.
Six MLST markers were sequenced in 78 T. cruzi isolates, of which 60 consistently produced amplicons and sequences of acceptable quality. Concatenated gene fragments comprised a total of 2571 bp for each isolate. No single gene was able to differentiate all 60 isolates on the basis of TE and DP. Table 1 describes the level of diversity seen in each gene fragment; the number of polymorphic sites ranged from 10 (RHO1) to 1 (LAP). Typing efficiency (number of ST/number of polymorphisms) was variable among loci and the gene fragment distinguishing the highest number of genotypes per polymorphic site was RB19 (TE = 2.25). In contrast, CoAR showed the lowest efficiency (TE = 1). RHO1 demonstrated the highest DP (0.935) for our cohort; and LAP, the lowest (DP = 0.383).
All fragments met the criterion for stabilizing selection (dN/dS<1) for conservation of metabolic function. FEL analyses detected signs of purifying (negative) selection in 10 sites across four gene fragments (LAP - 24th site; RB19 - 35th and 63th sites; RHO1 - 76th site; and GTP - 13th, 15th, 42th, 98th, 119th, 136th positions; p < 0.1).
A comparison of diplotypic phylogenies of individual gene trees revealed differences in topology and clustering between gene fragments (S1–S9 Figs). However, similarities did exist, most notably, the highly diverse loci RB19 and RHO1 possessed similar topologies (S8 Fig). Following concatenation of all six markers, DP increased to 0.997, differentiating 55 genotypes from 60 isolates.
The minimum number of loci needed to derive the maximum DP was assessed for all combination of fragments (2 loci to 6 loci) through MLSTest. A combination of 5 fragments: CoAR-GTP-LAP-RB19-RHO1 (S10 Fig) also yielded a high DP (0.995), discriminating 53 genotypes (2 genotypes less than yielded by the use of all 6 MLST markers).
Individual gene fragment trees revealed multiples polytomies in all six phylogenetic trees (S1–S9 Figs). Substantial congruences between the phylogenetic trees generated with SNP duplication (with Bayes) and Average State (with NJ) were observed (S1–S6 Figs), The two fragments with the most pronounced inconsistencies between Bayes and NJ were PDH and RHO1 (S4 and S6 Figs).
S7–S9 Figs, show the comparison between the six gene trees. RB19 and RHO1 each produced a cluster corresponding to isolates from Atlantic Forest, Cerrado and Pantanal, which are mostly congruent. However, no two gene fragments showed completely identical topologies. The remaining loci (CoAR, LAP, PDH and GTP), which had lower TE and DP values, generally yielded trees that were less congruent. None of the gene fragments showed 100% congruence between their clusters.
Topological incongruence analyses revealed a mean of 2.86 incongruences per branch and 25% of branches with at least n-1 incongruent fragments. These correspond to moderate levels of incongruence (S11 Fig), where moderate incongruence was defined as being between 20 and 40% [64]. The ILD tests of discrepancies were no higher than expected, indicating that the combination of six gene fragments produces reliable branches (ILD = 0.05).
To evaluate intra-DTU diversity of TcI, phylogenies were inferred from the concatenated alignment of six gene fragments (Fig 2). Both, NJ and Bayesian methods produced similar results, although NJ analysis showed lower bootstrap values. Clusters with >50% support in both analyses are indicated. The presence of several sub-clusters was observed, revealing considerable intraspecific diversity within TcI and also similar genotypes circulating sympatrically over large geographical areas.
Specific phyloepidemiological observations are as follows. Cluster A grouped isolates originating from very distant localities including the Atlantic Forest, Cerrado and Pantanal biomes (BPP = 81% and bootstrap <50%). Of particular note, cluster A (Fig 2) lacked genotypes present in the Amazon, in congruence with results from maxicircle phylogenies (below). Also, similar genotypes were isolated from different species. For example, isolates from Didelphis spp, primates, chiroptera, one rodent, and triatomine bugs grouped within cluster A (Fig 2).
Amazonian isolates were genetically diverse and were mostly contained within a single clade (Fig 2). Cluster B (Fig 2) contained genotypes from the Atlantic Forest and Cerrado biomes, comprising an enormous geographical distance (~1.130 km). Interestingly, cluster C comprised isolates from distant biomes, Amazon and Atlantic Forest. Likewise, isolates from Abaetetuba (11609) and Cachoeira do Arari (10272), separated by vast geographical distances (~78.38 km), were grouped in the same cluster. Of note, a single isolate FRN26, Oecomys mamorae from the Pantanal, was genetically dissimilar from all other TcI strains and placed in different topological positions in the context of MLST and maxicircle phylogenies.
Although isolates were collected in different years and localities (S1 Table), no clear clustering by collection date or biome was apparent; however, statistical tests were not able to rule out the existence of some association. Details of these tests are described below.
Haplotype analysis, applied to nuclear loci, was used to generate phylogenies and investigate the allelic origins of heterozygous isolates from homozygous putative donor genotypes (Fig 3 and S12–S16 Figs). Here isolates with haplotypes present in two different genetically clusters that also contained respective homozygous donor isolate genotypes were considered potential hybrids.
Three genetic loci (GTP, PDH and RB19) revealed heterozygous isolates and allelic profiles that could be derived from homozygous genotypes (Fig 3, S12 and S13 Figs, respectively). In more detail, Fig 3 shows the GTP locus and alleles from homozygous donor isolates: 10285, haplotypes 1 and 2, in one clade; and 14943, haplotypes 1 and 2, in another. Within GTP, two isolates contain heterozygous allelic profiles, 12630 and 12624, corresponding to one allele from each homozygous donor. For PDH, five isolates contain heterozygous allelic profiles: 2892, 2896, 10285, 17645 and G41 (S12 Fig). The potential parental alleles for PDH were: 2879 and 2869, in one clade, and 2880, in another clade (S12 Fig). Similarly, for RB19, eight isolates showing potential genetic exchange were identified. The most plausible parental alleles for each putative hybrid are shown in S13 Fig, while the SNP profile for putative homozygous donors and the corresponding heterozygous profiles are shown in the S4–S6 Tables. Putative recombinants were different in PDH, GTP, and RB19; possibly indicating that there have been multiple genetic exchange events over time. Although we detect the signature of genetic exchange through heterozygous genotypes and their associated homozygous “donor” isolates, we observe no evidence of genetic exchange at the level of individual alleles, since allelic mosaics were not detected using RDP3 software.
Sixty two COII sequences produced a 449 bp alignment, 10 unique haplotypes and 64 polymorphic sites. Maximum-Likelihood trees (Fig 4) revealed two major clades and almost complete congruence with cluster A derived from concatenated MLST (S17 Fig). This cluster contains strains from the Atlantic Forest, Cerrado and Pantanal with the notable exclusion of Amazonian isolates (bootstrap = 100%). Interestingly, isolate FRN26 from the Pantanal, and isolate G41 from the Atlantic Forest formed a strongly supported sub-clade (bootstrap = 100%). In contrast, nuclear phylogenies grouped G41 with Amazonian isolates. Also of note, isolates within sub-clusters were highly homogeneous. Analyses with MLST and maxicircle were congruent in relation to the isolates of Amazon, in which they formed a separate group that included a cluster with isolates from Cerrado and Atlantic Forest (S17 Fig).
The presence of genetically identical mitochondrial sequences despite a mutation rate one order of magnitude greater than that of nuclear genes provides support for the occurrence of multiple mitochondrial introgression events (Fig 4 and S17 Fig). Additionally, these sequences correspond to geographically dispersed isolates, obtained from different biomes and hosts and vectors, further supporting the case for introgression.
In total, 4595 alleles were identified, corresponding to 92 unique multilocus genotypes. Multiple (≥3) alleles were observed at 1.87% of markers. This is most likely attributable to aneuploidy in a small proportion of the loci (S7 Table). Bayesian clustering applied to 92 strains revealed the existence of four discrete phylogenetic groups without apparent association to the biome of origin (Fig 5). For example, isolates from Atlantic Forest clustered across three groups (Fig 5, yellow, green and pink label), in which specimens from the state of Rio de Janeiro are genetically similar to those from Posse, Goias (Cerrado biome) and specimens from the states of Minas Gerais and Santa Catarina clustered together with samples from Pantanal. Moreover, TcI specimens of the state of Bahia are genetically more similar to samples from the states of Piauí (Caatinga), Pará and Amazonas (Amazon) than to other samples from the Atlantic Forest biome. It is worth mentioning that, in general, samples from Cerrado, Caatinga and Amazon biomes were grouped together in two different groups (Fig 5, red and pink coloured groups).
The DAPC analysis with the 92 strains yields five genetic clusters, evidenced by a slight ‘elbow’ in the distribution of the BIC values across optimal cluster numbers at K = 5, once 22 principal components (PCs) were retained and analyzed (representing 80% of the total variation) (S18 Fig). DAPC-derived populations were broadly congruent with patterns of nuclear clustering identified by NJ and Bayesian clustering analysis. The five DAPC clusters, showed in S1 Table, corresponded to: Population 1 that includes Caatinga (n = 13) and Cerrado (n = 4); population 2, Atlantic Forest (n = 4), Pantanal (n = 10) and Cerrado (n = 1); Population 3, Amazon (n = 30), Atlantic Forest (n = 6), Pantanal (n = 1) and Caatinga (n = 1); population 4, Atlantic Forest (n = 14) and Cerrado (n = 3) and population 5, the remaining parasites principally from opossums and primates in the Atlantic Forest (n = 14) and bats and an opossum in Cerrado (n = 3). Similarly, the NJ tree (Fig 6) reveals that parasites from the Atlantic Forest, Cerrado and Pantanal were generally admixed together. We observe no strict specific association between biomes, species or collection years and the clusters based on DAPC; however, the chi square contingency test (p<0.05) can not completely exclude an association between these clusters and host/vector species, collection biome or dates.
Cluster A, derived from MLST data, was congruent with one MLMT cluster, the equivalent maxicircle cluster (S17 and S19 Figs paired trees). The isolate G41 (Atlantic Forest), grouped with isolates from Amazonia for MLST but was grouped with Atlantic Forest isolates with MLMT analysis. Similarly, topological positions for FRN26/26 were different for MLST and maxcircle trees (S17 Fig paired trees)
Population genetic parameters were calculated for strains grouped a priori according to their biome of origin, as well as a posteriori DAPC cluster assignments (Table 2 and S8 Table). Consistent results are observed when strains are grouped according to DAPC-assigned clusters. Table 2 shows high levels of genetic heterogeneity in Amazon (DAPC population 3), as well as excess homozygosity, high numbers of private alleles per locus and a low standard deviation associated with DAS value. T. cruzi strains from Atlantic Forest, Cerrado and Caatinga displayed similar, but lower levels of diversity, with comparatively lower numbers of private alleles per locus.
Three diverse populations (Atlantic Forest, Cerrado and Caatinga) were characterized by elevated standard deviations associated with DAS values and positive FIS values (Table 2). A hierarchical AMOVA demonstrated 83.1% of total genetic variation was present within populations, compared to 16.9%, among populations (p<0.0001 for both).
The observed subdivision between a priori populations suggests the existence of gene flow between T. cruzi from the Atlantic Forest and those of the Cerrado biome (FST = 0.067) (Table 3). Gene flow was also inferred to have occurred between T. cruzi populations of Caatinga and Cerrado (FST = 0.0982) (Table 3). The admixed character of these isolates was also supported by Bayesian assignment. More geographically-distant TcI populations display similar levels of subdivision, as observed between Caatinga and Pantanal, Caatinga and Atlantic Forest, Cerrado and Pantanal and Pantanal and the Atlantic Forest. T. cruzi isolates from the Amazon biome exhibited lower FST values than populations of all other biomes (Table 3). This observation is also supported by FST values calculated for the a posteriori populations (S9 Table).
At a local-level structure analysis (i.e. when samples were grouped using the collection site as prior information; Table 4), it is clear that some isolates from Atlantic Forest grouped with others from Cerrado and Pantanal due to the genetic similarity between samples of Rio de Janeiro and Possas, Goias (FST = 0.04), of Bahia and Tocantins (FST = 0.13), and of Santa Catarina and Mato Grosso do Sul (FST = 0.09). Similarly, samples from Cerrado (Piaui) and Amazon (Para) showed low levels of structure (FST = 0.09). The investigation of parasite population structure based on host taxonomy suggests that Didelphis marsupialis might play a role as the main disperser of TcI (S10 Table), since its overall pairwise FST values were lower than the others (FST ≤ 0.2; median = 0.15).
Finally, to determine the extent of spatial genetic structure, a Mantel test was conducted, demonstrating significant parasite isolation by distance across the sampled geographical range (RXY = 0.384; P = 0.01). (S20 Fig).
In this study we explored the genetic diversity of Brazilian TcI isolates, obtained from different vectors and mammal hosts, spanning 5 different ecological biomes. To this end, we analyzed data from six protein coding genes (MLST), 25 microsatellite loci and one mitochondrial locus. We observed substantial genetic diversity with no strict association of clusters with particular host/vector species or biomes. However, some degree of association to a cluster from MLMT was present in isolates from Amazon. No other noticeable relation between clusters and biomes was identified; nevertheless, statistical tests are consistent with the possibility of some form of association. In addition, we observed mitochondrial introgression events and evidence of intra-DTU genetic exchange. Previous works [29, 41, 81, 82] have studied the genetic diversity within DTU I using different methods encompassing nuclear and mitochondrial markers. However, this constitutes the first instance in which MLST analyses in combination with two other high resolution genetic markers (microsatellite and maxicircle sequencing) have been used to evaluate intra TcI diversity of isolates from Brazil.
The criteria for justifying the selection of MLST markers used were based broadly on Diosque et al. [18], and fragments were assessed, in order of importance, in terms of intra-DTU diversity (TE), genotype discrimination (DP) and statistical support in phylogenetic trees. MLST analysis using all six concatenated gene fragments discriminated 55 genotypes out of 60 isolates. The use of a combination of five fragments of concatenated genes (CoAr, GTP, LAP, RHO1 and RB19) also proved to be a viable alternative to the six genes, discriminating 53 of a possible 60 isolates. However, in light of the slightly reduced discriminatory power, we recommend the use of all six gene fragments.
There was significant variation in TE and DP among tested loci (Table 1). The most variable loci were RHO1 and GTP which possessed 17 and 13 polymorphic sites, respectively. This is in line with the observations made previously by Diosque et al. [18] and Ramirez et al. [36]. The LAP locus contained the least number of SNPs (4 polymorphic sites), in accordance with Ramirez et al. [36]. Some previous works have assessed RB19, GPI, LAP and TR and considered them non-informative when applied to typing schemes for cohorts spanning all six DTUs [18, 19, 36]. However, in the context of Brazilian isolates, RB19 proved to be a highly informative marker for investigating intra-TcI diversity (TE = 1.9). Variation in TE is likely the result of selective pressures on individual loci, genetic drift or differences in mutation rates. Additionally, non-synonymous SNPs in MLST fragments, contributing to amino acid alterations, have previously been reported in T. cruzi [36, 37]. Here all gene fragments met the criteria for stabilizing selection (<1) for conservation of metabolic function. FEL analyses provided evidence that 10 sites in four of the six gene fragments are under purifying selection.
Bayesian and NJ analyses of the trees generated with each individual gene showed the presence of polytomies (S1–S6 Figs). One possible explanation for the existence of polytomies is the relatively small number of informative polymorphisms in the markers analyzed. This type of structure (Genetic Structure Type 2) was previously observed by Tomasini et al [64] when studying A. fumigatus through the application of MLST.
We generated phylogenetic trees with NJ to assess the robustness of our findings, since Bayesian analyses with SNP duplication can lead to artificially high bootstrap values [64]. Indeed, the support values were higher in the Bayesian analysis with SNP duplication; nevertheless, the clusters in the concatenated tree were mainly consistent between both analyses. Concatenation across loci by MLST has been successfully applied to many prokaryotic and eukaryotic organisms [22, 61, 63, 83, 84–86]. Nevertheless, when using this methodology, high levels of inbreeding or genetic exchange at particular loci may confound true phylogenetic relationships; therefore, in the presence of these effects, results must be interpreted cautiously. In this study, concatenation of six genes resulted in two main groups: the first included all isolates from Amazon and some representatives from Cerrado and Atlantic Forest (Fig 2, clusters B and C), and the second group included the remaining isolates from Atlantic Forest, Cerrado and part of Pantanal (Fig 2, cluster A). The epizootiological significance of these findings are discussed below.
The BPP values supporting those clusters that show incongruences varied widely between individual gene phylogenies. Similar patterns of incongruence have been previously observed in nuclear genes [37, 87]. Such incongruence, where isolates differ in topological positions, are a classical marker in populations that have undergone genetic exchange. To investigate further, haplotypic phylogenies were constructed for each genetic locus in order to define heterozygous isolates and their potential homozygous allelic donors. (Fig 3, S12–S16 Figs) The results indicate potential allelic recombinants in 3 of the 6 loci. Putative recombinant isolates possessed heterozygous allelic profiles, each present in two different homozygous putative donor isolates, situated in different phylogenetic clusters. Potential allelic recombinant isolates across 3 genes is suggestive of multiple genetic exchange events. PHASE is a Bayesian method for the reconstruction of haplotypes. It is generally considered one of the most accurate haplotype reconstruction methodologies. However, there are potential confounders, for example, population size and frequency of recombination have the potential to skew outcomes. Furthermore, one must be cautious when using PHASE to infer frequency of genetic exchange, as this is one of the assumptions of the method. Nevertheless, the presence of recombinants and potential “donor” genotypes inferred in three independent nuclear markers is confirmed by heterozygous and homozygous SNPs derived from nuclear sequences (S4–S6 Tables). Together, these observations constitute evidence for the presence of genetic exchange at the nuclear level. Population structure of T. cruzi is frequently regarded as clonal [88]. This model does not exclude genetic exchange, but considers it to be infrequent [89]. However, exchange across DTUs has been demonstrated using MLST [18, 36]; and intra-TcI genetic exchange in a single isolate has been observed in a cohort of Colombian samples [36]. Similarly, Messenger et al. [41] and Ramirez et al [82] observed multiple incongruence and introgression events within TcI on the basis of MLMT, MLST and maxicircle phylogenies, concluding that genetic exchange within DTU I is frequent. Genetic exchange is inferred in the current data set, however the frequency of genetic exchange is presently unknown and a topic of enthusiastic debate.
In comparison with nuclear genes, remarkably low levels of intra DTU COII diversity were observed. Paradoxically, the mutation rate of mitochondrial genes is generally considered one order of magnitude higher than that of nuclear genes [90]. The spectrum of reduced diversity observed in maxicircle clades is consistent with introgression events as also reported in different TcI populations in South America [36, 41]. Two major clades were observed, the first consisting of all samples from the Amazon biome, together with a few samples from other biomes. The second clade grouped all of the remaining samples. This pattern was congruent across both nuclear and mitochondrial loci, and is indicative of genetically discrete populations. MLMT analysis (below) suggests limited gene flow (FST) between the Amazon biome and other studied areas. Although nuclear and mitochondrial phylogenies shared some topological characteristics, there were substantial incongruences between them. For example, isolate G41 clustered with Amazonian isolates at the nuclear level, but associated in mitochondrial phylogenies with isolates of non-Amazonian origin (S17 Fig). In the context of introgression, the discordance between nuclear and mitochondrial phylogenies is indicative of a prolonged and continuous association between populations from very distant localities [41]. This is consistent with the suggestion that genetic exchange in T. cruzi involves the independent exchange of kinetoplasts and nuclear genetic material [41]. Reciprocal nuclear genetic exchange among parasite strains undergoing mitochondrial introgression has not yet been detected, which may support an asymmetric, cryptic hybridization mechanism, or perhaps more likely, reflect the minor amount of nuclear genetic information sampled [81]. However, without the resolution of whole nuclear genome sequences, it is only possible to define the contributions of elements of meiosis, mitochondrial introgression and/or parasexual fusion [15, 82, 91]. The results presented here, include isolates from geographically distant sites (approximately 1790 km) and imply multiple introgression events occurring between different clades encompassing a large geographical area.
MLMT, the most sensitive method for assessing diversity, identified 4 groups when collection sites were used to group TcI specimens (Fig 5) or 5 clusters when no prior clustering was imposed. Three groups draw attention, one with isolates originating from Caatinga (gray branch), another from Pantanal (blue branch) and a third, consisting of an admixture of Atlantic Forest and Cerrado (Fig 6, orange and red branch). The third group contained genotypes that occurred in primates, bats, Didelphis and Rhodnius spp., in agreement with mitochondrial phylogenetic topology. There was a tendency for TcI isolates to cluster with other locally obtained isolates, which may reflect a sampling bias or clonal expansion. However, when samples were grouped according to their collection sites (Table 4), the analysis revealed specific examples of similar genotypes present across nearby states. Examples include Santa Catarina (Atlantic Forest) and Mato Grosso do Sul (Pantanal), Bahia (Atlantic Forest) and Tocantins (Cerrado), and Piaui (Caatinga) and Pará (Amazon). In stark contrast, Amazon demonstrated significant intraspecific heterogeneity (Table 2 and Table 3) and clustering indices suggest that parasites from Amazon (DAPC population 3) have undergone long-term, undisturbed, sylvatic diversification. The relative lack of human impact, particularly in some municipalities in the state of Para, may account for allelic richness evolving over time in a biome with an abundance of host species. [39]. Interestingly, DAS values from three diverse populations (Atlantic Forest, Cerrado and Caatinga) suggest the presence of intra-population sub-clusters, which is likely a consequence of the fragmentation due to intense human activity in these areas. Significant gene flow is observed over vast distances, for example between Cerrado and Atlantic Forest (Fig 1 and Table 3). The most parsimonious explanation is host movement, particularly aerial dispersion with bats, as exemplified over large distances in African clades of Trypanosoma sp. [92]. In South America, bats are known to harbor diverse trypanosome genotypes [92, 93], but their role in biogeography and dispersion is not fully understood. Unfortunately, TcI samples from Chiroptera species were collected from a single location (in Cerrado). A much more comprehensive effort to study Tc1 isolates in Chiroptera would be of interest to adequately address the nature of their role in dispersal in Brazil. Notwithstanding, we observed that D. marsupialis acts as a disperser of TcI genotypes across different biomes [94], this is evidenced by generally low FST values in pairwise comparisons with samples obtained from other hosts (FST ≤ 0.2, S10 Table). Isolates from Atlantic Forest, Amazon and Cerrado showed significantly low heterozygosity levels, which may be due to gene conversion or under sampling used in the study. In this case, processes such as inbreeding are expected to shape the genetic background of populations [94]. Indeed, isolates from the Amazon biome presented low gene flow and moderate levels of inbreeding (FIS = 0.194 ± 0.04), relatively to other biomes, indicating a degree of genetic isolation. (Table 2).
Our analyses of three classes of genetic markers revealed broadly similar patterns of intra-DTU diversity in Brazil. MLST and maxicircle marker analysis yielded two principal phylogenetic groups. One included all isolates from the Amazon region, with representatives from Cerrado and Atlantic Forest (Fig 2, clusters B and C). The second group included all other isolates from Atlantic Forest, Cerrado and part of Pantanal (Fig 2, cluster A). MLMT analysis comprising fast evolving markers, as expected, revealed the most diversity, five discrete populations and variable amounts of gene flow and fragmentation indicators. Among all biomes it is evident that Amazon harbors the most extensive diversity and comparatively low gene flow. High diversity and low fragmentation indicate a biome exposed to less ecological pressure and undisturbed sylvatic diversification. Generally, there was no clear evidence of specific host/vector associations. In particular, similar genotypes were represented in different vector/host species. For example, genotypes represented in cluster A (Fig 2) consisted of closely related genotypes observed in a diversity of hosts species including didelphids, rodents, chiroptera, primates and triatomines scattered across diverse municipalities within the Atlantic Forest, Cerrado and Pantanal biomes. Additionally, this cluster included hosts whose habitat is principally arboreal, with Didelphis spp occupying all strata. The presence of Didelphis spp. in all clades and low associated FST values (S10 Table) is compatible with the hypothesis that they are bioaccumulators of multiple genotypes [83, 94], they are highly permissive to infection and are known to move between all ecological strata from terrestrial to arboreal. The genealogical relationship of isolates in cluster A in MLST was preserved across MLMT and mitochondrial analyses (S17 and S19 Figs).
Evidence from all markers reveals that similar genotypes are found across vast geographical distances, over ecological barriers, diverse habitats, and different hosts species. Noticeably, isolates G41 (Atlantic Forest) and 2896, from Belem in the Amazon biome (Figs 1 and 2), have identical genotypes. Other examples include isolates 10272 and 11609, which possess identical genotypes despite being separated by the Marajo bay (a distance of 4500 km); and isolates from Belem (2855) and Abaetetuba (11606), which are genetically homogenous despite vast geographic separation. Human activity is likely to have an impact on the dispersal of genotypes. A case in point is provided by Combu and Murucutu, which are two island localities situated in the municipality of Belem (Amazon) that are sparsely occupied by humans and used primarily for açaí production [53]. They form a robust enzootic transmission cycle, and remote human infections are acquired by unwitting transport of infected triatomines in açai baskets [53]. Comparatively high indicators of gene flow between other biomes inferred by MLMT analysis are also compatible with the influence of human activity that may facilitate gene flow. Lima and collaborators [42] using MLMT, applied to Brazilian TcI, observed that isolates from Atlantic Forest and the Amazon formed distinct and separate clusters. Their proposition was that geographic distance separating biomes was the likely explanation for topological features. However, in this work, through the application of MLST, MLMT and maxicircle analysis, we find not only localized diversity but also genetic homogeneity over large distances. In summary, this study included a large number of samples and revealed extensive intra DTU diversity, an absence of strict associations to host/vector species, and similar genotypes circulating over vast areas. We provide evidence of genetic exchange based on phylogenetic incongruence among loci, haplotypic analysis of nuclear markers and also mitochondrial introgression. It is likely that gene flow between biomes is influenced by the movement of mammals and also facilitated by human activity.
|
10.1371/journal.ppat.1004420 | Characterization of Uncultivable Bat Influenza Virus Using a Replicative Synthetic Virus | Bats harbor many viruses, which are periodically transmitted to humans resulting in outbreaks of disease (e.g., Ebola, SARS-CoV). Recently, influenza virus-like sequences were identified in bats; however, the viruses could not be cultured. This discovery aroused great interest in understanding the evolutionary history and pandemic potential of bat-influenza. Using synthetic genomics, we were unable to rescue the wild type bat virus, but could rescue a modified bat-influenza virus that had the HA and NA coding regions replaced with those of A/PR/8/1934 (H1N1). This modified bat-influenza virus replicated efficiently in vitro and in mice, resulting in severe disease. Additional studies using a bat-influenza virus that had the HA and NA of A/swine/Texas/4199-2/1998 (H3N2) showed that the PR8 HA and NA contributed to the pathogenicity in mice. Unlike other influenza viruses, engineering truncations hypothesized to reduce interferon antagonism into the NS1 protein didn't attenuate bat-influenza. In contrast, substitution of a putative virulence mutation from the bat-influenza PB2 significantly attenuated the virus in mice and introduction of a putative virulence mutation increased its pathogenicity. Mini-genome replication studies and virus reassortment experiments demonstrated that bat-influenza has very limited genetic and protein compatibility with Type A or Type B influenza viruses, yet it readily reassorts with another divergent bat-influenza virus, suggesting that the bat-influenza lineage may represent a new Genus/Species within the Orthomyxoviridae family. Collectively, our data indicate that the bat-influenza viruses recently identified are authentic viruses that pose little, if any, pandemic threat to humans; however, they provide new insights into the evolution and basic biology of influenza viruses.
| The identification of influenza virus-like sequences in two different bat species has generated great interest in understanding their biology, ability to mix with other influenza viruses, and their public health threat. Unfortunately, bat-influenza viruses couldn't be cultured from the samples containing the influenza-like nucleic acids. We used synthetic genomics strategies to create wild type bat-influenza, or bat-influenza modified by substituting the surface glycoproteins with those of model influenza A viruses. Although influenza virus-like particles were produced from both synthetic genomes, only the modified bat-influenza viruses could be cultured. The modified bat-influenza viruses replicated efficiently in vitro and an H1N1 modified version caused severe disease in mice. Collectively our data show: (1) the two bat-flu genomes identified in other studies are replication competent, suggesting that host cell specificity is the major limitation for propagation of bat-influenza, (2) bat-influenza NS1 antagonizes host interferon response more efficiently than that of a model influenza A virus, (3) bat-influenza has both genetic and protein incompatibility with influenza A or B viruses, and (4) that these bat-influenza lineages pose little pandemic threat.
| Bats are present throughout most of the world and account for more than a fifth of mammalian species. They are natural reservoirs of some of the most deadly zoonotic viruses, including rabies virus, Ebola virus, Henipaviruses, and SARS coronavirus [1], [2]. Recently, nucleic acids obtained from bat samples indicated bats may be a reservoir of a new group of influenza viruses (bat-influenza) that are phylogenetically very distantly related to other influenza viruses [3], [4]. Type A, B, and C influenza viruses belong to the Orthomyxoviridae family and their genomes are composed of 7–8 negative sense RNA segments (vRNAs). While influenza B (IBV) and C viruses mainly infect human hosts, influenza A virus (IAV) has a broad host range; including humans, marine mammals, horses, pigs, waterfowl, and poultry. New subtypes of IAV, which have novel hemagglutinin (HA) and/or neuraminidase (NA) surface glycoproteins, are introduced into the human population by zoonosis and this periodically leads to devastating pandemics. Past pandemics include the “Spanish flu” (H1N1) in 1918, “Asian flu” (H2N2) in 1957, “Hong Kong flu” (H3N2) in 1968, “Russian flu” (H1N1) in 1977, and the recent “swine origin” flu (pH1N1) in 2009. Pandemic viruses are often reassortant viruses composed of vRNAs that are derived from multiple IAV lineages that previously circulated in swine and/or avian reservoirs (e.g., 1957 avian-human reassortant, 1968 avian-human reassortant, and 2009 avian-swine-human reassortant). The discovery of putative bat-influenza viruses expands the known host species reservoirs that may serve as a source of novel viruses, which is a major concern for public and animal health [4], [5].
Infectious bat-influenza viruses couldn't be isolated [3], [4] and although several structural and biochemical characterization studies have been conducted with the putative bat-influenza HA, NA, and part of PA, none of the vRNAs have been shown to be functional in a replicative virus [3], [4], [6]–[10]. This has led to questions such as: (1) are the putative bat-influenza vRNA sequences identified derived from infectious viruses or are they merely nucleic acid relics harbored in bats [5], (2) are the vRNA segments sequenced from a single bat-influenza virus or are they from multiple potentially incompatible viruses, and (3) were the sequences of the complete gene segments, which is a significant technical challenge, determined accurately. The inability to culture infectious viruses is the major hurdle to confirm the existence of these novel influenza viruses, and to answer important questions, such as pathogenicity in animal models, ability to reassort with other influenza viruses, and their potential risk to public health [5], [11]. The goals of this study were to synthesize the complete viral genome, characterize the bat-influenza virus using non-infectious approaches, then generate a replicative virus, and use it as a model to better understand bat-influenza viruses.
Lack of infectious particles in the original bat specimens is a potential factor in the inability to isolate/culture bat-influenza using multiple host cell substrates [3]. Based on digital sequence information published by Tong et al. [3], we synthesized the complete genome of A/little yellow-shouldered bat/Guatemala/164/2009 (H17N10) (Fig. S1) and cloned it into reverse genetics plasmids to rescue this putative bat-influenza virus (Bat09). Thousands of spherical influenza-like particles budded into the supernatants of human cells (293T) transfected with the Bat09 reverse genetics plasmids (Fig. 1A). The supernatants were inoculated into embryonated chicken eggs and cell lines derived from many species (canine (MDCK), mink (Mv1-Lu), swine (ST), African green monkey (Vero), human (A549, Calu-3), and free-tailed bat (Tadarida brasiliensis, Tb1Lu); however, none of the host cell substrates tested supported productive virus infection (determined by serial passage and subsequent real-time RT-PCR).
Previous biochemical and structural studies with purified proteins of Bat09 hemagglutinin (HA) and neuraminidase (NA) indicate that the HA doesn't bind to canonical sialic acid receptors of influenza viruses and the NA doesn't have neuraminidase activity, which is characteristic of IAV and IBV NAs [6]–[9]. To further examine if the HA and NA proteins are the major blocks to the propagation of the Bat09 virus, we attempted to rescue reassortant viruses that contained the 6 internal protein coding vRNAs (PB2, PB1, PA, NP, M, and NS) from Bat09 and the surface glycoprotein vRNAs (HA and/or NA) from a recombinant A/Puerto Rico/8/1934 (PR8). PR8 is a lab adapted H1N1 virus that has been used for many years in research and vaccine settings because it replicates efficiently in embryonated chicken eggs, cell lines (e.g., MDCK) and in the mice, but has low risk to humans. However, the three PR8-HA/NA reassortant genotypes containing the Bat09 internal protein vRNAs couldn't be rescued following transfection (Fig. 1B). While the Bat09 internal protein/vRNAs are capable of generating proteins and producing influenza-like particles, they may have critical mutations that were inhibiting infectivity, or they can't cooperate efficiently with the PR8-HA/NA proteins/vRNAs.
To further address the inability to rescue Bat09 or the Bat09:PR8-HA/NA reassortants, we created a modified HA vRNA (mH1) that contained the protein coding region from PR8-H1 flanked by putative cis-acting terminal packaging signals from Bat09 that we hypothesized would be similar to the regions known to be central to packaging of A/WSN/1933 and PR8 [12], [13] (Fig. 1C and Fig. S2). The Bat09 NA gene segment was modified using a similar strategy to replace the NA coding region with PR8-N1, while the putative bat NA packaging signals were retained (mN1) (Fig. 1C and Fig. S2). Co-expression of the mH1 and mN1 vRNAs with the six Bat09 internal protein vRNAs efficiently rescued a reassortant Bat09:mH1mN1 virus (Fig. 1B). The reassortant Bat09:mH1mN1 formed particles similar to that of Bat09 (Fig. 1A) and replicated robustly in vitro and in ovo (Fig. 1D, 1E). Next generation sequencing demonstrated that the consensus sequence of the virus stocks from 1 passage in MDCK cells or embryonated chicken eggs was identical to that of the reverse genetics plasmids. Furthermore, even after 3 passages in MDCK cells, we still didn't identify any nucleotide polymorphisms accounting for >10% of the genomic population that would suggest strong selective pressure on Bat09 genes or the modified HA/NA genes of PR8.
To investigate whether Bat09:mH1mN1 is able to infect and replicate in mice, a mouse study was performed using the mouse adapted PR8 IAV as a positive control. Bat09:mH1mN1 replicated efficiently in mouse lungs (Fig. 2A), and caused significant weight loss as early as at 4 days post inoculation (4 dpi) (Fig. 2B). The virulence of Bat09:mH1mN1 (75% mortality) was close to that of the PR8 virus (100% mortality) (Fig. 2C). Histopathological analysis showed that the Bat09:mH1mN1 virus caused typical influenza-like lesions characterized by a varying degree of broncho-alveolar epithelial degeneration and necrosis, and interstitial pneumonia. The peribronchiolar and perivascular areas were infiltrated by moderate numbers of lymphocytes and plasma cells (Fig. 2D). The histopathology identified correlates with presence of virus antigen in the mouse lungs (Fig. 2E).
Next generation sequencing was used to determine if the Bat09 vRNAs were genetically stable in mice. Although nucleotide polymorphisms (at the level of 12%–36%) were detected at sporadic loci throughout the Bat09 vRNAs, each lung sample only had one such polymorphism on average, and none of the mutations were found in more than one mouse. Nonetheless, serial passage of this virus in mice may identify mutations in the Bat09 backbone critical to replication/pathogenesis in mice. We did identify a low level nucleotide polymorphism in the modified PR8 HA at residue at 187 that emerged in multiple Bat09:mH1mN1 inoculated mouse lung samples collected at 3 and 5 dpi (HA-K187E, 10%–20% of the genomic population). This unanticipated result may have also occurred in PR8 inoculated mice; however the lung specimens from these mice were not sequenced.
The virulence of the Bat09:mH1mN1 in mice could partly result from the H1 and N1 of the mouse adapted PR8 virus. To further investigate pathogenicity of Bat09-like viruses we rescued another modified Bat09 virus that expresses H3N2 surface glycoproteins from A/swine/Texas/4199-2/1998 (H3N2) (TX98), which we have used in pigs previously [14]. The HA/NA vRNAs of Bat09:mH3mN2 were modified using a similar strategy used to generate the mH1/mN1, whereby the coding regions of Bat09 glycoproteins were replaced with TX98 H3N2, while the putative Bat09 packaging signals were retained (mH3/mN2) (Fig. 3A). The rescued Bat09:mH3mN2 virus replicated to peak titers close to that of TX98 (Fig. 3B) and both viruses were inoculated into mice to compare the morbidity (weight loss), mortality and virus replication at various times post inoculation. All mice survived infection and both viruses (Bat09:mH3mN2 and TX98) caused little effect on weight gain as compared to the mock inoculated animals (Fig. 3C), indicating little overall disease. Titration of virus in the lung tissues showed that the Bat09:mH3mN2 virus replicated as efficiently as the TX98 control in the mice at early time points, yet it appeared to be cleared more rapidly (Fig. 3D). This data suggests that some of the pathogenicity observed in the Bat09:mH1mN1 infected mice likely results from the mouse adapted HA/NA of PR8. However, it is clear that the bat influenza internal protein vRNAs do support replication of the modified viruses (Bat09:mH1mN1 and Bat09:mH3mN2) in vitro, in ovo, and in the mouse lungs. The slightly lower replication efficiency and pathogenicity of those two viruses compared to the corresponding PR8 and TX98 viruses could be ascribable to either the nature of the Bat09 internal protein vRNAs or the engineering of the modified HAs and NAs.
Bat-influenza viruses appear to have diverged from IAV a very long time ago and their internal protein vRNAs have many unique features that are not seen in IAVs [3], [4]. Therefore, the biological roles of the various vRNA segments and their protein products are likely to have both similarities and intriguing differences. Many deadly bat viruses (e.g., filoviruses) have evolved powerful molecular mechanisms that inhibit host (e.g., human) immune responses [15]–[18]. Therefore, to gain an understanding of how bat-influenza viruses may evade the host innate immune response we analyzed the Bat09 NS1 protein using interferon induction experiments and carboxy-terminal truncation mutations known to attenuate IAVs. The NS1 protein of IAVs is critical for pathogenicity of many strains because of its ability to antagonize the host interferon response [19]. To compare the direct effect of Bat09-NS1 and PR8-NS1 on interferon-β production, we expressed the proteins ectopically in human HEK-293T and then infected them with Sendai virus to stimulate the innate immune response. Activation of interferon-β promoter was determined by a luciferase mediated reporter assay [16]. Bat09-NS1 inhibited host interferon-β induction comparable to that of the PR8-NS1, and carboxy-terminal truncation of Bat-NS1 protein (NS1-128 and NS1-73, see Fig. S2C for diagram) decreased its ability to inhibit interferon-β production (Fig. 4A). These results are consistent with the attenuating effect that these NS1 truncations have on PR8 (Fig. 4A) and other IAV NS1 proteins; thereby, providing a strategy to generate live attenuated influenza vaccines [14], [20]–[22].
A VSV-luciferase virus mediated bioassay was also performed to compare the effect the NS1 truncations have on the Bat09 viruses' ability to inhibit host innate immune response [22]. The replication of the VSV-luciferase virus, which is sensitive to innate immune activation, is inversely correlated with type I interferon induced by influenza virus. Truncation of the Bat09-NS1 modestly reduced VSV replication, whereas truncation of the PR8-NS1 severely inhibited VSV replication (i.e., luciferase expression) (Fig. 4B). These results were confirmed by analysis of influenza virus replication kinetics in a human lung epithelial cell line (Fig. 4C). The Bat09-NS1 truncated viruses (Bat09:mH1mN1ss-NS1-128 and Bat09:mH1mN1ss-NS1-73) replicated to titers of 106–107 TCID50/ml (near wild type NS1; Bat09:mH1mN1ss), whereas the PR8-NS1 truncation mutants had 100–1000 fold lower titers than PR8 (Fig. 4C, Fig. S2 for gene and virus diagrams).
To analyze the impact of these Bat NS1 truncation mutations in vivo we inoculated mice with the same panel of modified Bat09 viruses, or the PR8-NS1-126 as a control. In contrast to the significant attenuation conferred by the truncated NS1 in PR8 (PR8-NS1-126), recombinant bat-influenza viruses with truncated NS1 genes (Bat09:mH1mN1ss-NS1-128 and Bat09:mH1mN1ss-NS1-73) replicated efficiently in the lungs (Fig. 5A), caused significant morbidity (Fig. 5B), and remained 100% lethal in mice (Fig. 5C). Altogether the NS1 studies show that the Bat09 NS1 protein inhibits host interferon-β production and carboxy-terminal truncation mutations reduce its ability to antagonize this response, likely through mechanisms similar to IAV (Fig. 4A). However, in contrast to IAV, truncation (NS1-128, NS1-73) of the Bat09 NS1 didn't dramatically impact the viruses' ability to antagonize the host innate response, or replicate and cause disease in mice (Fig. 4B, C and Fig. 5).
We analyzed the Bat09 PB2 gene because of its central role in the species specificity of IAVs, and some of the critical residues involved are known to be virulence determinants in mice and ferrets [23]–[28]. Asparagine (N) 701 in the PB2 protein is a mammalian-signature in IAVs and when this residue was mutated to aspartic acid (D, an avian-signature) in the modified Bat09 (Bat-701D), it decreased virus titers in lungs, morbidity (minor weight loss), and resulted in 100% survival (Fig. 6). The bat-influenza PB2 also has a serine (S) residue at position 627, which is unlike either mammalian or avian IAVs. Replacing the serine 627 with the mammalian-signature residue lysine (K) [24], [27] in the context of 701D (Bat-627K/701D) increased virus replication in the lungs but only caused slightly more weight loss (compared to the Bat-701D virus) and it remained attenuated in mice (Fig. 6). In contrast, introducing another virulence marker PB2-E158G [23] into the PB2-N701D virus (Bat-158G/701D) dramatically increased the pathogenicity of the Bat09 virus (100% mortality), which was higher than the Bat09 virus with wild type PB2 (Bat09:mH1mN1, Fig. 6). In addition, introducing the PB2-E158G (Bat-158G) into the wild type PB2 resulted the most significant increase of virus replication, morbidity, and mortality (Fig. 6), indicating there is an additive effect between the two virulence determinants (PB2-158G and PB2-701N) in the Bat09 PB2. All viruses collected from mouse lungs were deep sequenced to confirm the stability of the engineered mutations and although sporadic nucleotide polymorphisms (10% - 44%) were detected in the viral genomes (1 to 2 such polymorphisms per mouse sample on average), none of them occurred at the engineered loci. The high genetic stability of the modified Bat09 viruses in mice is consistent with the notion that the bat influenza viruses are mammalian viruses that have been evolving and adapting in the bats for a long period of time.
To determine the molecular basis for the altered pathogenicity imparted by the various mutations in the PB2 we examined their effects on the viral polymerase activity in human 293T cells using a luciferase-mediated mini-genome replication assay (Fig. 7). At all temperatures tested, the PB2-N701D mutation decreased the polymerase activity and the PB2-E158G mutation enhanced the polymerase activity, consistent with the decreased and increased pathogenicity in mice, respectively (Fig. 6). Interestingly, the PB2-627S showed intermediate polymerase activity compared to the PB2-627K and PB2-627E (Fig. 7). In addition, the polymerase activity of the PB2-158G and PB2-627E/K mutants decreased proportionally when they were combined with the PB2-701D mutation (Fig. 7). This result is consistent with the observation that Bat-158G/701D appeared to be less pathogenic than the Bat-158G virus (Fig. 6). Collectively, the data collected on the Bat09 PB2 show that amino acid residues known to be important in replication, species specificity, transmission, and/or pathogenesis of IAV are important in the replication and pathogenesis of Bat09.
Reassortment of IAVs is important in the evolution of IAVs and generation of panzootic and pandemic strains. Furthermore, efficient replication of bat-influenza internal protein vRNAs in human cells and mice, as well as their pathogenicity, necessitated an assessment of reassortment potential between Bat09 and other influenza viruses. Replication of vRNAs from different parental viruses is a factor critical in the generation of reassortant progeny. Transcription/replication of mini-genome reporter constructs showed that the viral RNA dependent RNA polymerase (RdRp), which is a heterotrimer of PB1, PB2, and PA, from bat-influenza, IAVs, and IBVs generally recognize and transcribe their cognate vRNAs more efficiently than non-cognate vRNAs (Fig. S3). Intriguingly, the Bat09 polymerase replicated the IBV reporter very efficiently (Fig. S3). Additionally, most RdRp combinations (PB2, PB1, PA) between bat-influenza and IAVs nearly abolished the polymerase activity in this very sensitive mini-genome reporter assay (Fig. 8A–I). Interestingly, the NP protein, which is a single-strand RNA-binding nucleoprotein, is completely compatible between Bat09 and IAVs (Fig. 8A–I), but it is incompatible between the bat-influenza and IBV (Fig. 8J).
Although some gene segment combinations showed limited polymerase activity in the mini-genome assays, we couldn't generate any reassortant viruses using reverse genetics between Bat09:mH1mN1 and PR8 that contain partly compatible RdRp components (e.g., Bat-PB2/PR8-PB1/PR8-PA), including the highly compatible NP vRNA/protein (Table 1 and Table 2). Instead, the PR8-M segment could unidirectionally substitute for the Bat09-M segment (Table 2). This likely results from the highly conserved nature of the M vRNA and proteins (M1, M2). Swapping the putative cis-acting packaging signals of the Bat-NP and known packaging signals of the PR8-NP, or between the Bat-NS and PR8-NS didn't enable rescue of viruses containing either the NP or NS vRNAs in a heterologous virus background (Table 3 and see Fig. S2 for diagrams).
Low efficiency of packaging at least some vRNA segments from the heterologous virus is also a major restrictive factor for reassortment. For instance, a reassortant virus containing six internal protein vRNAs from Bat09 and the HA and NA from PR8 couldn't be rescued, whereas the PR8 HA and NA coding regions flanked by Bat09 packaging regions (mH1 and mN1) can efficiently reassort with the Bat09 internal genes (Fig. 1B and Table 4). Nevertheless, PR8 HA and NA can individually reassort (7∶1) with the Bat09 six internal protein vRNAs when mN1 or mH1 were provided, respectively (Table 4). The inability to rescue the 6∶2 reassortant Bat09:PR8-H1N1 virus may result from compounding the low efficiency of packaging for each of the wild type PR8-HA and PR8-NA vRNAs into the bat-influenza backbone.
The mN1 can also reassort with the other seven segments from PR8, even when many silent substitutions (ss) were introduced into the N1 coding regions to disrupt the remaining PR8 packaging signals (Table 4). Actually, another modified NA that contains the coding region from IBV NA flanked by the putative packaging region of the Bat09-NA (Bat-N10ps-FluB-NA) can also be rescued in the PR8 background, strongly suggesting that the bat-influenza NA segment could be efficiently packaged into the PR8 virus, whereas the Bat09 HA packaging signal didn't mediate efficient packaging of the mH1 into the PR8 backbone (Table 4).
While the generation of reassortants through plasmid-based reverse genetics is a powerful and sensitive way to rescue influenza viruses, it's difficult to generate every possible gene constellation and accompanying minor nucleotide variations that could give rise to progeny reassortants during co-infection. Therefore, we attempted to generate reassortants between a modified Bat09 virus and PR8 using a classical co-infection approach. However, when MDCK cells were inoculated at a high multiplicity of infection (MOI) with both PR8 and Bat09:mH1mN1 viruses, reassortment between the two parental viruses was not detected. We plaque purified 118 progeny viruses from the co-infection and 53 of them were the parental PR8 virus and 65 of them were the parental Bat09:mH1mN1 virus. Although more exhaustive classical reassortant studies are needed to completely evaluate the generation of natural reassortants between these viruses, the data indicate that PR8 and Bat09:mH1mN1 don't efficiently reassort.
Recently, another bat-influenza virus A/flat-faced bat/Peru/033/2010 (H18N11) (Bat10) was identified in Peru and phylogenetic analysis indicated this virus diverged from the bat-influenza viruses identified in Guatemala (e.g., Bat09) so long ago that genetic diversity between these two bat-influenza viruses is higher than that of IAVs [4]. Reassortment of the PB2, PB1, PA, and NP segments in mini-genome polymerase activity assay demonstrated that the Bat09 and Bat10 viruses were fully compatible (Fig. 8K). Most importantly, successful reassortment between the two modified bat viruses (Bat09:mH1mN1ss and Bat10:mH1mN1ss) (Table 5 and Fig. S2 for diagrams of constructs) proved that these genetically divergent bat-influenza virus vRNAs were highly interchangeable, in contrast to their very low compatibility with IAV and IBV. Interestingly, classical co-infection of the Bat09:mH1mN1 and Bat10:mH1mN1 viruses in MDCK cells readily generated reassortant progeny viruses with various genotypes, and some were apparently preferentially selected (e.g., Bat10:Bat09-NS reassortant, Table S2), demonstrating the merit of classic co-infection strategy in identification of gene constellations that may have certain advantages. Collectively the mini-genome replication, reverse genetics reassortment, and co-infection reassortment experiments strongly suggest that two divergent bat-influenza viruses readily reassort with each other, whereas they won't reassort with canonical IAVs in the natural setting.
The generation of synthetic modified bat-influenza viruses (e.g., Bat09:mH1mN1) that grow to high titers in commonly used influenza virus culture substrates and mice is an important step toward understanding these novel bat-influenza viruses. The rescue of Bat09:mH1mN1 and Bat09:mH3mN2 viruses demonstrates that the putative vRNAs of Bat09 function efficiently together and are probably derived from either one virus, or a group of compatible viruses, whose PB2, PB1, PA, NP, M, and NS proteins efficiently replicate and package vRNAs in host cells commonly used to culture influenza viruses (Fig. 1). Importantly, the data also shows that the bat-influenza HA and NA were the sole determinants inhibiting Bat09 virus rescue, and that the terminal regions of HA and NA of bat-influenza viruses selected for our constructs contain cis-acting vRNA packaging signals. Although wild type bat-influenza virus (Bat09) couldn't be propagated in the human, canine, mink, avian, porcine or bat cell lines we tested, consistent with Tong et al. [3], it is likely that the bat-influenza virus can infect some other cell cultures from other species and/or tissues, especially cells derived from appropriate bat species.
Our Bat09:mH1mN1 studies provide other unique insights, which can't be gleaned from non-infectious assays. For instance, non-infectious assays (interferon-β reporter assay, Fig. 4A) showed the Bat09 NS1 carboxy-terminal truncationss (NS1-128 and NS1-73) were similar to the truncated PR8 NS1 (NS1-126 and NS1-73), which largely lost the ability to inhibit the host interferon response. However, mouse experiments with the replicative bat-influenza viruses revealed that the truncation of Bat09 NS1 had minimal effects on the viral pathogenesis compared to the truncation of PR8 NS1 (Fig. 5). Differences in the attenuating impact observed in the PR8-NS1 and the Bat09-NS1 truncated viruses suggests that Bat09 has novel molecular mechanisms that have evolved in the amino terminal portion of NS1 and/or other internal protein vRNAs to antagonize/evade the host innate immune response.
The PB2 of IAV plays important roles in replication, species specificity, transmission, and pathogenesis [23]–[29]. Our analysis of bat-influenza PB2 demonstrated that it is also a virulence determinant and as anticipated conversion of mammalian-signature residues at position 701 to avian-signature (N701D) attenuated the virus, and the E158G substitution [23] enhanced virulence. PB2-627 is one of the most studied positions differentiating avian viruses (glutamic acid) and mammalian viruses (lysine) [24], [27]. Intriguingly, the bat-influenza PB2 has a serine at position 627, which is unlike mammalian or avian IAVs. Our data show that PB2-627S has intermediate polymerase activity compared to PB2-627E and PB2-627K in mammalian cells, suggesting an alternative evolutionary pathway that avian influenza viruses may be able to take for mammalian adaptation.
Reassortment of the segmented genomes of Orthomyxoviruses is a powerful evolutionary mechanism that is central to the success of these pathogens. Viruses within a Genus readily reassort upon co-infection of a single host cell (e.g., avian and swine IAV); whereas, viruses from a different Genus (e.g., IAV and IBV) don't reassort. The factors important for generation of reassortant progeny from two parental influenza viruses include: recognition and replication of vRNAs by parental virus RdRp, protein-protein interaction/compatibility (e.g, heterotrimeric RdRp), and vRNA-protein interactions needed for virion morphogenesis. The RNA transcription/replication promoter of each influenza vRNA segment is formed by base pairing of highly conserved nucleotides at the 5′ and 3′ termini, which form a partially double-stranded structure. The IAV Genus has specific nucleotide variations within the termini that distinguish it from IBV. The termini of bat-influenza vRNAs also show conserved 5′ and 3′ complementarity; however, they also have distinct nucleotide variation. Therefore, we used mini-genome replication studies to analyze promoter recognition and RdRp activity of various combinations of the PB1, PB2, PA subunits in combination with various NPs from IAV, IBV, or bat-influenza. The data show that the wild type RdRp most efficiently replicate their cognate vRNAs, and that both IAV and IBV RdRp have 50–60% reduction in activity with the bat-influenza mini-genome. Many PB1, PB2, PA combinations between bat-influenza and IAV/IBV dramatically reduce activity, which demonstrates protein-protein incompatibility between the RdRp subunits. Interestingly, the bat-influenza NP and IAV NP were completely compatible in the mini-genome assay, however NP reassortant viruses could not be generated (Table 2 and Table 3) suggesting that the incompatibility of NPs may also involve complicated protein-vRNA interactions.
IAVs of various subtypes can infect and reassort in bat cell lines [30], [31], providing a permissive environment for them to reassort with bat-influenza viruses. However, our reassortant analysis indicates that while two divergent bat-influenzas readily reassort, bat-influenza and IAVs don't easily reassort in co-infection experiments. Reverse genetics reassortment studies showed the PB2, PB1, PA, NP, and NS vRNAs of bat-influenza don't efficiently reassort with the IAV or IBV, and provide many additional tantalizing results. For example, reassortants were not rescued from relatively compatible RdRp combinations in the mini-genome assay (e.g. Bat-PB2/PR8-PB1/PR8-PA, Fig. 8A) and demonstrate that divergent Bat09 and Bat10 can efficiently reassort with each other (Table 5). The M segment is the most highly conserved gene among influenza A and B viruses. We found that the PR8-M segment could substitute for the Bat09-M segment (Table 2), indicating that the M vRNAs/protein(s) of PR8 and Bat09 have enough conservation in both cis-acting packaging signals and functional domains of the proteins (M1/M2) to enable the replication of the modified Bat09 virus. In contrast, putative packaging signal swapping of the NP and NS segments didn't overcome reassortment defects suggesting that incompatibility at the protein-protein or protein-vRNA level is likely to be a critical factor inhibiting reassortment between the bat-influenza and other influenza viruses. Alternatively, one could argue that that since the vRNA packaging signals of bat-influenza NP and NS segments have not been delineated, the putative packaging regions incorporated in the Batps-PR8 constructs may not be sufficient for packaging the modified vRNAs. However, the well-defined PR8 packaging signals incorporated in our modified gene segments should be sufficient to package the corresponding bat-influenza NP and NS vRNAs (PR8ps-Bat-NP and PR8ps-Bat-NS, Fig. S2D) in the PR8 backbone. The failure to rescue the PR8ps-Bat NP or NS viruses, as well as the PR8:Bat09-M reassortant virus, strongly suggests protein-protein or protein-vRNA level incompatibility and provides a unique opportunity to better understand the functional domains of these proteins through characterizing chimeric/mosaic proteins containing motifs/domains from both viruses.
Another caveat with our bat-influenza reassortment experiments is the focus on interactions with the laboratory adapted PR8 virus, which was chosen primarily due to biosafety concerns. Reassortment between the Bat09:mH1mN1 virus and other IAVs, particularly avian viruses (e.g., H5N1, H7N9) that appear to be more compatible in the mini-genome assay (Fig. 8), are needed to fully assess reassortment potential of bat-influenza. However, based on our results from the NP reassortment and the Bat-PB2/PR8-PB1/PR8-PA reassortment experiments (Table 1 and Table 2), the likelihood of rescuing a reassortant with RdRp components from both Bat and IAVs is very low. Finally, since the HAs and NAs of the bat influenza viruses can't be used to rescue viruses using contemporary influenza virus host substrates, we were not able to fully assess the ability of the HA or NA to reassort with other influenza viruses (limited assessment provided in Table 4). However, the known bat influenza viruses (Bat09, Bat10) could pose a pandemic threat if their HA and NA acquire mutations that impart binding to canonical influenza virus receptors and rescuing the NA for neuraminidase activity, or acquisition of binding and entry through alternative human cell surface receptors.
Collectively, our experiments suggest that the bat-influenza virus is unlikely to reassort with an IAV or IBV and spread to other species even if they were to infect the same host cell. The restriction on reassortment appears to result from multiple levels of incompatibility (RNA-RNA, RNA-protein, and/or protein-protein) that are either additive or synergistic. Consequently, our data suggest that due to the extremely limited ability of genetic information exchange between bat-influenza and IAV or IBV, the International Committee on Taxonomy of Viruses could consider classifying these two bat-influenza virus lineages as a new Genus or Species within the Orthomyxoviridae.
This study also demonstrated the power of synthetic genomics in rapid characterization and risk assessment of an emerging virus, even when the virus itself is not readily cultured. The synthetic genomics/reverse genetics strategy employed provides an infinite supply of wild type bat-influenza particles that can be used to identify permissive cells or animals. The availability of our modified bat-influenza virus, opens many other avenues of investigation and discovery, including, for instance, to gain a better understanding of cis-acting signals in the vRNAs that are important in bat-influenza transcription, replication, packaging/particle morphogenesis, and to use forward genetics to elucidate viral protein-protein and/or viral protein-host protein interactions. Finally, continued study of bat-influenza viruses and integration of data from other contemporary influenza viruses is important in the elucidation of the evolutionary history of influenza viruses.
The study was reviewed and approved by the Institutional Biosafety Committee at Kansas State University (protocol #903), and by the institutional biosafety committee at the J. Craig Venter Institute (protocol # 3414). We conducted the initial studies using PR8 gene fragments to generate the modified bat-influenza viruses and to test the reassortment potential because PR8 is a widely used lab/mouse adapted BSL2 virus that poses very low risk to humans or livestock. Subsequently, TX98 H3N2 genes were used in a few experiments because this is a BSL2 swine virus, which we have used previously and the viruses generated were considered low risk.
The animal studies were performed in strict accordance with the recommendations in the Guide for the Care and Use of Laboratory Animals of the National Institutes of Health. The animal protocol (protocol #3339) was reviewed and approved by the Institutional Animal Care and Use Committee at Kansas State University. All animal studies were performed in a Biosafety Level 3 facility located at the Biosecurity Research Institute at Kansas State University under the approved protocol #3339 following the American Veterinary Medicine Association guidelines on euthanasia. For virus inoculation, each mouse was anesthetized by inhaling 4% isoflurane. Mice were euthanized if more than 25% of weight was lost after virus inoculation. Euthanasia of mice was conducted by inhaling 4% isoflurane followed by cardiac puncture and cervical dislocation. No survival surgery was performed, and all efforts were made to minimize suffering.
Human embryonic kidney 293T (HEK-293T) cells, mouse rectum epithelial carcinoma (CMT-93) cells, and African green monkey kidney (Vero) cells were maintained in Dulbecco's modified Eagle's medium (DMEM) supplemented with 10% fetal bovine serum (FBS). Madin-Darby canine kidney (MDCK) cells were maintained in minimum essential medium (MEM) supplemented with 5% FBS. Human lung epithelial (A549) cells, bat lung epithelial (Tb1Lu) cells, mink lung epithelial (Mv1Lu) cells and swine testis (ST) cells were maintained in MEM supplemented with 10% FBS. Human lung epithelial (Calu-3) cells were maintained in MEM supplemented with 10% FBS, 1% nonessential amino acids, and 1 mM sodium pyruvate.
Nucleotide sequences of the eight gene segments of A/little yellow-shouldered bat/Guatemala/164/2009 (H10N17) (Bat09) were retrieved from the GenBank database. A total of 472 oligonucleotides of 56–60 bases in length were designed for enzymatic assembly of the eight segments. The assembly and error correction processes were performed as recently described [32], [33], modified with increased time at all extension steps (from 72°C for 1 min to 72°C for 2 min) for efficient assembly of the polymerase segments. The synthesized segments (Fig. S1) were cloned into the modified bidirectional influenza reverse genetics vectors pBZ66A12 [34] using the recombination-based method [35] and transformed into Stella competent E. coli cells (Clontech). Colonies were selected and sequenced. The appropriate clones for each segment were propagated for plasmid preparation and verified by sequencing. The resulting plasmids are pBZ146A1 (PB2), pBZ147A11 (PB1), pBZ148A20 (PA), pBZ149A30 (HA), pBZ150A31 (NP), pBZ151A36 (NA), pBZ152A42 (M) and pBZ153A45 (NS). The whole process only took seven days to complete. The plasmids containing Bat09 PB2 mutations were constructed by site-directed mutagenesis using the pBZ146A1 as template. The NS1 truncation constructs were generated by Gibson assembly and details of the truncations are diagramed in Fig. S2C. The modified (m) Bat09 HA and NA (mH1, mN1, mH1ss, and mN1ss, see Fig. S2A, S2B for diagrams, and Fig. S4 for sequence alignment) were synthetized by Gibson assembly from oligonucleotides. Silent substitutions (ss) were introduced to disrupt the putative packaging signals in the PR8 HA and NA terminal coding regions. The mH1ss and mN1ss are thus more appropriate than the mH1 and mN1 to assess the HA and NA packaging signal compatibility between Bat09 and PR8. The Batps-PR8-NP, PR8ps-Bat-NP, Batps-PR8-NS, and PR8ps-Bat-NP constructs were constructed similarly and diagramed in Fig. S2D. As a comparison of the speed of different synthesis strategies, the eight gene segments of A/flat-faced bat/Peru/033/2010 (H18N11) (Bat10) were synthesized by Genewiz (NJ, USA) in the vector plasmid of pUC57 based on the GenBank database and subcloned into pHW2000 vector. The resulting plasmids (pHW-H18-PB2, pHW-H18-PB1, pHW-H18-PA, pHW-H18-HA, pHW-H18-NP, pHW-H18-NA, pHW-H18-M and pHW-H18-NS) were confirmed by sequencing. The whole process took more than one month. The PB2, PB1, PA and NP genes were also subcloned into the pDZ vector for use in the mini-genome assay. Diagrams of the mutant or modified genes of Bat09 and Bat10 are described in Fig. S2. The pPol1-NS-Luc reporters used in the mini-genome polymerase activity assay were described in Fig. S2E. Sequences of all constructs used in this study were confirmed to ensure absence of unwanted mutations and the GenBank accession numbers are KM203345-KM203356.
Briefly, 0.6 µg of plasmid for each gene segment was mixed and incubated with 15 µl of Mirus TranIT-LT1 (Mirus Bio, Madison, WI) at 20°C for 20 min. The transfection mixture was transferred to 90% confluent 293T/MDCK cell monolayers in a 35-mm tissue culture dish and incubated at 37°C with 5% CO2 for 8 h. The transfection supernatant was replaced with 3 ml of Opti-Mem I medium (Life Technologies) supplemented with 0.3% bovine serum albumin (BSA) fraction V (Life Technologies), 3 µg/ml tosylsulfonyl phenylalanyl chloromethyl ketone (TPCK)-trypsin (Worthington, Lakewood, NJ), and 1% antibiotic-antimycotic (Life Technologies). Three days post-transfection, culture supernatant (passage 0, P0) was collected and 0.5 ml of that was inoculated into MDCK cells in 6-well plates at 37°C. Supernatant (P1) was collected at 4 days post-inoculation (dpi), or when severe cytopathic effect (CPE) was observed. The P1 supernatant was further passaged blindly for two passage before determined to be negative for rescue. Titers of the viruses used in this study were determined by TCID50 assay in MDCK cells.
Rescue efficiency definition. Very easy (++++): P0 viral titer 106–108 TCID50/ml, or severe CPE observed in P1 within 1 dpi; Moderate (+++): P0 titer 104–106 TCID50/ml, or obvious CPE observed in P1 within 2 dpi; Difficult (++): P0 titer 102–104 TCID50/ml, or weak CPE observed in P1 within 4 dpi; Very difficult (+): P0 titer lower than 102 TCID50/ml, or CPE not observed until P2/P3; Negative (Neg): rescue failed, no CPE observed through passage 3.
Various transfection conditions including different transfection reagents, temperatures, and incubation time before supernatant collection were attempted to rescue the wild type Bat09 virus and the reassortants between Bat09 and PR8. However, none of them generated any positive rescue results if they were negative under standard rescue condition described above. Bat09 transfection supernatants were also transferred to various cells (MDCK, mink lung Mv1-Lu, swine testis, Vero, A549 cells, Calu-3, bat lung epithelial Tb1Lu) and embryonated chicken eggs and passaged at least three times. The real-time RT-PCR assays targeting Bat09 and PR8 M genes were used to confirm negative results (primers and probes are possible upon request).
To determine whether virus particles of Bat09 and other viruses can be produced by reverse genetics system, a total of thirty-five ml of transfected 293T cell supernatants for each virus were collected at 48 hours post transfection and centrifuged at 8000 rpm for 20 minutes to remove the cell debris. Then the clear supernatant was loaded on 30% (w/v) sucrose in centrifuge tubes and was concentrated at 27,000 rpm (Optima LE-80K ultracentrifuge, Beckman Coulter) for 2 hours. The virus pellets was dissolved in 100 µl of water and the viral particles were fixed by incubating with 0.2% paraformaldehyde at 37°C for 48 hours. The fixed particles were dipped on a 200 mesh copper grid and the grid was dried and stained with negative staining before observation under an electron microscope.
MDCK monolayers in 12-well plates were washed twice with PBS, and then 2 ml of virus growth medium (VGM) was added to each well. The cells were inoculated at a multiplicity of infection (MOI) of 0.01 TCID50/cell with the Bat09:mH1mN1 virus or PR8 virus (Bat09:mH3mN2 virus or TX98 virus) and incubated at 37°C. Supernatants were collected at 1, 2, and 3 days post inoculation (dpi). Inoculations of Calu-3 cells were performed similarly, except that an MOI of 0.02 TCID50/cell was used for the following viruses: Bat09:mH1mN1ss, Bat09:mH1mN1ss-NS1-73, Bat09:mH1mN1ss-NS1-128, PR8, PR8-NS1-73, and PR8-NS1-126. The VGM used for MDCK cells was EMEM supplemented with 0.15% BSA fraction V, 2 µg/ml TPCK-trypsin, and 1% antibiotic-antimycotic, and the VGM used for Calu-3 cells was EMEM supplemented with 0.3% BSA fraction V, 1 µg/ml TPCK-trypsin, and 1% antibiotic-antimycotic. All virus titers were determined by TCID50 assay using MDCK cells.
Six of 10-day-old embryonated chicken eggs were inoculated with Bat09:mH1mN1 or PR8 at 103 TCID50/egg. After 2 days incubation at 35°C, allantoic fluid was collected from each egg and titrated individually. The 4 eggs with the highest titers in each virus group was used to calculate the average titer and generate the graph in Fig. 1E.
A modified Multi-segment RT-PCR [35], [36] was used to amplify influenza-specific segments. The only modification to the procedure was the primers used for amplification were changed to match bat influenza termini. The oligonucleotide primers used were Uni12/Inf-5G (5′-GGGGGGAGCAGAAGCAGG-3′) and Uni13/Inf-1 (5′-CGGGTTATTAGTAGAAACAAGG-3′). The M-RTPCR amplicons were used for Illumina Miseq library construction via Nextera DNA sample prep kit (Illumina, Inc.) and sequenced using the Illumina MiSeq (Illumina, Inc.) according to manufacturer's instructions. SNP variations were identified using custom software that applies statistical tests to minimize false positive SNP calls that could be caused by the types of sequence-specific errors that may occur in Illumina reads identified and described in Nakamura, et al. [37]. To overcome this problem, the protocol requires observing the same SNP, at a statistically significant level, in both sequencing directions. Once a minimum minor allele frequency threshold and significance level are established by the user, the number of minor allele observations and major allele observations in each direction and the minimum minor allele frequency threshold are used to calculate a p-value based on the binomial distribution cumulative probability, and if the p-values calculated in each of the two sequencing directions are both less than the Bonferroni-corrected significance level, then the SNP call is accepted. For our analyses, we used a significance level of 0.05 (Bonferroni-corrected for tests in each direction to 0.025), and a minimum minor allele frequency threshold of 10% of the read population.
To measure the IFN-antagonist function of NS1, a luciferase-based, Sendai virus-mediated IFN-β promoter activation assay was conducted as previously described [16]. Briefly, 293T cells in 24-well plates were transfected with empty vector (200 ng) or increasing amounts of wild type (WT) or carboxyl terminal truncated NS1 from Bat09 and PR8 (2 ng, 10 ng, and 50 ng of NS1 expression plasmids supplemented with 198 ng, 190 ng, and 150 ng of empty vector, respectively). Also co-transfected were 200 ng of an IFN-β-promoter-luciferase reporter plasmid (pIFNβ-Luc) and 20 ng of a plasmid constitutively expressing Renilla luciferase (pRL-TK from Promega). At 18 hours post transfection, cells were infected with Sendai virus to induce the IFN-β promoter. A dual-luciferase assay was performed at 18 hour post virus inoculation, and firefly luciferase was normalized to Renilla luciferase activity. The relative luciferase activity of the group with empty vector was set as 100%, and the other groups were presented relative to that.
As previously described for the VSV-GFP virus mediated interferon bioassay [22], in the VSV-Luciferase virus mediated bioassay, A549 cells were inoculated with one of the wild type or NS1 truncated viruses at an MOI of 4 TCID50/cell, or were mock-inoculated; supernatants were then collected at 24 hpi. Supernatants were treated with UV irradiation to inactivate viruses and were then transferred to naïve A549 cells. Following 24 h of incubation at 37°C, supernatants were removed, and the cells were inoculated with VSV-Luciferase virus [38], at an MOI of 2 TCID50/cell. The firefly luciferase expression in the cells was measured using the Luciferase Assay System (Promega) at 4 hpi with VSV-Luciferase.
The luciferase-mediated mini-genome polymerase activity assay was performed as previously described, using a PolI-driven reporter plasmid and pDZ-based PB2, PB1, PA, and NP bidirectional expression plasmids [21], [35]. To determine the effects of PB2 mutations on polymerase activity (Fig. 7) 293T cells were co-transfected with 0.2 µg each of the PB2 (WT or mutant), PB1, PA, NP, and a pPol1-FluA-NS-Luc (firefly luciferase flanked by A/New York/1682/2009 [23]). As a control for transfection efficiency, 0.02 µg of the Renilla luciferase plasmid pRL-TK (Promega) was also co-transfected. After 18 hours of incubation at 33°C, 37°C, and 39°C, luciferase production was assayed using the dual-luciferase reporter assay system (Promega) according to the manufacturer's instructions. Firefly luciferase expression was normalized to Renilla luciferase expression (relative activity). The relative activity of the PB2-WT was set as 1 fold, and the relative activities of the PB2 mutants were presented relative to that (Fig. 7).
To test the compatibility between RNPs (PB2, PB1, PA, and NP) and viral RNA promoters from bat-influenza virus (Bat09) (Fig. S3), IAV (A/PR/8/1934), and IBV (B/Russia/1969), 293T cells were co-transfected with 0.2 µg each of the PB2, PB1, PA, NP, and a pPol1-NS-Luc reporter plasmid, followed by incubation at 37°C for 18 hours. Three reporters were used in this study, including pPolI-Bat-NS-Luc (firefly luciferase flanked by Bat09 NS non-coding regions), pPol1-FluA-NS-Luc, and pPolI-FluB-NS-Luc (firefly luciferase flanked by B/Russia/1969 NS non-coding regions) (Fig. S2D). For each combination of RNP and pPolI-NS-Luc reporter (from Bat09, A, or B Type), three independent replicates were conducted. For each RNP, the luciferase activity with the reporter from the same virus (e.g., Bat-RNP and pPol1-Bat-NS-Luc) was set at 100%, and the activities with the other two reporters (e.g., pPol1-FluA-NS-Luc and pPol1-FluB-NS-Luc) were presented relative to that (Fig. S3).
The PB2, PB1, PA, and NP compatibility between Bat09 and the following influenza viruses was examined in the study (Fig. 8): A/PR/8/1934 (lab adapted human H1N1), A/Ann Arbor/6/1960 (human H2N2), A/New York/238/2005 (human H3N2); A/New York/1692/2009 (human H1N1 seasonal), A/New York/1682/2009 (human H1N1 pandemic), A/canine/New York/6977983/2010 (canine H3N8), A/turkey/Ontario/7732/1966 (avian H5N9), A/Hong Kong/213/2003 (avian H5N1), A/Anhui/1/2013 (human H7N9), B/Russia/1969 (lab adapted human IBV), and A/flat-faced bat/Peru/033/2010 (bat H18N11). For the compatibility test between Bat09 and IAVs (Fig. 8A–I), 293T cells were co-transfected with 0.2 µg each of the PB2, PB1, PA, NP (from Bat09 or IAV), 0.1 µg of pPolI-Bat-NS-Luc plasmid and 0.1 µg of pPolI-FluA-NS-Luc. For compatibility test between Bat09 and IBV (Fig. 8J), 293T cells were co-transfected with 0.2 µg each of the PB2, PB1, PA, NP (from Bat09 or B/Russia/1969), 0.1 µg of pPolI-Bat-NS-Luc plasmid and 0.1 µg of pPolI-FluB-NS-Luc. For compatibility test between Bat09 and Bat10 (Fig. 8K), 0.2 µg each of the PB2, PB1, PA, NP (from Bat09 or Bat10), and pPolI-Bat-NS-Luc plasmids were used (The NS non-coding regions of Bat09 and Bat10 have the same sequence). Renilla luciferase was also co-transfected and dual-luciferase reporter assay system was used. For each combination of PB2, PB1, PA, and NP (from Bat09 or another influenza virus), three independent replicates were conducted at 37°C, the luciferase activity of the all-Bat09-combination (Bat09-PB2/Bat09-PB1/Bat09-PA/Bat09-NP) was set at 100%, and the activities of other 15 combinations were presented relative to that.
A total of 98 female BALB/c mice aged 6 to 7 weeks were randomly allocated to 7 groups (14 mice/group). Six mice were intranasally inoculated with 103 TCID50 of each virus (Bat 09:mH1mN1, Bat09:mH1mN1-PB2-701D, Bat09:mH1mN1-PB2-627K701D, Bat09:mH1mN1-PB2-158G701D, Bat09:mH1mN1-PB2-158G, PR8, or MEM Mock) in 50 µL fresh MEM medium while under light anesthesia by inhalation of 4% isoflurane. To determine the virus replication in mouse lungs, three mice from each group were euthanized on both 3 and 5 day post-inoculation (dpi). Another 8 mice from each group were intranasally inoculated with 104 TCID50 of viruses in 50 µL MEM medium; all eight mice were kept to monitor body weights and clinical signs. Weights were recorded daily and general health status was observed twice daily. After the onset of disease, the general health status was observed three times daily. Severely affected mice (i.e., more than 25% body weight loss) were euthanized immediately, and the remaining mice were euthanized on 14 dpi. All control mice were intranasally inoculated with 50 µL fresh MEM (mock group), three control mice were necropsied at 3 and 5 dpi, the remaining mice were kept until the end of the animal study.
During necropsy, the right part of the lung was frozen at −80°C for virus titration, and the left part of the lung was fixed in 10% formalin for histopathologic examination. For virus titration, the 10% lung homogenate was prepared in cold fresh MEM medium by using a Mini Bead Beater-8 (Biospec Products; 16 Bartlesville, OK). The homogenate was centrifuged at 6000 rpm for 5 minutes, and the supernatant was titrated by infecting MDCK cells in 96-well plates. For the histopathologic examination, lung tissues fixed in 10% phosphate-buffered formalin were processed routinely and stained with hematoxylin and eosin. The lungs were examined microscopically both for the percentage of the lung involved and for the histopathologic changes seen, including bronchiolar and alveolar epithelial necrosis, intraalveolar neutrophilic inflammation, peribronchiolar inflammation, and bronchiolar epithelial hyperplasia and atypia. For detection of virus NP antigens in lung sections on day 5 post infection, a rabbit anti-H1N1 (2009 flu pandemic) NP polyclonal antibody was used (Genscript, USA). A pathologist examined each slide in a blinded fashion.
A total of 70 female BALB/c mice aged 6 to 7 weeks were randomly allocated to 5 groups (14 mice/group). To determine virus replication, six mice were intranasally inoculated with 104 TCID50 of each virus (Bat09:mH1mN1ss-NS1-WT, Bat09:mH1N1ss-NS1-73, Bat09:mH1mN1ss-NS1-128, and PR8-NS1-126) in 50 µL MEM medium while under light anesthesia by inhalation of 4% isoflurane. Three mice from each group were killed on both 3 and 5 day post-inoculation (dpi). Another 8 mice from each group were intranasally inoculated with 105 TCID50 of each virus in 50 µL MEM medium for morbidity and mortality comparison. All the other procedures are same with described previously.
A total of 42 female BALB/c mice aged 6 to 7 weeks were randomly allocated to 3 groups (14 mice/group). To investigate virus replication in mice, six mice from each group were intranasally inoculated with 3×104 TCID50 of virus or mock-inoculated with 50 µL fresh MEM medium while under light anesthesia by inhalation of 4% isoflurane. Three of six inoculated mice from each group were euthanized at 3 and 5 day post-inoculation (dpi). To evaluate viral pathogenicity in mice, the remaining eight mice from each group were intranasally inoculated with 3×105 TCID50 of virus (Bat09:mH3mN2, and TX98) in 50 µL fresh MEM medium or mock-inoculated with 50 µL fresh MEM medium. The mice were monitored body weights and general health status daily. After the onset of disease, the general health status was observed twice per day. Severely affected mice (i.e., more than 25% body weight loss) were humanly euthanized, and the remaining mice were euthanized and bloods were collected from each mouse to isolate serum for the HI assay at 14 dpi. Sample collection and analysis, and virus titration were performed as described above.
To study the reassortment between Bat09:mH1mN1 and PR8 or Bat10:mH1mN1, confluent monolayer of MDCK cells in 6-well-plates were co-infected with both viruses (Bat09:mH1mN1 and PR8, or Bat09:mH1mN1 and Bat10:mH1mN1). Both modified Bat09:mH1mN1 and Bat10:mH1mN1 viruses showed similar replication kinetics in MDCK cells, whereas the PR8 replicated more efficiently than both modified viruses in MDCK cells. Therefore, for the co-infection study with PR8 and Bat09:mH1mN1 viruses, the cells were infected with the PR8 at MOI of 1 and with the Bat09:mH1mN1 at MOI of 4 (a ratio of both viruses is 1∶4). For the co-infection study with Bat09:mH1mN1 and Bat10:mH1mN1 viruses, the cells were infected with each virus at MOI of 1 (a ratio of both viruses is 1∶1). The co-infected MDCK cells were incubated at 37°C with 5% CO2 for 1 hour. After 1 hour of incubation, the supernatant was removed and the infected cells were washed with fresh MEM for 10 times. One mL of infection medium supplemented with 1 µg/mL TPCK-trypsin (Worthington, Lakewood, NJ) was added on cells. The supernatant containing progeny viruses was collected at 24 hours after inoculation. Plaque assays were performed in MDCK cells to select single virus from co-infected supernatants. The purified single virus (plaque) was amplified for further analysis. To identify the origin of each gene of the purified single virus, specific RT-PCR was used to differentiate internal genes from Bat09:mH1mN1, Bat10:mH1mN1 and PR8 viruses (primers for specific RT-PCR are available upon request). The surface HA and NA genes were differentiated by sequencing HA and NA non-coding regions (packaging signals) since three parental viruses contain identical HA and NA ORF sequences and different sequences in non-coding region (it is difficult to differentiate them by RT-PCR). For the RT-PCR, RNAs were extracted from each amplified single virus using a QIAamp Viral RNA Mini Kit (Qiagen). cDNA was synthesized by using the bat universal 12 primer (5′-AGCAGAAGCAGG-3′) for the samples from the co-infection study with Bat09:mH1mN1 and Bat10:mH1mN1 viruses, and by using a mixture of an IAV universal 12 primer (5′-AGCRAAAGCAGG-3′) and the bat universal 12 primer (5′-AGCAGAAGCAGG-3′) for the samples from the co-infection study with Bat09:mH1mN1 and PR8 viruses. If the origin of internal genes determined by the specific RT-PCR was inconclusive, sequencing was performed to confirm the results from specific RT-PCR (All sequence primers are available upon request).
Luciferase activity, virus titers, and mouse weights were analyzed by using analysis of variance (ANOVA) in GraphPad Prism version 5.0 (GraphPad software Inc, CA). One-way ANOVA with Dunnett's multiple comparison test was used to determine the significance of the differences (P<0.05) among different groups. For simple comparisons, Student's t test was used to examine the significance of differences observed. Error bars represent standard deviation (±SD).
|
10.1371/journal.ppat.1007381 | Intestinal Serum amyloid A suppresses systemic neutrophil activation and bactericidal activity in response to microbiota colonization | The intestinal microbiota influences the development and function of myeloid lineages such as neutrophils, but the underlying molecular mechanisms are unresolved. Using gnotobiotic zebrafish, we identified the immune effector Serum amyloid A (Saa) as one of the most highly induced transcripts in digestive tissues following microbiota colonization. Saa is a conserved secreted protein produced in the intestine and liver with described effects on neutrophils in vitro, however its in vivo functions remain poorly defined. We engineered saa mutant zebrafish to test requirements for Saa on innate immunity in vivo. Zebrafish mutant for saa displayed impaired neutrophil responses to wounding but augmented clearance of pathogenic bacteria. At baseline, saa mutants exhibited moderate neutrophilia and altered neutrophil tissue distribution. Molecular and functional analyses of isolated neutrophils revealed that Saa suppresses expression of pro-inflammatory markers and bactericidal activity. Saa’s effects on neutrophils depended on microbiota colonization, suggesting this protein mediates the microbiota’s effects on host innate immunity. To test tissue-specific roles of Saa on neutrophil function, we over-expressed saa in the intestine or liver and found that sufficient to partially complement neutrophil phenotypes observed in saa mutants. These results indicate Saa produced by the intestine in response to microbiota serves as a systemic signal to neutrophils to restrict aberrant activation, decreasing inflammatory tone and bacterial killing potential while simultaneously enhancing their ability to migrate to wounds.
| The intestine is colonized by complex microbial communities called the microbiota, which impacts diverse aspects of host physiology including development and function of the innate immune system. Neutrophils are phagocytic innate immune cells essential for host defense against infection. Neutrophil activity is strongly impacted by the microbiota but underlying mechanisms remain poorly defined. Here we show the evolutionarily-conserved secreted host protein Serum amyloid A (Saa) mediates microbiota-dependent effects on systemic neutrophil function. Saa is produced by the intestine and liver in response to the microbiota, but its in vivo functions have remained elusive. Using zebrafish, we demonstrate that Saa promotes neutrophil recruitment to peripheral injury yet restricts clearance of pathogenic bacterial infection. Analysis of isolated neutrophils revealed Saa reduces bactericidal activity and expression of pro-inflammatory genes in a microbiota-dependent manner. Transgenic expression of saa in the intestine and liver of saa mutants was sufficient to rescue mutant phenotypes, and intestinally-derived saa also alleviated defects in neutrophil recruitment to peripheral injury in germ-free zebrafish. Collectively, these data establish that Saa induced by the microbiota in the intestine signals systemically to neutrophils, tuning the extent to which they may be activated by other microbes or respond to injury.
| The vertebrate intestine is densely colonized with complex communities of micro-organisms, collectively referred to as the intestinal microbiota. Studies using gnotobiotic animals have demonstrated that microbiota colonization is required for the normal development of an innate immune system capable of mounting appropriate responses to diverse challenges such as infection and injury [1]. Despite being spatially confined to the intestinal lumen by physical and chemical barriers such as the intestinal epithelium and mucus, the microbiota influences both local and systemic host immune development and function [2, 3]. However, the specific molecular mechanisms by which the microbiota impacts local and systemic host immune responses remain poorly defined.
Intestinal epithelial cells (IECs) serve as the primary host interface with the intestinal microbiota and secrete a myriad of factors following microbial colonization [4]. We and others have hypothesized that these microbiota-induced IEC products may mediate the microbiota’s influences on the host immune system [5, 6]. Previous studies have identified a secreted host factor, Serum amyloid A (Saa), that is potently up-regulated in the intestine following microbial colonization in zebrafish and mice [7–11]. SAA is highly conserved amongst vertebrates, existing as a single gene in fishes and birds and as a multi-gene family in mammals [12, 13]. While basal SAA production is stimulated by the microbiota, SAA production is also markedly augmented following acute injury and infection as a part of the acute phase response, whereby circulating levels can reach 1 mg/mL [13–15]. Moreover, SAA is elevated in chronic pathological conditions where both local and circulating concentrations are positively correlated with inflammation. Accordingly, SAA is an established biomarker for chronic inflammatory diseases such as diabetes, atherosclerosis, and inflammatory bowel disease (IBD) [16–20]. Taken together, SAA’s high degree of evolutionary conservation coupled with its strong induction following inflammatory stimuli suggests important roles for SAA in health and disease.
Previous studies have reported both pro- and anti-inflammatory effects of SAA on host immune responses. The vast majority of these studies have been conducted in vitro using recombinant human SAA (rhSAA), which has been shown to directly influence granulocytes, including monocytes and neutrophils, promoting the production of inflammatory cytokines, reactive oxygen species (ROS), and directing motility [21–25]. Recent reports have also shown that mammalian SAAs can bind retinol and mediate host responses during infection [8]. Moreover, induction of SAA in the intestine following colonization with specific microorganisms such as segmented filamentous bacteria (SFB) can shape local adaptive immune cell development by promoting Th17 differentiation [26, 27]. However, a fuller assessment of SAA’s in vivo functional roles has remained elusive due to the existence of multiple SAA gene paralogs in mammals (3 in humans, 4 in mice) and the use of cell-culture based assays performed with rhSAA that behaves dissimilarly to endogenous SAA protein [28, 29].
The existence of a single SAA ortholog in fishes provides interesting opportunities to define SAA’s in vivo functional roles. We previously demonstrated that partial (~30%) knockdown of saa transcript in zebrafish influenced tissue-specific neutrophil behaviors in vivo, leading us to hypothesize that Saa regulates neutrophil activity in homeostasis [12]. However, Saa’s influence on systemic neutrophil function in homeostasis and in the relevant contexts of injury and infection remained unresolved. Neutrophils are professional phagocytic myeloid cells that play critical roles in host defense against pathogens. The most abundant immune cell in circulation and the first to be recruited to sites of injury, neutrophils eliminate microbial invaders and debris through a variety of mechanisms including phagocytosis, generation of ROS, and secretion of anti-microbial peptides [30, 31]. Neutrophils are conditioned by host and microbially derived signals, including pathogen associated molecular patterns (PAMPs) and damage associated molecular patterns (DAMPs), allowing for proper responses to inflammatory stimuli [32, 33]. Previous studies have shown that microbial colonization of the intestine promotes neutrophil differentiation, activation, and response to peripheral injury [10, 12, 34–39]. However the signaling molecules that mediate these interactions in vivo remain largely unknown.
Here, using zebrafish, we demonstrate that Saa is a host factor that signals microbial status in the intestine to extra-intestinal populations of immune cells and directs their responses to inflammatory stimuli. Zebrafish share highly conserved hematopoietic programs with other vertebrates, including specification of myeloid lineages, which can be coupled with optical transparency and genetic tractability to allow for high resolution in vivo imaging of innate immune processes [40, 41]. Moreover, the zebrafish genome encodes a single saa ortholog, allowing us to generate the first-ever saa null vertebrate model. By comparing wild-type (WT) zebrafish to those that lack saa or express it only in intestinal epithelial cells or hepatocytes, we reveal Saa’s impact on systemic neutrophil activity in homeostasis and following bacterial infection and wounding. Using tissue specific rescue, we demonstrate that liver- and intestinally-derived Saa can shape systemic neutrophil function and intestinal Saa can even restore neutrophil defects observed in germ-free zebrafish.
To investigate Saa’s effects on neutrophil function in vivo, we first generated saa mutant zebrafish, identifying three independent deletion alleles all resulting in frameshift mutations within saa exon 2 (S1A–S1C Fig). The largest saa deletion allele (22 bp, designated rdu60, homozygous mutants hereafter referred to as saa-/-) resulted in 90% reduced saa mRNA (S1D–S1F Fig). saa-/- zebrafish survived to adulthood and exhibited no gross developmental abnormalities (S1H–S1L Fig). We performed caudal fin amputations on WT and saa-/- Tg(lyz:EGFP) larvae and quantified neutrophil recruitment to the wound margin, observing fewer neutrophils at 6 hours post-wounding in saa-/- larvae (Fig 1A and 1B, S1N Fig). Further, in vivo imaging revealed that neutrophils in the vicinity of the wound moved with reduced mean speed in saa mutant larvae (Fig 1C). Importantly, saa mRNA was not upregulated at 6 hours post amputation in WT larvae (S1M Fig), demonstrating acute saa induction did not affect neutrophil activity. Thus, Saa is required for neutrophil mobilization to sites of injury independent of systemic induction.
Given that Saa affects neutrophil responses to injury, we tested if Saa regulates basal neutrophil behavior. Analysis of neutrophils in homeostasis revealed Saa promoted neutrophil speed in the trunk and linear migration in the intestine (Fig 1G–1J), consistent with our prior saa morpholino knockdown data [12]. Considering saa is expressed in IECs [10], we reasoned that Saa may promote neutrophil recruitment to the intestine. Indeed, we observed fewer intestine-associated neutrophils in saa-/- larvae (Fig 1D–1F). These data demonstrate that Saa promotes neutrophil recruitment to distinct tissues such as the intestine during homeostasis as well as to peripheral tissues following injury.
Given that saa loss is associated with impaired neutrophil recruitment to wounds and healthy tissues, we asked whether systemic neutrophil abundance is altered in saa mutants. We enumerated systemic neutrophils by flow cytometry (Fig 2A, S2 Fig) and consistently observed elevated neutrophil abundance in saa-/- larvae. This was corroborated by increased expression of the granulocyte marker genes lysozyme C (lyz), l-plastin (lcp), and the granulopoietic cytokine colony stimulating factor 3a (csf3a, also known as gcsf1a) in 6 days post fertilization (dpf) whole larvae (Fig 2D–2F). Morphological classification of lyz+ neutrophils into sub-populations from cytospin preparations (adapted from [42]) revealed an over-representation of immature lyz+ neutrophils in saa deficient animals compared to WT controls (Fig 2B and 2C). Together, these data demonstrate a novel role for Saa regulating neutrophil maturation in vivo.
Exposure to PAMPs or inflammatory host molecules can elicit defined neutrophil transcriptional responses, reflecting their activation state [43–47]. Gene expression analysis of FACS-isolated neutrophils revealed elevated expression of genes encoding pro-inflammatory cytokines (tnfa, il1b), antimicrobial peptides (pglyrp2, pglyrp5), and regulators of ROS production (mpx, ncf1) in saa-/- larvae (Fig 3A). These transcriptional differences suggest Saa restricts basal neutrophil activation. As neutrophils primarily function to clear microbial infections [48, 49], we co-cultured neutrophils isolated from adult zebrafish kidneys with non-pathogenic Escherichia coli then assessed bacterial viability. Isolated neutrophils from both WT and saa-/- fish were viable ex vivo and exhibited morphological responses to bacteria (e.g., extending cytosolic projections; Fig 3B, S3A–S3E and S3J Fig). Co-culture with E. coli induced il1b mRNA in WT neutrophils, demonstrating zebrafish neutrophils respond transcriptionally to bacteria ex vivo (Fig 3C). Moreover, il1b transcript levels were significantly increased in unstimulated saa mutant adult kidney neutrophils compared to WT, consistent with our observations from larval neutrophils (Fig 3A and 3C). However, following co-culture with E. coli, il1b in WT neutrophils reached similar levels measured in saa-/- neutrophils (Fig 3C).
To assess neutrophil bactericidal activity, we enumerated CFUs following 4 hours of co-culture and found saa-/- neutrophils killed significantly more bacteria than WT neutrophils (Fig 3D). To interrogate possible mechanisms of bacterial clearance, we labeled neutrophils with CellROX ex vivo and measured levels of intracellular ROS by confocal microscopy (S3F and S3G Fig). Bacterial exposure resulted in decreased ROS in both WT and saa-/- neutrophils, indicating bacteria stimulated neutrophil degranulation (Fig 3E). Interestingly, neutrophils from saa mutant animals had elevated levels of ROS relative to WT controls both at baseline and after bacterial stimulation (Fig 3E), which is consistent with their augmented bacterial killing activity (Fig 3D). Collectively, these data indicate neutrophils from saa mutants are aberrantly activated as evidenced by elevated pro-inflammatory mRNA expression, augmented bactericidal activity, and elevated ROS production, and suggest Saa restricts systemic neutrophil inflammatory tone in vivo.
Since saa is highly expressed in the larval zebrafish intestine and liver following microbiota colonization [10], we reasoned that Saa produced by specific tissues might differentially affect systemic neutrophil conditioning. Saa produced by the liver enters circulation and has systemic affects [50], however the distal influences of intestinally-derived saa remained unresolved. We generated a transgenic line to drive expression of zebrafish saa in hepatocytes using a 2.8 kb promoter fragment from the zebrafish fabp10a gene [Tg(-2.8fabp10a:saa);cmlc2:EGFPrdu66, hereafter referred to as Tg(fabp10a:saa)] [51–54]. We crossed this transgene into the saa mutant background and assessed whether hepatic saa was sufficient to rescue neutrophil defects observed in saa mutant animals.
Despite elevated levels of saa transcript in saa-/-;Tg(fabp10a:saa) relative to saa-/- mutant larvae (Fig 4A), we still observed reduced intestinal neutrophil recruitment compared to WT controls (Fig 4B). To test the impact of liver-derived saa on neutrophil function, we quantified neutrophil recruitment to the wound margin following caudal fin amputation. Neutrophil recruitment at 6 hours post wounding was restored to WT levels in saa-/-;Tg(fabp10a:saa) larvae, demonstrating that hepatic saa is sufficient to restore neutrophil response to peripheral injury (Fig 4C). To further investigate the effects of hepatic-derived Saa on neutrophil function, we quantified bactericidal activity of neutrophils from WT, saa-/-, and saa-/-;Tg(fabp10a:saa) zebrafish. Whereas E. coli is cleared quickly by the zebrafish immune system, Pseudomonas aeruginosa is capable of establishing systemic infections [55, 56]. Interestingly, transgenic hepatic saa was sufficient to restore P. aeruginosa killing activity of isolated neutrophils to WT levels in ex vivo co-culture (Fig 4D). These data establish that saa produced in the liver is capable of influencing a distinct subset of systemic neutrophil functions, promoting their ability to migrate to wounds yet restricting bacterial killing.
Given that the intestine serves as a primary interface between the host and microbiota, and that saa is transcriptionally upregulated in the intestine following microbial colonization, we sought to determine if intestinally-derived saa could also influence systemic neutrophil phenotypes. To do so, we first needed to engineer transgenic zebrafish in which saa is expressed specifically in IECs. While several zebrafish promoters have been identified with intestine-restricted activity, they only drive transgene expression in subsets of IECs (e.g., the fabp2/ifabp promoter is active in enterocytes in the anterior intestine) [10, 57]. Since zebrafish cldn15la is expressed more broadly and specifically in IECs [58, 59], we queried adult zebrafish IEC FAIRE-seq data and identified a 349 bp open chromatin region in the cldn15la promoter which contains predicted binding sites for intestine-specific transcription factors including Cdx2 (S4A and S4B Fig) [60]. To confirm intestinal specificity of this cldn15la promoter fragment, we used this element to drive expression of mCherry [Tg(-0.349cldn15la:mCherry)], and observed mCherry fluorescence restricted to IECs (Fig 4E and 4F, S4C–S4F Fig, S5A–S5E Fig) [60, 61]. Gene expression analysis of enterocyte marker fabp2 and pan-IEC marker cldn15la in FACS-isolated mCherry+ and negative cells from Tg(-0.349cldn15la:mCherry) larvae revealed strong enrichment in the mCherry+ fraction, further validating the IEC specificity of this promoter (Fig 4G, S4G and S4H Fig).
We next generated Tg(-0.349cldn15la:saa;cmlc2:EGFP) zebrafish [subsequently denoted Tg(cldn15la:saa)], which express zebrafish saa in IECs (S6A and S6B Fig). To confirm that saa expression was restricted to IECs, we crossed Tg(-0.349cldn15la:mCherry) and Tg(cldn15la:saa) adults, then sorted mCherry+ IECs and measured levels of saa expression in mCherry+ and negative fractions from WT and Tg(cldn15la:saa) 6 dpf larvae (S4G and S4H Fig). We observed significantly higher saa levels in mCherry+ IECs isolated from Tg(cldn15la:saa) larvae compared to WT IECs and both negative fractions (Fig 4H). This data indicates that saa expression in Tg(cldn15la:saa) larvae is driven specifically in IECs.
We crossed this transgene into the saa mutant background, and asked whether IEC-derived Saa could complement saa-/- neutrophil defects. As expected, we observed fewer intestine-associated neutrophils in each intestinal segment of saa-/- larvae (from anterior to posterior, segments 1–3). Intestinally-derived Saa was sufficient to partially complement this mutant phenotype, with an increased abundance of intestine-associated neutrophils in all segments (Fig 4I and 4J). Thus Saa produced in IECs, unlike hepatic Saa, is sufficient to promote neutrophil recruitment to the intestine, suggesting Saa is a neutrophil chemoattractant in vivo. To determine if systemic neutrophil function was altered by intestinally-derived Saa, we performed caudal fin amputations. At 6 hours post amputation, neutrophil recruitment to the wound in saa-/-;Tg(cldn15la:saa) larvae was equivalent to the WT response (Fig 4K), demonstrating intestinally-derived saa is sufficient to restore neutrophil mobilization to the caudal fin in otherwise saa deficient larvae.
To test the effects of intestinally-derived Saa on neutrophil function, we measured bactericidal activity of neutrophils from WT, saa-/-, and saa-/-;Tg(cldn15la:saa) zebrafish. We observed reduced survival of both P. aeruginosa and E. coli following co-culture with saa-/- neutrophils vs WT controls. However, neutrophils from Tg(cldn15la:saa)+ saa-/- zebrafish exhibited comparable bactericidal activity to WT neutrophils (Fig 4L, S6D Fig). Additionally, following P. aeruginosa co-culture, neutrophils from WT, saa mutant, and saa-/-;Tg(cldn15la:saa) zebrafish exhibited comparable il1b transcript induction (Fig 4M), demonstrating that neutrophils from each genotype are responsive to bacterial stimulation ex vivo. These findings demonstrate intestinally-derived Saa is sufficient to constrain bactericidal activity of neutrophils.
Considering the profound effects of saa levels and source on neutrophil antibacterial function ex vivo, we asked if intestinal Saa influenced bacterial clearance in vivo using systemic P. aeruginosa infection. As neutrophils play an important role in P. aeruginosa clearance in vivo [62, 63], we postulated that differences in bacterial clearance between WT and saa deficient larvae would be largely driven by differences in neutrophil activation and behavior. Larval zebrafish were injected with P. aeruginosa via the caudal vein to achieve systemic infection and bacterial burden was assessed from whole zebrafish larvae. Consistent with our ex vivo results, we observed enhanced bacterial clearance in saa-/- larvae which is restored to WT levels in saa-/-;Tg(cldn15la:saa) larvae (Fig 4N).
Flow cytometry revealed that transgenic intestinal Saa did not significantly alter systemic neutrophil abundance in saa-/- larvae whereas liver derived Saa returned neutrophil abundance to WT levels (S7A and S7C Fig). Moreover, qRT-PCR analysis of neutrophils isolated from both saa-/-;Tg(cldn15la:saa) and saa-/-;Tg(fabp10a:saa) larvae revealed persistently elevated expression of pro-inflammatory and anti-microbial effectors (pglyrp2, tnfa, il1b, ncf1) (S7B and S7D Fig). Thus, liver and intestinal expression of saa was sufficient to rescue only a subset of neutrophil defects observed in saa mutant animals. These observations highlight the potential requirement for other tissue sources, presentation, or temporal control of Saa to condition different aspects of systemic neutrophil function. Collectively, these data demonstrate that expression of saa in the intestinal epithelium and liver is sufficient to promote neutrophil recruitment to a peripheral wound and to restrict bactericidal activity in vivo and ex vivo, but is unable to dampen elevated neutrophil pro-inflammatory mRNA profiles observed in saa mutant zebrafish. These results further resolve the local and systemic roles for intestinal Saa, promoting neutrophil recruitment to the intestine as well as shaping systemic neutrophil migratory and bacterial killing activity.
We and others have shown that the intestinal microbiota influences a variety of neutrophil phenotypes both in homeostasis and following injury [12, 36–38]. Since our results indicate that Saa suppresses neutrophil activation (Fig 3A, 3D and 3E) and previous studies have reported Saa has direct bactericidal activity [64–66] we speculated that saa deficiency may impact microbiota composition. We used 16S rRNA gene sequencing to compare bacterial communities in the digestive tracts of co-housed saa-/- and WT sibling zebrafish at both larval and adult stages. We found that saa genotype had no significant effects on gut bacterial community composition as measured by alpha- or beta-diversity metrics at larval (6 dpf) or adult (70 dpf) stages (S2 and S3 Tables). These findings indicate that Saa does not broadly impact gut microbiota composition in zebrafish.
Since saa is potently induced following microbial colonization (Fig 5A), we asked if Saa regulates neutrophil activation in response to the microbiota. We isolated neutrophils from 6 dpf gnotobiotic WT and saa-/- larvae, and found expression of pro-inflammatory mRNAs was significantly elevated in WT neutrophils from conventionalized (CV) zebrafish vs germ-free (GF) controls, confirming microbiota-derived signals induce neutrophil pro-inflammatory mRNA expression (Fig 5B). Comparison of the same transcripts in neutrophils from saa-/- zebrafish reared under CV and GF conditions revealed augmented induction of mRNAs following microbiota colonization. Since neutrophil pro-inflammatory mRNA levels were comparable between WT GF and saa-/- GF larvae, we conclude the microbiota potentiates transcriptional activation observed in saa-/- larvae (Fig 5B). To assess the impact of the microbiota on neutrophil function in WT and saa mutant animals, we measured neutrophil recruitment to peripheral tail wound injury in GF larvae. Unlike our previous observations in conventionally reared larvae (Fig 1B), neutrophil recruitment to the fin wound was indistinguishable in GF saa-/- and GF WT larvae at 6 hours post wounding (S8A Fig). This indicates that neutrophil recruitment defects only manifest in saa-/- mutants when colonized with a microbiota. These data further demonstrate that Saa functions to restrict aberrant activation of neutrophils by the microbiota at homeostasis, thus allowing neutrophils to respond appropriately to injury.
Given that the microbiota induces saa expression in the intestine, and GF WT zebrafish larvae have diminished neutrophil responses to wounding compared to conventionalized WT siblings [10], we asked if transgenic saa expression in IECs was sufficient to rescue neutrophil deficiencies in GF animals. Consistent with previous findings, we observed reduced neutrophil wound recruitment to caudal fin amputation in GF WT larvae vs CV controls [12]. Transgenic intestinal saa expression in GF larvae rescued neutrophil mobilization to the wound margin 6 hours after injury (Fig 5C). These data support a working model wherein Saa produced by IECs in response to the microbiota promotes local recruitment of neutrophils to the intestine while limiting aberrant systemic neutrophil activation by the microbiota, thereby allowing neutrophils to effectively respond to subsequent challenges such as peripheral injury (Fig 5D).
Previous reports in mice and zebrafish have demonstrated that the microbiota influences diverse aspects of neutrophil biology, including increased abundance and longevity, enhanced wound recruitment, and elevated bacterial killing [10, 12, 34–36, 38, 39, 67]. However, few studies conclusively identify host or microbial factors that mediate the microbiota’s influences on neutrophils. Here we demonstrated that Saa induced following colonization transduces information regarding intestinal microbiota, regulating diverse aspects of neutrophil biology (Fig 5D). In colonized zebrafish at homeostasis, Saa promoted neutrophil maturation and mobilization to the intestine while suppressing systemic neutrophil abundance and pro-inflammatory gene expression. During inflammatory challenge, Saa restricted neutrophil anti-bacterial activity yet promoted neutrophil recruitment to a wound. While in vitro studies suggest rhSAA (rhSAA; Peprotech, Rocky Hill, NJ, USA) promotes pro-inflammatory cytokine (TNFα, IL-1β, and IL-8) and ROS production [23, 68–72], the pro-inflammatory effects of rhSAA on granulocytes contrasts with reports using purified endogenous SAA [28, 29]. Thus, our in vivo analysis clarifies conflicting reports of Saa’s effects on neutrophils.
Our results establish Saa as a major host factor mediating microbial control of neutrophil function. Expression of saa is negligible in germ-free zebrafish larvae, but is potently induced following microbial colonization [7, 10]. We therefore reasoned that neutrophil defects in colonized saa-/- larvae would phenocopy those exhibited in WT germ-free larvae [10, 12, 73, 74]. In accord, we observed decreases in neutrophil migratory behavior, intestinal association, and wound recruitment in GF WT and colonized saa-/- larvae. We demonstrated intestinally-derived Saa was sufficient to partially complement neutrophil deficiencies in saa-/- zebrafish and could restore neutrophil wound recruitment in defects GF WT larvae. These results indicate that intestinally-derived Saa conditions neutrophils in vivo following microbiota colonization. As we have previously shown that colonization with different bacterial taxa leads to varied levels of saa induction [11], this signaling axis may facilitate taxa-specific effects on host innate immune development. Further, it is possible that mammalian SAA paralogs similarly condition neutrophils, as a subset of Saa genes are induced in mouse intestinal tissues following microbial colonization [7–9, 64]. As broad-spectrum antibiotics usage can dramatically impact the microbiota [75–77] and antibiotic treatment results in reduced intestinal Saa in mice [78], our model predicts antibiotic treatment could be associated with Saa-mediated aberrations in neutrophil function. We speculate that secondary infections that can occur following antibiotic use [79] could be due in part to concomitant alterations in SAA production.
The mammalian SAA gene family is comprised of both constitutively-expressed and acutely-inducible forms. There are four described paralogs in humans and mice, with much focus placed upon the acute forms which are upregulated following inflammatory stimuli as part of the acute phase response [13]. Following injury or trauma, hepatic production of the canonical acute SAA paralogs, SAA1 and SAA2, is dramatically induced. These secreted proteins enter circulation and associate with high density lipoproteins (HDL) [80], although the functional consequences of this interaction remain unclear. Notably, these acute forms are not only expressed in the liver, as they have also been detected in tissues including the small and large intestine where IECs produce them in response to microbial and inflammatory stimuli [7, 8, 26, 64, 81–83]. In contrast, SAA4 is constitutively expressed in the liver and is not induced by the microbiota or injury [13, 81, 84]. Our previous and present analyses indicate that the single zebrafish saa homolog is inducible by microbial stimuli with negligible constitutive expression in germ-free animals, suggesting zebrafish have minimal or no constitutive Saa activity.
Whole animal Saa1-/-Saa2-/- double knockout mice (which still possess Saa3 –an acute extra-hepatic SAA that is a pseudogene in humans) and whole animal Saa3-/- knockout mice are both sensitive to chemically-induced enterocolitis [64, 85], suggesting SAA protects against intestinal inflammation. Moreover, following colonization with segmented filamentous bacteria (SFB), induction of SAA1/2 in mouse small intestinal epithelial cells stimulates expression of effector cytokines (IL-17) in T helper cells and promotes Th17 cell expansion and mucosal defense [26, 27, 86]. Our data indicate that IEC-derived Saa is important for neutrophil recruitment to the larval zebrafish intestine, which may play an important homeostatic role in host defense, mitigating breaches in barrier by commensal or pathogenic microbes. Mouse and human SAAs have also been described as retinol binding proteins that can bind dietary vitamin A with important implications in bacterial infections [8]. Furthermore, expression of SAA in IECs and the liver of mice requires dietary vitamin A [8]. We speculate that vitamin A, the microbiota, and perhaps other environmental factors interactively regulate SAA expression and subsequently influence the development and function of the innate immune system.
In this study we uncover tissue specific influences of Saa on both local and systemic neutrophil biology. Although our transgenic lines express saa at levels above endogenous saa transcript (Fig 4A and 4I, S6A and S6B Fig), we believe these transgenic levels are still within the physiologic range as circulating levels of Saa protein can reach 1 mg/mL [13–15]. Using these tissue specific promoters allowed us to disentangle the contribution of different tissue sources of Saa on host neutrophil responses. Our data indicate that Saa from the intestine and liver can impact neutrophil activities, such as promoting neutrophil recruitment to peripheral injury and suppressing bacterial killing. However, only Saa produced from IECs was sufficient to promote local neutrophil recruitment to the intestine. We speculate that both intestinal and hepatic Saa shape systemic neutrophil responses to injury by entering circulation from their respective origins. In zebrafish, microbiota induction of Saa is NF-κB and Myd88 dependent [10], and microbial factors, including LPS and flagellar function, are sufficient to induce Saa [11, 87]. Future studies aimed at delineating tissue specific induction of Saa by microbial signals are needed to understand how specific bacterial factors promote Saa production. Differential activity of SAA from the intestine versus liver may be due in part to differences in protein complex formation and downstream receptor recognition, as mammalian SAAs are known to associate with other proteins in HDL particles [88]. Our model suggests that in cases of chronic intestinal inflammation (e.g. inflammatory bowel disease) when SAA is expressed at high levels in the intestine, there may be SAA-mediated impacts on host systemic innate immune function [16–18].
Saa is thought to directly and indirectly influence in epithelial homeostasis in the mammalian intestine through direct bactericidal activity in the intestinal lumen and by shaping host mucosal innate and adaptive immunity [27, 64, 65]. Studies in mice have demonstrated that acute SAAs are expressed in IECs, and can be secreted both into the lumen and basolaterally, potentially aiding in the clearance of pathogenic bacteria [8, 64, 81, 83]. We did not detect significant alterations in intestinal microbiota composition in our saa mutant zebrafish when co-housed with WT controls. However, it remains possible that saa mutation may impact microbiota composition or density when genotypes are housed separately.
Our discovery that Saa functions in the microbiota-neutrophil axis motivates interest in underlying mechanisms. Neutrophil functions are largely regulated through “priming”, whereby microbial products (e.g. LPS, peptidoglycan, and flagellin) and host factors (e.g. TNFα, IL-1β, IL-8, GM-CSF, and ATP) signal to neutrophils, preparing them for response to additional stimuli [32, 89]. Intriguingly, many priming factors are induced in the zebrafish and mammalian intestine by the microbiota (e.g. TNFα, IL-1β, and GM-CSF) [7, 8, 10, 60, 81]. Primed neutrophils exhibit enhanced bacterial killing, altered motility, and transcriptional changes [33]. Once believed to be transcriptionally quiescent, several studies have reported neutrophil transcriptional responses to various priming stimuli in vitro [43–45, 47] and in vivo [46, 90, 91]. We demonstrate Saa restricts neutrophil pro-inflammatory gene expression and show induction of pro-inflammatory genes in neutrophils from colonized vs GF WT zebrafish, which is augmented in saa-/- larvae. Consistent with our gene expression results, Saa limits primed neutrophil phenotypes of ROS production and bacterial killing in vivo and ex vivo. We propose Saa is upregulated following microbiota colonization to temper aberrant neutrophil priming by microbial products and production of host inflammatory mediators, thus limiting collateral damage to host tissues [92–94]. Similarly, a recent report using Saa3 knockout mice demonstrated that SAA3 suppresses bone marrow derived dendritic cell response to LPS [95], mirroring the anti-inflammatory effects of Saa on zebrafish neutrophils demonstrated here. It will be interesting to determine if SAA’s pleiotropic effects on diverse cell types are mediated by different receptors, oligomeric state, or binding other molecules such as retinol [96].
Our data reveal that Saa promotes neutrophil maturation in adult zebrafish. A recent study showed that mature neutrophils exhibit increased motility and response to stimuli [97]. While it is possible that altered neutrophil maturation may underlie the phenotypes we observe in saa mutants, some maturation-associated phenotypes (e.g. ROS production and bacterial killing) were elevated in neutrophils from saa-/- zebrafish. Thus, Saa may differentially affect neutrophils at different stages of development and maturation.
Collectively, our findings highlight the importance of intestinal and hepatic Saa effecting systemic neutrophil development and function, suppressing their inflammatory tone and increasing mobilization to wounds. More broadly, our findings suggest that the ontogenetic and microbial control of priming factors is important for vertebrate immunological development.
Zebrafish studies were approved by the Institutional Animal Care and Use Committees of Duke University Medical Center (protocol number A115-16-05) in accordance with the Public Health Service Policy on the Human Care and Use of Laboratory Animals under the Unites States of America National Institutes of Health (NIH) Office of Laboratory Animal Welfare (OLAW).
All zebrafish lines were maintained on a mixed Tübingen (Tü) / TL background on a 14:10 hour light:dark cycle in a recirculating aquaculture system. From 5 dpf to 14 dpf, larvae were fed Zeigler AP100 <50-micron larval diet (Pentair, LD50-AQ) twice daily and Skretting Gemma Micro 75 (Bio-Oregon, B5676) powder once daily. Beginning at 14 dpf, larvae were fed Artemia (Brine Shrimp Direct, BSEACASE) twice daily, supplemented with a daily feed of Skretting Gemma Micro 75. From 28 dpf, the Gemma Micro 75 diet was replaced with Gemma Micro 300 (Bio-Oregon, B2809). At the onset of breeding age or sexual maturity, adult fish were given a 50/50 mix of Skretting Gemma Micro 500 (Bio-Oregon, B1473) and Skretting Gemma Wean 0.5 (Bio-Oregon, B2818) and one feeding of Artemia daily.
Larvae were also maintained on a 14:10 hour light:dark cycle in a 28.5°C incubator, and are of indeterminate sex. Gnotobiotic zebrafish were generated following natural mating and reared as described previously [98] with the following exception: GZM with antibiotics (AB-GZM) was supplemented with 50 μg/ml gentamycin (Sigma, G1264). Conventionally raised zebrafish were maintained at a density of ≤ 1 larva / mL, and at 3 dpf groups of 30 larvae were transferred to 10 cm petri dishes containing 20 mL gnotobiotic zebrafish media (GZM), inoculated with 3 mL filtered system water (5μm filter, SLSV025LS, Millipore) and fed autoclaved ZM-000 zebrafish diet (1% w/v stock concentration (in RO H2O), 0.0025% w/v final concentration, Zebrafish Management Ltd.) [10]. TgBAC(cldn15la:EGFP)pd1034Tg (which expresses a Cldn15la-EGFP fusion protein), Tg(lyz:DsRed)nz50 and Tg(lyz:GFP)nz117 have been previously characterized [40, 58].
Targeted deletion of the saa gene was performed using CRISPR/Cas9 gene editing targeting the second exon of saa as described [7]. Briefly, the guide RNA sequence was designed using the “CRISPR Design Tool” (http://crispr.mit.edu/). Guide RNA oligos (S1 Table, primers P1 and P2) were ligated into pT7-gRNA plasmid (Addgene, 46759), which, following BamHI (New England Biolabs, R0136L) digest, was in vitro transcribed using MEGAshortscript T7 kit (ThermoFisher, AM1354) [99]. Cas9 mRNA was generated from XbaI (New England Biolabs, R0145S) digested pT3TS-nls-zCas9-nls plasmid (Addgene, 46757), and in vitro transcribed using mMESSAGE mMACHINE T3 kit (ThermoFisher, AM1348) [99]. A cocktail consisting of 150 ng/μL of nls-zCas9-nls and 120 ng/μL of gRNA, 0.05% phenol red, 120 mM KCl, and 20 mM HEPES (pH 7.0) was prepared, and approximately 1–2 nL was injected directly into the cell of one cell stage Tü zebrafish embryos. Mutagenesis was initially screened using Melt Doctor High Resolution Melting Assay (ThermoFisher, 4409535), and identified three independent alleles which were confirmed as deletions by Sanger sequencing of TOPO-cloned PCR products. Subsequent screening of the Δ22 (allele designation rdu60) was performed by PCR (primers P3 and P4, S1 Table) and products were resolved on 2% agarose TBE gels. Screening of the Δ2 and Δ5 alleles (allele designations rdu61 and rdu62 respectively) was performed by PCR amplification using primers P3 and P4 (S1 Table), followed by purification and Hha1 digest (New England Biolabs, R0139S). All mutant alleles (rdu60, rdu61, and rdu62) result in the loss of a single Hha1 restriction site present in the WT sequence.
For generation of transgenic zebrafish expressing saa in intestinal epithelial cells (IECs) or liver hepatocytes, the following strategy was employed utilizing Tol2 mediated transgenesis (https://onlinelibrary.wiley.com/doi/abs/10.1002/dvdy.21343). A 349 bp region of the zebrafish cldn15la promoter was PCR amplified from Tü genomic DNA using primers P5 and P6 (S1 Table), digested with FseI and AscI (New England Biolabs, R0588 and R0558), and ligated into the p5E 381 vector that had been linearized with the same enzymes to generate p5E-0.349cldn15la. The -2.8 kb fabp10a p5E vector was kindly provided by Brian Link. The full-length zebrafish saa coding sequence was PCR amplified from Tübingen whole-larvae cDNA using primers P7 and P8 (S1 Table) and subcloned into the plasmid pENTR-AleI using In Fusion (Takara Bio, 638909) to generate pME-saa. Both p5E-0.349cldn15la and pME-saa were verified by PCR and Sanger sequencing. A 4-way Gateway LR reaction was performed using LR Clonase II (ThermoFisher, 12538120) to recombine p5E-0.349cldn15la or p5E-2.8fabp10a, with pME-saa, and p3E-polyA 229 into pDEST 395 (which contains a bicistronic cmlc2:EGFP reporter), yielding the following constructs: -0.349cldn15la:saa:polyA;cmlc2:EGFP and -2.8fabp10a:saa:polyA;cmlc2:EGFP. These plasmids were co-injected with transposase mRNA into Tü embryos at the single cell stage respectively, as described elsewhere [60]. Injected F0 larvae were subsequently screened for mosaic EGFP expression in the heart, raised to adulthood, and used to establish stable lines for three independent alleles of Tg(-0.349cldn15la:saa;cmlc2:EGFP) (rdu64, rdu67, and rdu68) and Tg(-2.8fabp10a:saa;cmlc2:EGFP) (rdu66). Experiments were conducted with both larval and adult zebrafish positive for rdu64, rdu67, rdu68 and rdu66 and non-transgenic siblings used as controls.
To confirm IEC specificity of the 349 bp cldn15la promoter fragment, we also recombined p5E-0.349cldn15la, pME-mCherry 386 (cytosolic mCherry), and p3E-polyA with pDEST 394, and injected this construct into single cell Tü embryos. Mosaic F0 larvae were raised to adulthood, screened for germline transmission, and used to establish stable lines for three independent alleles of Tg(-0.349cldn15la:mCherry:polyA). All three alleles displayed mCherry expression restricted to the intestine, and allele rdu65 was maintained for further analysis.
Larval zebrafish were anaesthetized with 0.75 mM Tricaine and mounted in 3% (w/v) methylcellulose (in GZM). Caudal fins were amputated using a surgical scalpel (Surgical Specialties Sharpoint, 72–2201) and fish were revived into 10 cm dishes containing either GZM (for experiments with conventionally reared larvae) or AB-GZM without gentamycin (for experiments with gnotobiotic larvae). At 15 minutes, 3 hours, and 6 hours post wounding, animals were euthanized and fixed in 4% PFA / 1x PBS overnight at 4°C on an orbital platform. Larvae were subsequently washed with 1x PBS 3–5 times at room temperature. Imaging was performed with a Leica M205 FA equipped with a Leica DFC 365FX camera using either GFP or mCherry filters. Lyz+ cells were enumerated at the tail wound margin using the Fiji Cell Counter plugin. Fish that were wounded at the notochord or were moribund were excluded from analysis.
Larval zebrafish were anaesthetized and mounted in 100 μL 0.75% (w/v in GZM) low melt agarose (Fisher Scientific, BP165-25) with 0.6 mM Tricaine in 96-well clear-bottom black-walled plates (Greiner Bio-One, 655090), and overlaid with 100 μl GZM containing 0.375 mM Tricaine. For homeostatic behavioral analysis, time lapse imaging was performed for 10 or 15 minutes and frames acquired at 30s intervals on a Zeiss Axio Observer with a Photometrics Evolve EMCCD camera and a 5x objective (NA 0.16, WD 18.5 mm). For live imaging following caudal fin amputation, larvae were mounted as described above and imaged for 6 hours at 2 or 5 minute intervals on an inverted Zeiss Axio Observer Z1 microscope equipped with an Xcite 120Q light source (Lumen Dynamics), an MRm camera (Zeiss), and a 20X objective lens (NA 0.4, WD 7.9 mm). Fish that were wounded at the notochord, moribund, or damaged during mounting were excluded from analysis. Automated cell tracking was performed using the MTrack2 plugin for Fiji. For cell tracking following caudal fin amputation, a region of interest (ROI) was drawn from the posterior end of the notochord to the wound margin. For neutrophil behavior analysis in homeostasis, two distinct ROIs were analyzed. An ROI (396 pixels by 52 pixels, w x h) was drawn over the intestine ending at the cloaca, and another ROI (364 pixels by 152 pixels, w x h) was positioned dorsally to the intestine in the trunk. We subsequently filtered tracking results to include cells that were tracked for ≥ 3 consecutive frames. Speed and meandering index were calculated as previously described [12].
Larval zebrafish from pooled clutches were anaesthetized in 0.75 mM Tricaine and mounted in 3% methylcellulose. Between 15 and 30 fish / genotype were imaged on a Leica M205 FA microscope equipped with a Leica DFC 365FX camera using identical magnification and exposures. Neutrophil recruitment to the intestine was quantified from 8-bit images with Fiji software using the Cell Counter plug in as described previously [12].
For quantifying mCherry fluorescence mean grey values were measured in FIJI. Equal size ROIs were drawn in the liver, trunk, or eye. To quantify intestinal fluorescence, an ROI was drawn around the entire intestine. All mean grey values were normalized by subtracting background signal.
For in vivo light sheet microscopy, 6 dpf zebrafish larvae were anesthetized in 0.75 mM Tricaine in GZM, mixed with 1% LMP Agarose in GZM supplemented with 0.75 mM Tricaine, and drawn up into a glass capillary tube. The agarose was allowed to polymerize for at least 10 minutes prior to imaging. Agarose-embedded larvae were extruded from the capillary into an imaging chamber filled with GZM supplemented with 0.75 mM Tricaine and heated to 28.5°C. Single Plane Illumination Microscopy (SPIM) was performed with a Zeiss Lightsheet Z.1 Microscope equipped with a 20x aqueous immersion objective (1.0 NA, 2.4 mm WD). Two channel acquisition was performed with frame switching using 488 nm and 561 nm excitation lines, and z-series were acquired with a 7.32 μm interval. Image-processing was performed using 64-bit FIJI.
For FACS, replicate pools of 60–90 lyz+ larvae or 30 Tg(-0.349cldn15la:mCherry)+ larvae of the indicated genotypes were euthanized with 3 mM Tricaine and washed for 5 minutes with deyolking buffer (55 mM NaCl, 1.8 mM KCl and 1.25 mM NaHCO3). Larvae were transferred to gentleMACS “C” tubes (Miltenyi Biotec, 130-096-334) containing 5 mL Buffer 1 [HBSS supplemented with 5% heat-inactivated fetal bovine serum (HI-FBS, Sigma, F2442) and 10 mM HEPES (Gibco, 15630–080)]. Larvae were dissociated using a combination of enzymatic and mechanical disruption. Following addition of Liberase (Roche, 05 401 119 001, 5 μg/mL final), DNaseI (Sigma, D4513, 2 μg/mL final), Hyaluronidase (Sigma, H3506, 6 U/mL final) and Collagenase XI (Sigma, C7657, 12.5 U/mL final), samples were dissociated using pre-set program C_01 on a gentleMACS dissociator (Miltenyi Biotec, 130-093-235), then incubated at 30°C on an orbital platform at 75 RPM for 10 minutes. The disruption-incubation process was repeated 4–6 times, after which 400 μL of ice-cold 120 mM EDTA (in 1x PBS) was added to each sample. Following addition of 10 mL Buffer 2 [HBSS supplemented with 5% HI-FBS, 10 mM HEPES and 2 mM EDTA], samples were filtered through 30 μm cell strainers (Miltenyi Biotec, 130-098-458) into 50 mL conical tubes. Filters were washed with 10 mL Buffer 2, and samples were centrifuged at 1800 rcf for 15 minutes at room temperature. The supernatant was decanted, and cell pellets were resuspended in 500 μl Buffer 2. For CellROX (Invitrogen C10491 or C10444) labeling experiments, cells were then resuspended in 500 μL Buffer 2 and transferred to individual wells of a 24-well plate. CellROX was added to a final concentration of 1 μM, and samples were incubated for 45 minutes in the dark at 28.5°C on a tilting platform. Samples were transferred to FACS tubes (Falcon, 352052), and DNaseI (5 μg/mL; Sigma, D4513) and 7-AAD (Sigma, A9400, 5 μg/mL) were added. FACS was performed with either Beckman Coulter MoFlo XDP, Beckman Coulter Astrios, Becton Dickinson Diva, or Sony SH800S cell sorters at the Duke Cancer Institute Flow Cytometry Shared Resource. Single-color control samples were used for compensation and gating. Viable neutrophils or IECs were identified as 7-AAD- lyz+ or 7-AAD- mCherry+ respectively. Data were analyzed with FloJo v10 (Treestar, CA).
Kidneys were dissected from adult male and female transgenic lyz+ zebrafish of the indicated genotypes. Fish were of standard length 25.72mm±1.18mm (mean±S.D.). Single cell suspensions were generated by enzymatic treatment of dissected kidneys with DNaseI (2 μg/mL) and Liberase (5 μg/mL final) with gentle agitation on a fixed speed orbital platform (VWR, 82007–202) for 20 minutes at room temperature. Enzymes were deactivated by the addition of EDTA as described above and cell suspensions were filtered through 30 μm filters and either stained with 1 μM CellROX for 1 hour at 28°C and/or stained with 7-AAD (5 μg/mL) or Propidium Iodide (PI, Sigma, P21493, 5 μg/mL). Viable (7-AAD- or PI-) lyz+ cells were collected into poly-D-lysine (Sigma, P7280-5MG, 33 μg/mL) coated 96 well black-wall clear-bottom plates (Corning, 3603) containing 100 μl RPMI1640 (Gibco, 11835030) supplemented with 10% HI-FBS at a density of 15,000 cells per well. A total of 6–9 kidneys per genotype were pooled and used to seed 4–6 wells. DsRed-expressing Escherichia coli MG1655 (pRZT3, [87]) or GFP-expressing Pseudomonas aeruginosa PA01 (pMF230, [62]) were cultured aerobically overnight shaking at 37°C in LB supplemented with either tetracycline (10 μg/mL) or carbenicillin (100 μg/mL), respectively. Bacteria (100 μL) were subsequently sub-cultured into 5 mL selective LB media and grown at 37°C with shaking under aerobic conditions to an OD600 of 0.7–1, diluted in sterile 1x PBS (Gibco, 14190) to a concentration of 104 bacterial / μL (E. coli) or 103 bacteria / μL (P. aeruginosa), and added isolated neutrophils at an MOI of 2 or 0.2 respectively. Neutrophils were co-cultured with bacteria for 2 to 4 hours at 28°C with gentle agitation in the dark. Serial dilutions of co-culture supernatants were prepared in sterile 1x PBS and plated on selective media. Plates were incubated aerobically at 28°C for 24 hours and CFUs were enumerated.
For imaging studies, neutrophils were collected as described above with the following modification: cells were collected into poly-D-lysine coated thin-bottom 96 well plates (Greiner, 655090). Neutrophils were imaged on a Zeiss 710 inverted confocal microscope with 10x (NA 0.45) or 63x oil objectives (NA 1.40) with or without addition of bacteria.
To measure neutrophil viability, 96 well black wall clear bottom plates containing 15,000 cells / well were incubated with or without E. coli MG1655 (MOI 2, grown as described above) for 3.5 hours. Propidium Iodide (PI) was added to a final concentration of 6 μg/mL to each well, and plates were incubated for an additional 30 minutes at 28.5°C before reading fluorescence at 535/620 (Ex/Em, nm) with a BioTek Synergy2 plate reader. As a positive control, lyz+ cells were incubated with Triton X-100 (20% v/v) for 3 hours, then incubated at 65°C for 30 minutes prior to PI staining. Data are shown as % of maximum PI signal.
For morphological assessment, 15,000–30,000 viable lyz+ cells isolated by FACS from adult dissected kidneys as described above were sorted into 500 μl Buffer 2. Cytospins were performed immediately following collection with a Cytospin 3 (Shandon) by centrifuging cell suspensions for 3 minutes at 800 rcf. Slides were dried overnight at room temperature, fixed with absolute methanol, then stained with Wright Giemsa (Sigma, WG16) according to the manufacturer’s instructions. Slides were imaged with a Leica DMRA2 compound microscope with a PL APO 40x air objective (NA 0.85, WD 0.11mm) and Q Imaging Micropublisher Digital color camera. Twenty individual ROIs were imaged per genotype, and cells were classified based on distinct nuclear morphology by a blinded investigator as described [42].
Pools of 20–30 whole larvae, dissected digestive tracts, or carcasses were collected into 1 mL of TRIzol (ThermoFisher, 15596026) and stored at -80°C. For experiments measuring transcripts from digestive tract versus carcass, dissections were performed as described previously [73] on 6 dpf larvae to remove the digestive tract (including intestine, liver, and pancreas). Digestive tracts were dissected and pooled into TRIzol (15–20 per replicate). The remaining carcasses were collected and pooled into TRIzol. Tissues were homogenized by passing samples 10–15 times through a 27-gauge needle. RNA was isolated following the manufacturer’s protocol with the following modification: a second wash with 70% ethanol (prepared in DEPC-treated H2O) was performed. For gene expression analysis of sorted neutrophils, viable lyz+ (4000–6000 cells / replicate) cells were collected into 750 μL TRIzol LS (ThermoFisher 10296010). For validation of the -0.349cldn15la promoter, 13,000 Tg(-0.349cldn15la:mCherry)+ and mCherry negative cells per replicate were collected into 750 μL of TRIzol LS. RNA was isolated using a NORGEN RNA Cleanup and Concentrator Micro-Elute Kit (Norgen Biotek, 61000), and samples eluted in 10 μL from which 8 μL of RNA was treated with DNaseI (New England Biolabs, M0303L). cDNA was synthesized using the iScript kit (Bio-Rad, 1708891). Quantitative PCR was performed in duplicate 25 μl reactions using 2X SYBR Green SuperMix (PerfeCTa, Hi Rox, Quanta Biosciences, 95055) run on an ABI Step One Plus qPCR instrument using gene specific primers (S1 Table). Data were analyzed with the ΔΔCt method.
Pseudomonas aeruginosa PA01 carrying a constitutively expressed GFP plasmid (pMF230) [62] was grown in LB media supplemented with 100 μg/mL carbenicillin overnight shaking at 37°C. Overnight culture was concentrated to an OD600 of 5 (approximately 2x109 bacteria / mL) then frozen in aliquots. At 1 dpf, larvae were treated with 45 μg/mL 1-phenyl-2-thiourea (PTU) to inhibit melanization. Larvae were infected at 2 dpf by a genotype-blinded investigator. To achieve systemic infection, bacteria were injected into circulation via the caudal vein with borosilicate needles along with a phenol red tracer (3% w/v) as described previously [100], and any larvae that were damaged by injections were excluded from analysis. Approximately 150–300 CFU of P. aeruginosa PA01 GFP was injected per larvae. To enumerate CFUs in the inoculum, the injection dose was plated before and after infections on LB agar plates supplemented with 100 μg/mL carbenicllin. Immediately following infection larvae were screened for even dosing by fluorescence microscopy, and significantly under- or over-infected larvae were excluded from further analysis by an investigator blinded to genotype. To quantify bacterial burden, individual larvae were homogenized in 500 μL sterile 1x PBS using a Tissue-Tearor (BioSpec Products, 985370) at 24 hour intervals, and serial dilutions were plated on selective media (LB agar supplemented with 100 μg/mL carbenicillin). P. aeruginosa CFUs were enumerated following aerobic incubation for 24 hours at 28°C.
For immunostaining, 6 dpf larval zebrafish were fixed in 4% PFA/1x PBS overnight at 4°C on an orbital platform, then washed 3-5x with 1x PBS the following day. Larvae were mounted in 4% low melt agarose in cryo-molds molds (Tissue-Tek) and 200 μm transverse sections cut using a Leica VT1000S vibratome. Sections were transferred to 24-well plates, washed 3x in 1x PBS at room temperature, and permeabilized with 1x with PBS containing 0.1% (v/v) Triton x-100 (PBS-T) for 30 minutes at RT. Sections were blocked in 5% (v/v) donkey serum in PBS-T for 1 hour at room temperature, then incubated with primary antibodies diluted in blocking buffer overnight at 4°C with agitation (mouse anti 4E8, Abcam, ab73643, 1:200; mouse anti 2F11, Abcam, ab71286, 1:200; rabbit anti DsRed, Clontech, 632496, 1:200; chicken anti GFP, AVES, GFP-10x0, 1:200). Sections were washed with PBS, then incubated with species-specific secondary antibodies [(ThermoFisher, A10042, A32728, A11039, 1:200) and Hoechst 33258 (ThermoFisher, H3569, 1:1000)] diluted in PBS-T for 2–4 hours at room temperature on a tilting platform. Sections were then washed 3x with 1x PBS then mounted and coverslipped on slides using mounting media containing DAPI (Vector Laboratories, Inc, H-1200). Slides were imaged with a Zeiss LSM 780 upright confocal microscope equipped with a GaAsP array detector using a 63x oil objective (NA 1.4, WD 0.19 mm).
Adult heterozygous saardu60/+ zebrafish were bred naturally and the resulting embryos were collected into system water and pooled at 0 dpf. Fertilized embryos were sorted at equal densities into autoclaved 3L tanks filled with system water at 1 dpf (50 embryos per 3 liter tank). Zebrafish of both WT and saa mutant genotypes were co-housed for the duration of the experiment. Larvae were maintained under static conditions until 6 dpf, at which time water flow and feeding were begun. Dissected digestive tracts and environmental water samples were collected at two time points: 6 dpf and 70 dpf. At 6 dpf zebrafish were sampled prior to first feeding, and fish sampled at 70 dpf were fed using the standard facility diet (as described above in animal husbandry) beginning at 6 dpf. For each timepoint, fish were sampled from a minimum of 5 tanks. For the 6 dpf timepoint DNA was isolated from 14 WT and 15 saa mutant intestines. For the 70 dpf timepoint, DNA was isolated from 15 WT and 12 saa mutant intestines. For water samples, 50 mL of water was collected from each tank and filtered using 0.2 μm MicroFunnel Filter Units (Pall Corporation, 4803). Filters were removed from filter units with sterile forceps, transferred to Eppendorf tubes and snap frozen in a dry ice/ethanol bath. For intestinal samples, digestive tracts were dissected and flash frozen and either carcasses or fin were reserved for genotyping. Fish samples were genotyped to identify homozygous saa+/+ and saardu60/rdu60 zebrafish prior to submission for genomic DNA extraction.
The Duke Microbiome Shared Resource (MSR) extracted bacterial DNA from gut and water samples using a MagAttract PowerSoil DNA EP Kit (Qiagen, 27100-4-EP) that allows for the isolation of samples in a 96 well plate format using a Retsch MM400 plate shaker. DNA was extracted from ≥12 fish per genotype per time point, and from 5 to 8 different tanks per timepoint to control for tank effects. Sample DNA concentration was assessed using a Qubit dsDNA HS assay kit (ThermoFisher, Q32854) and a PerkinElmer Victor plate reader. Bacterial community composition in isolated DNA samples was characterized by amplification of the V4 variable region of the 16S rRNA gene by polymerase chain reaction using the forward primer 515 and reverse primer 806 following the Earth Microbiome Project protocol (http://www.earthmicrobiome.org/). These primers (515F and 806R) carry unique barcodes that allow for multiplexed sequencing. Equimolar 16S rRNA PCR products from all samples were quantified and pooled prior to sequencing. Sequencing was performed by the Duke Sequencing and Genomic Technologies shared resource on an Illumina MiSeq instrument configured for 150 base-pair paired-end sequencing runs.
Subsequent data analysis was conducted in QIIME2 (https://qiime2.org), the successor of QIIME [101]. Paired reads were demultiplexed with qiime demux emp-paired, and denoised with qiime dada2 denoise-paired [102]. Taxonomy was assigned with qiime feature-classifier classify-sklearn [103], using a naive Bayes classifier, trained against the 99% clustered 16S reference sequence set of SILVA, v. 1.19 [104]. A basic statistical diversity analysis was performed, using qiime diversity core-metrics-phylogenetic, including alpha- and beta-diversity, as well as relative taxa abundances in sample groups. The determined relative taxa abundances were further analyzed with LEfSe (Linear discriminant analysis effect size) [105], to identify differential biomarkers in sample groups.
All experiments were repeated at least two times and statistical analyses were performed with GraphPad Prism v.7. Data are presented as mean ± SEM. For comparisons between 2 groups a two tailed student’s t-test or Mann-Whitney test was applied. For comparisons between 3 or more groups, a one-way ANOVA with Tukey’s multiple comparisons test was used. For experiments with 2 independent variables, a two-way ANOVA was performed. Significance was set as p < 0.05, and denoted as: * p < 0.05, ** p < 0.01, *** p < 0.001, **** p < 0.0001. Sample sizes are indicated in the figure legends.
|
10.1371/journal.pcbi.1005754 | Spatiotemporal network coding of physiological mossy fiber inputs by the cerebellar granular layer | The granular layer, which mainly consists of granule and Golgi cells, is the first stage of the cerebellar cortex and processes spatiotemporal information transmitted by mossy fiber inputs with a wide variety of firing patterns. To study its dynamics at multiple time scales in response to inputs approximating real spatiotemporal patterns, we constructed a large-scale 3D network model of the granular layer. Patterned mossy fiber activity induces rhythmic Golgi cell activity that is synchronized by shared parallel fiber input and by gap junctions. This leads to long distance synchrony of Golgi cells along the transverse axis, powerfully regulating granule cell firing by imposing inhibition during a specific time window. The essential network mechanisms, including tunable Golgi cell oscillations, on-beam inhibition and NMDA receptors causing first winner keeps winning of granule cells, illustrate how fundamental properties of the granule layer operate in tandem to produce (1) well timed and spatially bound output, (2) a wide dynamic range of granule cell firing and (3) transient and coherent gating oscillations. These results substantially enrich our understanding of granule cell layer processing, which seems to promote spatial group selection of granule cell activity as a function of timing of mossy fiber input.
| The cerebellum is an organ of peculiar geometrical properties, and has been attributed the function of applying spatiotemporal transforms to sensorimotor data since Eccles. In this work we have analyzed the spatiotemporal response properties of the first part of the cerebellar circuit, the granule layer. On the basis of a biophysically plausible and large-scale model of the cerebellum, constrained by a wealth of anatomical data, we study the network dynamics and firing properties of individual cell populations in response to 'realistic' input patterns. We make specific predictions about the spatiotemporal features of granule layer processing regarding the effects of the gap junction coupled network of Golgi cells on a spatially restricted input, in an effect we denominate first-takes-all. Furthermore, we calculate that the granule cell layer has a wide dynamic range, indicating that this is a system that can transmit large variations of input intensities.
| The granular layer of the cerebellar cortex consists of populations of granule cells (GrCs), Golgi cells (GoCs), unipolar brush cells, and Lugaro cells [1–3]. GrCs are excitatory [1] and form the largest population of neurons not only in the cerebellum, but also in the entire brain [4]. GoCs are inhibitory and are known as interneurons of the granular layer [1,5]. The granular layer of the cerebellar cortex receives its input from different parts of the brain primarily through mossy fibers [6]. The mossy fibers excite both GrCs and GoCs through their typical axonal boutons called ‘rosettes’ [7,8]. Within the cerebellar cortex, the GrCs excite GoCs through parallel fibers [9–11] and ascending axons [12] and the GoCs in turn inhibit numerous GrCs through sagittal branching of their axons [5,13]. So there exists a feedback loop between the GoCs and GrCs [14], which has a similar structure to that of the pyramidal-interneuron gamma rhythm generation (PING) model of the neocortex [15]. In addition, the GoCs are connected together by gap junctions [16–18] and have also been reported to inhibit each other sparsely [19].
Previous studies have proposed different roles for cerebellar GrCs. Jörntell and colleagues have suggested that GrCs function as signal-to-noise enhancing elements [20,21], based on their observation that GrCs in the C3 zone of decerebrated cats receive identical inputs through mossy fibers that are modality specific, have the same receptive field type and are similarly encoded. Other studies proposed that GrCs provide a bank of various temporal patterns (tapped delay line model, spectrum models) that can be used to generate learned temporal responses such as in classical conditioning experiments [22–25]. According to this view, the GrC population is endowed with a variety of time constants so that the different GrCs are active at different moments during conditioned stimuli [26].
One of the earliest proposals for the GrC function was by David Marr, suggesting that they act as low noise sparse encoders [27], a popular hypothesis supported by some recent electrophysiological and modeling studies [28–30]. In this theory, each GrC represents a combination of a few mossy fibers that provide diverse input, where the number of activated GrCs at any time is small compared to the total number of GrCs, i.e. sparse firing, to facilitate discrimination of binary input patterns by Purkinje cells. In support of this, GrCs receive only a few mossy fiber inputs [31] and exhibit low background firing rates partly due to the presence of tonic GABAergic input [29]. With such a synaptic structure, sparsely firing GrCs could losslessly encode a wide range of spatial input patterns [30].
Sparse encoding assumes relatively uncorrelated GrC activity, however, there are multiple anatomical and physiological mechanisms that promote correlation of GrCs. For instance, it has been demonstrated that bursting of a single mossy fiber afferent can lead to bursting of many GrCs [32]. Furthermore, granule cells in the flocculus respond with high mutual correlation during vestibule-ocular reflex tasks, due to the activation by unipolar brush cells [33]. In addition, spillover mechanisms could conceivably create spatially correlated GrC input [34]. The sagittal arrangement of mossy fiber rosettes are likely to create anisotropic spatial correlations [7,35]. Recent in vivo imaging of the granular layer reports a lack of sparse activity in GrCs [36]. Hence, a detailed analysis on how granular layer network mechanisms contribute to spatiotemporal encoding of parallel fiber activity is timely, particularly considering the anatomical interactions between the cell populations characterized by the high density of GrCs [4] (Fig 1A–1F).
Early physiologically detailed models of the granular layer, constructed in 1D and 2D, have suggested the presence of robust oscillations in the granular layer of the cerebellar cortex due to the feedback loop between GoC and GrC’s [14,37]. The oscillations cease if there is very low mossy fiber activity, or a dominant excitation of GoCs by mossy fibers or high tonic inhibition of GrCs in the network [14,37,38]. The 2D model suggested that gap junctions between GoCs increase the power of feedback loop driven oscillations [37]. Recently it has been suggested that the emergence of network oscillations can also be linked to NMDA receptors at parallel fiber-GoC synapses [39]. Comparable oscillations have also been experimentally observed in the local field potentials (LFP) recorded in the granular layer, in the 10–25 Hz range in the paramedian lobule of primates [40] and in the range of 7–8 Hz in Crus IIa of awake rats [41].
While previous detailed computational models were studied with a limited repertoire of mossy fiber stimuli such as spatially uniform and monotonic ones, etc. [14,37,39], the mossy fiber firings in vivo exhibit a variety of temporal and spatial patterns. Vestibular mossy fibers provide slow rate-coded inputs that linearly encode head velocity [33,42,43]. In response to sensory stimulation, mossy fibers in Crus I and Crus IIa generate high frequency bursts [29,32], and metronome mossy fibers of the lateral reticular nucleus (LRN) spike synchronously [44,45]. Furthermore, in response to peripheral stimulation, each body part is represented multiple times in the form of patches in the granular layer, where adjacent patches represent non-adjacent body parts, forming a so called fractured somatotopy [46,47]. Here we simulate a 3D large-scale network model of the granular layer activated by patches of mossy fibers inputs with realistic firing patterns such as slow rate modulation or rapid bursting, to study how spatiotemporal interactions between the neurons determines holistic network dynamics.
First, we simulated the granular layer network (Fig 1) model with spontaneous background firings of all mossy fibers. The mossy fiber firing rate was 5 Hz, which is comparable to experimental observations: Cuneate mossy fibers fire spontaneously around 9 Hz in vivo [21]. Mossy fibers of the LRN of the brainstem fire regularly (spontaneous) in a wide range from 2–23 Hz [21,45]. Mossy fiber boutons in vivo from crus I and crus II of cerebellar cortex are spontaneously active around 4 Hz [32]. With the background inputs, the GrCs and GoCs in the model fired with a mean frequency of 1.01±0.09 Hz and 8.18±0.61 Hz respectively, and these match values from in vivo recordings, ~1 Hz (GrCs in Crus I-IIa anaesthetized of rats) [29] and ~8 Hz (GoCs in Crus I-II of anaesthetized rats) [48].
A characteristic feature of the network activated with diffuse mossy fiber input is widely distributed oscillations of GoCs and GrCs (Fig 2), driven by the feedback loop from GoCs onto GrCs, and vice-versa. The loop consists of AMPA and NMDA receptors of the GrCs activated by the mossy fiber input, AMPAergic receptors in the GoC population activated by the parallel fiber/ascending axon input, and GABAergic receptors in the GrCs activated by the GoCs. Oscillations in baseline could readily be seen in single cell activities of GoCs (Fig 2A and 2E), but were less obvious in single GrC firing for background input, as they are sparsely active (Fig 2C). As observed in a previous 2D network model of the granular layer of cerebellar cortex [37], gap junctions between GoCs increased the synchrony of GoC and of GrC firing in case of low frequency diffuse mossy fiber input but had less effect when in addition a patch of mossy fibers was activated more strongly (compare Fig 2B and 2D).
The mossy fiber firing rate was vital in controlling the firing rates of GrCs and GoCs and of the network oscillation frequency. In addition to the baseline input of 5 Hz, we activated the mossy fibers in patches of 100 or 200 μm in radius, which we will call the ON patch, over a range of input frequencies. This protocol simulated the patch-like mossy fiber activations observed in vivo [49]. In those simulations, GoCs showed highly synchronized oscillations (Fig 2F) while GrCs exhibited more loose synchronization (Fig 2H). Oscillation frequency increased with the frequency of the activated mossy fibers (Fig 3A) together with the firing frequency of GoCs (Fig 3B) and GrCs (Fig 3C), regardless of the patch size. The presence of gap junctions slightly increased the firing frequency of GoCs (Fig 3B) and GrCs (Fig 3C) for high mossy fiber firing rates. Unlike the previous one-dimensional model of granular layer [14] where oscillations disappeared for mossy fiber firing rate below 15 Hz, oscillations were still observed in our model for 5 Hz mossy fiber background firing rate.
Each model GoC needed to receive inhibition on its apical dendrites to obtain experimentally observed firing rates (see Methods). The origin of this inhibition is not conclusively known, but we postulated that it largely originates from nucleocortical neurons [50,51], with a total conductance for each GoC of 2160 pS based on experimental results. This made GoCs fire in the experimentally observed range in vivo. Each GoC also received inhibition on basal dendrites from other GoC cells [19]. This basal inhibition was about ~10 pS/Hz*(average firing rate of presynaptic GoCs), which is much weaker than the inhibition on apical dendrites. However, the effect of GoC-GoC inhibition remained limited even when it was artificially increased. Increasing the synaptic conductance of GoC-GoC inhibition 8 times decreased the GoC firing rate only moderately from ~64 Hz to ~54 Hz with mossy fibers constantly firing at 80 Hz. The same condition changed oscillation frequency barely (from 52.1 Hz to 54 Hz), but the oscillation power decreased to 45%. Therefore, increased GoC-GoC inhibition weakened the oscillations, with small effect on their frequency.
At the single neuron level, the response to mossy fiber input was quite stochastic (Figs 3E and S1. See also S1 Movie). This paper, an initial description of our results, will mostly emphasize the average behavior of the network, which is quite complex. But this only summarizes the rich and stochastic network dynamics, shown in the Supplementary Movies.
The synchronization index of the GoCs and GrCs increased with the firing frequency of mossy fibers in the activated patch and with the size of the patch (Fig 3D), as a larger number of GoCs became involved in each oscillation. Elimination of gap junctions reduced the synchronization index but did not eliminate synchronization at higher mossy fiber firing rates.
The divergence rate from a single mossy fiber to GCs is quite large (see S1 Table). Moreover, mossy fiber rosettes occupy a restricted volume of the granular layer allowing for activation of circumscribed patches (Fig 1C). The sparse coding hypothesis predicts that GrC firing should remain sparse, also for high input conditions. However, in our model the GrC activity within the activated volume climbed quickly with increasing mossy fiber input frequency, from just a few cells to about half the cells in a patch (in a long integration window of 100 ms, Fig 4A). We defined the dynamic range as the ratio of maximal to minimal activation of GrCs (see Methods). The largest dynamic range was found for a physiologically relevant integration window of 1 ms, where increasing the mossy fiber firing rate from 10 to 80 Hz caused 10.1 times more GrCs to fire in the ON patch. The sparse baseline firing rate of the GrCs likely represents an almost quiescent network, and this sparse firing quickly transforms into dense firing upon stimulation. Adding tonic inhibition (see the last section of Results) slightly increased the dynamic range.
Figs 5–7 and S2 Fig demonstrate how the network responded when the input was time dependent, particularly when firing rates were slowly modulated (Fig 5A). For this, we stimulated the mossy fibers in single or double ON patches of 100 μm radius in various configurations that were 1) single patch (Fig 6C, inset) 2) double patches along the transverse axis with 800 μm of the center-to-center distance (Fig 6G and 6K, inset), 3) double patches along the sagittal axis with 400 μm distance (S2H and S2L Fig, inset). Volumetric maps of the network activity in response to a double patch input along the transverse axis are shown in S3 Fig and S2 Movie.
In both the single and double-transverse patch paradigm, GoCs along the parallel fiber axis showed a high degree of oscillatory synchrony with little effect of firing rate co-modulation (Fig 5B). This was observed not only in the cross-correlation between two ON patches (Fig 5D) but also in the cross-correlation of an ON patch with a non-stimulated patch, which we will call an OFF patch (Fig 5E). This demonstrated the effectiveness of the common parallel fiber input to the GoCs along the transverse axis. On the other hand, the effect of firing rate modulation was much more pronounced in the population activity of the GrCs (Fig 5C). In particular, the GrCs in the OFF patches (along the transverse direction) showed anti-correlation of their firing rate with the ON patch GrCs, while the synchronized oscillations could still be observed on a shorter time scale (Fig 5G). Therefore, the spatial structure of the average firing rate and the correlations was strikingly different between the GoCs and GrCs. There was only a small spatial dependence of the GoC firing rates along the transverse axis (Fig 6A, 6E and 6I). They exhibited a stable and high cross-correlation along the transverse axis and showed only a slight decay with distance even after discounting the effect of firing rate co-modulation (Fig 6B, 6F and 6J), indicating that the correlation was due to a high degree of synchronization (Fig 5E) driven by the parallel fiber inputs. On the other hand, the GrC firing rates displayed an on-beam inhibition [52] featuring an activated ON patch surrounded by laterally inhibited cells along the transverse axis (Fig 6C, 6G and 6K). In a single ON patch paradigm (Fig 6C), the ON patch GrCs fired at 47.7±1.9 Hz while those in the OFF patch (separated by 500 μm) fired at a below-baseline average rate of 0.86±0.11 Hz, due to increased GoC inhibition in OFF patches. The GrCs along the transverse axis were more strongly correlated within and between two ON patches than between the ON-OFF pairs (Fig 6D, 6H and 6L), with a stronger effect of firing rate modulation (Fig 5F). We compared the two-patch condition for identical (Fig 6E–6H) mossy fiber input frequency with simulations where the patches received different input frequencies (Fig 6I–6L), but overall there was little difference suggesting that the spatial input pattern is more important than the input frequencies.
Spatial profiles of cross-correlations of long term firing rates (red curves in Fig 5D–5G) also clearly exhibited these structures for both GoCs (Fig 7A) and GrCs (Fig 7B) along the transverse axis. In the single ON patch stimulation paradigm, the on-beam inhibition is clearly seen in the GrC rate correlations along the transverse axis (Fig 7B). This is a little more pronounced without gap junctions between GoCs since the GrC firing rates becomes less variable due to a decrease of spike synchronization, but gap junctions did not alter the spatial structure qualitatively. Therefore, the synaptic connectivity in the granular layer results in a winner-take-all mechanism where OFF patch GrCs get inhibited by GrCs that receive strong mossy fiber input (ON patch) through parallel fiber mediated feedback inhibition.
In the absence of common parallel fiber input, when activity was measured in patches along the sagittal axis in response to slow rate modulation, GoCs were not activated and GrCs were not inhibited in the OFF patches (S2A and S2C Fig). When two patches were activated in this configuration, GrC correlation reduced as the distance between the patches increased (S2H and S2L Fig) and the GoCs exhibited less synchrony (S2F and S2J Fig) than for the transverse configuration (Fig 6F and 6J). Patches of GoCs in this configuration were correlated if they were close to each other (100 μm) but the correlation between the patches rapidly decreased with distance.
As reported previously [37], gap junctions had a limited effect on the synchrony between GoCs along the parallel fiber axis, evoked by patch activation. Without gap junctions between GoCs, the synchrony was reduced among GoCs (for the single patch condition, Fig 6B) along the parallel fiber axis when compared with the control condition but was maintained all along the parallel fiber axis and didn’t exhibit any decay with distance. This was observed even when we activated two mossy fiber patches along the transverse axis with different firing rates (Fig 6J). The difference in correlation between control and gap junction block conditions decreased for two-patch activation due to increased overall parallel fiber activity (Fig 6F and 6J). Removal of gap junctions between GoCs reduced the synchrony of both GrCs (S2D, S2H and S2L Fig) and GoCs (S2B, S2F and S2J Fig) along the sagittal axis.
To summarize, the application of slow rate coded input to the granular layer model of the cerebellar cortex synchronized the GoCs along the transverse axis, resulting in an on-beam inhibition that makes OFF patch GrCs exhibit anti-correlation on a longer time scale and increases the separation between ON patch and OFF patch GrCs with respect to firing rate.
GrC ascending axons, which excite the basolateral dendrites of GoCs [12], play an important role in shaping the activity of the network, especially the GoC synchronization.
In Fig 6 (see also Fig 8A), the cross-correlation of GoCs between two remote ON patches (separated by 800 μm) along the transverse axis (0.64±0.02) is less than that of the corresponding patches (0.69±0.02) for a single patch activation configuration. This is contrary to the intuition that cross-correlation should be larger in the two-patch activation configuration due to an increased activation of shared parallel fiber inputs.
We first checked whether this feature is due to differences in the temporal structure of mossy fiber inputs between the activated patches. We eliminated the difference in temporal input structure between the two activated ON patches (see Methods) and calculated the cross-correlation. This procedure did not affect the cross-correlation in the two-patch activation condition (0.64±0.01; Fig 8B).
Next, we eliminated the ascending axon inputs to GoCs (reduced their peak synaptic conductance to zero) and calculated the cross-correlation in the same manner as above. Fig 8C shows that when ascending axon inputs are blocked, the cross-correlation between the ON patches (0.73±0.03) becomes higher than that of the corresponding patches for single patch activation configuration (0.70±0.02). Moreover, the removal of ascending axon inputs to GoCs, resulted in an overall increase in the cross-correlation for both activation paradigms (Fig 8C).
Because ascending axon inputs represent highly 'localized' input sources to GoCs in both ON patches, they result in reduction of the parallel fiber mediated synchronization. In a single patch activation configuration, cross-correlation varied non-monotonically with distance along the transverse axis and this is due to ascending axon inputs to GoCs (Fig 8A). Here the correlation decreases steeply until 400 μm, which is the last OFF-ON pair, and therefore the ascending axons in the ON patch resulted in a reduced correlation (0.55±0.02). Beyond this point, the correlation is a measure between OFF-OFF pairs and recovers up to 0.69±0.02 at a distance of 800 μm. For the two ON patch configuration, cross-correlation reaches a plateau for distances beyond 400 μm due to the localized ascending axon inputs in the second ON patch (Fig 8A). Increasing the strength of the ascending axon to GoC connections, by increasing their synaptic weights by 20% or 40% of their original values led to a small decrease in cross-correlation of 2–5% along the transverse axis (not shown). We conclude that ascending axon inputs to GoCs reduce the parallel fiber mediated synchronization of GoCs.
Besides the slow rate modulation that we have used so far, some mossy fibers can also respond to sensory stimulation in vivo on a much shorter time scale by bursting at an extremely high frequency of a few hundred hertz [32], sustained for a few tens of milliseconds. This signal is reliably transmitted to the GrCs that show similar bursting [29].
We simulated this by activating the mossy fibers in the selected ON patch(es) with burst type inputs. Each mossy fiber in the patch was given nine input bursts, with a duration of 10 ms and a firing rate of 500 Hz (Fig 9A). During the 10 ms burst, all the GoCs that fired emitted only one spike (S1I and S1J Fig) and the GrCs spiked 3–4 times (S1K and S1L Fig). As a result, the GoC PSTH showed a sharp peak (Fig 9B, inset), while the GrC PSTH exhibited a prominent broad peak (Fig 9C inset). GrCs were strongly correlated between the activated patches (Fig 9F). GoC firings were highly synchronized for both ON-ON (Fig 9D) and ON-OFF (Fig 9E) patch configurations. All along the transverse axis, GoC firing was sharply synchronized for both single and double patch conditions (Fig 10B and 10F). The ON patch GrCs along the transverse axis were strongly correlated (Fig 10H). The correlation decreased between the ON and OFF patch GrCs and as a result the correlation vs. distance relationship shows the shape of an inverted bell. Volumetric maps of the network activity in response to a single burst activation of a single patch are shown in S4 Fig and S3 Movie.
Along the sagittal axis, the correlation decreased with distance for both GoCs (S5B and S5F Fig) and GrCs, (S5D and S5H Fig) except when the correlation was measured between the two ON patches (S5F and S5H Fig), which showed strong stimulus driven correlations even with a distance of 400 μm between the two patches.
We observed that the on-beam inhibition of GrCs also emerged with the bursting input (Figs 10C and volumetric representation of cell activities in S4C and S4D) since OFF patch GrCs were silenced due to inhibition by strongly firing GoCs that were activated by the parallel fibers. This on-beam inhibition predicts that GrC spikes can be gated by synchronous GoC firing along the transverse axis. We examined how strongly synchronized inhibition can regulate GrC firing if multiple sets of the mossy fibers along the transverse axis are bursting, particularly when there are relative time delays between the bursts. This replicates in vivo conditions, where the patches of mossy fibers could get activated along the transverse axis at various latencies in response to peripheral activation.
We activated two mossy fiber patches separated by 500 μm along the parallel fiber axis (Fig 11, inset) with the same double patch burst activation paradigm but with different latencies between them. We discovered that the feedback inhibition due to the first patch parallel fibers inhibited the GrC excitation in the second patch when the arrival of the feedback inhibition coincided with their mossy fiber excitation of GrCs. Therefore, when the latency of mossy fibers in the second patch was around 5 ms, the GrC excitation was less effective and GrC firing rate decreased by 20% (Fig 11) compared to synchronous activation of the patches. Because of the slow spillover component of GoC inhibition this effect persisted for intervals up to 15 ms and then slowly declined with a return to baseline firing responses at 30 ms intervals. We conclude that synchronized GoC inhibition can strongly regulate GrC firing, even in the presence of excitatory drive, and this mechanism makes the earliest firing GrCs dominate the network activity. Therefore, it may be more appropriate to describe the competition among inputs in the granular as first-take-all instead of winner-take-all.
The activity of ON patch GrCs in response to the burst input was not only characterized by firing during the input, but also showed a long transient even after the offset of the mossy fiber burst (black line in Fig 12A). This was unexpected since there are no other sources to excite GrCs other than mossy fibers in our model. We found that the long transient was due to a long-term gain increase in the firing rate by activation of NMDA receptors.
GrCs express NMDA receptors [53] that contain the GluN2C subunit [54] and NMDA mediated currents are known for their non-linear voltage dependence and slow kinetics [55]. The NMDA receptors exhibit voltage dependent block at hyperpolarized membrane potentials due to partial block by magnesium ions, and the membrane needs to be depolarized enough to remove this block. Supralinear synaptic summation [56] can be a mechanism to deliver such depolarization to the NMDA receptors. Therefore, the spiking of GrCs caused by strong mossy fiber input could result in NMDA receptor unblocking and their slow decay kinetics caused effective elevation of the resting membrane potential for a considerable period of time (Fig 12A, inset). In this way, the NMDA receptors can implement a winner-keep-winning mechanism, enabling the GrCs that have already spiked to fire again more easily by improved integration of the subsequent inputs.
To further investigate how the effectiveness of this gain change, we delivered a weak probe asynchronous mossy fiber inputs (rate: 20 Hz, duration: 20 ms) with different time delays after the burst offset (Fig 12D). Then, we measured GrC firing during the probe inputs and compared it the control conditions where there is no probe input or when the probe inputs were delivered to an OFF patch. If there is no ON patch specific gain change, the firing rate change in the ON patch GrCs with a probe input should be equal to the rate increase in the OFF patch GrCs due to a probe input. Instead, we found a supra-linear increase in GrC firing, even up to 150 ms after a burst offset (Fig 12E). Therefore, the mossy fiber burst input induced a long lasting gain increase in ON patch GrCs.
The same simulation was repeated with a reduced conductance of NMDA receptors at the mossy fiber-GrC synapse (Fig 12A and 12C). In this case, while there was little change in the response of the ON patch GrC population during the burst, the rebound activity was significantly reduced (Fig 12A and 12C). In response to the probe mossy fiber input, ON patch GrCs exhibited reduced firing rate (Fig 12F) with reduced conductance of NMDA receptors (decrease in firing rate from 44.4 Hz to 10.8 Hz in response to an asynchronous input at 2 ms after the burst offset). The ON patch response amplitudes to the probe stimulus were now closer to those of the OFF patch GrCs. The winner-keep-winning mechanism was robust to changes in the strength of GoC to GrC synaptic inhibition (not shown).
Our results show that the NMDA receptors in GrCs cause a long-term change in their input/output function, which implements a winner-keep-winning mechanism in bursting GrCs. GrCs reliably burst with a single bursting mossy fiber and this has been proposed as a mechanism for reliable signal transmission [32]. Here our results demonstrate that mossy fiber bursts can also be a mechanism for regulating the signal processing property of GrCs and changing how the GrC population filters subsequent mossy fiber inputs.
GrCs possess extra-synaptic GABAA receptors (receptors with δ subunit) [57,58]. In many neurons of the central nervous system, extra-synaptic GABA receptors mediate a form of 'tonic' GABAergic current and play an important role in their baseline excitability [57].
We simulated the effect of tonic inhibition on network oscillations by quantifying the power and frequency of ON patch network oscillations when activated with constant mossy fiber input of different firing frequencies. We modeled the tonic inhibition as a tonic Cl- conductance of 88 pS in the GrCs (see Methods). We found that the tonic inhibition reduced the power of network oscillations (Fig 13A, 13B and 13D). For a mossy fiber frequency of 30 Hz, tonic inhibition reduced the peak power of GoC network oscillations from 336±34 to 154±22 (Fig 13A). We observed a fairly constant reduction in power of network oscillations for all values of input mossy fiber firing rate (Fig 13D). In contrast, the effect of tonic inhibition on oscillation frequency was small (Fig 13C). For a mossy fiber firing frequency of 30 Hz, oscillation frequency for control condition was 40.4±0.5 Hz while that in the presence of tonic inhibition was 38.2±0.8 Hz (Fig 13C). Therefore, although tonic inhibition of cerebellar GrCs reduces the power of network oscillations, robust oscillation still arise with sufficiently strong mossy fiber input without any effect on the oscillation frequency.
The neural network of the cerebellar granular layer is known as an anatomically well-studied circuitry that only contains a few neuron types, but how it transforms mossy fiber inputs is still being actively debated. Challenges in investigating this question with computational modeling come from the strikingly large imbalance between the number of excitatory and inhibitory neurons and the peculiar geometric properties governing synaptic connectivity. To address these problems, we constructed a large scale and physiologically realistic model of the granular layer neural network that emulates tissue properties of the granular layer of the cerebellum. Rather than introducing assumptions about cerebellar function, our bottom-up approach investigates the dynamical patterns emerging from the known neurocircuitry and physiology.
Another strain of cerebellar models, commonly denominated top-down approaches, aims at capturing functional dynamics that enable motor control [22,23,25,59]. Though some more recent models also include some biophysical and anatomical details, their function relies primarily on the ability of the climbing fiber to modify strengths of parallel fiber synapses to train Purkinje neurons for perceptron-like pattern classification. It is worth noting that other models of sensorimotor control attribute the same abilities to the neocortex [60]. In many models, details are introduced to address limitations of function, such as correcting for saturating activation and synaptic plasticity. For instance, GoCs [27,28,30] and molecular layer interneurons (MLIs) [61] have often been assumed to act like level setting systems or GrCs as tapped delay lines [22–25], a role that may or may not be compatible with their known physiology as described in the Introduction. Such models add biological details to improve the predefined performance of the model.
Our analysis instead focused on spatiotemporal dynamics in response to physiologically plausible mossy inputs and investigated which output patterns emerge in our physiologically detailed model. Through analysis of input/output relationships we showed that synaptic and cellular mechanisms in the cerebellar granular layer enable the network to stochastically transform and integrate information over multiple spatial and temporal scales of mossy fiber input.
Our model can be considered a superset that reprises findings of previous 1D [14], 2D [36,38] and 3D [38] network models such as feedback oscillations, while suggesting new dynamical phenomena implied by physiology and anatomy. Previous models were significantly smaller and did not analyze network responses as a function of complex mossy fiber activation. Earlier models in [14] and [37] lacked the fine spatial and temporal structure of mossy fiber activation in the cerebellar cortex and did not include the ascending axon input to GoCs. Similarly, the model in [39] approximated mossy fiber input by current injection. A 3D network model in [38] described network dynamics of the granular layer in response to spontaneous and burst input patterns and replicated the center-surround inhibition observed in experiments in sagittal slices, where the parallel fibers are cut [62]. Because of the absence of any significant parallel fiber contribution, this model could not produce the spatial interactions we described here.
Our network model is based on recent conductance based models of individual neuron types [38] and, network topology including both the long folium axis and parasagittal axis in the cerebellar cortex. A potential limitation of our network model is that it extends only for 700 μm along the sagittal axis and therefore does not capture rostro-caudal distribution of a number of structures (e.g., rostro-caudal extent of mossy fiber axonal arborization is usually greater than 1000 μm in the granular layer [7]). The model does not include NMDA receptors at parallel fiber-GoC synapse because they caused depolarization block in the GoC model and have been reported to be absent in adult animals [11,12]. A previous 2D model suggested that NMDA receptors at parallel fiber-GoC synapse can cause activity dependent state transitions in the granular layer [39]. Moreover, the GoC model used in our study does not incorporate the fine branched morphology of GoC dendrites found in the granular layer [11]. Gap junctions between GoCs probably occur more frequently along the sagittal plane of the folia as GoC dendrites follow the zebrin boundaries of Purkinje cells above them [63]. But lack of proper experimental data forced us to model them without any directional dependence. Finally, to achieve physiological low GoC firing rates it was necessary to include dendritic inhibition from sources outside of the granular layer. Inhibition of GoCs has been an unresolved issue since a recent study claimed that MLIs do not inhibit GoCs [19,51] and our results emphasize the importance of extracortical inhibition for normal GoC function. Conversely, the effect of GoC to GoC inhibition is modest due to its weak strength [19].
One of the strongest features of the simulated network dynamics is the network wide oscillation driven by a feedback loop between the GrCs and GoCs, and mediated by the parallel fibers, which synchronize the GoCs. The synchronized activity of GoCs, were stronger all along the transverse axis as in experimental studies [10]. Consistent with this result [10], we found that common parallel fiber input drives the synchronized spiking in GoCs, but the gap junctions also contributed significantly [18,64] particularly when mossy fiber firing frequency was low. The GrCs in the network, whose spike timings are controlled by cycles of GoC inhibition, exhibit a less precise synchronization.
Afferent mossy fibers that project to the granular layer exhibit a wide variety of firing patterns. Experimental studies have shown that mossy fibers exhibit slow rate modulation during a variety of behaviors [42,65], but can also exhibit burst activity in response to sensory stimulation [20,21,66] firing at more than 700 Hz [32]. We studied how the network responds to and encodes these different physiological input patterns. With one exception, the observed synchronization patterns differed little. This may seem surprising because the mechanisms are fundamentally different, driven by feedback inhibition for the slow rate modulated input, and caused by locking to the strong stimulus for burst input. In response to both types of input, GoCs exhibited parallel fiber mediated synchronization extensively along the transverse axis and this GoC activity powerfully regulated GrC firing. As a consequence, the GrCs correlations showed much more dependence on their location relative to the stimulus than on stimulus type or frequency. Separate GrC populations along the transverse (parallel fiber) axis, fired with significant correlations only when they both received mossy fiber inputs, regardless of whether the inputs were slow rate modulated or bursting. However, along the sagittal axis, the stimulus driven correlations in the GrCs became strong only with simultaneous bursting mossy fiber inputs. On the other hand, input by the ascending axons of GrCs [12] was found decrease synchronization of GoCs along the parallel fiber beam because they are highly local.
Additionally, the coherence of granular layer network oscillations was affected by tonic inhibition, which is present only in the GrCs [57,58], without any effect on the oscillation frequency. Extra-synaptic GABAA receptors that mediate tonic inhibition are known to be involved in many neuro-psychiatric disorders and also in memory and cognition [57], such as hippocampus-dependent learning and memory [67]. In the cerebellum, tonic inhibition improves the representation of sensory information in granule cells [68], whereas it is unclear how it affects motor learning [69].
In experiments, network oscillations in the granular layer have been probed by the LFP (reviewed in [40]). Our simulation can be augmented by recently developed softwares to compute the LFP directly [70] or via hybrid schemes [71–73], to predict how the LFP signal depends on physiological factors, which can be verified in extracellular recording experiments [74].
GrC population activity is characterized by two distinctive patterns at two time scales. On long time scales, there is an on-beam inhibition effect due to global inhibition of the unstimulated GrCs along the parallel fiber axis, which implements a first-take-all type mechanism. On shorter time scales, the GrC activities are regulated by the time window imposed by the timing of synchronized GoC spikes, which can regulate precision in timing, particularly regarding different latencies in the onsets of mossy fiber inputs. This is in contrast to a recent modeling study [38] that showed a much smaller inhibitory surround (< 100 μm in diameter) around an excited center. However, since that model was limited in space and had few parallel fiber contacts per GoC (~100), this was probably due to limitation of the model size. A recent in vitro study [75] also suggested that GoCs provide fast feedback inhibition to GrCs, based on the observation that a GoC receives inputs mostly from nearby GrCs but also some input from distant GrCs. Parallel fiber synapses may deliver much smaller input to a GoC soma compared to ascending axon synapses [12]. However, it has been observed that weak common inputs to individual cells can lead to robust synchronization, not only in the cerebellar network [14] but in many contexts [76]. Furthermore, our model predicts that the earliest GoC inhibition should dominate and this coincides with a recent experimental observation that the majority of GrCs receive early, not late, GoC inhibition [77].
We also observed that NMDA receptors in the GrCs play an important role by inducing a long-term increase in the GrC output gain after (burst) spiking, even in the presence of the lower voltage-dependence due to their GluN2C subunits [54]. Therefore, the GrCs that have already fired upon early mossy fiber inputs can integrate subsequent inputs much better than other cells in the network, which we called the winner-keep-winning mechanism. NMDA receptors have been well known for their role in supralinear synaptic integration in many systems including GrCs [56], which can contribute to information gating (e.g., [78]). The winner-keep-winning mechanism is a combined effect of two phenomena due to NMDA receptors, sustained depolarization [56] and voltage dependent synaptic integration [79], that gives an additional advantage (long-term gain upregulation) to GrCs that respond to bursting inputs. Furthermore, NMDA receptors in GrCs are known for their roles in synaptic plasticity. It has been proposed that this plasticity can tune the relative latency between the GrC firing and mossy fiber input, which in turn dictates whether the GrC firings can pass the time window imposed by the GoC feedforward inhibition[62,80].
All the mechanisms that we have discussed, the network mediated first-take-all and the cell intrinsic property based winner-keep-winning, give a predominant advantage to the GrCs that are activated earlier by the mossy fiber inputs while the others are suppressed. While this leads to a sparse spatial organization of the granular layer output, the activity within activated regions of the GrC can be quite dense due to the high dynamic range of the GrC population. This pattern of activation is compatible with the described fractured somatotopy of tactile inputs in crus II of the cerebellum [46,47]. The response to the two-patch configuration can be considered a simulation of patches activated by the same tactile input. Moreover, the larger amplitude of responses observed to the late input from sensory cortex, compared to the preceding trigeminal input [46], could be explained by the NMDA mediated increase of the GrC gain if the respective mossy fibers synapse onto the same GrCs. Note, however, that our two patch simulation results also apply to co-activated mossy fiber inputs carrying different modalities [81].
In Marr’s pioneering theoretical work [27] and following studies [30], the GoCs also play the role of regulating how many GrCs activate, but the spatiotemporal aspect of GoC firing has been largely ignored. In the mushroom body in the insect olfactory system, the synchronous and oscillatory firing of inhibitory interneurons maintain sparse firing of excitatory neurons [82]. In our model, the GoCs are governed by a similar principle since they oscillate, discharge synchronously over an extended spatial scale, and impose a narrow time window leading to effectively inhibiting a large number of GrCs, contributing to strongly spatially restricted activation. However, contrary to insect olfaction and to Marr’s theory [25], our model predicts that GrC activity within the activated patch depends on the strength of the mossy fiber stimulus and is often not sparse.
The nature of coding by the granular layer has been actively debated: Jörntell and colleagues in their study in C3 zone of decerebrated cats have found little evidence of sparse coding [20,21,83]. In C3 zone of cats, the authors reported that GrCs receive similar kind (unimodal) of mossy fiber inputs [20], whereas diverse mossy fiber inputs should converge at a GrC (multimodal) for sparse coding to work effectively. Also, GrCs in their study were not silent at rest and fired a barrage of spikes in response to peripheral activation [20]. However, other studies in mouse cerebellar cortex demonstrated that GrCs receive multimodal mossy fiber input. For example, Huang et al reported convergence of proprioceptive (external cuneate nucleus) and pontine (basilar pontine nucleus) inputs in various regions of the cerebellar cortex [84]. Convergence of multimodal mossy fiber inputs (vestibular, visual) is found in the GrCs of the vestibulocerebellum [85] and in the hemispheres (tactile, auditory and visual) [81]. Our model is neutral towards the convergence discussion because we did not specify what information is carried by the mossy fiber input.
The sparse coding by the GrCs hypothesis has recently also been challenged based on experimental observations of dense coding by GrCs [36] and that the GrCs also rate code the rate modulated MF inputs [83,86] (see also [87]). Similarly, our model showed that, despite strong temporal patterning by the GoCs, the GrC population rate follows the rate modulated mossy fiber input quite well, resulting in a large dynamic range. Furthermore, spatially separated GrC populations can co-activate, when each of them are stimulated by a different mossy fiber group. Note that this would be impossible if GoC inhibition is purely based on an asynchronous rate code, since no time window for co-activation would be allowed. Therefore, the rich spatiotemporal dynamics of our model provides a unified viewpoint for the resolution of experimental controversies about coding in the cerebellar granular layer.
Our simulations suggest that oscillations characterize the basic network activity of cerebellar granular layer network along with stochastic spiking of GoCs and GrCs and rich spatio-temporal dynamics. A first-take-all mechanism based on the network structure and NMDA receptor mediated winner-keep-winning mechanisms further characterize the spatiotemporal dynamics of granular cell firing. Wide dynamical range indicates a large flexibility in the allocation of granule cells, ranging the encoding from sparse to dense. Based on our results, we suggest that the unique anatomy of the cerebellar granular layer, coupled with cellular and network mechanisms promote spatial group selection of GrC activity as a function of MF input timing and spatial organization.
All simulations were carried out using the NEURON simulation platform (version 7.4) [88] on the OIST high-performance computing cluster, running on 200 cores. The mean time taken to run a “biological millisecond” was 3.00±0.03 seconds. The model is made publicly available at ModelDB (http://senselab.med.yale.edu/modeldb) under the accession number 232023.
We used previously published models of GrCs and GoCs [38] except that the dendritic morphology of a GoC was modified: two shorter (60 μm long) baso-lateral dendrites were constrained to the granular layer and the other two, longer (~166 μm long), apical dendrites extended into the molecular layer as in [1]. We also reduced the diameter of the dendrites to 2.4 μm to match the electrical and firing properties to the original model. All the cell and synapse models (see below) were simulated at the temperature of 37°C. For simulations with tonic inhibition, we included a tonic conductance of 88 pS with a reversal potential at -73 mV in a GrC model, which resulted in ~260 pS of total tonic conductance [32], which includes stationary activation of GABAergic synapses in the baseline condition.
Our granular layer network model is based on detailed anatomical information previously published [1,4,5,7,16,18,19,31,35,89,90]. The 3D network model has dimensions of 1500 μm along the transverse axis, 700 μm along the sagittal axis and 430 μm along the vertical axis (Fig 1). The granular and molecular layers were 200 μm thick each, with a 30 μm thick Purkinje cell layer between them. The number of neurons in the network was determined in the following way: We first calculated the number of GoCs in the network using the anatomical GoC density (9500 cells/mm3) [18]. From this we calculated the number of GrCs in the network using the GrC to GoC ratio [4]. The total number of GoCs in the network for the above-mentioned network dimension was 1,995 and total number of GrCs amount to 798,000. The somatic centers of all the neurons were uniformly distributed in the granular layer.
After this, we determined the connectivity between the neurons based on connectivity rules that we will explain in the following section. The neurons were then connected with experimentally validated synapses and gap junctions with corresponding conduction delays depending on their mutual distances. The conduction velocity of parallel fiber axons was set to 0.3 m/s [91,92], while that of mossy fiber and GoC axons was 2 m/s [93].
Mossy fiber rosette distribution was based on that of LRN mossy fiber axons [7]: Rosettes of a single primary collateral of LRN axon distribute widely along the parasagittal axis, but along the transverse axis the spread is limited. As a result, rosettes of a single LRN axon are arranged in sagittal strips parallel to each other along the transverse axis. A similar parasagittal arrangement of mossy fiber rosettes is also reported in other studies [35].
We first constructed mossy fibers with a density of 5000 fibers/mm2, which is based on the projection density of mossy fibers in C1 zone of Paramedian lobule of the cerebellum [94,95]. Due to network size limitations, the distribution of mossy fiber rosettes in the model is based on that of a single primary collateral of a LRN axon [7]. For each mossy fiber in the model, the rosettes were distributed according to the rosette cluster distribution of primary collaterals of LRN axon [7]. We used another experimental data set about the distribution of pontine mossy fiber rosettes ([96], private communication with Daria Rylkova) to optimize their distribution in the model. For each mossy fiber in the model, we adjusted the extent of spread of rosettes both along the long axis and sagittal axis until the amount of overlap between mossy fibers closely matched that of the pontine mossy fiber data. We measured the amount of overlap between different mossy fibers in the model and experimental (pontine) data as follows: We divided the entire volume into a number of small cubes and counted the number of distinct mossy fibers represented in each cube. From this data, we calculated the relative number of cubes representing 0,1,2,3 and 4 distinct mossy fibers. This was then repeated for different cube sizes. We computed the final mossy fiber density using anatomical ratio of glomerulus to GrC [31]. In order to eliminate boundary effects, we also instantiated mossy fibers around the network when the rosettes projected into the model. Total number of mossy fibers that project at least one rosette into the model was 2109 and total number of rosettes was 29519.
The connectivity between neurons in the network is based on anatomical connectivity patterns observed in the cerebellar granular layer [5,31,89,97,98]. The model has synapses projecting from excitatory mossy fibers to GrCs, mossy fibers to GoCs, inhibition by GoCs of GrCs, excitation by GrCs of GoCs through ascending axons and parallel fibers (Fig 2). In addition to the synapses listed above, GoCs are connected by gap junctions [16,18] and inhibitory synapses [19]. Convergence, divergence and synaptic parameters for each synapse in the model are described in S1 Table.
Except for synapses in GrCs (see below), the time course of synaptic conductance Gsyn(t) was modeled according to the standard double exponential equation [99]
Gsyn(t)=gmax×N×[exp(−tτdecay)−exp(−tτrise)]
(1)
where τrise and τdecay are rise and decay time constant respectively. gmax is peak synaptic conductance, and N is a normalization factor that makes the maximum of Gsyn(t) equal to gmax. τrise and τdecay were obtained by fitting Eq 1 to the respective experimental traces.
The connectivity between mossy fibers and GrCs is based on the maximum length of the GrC dendrite [31]. For each GrC, we formed a sphere of radius of 30 μm around its center and connected to rosettes within that sphere in a probabilistic manner. On average, each GrC received 4.5±1.5 (2–7) distinct mossy fiber connections.
We used the mossy fiber-to-GrC synapse model of [38] (deterministic version) with the following modifications: First, neurotransmitter diffusion is approximated by a cascade linear process [100],
dPdt=−rfastP,
dTdt=−rTT+P−rT1(T−I1),
dI1dt=r1T(T−I1)−r12(I1−I2),
dI2dt=r21(I1−I2)−r23(I2−I3),
dI3dt=r32(I2−I3),
(2)
where T is the concentration of diffused neurotransmitter. At each presynaptic spike, P is transformed as P→P+y where y represents a diffusing fraction of released neurotransmitter, controlled by a synaptic facilitation/depression mechanism. The parameters are given as rfast = 4/τD, rT = 6.2/τD, rD1 = 20/τD, r1D = 9.09/τD, r12 = 4.9/τD, r21 = 1.71/τD, r23 = 0.55/τD, and r32 = 0.333/τD. The diffusion time constant τD is given by τD=100Rd2/4D where Rd = 1.03 μm and D = 4 μm2/ms [38,79]. This scheme provided a good approximation of the AMPA and NMDA activation over a wide range of presynaptic inputs (S6A Fig). Second, we set the desensitization constant of the NMDA receptor to 12×10−4 ms-1 [79]. Finally, the voltage dependence of the NMDA receptors is modeled as
f(V)=11+[Mg]oKMgexp(−V/δV)
(3)
where [Mg]o = 1 mM, KMg = 1.77 mM, and δV = 22.4 mV [54].
Conductance parameters of the receptors were adjusted to have GrC firing ~1 Hz in the baseline condition (mossy fiber firing at 5 Hz) and also its ~6 fold increase in the absence of inhibition [68].
For connectivity between mossy fibers and GoCs, we assumed a sphere of radius 100 μm and connected GoCs and rosettes within that sphere probabilistically. Each GoC in the model received an average of 13.7±6.5 (1–36) distinct mossy fiber connections. The mossy fiber to GoC synapse is glutamatergic (only AMPA receptors) whose synaptic parameters were obtained from the experimental recordings of GoC EPSCs [101].
Inhibitory connections between GoCs and GrCs were based on the extent of axonal arborization of the former. GoC axons exhibit a parasagittal organization (Fig 2) [5]. Distribution of their axonal boutons is about 650 μm along the parasagittal axis and about 180 μm along the medio-lateral axis. We assumed a connection probability that generated 8.4±3.2 (1–22) GoC synapses per GrC on average.
Synaptic parameters were based on experimental data [13] and included an indirect spillover component (S6B Fig). The IPSC decay consisted of two components: the transient component with a time constant of 5 ms and an indirect spillover component with a time constant of 35 ms (that contributed to 10% peak amplitude). The IPSC rise time constant was 3 ms.
Connections from GrCs to GoC via parallel fibers/ascending axons were generated using our custom tool, the Boundary Representation Language (BREP) [102]. In this method, the geometric structures associated with the connectivity (parallel fibers/ascending axon and GoC apical dendrites) were described as points in space along a straight line in three dimensions. The ascending axon of each GrC was modeled as a straight vertical line of length 200 μm with points separated by 50 μm. The parallel fibers of each cell were modeled as two straight lines of length 1000 μm each (extending on either side of their bifurcation from the ascending axon in the molecular layer) with points separated by 7.5 μm. Small random perturbations were added to both.
GoCs also had random angular displacements of their dendritic points. In each GoC, the dendritic elements were modeled as lines that lie on the surface of an inverted cone of height 332 μm (apical dendrite) or 6 μm (baso-lateral dendrite). Each dendritic element was created with a randomly chosen angle from a normal distribution with mean (30°, 120°) for apical and (-20°, -240°) for basolateral and standard deviation of 10°. The elements were first rotated on to the circumference of a circle of radius 100 μm and raised (apical in molecular layer) or lowered (basolateral in granular layer) thereby forming an inverted cone. The GoC axons were represented as uniformly distributed random points in a rectangular area (boundaries in μm: transverse [-45:45], sagittal [-160:160], vertical [-75:75]) relative to the soma position.
Once the points associated with each geometric cell structure were generated, we used a K-d tree data structure [103] to order the points and performed fast nearest neighbor searches. We assumed a connectivity radius of 30 μm and 5 μm for ascending axon and parallel fiber connections, respectively. The connectivity probability was chosen to achieve the target number of connections. On the average, each GoC in the model received about 554±302 (55–1245) ascending axon connections. The number of parallel fiber synapses (4759±1037 (2512–6582)) on a single GoC in the model was calculated based on the density of parallel fiber synapses in the molecular layer [98] and also based on the fact that approximately 9% of them are formed on structures other than Purkinje neuron spines [97]. Both ascending axon and parallel fiber synapses on GoC dendrite are AMPAergic with time constants and maximal synaptic conductance described in S1 Table.
We connected the GoCs with gap junctions [16,18] and inhibitory connections [19]. Inhibitory and gap junctional connectivity between GoCs were also generated by BREP. The probability distribution function (Boltzmann function) for gap junction connectivity was based on an experimental published data [16] and the conductance decayed as a function of distance [18] as g = β exp(-λx) where β = 1.659 nS and λ = 0.01259 μm-1. Each GoC had about 13.7±4.6 (1–31) gap junctions on average. For inhibitory connections between the GoCs, we used the experimental measurements (20% connectivity probability at 50 μm) from ref. [19], coupled with the gap junction connection probability data. Each GoC received inhibitory input from 2.2±1.6 (0–10) GoCs on average.
As we tuned the model with the background mossy fiber firing of 5 Hz by varying synaptic conductances, we discovered that the firing rate of GrCs and GoCs tended to covary with a ratio of GrC:GoC ≈ 1:30. This suggested that GoCs needed extra inhibitory inputs to reproduce in vivo observations of GrC:GoC ≈ 1:8–18. Recent studies also suggested that the inhibitory inputs mostly originate from extracortical neurons [50,51], which are not in our model.
To simulate the effect of those inhibitory inputs, we included tonically active GABA receptors in the apical dendrites of the GoCs. We estimated that they should roughly correspond to ~150 synapses with a peak conductance ~180 pS [51] and also a 5 ms decay constant in in vivo-like conditions. A range of activation rates from 15 to 20 Hz robustly resulted in ~1 Hz and ~10 Hz firing of GrCs and GoCs, respectively, with a 5 Hz mossy fiber input, and we chose 16 Hz, which led to the resulting total conductance of 2160 pS.
We introduced random variations in the cellular and synaptic parameters as follows: GoC and GrC soma diameters were randomly varied by up to 20%, and their initial resting membrane potential was also varied in the range -60 to -75 mV. For each type of synapse, the peak conductances were varied in a manner so that they had a coefficient of variation of 0.25.
We generated firing of each mossy fiber by using a leaky integrate-and-fire (LIF) neuron model driven by a noisy current input: the membrane voltage of the model was given by
dVdt=−V−Eτ+βμ(t)+σ(t)ξ(t),
where τ = 1 ms, E = -70 mV and ξ(t) was a Gaussian white noise updated every 1 ms. The spike threshold was at V = -60 mV. After a spike, a refractory period of 1.1 ms was imposed and then V was reset to E. μ(t) and σ(t) were controlled by a common parameter ν(t) as μ(t) = gNν(t) and σ(t)=gNν(t) where N = 1000 and g = 5 μV/ms. We chose β = 0.01 to ensure that the model would fire mostly due to noisy fluctuations in the input. We first generated a table of constant ν vs. the output firing rate, and used it to calculate ν(t) for a certain target firing rate by linear interpolation.
The background firing of each unstimulated mossy fiber was generated by the LIF neurons firing at 5 Hz as in the in vivo recordings [32]. The stimulated mossy fibers were activated in patches of 100 or 200 μm radius, with different input patterns such as slow rate modulation or bursting. In the slow rate modulation paradigm, the input was defined by upper and lower bound frequencies, each lasting 300 ms. The lower bound frequency was kept constant at 10 Hz for all simulations. Upper bound frequency was varied between 50–60 Hz. A combination of an upper bound and lower bound is an epoch, which lasts for 600 ms and each of the rate modulated mossy fiber stimuli consists of five epochs (Fig 5A). The firing rate epochs were smoothed with a Gaussian kernel (σ = 50 ms) and the LIF spike trains were generated based on the rate. In the burst input paradigm, we activated mossy fibers in patches of 100 μm with bursts of frequency 500 Hz and duration 10 ms.
Mossy fibers were activated in single or two patches either along the parallel fiber axis or sagittal axis. For simulations involving ascending axon mediated de-synchronization of GoCs along the transverse axis, we eliminated the difference in temporal structure in mossy fiber input between the two activated patches (patch 1 and 2) in the following way: For each GoC in patch 1, a corresponding GoC was randomly picked (without replacement) from patch 2. Mossy fiber connectivity to the GoC from patch 2 was made identical to that of the GoC from patch 1.
We recorded spike times of all neurons during the course of simulation. Simulations were repeated 5 times with different global random seeds, which also affected the network structure. Data was then analyzed using MATLAB version R2011b (Mathworks, MA, USA) software. Spike times were transformed into spike trains with 1 ms long time bins. We often evaluated the average activity of specific neurons within a certain region by taking the average of the corresponding spike trains.
Oscillations were measured by binning the spikes of the population (GoC or GrC) in 1 ms long time bin. Power spectral density of the resulting oscillations was calculated and oscillation frequency was taken as the frequency corresponding to peak power in the power spectral density. The synchronization index for GoC oscillations was calculated as the proportion of total number of GoCs involved in each oscillatory cycle, calculated by integrating the area under each oscillatory cycle. The firing rate of ON patch neurons in Figs 9, 10 and 11 was computed for a time period of 30 ms from the burst onset. For Figs 6 and S2 (slow rate coded input), the firing rate of ON patch neurons was computed for a period of 100 ms (during the upstroke of the epoch when mossy fiber firing rate is maximum). The firing rate of OFF patch neurons was computed during the corresponding time period. All indicated values in the study represent mean ± standard deviation.
Dynamic range of GC activation (Fig 4) was quantified for both ON patch and OFF patch GrCs (both transverse and sagittal axis) when simulated with different mossy fiber firing rates. This was done by calculating the percentage of active GrCs in different time windows (1,10,100 ms) over the course of the entire simulation and averaging it across different data sets. Dynamic range was then calculated using the formula, Dynamicrange=maxvalueminvalue.
The volumetric reconstructions of cell activity are achieved by binning the cells spiking within any millisecond of simulation in a voxel of 10 μm. This produces a 3D histogram of spike counts per voxel. This volume is then convolved with 3D Gaussian kernels normalized by the maximum value of the kernel. The resulting voxel value is proportional to the maximum number of cells active in the voxel. The result is passed as color and alpha components to the MATLAB function Vol3D (http://www.mathworks.com/matlabcentral/fileexchange/22940-vol3d-v2).
Cross-correlations were computed between the average activities of neurons in two selected patches, regions of 100 or 200 μm radius in the model. We first formed the spike trains of all the neurons with 1 ms time bins. The average activity yA for patch A is given by,
yA=(∑i=1NAAi)NA
(4)
where NA is the number of the GrCs or GoCs in A.
The cross-correlation function (CCF) between region A and B is given by
CCFAB(t)=1L×ZAB∑s=1L−t(yA(s+t)–yA¯)×(yB(s)−yB¯)ift≥0,
CCFAB(t)=1L×ZAB∑s=1−tL(yA(s+t)–yA¯)×(yB(s)−yB¯)ift<0,
(5)
where ZAB=(Var[yA]×Var[yB]) and L is the length of yA,B.
CCF was computed for t = ±300 time lags.
We defined cross-correlation ‘c’ as oscillatory synchrony after discounting the effect of firing rate modulation. We denote the measured correlation coefficient by ‘a’ (zero-time lag correlation CCF (t = 0)), the expected coefficient from firing modulation only by ‘b’, and compute c as c = a-b. In all sections of the results we report the value of ‘c’ as cross-correlation. To find the effect of firing rate co-modulation, the average population activities for each patch were low pass filtered below 10 Hz, which was above the frequency range of our input firing rate modulation. We computed CCFs based on them according to Eq 4, except that the normalization factor ZAB is still based on the unfiltered spike trains. This scheme made it easier to compare cross-correlations at two different time scales (e.g., blue and red lines in Fig 5D–5G).
Statistical significance of CCF (t = 0) is non-parametrically evaluated by counting the number of the outliers nout in {CCFshuffled} whose amplitude exceeded that of CCF (t = 0). CCFshuffled was computed for t = ±300 time lags in the following way: In each of the patches, we divided each simulation epoch, which can be the period of the entire simulation or each stimulation protocol depending, into a number of small sub-epochs (of length 10 ms) and randomly shuffled the GrC and GoC spike trains in the divided sub-epochs. This gave us the average activity zA and zB from the shuffled spike trains of two patches. Nshuffle was chosen to be 50. CCFshuffled was calculated from zA and zB according to Eq 4. Then, the p-value of CCF(t) was estimated by an empirical type-I error rate, p=noutNtotal, where Ntotal = Nshuffle × (2 × nlag + 1). We assumed a confidence interval of 99% and p<0.01 was considered to be significant.
We also tried to increase the statistical power by combining the results from multiple simulations. In this case, CCFs were appropriately averaged and the p-values were obtained from combined observations. Error bars in cross-correlation were obtained by bootstrap resampling.
|
10.1371/journal.pgen.1004998 | Synergistic Interactions between Drosophila Orthologues of Genes Spanned by De Novo Human CNVs Support Multiple-Hit Models of Autism | Autism spectrum disorders (ASDs) are highly heritable and characterised by deficits in social interaction and communication, as well as restricted and repetitive behaviours. Although a number of highly penetrant ASD gene variants have been identified, there is growing evidence to support a causal role for combinatorial effects arising from the contributions of multiple loci. By examining synaptic and circadian neurological phenotypes resulting from the dosage variants of unique human:fly orthologues in Drosophila, we observe numerous synergistic interactions between pairs of informatically-identified candidate genes whose orthologues are jointly affected by large de novo copy number variants (CNVs). These CNVs were found in the genomes of individuals with autism, including a patient carrying a 22q11.2 deletion. We first demonstrate that dosage alterations of the unique Drosophila orthologues of candidate genes from de novo CNVs that harbour only a single candidate gene display neurological defects similar to those previously reported in Drosophila models of ASD-associated variants. We then considered pairwise dosage changes within the set of orthologues of candidate genes that were affected by the same single human de novo CNV. For three of four CNVs with complete orthologous relationships, we observed significant synergistic effects following the simultaneous dosage change of gene pairs drawn from a single CNV. The phenotypic variation observed at the Drosophila synapse that results from these interacting genetic variants supports a concordant phenotypic outcome across all interacting gene pairs following the direction of human gene copy number change. We observe both specificity and transitivity between interactors, both within and between CNV candidate gene sets, supporting shared and distinct genetic aetiologies. We then show that different interactions affect divergent synaptic processes, demonstrating distinct molecular aetiologies. Our study illustrates mechanisms through which synergistic effects resulting from large structural variation can contribute to human disease.
| Autism spectrum disorders (ASDs), which are characterised by poor social interaction and repetitive behaviours, are in part caused by genetic variation. A number of genes that vary in copy number in ASD patients have been identified, many of which were known to function at the neuronal synapse. We theorised that in some cases the dosage change of multiple genes simultaneously, rather than singularly, may lead to faulty neuronal development, and contribute to ASD. To test this, we asked whether alterations in these candidate genes would cause neuronal synapse and sleep/rest changes using the fruit fly Drosophila, and validated this model using single-gene models. We considered the simultaneous change of pairs of genes that were jointly affected by a large human copy number variant (CNVs), which are structural changes in the genome. In three of four CNVs, mutations in subsets of genes synergistically interacted to cause neuronal changes comparable to the single gene candidates. We also observed that the changes in synapse size followed the direction of the human gene copy number change. Finally, we show that different interactions affect the development of the synapse through different mechanisms, allowing us to identify distinct molecular alterations that illuminate the etiological heterogeneity of ASD.
| Autism spectrum disorders (ASDs) comprise a large group of complex neurodevelopmental diseases that are influenced by genetic and environmental factors [1]. They are characterised by altered cognitive function including poor social and verbal interaction capability, and repetitive and stereotyped verbal and non-verbal behaviours [1]. ASDs are highly heritable (∼90% monozygotic twin studies); however, the genetic cause has been identified in less than 30% of cases, while the increase in risk between di-zygotic twins is comparable to that of first degree siblings [2], suggesting that ASD-causative alleles are likely to be both numerous and rare [3].
Recently, large numbers of autistic individuals, with unaffected family members, have been shown to possess de novo copy number variants (CNVs) [2,4–6]. In addition, many rare variant studies have identified pathways or processes that are commonly contributed to by significant proportions of those genes found to be disrupted [7–9]. Two additional striking findings from a recent study into the genes affected by 192 de novo CNVs identified in individuals with ASD have also been identified [9]. Firstly, many of these CNVs affect genes that appear to operate in the same functional pathway/network and, secondly, a significant proportion of individual CNVs (33%) simultaneously affect multiple genes whose proteins interact within that functional pathway [9]. This raises the possibility that it is the combined effect of these genes’ copy number change that causally contributes to these patients’ autistic phenotypes. Combinatorial effects have also been observed beyond de novo variants, where an increased risk of ASD resulting from multiple distinct and inherited CNVs has been reported [10]. However, while the contribution from combinatorial effects of genetic variation has been proposed by computational and statistical analyses, these hypotheses have yet to be validated in vivo. Here, we use Drosophila as an in vivo system to examine genetic interactions that may contribute to neurological phenotypes like ASD.
Understanding the interactions between genes implicated in autism requires a tractable, high-throughput in vivo system. This is particularly important as patient genotypes possess variants affecting many genes, thus generating an exponential number of potential interactions. To this end, the fruit fly Drosophila melanogaster offers a versatile tool in which neurodevelopment and behaviour can be studied in isogenised genetic backgrounds, and under controlled environmental conditions [11–13]. To detect single and combinatorial gene dosage effects in the fly, we examine two neurological phenotypes, namely (1) abnormalities in larval neuromuscular junction (NMJ) bouton number and (2) circadian defects apparent through abnormalities in adult sleep rest cycles. The NMJ offers a sensitive in vivo system to identify interactions that alter synaptic growth and maturation [14] and has proved a valuable tool for studying genes associated with neurodevelopmental disorders including autism spectrum disorders, intellectual disability and neuropsychiatric diseases [15–19]. For example, mutations in Neurexin IV, the Drosophila orthologue of the autism gene CTNAP2, have been shown to decrease NMJ bouton number and the abundance of glutamate receptors that oppose the active zones. Circadian rhythm activity defects have been previously reported in Drosophila neurodevelopmental models, including fragile X syndrome and Angelman syndrome, and can be an indicator and causative factor of neurodevelopmental and neurodegenerative disorders in humans [20–23]. Recent studies in Drosophila have also identified sleep abnormalities in mutations of the candidate ASD gene cullin 3 (CUL3) [24–26]. Furthermore, sleep and circadian abnormalities are both significantly associated with ASD: Sleep disturbance is experienced by up to 80% of individuals with ASD, and while more strongly associated with ASD than other neurodevelopmental disorders it is not associated with intellectual disability, which is however frequently comorbid with ASD [22].
In this study, we modelled the effects of gene dosage changes on Drosophila neurological readouts using gene sets derived from multigenic de novo CNVs that had been identified in patients with autism [5,27–29]. We focussed our attention on the unique Drosophila orthologues of genes affected by these CNVs whose protein products had previously been found to participate in an ASD-associated interaction network, and which had a role in neural functioning [9]. To do this, we first considered those CNVs that changed only a single gene in the ASD-associated network, and show that the dosage alterations in the Drosophila orthologue yields neurological defects similar to those previously reported in Drosophila neurodevelopmental disease models [30,31]. We next looked at CNV gene sets that affected multiple genes in the ASD-associated network. Amongst these genes, no heterozygous mutation in a single gene led to significant synaptic defects in the fly. However, pairwise crosses between heterozygously-mutated genes yielded neurological defects comparable to the monogeneic models. We observe that (i) pairwise combinatorial dosage effects amongst these genes are not additive, but clearly synergistic, and (ii) that when the direction of copy change of the orthologues in individuals with ASD is considered, the observed effect at the Drosophila synapse supports a model of convergent phenotypic outcome between distinct synergistically-interacting gene pairs. No effects were observed among gene pairs that included neuronally-expressed Drosophila genes whose orthologues were affected by these CNVs but that were not part of the ASD-associated network. We show that the combinations of genes drawn from these CNVs that interact are specific, supporting distinct molecular aetiologies underlying ASD. We also show that these specific interactions affect different molecular processes at the Drosophila synapse, supporting the role of distinct molecular ASD related aetiologies. In total, we identified synergistically-interacting orthologous pairs among 3/4 of the CNVs considered, demonstrating novel synergistic interactions that may contribute to the aetiology of autism.
Previous studies applying network analyses to rare ASD associated genetic variants have proposed that these variants may genetically interact to exert their phenotypic influence in a patient [9,10]. To investigate the proposition of gene-gene interactions in these ASD cases, we used the fruit fly Drosophila. In particular, we modelled the effects of combinatorial heterozygous dosage changes of pairs of candidate genes, in the fly, and looked for synaptic and circadian defects. A schematic of our method is also set out in Fig. 1. Candidate genes were defined as those genes that had both (1) been identified to be previously affected in individuals with ASD by de novo CNVs, and additionally (2) those contributing to a large network of interacting proteins with roles in neural functioning, herein termed as an “ASD-associated network” [9]. Firstly, two CNVs were identified that affected only a single gene within the ASD-associated network: Specifically, 1 CNV affected CTNND2 while another CNV affected NOTCH1 (Table 1). These were brought forward as ‘monogenic’ candidates. Four additional de novo human CNVs were identified that each overlapped multiple ASD-associated network candidate genes, and where every candidate gene possessed a unique Drosophila orthologue. These CNVs gene sets were also taken forward for in vivo study (Table 1). In addition, from each of these 4 CNVs, two control genes were randomly selected and taken forward. These were genes that again possessed a unique Drosophila orthologue, and which were expressed in both the larval and adult nervous system (Table 1). Of the final 6 CNV gene sets taken forward for in vivo modelling, 2 sets were monogenic while 4 sets were polygenic. 4 were derived from copy number losses while 2 were derived from copy number gains. Table 1 details the CNVs, directionality, human genes and corresponding Drosophila ortholouges for all experiments.
Singular and combinatorial effects resulting from the simultaneous dosage change of ASD-candidate genes were investigated by identifying changes in neuromuscular junction (NMJ) bouton number, and circadian rhythms (specifically alterations in the light/dark bias towards sleep and rest). As a complex disease with behavioural deficits relating to alterations in the human brain, ASD may not be wholly modelled in Drosophila. However, by enabling the rapid screening of multiple target genes the fly is a powerful model to test gene-gene interactions in vivo. It thus offers a tractable method to better understand the gene-gene interactions postulated to occur as a result from these large de novo CNVs. We believed bouton number, and circadian rhythms to be relevant because phenotypes because: (i) The fly NMJ, a tractable and highly characterised glutamatergic synapse, has been successfully used to detect synaptic defects in models of ASD, neuropsychiatric disease and intellectual disability [15,31]; (ii) circadian rhythm defects are associated with ASD and several fly ASD models [21,32,33].
Two of the six CNV gene sets considered contained only one candidate gene. One monogenic gene set is derived from a loss CNVs that affected the orthologue of the Drosophila gene p120 catenin (p120ctn) with roles in cell adhesion and signal transduction, while the other CNV contained the evolutionarily conserved signalling molecule Notch, whose human orthologue was found to be copy number increased (Table 1). Mutants for neurexin IV (using Nrx-IV4304), the orthologue of the autism gene CNTNAP2, and w1118 were used as positive and wild type controls, respectively. All Drosophila stocks were isogenised to the w1118 wild type background for 7 generations for this study. As previously described, we found that NrxIV homozygous null mutants display reduced bouton numbers, while heterozygote nulls have no observable difference when compared to wild type (Fig. 2A) [34].
The first monogenic CNV gene set we analysed was Drosophila Notch, the orthologue of human Notch1, derived from a human Chromosome 9 copy number gain CNV (Table 1). To investigate the increased expression of Drosophila Notch, we overexpressed Drosophila Notch (using UAS-Notch-Full) driven by the ubiquitous expression GAL4 driver 1032-Gal4 (Fig. 2B). While larvae overexpressing Notch had no overt effect on early larval survival, we observed reduced NMJ bouton numbers (n>20, Kruskal-Wallis test, ** P<0.01; Fig. 2B) showing that dosage increase in this gene yields synaptic phenotypes in Drosophila.
Next, we considered the monogenic CNV gene set corresponding to the loss of the Drosophila orthologue p120ctn. The previously described null mutant p120ctn308 was isogenised to analyse hemizygous p120ctn loss [35]. However, p120ctn heterozygous null mutants displayed no significant change in NMJ (Fig. 2C) although homozygous p120ctn null mutants were found to display a significantly reduced bouton number (n>20, Kruskal-Wallis test, ** P<0.01, Fig. 2C). We note that, unlike in vertebrates, Drosophila p120ctn homozygous null mutants are viable [35].
We next looked for circadian rhythm defects in the monogenic CNV gene set orthologues Notch and p120ctn mutants. Notch overexpression larvae were reared at 16°C, and were transferred to 25°C during pupation, so to mitigate gross developmental defects. We analysed sleep/rest periods (measured as a contiguous 5 minute periods of inactivity) as a surrogate for looking at gross defects in rhythmicity. While all negative control and single mutants displayed normal light/dark differences in sleeping patterns (i.e more sleep/rest periods during the dark 12hrs; Fig. 2D, E), both p120ctn homozygous nulls (Fig. 2E) and the Notch (Fig. 2D) overexpressing flies all lost the dark bias and displayed no significant difference between light/dark sleeping patterns.
Taking these monogenic models together, we show that dosage change in Drosophila of the orthologues of known ASD diseases genes (NrxIV), and of ASD-candidate genes subject to de novo copy number increase (notch) and decrease (P120ctn) in human, all yield abnormalities at the NMJ, and in circadian rhythms (notch and p120ctn) (Fig. 2). We also find that despite differences in the direction of dosage change in Drosophila that are consistent with the copy change observed for these 3 genes in individuals with ASD, the bouton count at the NMJ is reduced in all models, supporting a convergent phenotypic outcome in both Drosophila and human.
We next considered the four CNVs that each affected multiple genes within the ASD-associated network. For each, we asked whether the dosage change of their Drosophila orthologues singularly or in pairwise combination yielded NMJ synaptic or circadian abnormalities. The number of ASD-associated network candidate genes in each of the five CNVs with multiple candidate genes ranged from 2–6, with a mean of ∼4. The four CNVs consisted of three loss CNVs (11079_chr3_loss_197208363_l with two candidate genes; 12239_chr22_loss_17249508_l with five candidates; 12691.p1_chr16_loss_68529466_s with six candidates) and one gain CNVs (12235_chr9_gain_129907917_l with two candidates) (Table 1)
The first multiple candidate gene CNV studied, human de novo loss CNV 11079_chr3_197208363 (Fig. 3A), contained two candidates: the septate junction protein discs large (dlg) and p21-activated kinase (pak), a serine/threonine-protein kinase [36], which has been previously shown to control the synaptic Dlg localisation. Isogenised transheterozygotes of the mutants dlg (dlg1) and pak (pak6) were used and bouton number analysed for synaptic alterations. Single dlg and pak heterozygous mutants alone displayed no significant change in NMJ morphology when compared to controls (Fig. 3B, C), whilst homozygous mutants are lethal, as previously reported [37,38]. However, dlg/pak transheterozygotes (although the correct full geneotype is w1118, dlg1; +/+; pak6/+ for this example, all transheterozygtes will be represented in the ‘gene/gene’ format going forward, for simplicity) displayed significant bouton number reductions (n>20 Kruskal-Wallis test, ** P<0.01; Fig. 3C). For additional controls, Fsn (using FsnKG08128) and CG5359 (using CG5359e03976), which were selected from genes found within CNV 11079_chr3_197208363 but did not participate in the ASD-associated network, were crossed to dlg and pak heterozygotes but no significant NMJ morphology changes were observed (Fig. 3D). To look for circadian behavioural phenotypes, day/night sleep patterns of adult flies were again analysed. Wild type flies and all negative controls (transheterozygote crosses with FsnKG08128 and CG5359e03976; Fig. 3E, F) and single mutants displayed normal light/dark differences in sleeping patterns, with more sleeping periods in the dark. However, dlg/pak flies lost the dark bias (Fig. 3E), displaying no significant difference between light/dark sleeping patterns. Thus, dlg/pak flies demonstrated synergistic effects, displaying both reduced NMJ bouton number and circadian rhythm defects only in the transheterozygotes.
Analysis of a second human de novo loss (12239_chr22_loss_17249508_l; Table 1; Fig. 4A), covering the recurrent 22q11.2 microduplication critical region [39], found no evidence of abnormalities in NMJ bouton count nor circadian cycle in the single heterozygote mutants of any of the 7 genes examined (5 candidates and 2 controls; Fig. 4). However, the two transheterozygous combinations of partner of drosha (pasha; using pashaLL03360) [40] with optomotor-blind-related-gene-1 (org-1, using org-1MB01466) [41] and that of pasha with Septin4 (Sep4, using Sep4NP7170) were both found to have reduced bouton numbers (Fig. 4B; n>20, Kruskal-Wallis test, ** P<0.01, *p<0.05). These relations, however, were not transitive as the combination of org-1 and Sep4 (org-1/Sep4) did not yield these phenotypes. Similarly, only the org-1/pasha and pasha/Sep4 transheterozygote flies also lost the dark bias, displaying no significant difference between light/dark sleeping patterns while org-1/Sep4 did not (Fig. 4D). No significant NMJ morphology or sleep/rest changes were seen when negative controls hira (using hira185b) and sea (using seaEP3364) were crossed to form transheterozygotes with the candidates (Fig. 4C, E). Thus, again, we observe synergistic combinatorial effects, with both NMJ bouton number and circadian rhythm defects apparent only in the transheterozygotes for this second multigenic loss CNV gene set. However, a final multigenic loss CNV gene set, 12691.p1_chr16_loss_68529466_s, failed to yield any significant NMJ bouton number or circadian defects amongst single or pairwise heterozygotes (Table 1; S1 Fig.).
We next analysed a gene set derived from a copy number gain (12235_chr9_gain_129907917_l, Fig. 5A), by generating constructs for overexpression, and by employing the UAS-GAL4 over-expression system. The two ASD-associated network genes, dynamin (Shibire) and alpha spectrin, when over-expressed together display a decreased NMJ bouton number (Fig. 5B) and lost the dark bias to sleep (Fig. 5C). The observed decrease in bouton number following pairwise over-expression of these candidate genes duplicated in humans with ASD is consistent with the bouton number decrease also observed among the pairwise disruptions of candidate genes found to be deleted in humans with ASD. Although the dynamin (Shibire) over-expresser alone also showed a loss of dark sleep bias in this case, individually-driven genes displayed no significant change in NMJ morphology over w1118 controls. No significant NMJ morphology changes are seen when non-ASD-network controls from the CNV gene set Su(P) (using Su(P)EY13245) and CG14104 (using CG14104f07593) are crossed into the overexpressing backgrounds (Fig. 5B for NMJ analysis and Fig. 5C for sleep/rest analysis).
Taking all the polygenic models together, with one exception (dynamin (Shibire) dark bias; Fig. 5B), we show that only particular pairwise combinations of dosage change generate interactions that yield neurological phenotypes comparable to those observed in the monogenic models (Figs. 2–5). As with the mongenic CNV gene sets examined, among the 3 CNV gene sets that demonstrate pairwise interactions, we observe directionality effects in NMJ bouton count that are consistent with a convergent phenotypic outcome. Finally, singularly or in pairwise combinations, we observed no phenotypes for any model involving non-ASD-associated network genes.
Understanding the functional relationships between genes underlying ASD will help elucidate the processes that lead to neurological dysfunction and ultimately may pinpoint common mechanisms that lead to the disorder. To test the relationship between our candidate genes and a known ASD candidate we crossed subsets of our candidates with neurexin IV, the orthologue of the autism gene CNTNAP2. From our candidate list we selected dlg, pak and p120ctn which have functional roles in cell adhesion processes that may involve neurexin IV [35]. We crossed heterozygous dlg (dlg1), pak (pak6) and p120ctn homozygotes (p120ctn308/p120ctn308) to a sensitised background of NrxIV (Nrx-IV4304/+) and analysed bouton number. In all three cases, the transheterozgotes of each of dlg, pak and p120ctn/p120ctn in combination with NrxIV (Nrx-IV4304/+) synergistically yielded reduced bouton number and displayed a loss in the dark bias to sleep suggesting that these proteins may act in the same pathway (Fig. 6A, B for NMJ analysis and Fig. 6D, E for sleep/rest analysis). It is worth noting that p120ctn/p120ctn flies in combination with NrxIV had a significantly reduced survival.
We next crossed pasha, sep4, and org-1 heteozygotes with neurexin IV to see if modification of the NMJ and dark sleep bias was a common feature when alleles were present in the sensitised NrxIV (Nrx-IV4304/+) background. In these cases no significant changes to the NMJ phenotypes (Fig. 6C) or sleep/rest rhythms (Fig. 6F) were observed suggesting that pasha, Sep4 and org-1 are acting on non-converging pathways.
To better understand how these interacting and non-interacting gene pairs exert common circadian and synaptic phenotypic effects, we next looked for molecular defects at the synapse. Single homozygous mutations of Drosophila ASD gene orthologues display defects in synapse development [18,42]. Examples of these defects include alterations in glutamate receptors abundances, active zone numbers, and presynaptic and postsynaptic structural defects at the larval NMJ [15,17,18,21,30,42]. To investigate whether gene dosage changes from the transheterozygote subsets above cause molecular synaptic defects, we looked for alterations in active zone localisation and glutamate receptors abundance at the Drosophila larval NMJ. Drosophila active zones are identified by staining with the protein bruchpilot (BRP, Fig. 7A), which is positioned presynaptically and opposite to the postsynaptic neurotransmitter receptors. We measured BRP foci and normalised them to bouton area from transheterozygotes of NrxIV, dlg, pak, pasha, Sep4 and org-1. We found that all transheterozygous crosses between nrxIV, dlg and pak, which we have shown to genetically interact, displayed a reduction in BRP localisation at the synapse (Fig. 7B, C). However, this was not observed for the genetically-interacting transheterozygotes pasha/Sep4 or pasha/org-1 or single heterozygous mutants and controls. We next explored whether dosage changes in our candidate genes might lead to the destabilisation of the clustering of the postsynaptic glutamate receptors, by studying the levels of GluRIIA at the synapse. Again, alterations in glutamate receptor subtypes have been discovered in single homozygous mutations of Drosophila ASD gene orthologues [42,43]. In this case, we found that out of all single and transheterozygote crosses, only pasha/Sep4 displayed a reduction in the levels of GluRIIA (Fig. 7D, E). Taken together, our findings demonstrate that distinct molecular developmental alterations are associated with the different genetically interacting gene combinations, supporting the idea that distinct molecular aetiologies may contribute to ASD by converging on common phenotypic outcomes (Fig. 7F).
In this study we have developed an in vivo model system in Drosophila to determine how genes can synergistically interact within ASD associated de novo CNVs. Specifically, we have shown that (i) of the 4 human CNVs containing 2, 2, 5 and 6 network-identified candidate genes respectively (from a combined total of 114 copy-changed protein-coding genes), pairwise interactions between Drosophila orthologues yielding changes in the neuromuscular junction (NMJ) bouton number and circadian rhythms were observed for 3 CNVs; (ii) that the interactions observed are synergistic, as opposed to additive, in nature, and (iii) that the synaptic bouton counts observed following the simultaneous dosage change of all 5 pairs of interacting CNV candidate genes’ orthologues within Drosophila support a convergent phenotypic outcome arising from these genes’ dosage change for the individuals with ASD within whom they were identified (Figs. 2–5, S2). We show that the combinations of genes drawn from these CNVs that interact are specific, both within a CNV (Fig. 4) and between CNVs (Fig. 6), supporting distinct aetiologies underlying ASD. Finally, we go on to show these specific interactions act through different molecular aetiologies, supporting the role of distinct molecular aetiologies in ASD (Fig. 7).
The synergistic, as opposed to additive, nature of the pairwise genetic interactions that we observe in Drosophila has important consequences for identifying the genetic causes of ASD, and (i) the conserved orthology of the interactors, (ii) the human orthologues’ participation in an ASD-relevant network constructed from known mammalian interactions, and (iii) the concordance between the direction of dosage change and phenotype all support the inter-species relevance of our findings. Although there are over 100 ASD candidate genes currently identified, at least 70% of the genetic causes remain to be explained [9,44]. The presence of multiple genetic variants in many patients [29,45] suggests that inherited variants might lead to ASD through the combinatorial effects of distinct deleterious variants which affect a shared biological pathway (Fig. 6) [10]. Where variants that act additively to cause ASD in a proband are inherited from each parent, those variants individually may cause detectable ASD-relevant traits in the parents [10,46,47]. However, if combinations of variants act only synergistically to cause ASD, there would be no expectation of ASD-relevant traits in either parent. Importantly, if sub-threshold ASD traits affect fecundity then variants that are only deleterious in combination may rise to a higher frequency in the population. Our results in Drosophila show that only particular combinations of dosage variants act together to yield an abnormal phenotype (Fig. 4 and Fig. 6). Identifying those variants that contribute to ASD only in combination with other specific variants, amongst a background of large amounts of non-contributing genetic variation, will be challenging because the variety of gene variant combinations is extremely large, and allele frequencies are likely very rare.
The genes participating in the pairwise genetic interactions identified by our screen were discs large (dlg: human orthologue (h.o) DLG1), p21-activated kinase (pak: h.o. PAK2), p20 catenin (p120ctn: h.o. CTNND2), Notch (N: h.o Notch 1), shibire/dynamin (shi/dynamin: h.o. DNM1), alpha-Spectrin (α-spec: h.o. SPTAN1), optomotor-blind-related-gene-1 (org-1: h.o. TBX1), partner of drosha (pasha: h.o. DGCR8) and Septin 4 (Sep4: h.o. SEPT5). An examination of CNVs listed in the Database of Genomic Variants (DGV) [48] reveals that most of these genes are found to be individually dosage changed in the same direction in apparently healthy individuals (DLG1, 7 CNVs; PAK2, 1 CNV; DNM1, 1 CNV; SPTAN, 1 CNV; SEPT5, 2 CNVs; TBX1, 9 CNVs; DGCR8 5 CNVs). However, only one of these CNVs might simultaneously change two genes that our study demonstrate genetically-interact in the fly (variant nsv828939; [49]) and CNVs strongly implicated in ASD have previously been reported in apparently healthy individuals [47,50]
Many of the interacting genes have known functions in the nervous system. For example the localisation of the septate junction and neuronal adhesion protein Dlg at the NMJ has been shown to be regulated by Pak serine/threonine-protein kinase activity [36]. In addition, it is interesting to point out that p21-activated kinase (PAK) has been shown to interact with the protein SHANK3 in rat, whose disruption can also cause ASD, with mutant Shank3 altering actin dynamics driven by PAK signalling [51]. Destabilisation of the actin filaments at the NMJ leads to defective NMDAR-mediated synaptic current in neurons. PAK inhibitors have also been shown to rescue fragile X syndrome phenotypes in Fmr1 KO mice [52], suggesting an important role for Pak serine/threonine-protein kinase activity in ASD and ID. The gene alpha-spectrin, which we show genetically interacts with the dynamin protein shabire [53], is known to cross link actin, and has been shown to be important for the localisation of Dlg at the synapse [54]. The phenotypes resulting from the combination of these genes’ variants suggests an important role for the control of synapse integrity via actin stabilisation in ASD [55]. This again is supported up by a particular enrichment for genes directly and indirectly associated with both cell adhesion and cytoskeletal associated cell membrane proteins in our interacting genes (5 out of 9; discs large, p120 catenin, Notch, alpha-spectrin, pak), several of which have been identified to have properties in the neuron [54,56–59]. Many studies have linked neurodevelopmental disorders, including ASD, to mutations in synaptic adhesion proteins, including the neurexins and neuroligins, and mutations in these in Drosophila have yielded both behavioural and larval NMJ defects [30,31,60]. We show specific interactions between P120ctn, dlg and pak with Drosophila neurexin IV, which has been shown to be involved in the maturation of the Drosophila NMJ. [34,61,62]. Notably, the ASD-network orthologues (namely org-1, pasha and sep4) that contribute to the interactions modelling the CNV 12239_chr22_loss_17249508_l that covers the 22q11.2 microdeletion critical region, did not yield phenotypes in the sensitised NrxIV background (Fig. 6) suggesting that these intracellular genes may be exerting phenotypic effects through an alternative process. While other (non-ASD network) genes in this 22q11.2 critical region have received interest in effecting the many associated phenotypes, our study suggests that interactions between the human genes TBX1, DGCR8 and SEPT5 may play a significant causal role [39].
Alterations in active zone structures have been connoted in ASD [63]. Moreover, neuron specific knockdown of the Drosophila orthologues of the ASD genes CNTNAP2 and NRXN1, NrxIX and Nrx-1 (dnrx), have been shown to alter the levels of the active zone protein BRP [18]. BRP shows both sequence and functional homology with the mammalian ELKS/CAST proteins that are structural components of the vertebrate active zone [64,65]. Here we show that dosage changes created by transheretozygotes between NrxIV, dlg and pak lead to a reduction in BRP foci. Dlg is a postsynaptic anchoring protein which is required for the development and stability of the postsynaptic subsynaptic reticulum (SSR), whilst Pak is known to phosphorylate Dlg and control its abundance at the synapse [36]. NrxIV is predominantly presynaptic, but is required for the cell-cell contacts that influence synaptic development [66], and govern the interconnectivity between both neurons, glial cells and the pre- and postsynapse [30]. Dosage alterations in NrxIV with Dlg, Pak and p120 catenin may lead to alterations in adhesion protein interactions, causing the destabilisation of the synaptic architecture in both the pre- and postsynapse, ultimately leading to defective synaptic maturation. In the null mutant of the Drosophila orthologue of NRXN1, Nrx-1 (dnrx), GluRIIA subunit fluorescence and BRP active zone density were increased, although bouton numbers still remain reduced [62]. It has been suggested that interactions between Drosophila neurexins and neuroligins may synchronise GluRIIA, and presynaptic active zone neurexin and neuroligin may be involved in the link between GluRIIA expression and presynaptic active zone dynamics [30,62]. The interactions observed between P120ctn, NrxIV dlg and pak also result in synaptic maturation defects. Null mutants in pak and dlg have also been shown to lead to alterations in glutamate receptor subunits (GluRIIA) [36], however, here we did not see a significant interaction between the dlg/pak transheterozygotes, or the interactions with NrxIV. GluRIIA levels were affected in the pasha/Sep4 cross. Reductions in GluRIIA have been found to lead to a compensatory increase in active zone size [67]. We did not observe a change in active zone puncta in the pasha/Sep4 cross, suggesting that these compensatory mechanisms may be compromised in this case. It is also worth noting that, through changes in the mammalian target of rapamycin mTOR, altered eIF4E-dependent translation results in ASD-relevant phenotypes in mouse [68] and altered regulation of the synthesis of neuroligins. Mutations in Drosophila TOR and eIF4E alter levels of GluRIIA but do not alter the active zones [69]. Interestingly, the fragile X syndrome associated protein FMRP (fragile X syndrome has 30% co-morbidity with ASD) and the miRNA pathway are known to mechanistically interact [70] (Pasha, is part of the miRNA microprocessor complex), while the mRNA of the Sept4 human orthologue (SEPT5) is an FMRP target [10]. Both FMRP, which is known to pause ribosomal translocation [71], and Pasha are involved in translational repression [72,73]. In addition, both mutations in FMRP and the microRNA processing machinery affect the ratios of GluR subunits [43,74]. It may be that pasha/Sep4 deficit leads to the suboptimal translation of Sep4, which functions in complexes that associate with cellular membranes and actin filaments. This may lead to inefficient synaptic anchoring. Further analysis of this process, and those arising from the gene-gene interactions in this study, can now be performed. In summary, our in vivo model system may be well suited to rapidly evaluate how combinations of genes may contribute synergistically to the neurological defects that, in turn, may contribute to ASD.
Although our data strongly supports a significant causal role for synergistic effects underlying ASD, our current study design is unable to reliably estimate the extent as it was limited to (i) considering only pairwise interactions among sets of candidate genes, defined as those genes whose protein products were identified as participating in an ASD-associated interaction network [9], (ii) considering a limited number of neurological phenotypes studied in the model organism Drosophila [11] and (iii) our study considered only those 4 multigenic de novo CNVs identified in individuals with ASD in previous studies where each candidate gene possessed a unique Drosophila orthologue (see Methods). Given that each CNV in those previous studies affected on average 16 protein-coding genes (including non-network genes), we might only expect only 4 genes to possess unique human:Drosophila orthologues (see Methods), severely limiting the ability of this model to examine all combinations of affected genes. However, given that even 16 genes per CNV would generate 240 pairwise gene combinations, it is difficult to imagine the extent and nature of these interactions being examined in a less tractable model with a higher ratio of unique orthologues. While we employed NMJ analysis as a tractable system for studying synaptic function, and circadian analysis to provide a high throughput method for studying behavioural deficit, it would be interesting to expand the behavioural assays to include those which studied social interaction, such as the social space index [75], and also courtship analysis [76]. Nonetheless, the relevance of our findings in Drosophila to humans is supported by the consistent directional effects observed between the increased or decreased bouton counts, which correspond well with the direction of gene dosage change in the human CNV. Taken together with the fact no non-ASD-associated network gene examined yielded abnormal phenotypic effects, when disrupted singularly or in combination (Figs. 2–5), the development of an informatics-targeted Drosophila-screen presents a rapid approach for identifying disease-relevant candidate interactions.
We considered the four sets of CNVs we informatically examined previously: (1) 73 de novo CNVs from the Autism Genome Project study [5], (2) 28 de novo CNVs from the Marshall et al. study [27], (3) 94 de novo or rare CNVs from the Levy et al. study[28] and (4) 67 de novo or rare CNVs from the Sanders et al. study [29]. On average each CNV overlaps 16 genes with an s.d. of 23 showing wide variation. In order to reduce the combinatorial search space, we considered only those 210 genes whose protein products had been identified in a previous CNV study to participate in a large and highly-significant network of interacting proteins with roles in neural functioning (herein termed the ASD-associated network) [9]. This network provides an aetiological basis through which genetic interactions might be mediated. We downloaded the set of the unique human:Drosophila orthologues as determined by the InParanoid tool [77]. Although our study has strictly focused on unique (1:1) orthologues, we note that a much larger number of Drosophila orthologues could be identified by relaxing the requirement of only a unique human orthologue [78]. Nonetheless, examining the 95 de novo CNVs that harboured genes from the ASD-associated network [9], we identified 7 CNVs for which a unique fly orthologue could be identified for every CNV-overlapped network gene (Table 1; Figs. 2–6, S1, S2). In addition, we selected two non-network genes from each CNV with multiple candidate genes, whose unique fly orthologues were neuronally-expressed in the larval stage. Acknowledging the limited number of unique human: Drosophila orthologues, we were not seeking here to exhaustively ascertain combinatorial effects in Drosophila between all simultaneously copy number changed genes in individuals with autism but rather to investigate the informatically-proposed presence of such effects in vivo (see Discussion). All selected genes were completely overlapped by their respective genes.
All Drosophila stocks were isogenised to the w1118 wild type background for 7 generations. Where possible, previously described amorphic mutants were selected for analysis. Uncharacterised insertions were validated using deficiencies. Stocks were acquired for positive mutation hits from the Bloomington Drosophila Stock Center (BDSC, Indiana University) unless otherwise stated and contained the following insertions or lesions: p120ctn308, dlg1, pak6, htsk06121, locoKG02176, aux727, org-1MB01466 and pashaLL03360, α-Specrg41, Sep4NP7170 (Drosophila Genetic Resource Center, Kyoto Institute of Technology), CG13192EY07746, CG34449d00976, CtBP87De-10, CG8507G4779, httMB03997. w1118; UAS-notchFull, UAS-alpha-spectrin and UAS-Dynactin (Shabire) were used for overexpression relating to gains. To generate these, the coding sequences were amplified using primers containing KpnI sites, subcloned into pUAST, and injected into embryos. The 1032-GAL4 ubiquitous driver was used for overexpression due to its moderate ubiquitous expression. NrxIV4304 (BDSC, Indiana University) was used as a positive control. Negative controls were randomly selected from genes that were not picked as candidates from the CNV set. All negative controls selected displayed both larval and adult neuronal expression (BDSC, Indiana University). w1118; FsnKG08128, w1118; cg5359e03976, w1118; Hira185 w1118; seaEP3364, w1118; Su(P)EY13245, w1118; CG14104f07593, w1118; nelf-aKG09483, w1118; CG8507G4779 and CG3321c00226 were used for negative controls.
All stocks were cultured on standard molasses/maize meal and agar medium in plastic vials or bottles at 25°C. Larvae were reared on apple juice plates supplemented with molasses/maize meal and yeast as previously described [79]. Larvae were selected for NMJ analysis at 5 days post egg laying. For analysis of bouton number was performed on the NMJ innervating muscles 6 and 7 from hemisegment A2 (1). Over 15 larvae were analysed for each genotype. For immunohistochemistry larvae were fixed for 20mins in 4% paraformaldehyde, or Bouin’s fixative for 30 minutes (GluRIIA). Primary antibodies used were anti-discs large (DLG, Developmental Studies Hybridoma Bank (DSHB), Iowa City, Iowa, USA),anti-Fas2 (DSHB), anti-GluRIIA (DSHB) and anti-BRP (DSHB), all used at 1/100. Secondary antibodies used were AlexaFluor 488 goat anti-rabbit and AlexaFluor 633 goat anti-mouse (Invitrogen) at 1/1000, and anti-HRP-TRITC (The Jackson Laboratory, Bar Harbor, Maine, USA). Z-stacks were taken using a laser-scanning confocal microscope (Leica TCS SP5 II confocal microscope) and analysis performed using ImageJ and Adobe Photoshop. For statistical analysis of the genetic interactions, ANOVA was performed between the control, the two single heterozygous mutations and the transheterozygotes.
For GluRIIA and BRP analysis at the NMJ, synapses were analysed with optical sections of 0.2μm using a laser-scanning confocal microscope (Leica TCS SP5 II confocal microscope) All digital analysis performed using ImageJ. For BRP staining the number of puntcta was scored over the synapse and normalized to synapse area. For GluRIIA analysis the average fluorescence intensity was analysed over the whole synapse (marked by HRP staining) and then normalized to HRP intensity. No alterations in HRP levels were observed in any genotypes.
All stocks and F1 crosses were cultured on standard molasses/maize meal and agar medium in plastic vials or bottles at 25°C within a light/dark cycle of 12 hrs light/ 12 hrs dark (12:12 LD). For overexpressions, flies were reared at 16°C and then switched during late pupation so to mitigate gross developmental defects. The flies were then transferred to 25°C within a light/dark cycle of 12 hrs light/12 hrs dark (12:12 LD). Flies selected for analysis were between 3 and 5 days old. Flies were the transferred to activity tubes containing 5% sucrose and 2% Bacto agar at one end and were continually synchronized and entrained using a light/dark cycle of 12 hrs light/12 hrs dark (12:12 LD) at 25°C in the circadian incubator for 3 days before data collection. The flies were then switched analysed for experimentation and data collection. Sleep/rest periods were identified as contiguous 5 minute periods of inactivity and were scored and averaged over 2 day period for both dark ‘day’ and ‘night’ cycles. The raw binary data is processed using DAM Filescan102X (Trikinetics, Inc.) and summed into 5 minute bins when analysing sleep/rest parameters. Data analysis was performed within Excel. Statistics were performed using student’s t-tests between ‘day’ and ‘night’ activity.
|
10.1371/journal.ppat.1007415 | Characterization of Haartman Institute snake virus-1 (HISV-1) and HISV-like viruses—The representatives of genus Hartmanivirus, family Arenaviridae | The family Arenaviridae comprises three genera, Mammarenavirus, Reptarenavirus and the most recently added Hartmanivirus. Arenaviruses have a bisegmented genome with ambisense coding strategy. For mammarenaviruses and reptarenaviruses the L segment encodes the Z protein (ZP) and the RNA-dependent RNA polymerase, and the S segment encodes the glycoprotein precursor and the nucleoprotein. Herein we report the full length genome and characterization of Haartman Institute snake virus-1 (HISV-1), the putative type species of hartmaniviruses. The L segment of HISV-1 lacks an open-reading frame for ZP, and our analysis of purified HISV-1 particles by SDS-PAGE and electron microscopy further support the lack of ZP. Since we originally identified HISV-1 in co-infection with a reptarenavirus, one could hypothesize that co-infecting reptarenavirus provides the ZP to complement HISV-1. However, we observed that co-infection does not markedly affect the amount of hartmanivirus or reptarenavirus RNA released from infected cells in vitro, indicating that HISV-1 does not benefit from reptarenavirus ZP. Furthermore, we succeeded in generating a pure HISV-1 isolate showing the virus to replicate without ZP. Immunofluorescence and ultrastructural studies demonstrate that, unlike reptarenaviruses, HISV-1 does not produce the intracellular inclusion bodies typical for the reptarenavirus-induced boid inclusion body disease (BIBD). While we observed HISV-1 to be slightly cytopathic for cultured boid cells, the histological and immunohistological investigation of HISV-positive snakes showed no evidence of a pathological effect. The histological analyses also revealed that hartmaniviruses, unlike reptarenaviruses, have a limited tissue tropism. By nucleic acid sequencing, de novo genome assembly, and phylogenetic analyses we identified additional four hartmanivirus species. Finally, we screened 71 individuals from a collection of snakes with BIBD by RT-PCR and found 44 to carry hartmaniviruses. These findings suggest that harmaniviruses are common in captive snake populations, but their relevance and pathogenic potential needs yet to be revealed.
| From the 1930s to 2015 arenaviruses were known as mainly rodent-borne viruses, which occasionally infect humans, causing a severe disease. After isolation of novel arenaviruses from snakes, the family Arenaviridae now comprises three genera, Mammarenavirus, Reptarenavirus, and Hartmanivirus. Here we characterize a pure isolate of the putative hartmanivirus type species, Haartman Institute snake virus-1 (HISV-1) and report three new hartmanivirus species: Old Schoolhouse virus, Dante Muikkunen virus and Veterinary Pathology Zurich virus. The genomes of all these hartmaniviruses lack the matrix protein found in other arenaviruses, and using HISV-1 as a model we demonstrate that hartmaniviruses can nonetheless form infectious particles. Although HISV-1 induces cytopathic changes in cell culture, there was no evidence of a cytopathogenic effect in infected snakes, where hartmaniviruses have a restricted cell tropism. We further identified hartmanivirus infection in over 60% of snakes in a single breeding collection, indicating that hartmaniviruses are relatively common in boid snakes. Several animals of the collection studied were diagnosed with BIBD, and it is not clear whether hartmaniviruses play a role in the development of BIBD in reptarenavirus-infected snakes. The co-infection with viruses from two genera of the same virus family provides an interesting model for future studies.
| The first member of the family Arenaviridae, lymphocytic choriomeningitis virus (LCMV), was identified and isolated already in the 1930s [1]. During the following four decades several novel members of the family were identified including human pathogens such as Junin (JUNV), Machupo (MACV), and Lassa (LASV) viruses [1]. For decades, arenaviruses were known as rodent-borne viruses with the exception of Tacaribe virus (TCRV), which was isolated from a bat [1]. In the early 2010s, three independent groups identified novel arenaviruses as the potential causative agents for boid inclusion body disease (BIBD) [2–6]. BIBD is characterized by intracellular cytoplasmic inclusion bodies (IB) within almost all cell types of affected snakes [4, 7, 8]. The IB mainly (or solely) consist of arenavirus nucleoprotein (NP) [4, 8], and BIBD was recently successfully reproduced by experimental reptarenavirus infection [9]. The identification of the “snake arenaviruses” prompted the establishment of two new genera, Mammarenavirus and Reptarenavirus, within the family Arenaviridae [1]. Snakes with BIBD often, if not always, carry L and S segments of several reptarenavirus species [10, 11]. Furthermore, infected snakes usually harbor more L than S segments, which significantly hampers the taxonomic classification of reptarenaviruses [10–12]. The International Committee on Taxonomy of Viruses (ICTV) Arenaviridae study group has recommended that the PAirwise Sequence Comparison (PASC, available at (https://www.ncbi.nlm.nih.gov/sutils/pasc/viridty.cgi?textpage=overview) tool should be used for genus and species determination [1]. The PASC tool classifies arenaviruses to the same genus if the nucleotide sequence identity in the S segment is >29–40% and >30–35% in the L segment [1, 13]. When analyzing some of the virus isolates of our first paper on BIBD [4], we found a virus genome with coding strategy similar to arenaviruses and named the isolate Haartman Institute snake virus-1 (HISV-1) [10]. Analysis of the HISV-1 genome with the PASC tool showed that HISV-1 represents a novel arenavirus genus, and in 2018 the third genus, Hartmanivirus, was established in the family Arenaviridae [14].
Arenaviruses are RNA viruses with a single-stranded, bisegmented, negative-sense RNA genome and an ambisense coding strategy [1]. The large (L) genome segment encodes matrix/Z protein (ZP) and RNA-dependent RNA polymerase (RdRp) and the small (S) segment encodes glycoprotein precursor (GPC) and nucleoprotein (NP)[1]. Arenaviruses replicate in the cytoplasm of the infected cells, the genome replication and transcription requires both RdRp and NP [15]. Initially, the ZP was also thought to contribute to the latter processes [16] but later studies have demonstrated that ZP rather acts to suppress both [15, 17]. All structural proteins of arenaviruses have essential roles in the arenavirus life cycle: RdRp is required for genome replication, GPC for spike formation to gain cell entry, NP for genome packaging and replication, and ZP for budding and regulation of replication [15, 18]. Additionally, the NPs of all mammarenaviruses but TCRV inhibit type I interferon (IFN-I) induction [19] at multiple steps of the signaling pathway [20]. Likewise, the ZPs of mammarenaviruses that are pathogenic in humans inhibit IFN-I signaling by targeting RIG-I and MDA5 [20]. The ZPs also interact with cellular components such as PML (promyelocytic leukemia protein), eIF4E (eukaryotic translation initiation factor 4E), and the ESCRT (endosomal sorting complexes required for transport) system required for budding [20].
When assembling the genome of HISV-1 we observed that the L segment lacks an open-reading frame (ORF) for the ZP [10]. At that point, we did not have a pure HISV-1 isolate and we could thus neither confirm the latter finding nor could we investigate whether HISV-1 would survive without co-infecting reptarenavirus(es). Herein, we report the isolation and characterization of a pure HISV-1 cell culture isolate demonstrating that infectious virions are produced despite the lack of ZP. We also identified three additional hartmanivirus species and provide the complete coding regions for their genomes along with a number of nearly complete reptarenavirus genome segments. We identified hartmaniviruses in snakes with BIBD, but could also detect hartmanivirus infection in apparently healthy snakes, suggesting that these viruses are not directly linked to BIBD pathogenesis. Even though we could not associate hartmanivirus infection with pathological changes in vivo, we observed cytopathic effects of HISV-1 infection in vitro.
Previously we used next-generation sequencing (NGS) for the characterization of reptarenavirus isolates [4]. This led to the identification of HISV-1, a putative representative for novel arenavirus genus, in addition to several reptarenavirus S and L segments [10]. Even though we had fairly high coverage of the S (11–27092 fold) and L (357–4493 fold) segments of HISV-1, we were unsure whether these represented the full length segments since the L segment comprised an ORF for the RdRp [10], but the ORF for the ZP found in other arenaviruses was missing. Since the original HISV-1 preparation contained also a reptarenavirus (UHV-2, University of Helsinki virus-2), we used serial dilution to obtain single virus isolates of both UHV-2 and HISV-1 (S1 Fig). Successful production of a clean HISV-1 isolate indicated that even though an ORF for the ZP, which functions as the matrix protein, cannot be found in the L segment, HISV-1 is able to replicate without a co-infecting reptarenavirus. For most subsequent comparative experiments we used UHV-2 as the reference reptarenavirus, since it was the co-infecting reptarenavirus of the original HISV-1 isolate. For a few experiments we used UGV-1 (University of Giessen virus-1) instead, as this is a reptarenavirus grows to high titers in our cell culture model. For the sequence analyses we chose to use the type species of each arenavirus genera.
To confirm the lack of an ORF for the ZP in the L segment of HISV-1, we isolated RNA from a batch of purified HISV-1 and sequenced the ends of the S and L segments using T4 RNA ligase to generate cyclic RNAs. The latter then served as templates for RT-PCR over the genome ends (S2 Fig) and yielded several sequences that indeed covered the genome ends of both the S and L segment. In addition to the S and L segment specific primer pairs, we also successfully applied primer combinations, i.e. L segment forward-S segment reverse and S segment forward-L segment reverse, in RT-PCRs, which suggested that T4 RNA ligation also produced S and L segment chimeras. This was confirmed by sequencing. The genome end sequencing then confirmed that the L segment of HISV-1 indeed lacks the matrix protein/ZP found in other arenaviruses. Subsequent NGS and de novo genome assembly for the purified HISV-1 preparation did not identify additional genome segments. The consensus sequence of the S segment revealed a nucleotide insertion (a stretch of 7 instead of 6 adenines at 575–581) in the GP1 ORF of our original submission [10], which led to incomplete “in silico” translation of the GPC.
Obtaining the full length genome segments of HISV-1 allowed us to compare the genomes of the three arenavirus genera, Mammarenavirus, Reptarenavirus, and Hartmanivirus. Since this is the first time that full length genome segments of all arenavirus genera are available, we decided to perform a bioinformatics-based comparison of their genomes and proteomes, and selected the type species of each genus for these “in silico” comparisons. The S segments of all three genera, schematically depicted in Fig 1A, are identical in their coding strategy and similar in size. However, while the L segments of mammarenaviruses and reptarenaviruses share the same coding strategy and are similar in size, the L segment of hartmaniviruses lacks the ZP ORF (Fig 1A).
As shown in Fig 1B, the genome ends–represented by 21 terminal nucleotides–of all arenavirus species have the ability to form a panhandle structure. The panhandles of LCMV and Golden Gate virus (GGV) L segments comprise 18 consecutive complementary nucleotides, while the corresponding region in the HISV-1 L segment contains two non-paired nucleotides. Comparison of the genome segment ends revealed a conserved CGCACxGxGxA motif at the 5´ end of mammarena-, reptarena-, and hartmanivirus S and L segments (shown in bold in Fig 1C). Similarly, the 3´ends show conserved nucleotides, GCGUGxCxCCU (shown in bold in Fig 1C), complementary to those found at the 5´ end with the underlined residue making an exception. While the predicted panhandle structures may differ even between viruses of the same species, the overall panhandle structure is maintained throughout arenavirus genera by the aforementioned conserved nucleotides. The terminal complementarity of the RNA is essential for replication and transcription [21] which, at least partially, may explain the conservation of the segment ends throughout the family. The non-conserved sites at both ends are speculated not to contribute to sequence-specific interaction with the RdRp [22], thus explaining the observed variation at these sites.
The proteomes and the amino acid identities between the corresponding proteins (based on MAFFT alignment) of the three genera are presented in S2 Table. The major difference in the proteomes is the lack of ZP in hartmaniviruses. While the ZPs of reptarenaviruses and mammarenaviruses share only 16% amino acid identity, their functions are assumed to be similar [2]. Interestingly, reptarenavirus ZPs have an N-terminal transmembrane helix (TM) [2, 4], whereas the N-terminus of mammarenaviruses is myristoylated [2]. The structural proteins of HISV-1 are all 20 to 23% identical to their LCMV and GGV counterparts. The RdRp and NP of GGV are slightly closer to LCMV (28% and 32% identical) than to HISV (20% and 20% identical). However, the GPCs of HISV-1 and LCMV (23% identity) are more similar than the GPCs of LCMV and GGV (16% identity). See S2 Table for more detail.
The GPC of each arenavirus genera are schematically presented in Fig 2A. By prediction the GPCs contain several N-glycosylation sites: 7 in HISV-1 (4 in GP1 and 3 in GP2) and LCMV (5 in GP1 and 2 in GP2) and 9 in GGV (7 in GP1 and 2 in GP2) (Fig 2A). The cleavage between GP1 and GP2 is mediated by subtilisin-kexin isozyme-1/site-1 protease (SKI-1/S1P) for mammarenaviruses (Fig 2A). Using ELM (eukaryotic linear motif resource, http://elm.eu.org/, [23]) we identified a potential SKI-1/S1P cleavage site in GGV GPC, and the GPC alignment of known reptarenavirus species suggests that the site is conserved (Fig 2A). For HISV-1 we could not detect a SKI-1/S1P cleavage site but rather identified a potent furin cleavage site in the same region. By comparing the region to GPCs of the other hartmanivirus species (described later in the manuscript) we could show that the furin cleavage site is preserved among the hartmaniviruses found thus far (Fig 2A).
A more thorough investigation of the GPCs indeed shows similarities between HISV-1 and LCMV. To begin with, both HISV-1 and LCMV have long (55 and 58 residues, respectively) signal peptides (SP) that by prediction are myristoylated at position 2, while the SP of GGV is only 23 residues and lacks the myristoylation site (Fig 2A). For mammarenaviruses the SP remains in the virion and is known as a stable signal peptide (SSP) that interacts with GP2 via an intermolecular zinc-binding motif [24] (Fig 2C). Sequence comparison between mammarenavirus and hartmanivirus SSPs and GP2s (Fig 2B) shows conservation at the histidine and cysteine residues required for the SSP-GP2 interaction. Additionally, the GP2s of both mammarenaviruses and hartmaniviruses have a relatively long cytoplasmic tails (48 and 42 residues respectively) while the GP2 tail of reptarenaviruses only comprises a few (two in GGV) residues (Fig 2A and 2B). The above suggests that there might be differences in the organization of the spikes complex between reptarenaviruses and the other two arenavirus genera. Of note, the GP2 is the most conserved reptarenavirus protein, sharing 68–99% sequence identity between all the known species and 87–99% identity when CASV and UHV-1 are excluded. The GP2s of both mammarenaviruses and hartmaniviruses appear to be more variable.
To monitor and characterize HISV-1 infection in cell culture, we produced an antiserum against the HISV-1 NP which we refer to as anti-HISV NP antiserum throughout the manuscript. We used the C-terminal half of the NP as the antigen, since this strategy had produced a broadly reactive anti-UHV-1 NP antiserum [10] referred to as anti-UHV NP antiserum throughout the manuscript. To test the anti-HISV NP antiserum, we performed infection and co-infection experiments with UHV-2 and HISV-1 on cultured boid kidney cells (I/1Ki), and analyzed the infected cells by western blotting, immunofluorescence (IF) staining, and immunohistology (IH) (Fig 3). The western blot shows that the anti-HISV NP antiserum does not cross-react with UHV-2 NP, and vice versa (Fig 3A). The IF staining of infected cells concurs with the western blotting results, and also indicates that neither of the antisera reacts with cellular proteins at the concentrations applied (Fig 3A). A comparison of the IF staining patterns in infected cells showed that HISV-1 infection produces large fluorescent foci (i.e. infection of several cells in close proximity to each other), whereas UHV-2 infection resulted in scattered individual positive cells (100x magnification in Fig 3B). At a higher magnification, UHV-2 infected cells were found to exhibit the punctate NP staining pattern typical for reptarenavirus infected cells with inclusion bodies (IB)[4, 25], whereas HISV-1 NP appeared to be more diffusely distributed in the cells (400x magnification in Fig 3B). A similar staining pattern was seen in HISV-1 infected I/1Ki cell pellets, indicating that the antiserum is suitable for immunohistology (IH) (Fig 3C).
Since we did not see a marked difference in the amount of viral NP in western blots when comparing single-virus infection to co-infection (Fig 3A), we decided to study the effect of co-infection on virus replication by quantifying the amount of viral RNA released from infected cells, using qRT-PCR. We initially used UHV-2 and HISV-1 for the experiment, since these viruses originated from the same isolate. Fig 4A shows representative results of one of the three consecutive experiments. We found that the amount of UHV-2 RNA released in co-infection was not reduced as compared to the single virus infection (Fig 4A), indicating that co-infection does not markedly affect the replication rate of reptarenaviruses and hartmaniviruses. To provide further evidence for this observation, we performed a similar co-infection experiment using HISV-1 and UGV-1. We decided to use UGV-1, the virus of which S segment we most often find in snakes with BIBD (other authors call this S segment S6 [11]). Also, UGV-1 grows to high titers in our cell culture model. We obtained results comparable to those obtained with UHV-2 (Fig 4A and 4B), indicating that reptarenaviruses and hartmaniviruses do not interfere with each other’s replication during co-infection. We also tried infecting cultured mammalian cell lines (Baby hamster kidney, BHK-21; African green monkey kidney, Vero E6) with HISV-1 at 30°C and 37°C, however, we could not detect viral antigen or replication.
As indicated by IF staining (Fig 3B) the reptarenavirus NP forms large IB in infected cells, which we and others have described in detail [2, 4]. However, the IF staining suggested that such IB would not be present in HISV-1 infected cells. Therefore, we performed an ultrastructural study of HISV-1 infected snake cells. Indeed, we could not demonstrate IB in HISV-1 infected cells by electron microscopy (EM). Instead, the HISV-1 infected cells generally exhibited extensive cytoplasmic vacuolization and blebbing of the plasma membrane (PM) (Fig 5A to 5C). We also observed tubular structures within the cells (Fig 5B), which appeared to contain electron electron-dense material similar to the blebs at the PM (Fig 5B to 5D). For comparison, an example of reptarenavirus infection-associated electron dense IB is presented in Fig 5E. We assumed that the electron-dense structures observed in HISV-1 infected cells would contain and/or consist of viral proteins. Indeed, immuno-EM revealed that both the tubular structures and the blebs at plasma membrane contain HISV-1 NP (Fig 6); the latter often appeared to be continuous between two adjacent cells (Fig 6E, inset) suggesting direct cell-to-cell contact. We found HISV-1 NP also associated with vacuoles (Fig 6B and 6C) and in the nucleai/along the nuclear membrane (Fig 6G), which is in line with the IF staining (Fig 3B). The HISV-1 NP was found to be abundant in the tubular structures with the infected cells (Fig 6D, 6F and 6G). Viral antigen expression in the blebs at the PM and in the intracellular tubular structures together with the detection of large foci of infected cells in IF staining could suggest that cell-to-cell spreading plays a greater role in the spread of hartmaniviruses than of reptarenaviruses. For comparison, the NP of the reptarenavirus (UGV-1) accumulated in the cytoplasmic inclusion bodies (Fig 7).
Since we could not find evidence of ZP at genome level, we decided to analyze purified HISV-1 virions in comparison to those of purified reptarenavirus (UGV-1). We used UGV-1 for this approach, as it produces abundant virions in our cell culture model. For the structural study, we purified the virions by density gradient ultracentrifugation (Fig 8A). The major band in the main fractions containing HISV-1 (F6-F8 in Fig 8A) represents the NP, as demonstrated by western blotting (Fig 8B). We then compared the SDS-PAGE patterns of purified UGV-1 and HISV-1 to find more evidence on the lack of ZP in hartmaniviruses. The NP appears as the most abundant protein for both UGV-1 and HISV-1 with an approximate size of 70 kDa by SDS-PAGE mobility (Fig 8B). The exact sizes of GP1 and GP2 are not known, however, by comparing the concentrated supernatants produced by the same cell line infected with reptarenavirus vs. hartmanivirus, the indicated bands most likely represent the viral glycoproteins (Fig 8B). A doublet band at around 16–17 kDa is evident in the UGV-1 preparation, which is slightly larger than the calculated molecular weight, 12.7 kDa, of the ZP. The observed difference between predicted molecular weight (M.W.) and SDS-PAGE mobility could be due to palmitoylation, since the ZP is by prediction palmitoylated at 3 sites (http://csspalm.biocuckoo.org/online.php). As expected, the corresponding bands are missing from the HISV-1 preparation (Fig 8B), providing further evidence that hartmaniviruses lack the ZP. The result also suggests that the lack of ZP is not complemented by a similar(-sized) cellular protein. We then studied the HISV-1 virions in EM under negative staining. The virions appeared pleomorphic to roundish with an approximate diameter of 120–150 nm and most of the particles seemed not to be intact (top panels in Fig 8C). We then analyzed HISV-1 virions in cryo-EM, which provided, in agreement with the SDS-PAGE analysis, no evidence of continuous density lining the inside of the membrane (Fig 8C). A comparison of the appearance of HISV-1 and UGV-1 virions in cryo-EM confirmed that the overall morphology and size (roughly 120–150 nm in diameter) of the virions is similar (Fig 8C).
We recently studied several Boa constrictor clutches born to parental animals with BIBD, aiming to demonstrate or rule out vertically transmission of reptarenaviruses [12]. We used NGS to initially define the reptarenavirome of each clutch, and confirmed our findings by species-specific RT-PCRs. However, for one of the five clutches we used only RT-PCR to demonstrate the virus transmission, since we already had analyzed three clutches from the same breeder by NGS [12]. Some of our species-specific RT-PCRs designed based on the NGS results of other clutches did not work optimally with the samples from clutch #5 [12], which prompted us to study some of the animals (snakes 2.1–2.5, Table 1) by NGS and de novo genome assembly. We found a hartmanivirus, designated as Veterinary Pathology Zurich virus-1 (VPZV-1), accompanied by three reptarenavirus L and two S segments in the liver of the father of this clutch (snake 2.1, Table 1) and in some of the 12 to 20-month-old offspring (snakes 2.2–2.5, Table 1) as well as one additional reptarenavirus L segment in a pooled sample of the offspring (2.2, Table 1). The GenBank accessions for the hartmaniviruses and reptarenaviruses identified in this study are provided in S1 Table. These findings suggest that also hartmaniviruses can be vertically transmitted, although we cannot entirely rule out the transmission after birth for these juvenile snakes. We also identified VPZV-1 by NGS in liver and brain samples of an adult snake from a different breeder (2.6, Table 1), again together with several reptarenavirus L and S segments. By studying some of the cell culture isolates of an earlier study [4] (animal 2.7, Table 1) with the NGS approach, and found a virus which according to ICTV criteria represents a genetically distinct lineage of the VPZV species (designated ad VPZV-2). This virus was accompanied by three reptarenavirus L and one S segments.
In another snake (1.4, Table 1) from the breeder of snake 2.6, we found a virus that by sequence comparison represents the same species as HISV-1 and thus named it HISV-2. The next hartmaniviruses we identified in a pooled blood sample of snakes (3.1, Table 1) with confirmed BIBD. According to ICTV criteria the identified viruses represent yet another hartmanivirus species, and were designated as Old Schoolhouse virus-1 and -2 (OScV-1 and -2). Subsequently, we found another representative of OScV-2 in a pooled blood sample (3.2, Table 1) of snakes with BIBD from another breeding colony. Finally, we identified a putative representative of a fourth hartmanivirus species (4.1, Table 1), which we named Dante Muikkunen virus-1 (DaMV-1), in a snake with mild neurological signs suspected to be associated with BIBD. In addition to identifying DaMV-1 and reptarenavirus segments, we also found the snake to carry a novel deltavirus, which we describe outside this report [26].
Congruently with PASC analysis (S3 Table), the phylogenetic analyses suggested that the novel hartmaniviruses clustered according to their tentative species designations on the basis of their L and S segments (Fig 9). The phylogenetic analysis of the RdRp amino acid sequences of the representatives of all known arenavirus species suggested that the genus Hartmanivirus forms an outlying group separate from mammarenaviruses and reptarenaviruses, whereas recently found Wenling frogfish arenaviruses [27] form an outlying group separate from the hartmani-, reptarena- and mammarenaviruses (Fig 9).
To gain some information on the prevalence of hartmaniviruses in larger populations, we designed primers based on the L segments of OScV-1 and -2, and used RT-PCR to screen a set of 71 blood samples collected from the same breeding colony, in which we initially detected these viruses. Interestingly, 44/71 snakes were RT-PCR positive for OScV-1 and/or -2; of these, one snake was positive for both OScV-1 and -2. Thirty-four snakes in the collection were diagnosed with BIBD as they exhibited IBs in blood cells, of these, 23 had hartmanivirus infection. The fact that close to 70% of snakes with BIBD had an accompanying hartmanivirus infection indicates that the role of hartmanivirus infection in the pathogenesis of BIBD needs to be further investigated.
Production of the anti-HISV NP antibody enabled us to study the tissue and cell tropism of hartmaniviruses, using immunohistology. We initially screened the tissues of the snake from which UHV-2 and HISV-1 originate (animal 1.1, Table 1). In comparison to reptarenavirus NP, which can be detected in most tissues as cytoplasmic IB [28], HISV NP expression was very limited and most consistent in the brain. The neurons exhibited a diffuse, finely granular, cytoplasmic and/or axonal reaction (Fig 10A), which clearly differed from the reptarenavirus NP expression pattern (Fig 10B). Additionally, HISV-1 NP expression was occasionally seen in a range of other cell types: smooth muscle cells in the lung and respiratory epithelial cells (Fig 10C), endothelial cells and medial smooth muscle cells in arteries (Fig 10D), ependymal cells in the brain (Fig 10E), and axons of peripheral nerves (Fig 10F). Examination of another three snakes that harbored HISV-1 or -2 (animals 1.2–1.4) revealed a similar HISV-1 NP expression pattern, though with some variation in the range of cell types (Table 1). In animal 1.4, infected with HISV-2, a broader range of epithelial cells was found to be occasionally positive (pulmonary epithelial cells (Fig 10G), glandular epithelial cells in trachea and stomach (Fig 10H), intestinal epithelial cells, acinar epithelial cells in the exocrine pancreas (Fig 9I) and tubular epithelial cells in the kidney (Fig 10J)). Also, smooth muscle cells in the muscular layers of stomach and intestine as well as dendritic cells in the spleen were found to express the viral antigen (Figs 9 and 10). In animal 1.3, the liver exhibited HISV-1 NP expression in sinusoidal endothelial cells. In none of the hartmanivirus negative animals was there any evidence of NP expression (Table 1). We tested some of the snakes found to be infected with the above HISV-like viruses (VPZV-1: 2.1, 2.3–2.6; DaMV-1: 4.1) by IH for HISV-1 NP. The antibody appeared to cross react only minimally, as the reaction was restricted to a few individual cells in two animals (Table 1). We then tested six snakes that had been HISV-1 negative by RT-PCR (5.1–5.5) or hartmanivirus negative by NGS (5.6) for the expression of HISV-NP, all with a negative result (Table 1).
After a most recent revision the family Arenaviridae currently comprises three genera Mammarenavirus, Reptarenavirus and Hartmanivirus [14]. Until now the genus Hartmanivirus only contained HISV-1 [14], a virus that was only characterized at nucleotide level [10]. The genome of HISV-1 appeared to lack ORF for the ZP, i.e. the matrix protein of mammarenaviruses and reptarenaviruses [10]. We had isolated HISV-1 in a permanent boid kidney cell culture, however the isolate contained two viruses, UHV-2 and HISV-1. Hence, we could not confirm the lack of ZP in the initial report. It was also unclear whether HISV-1 would survive without a co-infecting reptarenavirus. Herein, we report the production of a pure HISV-1 isolate, which represents the type species of the genus Hartmanivirus. The successful generation of a pure HISV-1 isolate indicates that hartmaniviruses can grow in the absence of a co-infecting reptarenavirus and allowed us to study the physical properties of HISV-1 virions at nucleotide, protein and structural level. We then used HISV-1 as the model to characterize hartmanivirus infection at both in vitro and in vivo level. Furthermore, we can expand the genus Hartmaniviruses to four known species, by obtaining complete or near complete genome segments for three new hartmanivirus species. Finally, by screening a snake collection for two hartmaniviruses, we provide first evidence that hartmanivirus infections are rather common in captive snakes.
Comparison of the HISV-1 genome to those of viruses in the other genera of the family Arenaviridae shows similarities, for example in the genome ends, but also a striking difference, namely the lack of an ORF for the ZP, the matrix protein present in the other arenaviruses. We did not find additional genome segments for HISV-1 when we performed NGS on a pure preparation of HISV-1, indicating that HISV-1 comprises an S and L segment like the viruses of other arenavirus genera. We tried to seek evidence for ZP or a ZP surrogate by analyzing a pure HISV-1 preparation by SDS-PAGE and cryo-EM, but again found no evidence. The fact that we found three further hartmanivirus species (with similar coding strategy) provides further support that the lack of ZP is a general feature of the genus Hartmanivirus.
The ZP drives the budding of mammarenaviruses which is mediated by proline-rich late domain motifs PTAP or PPPY [19, 29]. Curiously, the ZP of the bat-borne Tacaribe virus is devoid of the late domains but does still efficiently mediate budding [30]. The ZPs of reptarenaviruses also lack the late domain, which could imply that the putative budding function of reptarenavirus ZP resides in a yet unknown motif. Interestingly, a sequence comparison reveals a conserved late motif, PPPY, in the reptarenavirus NPs, which is not found in the NPs of hartmaniviruses and mammarenaviruses. Also, the C-terminus of reptarenavirus NPs has been reported to contain late domain like motifs [2]. Thus one could speculate that also the NP associates with the budding of reptarenaviruses. Experimental studies are needed to demonstrate the role of individual proteins in the budding of reptarenaviruses, but these are beyond the scope of this report. The fact that hartmaniviruses lack a ZP, but nonetheless efficiently produce infectious particles suggests that the budding function resides in some other structural protein. Alternatively, the virus could induce a cellular protein that aids to virus budding, however we found no evidence of ZP sized proteins in purified HISV-1 preparations. Hantaviruses are similar to hartmaniviruses with regards to their proteins; they also encode RdRp, GPC and NP, but lack a matrix protein. The cytoplasmic tail of the Gn glycoprotein of hantaviruses is suggested to act as a matrix protein surrogate [31]. Thus, one could speculate that the GP2 tail of hartmaniviruses contains a motif that mediates budding. And indeed, it harbors a conserved P-Y/F-P-H-Y-P stretch (Fig 2B), which by ELM (eukaryotic linear motif resource, http://elm.eu.org/ [23]) prediction binds to the apoptosis-linked gene 2 (ALG-2) protein which bridges ALIX (ALG-2-interacting protein X) and ESCRT-I complex via interactions with ALIX, TSG101 and VPS37 [32]. Thus, hartmaniviruses might utilize the interaction with ALG-2 to gain access to the ESCRT pathway for egress, analogously to HIV-1 [33]. Our immuno-EM data suggest that the NP of hartmaniviruses is included in the buds that form along the plasma membrane of infected cells. One could thus speculate that at some point the NP of reptarenavirus ancestors obtained the budding function via emergence of late domains. This would have made the GP2 tail redundant for budding, thus explaining the lack of it in reptarenaviruses. Alternatively, it has been proposed that the GP2 of reptarenaviruses evolved from a recombination event with a filovirus or retrovirus that provided the new gene [2]. Whatever the chain of events, it seems that the ZP could have emerged around the same time to regulate replication and to bridge between the NP and the GPs to facilitate efficient genome packaging. The ZP may also have emerged before speciation of mammarenaviruses and reptarenaviruses, perhaps initially without the late domains. These hypotheses provide interesting topics for further studies, and identification of arenaviruses from other animals could help to shed light on the evolution of arenaviruses.
Comparison of the GPCs between arenavirus genera revealed that mammarenaviruses and hartmaniviruses harbor both SSP and a cytoplasmic tail in their GP2, features missing from the GPC of reptarenaviruses. The N-terminal halves of the mammarena- and hartmanivirus SSPs are more similar than the C-terminal halves. The N-terminus of the SSPs contains a myristoylation motif/site followed by a hydrophobic stretch until a conserved lysine residue in mammarenavirus or RGR motif in hartmanivirus SSPs (Fig 2B). The SSP of mammarenaviruses is suggested to span the viral membrane twice, leaving the conserved lysine residue on the virion surface [24]. Even though the C-terminal halves of hartmanivirus SSPs are less hydrophobic mammarenavirus SSPs, we hypothesize the SSPs to have a similar topology. Supporting the above, we identified a conserved cysteine residue close to the SSP C-terminus (Fig 2B), which participates in formation of an intersubunit zinc-finger structure between two conserved histidine and four cysteine residues of the GP2 [24] (Fig 2C) that are also conserved in both mammarena- and hartmaniviruses. We hypothesize these to indicate similar spike structure between mammarena- and hartmaniviruses (Fig 2C). The fact that phylogenetic analysis, as discussed below, suggests hartmaniviruses to represent the ancestors of mamm- and reptarenaviruses renders the observed differences in the GPC ORF interesting. Because the SSP and GP2 cytoplasmic tail are found in mammarena- and hartmaniviruses but not in reptarenaviruses, the reptarenaviruses seem to have lost these features at some point during their evolution, perhaps in a suggested recombination event with filo- or retroviruses [2]. The fact that GP2s of reptarenaviruses are more conserved than GP2s of mammarena- and hartmaniviruses would support both adaptation to a new niche or the speculative recombination event.
In line with the previous studies [10, 27] the phylogenetic analysis suggested that the hartmaniviruses form a basal lineage for both reptarena- and mammarenaviruses while the Wenling frogfish viruses form an outgroup to the other arenaviruses (Fig 8). Therefore, the phylogeny of arenaviruses resembles, but does not recapitulate the evolution of the respective host species. This suggests that in addition to the apparent co-evolution between arenaviruses and their hosts [27] at least one host-switch event has occurred during the evolution of arenaviruses, potentially from snakes to mammalia. Discovery of arenaviruses from other reptiles or from amphibia would shed more light on the extent of co-evolution and frequency of cross-species transmission events among arenaviruses.
Our IF studies on hartmanivirus (HISV-1), reptarenavirus (UHV-2 and UGV-1) and co-infected cell cultures showed that the distribution of NP within the infected cells varies distinctly between the two genera. We also observed that HISV-1 rapidly induced the occurrence of large foci of infected cells in our cell culture system, while UHV-2 infected cells are scattered and mostly individual. We interpreted this as evidence of more pronounced cell-to-cell spreading of HISV-1. Potential support of this interpretation is the observed bleb formation of the plasma membrane with HISV-1 (Fig 5A–5D; Fig 6A and 6E), but not with reptarenavirus (UHV-2) infection (Fig 7). The membrane blebs contained electron dense material which we assumed to be of viral origin. Indeed, immuno-EM showed that the membrane blebs contain HISV-1 NP (Fig 6E). The latter also accumulated along cytoplasmic nanoscale tubules (Fig 6D, 6F and 6G). Interestingly, influenza viruses were recently shown to utilize tunneling nanotubules (involved in intercellular communication) to transfer viral proteins and genome from infected to naive cells [34]. It is thus tempting to speculate that the NP-loaded (and likely also RNA containing) membrane blebs and cytoplasmic tubules in HISV-1 infected cells are indications that hartmaniviruses employ a similar strategy. Further studies are needed to address the above hypotheses.
The comparative investigation of snakes with BIBD by IH for reptarenavirus and HISV-1 NP provides evidence that hartmaniviruses have a more restricted cell tropism than reptarenaviruses. The mammalian kidney cells (BHK-21 and Vero E6) commonly used for propagation of mammarenaviruses were not permissive for HISV-1. However, both cell lines are permissive for reptarenaviruses when cultured at 30°C, [4, 25], which could indicate broader tissue tropism of reptarenaviruses. The frequency and extent to which hartmaniviruses were detected in neurons of infected snakes suggests a pronounced neurotropism, it therefore seems worth testing if mammalian neuronal cells would be more permissive for HISV-1. Also, further studies are needed to demonstrate the presence or absence of hartmaniviruses in snake secretions. Coincidentally, we found hartmanivirus in “clutch 5” of our previous study [12], and could demonstrate that the father and some of the juvenile offspring were carriers of the same virus. These results are indicative of vertical transmission.
We identified HISV-1 accidentally while aiming to obtain full length genomes for reptarenavirus isolates [10]. Similarly, the viruses identified herein were in the vast majority found in snakes with BIBD. Our earlier observation was that identification and full genome sequencing of reptarenaviruses works very well from brain-derived total RNA. However, in a previous study we noticed that the brain might display only a fraction of the reptarenavirus S and L segments found in the blood [12]. Due to this we have recently focused on studying liver- and/or blood-derived RNA for NGS studies, which has also led to identification of more hartmaniviruses. IH analysis of tissues from snakes with hartmanivirus infection showed that viral antigen is not abundantly expressed in the brain, which suggests that the amount of RNA in the brain is indeed low. So far we have only studied snakes with either suspected or confirmed BIBD, in which the hartmaniviruses were always accompanied by reptarenaviruses. However, by producing a pure isolate of HISV-1, we could demonstrate that hartmaniviruses do not require reptarenavirus co-infection for their infectious cycle in vitro. We further show that reptarena- and hartmanivirus co-infection does not negatively affect the replication of either virus. While we could not associate hartmanivirus infection with any pathological changes, further studies are needed to confirm if hartmanirviruses are apathogenic in snakes. Future studies are also needed to identify the natural host(s) of hartmaniviruses. Also, the recent discovery of a three segmented arenavirus in fish [27] indicates that more arenaviruses are yet to be found with potential to alter the understanding of arenavirus evolution.
The samples included in this study originated from animals submitted by their owners either to the Institute of Veterinary Pathology, Vetsuisse Faculty, University of Zurich, Switzerland, or to the Department of Veterinary Biosciences, Faculty of Veterinary Medicine, University of Helsinki, Finland, for a diagnostic post mortem examination. An Animals Scientific Procedures Act 1986 (ASPA) schedule 1 (appropriate methods of humane killing, http://www.legislation.gov.uk/ukpga/1986/14/schedule/1) procedure was applied to euthanize the snakes. Full diagnostic post mortem examination, blood sampling and diagnostic testing of collected samples were performed with full owners' consent. Ethical permissions for euthanasia and diagnosis-motivated necropsies (both routine veterinary procedures) were not required due to suspicion of a lethal disease, BIBD.
The study was performed on tissues or full blood of 23 snakes that were suspected to suffer from BIBD. A further 68 blood samples from snakes in a private breeding collection which previously had animals dying with BIBD were screened for hartmanivirus infection upon the owner’s request. All animals were captive snakes from breeding collections in Germany and Switzerland, ranging in age from juvenile to more than 12 years (Table 1).
The Boa constrictor kidney cell line, I/1Ki, was used for virus propagation and virus isolation attempts as described [4]. A virus preparation containing HISV-1 and UHV-2 described in [10] was used as the source of pure isolates. The isolation strategy is depicted in S1 Fig. Briefly, 10-fold dilution series of the virus stock were prepared on I/1Ki cells grown on a 96-well plate, and at 14 days post infection (dpi) the cells inoculated with virus dilutions 1:107 and 1:108 were transferred onto a 24-well plate. The cell culture medium was collected at 7 and 14 dpi, and the pooled cell culture supernatants were analyzed by virus species specific RT-PCR as described [10, 12]. For production of HISV-1 stock, a 75-cm2 flask of semi-confluent I/1Ki was inoculated with 500 μl of the pooled supernatant from the 24-well plate, the cell culture medium was collected and replaced at 2–3 day intervals until 14 dpi, and the pooled supernatants were filtered through a 0.45 μm syringe filter (Millipore) and stored at -80°C for further use. Large quantities of HISV-1 were produced by inoculating semi-confluent I/1Ki 75-cm2 flasks with 1 ml of 1/50-1/200 diluted HISV-1 stock, followed by supernatant collection as described above. Viruses were concentrated by pelleting through a sucrose cushion and more pure virus preparations were obtained by sucrose density gradient ultracentrifugation as described [4, 35, 36]. For cryo-electron microscopy (cryo-EM) the virus-containing fractions were pooled and dialyzed against phosphate-buffered saline (PBS). For transmission electron microscopy and immunocytology, cells were infected and harvested six days after inoculation, pellets prepared and fixed as described [4]. For co-infection experiments I/1Ki cells were inoculated either with equal amounts of HISV-1 and UHV-2, or HISV-1 and UHV-2 alone as controls (multiplicity of infection > 1). The infection vs. co-infection experiments were done in duplicate.
Trizol and Trizol LS isolation reagent (Life Technologies) in combination with QiaGEN RNeasy Mini Kit (Qiagen) was used for RNA isolation as described [12]. RNA isolation from cell culture supernatants was done with either the QIAamp Viral RNA Mini Kit (Qiagen) or the GeneJET RNA Purification Kit (Thermo Fisher Scientific) following the manufacturer’s instructions. No carrier RNA was used during RNA isolation for samples analyzed by NGS. RT-PCR served to detect viral RNA from cell culture supernatants as described [10, 12].
HISV-1 RNA isolated from pelleted virus was treated with T4 polynucleotide kinase (Thermo Fisher Scientific) according to the manufacturer’s protocol, purified using the QiaGEN RNeasy Mini Kit (Qiagen), and circularized with T4 RNA ligase (Thermo Fisher Scientific). The RNA circularization reaction (2 h at 25°C) mix (20 μl in DEPC-treated water) included: 2 μl of 10X reaction buffer, 5 μl of isolated RNA, 1 μl of T4 RNA ligase, 0.5 μl RNAse inhibitor (40 μ/μl), 10% PEG 8000, and 100 μM ATP. The reaction ligase was inactivated by heating the reaction mix to 70°C for 10 min. The circularized RNA was reverse transcribed using RevertAid H Minus Reverse Transcriptase (Thermo Fisher Scientific) following the manufacturer’s protocol for specific primers. The S segment primers were 5´-CTCCATTTACTCGAACAAGCTCAC-3´ and 5´-CAGGTTAAATTCATTGTTGGAGCA-3´, the L segment primers were 5´-GCACAACAATCTTTCTGCGAT-3´ and 5´-CAGGGCTTTGTTTTGTCCAG-3´. Phusion Flash High-Fidelity PCR Master Mix (Thermo Fisher Scientific) was used for PCR amplification. The reaction mix consisted of: 1 μl of cDNA, 10 μl of Master Mix, 1 μl of forward and reverse primer (10 μM stocks), and 7 μl of molecular grade water. The cycling conditions were: 1) 10 s at 98°C; 2) 1 s at 98°C; 3) 5 s at 60°C; 4) 7 s at 72°C; 5) 1 min at 72°C; steps 2 to 4 were repeated 35 times. The PCR products were separated by agarose gel electrophoresis, purified with the QIAquick gel extraction kit (Qiagen) following the manufacturer’s instructions, and cloned into plasmid using Zero Blunt TOPO PCR Cloning Kit (Thermo Fisher Scientific) following the manufacturer’s recommendations. Plasmid minipreps were purified using the GeneJET Plasmid Miniprep Kit (Thermo Fisher Scientific) and the purified plasmids were sent for sequencing (with M13 forward and reverse primers) to Microsynth (Zurich, Switzerland).
To study the secondary structures formed by the genome ends of LCMV (strain Armstrong 53b, S segment GenBank accession NC_004294, L segment NC_004291), GGV (S segment NC_018483, L segment NC_018482), and HISV-1 (S segment KR870017, L segment KR870031) we used DuplexFold Web Server of RNA structure at RNAstructure (Web Servers for RNA Secondary Structure Prediction, available at https://rna.urmc.rochester.edu/RNAstructureWeb/Servers/DuplexFold/DuplexFold.html) [37, 38]. We applied standard parametrization for the predictions, the folding free energies for the models chosen for presentation are: LCMV S segment -30.9 kcal/mol; LCMV L segment, -41.4 kcal/mol; GGV S segment -33.4 kcal/mol; GGV L segment -35.6 kcal/mol; HISV S segment -34.3 kcal/mol; and HISV L segment, -29.9 kcal/mol.
NGS and de novo assembly was done as described [12, 39].
The sequences were aligned with Clustal Omega algorithm [40] implemented in EMBL-EPI webserver [41]. The phylogenetic trees were constructed using Bayesian Monte Carlo Markov Chain (MCMC) method implemented in BEAST version 2.4.7 [42] using LG or HKY-G-I substitution models for amino acid and nucleotide sequences, respectively. The analyses were run for 50 million states and sampled every 5000 steps. They were carried out on the CSC server (IT Center for Science Ltd., Espoo, Finland). Posterior probabilities were calculated with a burn-in of 10% and checked for convergence using the Tracer version 1.6.
Full-length NP (amino acids 1–582), and N- (aa 1–295) and C-terminal (aa 296–582) parts of it were PCR cloned for E. coli expression using primers (forward for full length and N-terminal portion 5´-CACCATGTCCTTGAACAAGGACCTT-3´; reverse for N-terminal portion 5´- TCTGTCGCTGGTGCAACC-3´; forward for C-terminal portion 5´- CACCATGATCTCATCTCAAAACATACC-3´; reverse for full length and C-terminal portion 5´- GTTGTTCATTATGTAGTTGAA-3´) designed according to the Champion pET101 Directional TOPO Expression Kit with BL21 Star (DE3) One Shot Chemically Competent E. coli manual (Thermo Fisher Scientific). Protein production and purification was done as described [39]. Full length HISV NP could not be recovered by this strategy, but the N- and C-terminal portions were recovered in moderate and good amount, respectively. The purified C-terminal portion of HISV NP was dialyzed against PBS. A rabbit polyclonal antiserum against C-terminal portion HISV NP (anti-HISV NP-C) was produced by Biogenes GmbH (Berlin, Germany).
IF staining was done on cells grown on 24-well Glass Bottom Plates (In Vitro Scientific) as described [39]. The primary antibodies, anti-UHV NP-C or anti-HISV NP-C antisera, were used at 1:2,000 dilution in PBS.
Routine protocols, described in [43], were utilized for SDS-PAGE and western blotting, the results were recorded with the Odyssey Infrared Scanning System (LI-COR).
A Taqman qRT-PCR assay for quantifying the S and L segments of UHV-2 and HISV-1 served to monitor the growth of UHV-2 and HISV-1 in cell culture. The primers and probes were: UHV-2 S segment forward primer (FWD) 5´-GCAAAACAGAACTGCTGCAGTC-3´, reverse primer (REV) 5´-TGCGATACAGACATAATTAGAGACATTG-3´, and probe 5´-6-Fam(carboxyfluorescein)-GTCACCATGTGTCCCTCAGAACTCATTCA-3´-BHQ-1 (Black Hole Quencher); UHV-2 L segment FWD 5´-TTGGGGAGTTTGTTACCAATGT-3´, REV 5´- GTGGGCCCAAATAACAAACCT-3´, and probe 5´-6-Fam- CTCTCTCGGACCTCCCACTTGTTCCTTTATG-3´-BHQ-1; UGV-1 S segment forward primer (FWD) 5´- CAAGAAAAACCACACTGCACA-3´, reverse primer (REV) 5´- AACCTGTTGTGTTCAGTAGT-3´, and probe 5´-6-Fam(carboxyfluorescein)- CTCGACAAGCGTGGGCGGAGG-3´-BHQ-1 (Black Hole Quencher); UGV-1 L segment FWD 5´- TCATAAAAGCTTCTAGCTATTCTTTTCAT-3´, REV 5´- CAAGTTGGAGGCCCAAGAG-3´, and probe 5´-6-Fam- TGAAGTCTCCTCCAAGACCCTGGTTATCAG-3´-BHQ-1; HISV-1 S segment FWD 5´-CTCAAAATCTTACCGAAGTTGTATGTAC-3´, REV 5´-CACTTTCCCTTTTGGATCTTTG-3´, and probe 5´-6-Fam-GTGACGACCAAGTGTCGGGTCACAC-3´-BHQ-1; HISV-1 L segment FWD 5´-GAGTCTTTGTTTGATAATGGTGGTT-3´, REV 5´-ATTGAAGACTACAGAACCATATC-3´, and probe 5´-6-Fam-TCATTTGATTCAAGTGTTCTCAGGGCA-3´-BHQ-1 (Metabion International Ag). RNA isolation for Taqman assays was done with the GeneJET RNA purification kit (Thermo scientific) with carrier RNA following the manufacturer’s protocol. Taqman Fast Virus 1-step master mix (Thermo scientific) was used for qRT-PCR, 10 μl reactions were run with the AriaMX real-time PCR System (Agilent) in duplicate with recommended cycling conditions: 1) 50°C 5 min; 2) 95°C 20 s; 3) 95°C 3 s; 4) 60°C 30 s (steps 3 and 4 were repeated 39 times).
Pellets from HISV-1 or UGV-1 infected I/1Ki cells were fixed in 1.5% glutaraldehyde, buffered in 0.2 M cacodylic acid buffer, pH 7.3 for 12 h at 5°C and embedded in epoxy resin. Toluidin blue stained semithin (1.5 μm) sections and, subsequently, ultrathin (100 nm) sections were prepared and the latter contrasted with lead citrate and uranyl acetate and examined with a Philips CM10 transmission electron microscope at 80kV.
For immuno-EM, cell pellets were fixed in 2.5% glutaraldehyde in 0.5xPBS and epoxy resin embedded. Thin sections (100 nm) were prepared and incubated for 30 min at RT in PBS with 1% BSA, followed by overnight incubation with rabbit anti-HISV NP antiserum (diluted 1:1,000 in PBS with 1% BSA) at 4°C. After washing with PBS, sections were incubated with 18 nm gold-conjugated goat anti-rabbit IgG antibody (Milan Analytica AG, Rheinfelden, Switzerland; diluted 1:20 in PBS with 1% BSA) for 2 h at RT. Sections were then contrasted and examined as descrived above.
Immunohistology with anti-HISV NP antiserum at 1:6,000 dilution was performed on formalin-fixed, paraffin embedded (FFPE) tissue sections, following previously published protocols [4, 12, 39]. Refer to Table 1 for animals examined. Consecutive sections incubated with a non-reactive rabbit polyclonal antibody instead of the specific primary antibody served as negative controls. A further section of each block was stained for reptarenavirus NP as described [4]. A FFPE pellet prepared from each an HISV- and reptarenavirus-infected cell sample served as positive control for the immunohistological stains. All snakes were also examined for any histopathological changes, using hematoxylin-eosin stained consecutive sections.
The virus sequences obtained in this study are made available via GenBank, the accession numbers are provided in S1 Table.
|
10.1371/journal.pgen.1001173 | DSIF and RNA Polymerase II CTD Phosphorylation Coordinate the Recruitment of Rpd3S to Actively Transcribed Genes | Histone deacetylase Rpd3 is part of two distinct complexes: the large (Rpd3L) and small (Rpd3S) complexes. While Rpd3L targets specific promoters for gene repression, Rpd3S is recruited to ORFs to deacetylate histones in the wake of RNA polymerase II, to prevent cryptic initiation within genes. Methylation of histone H3 at lysine 36 by the Set2 methyltransferase is thought to mediate the recruitment of Rpd3S. Here, we confirm by ChIP–Chip that Rpd3S binds active ORFs. Surprisingly, however, Rpd3S is not recruited to all active genes, and its recruitment is Set2-independent. However, Rpd3S complexes recruited in the absence of H3K36 methylation appear to be inactive. Finally, we present evidence implicating the yeast DSIF complex (Spt4/5) and RNA polymerase II phosphorylation by Kin28 and Ctk1 in the recruitment of Rpd3S to active genes. Taken together, our data support a model where Set2-dependent histone H3 methylation is required for the activation of Rpd3S following its recruitment to the RNA polymerase II C-terminal domain.
| Acetylation of histone N-terminal tails occurs on nucleosomes as a gene is being transcribed, therefore helping the RNA polymerase II traveling through nucleosomes. Histone acetylation, however, has to be reversed in the wake of the polymerase in order to prevent transcription from initiating at the wrong place. Rpd3S is a histone deacetylase complex recruited to transcribed genes to fulfill this function. The Rpd3S complex contains a chromodomain that is thought to be responsible for the association of Rpd3S with genes since it interacts with methylated histones, a feature found on transcribed genes. Here, we show that the recruitment of Rpd3S to transcribed genes does not require histone methylation. We found that Rpd3S is actually recruited by a mechanism implicating the phosphorylation of the RNA polymerase II C-terminal domain and that this mechanism is regulated by a transcriptional elongation complex called DSIF. We propose that the interaction between the Rpd3S chromodomain and methylated histones helps anchoring the deacetylase to its substrate only after it has been recruited to the elongating RNA polymerase.
| Histone acetylation was the first covalent histone modification shown to be involved in transcription regulation. Indeed, histones at the promoter of active genes tend to be hyper-acetylated while repressed genes have promoters with hypo-acetylated nucleosomes. It is now well established that this is due to the recruitment of histone acetyltransferases (HATs) and histone deacetylases (HDACs) by transcriptional activators and repressors, respectively [1]. The first described and best characterized HDAC is Rpd3. The repressive effect of yeast Rpd3 on transcription has been well studied over the last 15 years, paving the way for the characterization of its mammalian orthologs [2]. Yeast Rpd3 is recruited to the promoters of specific genes by DNA-binding repressors, leading to the repression of many important pathways such as stress response, meiosis, the cell cycle and others [3]–[19]. Moreover, Rpd3 also plays roles in silencing [20]–[27], DNA replication [28]–[31] and recombination [32], [33].
Recent proteomic studies have determined that Rpd3 is found in two distinct complexes: the large (Rpd3L) and the small (Rpd3S) complex [34], [35]. Both complexes share a core composed of Rpd3, Sin3 and Ume1. The large complex is composed of 11 additional proteins whereas the small complex contains only two additional subunits, namely Rco1 and Eaf3. While Rpd3L is likely responsible for the repressive function of Rpd3, the function of Rpd3S remains far less understood. Recent work by several groups has shown that Rpd3S is involved in the suppression of cryptic transcription [34], [36], [37] and that its activity is linked to the Set2 histone methyltransferase (HMT). Furthermore, in vitro studies have shown that Rpd3S is recruited to H3K36-methylated nucleosomes and that its Rco1 and Eaf3 subunits are essential for this recruitment [34], [35], [37], [38]. Rco1 mediates interactions with histones in a modification-independent manner through a PHD zinc finger domain, while Eaf3 contains a methyl-lysine binding chromodomain (CHD) that is essential for recognition of H3K36-trimethylated (H3K36me3) nucleosomes in vitro. These studies also indicate that genome-wide histone acetylation levels on promoters and coding regions are altered when either H3K36 methylation or Rpd3S is disrupted.
Taken together, these data lead to a model where Rpd3S is recruited to coding regions through the interaction of Eaf3 with H3K36me3 in order to deacetylate nucleosomes after the passage of the transcriptional machinery [39]. Deacetylation would allow chromatin disrupted by elongating RNA polymerase II (RNAPII) to return to a more ordered and compact structure, thereby restoring an environment hostile to cryptic transcription initiation within the coding region. However, this model relies heavily on the in vitro observation that the Eaf3 CHD binds preferentially to H3K36me3 peptides or nucleosomes [34], [35], [37], [38]. It was never formally demonstrated that this interaction is required for the targeting of Rpd3S to coding regions in vivo. In addition, whether Rpd3S is recruited to all transcribed genes or to only a subset of them has never been assessed. In order to address these questions, we performed genome-wide ChIP-chip experiments looking at both Rpd3S- and Rpd3L-specific subunits in wild type cells and in various mutants, including set2Δ and H3K36A (Table S1 and S2).
Quite interestingly, our data show that Rpd3S specifically binds to the coding region of actively transcribed genes whose promoters are also bound by Rpd3L. Surprisingly, the binding of Rpd3S to active genes is not dependant on Set2-mediated H3K36 methylation. However, methylation by Set2 is required for the activity of Rpd3S, as assayed by histone acetylation and RNAPII levels. We also provide in vivo evidence that Rpd3S is recruited to active genes via the phosphorylation of the RNAPII C-terminal domain (CTD). Finally, we show that the yeast DSIF transcription elongation factor negatively regulates Rpd3S recruitment. Based on these results, we propose that the recruitment and activity of Rpd3S on ORFs depend on a two steps mechanism: an initial recruitment to the elongation complex -coordinated by DSIF and RNAPII phosphorylation-, followed by the H3K36me3-dependant modulation of Rpd3S activity through the Eaf3 chromodomain.
While it is generally accepted that Rpd3L negatively regulates specific sets of genes, the specific function of Rpd3S remains largely unaddressed. However, it is expected to be ubiquitously recruited to active genes since it interacts with methylated H3K36. In order to address the specificity of Rpd3S and Rpd3L in a systematic manner, we performed ChIP-chip experiments on myc-tagged Rpd3, Rco1 and Sds3, the last two being specific subunits of Rpd3S and Rpd3L, respectively. As shown in Figure 1A, Rpd3 binds to a large subset of promoters (Figure 1A, clusters 1 and 2). In addition, a subset of these genes also exhibit Rpd3 binding on their coding regions (Figure 1A, cluster 2). Figure 1B and 1C show the average signal of Rpd3, Rco1 and Sds3 over the genes from cluster 1 and cluster 2, respectively. The data for cluster 3, representing the genes not bound by Rpd3, are also shown. Cluster 1, which is enriched for genes previously demonstrated to be repressed by Rpd3 (genes involved in M phase (p-value 10−8), cell cycle (p-value 10−9), etc.), shows binding of both Rpd3 and Sds3 to the promoter. The presence of Rco1 is not detectable on these genes. This cluster therefore represents genes repressed by Rpd3L, which is consistent with the fact that these genes have low level of RNAPII (Figure S1A). Cluster 2, however, shows evidence for the presence of both Rpd3L and Rpd3S since all three subunits tested for are detected. Rco1 is restricted to the coding region of these genes, consistent with the fact that they are actively transcribed (Figure S1A). Sds3 is present at the promoter, which is expected since Rpd3 binds to these promoters. More surprisingly, however, some level of Sds3 is also detected on the coding region of these genes. These data - also observed with another subunit of Rpd3L (Rxt2; Figure S1B) - suggest that the large complex may play some role during transcriptional elongation, perhaps in conjunction with the small complex (see Discussion). Nevertheless, the data presented here clearly show that Rpd3S binds to the coding region of active genes.
Strikingly, these experiments also show that Rpd3S preferentially associates with genes that are also bound by Rpd3L. In fact, we found no clusters of genes where Rpd3 binds in the coding region but not in the promoter. This suggests that Rpd3S, contrary to what has been expected, does not ubiquitously bind to active genes but rather targets some of them, namely a subset of those that are bound by Rpd3L at their promoter. In order to test if Rpd3S ubiquitously binds active genes, we performed ChIP-chip experiments of RNAPII and H3K36me3, two proxies for active gene expression. Figure 1D (cluster 4) clearly shows that many genes, despite having strong enrichment for RNAPII and H3K36me3, show no evidence of Rpd3S binding. Three conclusions can be drawn from these results; firstly, they confirm that the recruitment of Rpd3S is not a general phenomenon occurring on all transcribed genes; secondly, they suggest that the methylation of H3K36 by Set2 is not providing the specificity for the recruitment of Rpd3S; and finally they suggest that the large complex may play a role in the recruitment of the small complex as Rpd3S appears to target primarily genes also bound by Rpd3L.
The idea that H3K36me3 recruits Rpd3S through the Eaf3 subunit is well established in the literature [39]. Work done in vitro by several groups has clearly shown this using peptides and nucleosomal substrates [34], [35], [37], [38]. Our ChIP-chip experiments, however, clearly demonstrate that many genes harboring high levels of H3K36me3 are free of Rpd3S. We have therefore endeavored to examine whether the well characterized interaction between the Eaf3 CHD and H3K36me3 is responsible for the targeting of Rpd3S in vivo.
First, we looked at Rco1 binding in a set2Δ mutant. Since this mutant cannot methylate H3K36, the current model predicts that Rpd3S should not bind to ORFs under these conditions. To our considerable surprise, Rco1 enrichment on ORFs in this mutant is not significantly altered for about two-thirds of Rpd3S target genes (Figure 2A, cluster 5). For other genes, Rpd3S occupancy is decreased significantly, although not completely abolished (Figure 2A, cluster 6). We next repeated these experiments in a H3K36A mutant (where lysine 36 is mutated into an alanine) with similar results (Figure 2A). The deletion of the Rco1 PHD domain had no effect on Rpd3S occupancy (despite the fact that it destabilizes the Rco1 protein; Figure 2E), while deletion of the Eaf3 CHD domain phenocopied set2Δ and H3K36A. Interestingly, a set1Δ/set2Δ/dot1Δ triple mutant (abolishing all histone methylation activity in yeast) was similar to wild type, suggesting that deletion of SET1 and/or DOT1 can partially suppress the set2Δ phenotype. Taken together, these data demonstrate that H3K36 methylation is not required for the recruitment of Rpd3S to most genes.
Since Rpd3S occupancy seems to be more dependent on H3 methylation at some genes than others, we looked more closely at clusters 5 and 6. As shown in Figure 2B and 2C, clusters 5 and 6 are markedly different with regards to transcription levels. While cluster 5 is highly transcribed (as shown by the presence of high levels of both RNAPII and H3K36me3), cluster 6 is less so. Next we looked at the distribution of Rco1 on genes contained within these clusters in all strains shown in Figure 2A. As expected from data shown in Figure 2A, Rco1 occupancy on ORFs is not (or only slightly) affected in these mutants for the cluster 5 genes, but it is reduced for the genes from cluster 6 (Figure 2D and Figure S2). In addition, for all Rpd3S-bound genes, a redistribution of Rco1 to the promoter region was observed in all mutants tested. This redistribution towards the promoter remains unexplained but correlates with our observation that histone acetylation is decreased at promoters in these same mutants (Figure S3). Collectively, these data clearly demonstrate that H3K36 methylation has no impact on Rpd3S occupancy at highly transcribed genes (genes from cluster 5). However, the methyl mark, or the ability to recognize it through the Eaf3 chromodomain, is important for optimal Rpd3S association to ORFs with lower levels of RNAPII (genes from cluster 6).
It is known that set2Δ mutants, along with null mutants of Rpd3S subunits Eaf3 or Rco1, exhibit a cryptic transcription phenotype [34], [36], [37]. This phenotype is thought to be due to the improper deacetylation of transcribed ORFs by Rpd3S after each round of transcription [39] because of a lack of Rpd3S recruitment. Other groups that have characterized acetylation levels on coding regions in Set2 and Rpd3S mutants either looked at bulk chromatin by western blotting [37], or at specific genes by ChIP [34], [35], [38], [40] and have come to the conclusion that acetylation levels increase on ORFs when Set2 or Rpd3S is disrupted.
Since we—quite surprisingly—observed, however, that Rpd3S binding to genes is mostly independent of histone methylation by Set2, we decided to test whether the activity of Rpd3S requires methylation of H3K36 by Set2. To do so, we looked at H4K5 acetylation (H4K5ac) by ChIP-chip. We used H4K5 acetylation to score for Rpd3S activity because it was shown previously to be a robust read out for Rpd3 activity in ChIP-chip assays [41]. Similar to other groups [34], [35], [38], [40], we observed decreased acetylation on promoters in set2Δ, H3K36A or Rpd3S mutants (Figure S3). Histone acetylation is also dramatically affected across ORFs in these mutants. As shown in Figure 3A, we observed a loss of acetylation for normally highly acetylated ORFs, and a gain in acetylation for ORFs that exhibit low levels in the wild type. Because Set2 and Rpd3S are both known to prevent cryptic initiation within ORFs, we repeated the same analyses on RNAPII ChIP-chip results, and found a similar pattern to that observed for H4K5ac, namely that ORFs with high RNAPII enrichment show decreased RNAPII levels in the absence of Set2 or Rpd3S, whereas ORFs with low RNAPII tend to display higher levels of polymerase (Figure 3B). These results clearly show that the activity of the Rpd3S complex requires methylation of H3K36 by Set2.
The effect of the loss of Rpd3S activity on histone acetylation and RNAPII distribution on ORFs is more complex than previously described. Our data indeed suggest that both histone acetylation and RNAPII occupancy are redistributed in a more even manner across the genome than expected. A plausible explanation of the genome-wide averaging of RNAPII and histone acetylation levels in Set2 and Rpd3S mutants would entail aberrant recruitment of the transcriptional apparatus to low-expression genes whose coding regions were not reset properly by a, now inactive, Rpd3S complex. Assuming a limited pool of transcription machinery in a cell, this would result in a lower abundance of RNAPII at the more active ORFs, and aberrant genomic acetylation levels.
These results, combined with the Rpd3S occupancy profiles in rco1Δ, rco1-PHDΔ and eaf3-CHDΔ mutants, suggest that the loss of histone H3 methylation at lysine 36 affects ORF identity through a modulation of Rpd3S deacetylase activity rather than through altered recruitment of the Rpd3S complex on coding regions as previously thought.
Since the interaction between the Eaf3 CHD and H3K36me3 does not account for the initial recruitment of Rpd3S to active genes, we decided to look for a factor that would fulfill that role. Quan and Hartzog have shown genetic interactions between H3K36 methylation and Rpd3S with Spt5 [42]. Their data suggest that Rpd3S opposes the function of the elongation factor Spt4/5, which is the yeast ortholog of the human elongation factor DSIF [43]. DSIF negatively regulates elongation in its non-phosphorylated form, but is turned into a positive elongation factor upon phosphorylation by P-TEFb (Bur1 in yeast) [44]–[46]. The genetic interaction between Rpd3S and Spt5 led us to test whether Spt4/Spt5 is involved in the recruitment of Rpd3S. We therefore performed ChIP-chip experiments of Rco1 in spt4Δ cells (the deletion of SPT5 is lethal). As shown in Figure 4A, deletion of SPT4 leads to massive changes in Rco1 binding across the genome. Notably, the effect is far more dramatic compared to the deletion of SET2 (Figure 4A). Importantly, the level of RNAPII observed on these genes is not significantly affected in the mutant, ruling out the possibility that the effect is solely due to a reshuffling of the transcriptome (Pearson correlation = 0.94, Figure S4A). The deletion of SPT4 causes a decrease of Rco1 binding at some transcribed genes normally strongly associated with Rco1 (Figure 4A, cluster 7), as well as an increase at others where Rco1 is otherwise only found at low levels (cluster 8). Deletion of SPT4 even causes a slight increase of Rco1 occupancy at genes where it is normally undetectable (cluster 9). Overall, these effects lead to a distribution of Rpd3S that correlates better with RNAPII occupancy than in wild type cells (Figure 4B). To test the possibility that, in the absence of Spt4, Rpd3S is recruited via H3K36 methylation, we profiled Rco1 binding in a spt4Δ/set2Δ double mutant. As shown in Figure 4A, deleting both SPT4 and SET2 leads to a similar Rpd3S localization phenotype compared to the single spt4Δ mutant, giving further evidence that Set2 does not play a large role in Rpd3S recruitment, even in the absence of Spt4.
To distinguish the direct effect of the loss of SPT4 from eventual indirect effects of the mutation, we localized Spt4 in wild type cells by ChIP-chip using a strain carrying a myc-tagged SPT4 gene. Spt4 associates with genes in a manner that correlates with levels of RNAPII (Pearson correlation = 0.84) suggesting that DSIF acts as a general elongation factor. Moreover, it is present across the whole ORF, indicating that it travels with RNAPII, but is further enriched in the 3′ end of genes (Figure S4B), suggesting that it may also regulate the elongation-termination transition, as shown by others [47]–[49]. This is also consistent with the fact that Spt5 interacts with components of the capping and termination machineries [50]. Even more interestingly, the more a gene is occupied by Spt4 in wild type cells, the more Rco1 we detect in the spt4Δ mutant (Figure 4C), suggesting that the direct effect of the loss of Spt4 is an increase in Rpd3S binding (as observed for cluster 8). Consequently, the decrease in occupancy observed in cluster 7 is most likely indirect since Spt4 is barely detectable at these genes (Figure 4A). Similarly, the level of Rco1 increases dramatically in the spt4Δ mutant on cluster 4 genes (from Figure 1D), representing transcribed genes highly occupied by Spt4 where Rco1 is absent in wild type cells (Figure S4C). In general, genes with a higher Spt4/RNAPII ratio tend to have less Rco1 than genes with lower Spt4/RNAPII ratios (Pearson correlation = −0.42, Figure S4D). Taken together, these data suggest that Spt4 acts as a negative regulator of Rpd3S binding and that its presence prevents the HDAC from freely associating with transcribed genes. This model is in agreement with genetic data showing that Rco1 opposes the function of Spt4/5 [42].
We then looked at the distribution of Rpd3S on transcribed genes where Spt4 is also bound (cluster 8) and found strong differences between the wild type and spt4Δ mutants. While Rco1 occupies the whole ORF at a constant level in wild type cells, it accumulates to abnormally high levels towards the 3′ end of the gene in spt4Δ cells (Figure 4D, dashed line). This binding pattern is also found in the spt4Δ/set2Δ double mutant (compare dashed and dotted lines in Figure 4D). The occupancy profile of Rco1 in a spt4Δ mutant shows a strong similarity to the occupancy profile of RNAPII with a CTD phosphorylated at Ser2 (Figure 4D, red solid line). This led us to hypothesize that CTD phosphorylation by Ctk1 (the major serine 2 kinase) might be implicated in Rpd3S recruitment in the absence of Spt4/5. To test this hypothesis, we profiled Rco1 occupancy in a spt4Δ/ctk1Δ background. Surprisingly, ctk1Δ partially suppressed the spt4Δ Rpd3S binding pattern phenotype (Figure 4A) leading to an intermediate binding profile between wild type and spt4Δ. Looking at it more closely, we observed that short genes fail to accumulate Rco1 in spt4Δ/ctk1Δ cells (Figure 4E), whereas Rco1 is observed at genes irrespective of their length in wild type (Figure S5) or spt4Δ cells (Figure 4F), suggesting that Rpd3S accumulates more slowly in spt4Δ/ctk1Δ cells. We therefore conclude that in the absence of Spt4, Rpd3S binds to the RNAPII CTD and that the phosphorylation of the CTD at serine 2 contributes to that phenomenon.
The data presented above demonstrate that Spt4 negatively regulates the association of Rpd3S to highly transcribed genes. The data also suggest that phosphorylation of the RNAPII CTD, notably at serine 2, plays some role in the recruitment of Rpd3S to active genes in the absence of Spt4. We next tested whether the phosphorylation of the CTD is implicated in the association of Rpd3S with transcribed genes in wild type cells and performed ChIP-chip experiments of Rco1 in mutants for the known CTD kinases. As shown in Figure 5A, deletion of CTK1, the major serine 2 kinase, has a clear effect on Rco1 occupancy (compare solid line with dashed line). To test the effect of serine 5 and 7 phosphorylation, we used a strain carrying ATP-analog-sensitive alleles of KIN28 since the deletion of the KIN28 gene is lethal. As shown in Figure 5A (dotted line), inhibition of Kin28 has a dramatic effect on Rco1 occupancy. Indeed, no Rco1 can be detected on ORFs in that mutant. This clearly demonstrates that phosphorylation of serine 5 and/or 7 by Kin28 is a major element in the recruitment of Rpd3S to active genes. These results are supported by data from the Hinnebusch lab who have shown that Rpd3S interacts with the phosphorylated form of RNAPII and has high affinity for doubly phosphorylated (serine 2/5) CTD peptides in vitro [51]. Interestingly, CTD peptides carrying a single phosphate group on serine 5 or serine 2 respectively have a much weaker or no affinity to Rpd3S compared to doubly phosphorylated CTD peptides. Also noteworthy is the fact that Rpd3S is redistributed to promoter regions when Kin28 is inactive (Figure 5A). As we will be discuss below, this might be due to a decrease in H3K36 methylation in that mutant, most likely due to a defective Bur1/2 recruitment, as suggested by [52] and [53], leading to a defect in Rad6 phosphorylation. Our genome-wide data, together with these in vitro experiments, suggest that phosphorylation of the RNAPII CTD stimulates the recruitment of Rpd3S to transcribed genes while the elongation factor DSIF counteracts this recruitment.
In mammalian cells, DSIF is phosphorylated by Cdk9, a cyclin-dependent kinase associated with the elongation factor P-TEFb [54]. Cdk9 also phosphorylates the RNAPII CTD on serine 2 as well as other proteins including NELF. In yeast, the function of Cdk9 is fulfilled by two distinct kinases. Ctk1 mainly phosphorylates the RNAPII CTD and Bur1 targets Spt5 and Rad6 [55]–[57]. Inactivation of Bur1 has modest effects on phosphorylation of the RNAPII CTD, as shown by western blot and by ChIP-chip experiments (data not shown; see also [52], [57], [58]). Because Bur1 phosphorylates Spt5, the partner of Spt4, we tested the effect of inactivating Bur1 activity on Rco1 recruitment. Not surprisingly, inhibiting Bur1 using an ATP-analog-sensitive strategy (bur1AS) has a profound effect on Rco1 occupancy. In the absence of a functional Bur1, Rco1 is depleted from the coding region and redistributed to promoter regions (Figure 5B, dashed line) as was observed in the Kin28 mutant while the RNAPII level is mostly unchanged (data not shown). Deleting the CTD of Spt5 (spt5ΔC), the region phosphorylated by Bur1, caused a similar, although milder, phenotype (Figure 5B, dotted line). The stronger effect observed in the Bur1 inactivation experiment, compared to truncation of Spt5, is most likely due to the fact that Bur1 has additional targets. For example, Bur1 phosphorylates Rad6, an event that is required for the methylation of H3K4 and K79 by Set1 and Dot1 respectively [59]–[61]. In a triple mutant for Set1, Set2 and Dot1, we observed a shift of Rpd3S toward intergenic regions (Figure S2B), suggesting that the redistribution of Rpd3S to promoter regions in Bur1-impaired cells is at least in part a consequence of Bur1's activity on Rad6. Nevertheless, on the coding region, where our spt4Δ data showed that DSIF negatively regulates Rpd3S binding, we observed a clear decrease in Rco1 occupancy in both bur1AS and spt5ΔC. These data strongly suggest that phosphorylation of Spt5 by Bur1 negatively regulates the activity of DSIF on Rpd3S recruitment. While we cannot completely rule out the possibility that some of the effect observed in the bur1AS strain is due to an effect of Bur1 on the RNAPII CTD, our data on spt5ΔC rather suggest that Bur1 regulates Rpd3S recruitment by regulating DSIF. Interestingly, phosphorylation of Spt5 by Bur1 was previously shown to stimulate its activity as an elongation factor while our data suggest that it inhibits its activity as a negative regulator of Rpd3S recruitment. It will therefore be interesting to see whether these activities are linked.
The Rpd3S complex is recruited to the coding region of transcribed genes where it represses cryptic transcription by deacetylating histones in the wake of the elongating polymerase, therefore resetting chromatin to its pre-transcriptional state. The recruitment of Rpd3S to transcribed regions was thought to be mediated by the interaction between the Eaf3 CHD and H3K36me. However, the interaction of Rpd3S on the coding regions of actively transcribed genes has never been directly demonstrated in vivo. Here, we build on that model and show that: 1) Rpd3S targets a subset of transcribed genes; 2) Rpd3L is present at the promoter of genes where Rpd3S is bound; 3) The activity of Rpd3S requires methylation of H3K36 by Set2 but its association with active genes does not; 4) The recruitment of Rpd3S on ORFs requires the phosphorylation of the RNAPII CTD by Kin28 and Ctk1; and 5) The DSIF elongation complex counteracts the recruitment of Rpd3S to transcribed genes, a phenomenon that is regulated by the phosphorylation of Spt5 by Bur1. We therefore propose a model where the opposing effects of CTD phosphorylation and DSIF on Rpd3S recruitment together allow for the complex occupancy profile of that HDAC in vivo (Figure 6).
Because H3K36 methylation correlates with transcription, it is generally accepted to be present at all transcribed genes; Rpd3S was therefore expected to follow the same pattern. Our finding that many transcribed genes do not show any signs of Rpd3S binding despite the presence of H3K36me3 was therefore a considerable surprise. This has important implications since it suggests that not all genes are equally protected against cryptic initiation. Other mechanisms exist that prevent cryptic transcription [62], so it will be interesting to investigate how these various activities share the labor of protecting the genome from aberrant transcription.
Another peculiarity of Rpd3S is that it is preferentially enriched on the coding regions of genes where Rpd3L is found on the promoter: we rarely find Rpd3S on genes where Rpd3L is not present. The presence of Rpd3L on the promoter of active genes is counterintuitive given its role as a co-repressor but it has nevertheless being observed before [17]. The co-occurrence of Rpd3L at promoter and Rpd3S on the ORFs of the same genes suggests that Rpd3L may play a role in targeting Rpd3S. One possibility would be that Rpd3S emerges from Rpd3L during the transition from initiation to elongation. The transition to elongation may trigger the exchange of subunits to transform Rpd3L into Rpd3S. Our finding that DSIF and RNAPII CTD phosphorylation are involved in the proper recruitment of Rpd3S to chromatin suggests that they may play a role in that process.
Unexpectedly, we also observe Rpd3L on the coding regions of the genes where Rpd3S is also present. This last result was surprising in that it does not agree with the generally accepted function of Rpd3L as a promoter-recruited co-repressor. These data argue for a new model where subunits of both Rpd3L and Rpd3S coexist on actively transcribed coding regions. According to the model described above, the presence of Rpd3L subunits may reflect an imperfect exchange of subunits during the transition from initiation to elongation. Alternatively, it may reflect the presence of an Rpd3 “super large” complex that contains subunits of both Rpd3S and Rpd3L. In such a model, Rpd3S (as we normally see it) would only exist in solution and would represent a module that joins Rpd3L after initiation. This hypothesis is supported by data provided by Collins et al. [63] who combined and re-analyzed previously published mass spectrometry data [64], [65] to generate a high-accuracy yeast protein interaction dataset. They found that several subunits of Rpd3L can be co-purified using Rco1 as bait. Reciprocally, subunits of Rpd3S can be co-purified with Rpd3L subunits used as baits. These data argue that some interactions between Rpd3S and Rpd3L do exist in the cell. Since the complexes analyzed in these studies were purified from the soluble cellular fraction, the relative low abundance of these inter-complex interactions may be explained by the fact that these complexes normally exist together only on chromatin. The recent development of techniques to purify protein complexes from chromatin [66], [67] may help testing these intriguing possibilities.
We show that the recruitment of Rpd3S to active genes does not require H3K36 methylation, Set2 or the Eaf3 chromodomain. Interestingly, Rco1 binding is not abolished in a triple deletion mutant for all the known yeast HMTs (Set1, Set2 and Dot1), ruling out the possibility that the previously described interaction between H3K4me3 and Eaf3 [38] (or an eventual interaction with methylated H3K79) is compensating for the loss of H3K36 methylation in set2Δ and H3K36A mutants. Moreover, although the methylation state of H3K36 does not affect Rpd3S binding at highly transcribed genes, it has an effect on the occupancy of Rpd3S at ORFs showing lower transcription levels. This suggest that, despite not providing the main recruitment signal, H3K36 methylation provides some stabilizing effect on the association of Rpd3S with chromatin. While Rpd3S recruitment is largely independent on H3K36 methylation, the activity of Rpd3S, however, depends on the integrity of the Set2/Rpd3S pathway, namely Set2-dependant H3K36 methylation, the Rco1 PHD and the Eaf3 CHD. We therefore propose that the main function of H3K36 methylation is to “activate” Rpd3S after it has been recruited by virtue of its association with the RNAPII CTD. This may involve the anchoring of Rpd3S on chromatin via the Eaf3-H3K36me interaction.
Our data clearly show that phosphorylation by Kin28 is absolutely required for the association of Rpd3S with the coding region of transcribed genes. This suggests that phosphorylation of serine 5 provides the signal for the association of Rpd3S with early elongating RNA polymerases II molecules. Since Kin28 was recently shown to phosphorylate serine 7 in addition to serine 5, we cannot rule out the possibility that serine 7 phosphorylation also contributes to the binding of Rpd3S to RNAPII. Using a ctk1Δ strain, we also show that phosphorylation of serine 2 also contributes -although to a lesser extend than serine 5- to the association of Rpd3S with transcribed genes. These results are in perfect agreement with a recent paper from the Hinnebusch group who showed that RNAPII CTD peptides harboring a phosphate group on both serines 5 and 2 have a high affinity for Rpd3S in vitro while a single phosphate on serine 5 has a reduced affinity [51]. More importantly, they were not able to detect any interaction of Rpd3S with non-phosphorylated CTD peptides, and peptides carrying a single phosphate on serine 2 barely show any affinity with Rpd3S [51]. We therefore propose that Rpd3S is recruited to active genes via interaction with the phosphorylated RNAPII CTD. However, this complex remains inactive until it has been “anchored” on chromatin via H3K36 methylation. As discussed above, this phenomenon does not appear to occur equally on all genes. Negative regulation of these interactions by the DSIF elongation complex modulates association of Rpd3S with genes, therefore creating situations where active genes with high level of Spt4 carry much less Rpd3S than expected from their transcriptional level.
Interestingly, we found that deletion of SPT4, a subunit of DSIF, has a profound effect on Rpd3S occupancy in vivo. In the absence of Spt4, Rpd3S associates with the genome in a manner that correlates better with transcription than in wild type cells. This suggests that DSIF is involved in regulating the amount of Rpd3S on transcribed genes. DSIF appears to prevent the association of Rpd3S to a subset of transcribed genes, and even at genes occupied by Rpd3S, DSIF also plays a role since it prevents the hyper-accumulation of the HDAC in the 3′end of the gene. How exactly an elongation factor can regulate the association of a HDAC with elongating RNAPII remains obscure but we envision several mechanisms by which it may operate: 1) Rpd3S may directly or indirectly interact with DSIF; 2) The association of DSIF with the elongation complex may prevent the association of the HDAC; 3) DSIF may modulate the speed of the elongation complex in a way that makes it less favorable for Rpd3S binding; 4) DSIF may also impinge on the phosphorylation of the RNAPII CTD. More complex mechanisms may also be envisioned. For instance, Pin1, a proline isomerase that may modify the RNAPII CTD, binds to both, phosphorylated DSIF [68] and Rpd3/Sin3 [69]. CTD isomerisation may therefore be involved in the regulation of the recruitment of Rpd3S.
In human and yeast, DSIF is regulated by phosphorylation of the CTD of its Spt5 subunit. In yeast, this phosphorylation is mediated by the Bur1 cyclin-dependent kinase [56], [57]. We therefore tested whether phosphorylation of Spt5 by Bur1 affects Rpd3S occupancy in vivo. As expected, both the catalytic inactivation of Bur1 and the removal of its substrate (by deleting the CTD of Spt5) have a dramatic impact on Rpd3S occupancy. Both these mutants indeed show a decreased in Rpd3S occupancy along genes. These data suggest that phosphorylation of Spt5 by Bur1 negatively regulate the activity of DSIF on Rpd3S recruitment. In addition, the inactivation of Bur1 also causes Rpd3S to redistribute to promoter regions, a phenomenon that we also observed in set2Δ cells. Since Bur1 phosphorylates Rad6, which is also required for H3K36 methylation by Set2, it appears likely that this redistribution in Bur1 mutants is due to its effect on H3K36 methylation via Rad6. Why a lack in H3K36 methylation leads to an association of Rpd3S with promoters remains unknown, but is in agreement with the previous observation that histone acetylation decreases at promoters in these mutants.
Taken together, our in-depth analysis of Rpd3S genomic occupancy has revealed several key insights about its recruitment to genes in vivo. Our study highlights a complex network of protein-protein interactions mediated by phosphorylation of several substrates by at least three kinases. Interestingly, there is previous evidence in the literature linking these kinases to the function of DSIF. First, P-TEFb (Cdk9) and Bur1 can phosphorylate DSIF [56], [57], [70]. Second, Kin28, Ctk1 and Bur1 exhibit synthetic genetic interactions with Spt4 and Spt5 [71]. And third, the recruitment of Bur1 is stimulated by Kin28 [52], [72]. Finally, both DSIF and Kin28 have been shown to stimulate the recruitment of the Paf1 complex to the elongation complex [57], [73]. A lot more work will be required before we completely understand the interplay between these factors and Rpd3S.
All strains used in this study are listed in Table S1. All strains were grown in 50mL of YPD to an OD600 of 0.6–0.8 before crosslinking, unless otherwise indicated. For ChIP-chip, most strains were crosslinked with 1% formaldehyde for 30 min at room temperature on a wheel. The Rco1-9myc strains were crosslinked with 1% paraformaldehyde for 30 min at room temperature followed by 90 min at 4°C on a wheel. ATP analog-sensitive strains were treated with 6 µM of NAPP1 for 15 minutes prior to crosslinking.
ChIP experiments were performed as per [74], with minor modifications. For myc-tag ChIP, we used 5µg of 9E11 antibody coupled to 2×107 pan-mouse IgG DynaBeads (Invitrogen) per sample. For histone H4K5 acetylation ChIP, we used 4µL of a rabbit serum (Upstate 07-327) coupled to 2×107 protein G DynaBeads per sample. Histone H3K36me3 was immunoprecipitated with 4µL of antibody (Abcam Ab9050) coupled to 2×107 protein G DynaBeads per sample. Histone H4 levels were assayed with 2µL of an antibody raised against recombinant yeast histone H4 (a gift from Alain Verreault) coupled to 2×107 protein G DynaBeads per sample. RNAPII ChIPs were done using 2µL of 8WG16 antibody coupled to 2×107 pan-mouse IgG DynaBeads per sample. Note that in our ChIP-chip assays, 8WG16 generates profiles that are nearly identical as using a tagged RNAPII (data not shown). 8WG16 is therefore used here to measure total RNAPII levels on genes. RNAPII CTD serine 2 phosphorylation was assayed using 5µL of H5 antibody (Covance Research MMS-129R-200) coupled to 2×107 protein G DynaBeads per sample.
The microarrays used for location analysis were purchased from Agilent Technologies (Palo Alto, California, United States) and contain a total of 44,290 Tm-adjusted 60-mer probes covering the entire genome for an average density of one probe every 275 bp (±100 bp) within the probed regions (catalog # G4486A and G4493A). Myc-tag ChIPs were hybridized against ChIPs from isogenic strains that did not contain the tag as controls. Acetylation, RNAPII and histone H4 ChIPs were hybridized against a sample derived from 400ng of input (non-immunoprecipitated) DNA. Acetylation levels were normalized to histone H4 levels by subtracting Log2 (Histone H4/input) from Log2 (H4K5 acetyl/input). All microarray experiments described in this work are listed in Table S2, the processed data are available in Datasets S1, S2, S3, S4, and the raw data have been deposited into the GEO database (Accession # GSE22636).
The data were normalized and biological replicates were combined using a weighted average method as described previously [74]. The log2 ratio of each spot of combined datasets was then converted to Z-score, similar to Hogan et al. [75], to circumvent the large differences in the immunoprecipitation efficiencies of the different factors. Visual inspection of the Z-scores was carried out on the UCSC Genome Browser (http://genome.ucsc.edu/). All data analyses described here were done using data from protein-coding genes longer than or equal to 500 bp. Median Z-score values for promoter and complete length of each annotation from SGD (version Feb. 02 2008) were calculated without interpolation (Dataset S5) and used in our clustering and Pearson correlation analyses. Promoters are defined as the shortest of either 250bp or half the intergenic region (half-IG) relative to the reference gene's 5′ boundary. Self-organizing map (SOM) clustering was done with the Cluster software [76] and visualized with Java Treeview [77]. Only genes with no missing value were used for clustering.
Gene mapping was performed as in Rufiange et al. [78] on selected groups of genes described in the text. Briefly, data were mapped onto the 5′ and 3′ boundaries in 50 bp windows for each half-gene and adjacent half-IG regions. A sliding window of 300 bp was then applied to the Z-scores to smooth the curve.
|
10.1371/journal.ppat.1006213 | A calmodulin-like protein suppresses RNA silencing and promotes geminivirus infection by degrading SGS3 via the autophagy pathway in Nicotiana benthamiana | A recently characterized calmodulin-like protein is an endogenous RNA silencing suppressor that suppresses sense-RNA induced post-transcriptional gene silencing (S-PTGS) and enhances virus infection, but the mechanism underlying calmodulin-like protein-mediated S-PTGS suppression is obscure. Here, we show that a calmodulin-like protein from Nicotiana benthamiana (NbCaM) interacts with Suppressor of Gene Silencing 3 (NbSGS3). Deletion analyses showed that domains essential for the interaction between NbSGS3 and NbCaM are also required for the subcellular localization of NbSGS3 and NbCaM suppressor activity. Overexpression of NbCaM reduced the number of NbSGS3-associated granules by degrading NbSGS3 protein accumulation in the cytoplasm. This NbCaM-mediated NbSGS3 degradation was sensitive to the autophagy inhibitors 3-methyladenine and E64d, and was compromised when key autophagy genes of the phosphatidylinositol 3-kinase (PI3K) complex were knocked down. Meanwhile, silencing of key autophagy genes within the PI3K complex inhibited geminivirus infection. Taken together these data suggest that NbCaM acts as a suppressor of RNA silencing by degrading NbSGS3 through the autophagy pathway.
| Post-transcriptional gene silencing (PTGS) is an elaborately regulated process for defense against virus infection in plants. To achieve effective infection, a betasatellite molecule associated with geminivirus induced high levels of an endogenous RNA silencing suppressor, calmodulin-like protein (CaM), to counter host defenses. However, although CaM is one of the first identified cellular suppressors of RNA silencing, the mechanism of PTGS suppression is still poorly understood. This study demonstrates that CaM interacts with and degrades Suppressor of Gene Silencing 3 (SGS3) in Nicotiana benthamiana. We found that domains essential for the interaction between NbSGS3 and NbCaM are also required for the subcellular localization of NbSGS3 and for NbCaM suppressor activity. Moreover, NbCaM mediated NbSGS3 protein degradation can be blocked using the autophagy inhibitors 3-methyladenine and E64d, and by knock-down of key autophagy-related genes within the phosphatidylinositol 3-kinase (PI3K) complex. Silencing of the PI3K complex also inhibited geminivirus infection, which is consistent with autophagy playing an important role in RNA silencing suppression pathway and geminivirus infection.
| Post transcriptional gene silencing (PTGS) is an important RNA interference (RNAi)-based defense mechanism against foreign nucleic acid invasion and is involved in silencing a wide range of endogenous genes in plants. PTGS is triggered by double-stranded RNAs (dsRNAs), which are cleaved into 21- to 24- nucleotide (nt) small interfering RNA (siRNA) duplexes by Dicer-like (DCL) endoribonucleases. Subsequently, the siRNAs are loaded into an RNA-induced silencing complex (RISC), which contains an RNaseH-like Argonaute (AGO) enzyme, and one strand of the siRNA duplex is used to guide AGO to cleave homologous RNAs for degradation [1, 2].
In plants, overexpressed transgene transcripts, viral RNAs or their cleavage products can serve as the substrates for RNA-dependent RNA polymerases (e.g. RDR6) for conversion of single-stranded RNAs (ssRNAs) to dsRNAs, which further produce secondary siRNA molecules through DCL cleavage [3]. Therefore, RDR6 plays a key role in the sense RNA-induced PTGS (S-PTGS) pathway, the synthesis of trans-acting small-interfering RNA (ta-siRNAs), and anti-viral silencing pathways [4–6]. Recently, many reports have shown that RDR6-deficient plants (e.g. Nicotiana benthamiana) are more susceptible to infection by some positive-sense single-stranded RNA viruses, viroids and DNA viruses [7–10], strongly supporting the role of RDR6 in the host antiviral response. In these processes, a plant-specific RNA binding protein, Suppressor of Gene Silencing 3 (SGS3), functions together with RDR6 as a chaperone protein [4–6]. Arabidopsis SGS3 (AtSGS3) contains a zinc finger (ZF), rice gene X and SGS3 (XS), and coiled-coil (CC) domain. Among these, the XS and CC domains are involved in RNA binding and homodimer formation, respectively, and both are required for normal AtSGS3 localization and function in the synthesis of ta-siRNAs in plants [11–13]. Previous studies have suggested that AtSGS3 binds and stabilizes RNA templates during initiation of Arabidopsis RDR6 (AtRDR6)-mediated dsRNA synthesis [14], and AtSGS3 and AtRDR6 co-localize in certain cytoplasmic granules called SGS3/RDR6-bodies [13]. However, whether SGS3 from N. benthamiana plays a similar chaperone role with NbRDR6 is still obscure.
PTGS is an elaborately regulated process targeted against viral infection. However, most plant viruses have evolved viral suppressors of RNA silencing (VSRs) to counteract host antiviral silencing activity. Various VSRs have been identified in almost all plant virus genera, but they exhibit no obvious sequence similarities and interact with RNA-silencing pathways in multiple ways [15]. Recent reports show that several components of the Arabidopsis cytoplasmic exoribonuclease complex participated in mRNA quality control and mRNA processing, including FIERY1, XRN2, XRN3, XRN4, EIN5 and SKI2, which can also function as repressors of PTGS [16–18]. Moreover, impairing nonsense-mediated decay, deadenylation or exosome activity enhances S-PTGS in Arabidopsis, which requires the host RDR6 and SGS3 proteins for conversion of ssRNAs into dsRNAs to trigger PTGS [19]. Those endogenous RNA suppressors derived from mRNA decay pathways competed for SGS3/RDR6 RNA substrates to repress RNA silencing, suggesting the crucial role of SGS3/RDR6 in the endogenous RNA silencing pathway.
A calmodulin-like protein from Nicotiana tabacum (NtCaM) has been identified as an endogenous RNA silencing suppressor which interacts with the helper component-proteinase (HC-Pro) of a potyvirus [20]. However, follow-up work showed that NtCaM interacts with and directs degradation of several dsRNA binding VSRs likely through the autophagy-like protein degradation pathway, revealing a contradictory function for NtCaM in antiviral defense [21]. Nevertheless, a growing body of evidence published recently by different laboratories supports a role for the calmodulin-like protein as an S-PTGS suppressor [10, 22–24]. In the case of geminivirus infections, calmodulin-like protein from N. benthamiana (NbCaM) was up-regulated by Tomato yellow leaf curl China betasatellite (TYLCCNB)-encoded βC1 upon virus infection. Up-regulation of NbCaM by βC1 suppressed RNA silencing by repressing expression of RDR6 to promote viral infection [10]. Moreover, overexpression of Arabidopsis calmodulin-like protein 39 (AtCaM39) leads to increased susceptibility to infection by Tomato golden mosaic virus (TGMV) [22]. These studies indicate that calmodulin-like proteins are hijacked by plant viruses (at least geminiviruses, if not all) to counterattack the host defense response. However, the precise mechanism of calmodulin-like protein-mediated S-PTGS suppression is yet to be understood.
Autophagy is thought to be a nonspecific, bulk degradation process by which eukaryotic cells recycle intracellular components, such as protein aggregates and organelles [25]. There are at least three types of autophagy: macroautophagy, microautophagy and chaperone-mediated autophagy [26]. Macroautophagy (hereafter referred to as autophagy) is the major type of autophagy, and it occurs when cytoplasmic constituents are engulfed by double-membrane vesicles termed autophagosome and subsequently delivered to the vacuoles for breakdown and turnover in plants [27]. Autophagy is evolutionarily conserved from yeast to plants, and most of the essential components have been identified and characterized in plants through comparison to their homologs in yeast [26, 28–30]. Among these autophagy-related genes (ATGs), Beclin1 forms a complex with PI3K/VPS34, the class III phosphatidylinositol 3-kinase, as a first step in the initiation of autophagy, recruits other proteins to the complex and is required for autophagosome formation [26, 31].
Autophagy has also been shown to be important for anti-viral defense. In Drosophila, ATGs protect against Vesicular stomatitis virus (VSV) infection, and disruption of ATG5, ATG8, and ATG18 is associated with increased VSV RNA replication resulting in increased animal lethality [32]. Autophagy has also been reported to participate in antiviral defense in mammalian systems. For example, ATG5 is essential to protect mice against lethal infection of the mouse central nervous system by Sindbis virus [33]. Not surprisingly, viruses have developed strategies to subvert or use autophagy for their own benefit. For example, autophagy proteins are proviral factors that favor initiation of Hepatitis C virus infection and are required for translation of incoming viral RNA [34, 35]. In plants, deficiency in ATGs compromises plant vitality and disease resistance [29, 36, 37]. For example, N-gene mediated resistance against Tobacco mosaic virus (TMV) is dependent on autophagy genes, and plants deficient in the autophagy genes, Beclin1, PI3K/VPS34, ATG3, and ATG7, exhibit an unrestricted hypersensitive response (HR) in response to pathogen infection [29]. However, Arabidopsis mutant atg2-2 and several other ATG mutants, including atg5, atg7 and atg10, exhibit enhanced resistance to powdery mildew and dramatic mildew-induced cell death [38], providing new insights into the role of autophagy in disease resistance and cell death. A recent report showed that autophagy is possibly involved in the RNA silencing suppressor activity of P0, as P0-mediated degradation of AGO1 can be blocked by autophagy inhibitors [39]. Therefore, there is a question of whether autophagy is involved in the suppressor activity of NbCaM or in geminivirus infection.
In this study, we show that NbCaM interacts with SGS3 from N. benthamiana (NbSGS3), but not with NbRDR6. Furthermore, we found that NbCaM induces degradation of NbSGS3 by interacting with ATG factors, and silencing of ATG genes inhibits NbCaM-mediated NbSGS3 degradation and promotes resistance to infection by the geminivirus Tomato yellow leaf curl China virus (TYLCCNV) and its betasatellite (TYLCCNB). Together with previous results, these findings suggest that the endogenous RNA silencing suppressor NbCaM regulates RNA silencing and promotes geminivirus infection by repressing NbRDR6 expression and promoting degradation of NbSGS3, most likely via the autophagy pathway.
NbCaM suppresses sense RNA-induced PTGS and enhances geminivirus infection in N. benthamiana, similar to results observed when expression of AtSGS3 and AtRDR6 in Arabidopsis is reduced [4–6]. To explore the molecular mechanism of NbCaM-mediated suppression of RNA silencing and augmentation of geminivirus infection, potential interactions among NbCaM, NbSGS3 and NbRDR6 were analyzed initially using yeast two-hybrid (Y2H) assays. An expressed sequence tag (EST) for the NbSGS3 sequence was identified by aligning tobacco NtSGS3 and tomato SlSGS3 sequences (obtained from the GenBank database). Primers were designed to amplify the full-length coding sequence and the full-length gene encoding NbSGS3 was cloned from N. benthamiana. Sequence analysis revealed that the NbSGS3 open reading frame (ORF) contains 1908 nucleotides (nt) (GenBank accession number: KJ190939). Co-transformants of NbCaM cloned as a fusion with the GAL4 activation domain (AD-NbCaM) and NbSGS3 as a fusion with the GAL4 DNA binding domain (BD-NbSGS3) were plated on different selective media to detect activation of the reporter genes, HIS3 and ADE2. Yeast transformants containing AD-NbCaM and BD-NbSGS3 were able to grow on SD/Leu-Trp-His selection plates with 5 mM 3-amino-1,2,4-triazole (3-AT), whereas yeast transformants carrying AD-NbCaM with empty vector (AD-NbCaM + BD) or BD-NbSGS3 with empty vector (BD-NbSGS3 + BD) were unable to proliferate (Fig 1A). Furthermore, yeast transformants containing AD-NbCaM and BD-NbRDR6 showed no interaction between the proteins tested (Fig 1B). In addition, yeast transformants containing AD-NbRDR6 and BD-NbSGS3 or AD-NbSGS3 and BD-NbSGS3 also grew on the selection plates (S1A and S1B Fig), consistent with an interaction between SGS3 and RDR6 and self-interaction of SGS3, which was observed for AtSGS3 and AtRDR6 [12, 13]. Expression of NbCaM, NbSGS3 and NbRDR6 proteins was verified by Western blot (S1C–S1F Fig).
The interaction between NbCaM, NbSGS3 and NbRDR6 was further investigated by bimolecular fluorescence complementation (BiFC) in leaves from transgenic N. benthamiana plants which expressed H2B-RFP as a nuclear marker. In this assay, NbCaM, NbSGS3 and NbRDR6 were fused to the N (2YN) and C-terminal (2YC) fragments of yellow fluorescent protein (YFP), generating the constructs 2YN-NbCaM, 2YC-NbCaM, 2YN-NbSGS3, 2YC-NbSGS3, 2YN-NbRDR6 and 2YC-NbRDR6. Pairwise expression of 2YN-NbCaM and 2YC-NbSGS3, 2YC-NbCaM and 2YN-NbSGS3, 2YN-NbRDR6 and 2YC-NbSGS3, and 2YC-NbRDR6 and 2YN-NbSGS3 by agroinfiltration resulted in a clear YFP fluorescence signal in the cytoplasm of agroinfiltrated cells at 48 hours post infiltration (hpi) (Fig 1C). While no YFP fluorescence was observed when 2YN-NbCaM and 2YC-NbRDR6 or 2YC-NbCaM and 2YN-NbRDR6 were co-expressed together (Fig 1D and S2 Fig). The movement protein P3N-PIPO from Turnip mosaic virus (TuMV) [40] was used as a negative control, we found there was no YFP fluorescence in the pairwise expression of 2YN-NbSGS3 and 2YC-P3N-PIPO or 2YC-NbSGS3 and 2YN-P3N-PIPO (Fig 1D and S2 Fig). These results demonstrate that NbCaM specifically interacts with NbSGS3 in both yeast and plant cells.
In Arabidopsis thaliana, efficient RNA silencing requires RDR6 and its double-stranded RNA (dsRNA)-binding partner, SGS3 to amplify secondary siRNAs which allow plants to mount an effective defense response against transgene-induced aberrant RNAs or virus infection [4–6]. To better understand the role of NbCaM in the S-PTGS pathway and the potential effect of NbCaM on the function of NbSGS3, the sequence and biological features of NbSGS3 were analyzed. NbSGS3 cDNA encodes a 635-amino acid (aa) protein, with a structure similar to AtSGS3 and contains a zinc finger domain (ZF), a rice gene X and SGS3 domain (XS), and two coiled-coil domains (2*CC) (Fig 2A). A phylogenetic tree was constructed to compare the evolutionary relationships among SGS3 orthologs in tobacco (NtSGS3), tomato (SlSGS3) and Arabidopsis (AtSGS3) (Fig 2B). NbSGS3 is clustered with NtSGS3 and SlSGS3 and is distant from AtSGS3, sharing 94%, 82% and 51% aa identity with NtSGS3, SlSGS3 and AtSGS3, respectively.
To determine the expression pattern of NbSGS3, reverse transcription real-time quantitative PCR (RT-qPCR) was performed using total RNA isolated from different N. benthamiana tissues as template. NbSGS3 expression levels were very similar among different tissues, with the exception that the expression level in flower was higher than that detected in root tissues (Fig 2C, p<0.01). To examine the subcellular localization of NbSGS3, a green (GFP) or red (RFP) fluorescent protein reporter was fused to its C-terminus (NbSGS3:GFP or NbSGS3:RFP) under control of the Cauliflower mosaic virus (CaMV) 35S promoter. GFP or NbSGS3:GFP was transiently expressed in leaves of transgenic H2B-RFP N. benthamiana plants, and GFP fluorescence was examined in agroinfiltrated transgenic leaves at 48 hpi by confocal microscopy. Fluorescence in plants expressing GFP alone was observed in both the cytoplasm and nucleus, whereas fluorescence from NbSGS3:GFP was localized to granular-like structures in the cytoplasm (Fig 2D). These granular-like structures were also observed in N. benthamiana protoplasts (S3A Fig). Similar granular-like structures were also observed when a YFP was fused to the N terminus of NbSGS3 (YFP:NbSGS3, S3B Fig). A previous study reported that AtSGS3 localized to cytoplasmic granules, termed SGS3/RDR6-bodies [13]. To check whether NbSGS3 also localizes to SGS3/RDR6-bodies, NbRDR6:GFP and NbSGS3:RFP were co-expressed in wild type (Wt) N. benthamiana. As shown in Fig 2D, NbRDR6 alone formed irregular granules along the edge of the cell (the third line), but NbRDR6:GFP co-localized with NbSGS3:RFP to the cytoplasmic granules (the fourth line). In addition, we also examined whether NbSGS3 granules are related to cellular organelles, including chloroplasts, mitochondria, golgi bodies or peroxisomes, but no co-localization was found (S3A Fig and S4 Fig).
To map the protein domains required for the interaction between NbCaM and NbSGS3, three deletion mutants for NbSGS3 and four deletion mutants for NbCaM were constructed (Fig 3A and 3B). Mutants NbCaM-dX, NbCaM-dEFI, NbCaM-dEFII and NbCaM-dEFIV lacking the N-terminal 50 aa of an unknown domain, or the first, second and fourth Ca2+ binding EF-hand domain, respectively, were cloned into the plant BiFC vector 2YN [41]. Mutants NbSGS3-dZF, NbSGS3-dXS and NbSGS3-d2*CC, which lack the zinc finger, rice gene X and SGS3 domain, and two coiled-coil domains, respectively, were generated and cloned into the plant BiFC vector 2YC [41]. BiFC assays were performed using 2YN- and 2YC-tagged mutant proteins in transgenic H2B-RFP N. benthamiana plants. NbCaM deletion mutants NbCaM-dEFI and NbCaM-dEFII were unable to interact with NbSGS3, and NbSGS3 deletion mutants NbSGS3-dZF and NbSGS3-d2*CC failed to interact with NbCaM (Fig 3C). These results suggest that the EFI and EFII domains of NbCaM and the ZF and CC domains of NbSGS3 are essential for the interaction between NbSGS3 and NbCaM.
Localization of SGS3 to SGS3/RDR6-bodies is one of its basic features [13]. To investigate whether the domains essential for the interaction between NbCaM and NbSGS3 are also important for localization of NbSGS3, three deletion mutants of NbSGS3 were fused with GFP at their C-termini. Typical localization patterns of the deletion mutants and wild type NbSGS3:GFP (NbSGS3-Wt:GFP) are shown in Fig 3D. Expression of these proteins was verified by Western blot analyses (Fig 3E). In contrast to the granule localization of NbSGS3-Wt:GFP and the NbSGS3-dXS:GFP mutant, the NbSGS3-dZF:GFP or NbSGS3-d2*CC:GFP mutant exhibited no obvious granule localization (Fig 3D). These results suggest that the ZF and the 2* CC domains are important for localization of NbSGS3 to the granules.
We next tested whether domains essential for the interaction between NbCaM and NbSGS3 are also required for the PTGS suppressor activity of NbCaM. N. benthamiana leaves were co-infiltrated with agrobacteria harboring either an empty vector (Vec), 35S:NbCaM-dX (dX), 35S:NbCaM-dEFI (dEFI), 35S:NbCaM-dEFII (dEFII), 35S:NbCaM-dEFIV (dEFIV) or 35S:NbCaM (wild type NbCaM, Wt), together with 35S:GFP plus 35S:FP to trigger PTGS. Weak GFP fluorescence was observed in tissues co-infiltrated with empty vector, 35S:NbCaM-dEFI or 35S:NbCaM-dEFII together with 35S:GFP +35S:FP at 5 dpi. In contrast, strong fluorescence was evident in infiltrated patches where 35S:GFP +35S:FP were co-expressed with 35S:NbCaM-dX, 35S:NbCaM-dEFIV or 35S:NbCaM (Fig 3F). GFP fluorescence in infiltrated leaf patches was confirmed by the presence of GFP mRNA and protein, and expression of Wt or mutant forms of NbCaM was validated by the presence of NbCaM mRNA (Fig 3G). These results demonstrate that domains within NbCaM that are essential for the interaction with NbSGS3 are also indispensable for NbCaM suppressor activity.
It has been reported that the correct localization of SGS3 is important for its biological function [13]. To determine whether NbCaM affects the localization pattern of NbSGS3, GFP and NbSGS3:GFP fusion proteins were expressed alone or co-expressed with either empty vector (Vec) or Myc:NbCaM in N. benthamiana leaves. The distribution of GFP was very similar between tissue expressing GFP alone, or when co-expressing empty vector or Myc:NbCaM (Fig 4A). The NbSGS3:GFP fusion protein formed granules in the cytoplasm when it was expressed alone or together with empty vector. However, when NbSGS3:GFP was co-expressed with Myc:NbCaM, the number of NbSGS3:GFP granules was greatly decreased as compared to plants expressing NbSGS3:GFP alone or in conjunction with empty vector (Fig 4A and 4B). To further determine whether the weak fluorescence of NbSGS3:GFP in plants co-expressing Myc:NbCaM was due to decreased NbSGS3:GFP levels, Western blot analysis was performed to determine the accumulation of NbSGS3:GFP. As expected, levels of NbSGS3 decreased ~2-fold when co-expressed with Myc:NbCaM as compared to empty vector (Fig 4C). To exclude the possible influence of the tag, Myc:NbSGS3 was also expressed alone, or co-expressed with GFP or NbCaM:GFP. Protein and mRNA levels of Myc:NbSGS3 remained constant when expressed alone or together with GFP. In samples overexpressing NbCaM:GFP, no obvious changes in Myc:NbSGS3 mRNA were observed, but the amount of NbSGS3 protein was largely reduced (Fig 4D). It is worth mentioning that the NbCaM protein level was also reduced ~2-fold when co-expressed with NbSGS3 as compared to NbCaM alone (Fig 4C and 4D). These results suggest that overexpression of NbCaM leads to reduced NbSGS3 protein accumulation, and that both may be targeted for degradation after they form a complex.
A recent study showed that NtCaM is sensitive to 3-methyladenine (3-MA) and E64d, inhibitors of the autophagy pathway, and NtCaM seems to mediate degradation of VSRs [21]. To determine whether NbCaM-mediated degradation of NbSGS3 protein occurs via autophagy or other protein degradation systems, the sensitivity of NbCaM or NbSGS3 to 3-MA and E64d, two chemical inhibitors of autophagy, was tested. N. benthamiana leaves were agroinfiltrated with GFP, NbCaM:GFP or NbSGS3:GFP followed by infiltration of DMSO (control), 3-MA (10 mM) or E64d (100 uM) after 32 hours. Samples were collected from leaves after an additional 16 h incubation. No obvious changes in GFP, NbCaM:GFP and NbSGS3:GFP protein levels were observed in DMSO, 3-MA or E64d treated samples (S5 Fig). These results suggest that inhibition of autophagy did not affect accumulation of NbCaM or NbSGS3 when expressed alone. Similarly, the 26S proteasome inhibitor MG132 did not have an observable impact on the accumulation of NbCaM or NbSGS3 (S6 Fig). We next co-expressed Myc:NbSGS3 with GFP or Myc:NbSGS3 with NbCaM:GFP in plants treated with DMSO or 3-MA using different concentrations of Agrobacterium tumefaciens cultures carrying GFP or NbCaM:GFP and assessed protein levels by Western blot. Levels of GFP and Myc:NbSGS3 protein did not appear to change when co-expressed or when expressed in the presence of DMSO or 3-MA (Fig 5A). However, both Myc:NbSGS3 and NbCaM:GFP protein accumulated to higher levels in 3-MA treated plants as compared to DMSO treated plants (Fig 5B). The E64d had a similar role with 3-MA on the accumulation of NbSGS3:GFP when it was co-expressed with Myc:NbCaM (S7 Fig). These results suggest that NbCaM and NbSGS3 are most likely degraded by autophagy in plant cells. To confirm this assumption, we used YFP-tagged N. benthamiana ATG8a (YFP-ATG8a) as an autophagosome marker to monitor autophagy [36, 42–45]. In Wt or 35S:NbCaM transgenic N. benthamiana plants, we observed a low number of punctate YFP fluorescent structures (Fig 5C). However, when NbSGS3 was transiently over-expressed in 35S:NbCaM transgenic N. benthamiana plants via infiltration with TO:NbSGS3, there was a 3 to 4-fold increase in the punctate fluorescent structures, likely representing pre-autophagosome or autophagosome structures (Fig 5C and 5D). To check if NbSGS3 has any effects on YFP-ATG8a accumulation, YFP-ATG8a accumulation levels were compared by Western blot between co-expression of NbSGS3 with YFP-ATG8a or with empty vector and result showed that YFP-ATG8a accumulation level were similar between the two treatments (S8 Fig), indicating that NbSGS3 has no negative effect on YFP-ATG8a accumulation. To further assess induction of autophagy when NbCaM and NbSGS3 are co-expressed, transmission electron microscopy (TEM) was used to monitor autophagic activity. Co-expression of NbCaM and NbSGS3 induced a 4-fold greater number of double-membrane structures typical of autophagosomes in the cytoplasm, as compared to expression of NbSGS3 or NbCaM alone (Fig 5E and 5F). Taken together, these results indicate that NbCaM and NbSGS3 are likely degraded by autophagy after they form a complex.
To further understand the involvement of autophagy in NbCaM-mediated NbSGS3 degradation, the phosphatidylinositol 3-kinase (PI3K) complex containing Beclin1/VPS30/ATG6, PI3K/VPS34 and VPS15, which form phagophore to initiate autophagy [31], were analyzed. Predicted cDNA and protein sequences for Beclin1, PI3K, and VPS15 were identified in N. benthamiana (https://solgenomics.net/tools/blast/) through sequence similarity to homologs in A. thaliana and N. tabacum [29, 46, 47]. Partial-length cDNA sequences were isolated using N. benthamiana cDNA and cloned into a Tobacco rattle virus (TRV)-based VIGS vector. N. benthamiana seedlings were agroinfiltrated with recombinant TRV vectors carrying partial fragments of NbBeclin1, NbPI3K and NbVPS15, respectively to induce silencing of each gene. Silencing of these genes in N. benthamiana plants did not result in any distinct developmental defects in systemic leaves (S9A Fig), when compared to TRV-GUS-infected plants (negative controls). At 14 dpi, in plants infiltrated with ATG-silencing vectors, mRNA levels of each ATG were reduced by approximately 80% as compared to negative controls (infiltrated with TRV-GUS or mock) (S9B Fig). Newly formed upper leaves in ATG-silenced, TRV-GUS-treated or mock plants (no TRV infection) were infiltrated with NbSGS3:GFP and empty vector (Vec) or NbSGS3:GFP and Myc:NbCaM at 21 dpi. Two days later, the newly infiltrated leaf tissue was examined by confocal microscopy to compare the fluorescence strength of NbSGS3:GFP. Western blot analysis was performed to determine levels of NbSGS3:GFP. In mock or TRV-GUS-treated plants, overexpression of NbCaM decreased the fluorescence intensity of NbSGS3:GFP (S10 Fig) and reduced the accumulation of NbSGS3:GFP 3-fold (Fig 6A). Silencing either Beclin1, PI3K or VPS15 blocked degradation of NbSGS3 as determined by reduced levels of NbSGS3:GFP protein in infiltrated leaf patches (Fig 6A). These results demonstrate that the autophagy genes Beclin1, PI3K and VPS15, which constitute the PI3K complex, are required for NbCaM-mediated degradation of NbSGS3.
TYLCCNB-encoded βC1 up-regulates NbCaM to suppress RNA silencing and promote viral infection [10]. Given that NbSGS3 has an important role in defense against geminiviral infection and NbCaM-mediated NbSGS3 degradation appears to be dependent on the autophagy pathway, we next examined whether ATGs were also involved in geminivirus infection. Plants silenced for NbBeclin1, NbPI3K or NbVPS15 at 7 dpi were inoculated with equal amounts of TYLCCNV and its betasatellite (10Aβ) and symptoms induced by 10Aβ in mock, TRV-GUS-treated or ATGs-silenced plants were observed. Infection induced by 10Aβ in TRV-GUS-treated plants showed leaf curling symptoms similar to those observed in mock plants at 14 dpi. In contrast, NbBeclin1, NbPI3K or NbVPS15-silenced plants developed much milder symptoms with reduced leaf curling (Fig 6B). In agreement with these observations, Southern blot analysis of viral genomic DNA levels indicated almost undetectable amounts of viral DNA accumulation of both helper virus (10A) and betasatellite (10β) in NbBeclin1, NbPI3K or NbVPS15-silenced plants (Fig 6C). qPCR analysis of 10A and 10β DNA levels in the upper emerged infected leaves also showed a significant reduction in viral DNA levels in NbBeclin1, NbVPS15 or NbPI3K-silenced plants as compared to TRV-GUS-treated or mock plants (Fig 6D). These findings suggest that the PI3K complex is necessary for maximal symptom development and viral DNA accumulation, consistent with their role in the degradation of NbSGS3.
To determine whether these observations extend to geminivirus that lacks an associated betasatellite, mock, TRV-GUS-treated or ATGs-silenced plants were also inoculated with equal amounts of TYLCCNV alone (10A). No obvious viral symptom and viral DNA accumulation difference were observed in among mock, TRV-GUS-treated, and NbBeclin1, NbPI3K or NbVPS15-silenced plants at 14 dpi (Fig 6E and 6F). Similarly, the deficiency of NbBeclin1, NbPI3K or NbVPS15 also had no obvious effect on the infectivity and viral DNA accumulations of Tomato leaf curl Yunnan virus (TLCYnV) and Tobacco curly shoot virus (TbCSV) without betasatellite (S11 Fig). These data indicate that the proviral role of autophagy in geminivirus biology depends on the presence of betasatellite.
In plants, RNA silencing is a major defense mechanism against foreign genes or viral invasion [2, 3]. As a counter defensive strategy, plant viruses have evolved VSRs as potent molecular weapons to counteract antiviral RNA silencing by interacting with key components of the cellular RNA silencing pathway, such as binding long or short dsRNA duplex, interacting with or disrupting AGOs, DCLs, RDRs and their functional partners, or interfering with the assembly of RISC [1, 15]. The function of calmodulin-like protein is still in dispute and its suppression mechanism is not clear. To help clarify the mechanism by which calmodulin-like protein suppresses RNA silencing, we showed that NbCaM interacts with NbSGS3 in the Y2H and BiFC systems, but not with NbRDR6 (Fig 1).
AtSGS3 localizes to cytoplasmic granules (SGS3/RDR6-bodies), where RDR6-mediated dsRNA synthesis is thought to occur [13]. NbSGS3 also localizes to SGS3/RDR6-bodies, along with NbRDR6 (Fig 2D), indicating that SGS3/RDR6-bodies are likely conserved among different plant species. However, the domains that are necessary for localization of NbSGS3 and AtSGS3 appear to be different. We showed that the ZF and CC domains are required for NbSGS3:GFP localization, but the XS and CC domains are necessary for AtSGS3:GFP localization [13].
As the partner of RDR6, SGS3 can bind the 5’ overhang of dsRNAs and may prevent degradation of these dsRNAs, alter their localization, and/or recruit them as templates for dsRNA synthesis process [11, 14, 48]. Therefore, it is not surprising that SGS3 is targeted by several VSRs, including the V2 protein of Tomato yellow leaf curl virus (TYLCV) [49], p2 of Rice stripe virus (RSV) [50], the VPg protein of Potato virus A (PVA) [51], TGBp1 of Planta goasiatica mosaic virus (PlAMV) [52]. Our data showed that NbSGS3 is also a target of the endogenous RNA silencing suppressor, NbCaM. First, NbCaM interacted with NbSGS3 in yeast and in planta (Fig 1). Second, deletion mutants lacking the EFI and EFII domains lost the ability to interact with NbSGS3, and failed to suppress GFP-induced S-PTGS (Fig 3F and 3G). Finally, overexpression of NbCaM led to a reduced accumulation of NbSGS3 in granules and promoted its degradation (Fig 4). Our data also demonstrate that the interaction between NbCaM and NbSGS3 is required for the suppressor activity of NbCaM. Meanwhile, the ZF and CC domains of NbSGS3, which are necessary for localization to the SGS3/RDR6-bodies, are also required for the interaction with NbCaM. This suggests that these two domains play an important role in inducing RNA silencing.
Autophagy has been reported to play a central role in several physiological and developmental responses in plants, such as nutrient recycling, seed development and germination, nitrogen or carbon deprivation [44, 45, 53, 54], and plant immunity and programmed cell death [29, 55]. A recent study showed that NtCaM could mediate degradation of the dsRNA binding VSR 2b via the autophagy-like protein degradation pathway [21]. We found that overexpression of NbCaM induced degradation/reduction of NbSGS3 when co-expressed (Fig 4). Inhibition of autophagy (3-MA or E64d, S5 Fig) or the 26S proteasome (S6 Fig) did not have obvious effects on the accumulation of GFP, NbCaM:GFP or NbSGS3:GFP, when expressed alone. In contrast, 3-MA-treatment blocked degradation of both NbCaM and NbSGS3, when co-expressed and led to an increase in the level of their corresponding proteins (Fig 5A and 5B). This phenomenon suggests that individual expression of either NbCaM or NbSGS3 does not trigger autophagy-mediated degradation, but that interaction between NbCaM and NbSGS3 activates the autophagy system, possibly via recruitment of some ATG proteins. Indeed, degradation was blocked when NbBeclin1, NbPI3K or NbVPS15, which constitute the PI3K complex, was silenced. In addition, when NbBeclin1, NbPI3K or NbVPS15 was knocked-down, TYLCCNV and TYLCCNB (10Aβ) was unable to infect plants efficiently, showing very mild viral symptoms and almost undetectable viral DNA levels (Fig 6A–6D), suggesting a necessary role for the PI3K complex in viral replication and systemic infection. It is worthy to note that the knock-down of NbBeclin1, NbPI3K or NbVPS15 has no effect on the infection of this geminivirus that lacks an associated betasatellite (10A) (Fig 6E and 6F). These data indicate that the proviral role of autophagy in geminivirus biology depends on the presence of betasatellite, which is consistent with the fact that 10Aβ not 10A significantly up-regulates NbCaM expression and induces severe symptoms [10, 62]. Similarly, the other two geminiviruses in the absence of betasatellite showed no obvious differences in their infectivity in mock and ATGs-silenced plants (S11 Fig). Therefore, it seems that geminiviruses that lack an associated betasatellite fail to utilize autophagy factors to defend NbRDR6/NbSGS3-dependent resistance.
RDR6 and SGS3 play important roles in RNA silencing, and their expression can be effectively fine-tuned. For NbCaM to be an effective negative regulator of RNA silencing, it needs to repress both NbRDR6 and NbSGS3. In support of this, our previous study showed that NbCaM suppressed NbRDR6 mRNA levels [10]. Plant calmodulin-like proteins can directly bind to DNA and function as transcription factors (TF) to activate or suppress a target gene’s expression [56]. For example, an Arabidopsis calmodulin-binding transcription factor CAMTA3 functions as a suppressor of defense response and can activate gene expression by directly binding to promoters of suppressed genes or suppressing gene expression by activating expression of a repressor [57]. We found that NbCaM could activate expression of a reporter gene in yeast cells (S12 Fig), which indicates that NbCaM may also function as a TF. It is therefore possible that NbCaM suppresses NbRDR6 via binding to a promoter element and repressing its expression. However, further study is necessary to determine the mechanism of NbCaM suppression of NbRDR6 expression.
Together with previous studies, we conclude that the cellular suppressor NbCaM not only suppresses NbRDR6 transcription, but also interacts with the RNA silencing component NbSGS3 and mediates its degradation by recruiting autophagy factors. Geminivirus betasatellite appears to utilize NbCaM in suppression of plant antiviral defenses, which leads to successful viral infection and multiplication (Fig 7).
N. benthamiana seedlings were placed in soil and incubated in an insect-free growth chamber at 25°C and 60% relative humidity under a 16 h light/8 h dark photoperiod. The transgenic H2B-RFP line was gift of Michael M. Goodin (University of Kentucky, USA).
The tobacco NtSGS3 and tomato SlSGS3 sequences were used to identify orthologous sequences from available N. benthamiana ESTs. BLAST searches revealed high homology between SGS3 from tobacco and N. benthamiana (https://blast.ncbi.nlm.nih.gov/Blast.cgi?CMD=Web&PAGE_TYPE=BlastHome). Primers designed to anneal to conserved sequences in the 5' and 3' untranslated regions of tobacco SGS3 were used to amplify the coding region of N. benthamiana SGS3 by reverse transcription PCR (RT-PCR). Amplification with primer pairs NbSGS3-cds-F/NbSGS3-cds-R yielded a specific product of approximately 1900-bp, which was cloned into pMD18-T (TaKaRa, Dalian, China) and then sequenced. Detailed primer information is given in S1 Table. The full-length gene of NbSGS3 was deposited in GenBank under the accession number KJ190939. The full-length NbSGS3 was amplified using primer pair NbSGS3-F/NbSGS3-R and cloned into the binary vectors pCHF3-Flag, pCHF3-GFP or pCHF3-RFP between the BamHI and SalI sites. The resulting plasmids (pCHF3-35S-NbSGS3:Flag, pCHF3-35S-NbSGS3:GFP or pCHF3-35S-NbSGS3:RFP) were used for overexpression in transgenic plants or transient agroinfiltration assays using the CaMV 35S promoter. NbSGS3 was introduced into the 2YN or 2YC BiFC vectors between the PacI and AscI sites to generate 2YN-NbSGS3 or 2YC-NbSGS3 for BiFC analysis. Mutants of NbSGS3 were generated by overlapping PCR [58], using the corresponding primer pairs given in S1 Table and cloned into the 2YN, 2YC and pCHF3-GFP vectors.
To construct a TRV-based recombinant VIGS vector containing NbBeclin1, NbPI3K or NbVPS15, a partial fragment of each gene was generated by PCR amplification using the respective primer pair and cloned into the pTRV2 vector (a kind gift of Yule Liu) [59] using the restriction enzyme sites listed in S1 Table.
The coding sequence of NbCaM was amplified by PCR from N. benthamiana and introduced into the vectors pCHF3-Flag, pCHF3-GFP, 2YN or 2YC using the primers and restriction enzyme sites listed in S1 Table. pCHF3-based vectors were used for transient expression of NbCaM in N. benthamiana leaves. Construction of NbCaM mutants by overlapping PCR was similar to that described for NbSGS3 mutants [58], using the corresponding primers described in S1 Table. For the pCHF3-NbRDR6:GFP construct, the coding sequence of NbRDR6 was amplified by PCR from N. benthamiana and introduced into pCHF3-GFP between the SmaI and SalI sites using the corresponding primers described in S1 Table. For 2YN-NbRDR6 or 2YC-NbRDR6, the NbRDR6 coding sequence was introduced into the 2YN or 2YC BiFC vectors between the PacI and AscI sites. 2YN-P3N-PIPO and 2YC-P3N-PIPO have been described previously [40].
The pBA-Flag-Myc4:NbSGS3 (Myc4:NbSGS3), pEarleygate104:NbSGS3 (YFP:NbSGS3), pEarleygate101:NbSGS3 (NbSGS3:YFP), pBA-Flag-Myc4:NbCaM (Myc4:NbCaM) or pEarleygate104:NbATG8a (YFP:NbATG8a) construct was generated using gateway technology (Invitrogen, Burlington, Ontario, Canada) with the corresponding primer pairs given in S1 Table.
GenBank accession numbers for the genes analyzed in this study are as follows: NbSGS3 (KJ190939), NtSGS3 (NM_001325691), SlSGS3 (NM_001247782), AtSGS3 (NM_122263), NbRDR6 (AY722008), NbBeclin1 (AY701316), NbPI3K (AY701317), NbVPS15 (KU561371) and NbATG8a (KX120976).
Two-hybrid screen experiments to assess the different interactions between NbCaM, NbSGS3 and NbRDR6 in yeast were performed as described previously [60]. For BiFC and subcellular localization experiments fluorescence were examined in epidermal cells of 1–2 cm2 leaf explants by confocal microscopy (Leica TCS SP5, Mannheim, Germany) from 36 h to 72 h post infiltration as described [60].
For geminivirus agroinoculation, equal volumes of individual A. tumefaciens cultures at an OD600 of 1 were mixed prior to inoculations. Infectious virus clones, including TYLCCNV (pBinPLUS-Y10-1.7A), TYLCCNV/TYLCCNB (pBinPLUS-Y10-1.7A+Y10β), TbCSV (pBinPLUS-Y35-1.9A) and TLCYnV (pBinPLUS-Y194-1.4A) have been described previously [61–63]. Agrobacterium cultures carrying infectious virus clone(s) were infiltrated into N. benthamiana leaves and inoculated plants were photographed with a Canon 400D digital camera at different time periods.
For the TRV-VIGS assay, Agrobacterium cultures harboring pTRV1 and pTRV2-VIGS (TRV2-GUS, TRV2-NbBeclin1, TRV2-NbPI3K or TRV2-NbVPS15) were resuspended in infiltration buffer (10 mM MgCl2, 10 mM MES (pH5.6), and 100 μM acetosyringone) and mixed at a 1:1 ratio. After a 3 h incubation at room temperature, the mixed Agrobacterium cultures were infiltrated into leaves of N. benthamiana plants at the 5–6 leaf stage. A silenced phenotype appeared in the upper leaves at 2 weeks post infiltration.
Total DNA was extracted from infected plants using the CTAB method [64], and DNA blot hybridization performed to assess viral DNA accumulation essentially as described [65]. Total DNA electrophoresed through agarose gels was stained with ethidium bromide to ensure equal loading. After denaturation and neutralization, total DNA was transferred to Hybond N+ nylon membranes (GE Healthcare, Pittsburgh, PA, USA) by capillary transfer. Membranes were hybridized at 45°C to specific probes labeled with digoxigenin (Roche Diagnostics, Rotkreuz, Switzerland). Viral DNA levels were determined by qPCR using specific primers (S1 Table) and normalized to 25S RNA as an internal genomic DNA control [66].
Total RNA was isolated from virus-infected plants and different plant organs using Trizol reagent (Invitrogen, Carlsbad, CA, USA). For RT-qPCR analysis, 1 μg total RNA was firstly treated with DNase I, and then the first strand cDNA was synthesized from the treated RNA by using Oligo(dT)12-18 primer and SuperScript III reverse transcriptase (Invitrogen) following the recommended protocol. All primer information used in RT-PCR was given in S1 Table, and the specific primer pairs for qRT-PCR were designed by Primer Premier 5 software [10].
Total protein was extracted from infiltrated leaf patches as described previously [67]. Immunoblotting was performed with primary mouse monoclonal or rabbit polyclonal antibodies, followed by goat anti-mouse or anti-rabbit secondary antibody conjugated to horseradish peroxidase (Bio-Rad, Hercules, CA, USA). The GFP polyclonal antibody was obtained from Abcam (Massachusetts, US), and the Myc monoclonal antibody was obtained from Sigma (Los Angeles, CA, USA). Blotted membranes were washed thoroughly and visualized using chemiluminescence according to the manufacturer’s protocol (ECL; GE Healthcare).
PBS buffer containing 2% DMSO (control) or an equal volume of DMSO with 10 mM 3-MA and 100 uM E64d (Sigma) for inhibition of autophagy, or 100 μM MG132 (Sigma) for inhibition of the 26S proteasome was infiltrated into leaves 16 h before samples were collected. For TEM observation, detailed information has been described previously [46]. Vec and NbSGS3, or NbCaM, or NbSGS3 +Vec or NbSGS3 +NbCaM -infiltrated leaves pretreated with 10 mM 3-MA for 8 h, and then were cut into small pieces (1 mm × 4 mm). The sampled tissues were fixed in 2.5% glutaraldehyde and 1% osmium tetroxide (both in 100 mM phosphate buffer (PB), pH 7.0). The samples were then post-fixed in OsO4, dehydrated in ethanol, and then embedded in Epon 812 resin as instructed by the manufacture (SPI-EM, Division of Structure Probe, Inc., West Chester, USA). Ultrathin sections (70 nm) were cut with a diamond knife from the embedded tissues using the Ultracut E Ultramicrotome (Reichart-Jung, Vienna, Austria) and were collected on 3-mm copper (mesh) grids, and then stained with uranyl acetate and lead citrate before final examination under an electron microscope, Model JEM-1230.
|
10.1371/journal.pcbi.1004301 | Maintaining Homeostasis by Decision-Making | Living organisms need to maintain energetic homeostasis. For many species, this implies taking actions with delayed consequences. For example, humans may have to decide between foraging for high-calorie but hard-to-get, and low-calorie but easy-to-get food, under threat of starvation. Homeostatic principles prescribe decisions that maximize the probability of sustaining appropriate energy levels across the entire foraging trajectory. Here, predictions from biological principles contrast with predictions from economic decision-making models based on maximizing the utility of the endpoint outcome of a choice. To empirically arbitrate between the predictions of biological and economic models for individual human decision-making, we devised a virtual foraging task in which players chose repeatedly between two foraging environments, lost energy by the passage of time, and gained energy probabilistically according to the statistics of the environment they chose. Reaching zero energy was framed as starvation. We used the mathematics of random walks to derive endpoint outcome distributions of the choices. This also furnished equivalent lotteries, presented in a purely economic, casino-like frame, in which starvation corresponded to winning nothing. Bayesian model comparison showed that—in both the foraging and the casino frames—participants’ choices depended jointly on the probability of starvation and the expected endpoint value of the outcome, but could not be explained by economic models based on combinations of statistical moments or on rank-dependent utility. This implies that under precisely defined constraints biological principles are better suited to explain human decision-making than economic models based on endpoint utility maximization.
| Common decision-making models arise from firm axiomatic foundations but do not account for a variety of empirically observed choice patterns such as risk attitudes in the face of high-impact events. Here, we argue that one reason for this mismatch between theory and data lies in the neglect of basic biological principles such as metabolic homeostasis. We use Bayesian model comparison to show that models based on homeostatic considerations explain human decisions better than classic economic models—both in a novel virtual foraging task and in standard economic gambles. Specifically, we show that in line with the principle of homeostasis human choice minimizes the probability of reaching a lower bound. Our results highlight that predictions from biological principles provide simple, testable, and ecologically rational explanations for apparent biases in decision-making.
| Homeostasis is paramount to all living organisms [1]. Put simply, organisms have to maintain their internal milieu within certain boundaries to avoid dying. This homeostatic principle reverberates on the levels of molecular interactions [2], hormonal feedback loops [3,4], neural circuits [5], and psychophysiological processes [6]. Beyond the need for immediate regulation, many species face complex decisions with delayed and probabilistic consequences for long-term metabolic homeostasis. Here, we hypothesize that homeostatic requirements guide foraging decisions in humans. For example, hunting deer provides a large energy gain with low probability of obtaining it, while collecting berries provides a small energy gain with high probability. In order to minimize the probability of starvation, human agents should integrate the statistics of the available options with their current energy levels and with their time horizon.
Classical views of homeostasis [1–7] are often illustrated with a thermostat that senses the difference between a temperature set point and the current temperature. This deviation value elicits a change in heating levels. The thermostat is thus supposed to retrospectively compensate deviations that have already manifested themselves. In contrast, we propose that decision-makers can anticipate possible deviations and proactively minimize the probability of reaching a prohibitive boundary such as starvation. This extends established notions of homeostasis in (psycho)physiology [1–7] and is concordant with recent theoretical accounts of homeostasis as a principle explaining decision-making in healthy and psychiatric populations [8–10].
When applied to individual decision-making, predictions from this model are in contradistinction to economic models which firmly rest upon axiomatic foundations [11] and elegantly explain many types of monetary decisions [12]. These models posit that decision-makers base their choices on the utility assigned to the endpoint outcome of a choice irrespective of the trajectory to this endpoint [12–14].
In risk-return models and their variants [13,15], the endpoint utility is computed via statistical moments of an outcome distribution, usually expected value, variance, and in some models also skewness. A considerable literature has generalized risk-return models to include subjective transformations of statistical moments [15,16]. In behavioral economics, variants of risk-return models have been widely used to describe how humans choose between monetary gambles [17–19], how they assess real-life events [16], and how animals decide on primary reinforcers [20,21]. Expected utility theory and its derivations constitute another class of models in which values of possible outcomes are transformed into an internal utility measure by a decision-maker's individual utility function [12,15,22]. Rank-dependent utility models additionally supplement a non-linear weighting of the option’s outcome probabilities [23,24]. Similar to risk-return models, rank-dependent utility models have been used extensively to describe empirical data both from the lab and the field [15,23,24].
Empirically observed deviations from predictions of these microeconomic models are often framed as irrational biases, and additional parameters are included to absorb such biases, but often without making principled assumptions on why these influences should arise in the first place [25]. Here, we furnish a principled biological reason for deviations from economic principles. Critically, maximizing the utility of the endpoint outcome of a given set of options neglects the catastrophic consequences if the trajectory to this outcome reaches a lower bound of the internal milieu. We sought to show that even in a safe laboratory environment, a decision-making model based on homeostatic principles could explain foraging decisions better than economic models. Further, we hypothesized that homeostatic considerations would also guide human decision-making for simply structured lotteries without any reference to foraging, as often employed in behavioral economics [12–14,26].
To test these hypotheses, we developed a virtual foraging task. In each trial, human participants chose between two “foraging environments” with different possible “energy” gains and associated probabilities, in which they would forage for up to three consecutive “days” (see Fig 1A for illustration). At the time of choice, an energy bar depicted participants’ current internal state. Participants were instructed that on each foraging day, they would lose one energy point, and gain energy according to the statistics of the chosen environment. Successive days in the task required the integration of risks from multiple foraging attempts. For each trial, participants made a single decision between the two foraging environments for the indicated number of days. Losing all energy points on any day in a given foraging period was framed as “starvation” but was not explicitly punished. Each trial was independent. We did not give feedback on choice outcomes of their choices or intermediate states of the foraging sequence. At the end of the experiment, participants were rewarded for the endpoint foraging outcome of two randomly selected trials. Starvation meant that participants did not win anything from the trial. We hypothesized that participants would compute the probability of starvation for the foraging environments and base their decisions on this metric.
We used the mathematics of random walks to analytically derive the distributions of the endpoint outcomes of the foraging period, and of the probabilities of starvation during foraging (see Fig 1C for a graphical illustration, Fig 1D for an example of the variables derived by this procedure, Table 1 for a summary of the gambles, and Methods and S1 Text for mathematical details). Because participants were only rewarded for endpoint outcomes, homeostatic principles are irrelevant for maximizing utility, yet our task was suggestive of using them. Hence, we tested whether such principles also influence decisions when they are not invoked by the task frame. We presented participants with purely economic gambles, framed as wheel-spinning casino lotteries without any reference to foraging (see Fig 1B). These lotteries had identical endpoint outcomes as the foraging environments. The probability of starvation in the foraging frame corresponded to the probability of winning nothing in the casino frame. As in the foraging frame, participants did not receive feedback on the outcomes of their choices. In addition to the instruction, the two frames differed in the fact that options in the foraging frame were presented as gamble sequences when the number of days was greater than one, whereas options in the casino frame were always single-step gambles. To avoid foraging instructions influencing behavior in the casino frame, the casino frame preceded the foraging frame for all participants.
We first asked whether models based on homeostatic principles explain choice better than standard economic models—both in a foraging and in a casino frame. We combined choices from both frames and compared three families of formal decision-making models. The first two model families included variations of two types of economic models while the third family comprised models based on homeostatic considerations (see Methods and Table 2 for details).
In line with our hypothesis that participants’ decisions should take into account the probability of starvation (pstarve), the homeostatic model family provided the significantly best fit. Under the assumption that different participants may use different models (random-effects analysis), the exceedance probability that the homeostatic model family is the most frequent in the population was 0.9403 (Table 3). Under the assumption that all participants use the same model (fixed-effects analysis), the winning model belonged to the homeostatic family (see Fig 2A and Table 2 for log-group Bayes factors based on the Bayesian information criterion (BIC) relative to the simplest model; see S2 Text Section 7 and S1 Table and S2 Table for results based on the Akaike information criterion, AIC; see S3 Table for the fits of the different models for each individual participant). Thus, the overall comparison of model families confirmed our hypothesis that starvation probability and thus homeostatic principles provided explanatory power in explaining participants’ behavior, over and above economic variables, and although irrelevant for utility maximization in the laboratory.
Next, we separately analyzed choices in the foraging frame and in the purely monetary context of the casino frame, by comparing the three model families within each frame. The same overall pattern emerged. The homeostatic family, in which models included pstarve, had the highest exceedance probabilities in both frames independently (foraging: 0.9869; casino: 0.8675; Table 3). Also, the models winning in fixed-effects analyses belonged to the homeostatic family (see Table 2 for log-group Bayes factors based on BIC). Further, when we fitted the models separately for the first and second blocks of the foraging and casino frames, the same pattern emerged in all analyses. The homeostatic family had the highest exceedance probabilities and models belonging to this family won the fixed effects analyses (see Tables 2 and 3; see S2 Text Section 7, S1 Table and S2 Table for results based on AIC).
Within the winning model family, we analyzed which specific model best explained choices. In a random-effects analysis, the exceedance probability of the simplest homeostatic model was 0.9786 across both frames (Table 4). Similar results emerged when fitting the models separately within the two frames or separately to the first and second blocks (Table 4; see S2 Text Section 7 and S4 Table for results based on AIC; see S1 Fig for binned choice data). In this winning model, the decision variable was a linear combination of difference in starvation probability (pstarve) weighed by a parameter ξ, and difference in expected value (EV). This decision variable was transformed into a decision probability by a sigmoid function with another parameter, β.
The previous analyses showed that participants consistently used models based on homeostatic principles in both frames. This leaves open the question how participants used pstarve and whether this differed between the two frames. Hence, we added frame-specific free parameters to the winning model and then tested the parameter estimates across participants.
Specifically, we adapted the winning model (Model 7), which included two free parameters: a parameter β for the decision noise and a parameter ξ to quantify the impact of pstarve on participants’ decision. This parameter ξ was replaced by two frame-specific weighting parameters (ξforaging and ξcasino). We added this model (Model 10) to the initial set of three models in the third family. Despite being penalized for the additional free parameter, it explained choices better than the other three models considered. Its exceedance probability was 0.9983 in a random-effects analysis and it had the smallest log-group Bayes factor in a fixed-effects analysis (Table 5; see S2 Text Section 7 and S5 Table for results based on AIC). This indicates that participants weighted pstarve differently in the two frames.
Crucially, our prediction that participants’ choices minimize pstarve requires that weighting parameters of pstarve be negative. Indeed, the parameters for the frame-specific parameters ξforaging and ξcasino were significantly smaller than zero across participants (sign test on parameters in the overall winning model: ξforaging: p<.001; and ξcasino: p<.005; Fig 2B). That is, participants chose the gambles with the smaller pstarve and thus minimized pstarve. Additionally, across participants ξforaging was smaller than ξcasino (sign test comparing ξforaging and ξcasino: p<.05).
In line with the above analyses, supporting analyses showed that pstarve played a greater role than EV in the foraging frame, while in the casino frame EV played a greater role than pstarve, for explaining choices (see S2 Text Section 1 and S6 Table for details). Additionally, we devised a supplemental model to test whether the different number of foraging days led to a different weighting of pstarve but did not find evidence supporting this idea (S2 Text Section 2 and S7 Table). We also found no evidence supporting the hypothesis that different combinations of energy levels and foraging days lead to differential weighting of pstarve (S2 Text Section 4). Supplementary analysis showed that participants did not erroneously include values below zero in their estimation of the outcome distributions in the foraging frame (S2 Text Section 5 and S8 Table). In an exploratory analysis, we also found no evidence for a relationship of the model parameters to participants’ meta-cognitive risk assessments on the domain-specific risk-attitude scale (S2 Text Section 8).
Taken together, participants consistently minimized pstarve in both frames and did so more in the foraging than in the casino frame.
Can reaction times (RTs) as a tentative measure of choice difficulty give us additional evidence for the relevance of homeostatic principles? Since our model comparison showed that EV and pstarve explained participants’ choices, we tested whether EV and pstarve also related to RTs. That is, we tested whether RTs were faster for larger absolute differences between the two options in EV and pstarve. This was indeed the case as shown by a linear mixed effects model on log-transformed RTs (EV: t = -4. 98, p<.001; pstarve: t = -2.62, p<.05; significance levels were determined by log-likelihood tests, comparing the full model to a model without the respective factor). The interaction of EV and pstarve was significant and related to slower RTs (t = 3.52, p<.005). See S1 Fig for binned RT data. In sum, a combination of EV and pstarve was related to choice difficulty as indexed by RTs, which corroborates that homeostatic principles guided participants’ choices.
This study addressed whether homeostatic principles explain human decision-making over and above previously described economic models based on endpoint utility maximization. We found that human decisions minimized the probability of reaching a lower homeostatic bound on the trajectory to their endpoint outcomes, despite the fact that our tasks did not entail any explicit negative consequences of reaching this boundary. This was evident both in a virtual foraging task, in which the possibility of starvation was a salient task feature, and in a casino-like frame, in which only the endpoint outcomes of the gambles and their associated probabilities were explicitly stated. Our fine grained model comparison provided evidence that the decision variable in the most parsimonious model was based on a linear combination of the probability of starvation and endpoint expected value (EV), outperforming standard economic models.
The maximization of endpoint EV lies at the core of many variants of axiomatically derived microeconomic models. However, neither variants of risk-return models nor variants of expected utility theory predict that decisions minimize the probability of reaching a lower homeostatic bound before that endpoint is realized. The winning model family included the probability of zero outcomes although we did incentivize participants to avoid them, and although zero outcomes are already incorporated into the calculation of statistical moments and utilities. For a description of behavior, we could have used a very specific shape of the utility function which assigns a high negative utility to the zero outcome and positive utility to neighboring positive outcome, in contrast to typical utility functions in the economic literature. However, such a model would neither be more parsimonious than ours, nor offer any additional explanatory power.
We note that in the best fitting model, the decision variable was a linear mixture of outcome variables and thus it did not differ from previous risk-return models in its mathematical structure. Crucially, minimization of the probability of a zero outcome provided more explanatory power than risk-attitudes based on variance or skewness. Thus, our results are in line with previous accounts calling for more fine-grained and possibly context-dependent metrics within the framework of risk-return models [19,27]. Additionally, our model only makes meaningful predictions when the probability of threats to homeostasis is nonzero and thus our approach has the desirable feature that the scope of the model is under precisely defined constraints.
We provide evidence that the homeostatic principle of avoiding a lower boundary on energy levels pervades human decision-making. Classical descriptions often relate homeostatic processes to the actions of a thermostat [6,7]. The thermostat example best fits to physiological variables with a narrow homeostatic range, for which this range can be approximated by a set point (e.g., blood pH) [7]. For other variables the homeostatic range is larger. In the case of metabolic homeostasis, glycogen and fat buffers enlarge the homeostatic range and relevant homeostatic counter-measures occur at the boundaries of this range [7]. For simplicity, we assumed starvation to be a hard boundary but the same principle would apply for soft boundaries. More importantly, our results extend the notion exemplified by the thermostat analogy. In line with recent theoretical views on homeostasis in healthy and psychiatric populations [8,10], we conjectured that human decision-makers can estimate the probability of future disruptions to homeostasis. Thus, in contrast to a thermostat that can only react to homeostatic threats once they have occurred, human—and possibly many animal—decision-makers can proactively avoid threats to homeostasis.
The same model performed best in both the foraging frame and in the casino frame. We highlight this similarity between the two frames because it shows that the homeostatic principle of minimizing the probability of a zero outcome is at play even when participants are not primed by the task description to do so. Furthermore, the same model explains behavior in gamble sequences and in single-step gambles. In the casino frame, the probability of starvation was directly depicted by the size of the sector in the pie chart that indicated the probability for the zero outcome. Strikingly, in the foraging frame participants integrated the probabilities of gaining energy over the indicated number of days to compute the probability of starvation. Participants could not learn the outcome distributions through experience because we did not provide them with feedback. Thus, decisions in the foraging frame were not dependent on participants having directly experienced sequences in the virtual foraging environments. Risky decision-making differs depending on whether outcome distributions are described or learned from experience [22,28]. For example, rare events tend to exert less impact in decisions based on experience. Our results suggest that such an underweighting might not occur for the probability of starvation [28].
Within the winning model, more fine grained analyses revealed differences between the two frames in the best-fitting parameter estimates. The probability of starvation in the foraging frame had a greater impact on participants’ decisions than the corresponding probability of receiving nothing in the casino frame—an effect unrelated to the sequential versus single-step presentation of the gambles (see S2 Text Section 3). This was the case even though participants had to compute the probability of starvation in the foraging frame by combining information about internal state, foraging options, and time horizon. Approximating starvation probabilities may become more difficult and thus imprecise as the number of steps increases. In the current study, participants were able to approximate the probability of starvation with sufficient accuracy for at least three steps, as their decisions were based on this metric.
The structure of our tasks complies with the requirements of economic paradigms such as complete knowledge and incentive-compatibility [12]. Thus, specific task characteristics are unlikely to explain why our homeostatic model outperformed standard economic models, based on statistical moments [15,17,18] or non-linear probability weighting [23,24]. Instead, we reason that the biological constraints relevant in ecological contexts such as hunting or farming exert a prevailing impact on human decisions in the laboratory—even if apparently irrelevant to the task at hand. A similar rationale has recently been advocated in discussions of whether animal [29,30] and human [31–34] decision-making deviates from normative models. According to probabilistic accounts of brain function, the brain uses prior probabilities to perform probabilistic inferences [8,14,35,36]. These prior probabilities are tuned—by evolution and/or experience—to the natural statistics of real world environments [30,31]. Consequently, human and non-human decision-makers may behave rationally according to their beliefs but they appear irrational because those beliefs are not warranted in deliberately simplified laboratory tasks or in some other contexts [30,37]. Overall, this recent approach argues for complementing considerations about economic rationality (i.e., maximizing financial gain given currently available information) with considerations about ecological rationality (i.e., maximizing fitness given priors on environmental statistics). Its promise lies in unifying and explaining a diverse set of seemingly irrational behaviors while its challenge lies in identifying and testing the ecological principles on which to base such explanations [30].
The current study demonstrates that a basic biological principle about the internal milieu provides a refined and parsimonious explanation of human decisions under risk. Our virtual foraging task was specifically designed to test the influence of homeostatic principles on risky choice. It thereby relates to, and extends, previous tests of risk-sensitive foraging theory in animals [38–41]. Risk-sensitive foraging theory provides an account of how animal should choose between risky foraging options so as to maximize their fitness [40]. The crucial insight of risk-sensitive foraging theory is that foraging animals should choose options with higher variance if options with lower variance cannot provide a sufficient amount of energy to meet critical levels until a certain time point. For example, hungry birds in winter should become more risk-prone as nightfall approaches. Thus, risk-sensitive foraging theory provides an ecologically rational benchmark [38–40] although empirical evidence for it has been mixed [40]. Similar to risk-sensitive foraging theory, our model comprises a hard boundary that is relevant to the decision-maker within a given time horizon. Crucially, we introduce a novel and simple mathematical description for deriving sequential gambles that mirror foraging settings. Testing the model in a virtual setting in humans circumvents challenges of non-human animal research such as the need to impose actual threats onto participants or the need to impart outcome distributions through extensive training.
Risk-sensitive foraging theory has been related to loss aversion [40], which refers the empirical observation that humans seem to care more about losses than gains of equivalent magnitude [40,42]. One may speculate that loss aversion may be related to our finding that participants minimized the probability of starvation. However, loss aversion can only arise in mixed gambles (i.e., when options entail gains and losses) [23,42], and our gambles did not involve losses. Therefore, loss aversion cannot explain our findings.
Our approach is in line with some recent studies that have employed virtual foraging-like tasks to probe the psychological and neural mechanisms of complex decision-making in animals [41,43] and humans [44–46]. One notable study showed that humans adjust their risk-taking behavior dynamically over a sequence of gambles [44]. Another study provides evidence that humans continuously reassess the sequences of gambles available to them in the future although economically optimal strategies prescribe that decisions be independent of sequence order [47]. Our results complement these findings by suggesting that such behavior could be easily explained if people take into account the probability of “starvation” during the choice sequence. Overall, the current study makes detailed predictions for apparent irrationalities in dynamic foraging tasks that are consistent with earlier reports.
Our model of homeostatic decision-making lends itself to possible extensions. First, decision-makers usually have to maintain several variables in a homeostatic range. Our model can easily be extended to such situations with the prediction that decision-makers minimize the joint probability of starvation, which may imply giving up a large amount of one variable to avoid getting zero of another. When boundaries are soft rather than hard, this can be thought of as minimizing a constrained functional that describes a trajectory through homeostatic space. Second, risk preferences are often assumed to be rather stable personality traits [48] but our model implies that they should vary depending on threats to homeostasis [38]. Third, insurances for rare high-impact events have been a recent focus in economics [49]. The concept of starvation in our model may give a handle on investigating the impact of such events on human decisions.
Our results suggest that the pursuit of this fundamental biological goal translates into simple but specific predictions for decision-making that are amenable to empirical tests. Standard economic models provide an indispensable benchmark against which to test the inclusion of additional considerations about biological considerations [12,15,23]. Commonly, models of risky decision-making have to strike a balance between the elegance of axiomatic economic foundations that are at odds with empirical observations and the unwieldy ad hoc assumptions of irrational biases. Our results provide an example that models based on fundamental biological principles such as homeostasis can reconcile parsimony with an explanation for apparent irrationalities.
The study was conducted in accord with the Declaration of Helsinki and approved by the governmental research ethics committee (Kantonale Ethikkommission Zürich, KEK-ZH-Nr. 2013–0328). All participants gave written informed consent using a form approved by the ethics committee.
Twenty-two participants (15 female; age: mean = 25 years, SD = 5.0) were recruited from a student population via mailing lists of local universities. Participants were paid a show-up fee of CHF 15 plus a variable amount (see below).
Participants completed 960 trials in two variants (frames) of a binary choice task: the foraging and the casino frames (Fig 1A and 1B). The same list of 480 combinations of gambles was used for both frames (i.e., the outcome distributions were numerically equal; see Table 1 for an overview of the variables; see below and S1 Text for details on how gambles were derived). For both frames, participants received detailed written instructions and performed eight training trials followed by two blocks of the actual task. The task was presented using the MATLAB toolbox Cogent 2000 (www.vislab.ucl.ac.uk). The instruction for the foraging frame told participants to imagine themselves in a hunter-gatherer context. Since we wanted to exclude that putting participants into a foraging mindset influenced choices in the casino frame, all participants completed the casino frame before the foraging frame. The 480 gamble combinations for each frame were split into two blocks, which were counterbalanced for order. Each list contained 80 unique gamble combinations; the remaining 400 gamble combinations were included in both lists. During the game participants did not see the outcomes of their choices. That is, participants were given examples of possible outcomes in the written instruction but they did not directly experience them. At the end of the experiment, one trial from each of the four blocks was randomly chosen. The outcomes of these trials were determined based on participants’ choices and the corresponding amount was paid out (1 point was worth CHF 0.75). Thus, both frames were incentivized in the same way. See Fig 1 and S2 Text for further details.
We used the mathematics of random walks to derive outcome distributions for the 480 combinations of gambles (Table 1). We briefly introduce the basic logic (Fig 1C). For details see S1 Text. In a random walk an imaginary agent starts at a given position on a line of positive integers. The starting position corresponds to the initial number of energy points. The agent makes a number of steps on that line, which correspond to the number of days. In each step, the agent moves “right” with a certain probability p and “left” with the probability q = 1-p. Moving left corresponds to unsuccessful foraging and the step sizes correspond to a fixed cost of one energy point. Moving right corresponds to successful foraging and the step sizes correspond to the variable points to gain (minus the cost of one point). Zero represents an absorbing boundary (i.e., if the agent reaches zero, the random walk stops). The possible positions on the number line after a certain number of steps correspond to the range of outcomes. To obtain the probability of an outcome, all the probabilities of all “branches on the tree” toward that outcome have to be summed up. (The number of “branches” is calculated with a binomial coefficient.) Along a given branch the probabilities (i.e., p or q) have to be multiplied. We created different gambles by varying combinations of starting positions, probabilities of moving right and step sizes to the right. In the current study, we included gambles with four different combinations of starting positions and number of steps (a) starting position 1 and 1 step, (b) starting position 1 and 2 steps, (c) starting position 2 and 2 steps, and (d) starting position 2 and 3 steps (with each combination occurring in 120 combinations of gambles). We used the outcomes and their respective probabilities to calculate the statistical moments of the chosen gambles. The probabilities of reaching zero are denoted pstarve. Note that in the gambles included in the current study pstarve was never zero (see Table 1).
Log-transformed RTs were analyzed using a linear mixed effects model as implemented in the R package lmer [53] (http://cran.r-project.org/web/packages/lme4/index.html). Log-transformed RTs were approximately normally distributed. The independent variables in the mixed effects model were the variables that the comparison of choice models identified as relevant (i.e., EV and pstarve). Specifically, the fixed effects of the model included the difference between the two options in EV and pstarve as well as the interaction of the two. Random effects for participants included a random intercept and random slopes for EV, pstarve, and their interaction. The model is given by the following equation:
Significance levels of the fixed effects were determined by performing log-likelihood tests, which compared the full model to models without the respective factor.
|
10.1371/journal.pntd.0000103 | Targeted Screening Strategies to Detect Trypanosoma cruzi Infection in Children | Millions of people are infected with Trypanosoma cruzi, the causative agent of Chagas disease in Latin America. Anti-trypanosomal drug therapy can cure infected individuals, but treatment efficacy is highest early in infection. Vector control campaigns disrupt transmission of T. cruzi, but without timely diagnosis, children infected prior to vector control often miss the window of opportunity for effective chemotherapy.
We performed a serological survey in children 2–18 years old living in a peri-urban community of Arequipa, Peru, and linked the results to entomologic, spatial and census data gathered during a vector control campaign. 23 of 433 (5.3% [95% CI 3.4–7.9]) children were confirmed seropositive for T. cruzi infection by two methods. Spatial analysis revealed that households with infected children were very tightly clustered within looser clusters of households with parasite-infected vectors. Bayesian hierarchical mixed models, which controlled for clustering of infection, showed that a child's risk of being seropositive increased by 20% per year of age and 4% per vector captured within the child's house. Receiver operator characteristic (ROC) plots of best-fit models suggest that more than 83% of infected children could be identified while testing only 22% of eligible children.
We found evidence of spatially-focal vector-borne T. cruzi transmission in peri-urban Arequipa. Ongoing vector control campaigns, in addition to preventing further parasite transmission, facilitate the collection of data essential to identifying children at high risk of T. cruzi infection. Targeted screening strategies could make integration of diagnosis and treatment of children into Chagas disease control programs feasible in lower-resource settings.
| Chagas disease kills more people than any other parasitic disease in the Americas. The disease is caused by Trypanosoma cruzi, a single-cell parasite which is transmitted between people and other mammals by triatomine insects. Chagas disease control programs in Latin America have focused mainly on stopping transmission of T. cruzi rather than active surveillance for infection among human populations at risk. These programs have been very successful in controlling triatomine insects by spraying households with insecticides. This article is the first to describe T. cruzi transmission in an urban environment and show evidence that transmission may be epidemic in the city of Arequipa, Peru. The article also demonstrates how data easily collected during an ongoing insecticide spraying campaign in Arequipa might be used to identify children at greatest risk of infection with T. cruzi. The results of the analysis might aid in the optimal use of limited diagnostic resources by targeting screening efforts to those children in greatest need of diagnosis. Such targeted screening could facilitate the integration of diagnosis and treatment of children into Chagas disease control programs currently focused primarily on insects.
| An estimated 11 million people are currently infected with the causative agent of Chagas disease, Trypanosoma cruzi, in Latin America [1],[2]. T. cruzi is a protozoan parasite carried in the gut of bloodsucking triatomine bugs (Hemiptera, Reduviidae), and humans become infected with the trypanosome mainly through contamination with the insect's feces deposited on mucous membranes or broken skin. Many countries have implemented Chagas disease control activities, though most focus on interruption of T. cruzi transmission rather than surveillance for infection among human populations at risk. Triatoma infestans is the principal vector of T. cruzi in South America and the sole vector in southern Peru. A campaign to eliminate T. infestans, known as the Southern Cone Initiative, has been remarkably successful in interrupting vector-borne transmission of T. cruzi through household insecticide application, especially in Uruguay, Chile and Brazil [1]. Although the World Health Organization recommends serologic diagnosis and drug treatment of all T. cruzi-infected children in affected areas, national control programs in Peru and other countries have not had sufficient resources for comprehensive serological screening [3].
Anti-trypanosomal drug therapy can cure 60% or more of infected children aged < 13 years [4],[5],[6], however treatment efficacy apparently decreases with the duration of infection and side effects to anti-trypanosomal drugs increase with age [7]. Without timely diagnosis, children infected with T. cruzi prior to implementation of vector control may miss the window of opportunity for effective chemotherapy, and the Peruvian health system may be burdened with a generation of individuals aging with Chagas disease as has occurred in other countries [8]. Due to the significant labor and cost involved in diagnosis of T. cruzi infection, targeted screening strategies are needed to integrate diagnosis and treatment of children into Chagas Disease control programs in South America.
Vector-borne transmission of T. cruzi, typically confined to rural communities, has become an urban problem in Arequipa, Peru, a city of 850,000 people. Urban transmission cycles are also established elsewhere in the region (Region of Health, Arequipa, unpublished data and [9]), but little is known about Chagas disease transmission in or around cities. The Arequipa Regional Office of the Ministry of Health began a vector control campaign in the greater metropolitan area of Arequipa in 2002, and efforts continue. We accompanied the vector control campaign to one community on the outskirts of Arequipa and collected entomological and census information as insecticide was applied to each household. We then performed a cross-sectional serological survey for T. cruzi infection among the children of this community.
The aim of the study was to develop targeted screening strategies to detect T. cruzi infection in children from data collected during a vector control campaign. Drawing on methods from ecology, traditional epidemiology and Bayesian statistics we first describe the spatial patterns of T. cruzi transmission in a community and evaluate risk factors for infection in children. We then show how spatial and risk factor information can be used to identify high-risk children for targeted screening and evaluate alternative targeted screening strategies.
Arequipa is located at an elevation of 2300 meters in southern Peru. Arequipa's climate is arid most of the year, though there is rainfall between the months of January and March. Santa Maria de Guadalupe and Alto Guadalupe (hereafter referred to together as Guadalupe) are two of hundreds of communities located on hillsides on the outskirts of Arequipa (16.44°S, 71.59°W) and have been described previously [10]. Approximately 2550 people live in 397 houses of Guadalupe over an area of 14.1 hectares (2800 households/km2). Three hundred and seventy-four of the 397 households were sprayed with deltamethrin powder (Bayer K-othrina, Lima, Peru) suspended in water at an intended rate of 25 mg/m2 by the Arequipa Regional Office of the Ministry of Health in November and December of 2004. Twenty-three households either were closed or refused insecticide treatment. At the time of insecticide application, 194 (52.0%) households were found to be infested with Triatoma infestans, and 72 (19.3%) were infested with triatomines carrying T. cruzi [10]. Guadalupe was sprayed again in April of 2005.
During the course of the first insecticide application to households in Guadalupe, 2 trained triatomine collectors systematically searched each room of the human dwelling, each animal enclosure, and the remaining peridomestic area for a total of 1 person-hour. Triatomines captured from each site of collection were stored separately on ice packs and taken to the University of San Agustin where they were counted by site, stage, and sex (for adults). A sample of 10 live and moribund adult and 5th instar triatomines from each site of collection were examined microscopically for T. cruzi infection following procedures outlined in Gürtler et al. [11]. In order to evaluate household risk factors for T. cruzi infection in children, all sites of collection were classified as either domestic or peridomestic. Domestic sites include all sleeping, living, cooking and storage rooms of the human dwellings. Peridomestic sites include animal enclosures and all other structures in the enclosed yards surrounding the human dwellings. Household position was determined with a handheld global positioning system unit with an accuracy of 10 m (Garmin Corporation, Olathe, KS, USA). The entomologic collectors also gathered information on household materials and animal husbandry practices through a structured questionnaire. A census of the human population was performed separately by research nurses. Complete survey and laboratory sampling techniques are described in Levy et. al. [10].
Serologic testing was carried out between August and October of 2005. All children < = 18 years old were invited to participate in the study. Trained study staff explained the study to children and their parents or guardians in schools and over the course of several meetings in the community of Guadalupe. Participants 18 years of age provided informed consent. The parents or legal guardian of all participants under 18 years of age provided informed consent and each participant 7 years or older provided informed assent. The consent form was read aloud to all illiterate parents and participants, and in these cases consent or assent was indicated with the person's fingerprint rather than a signature.
After informed consent, 5 ml of venous blood was drawn by a trained research nurse from children over 5 years old; 3 ml was drawn from younger children. Blood was kept on ice and separated on the day of collection by centrifugation. Aliquots of sera were stored at −20°C until testing. Sera were tested by commercial ELISA (Chagatek, Biomerieux, formerly produced by Organon Teknika). All positive sera, and 10% of negative sera, were tested by immunofluorescence assay (IFA) at the Centers for Disease Control and Prevention. A specimen was considered positive by ELISA if absorbance was at least 0.100 greater than the average absorbance of three negative controls, following the manufacturer's indications. IFA titres of 1∶32 or higher were considered positive (F. Steurer, Division of Parasitic Diseases, CDC, Atlanta, GA). Children whose blood was positive by both ELISA and IFA were considered seropositive [3]; three children with discordant results were excluded from this analysis. All seropositive children were enrolled in Peru's integral health system (SIS-sistema integral de salud), and offered free, directly-observed, treatment by the Arequipa Regional Office of the Ministry of Health. The study protocol was approved by the Institutional Review Boards of A. B. PRISMA, Instituto Nacional de Salud, Peru (National Institute of Health, Peru), Johns Hopkins University Bloomberg School of Public Health, and the Centers for Disease Control and Prevention.
After examining the spatial patterns of vector infestation, parasite-infected vectors and seropositive children, we evaluated the association between the presence of a seropositive child and local covariates measured in each household. All data on local covariates were collected during the insecticide spray campaign. We divided these covariates into groups based on the amount of effort needed to collect each type of data. Census data, such as age and the presence of domestic animals, required the administration of a questionnaire. Routine spray data were those data collected by the Ministry of Health during spray campaigns, such as the presence or absence of vectors in the domestic and in the peridomestic area. Timed vector search data required a systematic timed entomologic search, and consisted of estimates of vector densities in the domestic and peridomestic areas. Microscopic examination data comprised the presence of T. cruzi-infected vectors among a sample of insects captured from the domestic or peridomestic areas.
In univariate analyses, associations between confirmed seropositivity and binary covariates were evaluated with the χ2 test. For continuous covariates a non-parameteric receiver operating characteristic (ROC) curve was plotted and the area under the curve (AUC) estimated by the trapezoid rule of integration. Larger areas (greater than 50%) indicate positive association between the covariate and confirmed diagnosis; smaller areas (less than 50%) indicate negative association. The area under the non-parametric ROC curve is equivalent to the Wilcoxon rank sum statistic (also known as the Mann-Whitney U statistic), and p-values for associations between continuous covariates and confirmed seropositivity were estimated by the Wilcoxon rank sum test [16]. All covariates with p value <0.05 in univariate analysis were considered in multivariate analyses.
We used Bayesian hierarchical mixed modeling techniques to estimate the effects of multiple covariates on the probability of confirmed seropositivity controlling for spatial autocorrelation of the outcome variable. The mixed model conditions inferences and predictions on an unknown underlying risk of each child (a random effect) [17]. We fit a spatial conditional autoregressive logistic model in which the underlying risk of children in each household was assumed to be a function of the infection status of children in neighboring households [15]. We considered two households neighbors if they were within 42 meters of each other because T. infestans nymphs are known to crawl at least 42 meters [18]. One child's home had no neighbors and was left out of the analysis.
Bayesian hierarchical models have been used for many years in epidemiology, and especially spatial epidemiology (reviewed in Boyd 2005 [17]). Inference in Bayesian hierarchical models is based on the joint posterior distribution, or the posterior, of model parameters. Bayes' theorem states that the posterior distribution is proportional to the product of a conditionally independent likelihood and the prior distribution of the parameters. Prior distributions, or priors, describe the conceivable values of a parameter before the collection of the data. We specified non-informative priors for model parameters so as to minimize the importance of our a priori assumptions. Priors for all fixed effect parameters were normally distributed with a mean of 0 and a variance of 106. We specified a Gaussian distribution for the random effects, and set a uniform prior distribution for the inverse of the standard deviation of the Gaussian distribution. Overly wide priors for the inverse standard deviation of the random effects led to numerical overflow errors; a uniform prior with a range from 0 to 15 was broad enough to avoid truncating the posterior estimates and narrow enough to avoid overflow errors. Models were fit in WinBugs 14.1; code is available upon request.
We used the fit Bayesian models to rank the children in the community based on their age and local covariates, but without including any information about the spatial component of risk, which is unknown prior to testing. Using the ranking from the fit models we plotted non-parametric ROC curves and calculated the area under these curves. We then considered two-step case detection strategies that reincorporated spatial information. In two-step strategies the fit multivariate models are used to rank the risk of seropositivity in children based solely on their age and local covariates collected from their households during the spray campaign. In the first step of screening a proportion of the highest-risk children are tested. The results of the preliminary screening are used to identify children living within a given distance of seropositive children, and in the second step of screening these children are tested (ring screening). We considered ring screening radii of 10, 20, 30, 40, 50, 60 and 70 meters. For each radius we plotted an ROC curve for the two-step screening strategy and calculated the area under the curve. We calculated the sensitivity (number of cases detected/total number of cases) and specificity (number of non-cases detected/total number of non-cases) for all potential cutoff points of the ROC curves. We report the percent of non-cases (1-specificity) that must be tested to identify >80% of cases for each fit model. ROC analyses were programmed in the R statistical environment (www.r-project.org), code is available from the authors upon request.
Specimens were tested for a total of 433 children. Of these, 26 (6.0% [95% CI 3.8–8.4]) were positive for antibodies to T. cruzi by ELISA, and 23 (5.3% [95% CI 3.4–7.9]) were confirmed positive by IFA. No ELISA-negative specimens were positive by IFA. ELISA-positive specimens from 3 children were IFA negative; these children were excluded from subsequent analysis. Thirty-two children either lived in households that refused spraying or could not be matched to households based on address information provided by their schools. The total sample size for risk factor analyses was 398 children, of whom 23 (5.8%) were confirmed seropositive for T. cruzi infection.
Of census data variables, age and the presence of animals, almost exclusively dogs and cats, sleeping in the domestic area of the household were significantly associated with confirmed T. cruzi seropositivity (Table 1). Both the presence of triatomines in the domestic and peridomestic areas were weakly associated with child seropositivity. Vector densities in the domestic and the peridomestic areas were associated with seropositivity, but the association was stronger for domestic vector density. The presence of T. cruzi-infected vectors in domestic sites of collection and in the peridomestic area were both significantly associated with seropositivity. Neither the presence of any specific animal species in the peridomestic area, nor the presence of any specific building material in the domestic area were significantly associated with child seropositivity (data not shown).
The difference in Ripley's K functions for households with and without triatomines never exceeded the 99% tolerance limits, suggesting no significant spatial clustering of vector infestation (Figure 1,I-3). The difference in K functions for households with T. cruzi-infected triatomines and remaining households exceeded 99% tolerance limits at all but 1 spatial scale (20 m) from 10 to 140 m indicating clustering of infection in vectors within the spatial distribution of the vectors (Figure 1,II-3). The difference in K functions for households with seropositive children and those with seronegative children exceeded tolerance limits at all spatial scales from 10 to 270 m indicating significant spatial clustering of seropositivity among children tested in the community (Figure 1,III-3).
After controlling for spatial effects, the risk of T. cruzi infection increased 20% with each year of age (Table 2, model A). The presence of animals sleeping inside the house did not significantly increase risk after controlling for age and spatial effects. The presence of triatomines within the household increased the risk of infection nearly two-fold after controlling for age and spatial effects, but the increase was not statistically significant (Table 2, model B). The estimated domestic vector density was significantly associated with the risk of seropositivity. A child's risk of infection increased by 4% for each bug captured in the domestic area after controlling for age and spatial effects (Table 2, model C). Domestic vector densities were very heterogeneous between households. Children who lived in households in which we detected a T. cruzi-infected triatomine were not significantly more likely to be seropositive than those in households without T. cruzi-infected vectors after controlling for age, nor after controlling for age and vector density inside the house (Table 2, model D).
The area under the curve for the predictive model based on age alone was 0.64 (Table 3, model A). This area increased with ring testing, and the maximum area was 0.81 when children living within 40 or 60 meters of identified cases were also tested (Table 3, model A). Testing 28% of seronegative children would be necessary to identify >80% of infected children (Figure 2, model A). The area under the curve for the model with age and the presence or absence of vectors within the household (routine spray data) was 0.68, and also increased with ring testing to a maximum of 0.85 when the testing radius was 10 or 20 meters (Table 3, model B); >80% of infected children could be identified while testing 22% of seronegative children (Figure 2, model B). Using the timed vector search data, the model with age and household vector density had an AUC of 71% with no ring testing, and a maximum AUC of 85% when children living within 20 meters of a case were tested (Table 3, model C). Only 19% of seronegative children would need to be tested to identify >80% of infected children (Figure 2, model C). The model which included age, vector density, and the presence of T. cruzi-infected T. infestans in the household had an AUC of 72% with no ring testing and an AUC of 85% with ring testing over a radius of 10 or 20 meters (Table 3, model D); >80% of infected children could be identified by testing 22% of seronegative children (Figure 2, model D).
One of the two-step screening strategies with an area under the ROC curve of 85% would begin by ranking children based on their age and the number of vectors captured within their houses. In the first round of screening 15% of the highest-ranked children would be tested, and the households of all seropositive children identified. In the second round of screening all children living within 20 meters of households with seropositive children would be tested. In total 23% of the population would be screened and 19/23 (83%) of seropositive children diagnosed.
Chagas disease transmission cycles have become established in communities on the outskirts of the city of Arequipa, Peru. A vector control campaign is currently disrupting transmission of T. cruzi, but we found 5.3% of children in Guadalupe had already been infected by the time their households received insecticide application. Many thousands of children live in similar communities in Arequipa and likely represent a significant proportion of the Chagas disease burden in Peru.
We observed spatial evidence that transmission of T. cruzi was in an epidemic phase in Guadalupe at the time household insecticide application was initiated. As noted previously [10], although households infested with vectors were distributed across the community, those with vectors carrying T. cruzi were significantly clustered. Such a pattern is consistent with recent introduction of T. cruzi into an established vector population, followed by vector-borne dissemination of the parasite to susceptible hosts in nearby houses. Here we show further that households with seropositive children were also significantly clustered around each other. Interestingly, clusters of households with infected children were well within the looser clusters of households harboring T. cruzi-infected triatomines. The pattern of infected children living in households at the heart of the cluster of households with infected vectors is further evidence of epidemic spread of T. cruzi in the community. If the parasite is actively spreading from one or many points of introduction in the community then we would expect exposure time of children living nearer to the site or sites of introduction to be much greater than that of children living at the periphery of the parasite's distribution. This spatial inequality of exposure would lead to the observed tight clustering of infection in children.
Traditional epidemiologic methods are not well suited for epidemics when infection is clustered in space [17]. We therefore relied on Bayesian mixed models to estimate associations between risk factors and T. cruzi infection status, controlling for the spatial aggregation of seropositive children. We found domestic vector density to be significantly associated with infection among children living in the house after controlling for age and spatial effects. The risk of infection increased by 4% per bug found within a child's house, such an increase in risk is very important given the heterogeneity in domestic vector densities across households. We had only cross-sectional entomologic data from 2004 with which to identify associations with infections which may have occurred years earlier. Although it is unlikely that vector density is constant over time, our finding may be significant because houses with a high density of bugs in 2004 likely also had a high density of bugs in previous years. Controlling for spatial effects and household vector density, a child's risk of infection increased by 22% for each year of age; this estimate was nearly constant across all multivariate models considered.
Although no published study has described risk factors for T. cruzi infection in an urban epidemic situation controlling for spatial autocorrelation, our results are qualitatively similar to findings from other foci of Chagas disease. In two papers, Gürtler and collaborators described risk factors for infection in children under 16 years of age in three rural towns in northwestern Argentina where T. infestans and Chagas disease were endemic. The authors also found the age of children and the density of insect vectors in their houses to be significantly associated with infection [19],[20]. Similar relationships between T. infestans density and risk of infection have been noted by Catala [21],[22]. As in the study by Gürtler [20], we found infected children living in households in which we caught few or no vectors. Gürtler suggests various possible explanations for infection in the apparent absence of vectors, including lack of sensitivity of timed entomologic collections and infection of children while staying in other households [20]. Unreported vector control measures taken by homeowners may also be an important explanation. In the urban environment where houses are contiguous, infection by vectors entering the household from neighboring households may also pose a risk to children in houses not colonized by the vector. In Guadalupe, as in the study site described by Gürtler, there does not seem to be a threshold vector density below which transmission does not occur.
We were not able to evaluate the sero-status of participant's mothers to consider the potential that children were infected congenitally [23]. In peri-urban Arequipa, where the prevalence of T. cruzi infection among women of childbearing age is low [24], the risk of congenital infection is likely negligible relative to the risk of vector borne transmission. In areas with higher rates of congenital transmission infection of children by this route may decrease the sensitivity of targeted screening strategies based on spatial and household risk factors alone. A second potential cause of error for the screening strategies is prior infection of immigrant children. We were not able to gather detailed migration histories through surveys conducted during the spray campaign. Most of the children in Guadalupe were born there, and the strong spatial pattern of infection suggests that cases in children are autochthonous. In other areas, where more cases are imported, targeted screening strategies would need to rely more on migration history data.
Knowledge of local risk factors [19],[25],[26] can be used to enhance the efficiency of serologic screening by identifying groups of persons at greatest risk of infection. Previous authors have suggested identifying high-risk children by diagnosing reservoir animals in their households[19], or calculating their Trypanosoma cruzi transmission risk index, which includes blood meal analysis of triatomines [21]. While very important to understanding the cycle of domestic T. cruzi transmission, these methods are too costly for routine use in low-resource settings. We show here how entomologic data collected easily during a vector control campaign could be used to identify children at increased risk of infection. We also show that a two-step screening strategy would be much more effective in detecting seropositive children than a screening strategy based on entomological data alone. Mott et al. showed household aggregation of seropositive children in Brazil [27], suggesting that in a targeted screening campaign all children living in a house with a seropositive child should be diagnosed. In Guadalupe aggregation of seropositivity extends beyond household boundaries and testing of children living in surrounding households is needed.
Economic analysis is needed to optimize targeted screening strategies for Arequipa given the real costs of gathering data and serologic testing. Measuring vector density by the person-hour method requires that an additional person accompany each spray worker at the time of application of insecticide. Alternative methods for estimating vector density, such as leaving collection bags with homeowners following spraying [28], or timed collection by the spray worker alone, may work nearly as well at a much lower cost. Data from microscopic examination of triatomines for T. cruzi infection did not result in an improvement in identification of seropositive children in Guadalupe, possibly due to our ability to routinely examine only a small proportion of captured vectors. Targeted screening might be significantly simplified with the utilization of rapid tests for T. cruzi infection. If test results are available immediately ring screening around cases could also begin immediately, decreasing the labor involved in the two-step screening process. Economic analysis should also take into account the increase in the positive predictive value of diagnostic tests that occurs when testing is limited to a small higher-risk population.
We are limited in terms of our ability to extrapolate the findings of our study to other areas. If, as we suggest, transmission is epidemic in peri-urban Arequipa, the results of our predictive models might be sensitive to the precise timing of insecticide application. Had Guadalupe been sprayed one year later not only might the prevalence of infection have been higher, but the associations between covariates and infection might have been quantitatively different. In an analogous analysis, Struchiner et al. demonstrate how estimates of the effect of a vaccine against malaria would change over the course of a malaria epidemic [29]. The authors show that the expected effect of a vaccine decreases over the course of an epidemic, and the same may be true for local risk factors for T. cruzi infection. More empirical data collection and mathematical modeling are necessary to elucidate the associations between spatial, temporal and entomologic variables and seropositivity as the parasite spreads through a peri-urban community. Until we have a better understanding of the associations between risk factors and infection at different stages of epidemic transmission we suggest using adaptive sampling methodology [30] to simultaneously implement and evaluate screening strategies.
In conclusion, our results suggest that peri-urban communities in Arequipa may be in the midst of an epidemic of vector-borne T. cruzi transmission. Like in the early HIV epidemic, climbing prevalence of T. cruzi infection has gone unnoticed due to the subclinical nature of most recent T. cruzi infections. Ongoing vector control campaigns, in addition to preventing further parasite transmission, facilitate the collection of data essential to identifying children at high risk of T. cruzi infection. Concentrating diagnostic resources on these high risk children will ensure that the greatest number of infected children receive treatment before their window of opportunity for effective chemotherapy closes. Targeted screening strategies could make integration of diagnosis and treatment of children into Chagas disease control programs feasible in lower-resource settings. |
10.1371/journal.ppat.1004977 | Transgenic Mouse Bioassay: Evidence That Rabbits Are Susceptible to a Variety of Prion Isolates | Interspecies transmission of prions is a well-established phenomenon, both experimentally and under field conditions. Upon passage through new hosts, prion strains have proven their capacity to change their properties and this is a source of strain diversity which needs to be considered when assessing the potential risks associated with consumption of prion contaminated protein sources. Rabbits were considered for decades to be a prion resistant species until proven otherwise recently. To determine the extent of rabbit susceptibility to prions and to assess the effects of passage of different prion strains through this species a transgenic mouse model overexpressing rabbit PrPC was developed (TgRab). Intracerebral challenges with prion strains originating from a variety of species including field isolates (ovine SSBP/1 scrapie, Nor98- scrapie; cattle BSE, BSE-L and cervid CWD), experimental murine strains (ME7 and RML) and experimentally obtained ruminant (sheepBSE) and rabbit (de novo NZW) strains were performed. On first passage TgRab were susceptible to the majority of prions (Cattle BSE, SheepBSE, BSE-L, de novo NZW, ME7 and RML) tested with the exception of SSBP/1 scrapie, CWD and Nor98 scrapie. Furthermore, TgRab were capable of propagating strain-specific features such as differences in incubation periods, histological brain lesions, abnormal prion (PrPd) deposition profiles and proteinase-K (PK) resistant western blotting band patterns. Our results confirm previous studies proving that rabbits are not resistant to prion infection and show for the first time that rabbits are susceptible to PrPd originating in a number of other species. This should be taken into account when choosing protein sources to feed rabbits.
| Prions, the infectious agents responsible for causing mad cow disease, amongst other diseases, can transmit from one species to another. For example, Bovine Spongiform Encephalopathy can transmit to humans resulting in invariably fatal variant Creutzfeldt-Jakob Disease. We wanted to study the susceptibility of rabbits as, until recently, they were considered a prion resistant species. Once proven otherwise, we wanted to know which particular prions rabbits were susceptible to. With this aim, a transgenic mouse was designed expressing the rabbit prion protein gene instead of the corresponding mouse gene to model the transmission barrier between rabbits and other species. The resultant mice where challenged with several field prion isolates including classical and atypical strains of Bovine Spongiform Encephalopathy, sheep Scrapie and cervid Chronic Wasting disease. The transgenic mice were susceptible to classical and atypical Bovine Spongiform Encephalopathy prions and also to mouse-adapted Scrapie prions. This information must be taken into account when assessing the risk of using ruminant derived protein as a protein source to feed rabbits.
| Prions are protein based, genome devoid, infectious agents causing Transmissible Spongiform Encephalopathies (TSEs), a group of diseases classified as transmissible protein misfolding disorders [1,2]. Prions show a remarkable ability for interspecies transmission. Initially, a species barrier was defined, but extensive field and experimental evidence has been published proving that interspecies prion transmission is not an isolated phenomenon [3–6]. Interspecies transmission of prions has resulted in the generation of significant prion strain diversity and its incidence has been documented worldwide [3,4,7–10].
The existence of prion diseases has been documented for centuries with the earliest reports of scrapie cases dating back to 1732 [11]. In the last seven decades prions were also reported in other animal species, usually in the form of outbreaks, which somehow involved human intervention. Namely classical bovine spongiform encephalopathy (BSE-C) [12], feline spongiform encephalopathy (FSE) [13] and transmissible mink encephalopathy (TME) [14]. Humans can also be included in the list of TSE susceptible species due to the Fore tribe from Papua New Guinea suffering from Kuru [15] or the relatively newly created variant Creutzfeldt-Jakob disease (vCJD)[3]. Cervidae is another family of animals currently affected by a, yet uncontrolled, epizooty: chronic wasting disease (CWD) [16]. Although classical animal prion disease strains, as opposed to the so called atypical prion disease strains [17–20], have been documented for at least three centuries [11], sporadic spontaneous generation of atypical prions has probably existed for as long as susceptible species have been present in large enough numbers for the spontaneous event to occur. Currently there is no evidence to suggest that any mammalian species cannot undergo a spontaneous disease-linked prion protein misfolding event [21] as long as there are sufficient numbers of individuals with the necessary lifespan.
Although the mechanisms of interspecies prion transmission remain unknown, in vitro and in vivo studies have shown that species particularly susceptible to certain prion strains can actually be resistant to others which originated in the same or different species [9,21–28]. The ability of prions to adapt to new species and even generate new strains with pathobiological properties different from the original one is not an isolated phenomenon [9,27,29,30]. Therefore new prion strains may arise with the ability to infect new species previously considered resistant.
Normal cellular prion protein (PrPC) is a host encoded protein, particularly abundant in nerve cells, which when misfolded is believed to acquire pathological properties leading to TSE neurodegenerative disease [1]. Several studies argue that certain species specific amino acid sequences of PrPC may render some species less susceptible to TSE [31–33] due to them being less prone to misfolding. This, along with absence of experimental evidence or TSE field cases described, led to belief that dogs, horses and rabbits (leporids) were more resistant to prion infection than other mammalian species [34,35].
Leporids have been the most intensely studied, both in vivo and in vitro, of the presumed prion disease resistant species. This is probably because rabbits are consumed by humans and also due to their comparatively small size and long lifespan which facilitates their use as experimental animals. Our group has proved recently, in contrast with the last three decades of reports, that rabbits are susceptible to prion diseases. Using protein misfolding cyclic amplification (PMCA), inocula where generated in vitro which were infectious and transmissible in this species [23] and more recent studies have proven that rabbit PrPC has a misfolding ability comparable to other species as BSE prions have been shown to retain their in vivo strain properties after misfolding rabbit PrPC [36]. Houdebine’s group studied whether the genetic background of rabbits was responsible for their apparent prion resistance generated transgenic rabbits expressing ovine PrPC. Upon inoculation with scrapie, these rabbits succumbed to prion disease further proving that leporids are not resistant to prion disease [37].
In the present paper we report an extensive evaluation of the susceptibility of TgRab mice to a variety of prion strains by means of in vivo experiments. A transgenic murine model has been generated ad hoc for this purpose which overexpresses the leporid PRNP on a mouse Prnp-null background. This model, denoted TgRab, has already been shown to correlate well with the rabbit model [23]. Our results show the susceptibility of rabbits has been vastly underestimated previously and that they behave similarly to other species whose vulnerability and/or resistance to prion disease also varies depending upon the prion disease strain encountered.
Even though the actual rabbit model would be more suitable for this purpose there are several significant limitations (size, cage space in biocontainment conditions, lifespan, expression levels, and budget required) that are easily overcome by using a transgenic mouse model and such models have been of great use within the field of prion research. Based on our previous experience a new mouse line was generated by pronuclear injection of a construct consisting of the moPrP promoter and the rabbit PrP sequence. From a total of seven positive animals identified from the 83 pups obtained, five animal founders transmitted the transgene to their progeny. After backcrossing to a line that did not express endogenous PrP (STOCK-Prnptm2Edin), expression levels of the transgene were analyzed by western blot. Two out of five transgenic lines expressed PrP at higher levels than the endogenous gene. However, only hemizygous line 58 showed a consistent expression pattern of 5x-6x that of the endogenous rabbit prion protein level and 10x-12x that of the endogenous mouse prion protein level (S2 Fig). This line was selected for further studies. The low expressing lines were discarded since PrPC expression levels were lower than those found in WT rabbits and this would predictably diminish their susceptibility to prions.
In previous experiments normal rabbit brain homogenate was seeded in vitro with different prion strains before applying serial automated PMCA (saPMCA) to determine the ability of rabbit PrPC to be converted by different PrPres conformations. The results of some of these experiments have been reported previously such as seeding with cattle BSE which generated BSE-RaPrPres [36]. Additional prion isolates were included in the present work, which successfully misfolded rabbit PrPC in vitro including SSBP/1 sheep scrapie, ME7 and RML murine adapted scrapie strains and CWD. The following rabbit adapted strains were generated respectively: SSBP/1-RaPrPres, ME7-RaPrPres, RML-RaPrPres and CWD-RaPrPres (S1 Fig) [23]. Spontaneously misfolded PrPres was also obtained from unseeded normal rabbit brain homogenates and named de novo RaPrPres. This spontaneous strain has been demonstrated to be infectious to rabbits [23]. Despite saPMCA not being a quantitative method, rabbit PrPC appeared to be quite susceptible to misfolding since all seeds tested were able to generate PK-resistant RaPrPres by or before round 7 and the unseeded homogenate produced RaPrPres by round 13 [23]. All in vitro-derived RaPrPres products were easily amplified further in vitro.
The western blotting migration pattern of the obtained RaPrPres, particularly the unglycosylated band, was similar to the strains of origin used in the bovine and ovine strains tested.
Accordingly, the following isolates were selected for in vivo challenge: BSE-C, SSBP/1, ME7, RML and CWD. Additionally, we included L-type atypical BSE (BSE-L) and Nor98 Atypical scrapie and the PMCA obtained de novo RaPrPres. The rationale for including the latter, in vitro generated, PrPres was that it was able to infect the natural host i.e. rabbits, our species of study [23], de novo RaPrPres was the only positive control available. Finally, de novo NZW prions (obtained from rabbits infected with de novo RaPrPres) were also inoculated into this transgenic mouse model and even though most of these results have already been published [23], some of them are discussed in the present paper.
As reported for rabbits showing very long incubation times (766 dpi) and a 33% attack rate (1/3) [23], TgRab mice were also susceptible to de novo RaPrPres with a low attack rate (1/11) and a rather long incubation period of 604 dpi. However, upon inoculation with rabbit in vivo-adapted de novo RaPrPres (de novo NZW) the TgRab mice developed a 100% attack rate (8/8) with a shortened incubation period of 256 (±5) dpi (Table 1 and Fig 1). The same rate was obtained by Chianini et al. [23] in rabbits inoculated (in second passage) with this prion strain (Table 1). In vivo experimental challenges in rabbits and TgRab mice have shown a good correlation making the transgenic mouse model overexpressing rabbit PrPC a valid model to study rabbit prion susceptibility.
Even though rabbits had been considered resistant to prion infection until recently [23], TgRab mice could be infected with a number of the prions tested. Prions originating from BSE, i.e. cattle BSE-C and sheep BSE-C, were both infectious (Table 2). Cattle BSE-C showed an attack rate of 44.4% with an incubation period of 551(±10) dpi. Interestingly Sheep BSE-C showed a 100% attack rate and a significantly shortened incubation period of 368(±12) dpi (P = 0.0069, Mann-Whitney test) without previous adaptation to rabbit compared to cattle BSE-C (Fig 1). This supports, once again, the idea that after passage through sheep BSE-C shows enhanced virulence [29].
The picture with scrapie-originating prion isolates was quite different. SSBP/1 prions were not able to infect TgRab (mice survived for longer than 750 dpi). Two other murine adapted classical scrapie prion sources were tested, ME7 and RML, and both strains readily infected TgRab mice with attack rates of 50% and 70% and incubation periods of 360(±41) and 371(±6) dpi, respectively (Fig 1). Therefore, prion strains originating from classical scrapie were transmissible to TgRab mice but only after being adapted previously to rodents. This situation is similar to that found with CWD which will infect hamsters readily after passage through ferrets [9].
The new TgRab model was further characterized by testing its susceptibility to atypical prion strains using the more frequent isolates for each species, BASE (BSE-L, cattle) and Nor89 (sheep). TgRab mice were resistant to infection on first passage with atypical scrapie prions (living up to 775 dpi) (Fig 1) with one exception: a single animal (euthanized at 742 dpi) showed a positive result for PrPd by ELISA but was negative when examined by western blotting and IHC and showed no TSE related spongiform change. A second passage is ongoing to determine if this animal was truly infected.
A 27% attack rate was present in the group inoculated with BSE-L with a mean incubation period of 280(±26) dpi, a similar rate to that of cattle BSE-C but with a much shorter incubation period (the number of positive animals per group was too low to assess statistical significance).
As mentioned before, in vitro adaptation of CWD prion to rabbit PrPc was successful which indicated a potential susceptibility of rabbits in vivo. However, TgRab mice inoculated with CWD did not show any indication of a TSE on first passage, living up to 825 dpi (Fig 1 and Table 2). A second passage is ongoing to confirm these results.
Biochemical and neuropathological characterization of the brains of the inoculated mice strongly suggests that TgRab mice are not only susceptible to multiple prion strains but are also able to maintain their strain features.
Western blotting analysis of TgRab brain homogenates after protease K digestion revealed the characteristic three-band pattern with a predominant diglycosylated band and a 19-20kDa unglycosylated band in mice inoculated with BSE-derived strains (Fig 2). The brains of mice inoculated with RML showed a typical predominance of the monoglycosylated band and a 21kDa unglycosylated band. As shown in the 12B2 antibody developed membrane only ME7 and RML inoculated mice fully maintained the N-terminus specific epitope after PK digestion (Fig 2). The migration pattern of the bands from mice inoculated with de novo strains, both with the in vitro generated (de novo-RaPrPres) and the one obtained from NZW rabbits (de novo NZW), was constant and showed a similar pattern to BSE-C even though, as shown later, the immunohistochemical features differed completely. No bands were observed in western blots of brains of mice inoculated with SSBP/1, Atypical scrapie nor CWD or in any of the negative controls.
Spongiform change and PrPd distribution throughout the brain was semi-quantitatively assessed in histological sections of the inoculated brains of TgRab mice (Figs 3 and 4). Classical BSE-derived strains, namely BSE-C and sheep BSE, yielded very similarly shaped curves characterized by a strong involvement of the medulla oblongata, mesencephalon and thalamus but sparing of the hypothalamus. Involvement of the cortices and hippocampus was less intense but present, particularly at the deeper layers of the parietal cortex, involving the corpus callosum and sometimes extending to the oriens layer of the hippocampal formation. This pattern is equivalent to the one observed for BSE-C in the botg110 mouse model previously published by our group [36]. Mice inoculated with BSE-L, in contrast, showed a widespread involvement of the neocortex and less so in the diencephalon, mesencephalon and medulla oblongata in accordance with the brain PrPd distribution observed in natural and experimental cases of BSE-L in cattle [18,38].
The type of PrPd deposits seen by immunohistochemistry was also distinct in all mice inoculated with classical BSE-derived strains and consisted of amyloid-like rounded plaques, often confluent, which were readily visible on HE stained sections and positively stained in sections subjected to immunohistochemistry for PrPd (Fig 4A). BSE-L inoculated mice lacked plaque type deposits and showed a very different punctate immunolabelling pattern in the neuropil and perikarya (Fig 4A). This was consistent with the pattern obtained in tgBov mice when inoculated with BSE-L.
The scrapie-derived strains RML and ME7 showed PrPd deposits with a tropism for the diencephalon, including a consistent involvement of the hypothalamus (distinct from BSE strains), the mesencephalon and the medulla oblongata and also showed tropisms for the neocortex and cerebellar cortex. The PrPd type, on immunohistochemistry, was distinguishable from that of BSE infected mice, as it was comprised of a fine punctate pattern in the neuropil and perikarya (Fig 4).
The lesion and PrPd distribution of the rabbit-obtained de novo NZW strain showed a tropism confined to the diencephalon, including a consistent involvement of the hypothalamus, mesencephalon and medulla oblongata while sparing the cortices and hippocampus. The PrPd type, on immunohistochemistry, consisted of a fine punctate pattern in the neuropil and perikarya resembling that observed in ME7 and RML infected mice.
The data presented validate the TgRab model to study rabbit susceptibility to prion strains. However, the TgRab line 058, chosen because it was the transgenic line showing the highest PrPC expression levels, also showed a spontaneous phenotype secondary to PrPC overexpression, as described by Westaway and coworkers [39], which needs to be taken into account when evaluating the results of any given experiment. Similar changes have been observed previously in other useful transgenic models overexpressing PrPc [40,41]. In this phenotype, between 300 and 400 days, the majority (over 80%) of hemizygous mice (5x-6x PrPc expression compared to normal levels; S2 Fig) developed gait abnormalities in the hindquarters that progressed slowly to complete hind-limb paralysis and atrophy of muscles (S3 Fig). The animals were able to feed, drink and groom normally and when it was not the case, as with any infected animal that reached the end point criteria, they were humanely euthanatized. See death time points for control groups in Fig 1. The same clinical presentation, but with enhanced severity, appeared in mice homozygous (10x-12x PrPc expression levels) for the transgene as early as 60 days of age.
Microscopically, the skeletal muscle tissue showed irregular diameter of the muscle fibers along with the presence of anguloid fibers, centralization of nuclei and substitution by adipose tissue proliferation in the endomysium (S3E Fig and Fig 3). These changes are compatible with neurogenic atrophy. Lesions were observed also in the central nervous system and consisted of an intense spongiform change in the white matter, particularly in the corpus callosum and internal capsule (S4 Fig). The remaining brain parenchyma also showed diffuse moderate spongiosis, which was more evident in the diencephalon and brainstem and particularly intense in mice euthanized at older ages.
Even though no PrPd was detected by western blotting or ELISA in any of the control animals, upon immunohistochemistry an intense PrPC background immunolabelling was present throughout the brain in agreement with the known overexpression of PrPC. Additionally, a more intense labelling was observed, that could be mistaken for PrPd signaling, which consisted of punctuate labelling around and within the cytoplasm of neurons, mainly located in the cortices but occasionally in the diencephalon and brain stem. Also, in the white matter, a punctate immunolabelling pattern was observed. Certain regions consistently showed strong immunolabelling of PrPC including the cochlear nucleus in the medulla oblongata and the cerebellar cortex where a diffuse labelling was observed in the molecular layer and intense labelling in the granular layer, depicting the synaptic glomeruli (S4 Fig). Despite most of the animals displaying an overt overexpression phenotype, characterization allowed clear discrimination of this from bona fide prion infection in this model.
Brain homogenates (10−1 in PBS) for use as seeds for PMCA or direct intracerebral inocula were prepared manually using a glazed mortar and pestle from brains of animals clinically affected by various TSE: BSE-C and BSE-L field cases supplied by the Laboratorio Central de Veterinaria (Algete, Madrid, Spain), SSBP/1 and ME7 supplied by Animal Heath and Veterinary Laboratory Agency (New Haw, Addlestone, Surrey, UK), CWD from the thalamus area of the brain of a female Mule deer, genotype 225SS, infected with CWD (04–22412 WSV2 EJW/JEJ), supplied by Department of Veterinary Sciences (Laramie, WY, USA), RML supplied by Rocky Mountain Laboratories (Hamilton, MT, USA) and Sheep BSE supplied by Ecole Nationale Vétérinaire (Toulouse, France). The atypical scrapie isolate was obtained from a field case diagnosed in PRIOCAT laboratory, CReSA-IRTA (Barcelona, Spain). Rabbit spontaneous prions were those obtained in the rabbit bioassays conducted in the Moredun Research Institute, Scotland [23].
The in vitro prion replication and PrPres detection of amplified samples was performed as described previously with minor modifications [23,42]. Briefly, rabbit brains used for substrate were perfused using PBS + 5 mM EDTA and the blood-depleted brains were frozen immediately until required for preparing the 10% rabbit brain homogenates (PBS + NaCl 1% + 1% Triton X-100). 50–60 μl of 10% rabbit brain homogenate, either unseeded or seeded with the corresponding prion strain were loaded onto 0.2-ml PCR tubes and placed into a sonicating water bath at 37–38°C without shaking. Tubes were positioned on an adaptor placed on the plate holder of the sonicator (model S-700MPX, QSonica, Newtown, CT, USA) and subjected to incubation cycles of 30 min followed by a 20 s pulse of 150–220 watts sonication at 70–90% of amplitude. Serial rounds of PMCA consisted of 24-48h of standard PMCA followed by serial in vitro 1:10 passages in fresh 10% rabbit brain homogenate substrate. An equivalent number of unseeded (4–6 duplicates) tubes containing the corresponding brain substrate were subjected to the same number of rounds of saPMCA in order to control cross-contamination and/or the generation of spontaneous PrPres. The detailed protocol for PMCA, including reagents, solutions and troubleshooting, has been published elsewhere [43].
PMCA treated samples were incubated with 85–200 μg/ml of protease K (PK) for 1 h at 42°C with shaking (450 rpm) as described previously [44]. Digestion was stopped by adding electrophoresis Laemmli loading buffer and the samples were analyzed by Western blotting.
After isolation by PCR amplification using 5’ CCGCCGTACGTCATCATGGCGCACCTCGGCTAC 3’ and 5’ GGGGCCGGCCTCATCCCACGATCAGGAAG 3’ as primers, the open reading frame (ORF) of the rabbit PRNP gene was cloned into the pGEM-T vector (Promega). The rabbit-PrP ORF was excised from the cloning vector by using the restriction enzymes BsiWI (Thermo Fisher Scientific Inc.) and FseI (New England Biolabs Ltd.) and then inserted into a modified version of MoPrP.Xho vector [45] as described previously [46], which was also digested with BstWI and FseI. This vector contains the murine PrP promoter and exon-1, intron-1, exon-2 and 3’ untranslated sequences. The transgene was excised using NotI and purified with Invisorb Spin DNA Extraction Kit (Inviteck) according to the manufacturer recommendations.
Transgenic mouse founders were generated by microinjection of DNA into pronuclei following standard procedures [40]. DNA extracted from tail biopsies was analyzed by PCR using specific primers for the mouse exon 2 and 3’ untranslated sequences (5’ GAACTGAACCATTTCAACCGAG 3’ and 5’ AGAGCTACAGGTGGATAACC 3’). Those which tested positive were bred to mice null for the mouse Prnp gene in order to avoid endogenous expression of mouse prion protein. Absence of the mouse endogenous Prnp was assessed using the following primers: 5’ ATGGCGAACCTTGGCTACTGGC 3’ and 5’ GATTATGGGTACCCCCTCCTTGG 3’. The rabbit PrP expression levels of brain homogenates from transgenic mouse founders were determined by western blot using anti-PrP MAb L42 antibody (RIDA-Biopharm, Darmstadt, Germany) and compared with the PrP expression levels from NZW rabbit brain homogenates.
Animals homozygous for the transgene showed a spontaneous clinical phenotype as early as 60 days old, resembling the one described in the results section, but more severe, requiring euthanasia at 60–120 days old. Due to this, hemizygous mice were maintained for subsequent studies. The international code to identify this transgenic mouse line is STOCK-Prnptm2Edin Tg(moPrpn rabPrP)58Bps although throughout the paper they are referred to as TgRab mice.
Mice of 42–56 days of age were intracerebrally inoculated under gaseous anesthesia (Isoflurane) through the right parietal bone. A 50 μl SGC precision syringe was used with a 25 G gauge needle and coupled to a repeatability adaptor fixed at 20 μl. A dose of buprenorphine was subcutaneously injected before recovery to consciousness to reduce post-inoculation pain.
Mice were kept in a controlled environment at a room temperature of 22°C, 12 h light-darkness cycle and 60% relative humidity in HEPA filtered cages (both air inflow and extraction) in ventilated racks. The mice were fed ad libitum, observed daily and their clinical status assessed twice a week. The presence of ten different TSE-associated clinical signs [47] was scored.
The experimental groups are listed in Table 2. As the hemizygous mice had a slight spontaneous phenotype due to PrPC overexpression (see Results), involving gait abnormalities, animals were euthanized following the end-point criteria (body weight, measurable clinical signs, physical appearance, unprovoked behavior and response to external stimuli). Positive TSE diagnosis relied principally on the detection of PrPd (either by immunohistochemistry and/or western blotting or ELISA) and associated spongiform change in the brain parenchyma.
All experiments involving animals were approved by the animal experimentation ethics committee of the Autonomous University of Barcelona (Reference number: 585–3487) in agreement with Article 28, sections a), b), c) and d) of the “Real Decreto 214/1997 de 30 de Julio” and the European Directive 86/609/CEE and the European council Guidelines included in the European Convention for the Protection of Vertebrate Animals used for Experimental and Other Scientific Purposes.
When the clinical end-point criteria were reached, mice were euthanized by an overdose of pentobarbital administered intraperitoneally followed by decapitation. The brain was immediately extracted and placed into 10% phosphate buffered formalin. From each mouse a rostral section of the brain (including olfactory bulbs and frontal cortex), a caudal fraction of the medulla oblongata and the whole spinal cord were kept frozen (for biochemical studies and second passage). Transversal sections of the remaining brain tissue were performed at the level of the piriform cortex, optic chiasm and medulla oblongata. Samples were embedded in paraffin-wax after dehydration through increasing alcohol concentrations and xylene. Four micrometer sections were mounted on glass microscope slides which were stained with hematoxylin and eosin for morphological evaluation. Additional sections were mounted in 3-trietoxysilil-propilamine-coated glass microscope slides for immunohistochemistry.
A pool of all frozen central nervous tissues samples was homogenized 1:10 (W/V) in PBS using closed tubes containing ceramic beads, placed in a ribolyzer (Precess, Bio-Rad) and subsequently analyzed either by western blotting, as described above, or by ELISA (IDEXX, Herdcheck). The latter is a commercial ELISA based on the affinity of misfolded prions to an anionic substrate (termed Seprion). A new threshold was defined to adapt to the higher densitometry readings obtained when working with samples with PrPc overexpression: only samples with a ratio spectrophotometry reading/cutoff over 5 were considered positive.
Immunohistochemistry (IHC) against PrPd was performed as described previously [48]. Briefly, deparaffinized sections were subjected to epitope unmasking treatments: immersed in formic acid and boiled at low pH (6.15) in a pressure cooker and pre-treated with proteinase K. Endogenous peroxidases were blocked by immersion in a 3% H2O2 in methanol. Then, the sections were incubated overnight with anti-PrP MAb 6H4 primary antibody (1:2000, Prionics AG) and subsequently visualised using the DAKO Goat anti-mouse EnVision system (Ref. K400111/0) and 3,3’diaminobenzidine as the chromagen substrate. As a background control, incubation with the primary antibody was omitted.
Histological lesions (i.e. spongiform change) and PrPd immunolabelling were evaluated under a light microscope by a pathologist. A semi-quantitative approach was used to obtain comparable data from the different prions used to challenge mice. Spongiform lesion and PrPd immunolabelling were scored separately. A total of 15 different brain regions were chosen: piriform cortex (Pfc), hippocampus (H), frontal cortex (Fc), parietal cortex (Pc), temporal cortex (Tc), occipital cortex (Oc), thalamus (T), hypothalamus (HT), mesencephalon (M), medulla oblongata (Mobl), cerebellar nuclei (Cm), cerebellar vermis (Cv) and cerebellar cortex (Cc). Scores ranging from (0) absence of spongiosis or immunolabelling: (1) mild, (2) moderate, (3) intense and (4) maximum intensity of lesion or immunolabelling were assigned to each brain area studied (Fig 3). Each area was investigated globally as region for the scoring. Brain profiles were plotted as a function of the anatomical areas which were ordered along the X axis in an attempt to represent the caudo-rostral axis of the brain. This methodology was adapted from a previous study performed on BSE-infected cattle [49]. Graphs were plotted using Microsoft Office 2007 Excel software.
This is the first report of in vivo evidence suggesting that TgRab mice are susceptible to cross species transmission of prion strains. This not only reinforces that rabbits can no longer be considered TSE resistant, but also that there is a possibility they could act as a reservoir for other prion strains. As such, rabbits must be taken into account when determining the epidemiology of several TSE both in relation to the species of origin, especially sympatric ones, but also to potential zoonotic transmission.
In previous studies we demonstrated that rabbits were able to propagate abnormal prions and that these were transmissible to other rabbits. However, this was only one prion strain which was generated de novo in an in vitro PMCA assay in rabbit brain homogenate (a spontaneous rabbit prion strain) and on first passage it had only a very limited attack rate [23]. This new mouse model, which responded in a comparable manner to rabbits when challenged with the same in vitro generated rabbit derived inoculum, has allowed us to evaluate the TgRab’s susceptibility to a number of actual field prions strains from a variety of different species. Although the use of rabbits would have been the most appropriate model there are strong, particularly budgetary, limitations due to the longer lifespan of rabbits and the need to use level 3 biosafety facilities. Thus, a transgenic mouse model overexpressing rabbit PrPC was designed to overcome these limitations and allow us to determine its susceptibility to different prion strains.
No polymorphisms have been described in the PRNP rabbit gene, therefore several mouse transgenic lines were generated expressing rabbit PrPC at different expression levels. The line with the highest possible PrPC expression levels was selected to allow for easier prion propagation capacity but the overexpression was not so high as to generate a spontaneous phenotype at an early age which would preclude the attainment of infectivity/susceptibility data. The hemizygous TgRab line met these criteria with levels of PrPC between 5 to 6 times those present in rabbits. The use of transgenic mice overexpressing ovine PrPC to obtain the infectivity titer of specific prion isolates has been shown to be equivalent to titrations obtained through bioassay in the natural host [50]. Phenotyping of the newly developed prion transgenic model was essential, especially as the levels of PrPc expression present have not been shown to be problematic in other models [41,46]. Eighty percent of the TgRab mice presented with a late onset spontaneous neurological disease phenotype (S3 Fig and S4 Fig) which, fortunately, did not interfere in the interpretation of prion susceptibility results. This allowed us to work with a model that faithfully reproduced the behavior in rabbits with respect to their capability to propagate different prion strains. One cannot exclude the possibility that the presence of spontaneous disease might create a toxic environment in the brain which artificially enhances the transmission of certain strains. Therefore a thorough knowledge of the PrPC overexpression-related changes in uninfected controls was essential to identify the true prion disease status and validity of susceptibility.
Lesion morphology and profiling within the brain and identification of specific PrPd deposition-types allowed unequivocal identification of infected animals (either spontaneous or as a result of an inoculation). Further biochemical detection of the presence of PrPres by western blotting confirmed the ability of morphological techniques to identify an infected animal. Additionally, as PrPC overexpression may mask an incipient infection, second passages are required to confirm if rabbits are totally resistant to those prion isolates to which they initially appeared to be, such as SSBP/1, atypical scrapie or CWD, and these experiments are ongoing.
Once validated the TgRab model was used to evaluate which TSE strains were able to infect the model (Table 2). Previous attempts in rabbits had concluded they were resistant, probably due to incomplete studies and the strong barrier of rabbits to propagate prions [34]. Initially classical cattle BSE, the most relevant field strain, was tested and found to be infectious on first passage with a low attack rate (4/9) and relatively long incubation period (551dpi±10). The strain properties observed in the infected TgRab mice (western blotting, brain lesion and PrPd deposition profiles) were typical of BSE and indistinguishable from those obtained in other BSE murine models [36]. Parallel bioassay studies were conducted with the BSE isolate previously amplified in vitro using rabbit normal brain homogenate as a substrate (BSE-RaPrPres, this inoculum was characterised previously in a TgBov mouse model by our group [36]). These animals showed a 100% (12/12) attack rate and a shortened incubation period (396dpi ±12 vs 551dpi ±10) compared to the cattle BSE inoculated TgRab mice. This reduction already indicated that a transmission barrier between species had been overcome thanks to the in vitro adaptation of the cattle BSE-C to rabbit PrPC, a second passage was performed from that isolate which also showed a 100% attack rate (3/3). Its incubation period was reduced to 322dpi ±12 (mean ± s.e.m.) indicating further host adaptation (S5 Fig).
SheepBSE, derived from BSE-C, infected TgRab mice with a 100% attack rate (9/9), a relatively short incubation time (368±10 dpi) and with lesion and PrPd brain profiles identical to those of BSE-C inoculated mice, suggesting that the same strain was being propagated through both isolates. This enhanced virulence of sheepBSE compared to BSE-C has been previously demonstrated in other experimental scenarios [29,51]. The results obtained with sheep scrapie differed completely as, in agreement with early experiments in rabbits [34], none of the TgRab mice inoculated with SSBP/1 showed any evidence of a prion disease on first passage. However, this result does not preclude that, if further in vivo SSBP/1 passages were to be performed, the transmission barrier would be crossed. As in the case of BSE in the bank vole (Myodes glareolus), where after an initial resistance a bank vole adapted BSE strain was obtained which was highly transmissible [52,53]. Conversely, ME7 and RML scrapie, both murine adapted sheep scrapie strains, infected TgRab mice on first passage and their incubation times, PrPres biochemical profiles, lesion profiles and PrPd deposition patterns were clearly distinguishable from cattle derived strains. Together these data are the first evidence that TgRab mice are not only able to propagate prions but they do it maintaining clearly the different distinguishing strain features (Figs 1, 3 and 4) which strongly suggests that rabbits may also.
It is noteworthy that both ME7 and RML, which originated from serial passages of SSBP/1 in different rodents [54,55], directly propagated in TgRab mice on first passage. Conversely, SSBP/1 did not infect TgRab mice on first passage. The murine adapted prion strains behaved differently to their parent strain, possibly because passage through rodents had selected for a strain capable of crossing the rodent species barriers. The situation is analogous to CWD which will infect hamsters after initial passage through ferrets [9]. In the present work, previous adaptation of scrapie to rodents, possibly resulting in a higher sequence identity in some specific and crucial PrP regions with rabbits compared to sheep, allowed rodent adapted scrapie prions to misfold rabbit PrPC. In previous studies ME7 did not infect rabbits after 4–5 years of incubation, with the exception of a single inconclusive case [23,34]. This result is difficult to extrapolate since we are discussing different species, of differing lifespans and with a species barrier between them. The PrPC overexpression in TgRab may have allowed ME7 to propagate more efficiently than in rabbits which suggests that if the original rabbit experiments had been performed over the maximum lifespan of rabbits ME7 may have propagated on first passage also.
Once BSE in cattle has been virtually controlled, CWD in cervids is the animal prion disease with the most repercussions, at least in the North American continent. The uncertainty of its transmissibility to humans [56] and its unique ability to spread through the free ranging cervid population make its study highly relevant with respect to transmissibility to other species. Moreover CWD prions are known to be shed and are highly persistent in the environment. Rabbits are a sympatric species with cervids. Even though CWD has been shown to transmit on first passage to many species including sheep, cattle [57], squirrel monkeys [58], cats [59], hamsters [60], ferrets [9], mink [61], bank voles and deer mice (Genus Peromyscus) [62] its transmissibility efficiency is relatively low with very long incubation periods and low attack rates. For instance, wild type mice could not be readily infected, so tga20 mice overexpressing murine PrPC were required to prove susceptibility to CWD [63] or required a second passage [64]. Another example is the transmission of CWD to cats, which required an incubation period of longer than 4 years [59]. The present study showed CWD was not able to infect TgRab on first passage (0/12). Further experiments are required to confirm the resistance of rabbits to CWD including a blind second passage and bioassays with CWD previously passaged in other species, especially rodents [9]. This will rule out an analogous situation as the one observed in this paper with sheep scrapie whereby SSBP/1 does not transmit to TgRab but murine passaged counterparts, ME7 and RML, do.
With respect to the atypical prion strains of purported spontaneous origin [18,65,66], BSE-L infected TgRab mice on first passage and, although the attack rate was low (3/11), they had the shortest incubation period observed in this model so far (221dpi for the first animal to die, mean 280±26dpi). The lesion and PrPd deposition brain profiles differed considerably from those of BSE-C. None of the TgRab mice inoculated with atypical scrapie showed evidence of a TSE with the exception of one animal, euthanized at 742 dpi which, even though no histological lesions nor PrPd deposits were present suggestive of infection, it was positive by PrPd ELISA. This result could not be confirmed by western blotting. However, this ELISA detects PrPd through its affinity to an anionic ligand not due to its resistance to protease K so we cannot rule out this single mouse was positive. A second passage is ongoing which will determine the result.
Initial in vitro experiments predicted that BSE as well as SSBP/1 and CWD isolates were able to missfold rabbit PrPC. However, a discrepancy was found with the bioassay results since neither SSBP/1 nor CWD infected TgRab mice on first passage. Several saPMCA rounds were needed in order to amplify the different isolates, varying in number depending of each strain. Thus, it is not surprising that on first passage some of the isolates do not transmit.
Besides the PRNP sequence, another component of the transmission barrier is the genetic background in which each PrPC is contained. This has been demonstrated by infectivity studies showing BSE propagated more efficiently in RIII mice than C57/Black mice, two mice strains of the same species with the same PRNP gene [67]. Or when the genetic background (i.e. passage through different inbred mouse lines) determined not only the incubation period but also the propagation of two biochemically different BSE-derived strains [68]. For these reasons the belief that rabbits were resistant to prion infection was not only attributed to the rabbit PrPC sequence but also to its genetic background. To study whether the genetic background of rabbits was responsible for the apparent prion resistance, Houdebine’s group generated transgenic rabbits expressing an ovine PrPC which was known to easily misfold. Upon inoculation with ovine prion strains these rabbits succumbed to prion disease further proving that rabbits are not resistant to prions (results published paired with this article) and that the genetic background is not a limiting factor [37].
The differential susceptibility observed between actual rabbits and the transgenic model presented here can be explained by the higher PrPC expression levels of TgRab mice. Lower expression mouse lines would probably only be susceptible on first passage to strains previously adapted to rabbit PrPC as occurs with rabbits. It has taken more than three decades to finally dismiss the rabbit as a prion resistant species. We believe that the studies presented here confirm that in vitro studies are of great help in interpreting in vivo results, leave no room for misinterpretation, and that it can be ascertained that rabbits, and probably all other mammal species [21], are susceptible to infection by specific prion strains. The prion strain and its species of origin determine the extent of susceptibility, but neither rabbit PRNP nor their genetic background suggest they are resistant to prion propagation. Unfortunately, as with other mammals, the exact molecular mechanisms governing the capricious choice of strains that can be propagated in a certain species is still unknown.
In light of our results, especially susceptibility to spontaneous cattle prions (BSE-L), the restrictions on rabbits being fed ruminant protein should be maintained sine die to minimize the chances of any prion strain transmitting to rabbits.
|
10.1371/journal.pgen.1005518 | Autophosphorylation of the Bacterial Tyrosine-Kinase CpsD Connects Capsule Synthesis with the Cell Cycle in Streptococcus pneumoniae | Bacterial capsular polysaccharides (CPS) are produced by a multi-protein membrane complex, in which a particular type of tyrosine-autokinases named BY-kinases, regulate their polymerization and export. However, our understanding of the role of BY-kinases in these processes remains incomplete. In the human pathogen Streptococcus pneumoniae, the BY-kinase CpsD localizes at the division site and participates in the proper assembly of the capsule. In this study, we show that the cytoplasmic C-terminal end of the transmembrane protein CpsC is required for CpsD autophosphorylation and localization at mid-cell. Importantly, we demonstrate that the CpsC/CpsD complex captures the polysaccharide polymerase CpsH at the division site. Together with the finding that capsule is not produced at the division site in cpsD and cpsC mutants, these data show that CPS production occurs exclusively at mid-cell and is tightly dependent on CpsD interaction with CpsC. Next, we have analyzed the impact of CpsD phosphorylation on CPS production. We show that dephosphorylation of CpsD induces defective capsule production at the septum together with aberrant cell elongation and nucleoid defects. We observe that the cell division protein FtsZ assembles and localizes properly although cell constriction is impaired. DAPI staining together with localization of the histone-like protein HlpA further show that chromosome replication and/or segregation is defective suggesting that CpsD autophosphorylation interferes with these processes thus resulting in cell constriction defects and cell elongation. We show that CpsD shares structural homology with ParA-like ATPases and that it interacts with the chromosome partitioning protein ParB. Total internal reflection fluorescence microscopy imaging demonstrates that CpsD phosphorylation modulates the mobility of ParB. These data support a model in which phosphorylation of CpsD acts as a signaling system coordinating CPS synthesis with chromosome segregation to ensure that daughter cells are properly wrapped in CPS.
| Bacteria utilize a multi-protein membrane complex to synthesize and export the polysaccharide capsule that conceals and covers the cell. In bacterial pathogens, the capsule protects the cell form opsonophagocytosis and complement-mediated killing. The mechanisms allowing the bacterial cell to maintain this protective capsule during cell growth and division remain unknown. The capsule assembly machinery encompasses a particular type of tyrosine-kinases found only in bacteria, which are called BY-kinases. These kinases are involved in the regulation of several cellular functions including polysaccharide capsule production. Studying the role of BY-kinase represents thus an interesting approach to decipher the mechanisms of capsule synthesis and export. Here, we study the role of the BY-kinase CpsD in the human pathogen Streptococcus pneumoniae. We show that CpsD plays a dual function in the pneumococcus. Indeed, CpsD captures the capsule assembly machinery at the site of division, but we also show that CpsD coordinates capsule production with the cell cycle by interacting with the chromosome segregation system. These features provide a simple mechanism to cover the complete surface of the pneumococcal daughter cells. This finding further opens a new view of the function of BY-kinases in the bacterial cell notably in localizing protein complexes in subcellular regions over the cell cycle.
| Streptococcus pneumoniae is a Gram-positive bacterium usually found as a commensal in healthy adults and children [1]. It does however have the potential to become pathogenic and is a frequent cause of community-acquired diseases. S. pneumoniae is associated with a variety of infections that can range in severity from otitis media to pneumonia or meningitis [2]. Despite the availability of antibiotics, pneumococcal infections still have high mortality rates and vaccine efficiency drops over time as new and infectious non-vaccine covered serotypes are emerging in clinical isolates [3]. Pneumococcal virulence is strictly dependent on the capsular polysaccharide (CPS) production: non-encapsulated mutants of clinical pneumococcal isolates are non-virulent [4]. The capsule plays a major role in both colonization and persistence of S. pneumoniae in the infected host due to its ability to form a shield that prevents antibodies and complement components from interacting with their receptors on the host phagocytic cells [5, 6].
In all serotypes, the cps operon includes serotype-specific genes, encoding enzymes required for the synthesis of specific sugar components, as well as conserved genes encoding proteins essential for capsular synthesis and export (Fig 1A) [7]. Export of the capsule across the plasma membrane occurs by a Wzy-dependent polymerization pathway, analogous to Group 1 CPS biosynthesis in Escherichia coli [8, 9] (Fig 1B). The 5’ region of the locus encodes the cpsA, cpsB, cpsC and cpsD genes, (also known as wzg, wzh, wzd and wze) (Fig 1A). CpsA was shown to interact with the pyrophosphoryl-lipid carrier of the polysaccharide precursor and is proposed to attach capsular polysaccharide to cell wall peptidoglycan [10]. cpsB, cpsC and cpsD constitute a phosphoregulatory system that controls the polysaccharide assembly machinery encompassing a glycosyl-transferase (CpsE), a flippase (CpsJ) and a polymerase (CpsH) (Fig 1B) [9]. CpsB is a metal-dependent phosphotyrosine-protein phosphatase of the PHP family [11] whereas CpsC and CpsD constitute a so-called BY-kinase, a particular type of tyrosine-autokinase, which shares no resemblance with eukaryotic tyrosine-kinase and is conserved among most bacterial phyla [12–14].
BY-kinases consist of two main structural domains: an N-terminal extracellular domain flanked by two transmembrane helices and a cytoplasmic C-terminal domain, harboring the kinase activity [15]. In Firmicutes, these domains are encoded by two successive genes, and are therefore present as separate polypeptide chains, one cytoplasmic and the other in the membrane (Fig 1C). The two polypeptides need to interact to form an active BY-kinase [16]. The crystal structure of the BY-kinase CapB from Staphylococcus aureus showed that the cytoplasmic C-terminal end of the transmembrane modulator CapA is required for the activation of the cytoplasmic kinase CapB [17]. More precisely, the C-terminal extremity of CapA forms a αA-ßA motif complementing the catalytic site of CapB and stabilizing the ATP molecule. The cytoplasmic domain of BY-kinases is able to autophosphorylate on several tyrosines forming a C-terminal tyrosine cluster motif (Fig 1C) [18, 19]. Although the detailed mechanisms by which BY-kinases promote CPS synthesis and export remain elusive, it has been proposed that cycling between phosphorylated and non-phosphorylated forms of the BY-kinase, regulated by the cognate phosphotyrosine-phosphatase, is required for proper synthesis and export of the polysaccharide polymer [20–23].
The single BY-kinase produced by most of the 93 S. pneumoniae serotypes [9] comprises the transmembrane modulator CpsC and the cytoplasmic kinase domain CpsD [12, 13]. Several studies reported that autophosphorylation of CpsD in the tyrosine cluster negatively regulates CPS production [18, 24]. Moreover, evidence was provided that CpsD tyrosine-kinase activity influences capsule production and modulates invasive pneumococcal disease [12, 25]. Interestingly, it was recently shown that CpsD and CpsC both localize at the division site in the serotype 14 strain ATCC6314 [13]. In addition, capsule is absent from the division site and detected only at the old cell halves in the absence of either CpsD or CpsC. Collectively, it is hypothesized that the CPS assembly machinery would adopt two distinct localizations: one around the cell and one at the division site captured by CpsC and CpsD for septal CPS production. In the latter, CpsC and CpsD would either act as activators of CPS export or function as the exporter [13].
Here, we have investigated the role of CpsD activation by CpsC as well as the impact of CpsD autophosphorylation on capsule production in the well-studied serotype 2 strain D39 [26]. We first show that the C-terminal cytoplasmic end of CpsC is required for CpsD autophosphorylation and localization at the division septum. Analysis of CPS production together with imaging of the polysaccharide polymerase CpsH localization demonstrates that CPS are exclusively produced at the division septum in WT cells and challenges the two-machine model for CPS assembly. Strikingly, we also observe that cells producing a non-phosphorylatable variant of CpsD display defective capsule production at the septum together with aberrant elongated shape with multiple non-constricted septa, nucleoid defects and reduced dynamics of the chromosome segregation protein ParB. In line with a role of CpsC and CpsD in controlling the pneumococcal cell cycle, it was shown that BY-kinases have homology to the large P-loop NTPase superfamily [27] that includes ParA-like proteins, which are involved in chromosome segregation by interacting with ParB [28]. Molecular modeling confirms that CpsD displays structural similarities with ParA. Interestingly, we found that CpsD interacts with ParB in vitro and that the stability of the CpsD/ParB complex is modulated by CpsD phosphorylation in vivo. These observations show that CPS production is tightly connected with the cell cycle and support a model wherein crosstalk between CpsD and ParB, modulated by CpsD autophosphorylation, signals the status of CPS production to the proteins in charge of chromosome segregation, thus ensuring coordination between encapsulation and cell division.
Structural studies of BY-kinases have established that the C-terminal peptide of the transmembrane modulator specifically interacts and activates the cytoplasmic catalytic domain [17]. We first tested whether the C-terminal end of CpsC (CpsC-Cter) is required for CpsD autophosphorylation as it is the case for BY-kinases from other Firmicutes [16, 29, 30]. To do that, we first analyzed by yeast two-hybrid assays the ability of a derivative of CpsC lacking the C-terminal 30 amino acids (CpsC-ΔCter) to interact with CpsD. We observed that full-length CpsC interacts efficiently with CpsD while this interaction was abolished in the absence of CpsC-Cter (Fig 2A). As expected for a BY-kinase [17], CpsD interacted with itself. Next, we constructed a nonpolar markerless mutant strain expressing CpsC-ΔCter (cpsC-ΔCter strain) and analyzed CpsD autophosphorylation using anti-phosphotyrosine antibodies (Fig 2B). Non-polarity of the deletion of cpsC-Cter was confirmed by analyzing the expression of CpsD and CpsH fused to GFP (see below) in WT and cpsC-ΔCter strains (S1A and S2 Figs) and partial restoration of capsule production in the cpsC-ΔCter mutant carrying a copy of cpsC at the ectopic amiF/treR locus under the control of the maltose inducible promoter PM [31] (S1B and S1C Fig). No phosphorylation signal was detected for CpsD in the cpsC-ΔCter mutant (Fig 2B). As controls, CpsD was efficiently phosphorylated in the wild-type strain whereas no phosphorylation signal was detected in a mutant deficient for CpsD (ΔcpsD) or a strain producing CpsD mutated on the three tyrosines of its C-terminal tyrosine cluster motif. In addition, CpsD phosphorylation was partially restored in the cpsC-ΔCter strain complemented with PM-cpsC (Fig 2B). The partial restoration of CpsD phosphorylation and capsule synthesis observed for the cpsC-ΔCter PM-cpsC strain suggests that production of CpsC-ΔCter interferes with the ability of native CpsC to interact with CpsD and/or to function in the CPS assembly machinery. Next, we analyzed the effect of the CpsC C-terminal deletion (CpsC-ΔCter) on CpsD localization. For this purpose, we constructed a C-terminal monomeric GFP fusion to CpsD. The CpsD-GFP fusion is stable and functional since cells grew normally and displayed normal CPS production patterns (S3 and S4 Figs) [24]. As previously shown in a serotype 14 strain [13], CpsD-GFP also localized at midcell in serotype 2 D39 cells (Fig 2C). However, CpsD delocalized to the cytoplasm in cpsC-ΔCter cells whereas CpsC-ΔCter-GFP still localized at midcell (Fig 2C). Altogether, these observations show that the C-terminal end of CpsC is required to position CpsD at the division site and trigger its autophosphorylation.
Our observations prompted us to analyze CPS synthesis in the cpsC-ΔCter mutant. For that, we used anti-serotype 2 capsule antibodies and localized CPS by immunofluorescence microscopy. Consistent with observations reported by Henriques and co-workers [13], CPS were not produced at the division site but only at the old cell halves in the ΔcpsD mutant while CPS were detected over the entire surface of wild-type cells (Fig 3A). We also observed that CPS were absent from the division septa in the cpsC-ΔCter mutant (Fig 3A). These data show that the absence of CpsC C-terminus, and consequently CpsD phosphorylation and localization at midcell, alters the localization and/or the activity of the capsule assembly machinery. To test these hypotheses, we quantified the CPS fluorescence signal in living cells and immunodetected the total fraction of CPS produced by our mutants using anti-type 2 capsule antibodies. We observed a striking CPS production and polymerization defect in both cpsC-ΔCter and ΔcpsD mutants compared to the WT strain (Fig 3B and 3C). This defect led to the accumulation of low molecular weight polysaccharides (Fig 3C). Next, we used the polymerase CpsH as a marker to localize the capsule assembly machinery. CpsH is a predicted membrane protein with both N- and C-terminal ends located outside the cell [9]. It was shown that superfolder GFP (sfGFP), in certain fusions, can fluoresce in the extracellular milieu [33]. We thus constructed a strain expressing CpsH fused to sfGFP at its C-terminus. WT cells producing CpsH-sfGFP as the only source of CpsH from their endogenous chromosomal locus grew and produced capsule similarly to WT cells attesting that the fusion CpsH-sfGFP is functional (S3 Fig). As shown in Fig 3D, we found that CpsH-sfGFP localized exclusively at the division septum in WT cells. Interestingly, CpsH partially delocalized in both cpsC-ΔCter and ΔcpsD mutants, and the signals were not exclusively present at the division septum (Fig 3D). CpsH-sfGFP was stable and produced at similar amount in WT and cpsC-ΔCter and ΔcpsD mutants (S2 Fig). These data suggest that the capsule assembly machinery localizes at the division site to assemble CPS. In addition, the localization and the activity of the capsule assembly machinery are both dependent on CpsD localization at the division site.
To better understand the role of CpsD phosphorylation at the division site in CPS production, we constructed two mutant strains expressing either non-phosphorylated CpsD (cpsD-3YF) or its phosphomimetic form (cpsD-3YE). For that, each of the 3 tyrosines of the tyrosine cluster of CpsD was substituted either for phenylalanine or glutamic acid. Then, we analyzed these strains for capsule production and CpsH localization. As shown in Fig 3A and 3B, cpsD-3YE cells displayed CPS localization pattern and quantification indistinguishable from that of WT cells even if the capsular halo around WT cells suggested that CPS could be in less tighter association with the cells than in cpsD-3YE cells. CpsH-sfGFP was also found to localize properly at the septum in cpsD-3YE cells (Fig 3D). However, by measuring cell lengths, cpsD-3YE cells were significantly shorter (1.88 μm +/- 0.33) than WT cells (2.02 μm +/- 0.32, p<0.0001, Mann Whitney rank sum test) (Fig 4). Furthermore, cpsD-3YF cells displayed capsule mainly at the old cell poles as observed in ΔcpsD cells (Fig 3A). Strikingly, 26.6% of cpsD-3YF cells possessed an elongated cell shape (> 3 μm). This aberrant cell length of > 3 μm was observed in less than 1.4% of WT or cpsD-3YE cells (Fig 4). Furthermore, CpsH-sfGFP partially delocalized and formed several foci around the cell in the cpsD-3YF mutant although CpsH-sfGFP was stable and produced at similar amount in WT and cpsD-3YF and cpsD-3YE mutants (S2 Fig). In agreement with these observations, quantification of CPS fluorescence signal and western immunoblotting using anti-serotype 2 capsule antibodies showed that CPS production was hampered in the cpsD-3YF mutant compared to that of WT and cpsD-3YE cells (Fig 3B and 3C). However, CPS polymerization remained clearly more effective than in cpsC-ΔCter and ΔcpsD mutants suggesting that beyond CpsD phosphorylation, deletion of cpsD or cpsC-Cter further alters the global functioning of the CPS assembly machinery (Fig 3B and 3C). These data show that permanent dephosphorylation of CpsD is detrimental for localizing the capsule synthesis machinery at the division site and causes problems with cell division, thus suggesting a link between defective capsule synthesis and aberrant elongation of cells.
We hypothesized that the absence of CpsD phosphorylation could be detrimental for its localization at midcell. To test whether the phosphorylation state of CpsD influences its localization, we constructed strains producing C-terminal GFP fusions to either CpsD-3YE (cpsD-3YE-GFP) or CpsD-3YF (cpsD-3YF-GFP). Both types of tyrosine substitutions did not affect the stability and the production of the CpsD-GFP fusion (S4 Fig). Fluorescence microscopy indicated that these mutations do not affect septal localization (Fig 5A). Therefore, we questioned why expression of cpsD-3YF induces cell elongation whereas cells expressing CpsC-ΔCter, in which CpsD autophosphorylation is prevented (Fig 2B), are not elongated (Figs 2B and 4). To reconcile this apparent contradiction, we constructed mutant strains expressing cpsC-ΔCter and cpsD-3YF fused or not to gfp (cpsC-ΔCter-cpsD-3YF-gfp and cpsC-ΔCter-cpsD-3YF). Consistent with our observations presented in Fig 2C, CpsD-3YF-GFP delocalized in the cytoplasm in the absence of the cytoplasmic C-terminal end of CpsC (Fig 5A). However, cpsC-ΔCter-cpsD-3YF cells did not display an aberrant elongated cell shape phenotype anymore indicating that the deletion of CpsC-Cter suppresses cell elongation of the cpsD-3YF mutant (Fig 4). Together, these data suggest that the presence of non-phosphorylated CpsD at the division site partially inhibits cell division, leading to an elongated cell shape phenotype.
To better analyze the ultrastructure of elongated cpsD-3YF cells, we examined these cells by transmission-electron microscopy (TEM). As controls, WT and cpsD-3YE cells were found to possess the characteristic ovoid cell shape (Figs 5B and S5). By contrast, examination of cpsD-3YF mutant confirmed the severe disturbed cell morphology observed by phase-contrast microscopy (Fig 5B). In addition, these images showed the presence of multiple septal initiations on each side of the long cell axis, indicative of impaired cell constriction (Fig 5B). In addition, and in agreement with CPS immunolabelling (Fig 3A), CPS were primarily detected at the pole in cpsD-3YF cells observed by electron microscopy (Fig 5B).
To further investigate the division defects of the cpsD-3YF mutant strain, we analyzed the localization of the major cell division protein FtsZ that forms contractile rings at mid-cell [34]. As expected, FtsZ-GFP [35] localized at mid-cell in WT cells as well as in cpsD-3YE cells (Fig 5C). In elongated cpsD-3YF cells, FtsZ-GFP fluorescence showed a ladder localization pattern indicating that it is properly recruited to each non-constricted septum (Fig 5C). This observation shows that the division defects of cpsD-3YF cells are not due to altered Z-ring assembly at the division site.
S. pneumoniae does not contain homologs of proteins involved in nucleoid occlusion, and it was shown that the cell division machinery assembles over the nucleoid and that septation occurs just after chromosomes splitting, indicating that DNA itself can act as a physical barrier for cell division [36]. In line with this idea, mutants affected in chromosome segregation are often elongated [36, 37]. To test whether the non-phosphorylated CpsD mutant induces chromosome segregation defects that could lead to the elongated cell phenotype, we analyzed nucleoid morphology using DAPI staining. As shown in Fig 6A, the nucleoid is properly condensed and segregated in WT and cpsD-3YE cells. However, DAPI staining revealed abnormal elongated nucleoids in cpsD-3YF cells (Fig 6A). To quantify these nucleoid defects, we used the histone-like HlpA-RFP fusion as a chromosome marker in live cells [36]. As expected, 93.7% of WT cells displayed normal, well-condensed nucleoid(s) (Fig 6B and 6C). The same observation (95.7% normal nucleoids) was made in the cpsD-3YE mutant confirming that the expression of CpsD-3YE as the sole source of CpsD does not impact the cell cycle. The situation differed in cpsD-3YF cells in which only 69.8% of cells displayed well condensed nucleoid(s). In addition, 3.2% of cpsD-3YF cells were either devoid of nucleoid or presented an uncondensed nucleoid (Fig 6B and 6C). These defects are significantly more abundant than in WT and cpsD-3YE cells in which they are observed in only 0.6% and 0.7% of cells, respectively. More importantly, we found that nucleoids are elongated and abnormally shaped in 27.0% of cpsD-3YF cells compared to 5.7% and 3.6% of WT and cpsD-3YE cells, respectively (Fig 6B and 6C). Altogether, these observations show that chromosome replication and/or segregation is impaired in the cpsD-3YF mutant, suggesting that expression of non-phosphorylated CpsD at the septum could interfere with these processes.
To test whether initiation of DNA-replication is perturbed when CpsD is not phosphorylated, we performed marker frequency analysis using quantitative real-time PCR [38]. Determining the origin-to-terminus ratios showed no significant difference between WT and cpsD-3YF cells (S6 Fig). Interestingly, the chimera CpsC/D composed by CpsD fused to the C-terminal end of CpsC exhibits 20% identity and 52.4% similarity to the Bacillus subtilis ParA protein Soj (Fig 7A) and molecular modeling using the calculated structure of the BY-kinase chimera CapA1/B2 from Staphylococcus aureus as template [17] suggests that the structure of the two proteins is similar (Fig 7B–7D). The accurate role of ParA in chromosome segregation remains unclear but it is proposed to interact with and assist the chromosome partitioning protein ParB during segregation of chromosomes to the daughter cells [28]. ParB specifically binds to centromere-like DNA sequences named parS that are located near the origin of replication [39, 40]. Therefore, to evaluate whether CpsD phosphorylation impact ParB dynamics in live cells, we constructed strains (WT, cpsD-3YE and cpsD-3YF) producing the functional fusion ParB-sfGFP [40] and we analyzed ParB dynamics by TIRF (Total Internal Reflection Fluorescence) microscopy, a technique in which only a thin section of the cell (approx. 100 nm) is excited, thus providing high axial resolution. ParB-sfGFP foci were categorized as stationary if they remained in TIRF focus during the 40 sec time-span of the experiment, while ParB-sfGFP foci appearing, disappearing or splitting were categorized as dynamic (Fig 8A and 8B). We observed that 66.9 +/- 2.8% of ParB-sfGFP were stationary in WT cells (Fig 8C). Interestingly, this fraction increased up to 80.6 +/- 2.4% in cpsD-3YF cells whereas it decreased down to 56.4 +/- 2.4% in cpsD-3YE cells (Fig 8C). These experiments thus suggest that CpsD phosphorylation could play a role in chromosome segregation by modulating ParB mobility.
The latter observation, combined with the fact that CpsD is homologous to ParA-type proteins, prompted us to analyze the timing of CpsD and ParB localization in living cells. To do so, we constructed a double-labeled strain expressing both ParB-sfGFP and CpsD-RFP and performed time-lapse-microcopy (Fig 9A). As shown in Fig 9A and S1 Movie, and consistent with previous observations [40], ParB-sfGFP localizes as single foci at cell equators (future division site). In these predivisional cells, CpsD-RFP did not co-localize with ParB and was exclusively detected at the current division site (Fig 9A). As the cell cycle progresses, CpsD-RFP displayed a dual localization pattern; the fusion protein remained at the division site until cell constriction was completed, but also localized at the future-division site where it co-localizes for a short time with ParB (arrows in Fig 9A). Then, as the new cell cycle began, the oriC-localized ParB-sfGFP moved, due to chromosome segregation, toward the daughter cell equator while CpsD-RFP remained at the division site. This transient co-localization of CpsD with ParB suggests that the two proteins might interact physically. To investigate this, we applied the approach commonly used for the purification of BY-kinases from Firmicutes: we purified the active and fully autophosphorylated CpsC/D chimera in which the cytoplasmic C-terminal end of CpsC is fused to the N-terminal end of CpsD (Fig 1C and S7A and S7B Fig). As shown in S7A Fig, ParB and the non-phosphorylatable CpsC/D-YF chimera were also purified to homogeneity. Then, microscale thermophoresis was used to determine the binding affinity of the ParB-CpsC/D complex. While no binding of ParB could be detected with the BSA (Bovine Serum Albumin) control (S7C Fig), an affinity constant of 7 ± 0.8 μM (n = 5) was obtained for the ParB-CpsC/D complex (Fig 9B). A two-fold higher affinity constant was calculated for the ParB-CpsC/D-YF complex (14 ± 0.9 μM (n = 5) suggesting that CpsD autophosphorylation influenced the interaction between ParB and CpsC/D. Then, we investigated the interaction between ParB and CpsD in vivo. For that, we constructed strains producing ParB-sfGFP together with either CpsD, CpsD-3YE or CpsD-3YF fused to a 6his-tag. We checked that fusions were produced at similar levels (S7D Fig). After immunoprecipitation of ParB-sfGFP, the presence of CpsD variants in the immunoprecipitated fractions was probed by Western blot using anti-His antibodies (Fig 9C). Western blots were also performed using anti-GFP antibodies to confirm that similar amounts of ParB-sfGFP were loaded. We observed that all CpsD constructs co-immunoprecipitated with ParB-sfGFP. Yet, we observed that immunoprecipitation of CpsD-3YF by ParB-sfGFP was less efficient than that of CpsD or CpsD-3YE. Altogether, these observations indicate that CpsD interact with ParB and that this interaction is modulated by CpsD phosphorylation.
Many reports have demonstrated that BY-kinases are key regulators of extracellular polysaccharides biosynthesis and export [22, 24, 41–44]. The current model proposes that they would function as co-polymerases assisting the polymerase of the capsule assembly machinery [8]. However, the detailed mechanism by which BY-kinases control this machinery remains unknown. It is proposed that BY-kinases could serve as a molecular scaffold for the other proteins of the machinery. Phosphorylation / dephosphorylation of BY-kinases would trigger a conformational switch affecting the functioning of the other protein components of the polysaccharide assembly machinery [17]. Alternatively, BY-kinases could form a channel across the cytoplasmic membrane, large and hydrophilic enough to allow the polysaccharide polymer to cross the membrane. Indeed, BY-kinases can form a ring-shaped octamer that upon autophosphorylation dissociates to monomers [21]. Recently, Henriques and co-workers have shown that the BY-kinase CpsD of S. pneumoniae ATCC6314 localizes at the bacterial division septum suggesting that BY-kinases might also function as spatial regulators of capsular polysaccharide biosynthesis [13]. They propose that in the pneumococcus, two types of CPS assembly machinery would allow to wrap the cell in CPS. The membrane machinery without CpsC and CpsD would allow production of CPS around the cell while the septal machinery, associated with CpsC and CpsD, would specifically produce CPS at the septum. Our finding that CpsH localizes exclusively at the septum in WT cells allows us to refine this model and strongly suggests that CPS are produced only at the division septum by a single machinery (Fig 3D). This observation is further supported by the septal localization of CpsJ (S8 Fig). The capsule detected at the old cell poles in the ΔcpsD mutant could thus reflect accumulation of basal amount of immature capsule produced by the defective and mislocalized CPS assembly machinery. To our knowledge, this represents the first study determining the cellular site of polysaccharide synthesis and export.
In this context, what is the contribution of CpsC and CpsD to CPS production at the division septum? Our data show that the C-terminal and cytoplasmic end of CpsC is required for the interaction between CpsC and CpsD and consequently, CpsD autophosphorylation and localization at the division septum (Fig 2). These data are in agreement with the existence of a conserved activation mechanism of BY-kinase autophosphorylation. We observe that both CpsD and CpsH delocalize in cells expressing CpsC devoid of its C-terminal cytoplasmic end (cpsC-ΔCter). Yet, CpsH also delocalizes in cells deficient for cpsD (Fig 3D). Collectively, these data are consistent with a sequential interaction model in which CpsC captures CpsD at the division septum allowing subsequent localization of CpsH at the septum. The absence of capsule at the division septum together with impaired polymerization of CPS in cpsC-ΔCter and ΔcpsD cells (Fig 3A–3C) support this model. One should, however, note that CPS are still detected at the old cell pole in cpsC-ΔCter and ΔcpsD cells (Fig 3A). This shows that the capsule assembly machinery is still able to export some polysaccharides at the surface of cells in these genetic backgrounds. This observation implies that some polysaccharide subunits produced in the cytoplasm are flipped across the membrane by the flippase CpsJ but not properly polymerized by CpsH. Supporting this, cells deficient for cpsJ do not produce CPS [45]. This indicates that CpsC and CpsD are unlikely to function themselves as the exporter as previously suggested by Henriques and co-workers and our previous work [13, 21]. More likely, CpsC together with CpsD capture the CPS assembly machinery at the division site and trigger CPS export and polymerization by the flippase CpsJ and the polymerase CpsH, respectively.
One may also wonder how CPS produced at the division site covers the whole cell. An interesting possibility lies in the mode of cell division and elongation of the pneumococcus. Contrary to rod-shaped bacteria, the pneumococcus does not perform lateral synthesis of peptidoglycan. Peptidoglycan is produced at mid-cell and serves both for cell elongation and division, resulting in its characteristic ovoid morphology [46]. More precisely, it was demonstrated that ongoing peptidoglycan synthesis pushes the previous synthesized peptidoglycan leading mechanically to cell elongation and then formation of the new cell pole [47]. Considering that CpsA localizes at midcell and likely ligates CPS to the cell wall [10, 48, 49], CPS concurrently produced with peptidoglycan at mid-cell could be shuttled by peptidoglycan as the cell elongates and constricts. Because CPS are required to avoid pneumococcus recognition by the host complement and immune systems, this would provide a very simple mechanism to conceal and cover the complete surface of the cell. To test this hypothesis, the generation of fluorescent CPS precursors might be useful, similar to the approaches used to image peptidoglycan synthesis [50].
An important feature of BY-kinases regulation of CPS synthesis and export is their phosphorylation on several tyrosine residues grouped in a C-terminal motif termed “tyrosine cluster” [19]. Here, we observe that the cpsD-3YF strain displays slightly reduced production of CPS (Fig 3C). This finding is consistent with previous observations made in E. coli and S. pneumoniae [20, 22, 23, 25]. In addition, we observe that the localization of CpsD remains unchanged whatever its phosphorylation state (Fig 5A). This suggests that CpsD autophosphorylation is not crucial for the activity of CPS assembly machinery per se. However, and strikingly, expression of CpsD-3YF hinders CPS production at mid-cell, alter CpsH septal localization and results in an elongated cell phenotype (Figs 3 and 4). Altogether, these data suggest that CpsD phosphorylation would coordinate CPS production with the pneumococcal cell cycle. In this context, the presence of permanently non-phosphorylated CpsD (CpsD-3YF) at the division site would interfere with cell constriction, leading to elongated cells. Our finding that the non-septal localization of CpsD-3YF (due to deletion of the CpsC C-terminus) (Fig 5A) suppresses the elongated phenotype of the cpsD-3YF mutant cells (Fig 4), together with the presence of non-constricted septa in cpsD-3YF cells (Fig 5B), support this hypothesis.
Our data also show that the cpsD-3YF elongated cells display nucleoid defects (Fig 6). Interestingly, BY-kinases have been grouped with ParA and MinD proteins in the same protein superfamily on the basis of sequence similarity-based clustering [27]. ParA and MinD are involved in chromosome segregation and positioning of the Z-ring at mid-cell, respectively [51]. The first structure of the BY-kinase CapB from S. aureus was solved by molecular replacement using the structure of MinD from Pyrococcus horikoshii, and the two structures are highly similar [17]. Interestingly, the pneumococcus lacks ParA and MinD proteins [51]. On this basis, it is tempting to speculate that CpsD could share some functional properties with either ParA or MinD. Positioning of the Z-ring at mid-cell has been recently elucidated in the pneumococcus and it relies on the protein MapZ [47, 52]. Together with our finding that Z-rings are recruited to each non-constricted septum of cpsD-3YF cells, it is unlikely that CpsD contributes to division site selection as is the case for MinD (Fig 5C). However, it would be interesting to see whether the CpsD-phosphorylation mutants (YE/YF), by affecting CPS production, also cause an imbalance in PG precursor levels and hence cell division defects. By contrast, the structural similarity between CpsD and ParA proteins (Fig 7) suggests that CpsD could behave as a ParA-like protein even if we did not detect any DNA-binding for CpsD (S9 Fig). The accurate role of ParA in chromosome segregation remains unclear but it is proposed to interact with and assist ParB during chromosome segregation [28]. Unencapsulated strains of pneumococcus do not possess CpsD, thus suggesting that CpsD is unlikely to represent an authentic ParA protein involved in chromosome segregation. However, Δcps unencapsulated cells are also significantly shorter (Fig 4), suggesting that the production of the large CPS structure may require additional checkpoints to ensure correct chromosome segregation and cell division. Several studies have reported that ParA-like proteins are actually involved in protein localization in connection with the cell cycle [53]. For instance, ParC and PpfA facilitate polar localization and segregation of chemotaxis proteins in Vibrio cholerae and Rhodobacter sphaeroides, respectively [54, 55]. Another example is the ParA-like protein FlhG that is required for the polar localization of flagellar assembly factors [56]. Considering the CPS and nucleoid defects of cpsD-3YF cells, together with the timing of localization of CpsD and ParB and the ability of CpsD to interact with ParB (Figs 6 and 9), it is tempting to speculate that a crosstalk between the ParA-like CpsD protein and ParB could coordinate septal CPS production with the cell cycle. Our finding that CpsD phosphorylation modulates the stability of the complex encompassing CpsD and ParB (Fig 9B and 9C) further suggests that CpsD phosphorylation constitutes a system for signaling the CPS synthesis status to chromosome segregation and ensuring that daughter cells are properly wrapped in CPS.
Collectively, these data fit into the model presented in Fig 10. ParB localizes first at the division site, rapidly followed by CpsD before chromosome segregation starts. When CpsD is not phosphorylated, reflecting that CpsD does not hydrolyze ATP and that the CPS assembly machinery is switched off, chromosome segregation is delayed due to reduced ParB mobility. By contrast, when the CPS assembly machinery is switched on, CpsD hydrolyzes ATP and autophosphorylates. Phosphorylation of CpsD then favors the mobility of ParB and chromosome segregation. CPS would thus cover the new cell halves as the cell elongates and constricts. Consistent with this model, cpsD-3YE cells are significantly shorter in length (Fig 4) reflecting that cell constriction likely occurs prematurely due to an enhanced crosstalk between CpsD and ParB.
Finally, our data question the raison d’être of such a phospho-regulatory process. While cpsD-3YE cells are slightly shorter in length, they divide properly and are covered by CPS (Fig 3A). An interesting and promising hypothesis concerns the life-style of the pneumococcus. CPS are required during infection for protection against the human immune system but they are also disadvantageous because of their inhibitory effects on adherence to the host cell [3, 6, 57, 58]. One can speculate that the presence of a regulatory mechanism based on CpsD phosphorylation coordinating CPS production with the cell cycle would allow modulating capsule production to satisfy optimal colonization and dissemination.
S. pneumoniae strains were cultivated at 37°C in Todd-Hewitt Yeast (THY) broth (Difco) or in C+Y medium [59]. S. pneumoniae mutants were constructed by transformation in D39 as previously described [60], using precompetent cells treated at 37°C with synthetic competence stimulating peptide 1 (CSP1) to induce competence. Transformants were selected on THY-agar or Columbia agar supplemented with either 3% (vol/vol) defibrinated horse blood or 2% defibrinated sheep blood and containing the appropriate antibiotic (streptomycin 200 μg.mL-1, kanamycin 250 μg.mL-1, spectinomycin 100 μg.mL-1, chloramphenicol 2 μg.mL-1). Strains complemented with cpsC or cpsD at the ectopic amiF/treR locus under the control of the maltose inducible promoter PM [31] were grown in C+Y medium containing 20% maltose. The E. coli XL1-Blue strain was used as a host for cloning. E. coli BL21 (DE3) strain was used as a host for overexpression. Luria-Bertani (LB) broth and agar supplemented with appropriate antibiotic (tetracycline 15 μg.mL-1, ampicillin 100 μg.mL-1) were used for routine growth at 37°C. Strains used in this study are listed in S1 Table.
To construct pneumococcus mutants (gene deletions, gfp/rfp/sfgfp fusions or site-directed mutagenesis), we used a two-step procedure based on a bicistronic kan-rpsL cassette called Janus [61], except for allelic replacement of parB by parB-sfGFP and hlpA by hlpA-RFP, where we used a one-step procedure with a spectinomycin or chloramphenicol resistance marker, respectively [36, 40]. Throughout this study, gene mutagenesis or fusion with fluorescent protein were constructed at each native chromosomal locus, expressed under the control of the native promoter and represented the only source of protein. Full description of primers used for the construction of strains is provided in S2 Table. The genes encoding monomeric sfGFP, monomeric GFP and RFP were from [33, 62, 63], respectively.
DNA fragments coding for CpsC, CpsC-ΔCter, CpsD and ParB were obtained by PCR using chromosomal DNA from S. pneumoniae D39 as template. For the chimera CpsC/D, we used DNA from strain TIGR4. Oligonucleotides used are described in S2 Table. The chimera DNA fragment was constructed by fusion of fragments obtained using primer pairs I-III and IV-II. The obtained DNA fragment was cloned between the BamHI and HindIII cloning sites of the pQE30 plasmid (Qiagen). parB was cloned between the NdeI and PstI cloning sites of the pT7.7 plasmid [64]. To construct plasmids for yeast two-hybrid, the PCR DNA fragments were digested by EcoRI and PstI and ligated either into pGAD-C1 or pGBDU-C1 vectors [65]. The nucleotide sequences of all DNA fragments were checked to ensure error-free amplification. Plasmids and primers used in this study are listed in S1 and S2 Tables, respectively.
Recombinant plasmids overproducing the chimera CpsC/D and ParB were transformed into BL21 (DE3) E. coli strain. The transformants were grown at 37°C until the culture reached an OD600 = 0.5. Expression was induced by adding IPTG to a final concentration of 0.5 mM and incubation was continued for 3 h. After 3 h culture at 37°C, cells were harvested and resuspended in buffer A (Tris-HCl 50 mM, pH 7.5; NaCl 300 mM; DTT 1 mM; imidazole 10 mM; glycerol 10%) containing 10 mg.L-1 of lysozyme and 6 mg.L-1 of DNase I and RNase A and sonicated. After centrifugation at 15000 g for 30 min, the supernatant was applied to a Ni-NTA agarose column (Qiagen) and extensively washed with buffer A supplemented with 30 mM imidazole. Samples were eluted with buffer A supplemented with 300 mM imidazole. The fractions corresponding to the pure protein were pooled and dialyzed against the following buffer: HEPES 50 mM, pH 7.5; NaCl 100 mM; DTT 1 mM; MgCl2 1 mM; glycerol 10%. The protein concentrations were determined using a Coomassie Assay Protein Dosage Reagent (Uptima) and aliquots were stored at -80°C.
Cultures of S. pneumoniae cells were grown at 37°C in THY medium until OD550 = 0.4. Cells pellets were incubated at 4°C for 15 min in buffer B (Tris-HCl 50 mM, pH 8.0; NaCl 150 mM; MgCl2 5 mM) containing 0.1 mg.mL-1 of lysozyme, 800 units of mutanolysine and 0.2 mg.mL-1 of DNase I and RNase A and sonicated. After centrifugation, the supernatant was incubated with the GFP-Trap resin suspension (Chromotech). Resin was washed according to the manufacturer’s instructions. Protein-bounded GFP-Trap resins were eluted with Laemmli buffer at 95°C for 10 min and analyzed by SDS-PAGE followed by an immunoblot directed against 6-Histidines tag or GFP.
The yeast two-hybrid phenotypic assays were performed as described previously [66]. Briefly, genes encoding for CpsD, CpsC and CpsC-ΔCter were fused to either the activating domain of Gal4 or the DNA-binding domain of Gal4. Resulting plasmids were inserted in yeast haploid cells and interactions were screened for ability to grow on the selective medium.
In vitro CpsC/D autophosphorylation and in vivo phosphorylated proteins in crude extract of S. pneumoniae were immunodetected using mouse anti-phosphotyrosine monoclonal antibody PY-20 (Sigma-Aldrich) at 1/2000. Detection of GFP fusions was performed using a rabbit anti-GFP polyclonal antibody (AMS Biotechnology) at 1/10000. Detection of Enolase was performed using a rabbit anti-enolase polyclonal antibody at 1/50000 [32]. Proteins fused to a 6-histidines tag were detected using a mouse anti-6His monoclonal antibody (Sigma-Aldrich) at 1/1500. Detection of capsular polysaccharides was performed using a rabbit anti-serotype 2 CPS polyclonal antibody (Statens serum Institute) at 1/2000. A goat anti-rabbit or anti-mouse secondary polyclonal antibody horseradish peroxidase (HRP) conjugated (Biorad) was used at 1/5000 to reveal immunoblots.
CPS were prepared as previously described with minor modifications [67]. Briefly, S. pneumoniae cultures were grown until OD550 = 0.3 and cells were harvested by centrifugation at 14,000xg for 20 min at 4°C. Pellets were then washed once with PBS and resuspended in buffer A (Tris-HCl 50 mM, pH 7,4; sucrose 20%; MgSO4 50 mM) at 1/100 of the original culture volume. The solution was then added with 400 units of mutanolysin and 6 μg of DNase and RNase per milliliter of solution and incubated overnight at RT. After centrifugation at 16,000xg for 20 min at 4°C, pellets were resuspended in the same volume of buffer A. 10 μL of the mixture were then mixed with 5 μL of buffer B (Tris-HCl 50 mM, pH 8.0; EDTA 50 mM; Tween20 0.5%; Triton X100 0.5%) and 20 μg of proteinase K, incubated 30 min at 37°C and analyzed by immunoblotting as previously described [68].
Protein sequence alignments were obtained using ClustalW and ESPript3 [69, 70]. Predicted secondary structures of CpsC/D and the B. subtilis Soj were made using Jpred3 [71]. The three-dimensional model of the chimera CpsC/D, based on the structure of the CapA1/B2 protein (PDB code 3BFV) was built using I-Tasser [72]. The visualization of 3D-molecules was performed using PyMOL (Schrödinger).
Microscale thermophoresis was used to test the interaction of ParB with the chimeras CpsC/D and CpsC/D-YF [73]. BSA (Bovine Serum Albumin) was used as negative control. Binding experiments were carried out with a Monolith NT.115 Series instrument (Nano Temper Technologies GMBH). ParB was labeled with the red dye NT-647. Briefly, 4 μl of sample containing 100 nM of labeled ParB and increasing concentrations of CpsC/D (from 4 nM to 92 μM) or BSA (from 5 nM to 180 μM) were loaded on K003 Monolith NT.115 hydrophilic treated silicon capillaries and thermophoresis was measured for 30 s. Each measurement was made in triplicates. Experiments were carried out at 25°C in MST optimized buffer (50 mM Tris-HCl, 150 mM NaCl, 10 mM MgCl2, 0.05% Tween-20). Analysis was performed with the Monolith software. Affinity KD was quantified by analyzing the change in normalized fluorescence (Fnorm = fluorescence after thermophoresis/initial fluorescence) as a function of the concentration of the titrated CpsC/D protein. The fraction of ParB bound was plotted against the concentration of CpsC/D.
Microscopy was performed on exponentially growing cells (OD550 = 0.3). For in vivo immunofluorescence microscopy, cells were mixed with an rabbit anti-serotype 2 CPS polyclonal antibody (Statens Serum Institute) at 1/1000, washed several times with THY at 37°C and then incubated with an anti-rabbit DyLight-549 (Jackson ImmunoResearch) at 1/2000. After a last wash with PBS, CPS were imaged. For DAPI, 10 μL of S. pneumoniae cell culture were mixed with 1 μL of DAPI at 2 mg.mL-1 (Molecular Probes) and incubated 5 min at room temperature. GFP, RFP and sfGFP fusions were visualized by fluorescence microscopy. Slides were visualized with a Zeiss AxioObserver Z1 microscope fitted with an Orca-R2 C10600 charge-coupled device (CCD) camera (Hamamatsu) with a 100× NA 1.46 objective. Images were collected with axiovision (Carl Zeiss), deconvolved with ImageJ (http://rsb.info.nih.gov/ij/) and analyzed with Coli-Inspector (detection was approved manually) running under the plugin ObjectJ (http://simon.bio.uva.nl/objectj/) to measure CPS fluorescence intensity or generate fluorescent intensity linescans sorted to cell length and MicrobeTracker suite [74] extended by custom MATLAB routines to generate cell length distribution histograms. Cell lengths were determined using MicrobeTracker. Images are representative of experiments made in triplicate.
Fluorescence time-lapse microscopy was performed on a DV Elite (Applied Precision) with a sCMOS camera using SSI Solid State Illumination (Applied Precision) through a 100× oil immersion objective (phase contrast), essentially as described before [36, 75]. During the experiment, cells were incubated on a slide of C+Y agarose and kept at 37°C in a temperature controlled chamber. Phase contrast and fluorescent (GFP and RFP) images were acquired every 5 min. Images were processed using softWoRx 5.5 (Applied Precision) and ImageJ.
For TEM, S. pneumoniae cells (wild type and mutants) were collected, centrifuged and pre-fixed 20 min on ice with the fixing mix (cacodylate 0.1 M, pH 7.4; glutaraldehyde 5%; lysine-HCl 0.075 M; ruthenium red 0.075%). After washing, cells were fixed overnight at 4°C with fixing mix without lysine-HCl. Cells were took onto 2% agarose and post-fixation with 1% osmium tetroxide in cacodylated buffer added to ruthenium red was carried out for 1 h at room temperature. These fixed cells were dehydrated using a graded series of ethanol and embedded in LR-White at 60°C for 48 h. Ultrathin sections (60 nm) were obtained using a Leica UC7 microtome and were counter-stained with uranyl acetate and lead citrate (Reichert Ultrostainer, Leica). Samples were examined with a Phillips CM120 transmission electron microscope equipped with a Gatan Orius SC200 CCD camera.
TIRF microscopy was performed as described before [36]. Cells were grown in C+Y until mid-exponential phase prior to examination on a DV Elite (Applied Precision) with a sCMOS camera using 50 mW laser illumination (488 nm and 561 nm) through a 100× oil 1.49 NA TIRF objective. Cells were imaged every 10 seconds for 40 seconds. The GFP foci in the cells were classified as stationary or dynamic (splitting, appearing or disappearing foci) using intensity line scans at different time points in Image J. Fluorescence foci were classified as “splitting” when one fluorescence peak splits in two peaks during the time span of the experiment and foci were classified as “disappearing” when the signal of a fluorescence peak was reduced >90% during the time span of the experiment.
|
10.1371/journal.pntd.0007357 | Loss of cytoplasmic incompatibility in Wolbachia-infected Aedes aegypti under field conditions | Wolbachia bacteria are now being introduced into Aedes aegypti mosquito populations for dengue control. When Wolbachia infections are at a high frequency, they influence the local transmission of dengue by direct virus blocking as well as deleterious effects on vector mosquito populations. However, the effectiveness of this strategy could be influenced by environmental temperatures that decrease Wolbachia density, thereby reducing the ability of Wolbachia to invade and persist in the population and block viruses. We reared wMel-infected Ae. aegypti larvae in the field during the wet season in Cairns, North Queensland. Containers placed in the shade produced mosquitoes with a high Wolbachia density and little impact on cytoplasmic incompatibility. However, in 50% shade where temperatures reached 39°C during the day, wMel-infected males partially lost their ability to induce cytoplasmic incompatibility and females had greatly reduced egg hatch when crossed to infected males. In a second experiment under somewhat hotter conditions (>40°C in 50% shade), field-reared wMel-infected females had their egg hatch reduced to 25% when crossed to field-reared wMel-infected males. Wolbachia density was reduced in 50% shade for both sexes in both experiments, with some mosquitoes cleared of their Wolbachia infections entirely. To investigate the critical temperature range for the loss of Wolbachia infections, we held Ae. aegypti eggs in thermocyclers for one week at a range of cyclical temperatures. Adult wMel density declined when eggs were held at 26–36°C or above with complete loss at 30–40°C, while the density of wAlbB remained high until temperatures were lethal. These findings suggest that high temperature effects on Wolbachia are potentially substantial when breeding containers are exposed to partial sunlight but not shade. Heat stress could reduce the ability of Wolbachia infections to invade mosquito populations in some locations and may compromise the ability of Wolbachia to block virus transmission in the field. Temperature effects may also have an ecological impact on mosquito populations given that a proportion of the population becomes self-incompatible.
| Aedes aegypti mosquitoes infected with Wolbachia symbionts are being deployed in the tropics as a way of reducing disease transmission. Some Wolbachia strains are vulnerable to high temperatures but these effects have not been evaluated outside of a laboratory setting. We reared Ae. aegypti infected with the wMel strain of Wolbachia in the field during the wet season in Cairns, Australia, where the first releases of Wolbachia-infected Ae. aegypti took place. wMel-infected mosquitoes became partially self-incompatible, with reduced egg hatch, when larvae were reared in partial shade where maximum daily temperatures exceeded 39°C. Under these conditions the amount of Wolbachia in adult mosquitoes was reduced to less than 1% of laboratory-reared mosquitoes on average, while some mosquitoes were cleared of Wolbachia entirely. In contrast, wMel was stable when mosquitoes were reared under cooler conditions in full shade. Field trials with the wMel strain are now underway in over 10 countries, but high temperatures in some locales may constrain the ability of Wolbachia to invade natural mosquito populations and block disease transmission.
| Aedes aegypti mosquitoes are the principal vectors of dengue and are widespread in the tropics where they live near humans [1, 2]. Chemical insecticides have historically been used to control Ae. aegypti populations during disease outbreaks, but this approach is unlikely to be sustainable as insecticide resistance is now widespread in many parts of the world [3, 4]. There is increasing interest in ‘rear and release’ programs where mosquitoes modified with desirable traits are released into natural populations as an alternative approach to disease control [5]. At the forefront of these programs is the deployment of mosquitoes infected with Wolbachia bacteria. Wolbachia occur naturally in many insects but have been introduced experimentally into Ae. aegypti where they can interfere with the transmission of dengue and other pathogens [6–8]. Wolbachia are transmitted maternally and typically affect host reproduction [9] or provide other advantages [10] to facilitate their spread into populations. These phenotypes have been utilized in disease control programs where Wolbachia-infected mosquitoes have been deployed to replace natural populations [11–13] or suppress populations through the release of only males [14, 15]. Both approaches rely on cytoplasmic incompatibility induced by Wolbachia, where uninfected females that mate with infected males do not produce viable offspring, but viability is restored if the female is also Wolbachia-infected [16, 17].
Over ten Wolbachia strain associations have now been generated in Ae. aegypti and they exhibit a diverse range of phenotypes. Some Wolbachia strains are relatively benign and have little impact on host fitness or virus blockage such as the wRi strain [18]. Others impose large fitness costs but also strongly reduce virus transmission including wMelPop [17] and wAu [19] while others like wAlbB fall somewhere in between [20, 21]. There are also superinfections where two or more Wolbachia strains infect the same host, which can have combined or unexpected effects [21, 22]. Aedes aegypti infected with the wMel strain of Wolbachia have been or are now being released in over ten countries (https://www.worldmosquitoprogram.org/) and have successfully established in suburban areas in Cairns and Townsville in Queensland, Australia [11–13] and in Brazil [23]. The wAlbB strain has also been deployed successfully in Malaysia for population replacement (http://www.imr.gov.my/wolbachia/) and in several countries for population suppression, where the release of only infected males has reduced population sizes by more than 80% due to cytoplasmic incompatibility (https://debug.com/; https://www.nea.gov.sg/corporate-functions/resources/research/wolbachia-aedes-mosquito-suppression-strategy).
Despite these successes, there are limitations of Wolbachia infections that may affect their utility as disease control agents (reviewed in Ritchie et al. [24]). The majority of Wolbachia infections in Ae. aegypti reduce mosquito fitness and these costs tend to be exacerbated in stressful environments such as when larvae are starved [25] or in quiescent eggs [18, 19, 26, 27]. Fitness costs can have enormous effects on invasion success. For example, the wMelPop infection failed to persist in release zones in Australia and Vietnam despite reaching frequencies above 90%, likely due to the massive fitness costs of this strain [28]. Wolbachia infections that occur naturally in mosquitoes can interfere with patterns of cytoplasmic incompatibility and limit the potential for population replacement and suppression [29]. Density-dependent interactions [30] and spatially heterogeneous environments [31] can also slow the rate of invasion, as can pesticide susceptibility in released mosquitoes [23].
For population replacement programs to be successful, Wolbachia infections should persist at high frequencies in the environment and block virus transmission under field conditions for many years following deployment [24]. There is a risk that Wolbachia infections, viruses or mosquitoes will evolve following the establishment of Wolbachia in populations, leading to less effective virus protection in Wolbachia-infected mosquitoes in the long-term [32]. However, the wMel infection has remained stable so far in terms of virus blockage [33] and its effects on fitness [34]. After seven years in the field, wMel has retained a high titer and continues to induce complete cytoplasmic incompatibility in the laboratory [35], indicating that attenuation is unlikely for at least several years following deployment.
While the wMel infection in northern Queensland Ae. aegypti populations does not appear to have changed phenotypically since release, environmental conditions in the field such as temperature can have transient effects on Wolbachia infections, influencing their ability to suppress virus transmission or establish in populations. This issue is particularly important as climate change is leading to warmer average conditions and higher temperature extremes, including in the tropics [36]. Wolbachia infections in Ae. aegypti are vulnerable to high temperatures; heat stress during larval development reduces Wolbachia density in adults [37] and decreases the fidelity of cytoplasmic incompatibility and maternal transmission [35, 38]. The fidelity of cytoplasmic incompatibility and maternal transmission are two key parameters for Wolbachia spread [39] while Wolbachia density is positively associated with the strength of virus blockage in both Drosophila [40, 41] and mosquitoes [42].
Wolbachia strains in Ae. aegypti differ in their response to heat stress; the wMel and wMelPop strains are relatively susceptible while wAlbA, wAlbB and wAu are more robust, retaining high densities when larvae are reared at cyclical temperatures of 26–37°C [19, 38]. These laboratory studies demonstrate the potential for heat stress to affect the success of Wolbachia interventions, but conditions experienced by mosquitoes in field situations are more complex than in an incubator. To understand the effects of high temperatures under more natural conditions, we reared Ae. aegypti larvae infected with wMel in the field with varying levels of exposure to sunlight. We then performed crosses to test for effects on cytoplasmic incompatibility and measured Wolbachia density. Finally, we performed experiments with Ae. aegypti eggs in the laboratory to determine the range of temperatures that adversely affect Wolbachia infections.
Blood feeding on human subjects was approved by the Human Research Ethics Committee, James Cook University (approval H4907). All adult subjects provided informed oral consent (no children were involved).
Aedes aegypti mosquitoes infected with the wMel strain of Wolbachia were collected in 2013 from locations near Cairns, Australia where wMel had successfully established [11]. Aedes aegypti infected with wMelPop were collected from Cairns, Australia in 2012 following field releases and local field breeding [28]. Aedes aegypti infected with wAlbB were derived from laboratory colonies and are described in Xi et al. [16] and Axford et al. [20]. Uninfected Ae. aegypti were collected in 2016 from locations where Wolbachia-infected mosquitoes had not been released. Mosquitoes were maintained in an insectary at the University of Melbourne according to methods described by Ross et al. [43]. Females with each Wolbachia strain were crossed to males from the uninfected population for at least three generations prior to the start of experiments to ensure that genetic backgrounds between populations were similar [27].
We conducted two experiments during the wet season in Cairns, Australia in 2018 to test the stability of the wMel Wolbachia infection under field temperature conditions. Experiments took place at James Cook University under a protective awning with different levels of shade as described in Ritchie et al. [44]. The relative reduction in light intensity compared to direct sunlight was determined using an EasyView EA30 light meter (Extech Instruments Corporation, Waltham, MA 02451 U.S.A.); two shade levels were chosen for the experiments which reduced light intensity by 50% and 99% respectively. We then established water-filled containers of various sizes to simulate a range of field larval habitats. The wMel-infected eggs were hatched in a single tray in 99% shade, then approximately 100 1st instar larvae were placed into each container. Larvae were provided with TetraMin tropical fish food tablets (Tetra, Melle, Germany) ad libitum throughout their development. Water temperatures were recorded every 30 minutes for the duration of larval development with data loggers (Thermochron; 1-Wire, iButton.com, Dallas Semiconductors, Sunnyvale, CA, USA) placed in zip-lock bags and submerged in the centre of each container. Pupae were collected one week after hatching and were returned to the laboratory for adult emergence.
In the first experiment, Ae. aegypti eggs infected with the wMel strain of Wolbachia were hatched on the 10th of January 2018 and pupae were collected on the 17th of January. We used containers of three types with a wide range of water volumes which were expected to experience a variety of temperature conditions. Black buckets (13 cm radius, 25 cm height) were filled with 8 L of tap water, plant pots (5 cm radius, 10 cm height) with 500 mL and cups (2 cm radius, 6 cm height) with 60 mL. Containers were covered with mesh or stockings to prevent wild mosquitoes from ovipositing, and experimental mosquitoes from escaping. This setup was repeated for both the 99% and 50% shade levels. wMel-infected larvae were also reared in a single bucket filled with 8 L of water and placed in direct sunlight.
In the second experiment, we repeated this procedure but used two container types: black buckets filled with 8 L of water and small round clear plastic containers (10 cm radius, 7 cm height) filled with 400 mL of water, with each container replicated three times at 99% and 50% shade levels. Eggs were hatched on the 26th of January 2018 and pupae were returned to the laboratory on the 1st of February. Populations of wMel-infected and uninfected Ae. aegypti were also reared in the laboratory concurrently at 26°C ± 1°C according to Ross et al. [43] for experimental crosses.
We tested the ability of wMel-infected males to induce cytoplasmic incompatibility and wMel-infected females to restore compatibility after being reared under field temperature conditions. Adults emerging from each container type and shade level were added to 15 cm3 cages (BugDorm-4M1515, Megaview Science Co., Taichung, Taiwan) where sexes were maintained separately. Crosses were performed two to three days after adults emerged by aspirating approximately 50 females into cages with an equal number of males. Females were blood fed three days later and then isolated in plastic cups containing 15 mL of larval rearing water and a strip of sandpaper for oviposition. Eggs were collected four days after blood feeding, partially dried, and then hatched four days after collection. Eggs were then counted under a dissecting microscope and hatch rates were determined by counting the proportion of eggs that had a clearly detached cap.
In the first experiment, we performed crosses with adults reared in buckets at either 99% or 50% shade. Field-reared wMel-infected males were crossed to uninfected females to determine their ability to induce cytoplasmic incompatibility. We also crossed field-reared wMel-infected females to either uninfected males or wMel-infected and laboratory-reared males to determine the ability of females to restore compatibility. Infected males and uninfected females both reared under laboratory conditions were crossed to each other to confirm that wMel induces complete cytoplasmic incompatibility at 26°C. Twenty females were isolated for oviposition in each of these crosses, but individuals that died or did not lay eggs were excluded from the analysis.
In the second experiment we performed a similar set of crosses but only for adults emerging from buckets held in 50% shade. In crosses where both sexes were infected we used males and females from the same container rather than using males reared at 26°C. This was done to see if populations became self-incompatible when both sexes were reared at warmer temperatures. Thirty females were isolated for oviposition in each cross. Crosses between males and females reared under the same conditions were also performed with adults that were reared in buckets held in 99% shade and small containers held at 50% and 99% shade. We determined egg hatch proportions and Wolbachia densities of females in each of these crosses to see if there was a relationship between Wolbachia density and hatch rate.
We performed two experiments using thermocyclers (Biometra, Göttingen, Germany) to test the thermal tolerance of Wolbachia-infected Ae. aegypti eggs and the density of Wolbachia under a range of temperature conditions. We followed methods described in Kong et al. [45] with some modifications. Eggs from uninfected, wMel, wAlbB and wMelPop colonies were collected on sandpaper strips which were then partially dried, wrapped in paper towel and held in sealed zip-lock bags. Four days after collection, eggs were brushed onto filter paper with a small paint brush and then tipped into 0.2 mL PCR tubes using a funnel. Batches of 15 to 39 eggs (mean 25.7) were added to each tube. Tubes were closed and then tapped on the bench to ensure that eggs sank to the bottom of the tube where temperature control in the thermocycler is the most accurate [45]. Tubes were then placed in heat blocks of Biometra TProfessional TRIO 48 thermocyclers with tubes from each population arranged randomly in each block.
In both experiments we used three thermocyclers, each with three heated blocks that can run independently for a total of 9 temperature regimes. In the first experiment we chose a broad range of temperature cycles to cover the entire range of temperatures that Ae. aegypti may experience in the field (http://www.bom.gov.au/climate/averages/tables/cw_031011_All.shtml). Each regime had a fluctuation of 10°C between the minimum and maximum temperature; the lowest being 8–18°C and the highest being 32–42°C, with a difference of 3°C between each regime (S3A Fig). In the second experiment we chose a narrower temperature range based on when egg hatch and Wolbachia density started to decline in the previous experiment. The lowest regime was set to 24–34°C and the highest was 32–42°C, with difference of 1°C between each regime (S3B Fig). In the first experiment there were six replicate tubes of eggs for each temperature cycle and Wolbachia infection type and the second experiment had 12 replicates. After all tubes were added to the thermocyclers we closed the lids and started programs simultaneously. Eggs in tubes were also maintained at 26°C in a controlled temperature room in both experiments.
After one week, tubes were removed from the thermocyclers and eggs were hatched by holding PCR tubes sideways above 70 mL specimen cups and then pipetting water into the tubes so that eggs fell into the cup. Each cup was filled with 40 mL of water and provided with a small amount of TetraMin and a few grains of yeast. Two days after hatching we determined egg hatch proportions by dividing the number of larvae by the number of eggs. We counted larvae again every 2 days as some eggs were slow to hatch, allowing one week in total for larvae to appear before we ceased counting. All larvae that hatched were added to plastic containers filled with 500mL of RO water and reared to adulthood. Multiple replicate cups of larvae were combined into trays for rearing, but the larval density was controlled to 100 larvae per tray or fewer to account for effects of larval competition and development time on Wolbachia density [46]. All adults were stored in ethanol for Wolbachia density measurements.
In each experiment, random subsets of adults were stored in ethanol within 24 hours of emergence for Wolbachia screening. For both field experiments we extracted DNA from 16 males and 16 females from each container type and shade level. For experiments with eggs held in thermocyclers we extracted DNA from up to 10 (first experiment) or 12 (second experiment) males and 10 or 12 females from each Wolbachia infection type and treatment. Some treatments had lower sample sizes due to low egg hatch proportions. DNA was extracted from whole adults with 150 μL of 5% Chelex 100 resin (Bio-Rad Laboratories, Hercules, CA) and 3 μL of Proteinase K. We then conducted qPCR to detect and estimate the density of Wolbachia in each whole adult using methods described previously [47]. Individuals were considered uninfected if the Ae. aegypti-specific marker amplified successfully (Cp value < 35) but the Wolbachia-specific marker did not (Cp value of 35 or no Cp value) in two independent runs. For individuals that were positive for Wolbachia, (Cp value < 35 for both markers), differences in Cp between the two markers were transformed by 2n to provide an estimate of Wolbachia density, averaged from at least two independent runs. For the second field experiment, we also estimated the Wolbachia density of females after they had laid eggs to see if there was a relationship between Wolbachia density and egg hatch rate when crossed to infected males reared under the same conditions.
All data were analyzed using SPSS statistics version 24.0 for Windows (SPSS Inc, Chicago, IL). Hatch proportion and Wolbachia density data were often not normally distributed, and we therefore compared treatments for these variables with Kruskal-Wallis and Mann-Whitney U tests. We also used Spearman’s rank-order correlation to test the relationship between egg hatch and female Wolbachia density.
We monitored water temperatures experienced by larvae in each container type at different levels of shade. Maximum temperatures differed between container types at 99% shade, with cups having average maximum daily temperatures that were 2.5°C higher than buckets, though average temperatures were similar because smaller containers reached cooler temperatures at night (S1 Fig). Temperature cycles were similar between containers in 50% shade, which was unexpected given the large differences in water volume. The level of shade affected temperature substantially, with buckets in direct sunlight experiencing average maximum temperatures of 38.3°C, while containers in 50% shade (average minimum: 23.7°C, average maximum: 35.3°C) were much warmer than containers in 99% shade (23.2–29.6°C) (Fig 1A).
We tested the ability of wMel-infected males reared under field temperature conditions to induce cytoplasmic incompatibility with uninfected females. In a control cross, wMel-infected males reared in the laboratory at 26°C caused complete cytoplasmic incompatibility (no eggs hatched) with uninfected females (Fig 1). wMel-infected males reared in buckets at 99% shade induced almost complete cytoplasmic incompatibility, though 2/18 females produced a single viable progeny each (Fig 1B). wMel-infected males reared in buckets at 50% shade induced weaker cytoplasmic incompatibility, with 9/16 females producing some viable progeny (Fig 1C).
We also tested the ability of wMel-infected females to retain their compatibility with wMel-infected males reared in the laboratory. When females were reared in buckets at 99% shade, there was no difference in hatch rate between crosses with uninfected males and crosses with wMel-infected males (Mann-Whitney U: Z = 0.348, P = 0.726, Fig 1B). In contrast, wMel-infected females reared in buckets at 50% shade had a 47.6% reduction in egg hatch rate when crossed to wMel-infected males (Z = 3.612, P < 0.001). This indicates partial incompatibility with Wolbachia-infected males, suggesting a substantial loss of Wolbachia infection.
We estimated the Wolbachia density of a subset of adults from each container type and level of shade (Fig 2). Wolbachia density was not consistently affected by container type for both females (Kruskal-Wallis χ2 = 2.598, df = 2, P = 0.273) and males (χ2 = 4.419, df = 2, P = 0.110), likely because the container types experienced similar temperature fluctuations at 50% shade. Conversely, Wolbachia density was affected substantially by shade level for both females (χ2 = 71.261, df = 1, P < 0.001) and males (χ2 = 68.563, df = 1, p < 0.001). Females reared under 50% shade had a median Wolbachia density that was just 0.32% of the laboratory control, while males had a density of 8.09% of the control. This reduction likely reflects the substantially higher maximum daily temperatures experienced in containers at 50% shade. In contrast, the Wolbachia density of adults reared at 99% shade was not significantly different to laboratory-reared adults (females: χ2 = 0.650, df = 1, P = 0.420, males: χ2 = 0.085, df = 1, P = 0.771).
All adults screened from containers in 99% shade, 50% shade and the laboratory were positive for Wolbachia. However, we were unable to detect any Wolbachia infection in a sample of 11 adults taken from a bucket placed in direct sunlight. This indicates a complete loss of infection which is likely due to the extreme temperatures experienced in that container (up to 43°C, Fig 1A). Though we did not score survival to adulthood in containers directly, the bucket placed in direct sunlight experienced high mortality since only 11 adults emerged out of the approximately 100 larvae added initially.
We conducted a second experiment later in the month where we also tested cytoplasmic incompatibility and measured Wolbachia density in adults. Temperatures were affected substantially by the location of containers where average maximum temperatures were nearly 7°C warmer in 50% shade compared to 99% shade (Fig 3A). Maximum temperatures also differed between container types; at 50% shade small containers reached 39.26°C on average while buckets reached 36.54°C, but average temperatures did not differ much between containers at the same level of shade because of warmer minimum temperatures in buckets.
We set up crosses with adults emerging from buckets held in 50% shade to test for any effects on cytoplasmic incompatibility. wMel-infected males induced strong but incomplete cytoplasmic incompatibility with uninfected females; 7/18 females produced some viable progeny, compared to 0/20 in the control (Fig 3B). wMel-infected adults reared in buckets at 50% shade that were crossed to each other experienced a 79.9% reduction in egg hatch rate when crossed to each other relative to crosses with uninfected males (Mann-Whitney U: Z = 5.615, P < 0.001), suggesting a greatly reduced ability of females to restore compatibility under these rearing conditions.
We estimated the Wolbachia density of adults that emerged from each treatment in the second experiment. We found that many individuals had lost their Wolbachia infection (no detectable infection) when reared in containers held in 50% shade (Fig 4), particularly in small containers where larvae experienced higher maximum daily temperatures (Fig 3A). Of the adults reared at 50% shade that were still infected with Wolbachia, their density had been reduced to 0.19% and 0.23% of the 26°C control in females and males respectively. In this experiment, Wolbachia density was also reduced at 99% shade relative to the 26°C control (females: χ2 = 14.828, df = 1, P < 0.001, males: χ2 = 16.519, df = 1, P < 0.001), with densities being approximately 50% the level of the control.
wMel-infected adults emerging from the two container types and shade levels were returned to the laboratory and allowed to mate with individuals from the same container. We then scored egg hatch proportions of individual females and measured their Wolbachia density after oviposition to determine the relationship between Wolbachia density and egg hatch proportion. Females with high Wolbachia densities exhibited high hatch proportions while females with lower densities tended to have very low hatch proportions or produced no viable offspring, with the correlation between density and egg hatch being highly significant (Spearman’s rank-order correlation: ρ = 0.899, P < 0.001, n = 65, Fig 5). This indicates that females with low densities had partially or completely lost their ability to restore compatibility, but males reared under the same conditions had largely retained their ability to induce cytoplasmic incompatibility. The strong relationship between Wolbachia density and egg hatch indicates that a high density in females is important for restoring compatibility with infected males.
We tested the tolerance of Wolbachia-infected and uninfected eggs to a broad range of temperature conditions. When eggs were held at 26°C for one week, Wolbachia-infected eggs did not differ from uninfected eggs in terms of hatch proportion (Mann-Whitney U: all P > 0.05). At higher temperatures fitness costs of Wolbachia infections were evident; wMelPop-infected eggs had lower hatch proportions than uninfected eggs under temperature cycles of 26–36°C and 29–39°C (both Z = 2.802, P = 0.005, Fig 6A). wAlbB-infected eggs also had reduced hatch proportions relative to uninfected eggs at 29–39°C (Z = 2.162, P = 0.031) but wMel-infected eggs did not differ from uninfected eggs under any temperature cycle (all P > 0.05).
We reared larvae hatching from eggs held at each temperature cycle and measured Wolbachia density in adults. Wolbachia density did not differ between males and females across all temperature conditions (wMel: Kruskal-Wallis χ2 = 0.271, df = 1, P = 0.603, wAlbB: χ2 = 2.398, df = 1, P = 0.122) except for wMelPop, where density was higher in females than in males (χ2 = 14.507, df = 1, P < 0.001). When eggs were held at 26°C, wMelPop-infected adults had the highest density of Wolbachia while wAlbB had an intermediate density and wMel had the lowest density (Fig 6B and 6C), consistent with previous studies [20, 38]. This pattern was consistent across the cooler temperature cycles (maximum daily temperatures of 18–33°C) where Wolbachia densities for wAlbB (χ2 = 6.505, df = 5, P = 0.260), wMelPop (χ2 = 9.108, df = 5, P = 0.105) and wMel males (χ2 = 2.950, df = 5, P = 0.708) were stable (Fig 6B and 6C). In contrast, wMel density in females declined with increasing maximum temperatures across this range (χ2 = 27.190, df = 5, P < 0.001). When eggs were held at 29–39°C, adult Wolbachia density declined steeply for both wMel and wMelPop infections (Fig 6B and 6C). Median wMel densities were reduced to only 0.41% and 0.14% of densities at 26°C in females and males, respectively. In wMelPop the relative loss was even steeper, with females reared from eggs held at 29–39°C having just 0.02% the density of females at 26°C, while Wolbachia was not detected in males. wAlbB density also declined when eggs were held at 29–39°C in both females (Mann-Whitney U: Z = 2.797, P = 0.005) and males (Z = 2.735, P = 0.006) but the effect was much weaker than the other strains, with median densities of 34.88% (in females) and 42.04% (in males) of eggs held at 26°C.
In a second experiment, we used a narrower temperature range to investigate egg thermal tolerance and the loss of Wolbachia infections on a finer scale and with greater replication. Egg hatch proportions declined for all Wolbachia infection types as maximum temperatures increased, with the effect being most severe for the wMelPop infection (Fig 7A). Egg hatch proportions of the wMel and wAlbB strains did not differ significantly from uninfected eggs at maximum temperatures of 37°C and below (Mann-Whitney U: all P > 0.05). However, at maximum temperatures of 38–41°C both the wMel and wAlbB strains had lower hatch proportions than uninfected eggs held at the same temperature (all P ≤ 0.026). This indicates that Wolbachia infections in Ae. aegypti lower the tolerance of eggs to high temperatures, particularly in the case of wMelPop.
Consistent with the previous experiment, Wolbachia density declined as eggs were exposed to increasing maximum temperatures, beginning at 35°C for wMelPop and 36°C for wMel (Fig 7B and 7C). The wMel and wMelPop infections were lost from some individuals when eggs were exposed to 29–39°C and absent from all adults at 30–40°C (Fig 7B and 7C). In contrast, all wAlbB adults were infected across all temperature cycles, though density was reduced in both females (Mann-Whitney U: Z = 3.204, P = 0.001) and males (Z = 3.897, P < 0.001) at 30–40°C relative to 26°C.
We tested the stability of the wMel Wolbachia infection in Ae. aegypti under field temperature conditions and performed laboratory experiments to determine the range of temperatures that affect different Wolbachia strains. Our experiments demonstrate three main outcomes of heat stress on Wolbachia-infected mosquitoes. Firstly, there are direct costs of Wolbachia infections on Ae. aegypti thermal tolerance, at least during the egg stage. Secondly, heat stress under partial shade conditions in the field reduces cytoplasmic incompatibility fidelity in wMel-infected males, while infected females become partially incompatible with infected males. Thirdly, heat stress reduces Wolbachia density and may impair the ability of Wolbachia to block virus transmission for a subsection of the mosquito population reared under specific field conditions. Heat stress could therefore adversely affect the success of disease control programs depending on the location and nature of the field breeding sites.
There are relatively few examples of symbionts affecting the thermal tolerance of their hosts [48]. In Drosophila melanogaster, the wMelCS strain of Wolbachia increases the survival of adults under heat stress while the wMelPop infection decreases survival [49], though wMel appears to have no effect on high temperature tolerance [50]. Here we show that Wolbachia infection reduces the tolerance of Ae. aegypti eggs to high temperatures, with the severity of the effect depending on Wolbachia strain. In addition, we have determined the temperature range where deleterious effects on Wolbachia infections start to occur and where the infections are lost, at least during the egg stage. For the wMelPop and wMel infections, Wolbachia density declined beginning at temperatures of 25–35°C (30°C mean) and 26–36°C (31°C mean) respectively, while for wAlbB this occurred at a much higher temperature range (30–40°C, 35°C mean). The higher tolerance of wAlbB to heat stress is consistent with prior studies in Ae. aegypti larvae [19, 35, 38], but the increased resolution in this experiment provides a better estimate of the maximum daily temperatures that could affect Wolbachia interventions. In field situations the temperature ranges where Wolbachia infections are adversely affected will depend on the duration and timing of heat stress (see below).
In our field experiments we found substantial effects of heat stress on cytoplasmic incompatibility that could limit the potential of wMel to invade natural populations during disease control programs and persist following releases. When reared in partial shade, wMel-infected males partially lost their ability to induce cytoplasmic incompatibility while females partially or completely lost their ability to restore compatibility when crossed to infected males reared in the lab. Infected females reared in partial shade had greatly reduced fertility in crosses with infected males from the same container. Female Wolbachia density was positively associated with egg hatch, consistent with a study in Drosophila [51]. High densities therefore appear needed for females to restore compatibility with infected males, but the density required for males to induce cytoplasmic incompatibility appears to be lower. Heat stress conditions in the field could greatly diminish or even reverse the reproductive advantage provided by Wolbachia, making invasion challenging, particularly when Wolbachia is at a low frequency, when its fitness relative to uninfected individuals is relatively lower and where it is susceptible to stochastic [52, 53] and density related [30, 54] effects. Where wMel has already established in a population, reduced egg hatch in wMel-infected mosquitoes that mate with each other could provide an opportunity for an increase in the frequency of uninfected mosquitoes although once wMel invaded areas of North Queensland it appears to have been stable [34]. Fitness costs and self-incompatibility between infected mosquitoes could also have unexpected ecological effects; a decline in the Ae. aegypti population could lead to shifts in species composition [24] which could be beneficial for disease control efforts.
Though we attempted to rear mosquitoes under realistic temperature conditions, our field experiments will only be relevant to a subset of natural breeding sites. We provided abundant food to speed up and synchronise larval development to facilitate experimental crosses between strains. Larval development times in nature are variable and can exceed two months under competitive conditions [30]. Increasing the rate of larval development in this experiment likely underestimated the effect of heat stress; longer development times increase the chance that larvae will experience a heat wave and increased durations of heat stress may further reduce Wolbachia density, though density may also recover over time in the absence of heat stress [37]. wMel-infected larvae provided with a low level of food have a greatly reduced Wolbachia density when reared at 26–32°C compared to 26°C, indicating that even moderate temperatures can reduce Wolbachia density when combined with nutritional stress (see Figure S4 of Ross and Hoffmann [35]). The effects of heat stress on Wolbachia density can carry over into the next generation [55] which may lead to reduced virus blockage or cytoplasmic incompatibility across a generation after a heat wave. In our laboratory experiments eggs were maintained for one week before hatching, but in the field the egg stage can be shorter or much longer. During the dry season eggs can remain quiescent for months before hatching [56], increasing their potential exposure to high temperatures. Wolbachia infections reduce the viability of quiescent Ae. aegypti eggs [20, 26, 27] and under high temperatures these fitness costs will likely be exacerbated.
A further limitation of our field experiment is that the containers used for larval rearing were not colonized naturally. Aedes aegypti seem to prefer laying eggs in shaded areas but will also utilize containers in sunlight [57–59]. Wolbachia infections may also affect thermal preference; adult Drosophila melanogaster infected with Wolbachia prefer cooler temperatures than uninfected flies [60, 61]. Nevertheless, data from sentinel containers indicates that wMel-infected mosquitoes will lay eggs in containers placed in direct sunlight. Sentinel buckets and small containers placed within the wMel release zone in Cairns were all colonized by Ae. aegypti despite some of these experiencing similar temperatures to the experimental containers held at 50% shade (S2 Fig). Ae. aegypti tend to lay eggs during cooler parts of the day [62, 63] and therefore may be unable to discriminate against habitats that reach high maximum temperatures later. Unlike adult mosquitoes, immature stages cannot easily escape heat stress as they are unable to move beyond the container. Since Wolbachia density and egg hatch in wMel-infected mosquitoes appears to depend strongly on the level of shade, temperature and productivity surveys of larval habitats could be conducted in release areas if there are concerns around heat stress impacts in a release area.
Despite the substantial effects on Wolbachia density and fertility in our experiments, Ae. aegypti mosquitoes infected with wMel have successfully established in Cairns [11, 12] and Townsville [13], Australia and in Brazil [23], with the infection persisting at a high frequency in most locations. In areas where the releases succeeded, the costs of heat stress observed here were clearly not prevalent or severe enough to prevent the establishment of wMel. Once a Wolbachia infection has attained a high frequency in a population it may stay high unless the fitness costs are extreme, as is the case for wMelPop [28]. Nevertheless, heat stress will likely slow the rate of Wolbachia invasion and spread, increasing the number of mosquitoes required for releases, and potentially creating an unstable situation around critical invasion points that must be exceeded for Wolbachia to invade [52]. Heat stress could partially explain why infection frequencies have persisted at intermediate levels in some suburbs [13] and may also contribute to the incomplete maternal transmission fidelity of wMel observed in Cairns [64] given that some individuals were cleared of their Wolbachia infections in our experiments. wMel-infected mosquito releases outside of Australia in locations where maximum daily temperatures are warmer may be more challenging. Reduced Wolbachia densities may also reduce virus protection provided by Wolbachia even if infection frequencies remain high in a population, though we do not demonstrate this effect directly. The wMel strain has retained its susceptibility to heat stress for seven years after field deployment in Australia [35], indicating that alternative strains may be needed in areas where wMel has difficulty establishing or where viral blockage is insufficient.
|
10.1371/journal.pntd.0001896 | Identification of Compounds with Anti-Proliferative Activity against Trypanosoma brucei brucei Strain 427 by a Whole Cell Viability Based HTS Campaign | Human African Trypanosomiasis (HAT) is caused by two trypanosome sub-species, Trypanosoma brucei rhodesiense and Trypanosoma brucei gambiense. Drugs available for the treatment of HAT have significant issues related to difficult administration regimes and limited efficacy across species and disease stages. Hence, there is considerable need to find new alternative and less toxic drugs. An approach to identify starting points for new drug candidates is high throughput screening (HTS) of large compound library collections. We describe the application of an Alamar Blue based, 384-well HTS assay to screen a library of 87,296 compounds against the related trypanosome subspecies, Trypanosoma brucei brucei bloodstream form lister 427. Primary hits identified against T.b. brucei were retested and the IC50 value compounds were estimated for T.b. brucei and a mammalian cell line HEK293, to determine a selectivity index for each compound. The screening campaign identified 205 compounds with greater than 10 times selectivity against T.b. brucei. Cluster analysis of these compounds, taking into account chemical and structural properties required for drug-like compounds, afforded a panel of eight compounds for further biological analysis. These compounds had IC50 values ranging from 0.22 µM to 4 µM with associated selectivity indices ranging from 19 to greater than 345. Further testing against T.b. rhodesiense led to the selection of 6 compounds from 5 new chemical classes with activity against the causative species of HAT, which can be considered potential candidates for HAT early drug discovery. Structure activity relationship (SAR) mining revealed components of those hit compound structures that may be important for biological activity. Four of these compounds have undergone further testing to 1) determine whether they are cidal or static in vitro at the minimum inhibitory concentration (MIC), and 2) estimate the time to kill.
| Human African Sleeping Sickness (HAT) is a disease caused by sub-species of Trypanosoma. The disease affects developing countries within Africa, mainly occurring in rural regions that lack resources to purchase drugs for treatment. Drugs that are currently available have significant side effects, and treatment regimes are lengthy and not always transferrable to the field. In consideration of these factors, new drugs are urgently needed for the treatment of HAT. To discover compounds suitable for drug discovery, cultured trypanosomes can be tested against libraries of compounds to identify candidates for further biological analysis. We have utilised a 384-well format, Alamar Blue viability assay to screen a large non-proprietary compound collection against Trypanosoma brucei brucei bloodstream form lister 427. The assay was shown to be reproducible, with reference compounds exhibiting activity in agreement with previously published results. Primary screening hits were retested against T.b. brucei and HEK293 mammalian cells in order to assess selectivity against the parasite. Selective hits were characterised by chemical analysis, taking into consideration drug-like properties amenable to further progression. Priority compounds were tested against a panel of protozoan parasites, including Trypanosoma brucei rhodesiense, Trypanosoma cruzi, Leishmania donovani and Plasmodium falciparum. Five new compound classes were discovered that are amenable to progression in the drug discovery process for HAT.
| Human African Trypanosomiasis (HAT) is caused by infection with either the trypanosome subspecies Trypanosoma brucei gambiense or Trypanosoma brucei rhodesiense. Decreasing numbers of reported new cases over the last 10 years have been reported - from over 25,000 in 2000 to 10,000 in 2009 - of which over 95% are caused by T.b. gambiense [1]. However, the World Health Organization (WHO) currently estimates the actual number of cases to be around 30,000 [http://www.who.int/mediacentre/factsheets/fs259/en/]. HAT is mainly confined within sub-Saharan Africa, where the vector, the parasite and the animal reservoirs co-exist [2]. HAT occurs in two stages, whereby the first stage, also called the haemolymphatic stage, corresponds to the invasion of lymph, blood and other tissues by the trypanosomes, and the second stage is associated with parasites crossing the blood-brain barrier and invading the central nervous system (CNS). Symptoms of the second stage of the disease include mental impairment, severe headaches, fever, chronic encephalopathy and an eventual, terminal somnolent state, if the disease remains untreated.
There are currently few drugs available for the treatment of HAT. For the first stage of the disease, suramin is used as the treatment for T.b. rhodesiense and pentamidine for T.b. gambiense infections. Neither of these drugs are able to cross the blood brain barrier and therefore are not effective against the CNS resident, second stage of the disease. In addition, both of these treatments have significant side effects, often resulting in reduced compliance. Suramin is associated with exfoliative dermatitis [2] and renal failure [3], whilst pentamidine use has been correlated with diabetes mellitus and nephrotoxicity [4]. Melarsoprol, an organoarsenic compound, is most frequently used for the treatment of the second stage of the disease as it is effective against both trypanosome subspecies. However, there have been reports of high failure rates with melarsoprol, and although resistance has not definitively been proven, this does highlight the need for alternative therapies [5]. As a consequence of treatment with melarsoprol, encephalopathic syndromes occur in 5 to 10% of all of treated patients causing between 10 to 70% fatality, depending on the literature source [6]–[10]. The alternative therapy for the second stage of the disease, eflornithine, is a less toxic and a safer alternative however it is unfortunately not effective against T.b. rhodesiense. There are also problems with affordability of eflornithine in many of the disease-endemic countries [11]. The recent inclusion of nifurtimox to the WHO Essential Medicine List in 2009 [11], [12], to be used only in combination with eflornithine for the treatment for the second stage of HAT caused by T.b. gambiense, is a significant milestone. Nifurtimox-eflornithine combination therapy (NECT) has a shorter and simplified administration regimen and is the only significantly improved treatment option made available to patients in the past 25 years. NECT is now used as the first line treatment for stage 2 HAT caused by T.b. gambiense [13], [14]. There was some hope for an oral drug for treating the first stage of HAT with the compound, pafuramidine (DB289). Unfortunately, in an extended phase III trial, liver toxicity and delayed renal insufficiency were observed in a number of patients and consequently the program was discontinued in 2008 [15]. Recent advances which hold promise include the identification of orally bioavailable oxaborole 6-carboxamides which have been shown to cure a murine model of late stage CNS HAT [16] and an orally active benzoxaborole has been selected to enter pre-clinical studies [17]. Despite this there is still a need for the discovery of additional trypanocidal compounds with the potential for further progression in the drug discovery pipeline for HAT. This is particularly evident when one takes into account the toxicity of traditional treatments, the inability of the newer less toxic combination therapies to treat both subspecies or both disease stages, and the historical 90% failure rate of drugs entering the clinic to reach the market [18].
One method for the identification of active compounds against HAT is the application of high throughput screening (HTS) methods. HTS against T.b. brucei targets, such as the enzyme TbHK1 (Trypanosoma brucei hexokinase 1) [19] have recently been reported. A potential drawback to target-based HTS is that screening hits may have to undergo significant medicinal chemistry optimisation to impart favourable properties for low serum binding, high membrane permeability and high aqueous solubility in order to register potent activity against the parasite. Whole cell screening is becoming increasingly popular, as although elucidation of the biological target requires deconvolution, active compounds are discovered under conditions that are already physiologically relevant. We have recently reported the development of a 384 Alamar Blue based 384-well viability assay for HTS screening of compounds against T.b. brucei [20]. For this assay, and indeed many in vitro models for studies of HAT, the human non-infective sub-species T.b. brucei blood stream form has been utilised [21]. Alamar Blue (containing resazurin) is a fluorometric/colorimetric REDOX indicator. In a reducing environment caused by metabolising cells, resazurin is converted to resorufin, a fluorescent end product. This reagent has been used routinely as an indicator of the viability of mammalian cells. It is thought that cells may induce a reduction in the medium or reduce Alamar Blue intracellularly [22]. We have shown that the fluorescent Alamar Blue signal is linear to the number of T.b. brucei cells in a well, therefore it provides a good indication of viable cell numbers [20]. For this reason we have used this assay to assess the activity of compounds against T.b. brucei whole cells.
Here we describe the HTS of a compound library (WEHI 2003 collection [23]) using a 384-well whole cell T.b. brucei assay, and the retesting of the identified active compounds against both T.b. brucei and a human cell line, HEK293, in order to assess mammalian cytotoxicity. The reproducibility of both the primary and retest assays were evaluated by the Z'-factor (Z'), a coefficient which reflects the reproducibility of the assay and is calculated using the positive and negative controls. The Z' takes into account the control signal range and variation, with a value close to 1 considered highly reproducible [24]. Reference compound activities for the T.b. brucei assay were compared with previously published results for the same assay format [20], [25]. Selectively active compounds were subjected to rigorous chemical analysis taking into account drug like and non-drug like structural properties. The selectivity index (SI) was defined as the HEK293 IC50 values divided by the T.b. brucei IC50 value. The compounds selected, with the initial criteria of an SI of greater than 10 times, were ultimately shown to have SI values ranging from 19 and a predicted value greater than 345. Further testing against T.b. rhodesiense revealed five new classes of active compounds that are recommended as chemical leads for the potential development of therapeutics against HAT. SAR mining revealed components of these hit compound structures that may be important for the observed biological activity, and these will be outlined. Based on compound availability, four compounds were selected for further biological profiling by estimating the time to kill and assessment if the compound action is cidal.
T.b. brucei lister 427 cells [26] were maintained in log phase growth in 25 cm2 tissue culture flasks (Corning, NY, USA) by sub-culturing at either 24 or 48 hour intervals. Cells were grown in HMI-9 medium [27], supplemented with 10% fetal calf serum (FCS) and 100 IU/ml penicillin/streptomycin (Invitrogen, Carlsbad, California, USA) with incubation at 5% CO2 at 37°C in humidified conditions. HEK293 cells were maintained in high glucose DMEM with L-glutamine, supplemented with 1× non-essential amino acids (NEAA; Invitrogen, USA) and 1 mM sodium pyruvate. Growth conditions were in 5% CO2 at 37°C, under humidified conditions.
All reagent and cell additions were made with a Multidrop liquid handler (Thermo Scientific, Newington, NH, USA) under sterile conditions. Fifty-five microliters of 2000 cells/mL of T.b. brucei in HMI-9 medium were added to a black, clear-bottomed 384-well lidded plate (BD Biosciences, Franklin Lanes, NJ, USA). Cells were incubated for 24 hours at 37°C in an atmosphere of 5% CO2 before addition of 5 µl of compounds/DMSO for control wells. Compounds suspended in 100% DMSO or 100% DMSO as controls were pre-diluted 1∶21 in high glucose DMEM without FCS by using a Minitrack robotic liquid handler (PerkinElmer, Waltham, MA, USA). Five microliters of diluted sample was added to the plate to give a final DMSO concentration of 0.417% in the assay. Cells were incubated for an additional 48 hours at 37°C. Ten microliters of 70% Alamar Blue (Biosource, Bethesda, MD, USA) was added to each well (diluted in HMI-9 medium supplemented with 10% FCS) to a final concentration of 10% in the assay. The plate was incubated for two hours under the same conditions, then incubated for 22 hours in the dark at room temperature. Wells were read at 535 nm (excitation) and 590 nm (emission) wavelengths on a Victor II Wallac plate reader (PerkinElmer, USA). Specific dilutions are explained further in the primary and retest assay methodology. Reference drugs used in the assay were pentamidine (Sigma-Aldrich, St Louis, MO, USA), diminazene aceturate (Sigma-Aldrich, USA) and puromycin (Calbiochem, San Deigo, CA, USA). Pentamidine is used to treat patients with HAT and diminazene is a veterinary drug used against T.b. brucei to combat infections in cattle. Puromycin is a non selective, protein synthesis inhibitor.
Cells at 80% confluence were harvested and diluted in growth medium (high glucose DMEM supplemented with 10% FCS) to 7.27×104 cells/ml. Under sterile conditions, 55 µl of diluted cells per well were added to a black, clear bottomed 384- well lidded plate (BD Biosciences, Bedford, MA, USA) with a Multidrop liquid handler (Thermo Scientific, Barrington, IL, USA). Incubation times, compound additions and plate read were as per the trypanosome viability assay, with the exception that Alamar Blue was diluted in HEK293 growth media before addition, and incubation of Alamar Blue at 37°C, in 5% CO2, was for 4 hours, followed by incubation at room temperature for 20 hours. The activity of compounds against HEK293 cells was used to calculate the SI of mammalian to T.b. brucei cells. The control compound for HEK293 cells was puromycin (Calbiochem, USA).
L6 rat skeletal myoblasts [28], [29] were purchased from the American Type Culture Collection (ATCC, Rockville, MD, USA; ATCC number CRL 1458). This cell line was used for cytotoxicity testing to calculate an SI against T.b. rhodesiense and screened alongside the T.b. rhodesiense, P. falciparum, T. cruzi and L. donovani assays. L6 were also the host cells for the T. cruzi assay. Assays were performed in 96-well microtiter plates, each well containing 100 µl of RPMI 1640 medium supplemented with 1% L-glutamine (200 mM), 10% FCS, and 4000 L6 cells. Serial drug dilutions of eleven 3-fold dilution steps covering a range from 100 to 0.002 µg/ml were prepared. After 70 hours of incubation the plates were inspected under an inverted microscope to assure growth of the controls and sterile conditions. Ten µl of resazurin solution (resazurin, 12.5 mg in 100 ml double-distilled water) was then added to each well and the plates incubated for another 2 hours. Then the plates were read with a Spectramax Gemini XS microplate fluorometer (Molecular Devices Cooperation, Sunnyvale, CA, USA) using an excitation wavelength of 536 nm and an emission wavelength of 588 nm. Data was analysed using the microplate reader software Softmax Pro (Molecular Devices, USA). Podophyllotoxin was used as a positive control in the assay.
T.b. rhodesiense STIB900 stock was isolated in 1982 from a human patient in Tanzania and after several mouse passages cloned and adapted to axenic culture conditions [30]. Fifty microliters of Minimum Essential Medium (MEM) supplemented with 25 mM HEPES, 1 g/l additional glucose, 1% MEM non-essential amino acids (100×), 0.2 mM 2-mercaptoethanol, 1 mM Na-pyruvate and 15% heat inactivated horse serum was added to each well of a 96-well microtiter plate. Serial drug dilutions of eleven 3-fold dilution steps covering a range from 100 to 0.002 µg/ml were prepared. Four thousand bloodstream form cells of T.b. rhodesiense STIB 900 in 50 µl were added to each well and the plate incubated at 37°C under a 5% CO2 atmosphere for 70 hours. Ten microlitres of resazurin solution (resazurin, 12.5 mg in 100 ml double-distilled water) was then added to each well and incubation continued for a further 2–4 hours [31]. Plates were then read with a Spectramax Gemini XS microplate fluorometer (Molecular Devices, USA) using an excitation wavelength of 536 nm and an emission wavelength of 588 nm. Data was analysed using the microplate reader software Softmax Pro (Molecular Devices, USA). The drug melarsoprol was a positive control against T.b. rhodesiense.
Rat skeletal myoblasts (L6 cells) were seeded in 96-well microtitre plates at 2000 cells/well in 100 µl RPMI 1640 medium with 10% FCS and 2 mM l-glutamine. After 24 hours the medium was removed and replaced by 100 µl per well containing 5000 trypomastigote forms of T. cruzi Tulahuen strain C2C4 containing the β-galactosidase (Lac Z) gene [32]. After 48 hours the medium was removed from the wells and replaced by 100 µl fresh medium with or without a serial drug dilution of eleven 3-fold dilution steps covering a range from 100 to 0.002 µg/ml. After 96 hours of incubation the plates were inspected under an inverted microscope to assure growth of the controls and sterility. Then 50 µl of the substrate, containing chlorophenol red-β-D-galactopyranoside (CPRG) and Nonidet, was added to all wells. A colour reaction developed within 2–6 hours that could be read photometrically at 540 nm. Data were transferred into the graphic programme Softmax Pro (Molecular Devices, USA), which calculated IC50 values. The drug benznidazole was used as a positive standard in this assay.
Amastigotes of L. donovani strain MHOM/ET/67/L82 were grown in axenic culture at 37°C in SM medium [33] at pH 5.4 supplemented with 10% heat-inactivated FCS under an atmosphere of 5% CO2 in air. One hundred µl of culture medium containing 105 amastigotes from axenic culture with or without a serial drug dilution were seeded in 96-well microtitre plates. Serial drug dilutions of eleven 3-fold dilution steps covering a range from 100 to 0.002 µg/ml were prepared. After 70 hours of incubation the plates were inspected under an inverted microscope to assure growth of the controls and sterile conditions. Ten µl of resazurin solution (12.5 mg resazurin dissolved in 100 ml distilled water) [34] were then added to each well and the plates incubated for another 2 hours. The plates were then read with a Spectramax Gemini XS microplate fluorometer (Molecular Devices, USA) using an excitation wavelength of 536 nm and an emission wavelength of 588 nm. Data was analysed using the software Softmax Pro (Molecular Devices, USA). Decrease of fluorescence ( = inhibition) was expressed as percentage of the fluorescence of control cultures and plotted against the drug concentrations. From the sigmoidal inhibition curves the IC50 values were calculated. Miltefosine served as a known drug control in this assay.
In vitro activity against erythrocytic stages of P. falciparum was determined using a 3H-hypoxanthine incorporation assay [35], [36] using the chloroquine and pyrimethamine resistant K1 strain that originates from Thailand [37]. Compounds dissolved in DMSO at 10 mg/ml were added to parasite cultures incubated in RPMI 1640 medium without hypoxanthine, supplemented with HEPES (5.94 g/l), NaHCO3 (2.1 g/l), neomycin (100 U/ml), Albumax (5 g/l) and washed human A+ red blood cells at 2.5% haematocrit (0.3% parasitaemia). Serial drug dilutions of eleven 3-fold dilution steps covering a range from 100 to 0.002 µg/ml were prepared. The 96-well plates were incubated in a humidified atmosphere at 37°C; 4% CO2, 3% O2, 93% N2. After 48 hours, 50 µl of 3H-hypoxanthine ( = 0.5 µCi) was added to each well of the plate. The plates were incubated for a further 24 hours under the same conditions. The plates were then harvested with a Betaplate cell harvester (Wallac, Zurich, Switzerland), and the red blood cells transferred onto a glass fibre filter then washed with distilled water. The dried filters were inserted into a plastic foil with 10 ml of scintillation fluid, and counted in a Betaplate liquid scintillation counter (Wallac, Zurich, Switzerland). IC50 values were calculated from sigmoidal inhibition curves using Microsoft Excel. Chloroquine was used as a positive control in the hypoxanthine assay.
Primary screening of the library, consisting of 87,296 compounds in two hundred and forty eight 384-well plates, was undertaken in single point against T.b. brucei. Stock solutions consisted of test compound at a concentration of 5 mM in 100% DMSO. One µl of each compound stock solution was diluted by the addition of 40 µl of dilution medium (high glucose DMEM without FCS) by a multidrop liquid handler (Thermo Scientific, USA). A 5 µl sample of this diluted solution was then added to the trypanosome assay plate. The final concentration of test compound in the assay was 10.2 µM and that of DMSO was 0.42% v/v. Compounds were screened over a total of 11 days, at an average of 80 plates per day, taking into consideration that the assay incubation was 3 days total. Test compounds were added to plates in batches of 20 at two hour intervals, to maintain the timing of additions and reads.
Compound activity was calculated as the percentage inhibition in relation to positive and negative controls. The positive control, pentamidine, was contained in whole control plates, separate to the plates containing compounds, and the negative control (no effect) comprised of 0.42% DMSO, in column 24 of each test compound assay plate. These in-plate negative controls were used in an effort to normalise compound activity in relation to any plate to plate variation in the assay signal. A whole 384-well control plate was included in each day's screening, one per 20 compound plates containing half a plate of 2 µM pentamidine for the positive assay control, and half a plate of 0.42% v/v DMSO as a negative control. The positive assay control was used to calculate compound activity for batches of 20 compound plates. As well providing the positive control data, these whole plate controls were used for the calculation of the Z' to measure the reproducibility of the assay. An active hit was defined as a compound that demonstrated greater than the mean percentage activity of the library, plus three times the standard deviation. A separate plate containing a 13 point dose-response of reference compounds in triplicate was also included per 20 test plates to calculate the sensitivity of the assay.
Compounds identified from primary screening were retested against both T.b. brucei and HEK293 cells in duplicate and at varying concentrations to obtain a dose-response curve. Thus, a 5 µl sample of fresh compound stock solution (5 mM in DMSO) was cherry picked into 384-well plates and diluted 1∶10 in dilution medium (high glucose DMEM without FCS). Serial dilutions of these samples were then prepared in the same media by a Minitrak robotic liquid handler (Perkin Elmer, USA). This resulted in a total 13 doses per sample with 41.7 µM as the highest concentration of test compound, for which the DMSO concentration was 0.83% v/v. A screening dose of 10.4 µM was included in the dilution series to enable reconfirmation of primary screening T.b. brucei activity. Compounds with activity against T.b. brucei ≤10 µM, which also displayed an SI of ≥10, were selected for medicinal chemistry analysis.
The DMSO working concentration in the serial dilutions was maintained at 5%, giving a final assay concentration of 0.42% DMSO, except for the 41.7 µM test compound solution, where as previously stated the corresponding DMSO concentration was 0.83% v/v. The concentration of DMSO that can be tolerated in the T.b. brucei assay has been previously determined as 0.42% [20]. Therefore the 41.7 µM test compound sample with 0.83% v/v DMSO was not used in the T.b. brucei assay and thus the top test compound concentration in this assay was 20.8 µM. However, as the HEK293 assay can tolerate 0.83% DMSO (results not shown) the highest test compound concentration of 41.7 µM with 0.83% DMSO was included in the HEK293 assay in order to maximise the chances of deriving an IC50 value for more weakly cytotoxic compounds.
Compound activity in the retest campaign was calculated as percentage inhibition in relation to positive and negative controls, in the same manner as the primary screening campaign. The positive controls, pentamidine (2 µM final concentration) for T.b. brucei and puromycin (8 µM final concentration) for HEK293, were both screened in whole 384-well control plates. Whole 384-well plate controls were included after every batch of 20 compound plates, and were comprised of half a plate of negative control and half a plate of positive control. The negative control was the vehicle, 0.42 µM DMSO. The negative control was also included in column 24 of each compound assay plate to determine signal variation from plate to plate and to calculate compound activity. The exception was the 41.7 µM compound dose used in the HEK293 assay. In these plates column 24 contained 0.83% DMSO as a negative control. A separate 384-well plate containing a 13 point dose-response of the reference compounds puromycin, pentamidine and diminazene, in triplicate, was also included per 20 compound test plate batch to estimate assay sensitivity.
Cluster analysis of the active compounds (n = 205) identified and confirmed from the primary and follow up retest campaign was performed using Pipeline Pilot. A predefined functional class fingerprinting set (FCFP_6, average number of molecules per cluster = 5, max distance to center = 0.6) was applied, followed by the removal of compounds carrying toxicophores (n = 35) or permanent charge (n = 25) based on filters developed in-house, which include a list of 110 undesirable chemical moieties. The remaining clusters (n = 93, total of 137 compounds) were then independently scored by 3 medicinal chemists with industrial experience. Scoring was based on criteria including activity and selectivity, number of active analogues in the cluster, drug-like structural features, chemical tractability, presence of additional toxicophores not detected by the previously applied filters, potential for CNS penetration, and possible overlap with scaffolds already considered for HAT development at DNDi, or the literature.
Following medicinal chemistry analysis, compounds deemed to be of most interest were re-purchased or re-synthesised and analysed by liquid chromatography-mass spectrophotometry (LCMS) to confirm expected molecular weight and acceptable purity (>85%) prior to retesting of biological activity. These compounds were retested as in dose for N of three replicates, as described above for both T.b. brucei and HEK293.
For SAR mining, hit compounds were compared structurally to the whole primary screening compound collection. A series of substructure searches, performed in ActivityBase, were defined and refined to retrieve analogues most relevant to SAR interpretation. We have undertaken SAR mining for more than 60 HTS campaigns and have found substructure searching to return more meaningful SAR-relevant analogues than similarity searching. This is not surprising as it is well known that fingerprint-derived structural recognition captures medicinal chemistry-based structural recognition in only a rudimentary fashion [38]. The substructures that were used for searches are shown in Figure 1. A1 and A2 were the basis for searches for analogues of compounds 1 and 2, B1–B3 were used for compound 3, C1 for compound 6, D1 for compound 8, and E1–E3 for compound 7.
For compounds 1, 2, 6 and 7, the minimum inhibitory concentration (MIC) was determined from a concentration response curve, generated using the Alamar Blue assay. The MIC was extrapolated as the minimum concentration at which there was a plateau of activity in the assay (>95% activity). For compound 1, this was 3.97 µM, compound 2 was 19.8 µM, compound 6 was 9.92 µM and compound 7 had a MIC of 0.99 µM. To determine cell counts at this MIC, compounds at their MIC concentration were added following incubation of 2×103 parasites per well for 24 hours in the absence of compound. Cell numbers were determined after 24, 48 and 72 hours exposure to compounds and compared to controls of puromycin, also at an MIC concentration (1.15 µM). Puromycin was used as a positive control for 100% cell death (cidal action), as since for this drug there were no parasites remaining in the treated wells following 24 hours. The MIC calculated for pentamidine was 0.04 µM.
IC50 values were determined for compounds 1, 2, 6 and 7 following exposure of T.b. brucei to each compound for 29, 48 and 72 hours. The starting dose was 40 µM, and the IC50 values were determined from a 16 point dose-response curve. The assay conditions were the same as previously described for the Alamar Blue assay, except that 10 µL of a final 10% concentration of Presto Blue in HMI-9 medium was added as the indicator of viable cells, at various time points. At the first time point, following 20 hours of incubation with compounds, Presto Blue was added to the wells and incubation at 37°C continued. Plates were read every hour and returned to continue incubation at 37°C. This was performed to determine at which time point there was a reproducible signal (Z' of >0.5), using puromycin as a negative control and 0.42% DMSO as a positive control. This corresponded to a 9 hour incubation, or 29 hours incubation in the presence of the compound. After 45 hours incubation, Presto Blue reagent was added, and the samples incubated for an additional 3 hours, thus read at 48 hours to give a reproducible signal. Similarly, at 70 hours, reagent was added, samples incubated for another 2 hours and read at 72 hours. If a compound reached a plateau of activity and no cells were identified at the MIC, compounds were considered to have been effectively cidal at that time point.
As a hit threshold, three times the standard deviation plus the mean of the activity of the compound collection was calculated at 50%, in an effort to reduce false positives in the assay. Compounds with ≥50% activity were therefore considered active. From the primary screening campaign, 1,980 compounds inhibited T.b. brucei growth by ≥50%, a hit rate of 2.27%. These were grouped into two classifications, the first containing those compounds that inhibited growth between ≥50% and <80% and the second consisted of compounds with inhibitory activity of ≥80%. Group two was comprised of 1,217 compounds and it was these that were progressed to initial retesting. In-plate controls revealed little variation in the assay signal expressed as a ratio of maximum signal to background, throughout the entire test period (Figure 2). From separate whole plate controls, the Z' was calculated as an average of 0.81±0.05 (Figure 3). IC50 values for each of the reference compounds determined over the four screening days are shown in Figure 4. For each screening day, there were four control plates containing a dose-response of each reference compound, in triplicate, starting at doses of: puromycin 120 µM, pentamidine 70 µM and diminazene 80 µM. One control plate was included for screening per 20 compound plate batch. Thus, there were 4 control plates each in the first 3 days of screening (80 library compound plates per day) and only 1 control plate for the last (8 library compound plates). Mean IC50 values and standard deviations were therefore calculated from 12 replicates each on days 1–3, and 3 replicates of dose-response of reference compounds on day 4. These values were not significantly different from one another, as determined by a one way ANOVA in GraphPad Prism, with a significant difference of P<0.05. An IC50 value was not considered reproducible if varying more than 3 times from the mean. All values fell within three times the mean. IC50 values for the reference compounds were 61.9±6.8 nM for puromycin, 65.4 nM±12.5 nM for diminazene and 14.7±4.7 nM for pentamidine.
Of the 1,217 primary actives that were retested from stock solutions in duplicate with dose-response curves, 822 compounds (67.5%) reconfirmed in duplicate to be ≥50% inhibitory in the T.b. brucei assay at the serial dilution concentration point of 10.4 µM (closest to the primary screening concentration of 10.2 µM). A dose-response plateau is necessary for IC50 values to be determined for these compounds. Hence, a compound needed to display ≥80% inhibitory at both 41.7 µM and 20.8 µM in duplicate (although one singleton was allowed to extend to ≥70%). There were 57.6% of the 1,217 compounds that passed these criteria. For all these compounds, titration data were imported into GraphPad Prism and the IC50 values estimated.
Similarly for the HEK293 assay, only data whereby an IC50 value could be estimated were imported in to GraphPad Prism. There were 700 compounds that displayed ≥80% inhibition at both 41.7 µM and 20.8 µM in duplicate (although one singleton was allowed to extend to ≥70%). As before, criteria included a plateau of activity necessary for the calculation of an IC50 value. The HEK293 IC50 value could be estimated for 10% of the 700 compounds in this manner and this allowed for the determination of the SI. For the remainder of compounds, an estimation of the IC50 against HEK293 cells was possible by observing the lowest concentration in the HEK293 assay that displayed ≥50% inhibition in at least one of the two replicates.
Using these analyses, there were 205 (29%) of the 700 re-confirmed compounds that had an estimated SI of ≥10. Of these compounds, 8 produced a non-sigmoidal curve in the T.b. brucei assay and therefore could not have an IC50 value, nor SI estimated. This may have been due to compound solubility, or the nature of the compound's action, and these compounds were de-prioritised. This left 197 hits that were progressed to medicinal chemistry cluster analysis.
Control plates, used as a measure of reproducibility, showed that the T.b. brucei assay had an average Z' of 0.74 (Figure 3). For the HEK293 assay, the mean Z' was 0.73 for both 0.42% DMSO and 0.83% DMSO final assay concentrations. Puromycin was active on both cell lines with an IC50 of 138.5±15.7 nM against T.b. brucei and 1123±155 nM against HEK293. Puromycin, a known cytotoxic compound therefore exhibited an SI of less than 10, supporting the use of the Alamar Blue for the identification of cytotoxic compounds. For T.b. brucei, diminazene exhibited an IC50 value of 29.5±5.8 nM and pentamidine 7.8±3.6 nM. Neither pentamidine nor diminazene displayed activity in the HEK293 assay at the doses screened (1 µM and 40 µM, respectively).
Scoring was attributed independently by 3 medicinal chemists with industrial experience and was based on criteria including criterion 1: activity and selectivity (compounds with IC50 values indicating good activity and high selectivity were favored); criterion 2: number of active analogues in the cluster (cluster with n>1 were preferred to singletons), criterion 3: drug-like structural features (based on Lipinski's rule of 5 scoring [39]), criterion 4: chemical tractability (based on personal experience, as well as availability of commercial analogs), criterion 5: presence of additional toxicophores not detected by the previously applied filters (personal experience), criterion 6: potential for CNS penetration such as molecules with low PSA, low molecular weight, low clogP and low number of H-bond donor/acceptors [40]. It is recognised that this method is internally consistent, however may differ from analysis undertaken by other medicinal chemists [41]. This analysis lead to the selection of 11 compounds for retesting.
The 11 compounds identified from medicinal chemistry analysis were either re-synthesised or re-purchased, and re-tested in both the T.b. brucei and HEK293 assays. Following this, the number was reduced to 8 (Table 1) after two resupplied compounds did not confirm activity in the T.b. brucei assay (<50% activity, results not shown). A third compound was found to only be >50% active at the top dose of 65 µM and therefore was unsuitable for IC50 or SI calculation. The IC50 values and calculated selectivity indices of the remaining 8 compounds are outlined in Table 1, and the structures in Figure 5. During rescreening of resynthesized compounds the Z' for the T.b. brucei assay was 0.81±0.02 and 0.88±0.01 for the HEK293 assay. In the T.b. brucei assay, pentamidine displayed an IC50 value of 3.52±0.36 nM, diminazene 121±9.04 nM and puromycin 58.4±0.77 nM (Table 1). In the HEK293 assay, puromycin was active at 518±28.1 nM, whilst as expected neither pentamidine nor diminazene displayed activity at the doses screened (1 µM and 40 µM, respectively). The selectivity index for puromycin was similar to that found at original retest, (8.9 fold, Table 1), as expected for a non-selective inhibitor.
The compounds identified by medicinal chemistry analysis as the most promising were also tested in dose-response against the human infective parasites T.b. rhodesiense, L. donovani and T. cruzi to estimate IC50 values. Data obtained is shown in Table 1. Rat skeletal L6 muscle cells were also used as an indicator of cytotoxicity and the SI was calculated against all species. Initial analysis of compound activity was made against the HAT reference strain, T.b. rhodesiense, taking into consideration the IC50 value and the SI. Criteria used were as described for the primary screening and retest campaigns, therefore for compounds to be initially considered as favourable hits for further progression, the IC50 cut off was <10 µM and the SI>10. Compound 5 had an IC50 value <10 µM and a corresponding SI of <10, and thus was de-prioritised. Compound 4 displayed an SI of 0.19 and therefore was also de-prioritised. This left a panel of 6 compounds to be considered for further progression. Table 2 shows the physiochemical properties of these 6 prioritised compounds: the molecular weight, aqueous solubility, polar surface area and cLogP.
Reference compounds were used as controls throughout testing with all of these assays and are also shown in Table 1. For the T.b. rhodesiense assay, the drug melarsoprol displayed an IC50 of 6.28±1.78 nM. Benznidazole, a drug used to treat Chagas disease, was 1680±1930 nM active and miltefosine, a treatment for Leishmaniasis had an IC50 value of 365±93.7 nM. The drug chloroquine was active against P. falciparum with an IC50 of 164±24.7 nM.
For the 6 hit compounds, ActivityBase was used for substructure searching to identify the relevant analogues to associate with the primary screening data. The refined substructures used for searches are shown in Figure 1. Tables S1, S2, S3 and S4 show structure and activities of these compounds over the T.b. brucei primary screening and retest campaigns. Identified analogues are shown in Supplementary Table S1 (compounds 1 and 2), Table S2 (compound 3), Table S3 (compound 6) and Table S4 shows analogues of compound 7. No analogues of compound 8 were found in the library using these methods, even using the relatively broad substructure definition D1 in Figure 1.
Measurements of the number of T.b. brucei cells, following exposure to the MIC of compounds 1, 2, 6 and 7 during a 72 hour period, are shown in Figure 6. Treatment with three of the 4 compounds at the MIC for 24 hours resulted in cell counts indicating the complete lack of viable trypanosomes. However, the compound pyrido-isoxazol-2-ylanilide (compound 7), only cleared parasites following 72 hours incubation. At the MIC of puromycin, no cells remained following 24 hours treatment, whereas with pentamidine this effect was not observed until 72 hours incubation at the MIC.
The IC50 values for all 4 compounds selected for cidal assessment did not differ between the Presto Blue assay at 72 hours and the Alamar Blue assay (total compound exposure in this assay is also 72 hours), as shown in Table 3. Thus the Presto Blue assay was considered to also be an accurate indicator of compound activity measured over time and the results were comparable to IC50 values determined in the Alamar Blue assay. All compounds were active at 29 hours, with a plateau of activity displayed in dose-response curves. Compounds 1, 2 and 7 showed similar IC50 value across all time points, while compound 6 reached a stable IC50 value at 48 hours incubation with the compound (Table 3). Puromycin and pentamidine were demonstrated to reach a maximum IC50 value after 48 hours exposure.
Due to the many problems associated with current existing treatments for HAT, in particular toxicity, treatment regimes and cost, there exists a tangible need for new compounds to be introduced into early HAT drug discovery. HTS has been utilised by a number of research groups for HAT to identify active compounds for the drug discovery process, however there are few incorporating the use, or development of HTS for whole cells [20], [42], [43]. The inclusion of an assay to estimate cytotoxicity as a part of a whole cell HTS campaign is an important consideration for the progression of potential compounds. Here we describe the utilisation of an Alamar Blue HTS assay [20] to successfully screen a library of almost 90,000 small molecules. Following medicinal chemistry analysis of the positive hits in the assay, eight compounds with activity against T.b. brucei were identified. These compounds had IC50 values ranging from 0.22 µM to 4 µM with associated selectivity indices ranging from 19 to greater than 345.
Both the primary and retest screening campaigns were reproducible as exemplified by the statistical coefficient of the Z'. For the primary screening campaign, the Z' was averaged at 0.81 for the T.b. brucei assay (Figure 3). At retest the respective Z' values were 0.74 and 0.73 for T.b. brucei and HEK293 assays. Throughout the campaign, reference compounds in the T.b. brucei assay were within the range of the IC50 value of previously reported results for the same assay format [20]. The reproducibility of the reference compounds over primary screening days is highlighted in Figure 4. The HEK293 assay was validated in this campaign as effective for the identification of cytotoxic compounds by the activity of the compound puromycin. Puromycin is a general cell growth inhibitor of both eukaryotic and prokaryotic cells which disrupts protein synthesis. It was active on both cell lines in the retest screening campaign with an IC50 of 138.5±15.7 nM against T.b. brucei and 1123±155 nM against HEK293. This compound would therefore have been correctly identified as non-specifically cytotoxic by our criteria that a potentially useful T.b. brucei active must have an initial SI >10. This was also shown through the data obtained for the controls from during screening of re-isolated compounds, where the SI for puromycin was 8.9 (Table 1). As anticipated, neither pentamidine nor diminazene, which are registered drugs against HAT and T.b. brucei, respectively, exhibited activity in the HEK293 assay at the doses screened. Diminazene is reported to have an SI of 692 [44], whilst pentamidine has low µM activity reported for some mammalian cell lines [45].
The 8 compounds identified following reconfirmation of actives from new solids and chemical clustering were subjected to testing against the human HAT infective species T.b. rhodesiense, as well as other protozoal species that cause disease such as T. cruzi (Chagas disease), L. donovani (Leishmaniasis) and a chloroquine and pyrimethamine resistant strain of P. falciparum (Malaria). The structures and chemical classes of these compounds, designated compounds 1 to 8, are shown in Figure 5. As an additional mammalian cytotoxicity control and one relevant when screening these additional assays for protozoal parasites, the rat skeletal myoblast L6 cell line was used as this cell line is the host cell line used for the T. cruzi assay. The biological activities of these 8 compounds against T.b. brucei, a panel of human infective parasite species, plus the L6 cytoxicity data, with corresponding HEK293 selectivity indices are shown in Table 1. The activity of the relevant control/reference drugs has also been included. On the basis of this data, 2 compounds displayed relatively low (Table 1, compound 5) or extremely low (Table 1, compound 4) SI and thus were not considered favourable for progression. This left 6 high priority compounds, representing 5 distinct structural classes that could serve as a basis for progression in the early drug discovery process for HAT. Structures and key physicochemical properties for selected compounds are listed in Table 2. For analysis of physicochemical properties, a cLogP of 1–4 is considered favorable; >4–6 is acceptable, while >6 is unfavorable. A preferred solubility is considered to be >10 µM. Polar surface area is considered to be good at less than 70 Å2 and acceptable less than 80 Å2. A molecular weight lower than 400 is preferred in terms of lead-likeness and blood brain barrier crossing properties.
The phenylthiazole amide (compound 1) was active against T.b. brucei with an IC50 value of 0.79 µM and an SI of >96. It was similarly active against T.b. rhodesiense with an IC50 of 1.5 µM and an SI of 42. This compound also demonstrated activity against T. cruzi with an IC50 of 2.3 µM. In terms of physicochemical properties, it has a low molecular weight of 306, predicted good aqueous solubility of 63 µM, a low polar surface area of 42 Å2 suitable for CNS penetration, and a favourable cLogP of 3.4. Phenyltriazol-5-yl-ethylamide (compound 2), although closely related, was significantly less active against T.b. brucei with an IC50 value of 4.0 µM, with also a weaker T.b. rhodesiense activity IC50 of 6.8 µM. The SI for compound 2 determined against both HEK293 and L6 cells was approximately 20. The physicochemical properties of this compound reveal it to be of low molecular weight, with an acceptably low polar surface area of 71 Å2 and a favourable cLogP of 2.9, although the calculated aqueous solubility is low at 8 µM. A literature search revealed no biologically active compounds closely related to these two hit compounds, suggesting that these compounds may represent starting points for novel trypanocides. There were approximately 3 dozen compounds related to compound 1 (Table S1), approximately two dozen of which (1, 4–7, 16–24, 26–28, 33–36) were structurally very similar. Few of these exhibited any activity, suggesting tight SAR around the core structure. The exception to this was the potent thiophene-containing compound (entry 23), that did not initially pass the medicinal chemistry functional group filters, because of the thiophene group. However, this compound still provides useful SAR and suggests that different hydrophobic amides may be tolerated in this region with retention of potent activity. Remaining compounds were more distant, conformationally constrained, or contained heterocyclic alternatives to the thiazole and none were active.
The phenoxymethylbenzamide (compound 3) had moderate activity against T.b. brucei with a retest IC50 of 1.1 µM and an SI of >67. It was similarly active against T.b. rhodesiense with an IC50 of 0.85 µM and an SI of 60. In terms of physicochemical parameters, this compound has a moderately low molecular weight of 353, a calculated aqueous solubility of 25 µM, a low polar surface area of 39 Å2 and an acceptable cLogP of 4.8. SAR mining revealed 34 analogues related to compound 3 (Table S2). Some of these compounds had only relatively minor changes (entries 5, 11, 13, 33) but of these, only one (entry 13) showed some activity (77% at 10.4 µM), suggesting both ends of the molecule (piperidine amide and p-alkoxyphenyl) are likely to be important for activity. The remaining compounds tended to have more significant changes to both ends or the central unit and none of these were active except for one (entry 3), suggesting the piperidine could be replaced with a diaminoethane though cytotoxicity would need to be monitored.
The activity of the pyrimidin-2-yl-pyrazol-5-ylamide (compound 4) was 3.1 µM for T.b. brucei, with an SI of 25. This compound also demonstrated activity against T.b. rhodesiense IC50 of 4.8 µM but with an extremely low SI of 0.19 to L6 cells, and it was for this reason that this compound was not included in the top 6 compounds to be considered further. The 7-aminotetrahydroquinoline bis sulfonamide (compound 5) had a moderate retest T.b. brucei IC50 value of 2.1 µM and an SI of 36 to HEK293 cells. However the low activity observed against the infective species (T.b. rhodesiense) of 14 µM rendered this compound de-prioritised.
None of the entries 1, 2, 3 or 5 belong to classes associated with any known biological activities as far as the authors can ascertain. However, this is not the case for compound 6, 6-aryl-3-aminopyrazine-2-carboxamide, which was moderately active with a retest IC50 of 1.2 µM and an SI of >65 when cytotoxicity is measured on HEK293 cells. It was similarly active against T.b. rhodesiense with an IC50 of 0.97 µM and an SI to L6 cells of 18. This compound is predicted to have a favourable aqueous solubility of 3.2 mM, has a low molecular weight of 270, an acceptably low polar surface area of 81 Å2 and a favorable cLogP of 2.0. This class is quite heavily patented and associated with numerous biological activities [46]–[50]. Only one compound was a close analogue of compound 6, a des-N-alkyl carboxamide (Table S3), however this was inactive, suggesting the alkyl group is essential for activity.
The pyrido-isooxazol-2-ylanilide (compound 7) is an isoxazol-2-ylanilide with a fused pyridine ring and displays the best biological activity profile of all compounds, with a T.b. brucei retest IC50 value of 0.22 µM and an SI of >345. It was similarly active against T.b. rhodesiense with an IC50 of 0.59 µM and an SI of 39. This compound also displayed activity against T. cruzi with an IC50 of 0.23 µM and an IC50 against L. donovani of 1.8 µM, suggesting potential as a broad spectrum anti-kinetoplastid. The physicochemical properties of this compound are favourable, with a moderately low molecular weight, an acceptable polar surface area of 81 Å2 and a favourable cLogP of 2.6. The calculated aqueous solubility is low (25 nM) and it is possible the actual solubility may be improved due to the ortho effect of the 2-chloro substituent. This compound belongs to a class with an isolated report of biological activity, activation of the NAD+-dependent deacetylase SIRT1 [51], a sirtuin, which also appears to be present and important in trypanosomes [52]–[54]. This compound would appear to present a promising starting point for drug development, though early investigation of aqueous solubility and its improvement could be important. For this compound, there were 19 analogues that provided useful SAR (Table S4). Several compounds suggested the furan was important for activity, as replacement with substituted phenyl ring (entries 6, 10, 11, 16, 17, 19) or extension (entries 5, 7, 9, 12, 15, 18) led to inactive compounds. However, replacement with a simple propyl group (2) led to an active compound suggesting smaller hydrophobics may be acceptable. Two compounds had small changes in other parts of the molecule and were also inactive, suggesting even simple substitution changes to the central phenyl ring (entry 3) are not necessarily tolerated nor small changes to the distal pyridine ring (entry 4). While more significant in their alterations, all other analogues are still clearly related to the parent compound yet inactive, suggesting tight SAR.
The aminoethyl benzoyl arylguanidine (compound 8) displayed a T.b. brucei retest IC50 value of 2.6 µM and an SI of >29. This compound displayed increased activity against T.b. rhodesiense with an IC50 value of 0.88 µM and an SI of 150, whereas the activity against T. cruzi was low (IC50 of 82 µM). The biologically active conformation of this molecule may adopt an intra-molecular hydrogen-bonded form as shown [55], similar to benzoylureas [56]. In terms of physicochemical properties, this compound has a moderate molecular weight of 380, a reasonable aqueous solubility of 32 µM, an acceptably low polar surface area of 67 Å2 and a favorable cLogP of 3.1. Closely related compounds are patented as inhibitors of human mitochondrial F1Fo- ATPase [57], the same molecular target that DB289 has been suggested to target in T. brucei [58]. Oligomycin A, which is known to inhibit mitochondrial membrane associated ATPases in mammalian cells [59] has also demonstrated potent activity against T.b. brucei [60]. Oligomycin sensitive ATPases have been found to be present in T.b. brucei [61]. The aminoethyl benzoyl arylguanidine represents a highly tractable and attractive structure for medicinal chemistry optimization, although consideration will need to be given to the potential for liver toxicity manifested in DB289, and how this may be overcome [9]. Data mining showed there were no analogues of this compound in the library screened.
From the hit chemical classes, compounds 1, 2, 6 and 7 underwent further biological profiling to ascertain whether their action was cidal or static at the MIC determined. Of the 4 compounds profiled, all had completely cleared parasites in wells by 72 hours incubation at the MIC (Figure 6) and were therefore considered to have a cidal action. Compounds 1, 2 and 6 and the control puromycin resulted in complete depletion of trypanosomes at this dose at 24 hours, whilst compound 7 and the control compound, pentamidine, required a 72 hour incubation to attain the same effect.
To determine the IC50 values of compounds 1, 2, 6 and 7 over time, as an estimation of the kill time, the resazurin-based reagent, Presto Blue, was used. In the presence of live cells this dye converts more rapidly to a fluorescent end product, in comparison to Alamar Blue (results not shown). Dose-response curves of these compounds showed a plateau of activity of the 4 compounds at 24 hours (considered as 2 doses or more at >90%), suggesting that all compounds were active ≤29 hours. Compounds 1, 2 and 6 at MIC resulted in complete clearance of all parasites at ≤29 hours, with compounds 1 and 2 displaying the fastest cidal activity, with a maximum IC50 value reached at this point (Table 3). Compound 7 had similar IC50 values over each time interval investigated however at the MIC not all parasites were cleared until 72 hours. Although the MIC would shift slightly over time, at 24 and 48 hours there were 0.41% and 0.06% of the population remaining, respectively. Additional profiling revealed these compounds were cidal in nature and the speed of action was either similar to, or faster than, the known drug, pentamidine. These compounds will be profiled at reduced exposure times to determine if the time to kill may be less than the exposure times studied here. Estimation of the MIC at each time point would clarify complete parasite clearance.
Collation of all of the analyses completed led to the selection of five priority classes: phenylthiazol-4-ylethylamide, phenoxymethylbenzamide, 6-aryl-3-aminopyrazine-2-carboxamide, pyrido-isoxazol-2-ylanilide and aminoethyl benzoylarylguanidine. In summary, these compounds are novel scaffolds for HAT early drug development and represent attractive templates for further biological analysis and medicinal chemistry optimization, to build structure-activity relationships for compounds active against T.b. brucei. Upon confirmation of SAR, the chemistry program would be extended to optimize potency and solubility, in conjunction with early in vitro absorption, distribution, metabolism, elimination (ADME) and toxicity assays. Early pharmacokinetic studies (PK), to measure of brain compound levels, as well as in vivo efficacy studies in HAT murine models, would follow upon identification of suitable candidates. Medicinal chemistry efforts are being actively pursued to synthesize new compounds from the starting points discussed here, in a bid to generate leads with improved physicochemical and biological properties. Chemical structures and biological activities of all compounds defined as actives in the T.b. brucei primary screening campaign at ≥80% activity (1217), which were retested in dose-response in both the T. b. brucei assay and the HEK293 cytotoxicity assay are available in the CHEMBL-NTD database https://www.ebi.ac.uk/chemblntd.
|
10.1371/journal.pbio.1000534 | A Post-Burst Afterdepolarization Is Mediated by Group I Metabotropic Glutamate Receptor-Dependent Upregulation of Cav2.3 R-Type Calcium Channels in CA1 Pyramidal Neurons | Activation of group I metabotropic glutamate receptors (subtypes mGluR1 and mGluR5) regulates neural activity in a variety of ways. In CA1 pyramidal neurons, activation of group I mGluRs eliminates the post-burst afterhyperpolarization (AHP) and produces an afterdepolarization (ADP) in its place. Here we show that upregulation of Cav2.3 R-type calcium channels is responsible for a component of the ADP lasting several hundred milliseconds. This medium-duration ADP is rapidly and reversibly induced by activation of mGluR5 and requires activation of phospholipase C (PLC) and release of calcium from internal stores. Effects of mGluR activation on subthreshold membrane potential changes are negligible but are large following action potential firing. Furthermore, the medium ADP exhibits a biphasic activity dependence consisting of short-term facilitation and longer-term inhibition. These findings suggest that mGluRs may dramatically alter the firing of CA1 pyramidal neurons via a complex, activity-dependent modulation of Cav2.3 R-type channels that are activated during spiking at physiologically relevant rates and patterns.
| The hippocampus is an essential structure in the brain for the formation of new declarative memories. Understanding the cellular basis of memory formation, storage, and recall in the hippocampus requires a knowledge of the properties of the relevant neurons and how they are modulated by activity in the neural circuit. For many years, we have known that various chemical neurotransmitters can modulate the electrical excitability of neurons in the hippocampus. Here, we report new experiments to reveal how the chemical neurotransmitter glutamate increases neuronal excitability. The effect we study is the conversion of the afterhyperpolarization (a cellular consequence of firing an action potential) to an afterdepolarization. We identified the metabotropic glutamate receptors involved in this conversion (receptors called mGluR1 and mGluR5) as well as the final target of modulation (R-type calcium channels composed of Cav2.3 subunits), which cause the neurons to exhibit altered excitability in the presence of glutamate. We also determined some of the intermediate steps between activation of the glutamate receptors and modulation of the calcium channels responsible for the change in excitability, offering further mechanistic insight into how synaptic transmission can regulate cellular and network activity.
| Metabotropic glutamate receptors (mGluRs) are a class of G-protein coupled receptors that may mediate a variety of effects through presynaptic and postsynaptic actions. Because these receptors are activated by glutamatergic neurons during network activity, they are in a position to regulate neural function in an activity-dependent manner. The effects of mGluR activation may be rapid or long-lasting, so they are important for short-term and long-term regulation of neural activity [1]. They have been implicated in physiological functions, such as learning [2]–[5], as well as in a number of neurological disorders [6], including mental retardation, epilepsy, and Alzheimer's disease [7]–[11].
Among the many effects of mGluR activation, the modulation of neuronal excitability has a direct effect on the response of cortical pyramidal neurons to excitatory synaptic input. The effects of mGluRs on excitability are commonly mediated by group I mGluRs, resulting in modulation of voltage-gated Na+, Ca2+, or K+ currents, as well as Ca2+-activated K+ currents, nonselective cation currents, or ion exchanger currents [1]. Modulation of these targets by group I mGluRs typically increases postsynaptic excitability [12],[13]. Thus, group I mGluRs may modulate network function through modulation of multiple ion channels, resulting in enhanced excitability of glutamatergic pyramidal neurons.
Activation of group I mGluRs in cortical and hippocampal pyramidal neurons has been reported to reduce the post-spike AHP and induce an ADP [14]–[18], but the receptors, signal transduction mechanisms, and ion channels responsible for this effect are incompletely understood. Because of the importance of these modulatory effects for hippocampus-dependent functions and diseases, we studied the effects of activating group I mGluRs on the excitability of hippocampal CA1 pyramidal neurons. We report here that activation of these receptors results in enhanced activity of Cav2.3 R-type calcium channels, thus producing a medium ADP lasting a few hundred milliseconds. A slow ADP lasting for seconds is mediated by different mechanisms.
We obtained whole-cell current-clamp recordings from CA1 pyramidal neurons in rat hippocampal slices. The effects of the group I mGluR activation on responses to intracellular current injection were examined following bath application of the group I mGluR agonist DHPG (2–4 µM, see Materials and Methods). Step current injections (0.6-s long) that were just above threshold for action potential firing in control elicited increased action potential firing in the presence of DHPG (Figure 1A). In response to longer current injections (4.5 s), DHPG converted a simple pattern of spike-frequency accommodation to a more complex pattern consisting of a high-frequency burst of action potentials, followed by a period of silence, and finally a continuous train exhibiting spike-frequency accommodation. In response to noisy current injections, the number of action potentials was also increased when DHPG was applied (Figure 1B and 1C). DHPG produced a small but statistically significant increase in the input resistance (54.7±3.0 MΩ for control, t = 0 min; 59.7±4.1 MΩ for DHPG, t = 15 min; n = 26, paired t test, p<0.01). This did not noticeably change the subthreshold response to noisy current injection, but enhanced spiking appeared to be attributable to a reduction in the post-spike AHP in the presence of DHPG (Figure 1D).
The effect of DHPG on the post-spike AHP was studied systematically by examining responses to bursts of action potentials evoked by five brief (2 nA, 2 ms) current injections. In normal artificial cerebrospinal fluid (ACSF), a 100 Hz burst of five spikes was followed by an AHP (−3.1±0.2 mV) that reached a peak at 59±5 ms after the last spike. Application of DHPG (or quisqualate, another group I mGluR agonist, Figure S1) eliminated the AHP, resulting in a post-burst ADP (+18.7±0.7 mV) that reached a peak at 34±3 ms after the last spike and decayed to 25% of the peak value in 200±23 ms (n = 20, Figure 1E). We refer to this ADP as a medium ADP to distinguish it from the fast ADP following a single spike in normal ACSF [19]–[23] and a slow DHPG-induced ADP described later in this article. Functionally, the change from post-burst AHP into a post-burst medium ADP made the pyramidal neurons more excitable; current injections that were subthreshold during the post-burst AHP evoked action potential firing during the medium ADP (Figure 1F).
Application of DHPG resulted in a gradual reduction of the AHP and conversion to an ADP. The medium ADP reached its maximum value after several minutes in DHPG and was fully reversible, with a similar time course, upon washout of DHPG (Figure S2); however, the slow onset and reversal of the medium ADP is attributable to the slow speed of the perfusion system, as rapid application of DHPG produced more rapid responses (see below). In all experiments, the amplitude of the post-burst potential (AHP or medium ADP) was quantified at a fixed time, corresponding to the peak of the AHP in normal ACSF (59±5 ms after the last spike). For some analyses, the effect of DHPG was quantified by the change in the post-burst potential (Δ post-burst potential) at this time point in the response (see Figure S3). This measure required comparison of the response in normal ACSF (AHP at t = 0 min) to the response at a fixed time after application of DHPG (e.g., ADP at t = 15 min). Importantly, this measure was not complicated by any slow, drug-independent effects, as the post-burst AHP was stable for typical recording duration in normal ACSF (Figure S3).
To determine which group I mGluR subtype was responsible for the DHPG-induced medium ADP, we used the subtype-selective antagonists LY367385 (25 µM, mGluR1 antagonist) and MPEP (10 µM, mGluR5 antagonist). LY367385 blocked the medium ADP (t = 15 min) by 16%, while MPEP blocked it by 73% (Figure 2A), suggesting that the effect of DHPG is mediated primarily by mGluR5 and partially by mGluR1. When applied together, the two drugs blocked the medium ADP by 88% (Figure 2A), suggesting that the effects of mGluR1 and mGluR5 are approximately additive. To determine the subcellular localization of the mGluRs mediating the medium ADP, DHPG was applied locally, by pressure application from a large patch pipette, to either the perisomatic region or the apical dendrites (see Materials and Methods for details). Application of DHPG (500 µM in application pipette) to the apical dendrites had no effect, while perisomatic application produced a medium ADP (Figure 2B). Similarly, when DHPG was present in the bath, application of normal ACSF reduced the medium ADP when applied to the soma, but not when applied to the apical dendrites (Figure 2C). The effects of DHPG were induced within seconds of its application and reversed rapidly when the DHPG application was terminated. Together, these results suggest that perisomatic mGluRs must be activated to elicit an medium ADP in response to somatic action potential firing and that the effect is rapidly induced and reversed. The lack of effect with application to the dendrites does not necessarily imply the absence of group I mGluRs, as backpropagating action potentials could have a different effect from action potentials in the soma (see Discussion).
To determine whether group I mGluR-mediated modulation of the post-burst potential could occur in response to synaptically released glutamate, the post-burst AHP/ADP was compared during control conditions and during high-frequency (50 Hz) activation of Schaffer collaterals. Fast synaptic responses were prevented by blocking glutamate and GABA receptors (see Materials and Methods). Under control conditions, each burst was followed by an AHP. During synaptic stimulation, however, each burst was followed instead by an ADP (Figure 3A). To determine the role of group I mGluR activation in the induction of the post-burst ADP, we applied blockers of mGluR1 and mGluR5 (Figure 3B and 3C). The results of these experiments were compared to a separate group of control experiments occurring over the same time course but without blocker application. In the absence of synaptic stimulation, the post-burst AHP was stable over time, both in control and in the presence of group I mGluR blockers (Figure S4). In the presence of synaptic stimulation, the size of the ADP in the control group gradually increased over the course of the experiment (156%±12% of initial value, n = 5). By contrast, blocking the group I mGluRs resulted in a decrease in the size of the ADP (66%±7% of initial value, n = 5). The magnitude of the ADP at the end of the experiment (60 min) in the presence of group I mGluR blockers was 42% of control, consistent with a substantial contribution of these receptors to induction of the ADP triggered by synaptically released glutamate. The long-term effects of synaptic stimulation on the ADP are interesting but are not considered here. Instead, our focus is on the acute modulation of the AHP/ADP during activation of group I mGluRs.
Bath application of DHPG increased the fast ADP following a single spike, and the size of the medium ADP increased with both the number and the frequency of action potentials, reaching medium ADP values of nearly 15 mV for 5 spikes at 100 Hz (Figure S3). Longer trains (20 or 50 spikes) did not increase the ADP further and in fact resulted in a decrease in the size of the ADP (Figure S5), perhaps due to inactivation of the ADP-producing current or enhanced activation of an AHP-producing current.
The medium ADP following a burst of spikes evoked by a step current injection was blocked by application of tetrodotoxin (TTX; 0.5 µM) to eliminate Na+-mediated spikes; however, increasing the magnitude of the current injection to elicit a Ca2+ spike [24] restored the medium ADP (Figure 4A and 4B). Under these conditions, the amplitude of the medium ADP increased with the magnitude and duration of the current injection, reaching a maximum of about 10–15 mV for current injections of at least 1.4 nA for 40 ms (Figure 4C and 4D).
The requirement for action potential firing or a Ca2+ spike suggests that the medium ADP may require Ca2+ entry through voltage-gated calcium channels (VGCCs). Consistent with this idea, we found that the medium ADP was eliminated by switching to a Ca2+-free ACSF (Figure 5A) or by bath application of micromolar concentrations of NiCl2 (Figure 5B–E; IC50 = 23 µM). Nimodipine (10 µM, an L-type calcium channel blocker) did not block the medium ADP (Figure 5B–D). To test whether the medium ADP required elevation of internal Ca2+ concentration, we performed experiments with patch-clamp electrodes containing BAPTA (10 mM) and found that this strongly reduced the medium ADP (Figure S6). Blocking Ca2+ release from internal stores with cyclopiazonic acid (CPA; 20 µM) also reduced the medium ADP (Figure S6), suggesting that this is another important mechanism for induction of the medium ADP.
The requirement for Ca2+ entry through VGCCs and the elevation of internal Ca2+ concentration is consistent with two different models of the medium ADP. In the first model, the medium ADP is mediated by Ni2+-sensitive VGCCs, with their modulation (enhanced activity) by DHPG requiring elevated intracellular Ca2+. In the second model, Ca2+ entry through VGCCs contributes little to the medium ADP directly but acts as a trigger for a downstream conductance that is modulated by DHPG. For example, Ca2+ entry and release from internal stores could activate nonselective cation currents, such as ICAN, which mediate the medium ADP. Alternatively (or in addition), downregulation of Ca2+-activated K+ channels by DHPG could unveil the medium ADP. To distinguish between these possibilities, we examined the voltage dependence of the medium ADP. We reasoned that in the first model, where the medium ADP is mediated by VGCCs directly, hyperpolarization should accelerate deactivation of the VGCCs, thus reducing the medium ADP. In the second model, hyperpolarization would not eliminate Ca2+ entry during the action potentials, and it would increase the driving force on the cation channels, thus increasing the medium ADP. We found that holding the cell at a hyperpolarized holding potential strongly reduced the amplitude and duration of the medium ADP (Figure 6A and 6B), a finding most consistent with the first model, in which the medium ADP is mediated by VGCCs directly. However, the slowest component of the medium ADP was not reduced by hyperpolarization (Figure 6A), suggesting a contribution of Ca2+-activated channels to a slow ADP (see below).
To further examine the voltage dependence of the medium ADP, we delivered short hyperpolarizing current injections (−6 nA, 2 ms) either 8 or 48 ms after the last action potential in a burst (Figure 6C). On its own, the brief current injections produced a hyperpolarization to about −120 mV (see Materials and Methods) and returned to rest in 46±2 ms (n = 12). Hyperpolarization at the early time point reduced the medium ADP amplitude more than expected by simply summing the medium ADP and the short hyperpolarization alone (Figure 6D and 6E), again consistent with the notion that the medium ADP is mediated by a voltage-dependent conductance that can be deactivated by hyperpolarization. This effect was measured at the time when the hyperpolarizing response on its own decayed back to rest (54±2 ms after the last spike). When the hyperpolarizing current step began 48 ms after the last action potential, its effect (again measured when the hyperpolarization on its own decayed to rest; 100±2 ms after the last spike) was not statistically significant (Figure 6D and 6E). This finding suggests that the VGCCs responsible for the medium ADP remain activated for at least 50 ms, but not longer than 100 ms, while the conductance responsible for the slow ADP is not deactivated by hyperpolarization.
The voltage dependence of the medium ADP, combined with its sensitivity to low concentrations of Ni2+, suggests that the medium ADP may result from upregulation of Cav2.3 R-type calcium channels. However, previous work has indicated that mGluR activation can downregulate K+ channels [17],[18],[25]–[28], which could result in inhibition of the AHP and activation of R-type channels, without any actual modulation of the calcium channels by mGluRs. To test this alternative hypothesis, we converted the medium AHP to a medium ADP by injecting a ramp current that followed action potential firing, resulting in an artificial medium ADP in control ACSF, which resembled the medium ADP following application of DHPG. This artificial medium ADP was unaffected by bath application of Ni2+, suggesting that the Ni2+-sensitive calcium channels are not significantly activated by a post-burst medium ADP in the absence of mGluR activation (Figure 7A and 7B). We also tested the voltage dependence of the medium ADP in the absence of DHPG by varying the initial amplitude of the ramp current. The relationship between the medium ADP amplitude and the current injection was linear (Figure 7C and 7D), suggesting that voltage-dependent conductances do not amplify the medium ADP over this voltage range in the absence of mGluR activation.
To further test the hypothesis that upregulation of R-type calcium channels is responsible for the DHPG-induced medium ADP, we performed experiments on Cav2.3 knockout mice [29]. The medium ADP induced by DHPG was significantly smaller in the knockout mice, compared to wild-type controls (Figure 8). Because the Cav2.3 knockout mice are a hybrid of C57BL/6J (black) and 129S1/SvImJ (brown) mice, pups had different coat colors (black, dark brown, light brown; see Materials and Methods). We analyzed the results from different-colored mice separately and found no differences between these groups. Furthermore, we performed control experiments using black and brown mice and found similar DHPG-induced medium ADP amplitude in each of the wild-type strains (Figure S7).
We also used voltage-clamp recording to measure isolated R-type calcium currents in CA1 pyramidal neurons (see Materials and Methods). The resulting currents were upregulated by DHPG when activated by large but not small depolarizing steps (Figure 9A and 9B), consistent with modulation of R-type (high voltage activated) but not T-type (low voltage activated) calcium currents. Finally, to explore the ability of mGluR5 activation to regulate current mediated by calcium channels, we co-expressed, in Xenopus oocytes, mGluR5 along with either Cav2.3 α1 (plus α2δ1 and β3) or Cav3.2 α1 (see Materials and Methods; [30]). We found that DHPG application upregulated barium currents in oocytes expressing mGluR5 and Cav2.3 but not mGluR5 and Cav3.2, suggesting subunit selective modulation of R-type, but not T-type, calcium channels by mGluR5 activation with DHPG (Figure 9C and 9D).
The observation that the slow ADP was not eliminated by hyperpolarization suggests that a different class of channels may contribute to the slow ADP. We therefore examined this component of the ADP pharmacologically. It was not blocked by NiCl2 (Figure S8A and S8B), consistent with the idea that it is not mediated by the VGCCs responsible for the medium ADP. The slow ADP was also unaffected by nimodipine (Figure S8A and S8B).
We performed a battery of pharmacological experiments to explore the signal transduction mechanisms responsible for the post-burst medium ADP and slow ADP. These experiments (Figure S8C and S8D) required either intracellular drug application or pre-incubation of the drug in the bath. Like the medium ADP, the slow ADP was blocked by group I mGluR antagonists, especially the mGluR5 blocker MPEP. The slow ADP was also reduced by intracellular BAPTA, but it was not blocked by drugs that interfere with Ca2+ release from intracellular stores (CPA, ruthenium red, or heparin). All of these drugs inhibited the medium ADP, suggesting the Ca2+ release is required for the medium ADP, but not the slow ADP. The medium ADP was blocked by intracellular GDP-β-S, which interferes with G-protein coupled signaling, or by the PLC inhibitor U73122 (but not the inactive analog U73343). None of these drugs blocked the slow ADP, however, suggesting further that distinct signaling mechanisms mediate the DHPG-induced medium ADP and slow ADP.
We examined the activity dependence of the DHPG-induced post-burst medium ADP by delivering pairs of bursts at intervals of 0.1 to 20 s. Three-spike bursts were used in order to limit the size of the medium ADP so that either facilitation or inhibition could be observed. At intervals up to 200 ms, the second burst evoked a medium ADP almost twice the size of the first; at intervals of 1–5 s, the second burst was reduced by about 25% (Figure 10). The data were well fit by a model consisting of two processes: a facilitation process with a decay time constant of 0.25 s and an inhibition process with a decay time constant of 10 s (Figure S9 and Text S1). In the model, inhibition affected the fraction of the current available to be activated and facilitation affected the probability of activation (by a burst) of the available current. A key feature of the model was that any portion of the current could be inhibited, regardless of whether or not it was activated. This model predicted that facilitation is not expected, but inhibition persists, for a third burst delivered following two bursts (Figure S9). We conducted this experiment and the results were consistent with the predictions of the model (Figure 10). By contrast, a model in which inactivation was limited to the activated channels could not explain the results of the three-pulse experiment (Figure S9 and Text S1).
The findings reported here suggest that activation of group I mGluRs, which can occur as a result of synaptically released glutamate, increases the excitability of CA1 pyramidal neurons primarily by converting the post-spike AHP to an ADP via group I mGluR-mediated upregulation of Cav2.3 R-type calcium channels. The largest component of this change is a medium ADP lasting a little over 200 ms. A longer-lasting slow ADP (seconds) was smaller and mediated by different ion channels and signal transduction pathways than the medium ADP.
The medium ADP required action potential firing, although calcium spikes also activated the medium ADP in DHPG. The medium ADP was not affected by blocking L-type VGCCs with nimodipine, but calcium entry through Ni2+-sensitive channels was required for the medium ADP, as was intracellular Ca2+ elevation and Ca2+ release from internal stores. Block of the medium ADP by micromolar Ni2+ and the strong reduction of mGluR-mediated modulation of the post-burst potential in Cav2.3 knockout mice suggest that activation of R-type VGCCs are required for conversion of the medium AHP to a medium ADP [31]–[33]. Although it is difficult to distinguish between direct and indirect contributions of these channels, the voltage sensitivity of the medium ADP—including inhibition of the medium ADP by hyperpolarization after the triggering action potentials—suggests that R-type VGCCs contribute directly to the medium ADP. Our voltage-clamp experiments suggest the existence of an R-type Ca2+ current even prior to activation of mGluRs. The presence of an AHP under control conditions suggests, however, that K+ currents are larger than Ca2+ currents. Activation of mGluRs may downregulate K+ currents, but this downregulation is not sufficient to explain a voltage-dependent ADP as indicated by the ramp current experiments (Figure 7), which show that upregulation of R-type Ca2+ current is required to produce the ADP.
We cannot rule out the possibility that Ca2+ entry activates a voltage-dependent cation current underlying the medium ADP. Indeed, a number of reports implicate the activation of cation currents following group I mGluR activation in hippocampal neurons [34]–[39]. Some of these currents are Ca2+ sensitive, some are voltage sensitive, and others are both Ca2+ and voltage sensitive. Expression of these currents varies between CA3 and CA1 pyramidal neurons [40]. All of these currents are slower than the medium ADP reported here. Furthermore, none of these currents have been reported as sensitive to micromolar concentrations of Ni2+. Thus, the most parsimonious explanation of our data is that DHPG upregulates Ni2+-sensitive R-type VGCCs, which remain active for about 100 ms following a burst of spikes.
At least some of the slow, group I mGluR-activated currents described previously may be responsible for the slow ADP we observed here. Other candidate mechanisms are inhibition of slow K+ currents, including Ca2+-activated K+ currents, which have been reported in hippocampal neurons [26]–[28],[35],[41]–[43], and activation of exchanger currents [1]. These effects may also contribute somewhat to the medium ADP, as DHPG induced a small medium ADP even at very negative holding potentials, where activation of VGCCs is limited. At normal (i.e., resting) membrane potentials, inhibition of K+ currents could enhance the contribution of VGCCs.
If activation of R-type Ca2+ channels were the only requirement for the medium ADP, we would not expect its inhibition by chelating intracellular Ca2+ or interfering with Ca2+ release from internal stores. Indeed, either of these findings could be presented in support of a Ca2+-activated cation current as the primary mechanism. However, it is possible that intracellular Ca2+ elevation is required for modulation of the VGCCs following activation of mGluRs by DHPG. Indeed, the pharmacology suggests that the medium ADP requires activation of mGluR5 (and to a lesser extent mGluR1) and activation of G proteins (likely Gq) and PLC. This pathway can lead to several other signal transduction events, including Ca2+ release via activation of IP3 receptors. Several previous studies have shown that activation of group I mGluRs triggers Ca2+ release from internal stores in hippocampal neurons [26],[36],[44]–[48]. In most cases the IP3 receptor is the primary mediator of Ca2+ release; however, our finding that the medium ADP is inhibited by heparin and ruthenium red suggests that both IP3 receptors and Ca2+-induced Ca2+ release via ryanodine receptors are involved (Figure S8D) [49]. The consequences of this Ca2+ release are unknown, but we postulate it is an essential step in the complex cascade of signal transduction events that ultimately result in modulation of the Cav2.3 subunits responsible for the medium ADP.
Our finding that both mGluR5 and mGluR1 activation are required for the full effect of DHPG is consistent with previous work demonstrating the effects of both receptor subtypes in CA1 pyramidal neurons [26],[41],[44], despite greater expression of the mGluR5 subunit [50]–[55]. PLC-dependent and PLC-independent effects have been reported [10],[16],[18], and tyrosine phosphatases have also been implicated in mediating the effects of group I mGluR activation [17]. Clearly, more work is required to elucidate all of the pathways involved in activation of the medium ADP. Even more work will be needed to uncover the transduction mechanisms of the slow ADP. One possible mechanism is a current similar to the inward current mediated by mGluR1 activation previously described in CA3 pyramidal neurons, which was not dependent on G-protein activation (like the slow ADP reported here) but required activation of a Src-family tyrosine kinase [39].
Although bath application of DHPG produced a gradual onset of the medium ADP, it appeared more rapidly (<3 s) when DHPG was applied by pressure application. This suggests that the signal transduction pathways can be activated and reversed very rapidly, an observation that has been used to suggest a membrane delimited mechanism [1]. The block of the medium ADP by chelating intracellular Ca2+ or interfering with Ca2+ release suggests, however, that membrane delimited signaling alone may be insufficient.
The lack of effect of DHPG when applied to the apical dendrites should be interpreted with caution. Although it is tempting to conclude that the relevant mGluRs may have a perisomatic location, it is also possible that direct activation of the dendrites (e.g., synaptic activation or dendritic Ca2+ spikes) could lead to a medium ADP when dendritic mGluRs are activated. More work is needed to determine the distribution of mGluR1 and mGluR5 in CA1 neurons and their physiological effects when activated in various cellular compartments.
An intriguing aspect of the medium ADP is its activity dependence. It was markedly enhanced when pairs of bursts were delivered at intervals of less than 1 s but suppressed during pairs of bursts at longer intervals or when triplets of bursts were delivered to the neurons. Similarly, while short bursts of action potentials produced an ADP, longer trains of spikes did not produce an ADP. The molecular steps responsible for these activity-dependent effects are unknown, but the data using pairs and triplets of bursts were well described by a model consisting of a short-lasting facilitation of unactivated channels and a longer-lasting inhibition of a fraction of all of the channels, independent of activation. These interesting properties may offer a clue to identification of the underlying currents in future voltage-clamp experiments.
Identifying the contribution of mGluR activation to neuronal excitability in vivo will be a crucial step for ultimately establishing the importance of this mechanism for hippocampal function. Accomplishing this task will require that the balance of two competing factors be determined: the enhanced activation of mGluRs during periods of high activity and the activity-dependent inhibition of the ADP during high-frequency spiking. In general, identifying the underlying conductances, their possible molecular composition, and the signal-transduction steps and molecular players involved in their activation and modulation will be critical for determining how excitability is regulated via changes in the AHP/ADP in vivo. This knowledge would facilitate the use of molecular genetics to study the effects of these mechanisms on hippocampal function in vivo with single-unit recordings and behavioral analysis.
Slice experiments were performed in the USA and approved by the Northwestern University Animal Care and Use Committee. Oocyte experiments were performed in Korea and approved by the Sogang University Animal Care Committee.
Hippocampal slices were prepared from male Wistar rats 25–45 d old or mice (C57BL/6J or 129S1/SvImJ or Cav2.3 knockout) 21–35 d old. Voltage-clamp experiments were done on slices prepared from younger rats (13–18 d old) in order to reduce space-clamp problems. Knockout mice were derived from 129S1/SvImJ (brown mouse) embryonic stem-cell injections in C57BL/6J (black) mice [29]. The founder mice were bred to C57BL/6J females; therefore, offspring of the knockout mice had either light brown, dark brown, or black coat color.
Following anesthesia with halothane or isoflurane, animals were perfused through the heart with ice-cold ACSF (see below). The brain was removed rapidly and mounted in a near-horizontal plane for preparation of 300 µm hippocampal slices using a vibratome. Slices were prepared in either ice-cold ACSF or sucrose-based solution, then transferred to a chamber containing oxygenated ACSF (no sucrose) at approximately 35°C for half an hour. The slice chamber was then maintained at room temperature and slices were removed individually for electrophysiological recordings.
Whole-cell current-clamp recordings were obtained at 33±2°C. Patch-clamp electrodes were pulled from 2.0 mm outer diameter borosilicate glass and filled with a K-gluconate-based intracellular solution (see below). Electrode resistance was 3–6 MΩ in the bath and series resistance was 5–20 MΩ during the recordings. Current-clamp recordings were obtained with Dagan BVC-700 amplifiers, using appropriate bridge balance and electrode-capacitance compensation. Voltage-clamp recordings with appropriate capacitance and series resistance compensation were performed at room temperature (23–25°C) and monitored with an Axopatch 200B amplifier (Molecular Devices, Union City, CA). Data acquisition and analysis was performed using custom software written for Igor Pro. Statistical tests included the paired or unpaired t test and analysis of variance (one-way ANOVA or repeated measures one-way ANOVA) with Tukey's post hoc comparisons. All statistical analyses were performed using Prism 4 software and in most cases detailed results are provided in the figure legends.
For the hyperpolarizations shown in Figure 6, the −6 nA, 2 ms current steps produced hyperpolarizations that briefly (1–8 ms) exceeded the −10 V limit of the analog-to-digital converter. The peak hyperpolarization was therefore estimated by extrapolating double-exponential fits of the response to the peak time predicted from linear fits of the rising phase. The estimated peaks were −119±3 mV for the early steps (n = 12) and −117±2 mV for the late steps (n = 11). The small clipping effect had an insignificant effect on the subtraction procedure used (see Results), because the effect is measured when the hyperpolarization had decayed back to rest, more than 40 ms after the clipping ended.
To test whether synaptic activation can induce the post-burst ADP, 5 brief action potentials were somatically injected either with or without synaptic stimulation and the responses were monitored once every 5 min with 1 min delay between two conditions. In both the MPEP/LY and control groups, experiments were performed in the presence of blockers of ionotropic glutamate receptors (30 µM CNQX and 50 µM D-AP5) and GABA receptors (2 µM SR95531 and 1 µM CGP55845). Bipolar borosilicate theta glass stimulation electrodes (Sutter Instruments) filled with ACSF were used in conjunction with Dagan BSI-950 biphasic stimulus isolator. Stimulating electrodes were placed in proximal stratum radiatum and at least 100 µm away from the recorded cell and toward CA3. Stimulus intensity was set to produce a 5–11 mV ADP during synaptic stimulation.
Normal ACSF had the following composition (mM): 125 NaCl, 2.5 KCl, 25 NaHCO3, 1.25 NaH2PO4, 1 MgCl2, 2 CaCl2, 25 Dextrose (Fisher Scientific; Sigma). In some cases slices were prepared in a modified ACSF in which 125 mM NaCl was replaced with 75 mM NaCl and 75 mM sucrose. In many experiments, drugs were added to the bath (see below). The bath perfusion rate was 2–3 ml/min.
The intracellular recording solution had the following composition (mM): 115 K-gluconate, 20 KCl, 10 Na2phosphocreatine, 10 HEPES, 2 MgATP, 0.3 NaGTP, 0.1% Biocytin (Fisher Scientific; Sigma). In some experiments, drugs were added to the intracellular solution (BAPTA, GDP-β-S, U73122, U73343, ruthenium red, and heparin; see below); for BAPTA-containing internal solution, the K-gluconate concentration was reduced to 100 mM. The K-gluconate based internal solution was used because the properties of CA1 pyramidal neurons are more stable with this solution than with K-Methylsulfate based solutions [56].
For voltage-clamp experiments in slices, patch electrodes (3–6 MΩ in bath) were filled with intracellular solution containing the following (in mM): 110 Cs-gluconate, 25 TEA-Cl, 10 HEPES, 2 EGTA, 4 Mg-ATP, and 0.5 Na-GTP, 5 Na2-phosphocreatine, pH 7.3 with CsOH. R- and T-type calcium currents were isolated pharmacologically by preincubating the slices in a mixture containing ω-conotoxin MVIIC (2 µM), ω-conotoxin-GVIA (2 µM), and ω-agatoxin IVA (0.2 µM) to block N-, P-, and Q-type Ca2+ currents and cytochrome c (0.1 mg/ml to block nonspecific toxin binding) for>1 h at room temperature. Nifedipine (20 µM) and TTX (1 µM) were bath applied to block L-type Ca2+ currents and Na+ currents, respectively. Recordings were performed in modified ACSF solution containing the following (in mM): 125 NaCl, 25 NaHCO3, 2.5 KCl, 1.0 MgCl2, 2.0 CaCl2, 1.25 NaH2PO4, 25 NaHCO3, and 10 dextrose, 2 CsCl, 5 4-AP, 10 TEA-Cl, pH 7.4.
The following drugs were obtained from Tocris: (S)-3,5-Dihydroxyphenylglycine (DHPG), (S)-(+)-a-Amino-4-carboxy-2-methylbenzeneacetic acid (LY367385), 2-Methyl-6-(phenylethynyl)pyridine hydrochloride (MPEP), Ammoniated ruthenium oxychloride (Ruthenium Red), 1,4-Dihydro-2,6-dimethyl-4-(3-nitrophenyl)-3,5-pyridine dicarboxylic acid 2-methyloxyethyl 1-methylethyl ester (Nimodipine), (6aR,11aS,11bR)-rel-10-Acetyl-2,6,6a,7,11a,11b-hexahydr o-7,7-dimethyl-9H-pyrrolo[1′,2′:2,3]isoindolo[4,5,6-cd] indol-9-one (CPA), D-(−)-2-Amino-5-phosphonopentanoic acid (D-AP5), 6-Cyano-7-nitroquinoxaline-2,3-dione disodium (CNQX disodium salt), (2S)-3-[[(1S)-1-(3,4-Dichlorophenyl)ethyl]amino-2-hydro xypropyl](phenylmethyl)phosphinic acid hydrochloride (CGP 55845 hydrochloride), and Octahydro-12-(hydroxymethyl)-2-imino-5,9∶7,10a-dimethan o-10aH-[1],[3]dioxocino[6,5-d]pyrimidine-4,7,10,11,12-pen tol citrate (Tetrodotoxin citrate).
The following drugs were obtained from Sigma: 1,2-Bis(2-aminophenoxy)ethane-N,N,N',N'-tetraacetic acid tetrapotassium salt (BAPTA), 1-[6-[((17β)-3-Methoxyestra-1,3,5[10]-trien-17-yl)amino]hexyl]-1H-pyrrole-2,5-dione (U73122), 1-[6-[((17β)-3-Methoxyestra-1,3,5[10]-trien-17-yl)amino]hexyl]-2,5-pyrrolidinedione (Uf73343), Guanosine 5′-[β-thio]diphosphate trilithium salt (GDP-β-S), Heparin sodium salt (from porcine intestinal mucosa average mol wt ∼3,000 kD), Nickel(II) chloride hexahydrate (NiCl2), 2-(3-Carboxypropyl)-3-amino-6-(4 methoxyphenyl)pyridazinium bromide (SR 95531), Dextrose, K-gluconate, sodium phosphocreatine, HEPES, MgATP, NaGTP, and biocytin.
In most experiments, DHPG was applied by bath perfusion at a concentration of 4 µM. Some batches of DHPG were more potent than others, so in some cases it was necessary to reduce the DHPG concentration to as low as 2 µM in order to prevent additional spiking following the triggered burst of action potentials. Experiments using 2–4 µM DHPG were pooled together for analysis. In most experiments, the membrane potential was held at −65 mV, which required very small current injections (< 50 pA). DHPG application resulted in a depolarization of 3–5 mV, so the holding potential was adjusted to −65 mV with hyperpolarizing holding current. In some experiments (as noted), DHPG (500 µM DHPG) or ACSF was applied to the cell via pressure application from a broken patch pipette. Pressure (10 psi, 0.2 s) was applied via a Dagan PMI-100 pressure micro-injector. Current was injected to the cell within 3 s of pressure application. For experiments with mouse slices, bath application was performed using 10 µM DHPG, yielding an ADP similar to that observed with 4 µM DHPG in rat slices.
The cDNAs for the Cav3.2 (accession number AF051946), Cav2.3 (L27745), β3 (M88751), α2δ1 (M86621), and mGluR5 (D10891) were subcloned into a high expression vector pGEMHEA, which contains the 5′ and 3′ untranslated regions of the Xenopus β globin gene, linearized, and transcribed into cRNA using T7 RNA polymerase according to the manufacturer's protocol (Ambion, Austin, TX, USA).
Oocytes were obtained from female Xenopus laevis (Nasco, WI, USA) using a standard procedure. Several ovary lobes were surgically removed from mature female Xenopus laevis and torn into small clusters in SOS solution (in mM: 100 NaCl, 2 KCl, 1.8 CaCl2, 1 MgCl2, 5 HEPES, 2.5 pyruvic acid, 50 µg/ml gentamicin, pH 7.6). The follicular membranes were removed by digestion in Ca2+-free Barth's solution (in mM: 88 NaCl, 1 KCl, 2.4 NaHCO3, 0.82 MgSO4, pH 7.4) containing 20 mg/ml collagenase (SERVA, Heidelberg, Germany). Oocytes were injected under a stereo-microscope with 2–20 ng of cRNA using a Drummond Nanoject pipette injector (Parkway, PA, USA) attached to a Narishige micromanipulator (Tokyo, Japan).
Barium currents were measured at room temperature 4 to 8 d after cRNA injection using a two-electrode voltage-clamp amplifier (OC-725C, Warner Instruments, Hamden, CT, USA). Microelectrodes (Warner Instruments, Hamden, CT, USA) were filled with 3 M KCl and their resistances were 0.2–1.0 MΩ. The 10 mM Ba2+ bath solution contained (in mM): 10 Ba(OH)2, 90 NaOH, 1 KOH, 5 HEPES (pH 7.4 with methanesulfonic acid). The currents were sampled at 5 kHz and low pass filtered at 1 kHz using the pClamp system (Digidata 1322A and pClamp 8; Axon instruments, Foster City, CA, USA). Peak currents and exponential fits to currents were analyzed using Clampfit software (Axon instruments, Foster City, CA, USA).
|
10.1371/journal.pgen.1004175 | The Caenorhabditis elegans Iodotyrosine Deiodinase Ortholog SUP-18 Functions through a Conserved Channel SC-Box to Regulate the Muscle Two-Pore Domain Potassium Channel SUP-9 | Loss-of-function mutations in the Caenorhabditis elegans gene sup-18 suppress the defects in muscle contraction conferred by a gain-of-function mutation in SUP-10, a presumptive regulatory subunit of the SUP-9 two-pore domain K+ channel associated with muscle membranes. We cloned sup-18 and found that it encodes the C. elegans ortholog of mammalian iodotyrosine deiodinase (IYD), an NADH oxidase/flavin reductase that functions in iodine recycling and is important for the biosynthesis of thyroid hormones that regulate metabolism. The FMN-binding site of mammalian IYD is conserved in SUP-18, which appears to require catalytic activity to function. Genetic analyses suggest that SUP-10 can function with SUP-18 to activate SUP-9 through a pathway that is independent of the presumptive SUP-9 regulatory subunit UNC-93. We identified a novel evolutionarily conserved serine-cysteine-rich region in the C-terminal cytoplasmic domain of SUP-9 required for its specific activation by SUP-10 and SUP-18 but not by UNC-93. Since two-pore domain K+ channels regulate the resting membrane potentials of numerous cell types, we suggest that the SUP-18 IYD regulates the activity of the SUP-9 channel using NADH as a coenzyme and thus couples the metabolic state of muscle cells to muscle membrane excitability.
| Iodotyrosine deiodinase (IYD) controls the recycling of iodide in the biogenesis of thyroid hormones that regulate metabolism. Defects in IYD result in congenital hypothyroidism, a multisystem disorder that can lead to growth failure and severe mental retardation. We identified the gene sup-18 of the nematode Caenorhabditis elegans as a regulator of the SUP-9/UNC-93/SUP-10 two-pore domain potassium channel complex and showed that SUP-18 is an ortholog of IYD, a member of the NADH oxidase/flavin reductase family. SUP-18 IYD is required for the activation of the channel complex by a gain-of-function mutation of the SUP-10 protein. SUP-9 channel activation by SUP-18 requires a conserved serine-cysteine-rich region in the C-terminus of SUP-9 and is independent of the function of the conserved multi-transmembrane protein UNC-93. We propose that SUP-18 uses NADH as a coenzyme to activate the SUP-9 channel in response to the activity of SUP-10 and the metabolic state of muscle cells.
| Hypothyroidism, one of the most common endocrine disorders, can cause many different symptoms and can lead to defects in brain development and maturation and retarded postnatal development [1]. For thyroid hormone biosynthesis, iodide is recycled by iodotyrosine deiodinase through the deiodination of monoiodotyrosine and diiodotyrosine, two byproducts in the generation of thyroid hormones [2]–[6]. In humans, this deiodination is catalyzed by human iodotyrosine dehalogenase (DEHAL1)/iodotyrosine deiodinase (IYD), an NADH oxidase/flavin reductase [7]–[10]. Mutations in IYD cause congenital hypothyroidism [11]–[13]. How the activity of IYD is regulated in vivo and whether IYD has other functions remain to be elucidated.
Four transmembrane/two-pore domain K+ channels play a key role in establishing the resting membrane potentials of many cell types and in modulating their responses to neurotransmitters and second messengers [14]–[16]. To date, 15 human two-pore domain K+ channels have been identified [14], [16], [17]. The activities of two-pore domain K+ channels can be regulated by multiple chemical and physical factors, including temperature [18], membrane stretch [19], [20], arachidonic acid [21], pH [22], [23], volatile anesthetics [24], [25] and neurotransmitters [26], [27].
The gene sup-9 of the nematode Caenorhabditis elegans encodes a two-pore domain K+ channel [28]. sup-9(n1550) gain-of-function (gf) mutants are egg-laying defective and display a flaccid paralysis and a rubberband uncoordinated (Unc) behavior: when prodded on the head, a sup-9(n1550gf) worm contracts and relaxes along its entire body without moving backwards, while a wild-type worm contracts its anterior end and moves away [29]. Loss-of-function (lf) mutations in sup-9 or two other genes, sup-10 and unc-93, completely suppress these sup-9(n1550gf) defects [29]–[31]. In addition, gf mutations in sup-10 and unc-93 themselves induce a rubberband Unc paralysis, which in turn are suppressed by lf mutations in sup-9, sup-10 and unc-93 [30]–[32]. lf mutants of unc-93, sup-9 and sup-10 do not have obviously abnormal phenotypes [29]–[31], [33]. The SUP-9 two-pore domain K+ channel is most closely related to human TASK-3 [28], [34], [35]. unc-93 encodes a conserved multi-pass transmembrane protein [33]. An UNC-93 homolog, UNC93b1, is involved in innate immune responses in mammals [36], [37]. sup-10 encodes a novel type-I transmembrane protein [35]. Genetic analyses and the molecular identities of these genes suggest that in vivo SUP-10 and UNC-93 form a protein complex with the SUP-9 two-pore domain K+ channel and modulate its activity as regulatory subunits [28], [33].
Mutations in the gene sup-18 suppress the muscle defects caused by gf mutations in these three genes, strongly suppressing the locomotory defects of sup-10(n983gf) mutants, partially suppressing the locomotory defects of the strong unc-93(e1500gf) mutants, the weak unc-93(n200gf) mutants and the strong sup-9(n1550gf)/+ heterozygous mutants, and suppressing only the lethality of sup-9(n1550gf) mutants [29], [30] (also see Table 1 below). In this study we report that sup-18 encodes the C. elegans ortholog of mammalian iodotyrosine deiodinase/dehalogenase (IYD) [7], [8], [10]. Our findings suggest that SUP-18 is a functional regulator of the SUP-9/SUP-10/UNC-93 two-pore domain K+ channel complex in vivo and that IYD might function with two-pore domain K+ channel complexes in mammals.
sup-10(n983gf) mutants have a reduced locomotory rate (Table 1). A loss-of-function mutation in sup-18, n1030, restores wild-type locomotion to sup-10(n983gf) mutants (Table 1) [30]. unc-93(n200gf) causes a less severe rubberband Unc phenotype than sup-10(n983gf), yet the unc-93(n200gf) phenotype is still only partially suppressed by sup-18(n1030) (Table 1). unc-93(e1500gf) mutants, which have a more severe rubberband Unc phenotype than sup-10(n983gf) mutants, similarly are only weakly suppressed by sup-18(n1030). These results suggest that the differential suppression of the rubberband Unc mutants by sup-18(n1030) is caused by gene-specific effects rather than by differential severity of paralysis in these mutants.
We further tested this notion using weakly paralyzed double mutants carrying the unc-93(e1500gf) mutation and a partial lf allele of sup-10. Introduction of the sup-18(n1030) mutation into partially suppressed unc-93(e1500gf); sup-10(n4025) or unc-93(e1500gf); sup-10(n4026) mutants only weakly improved their locomotory rates from approximately 14 to 19 body-bends/minute (Table 1). These results confirm that sup-18(n1030) only weakly suppresses gf mutations in unc-93.
The suppression of sup-10(n983gf) depends on the dosage of the sup-18 allele [29]. We found that sup-18(n1030)/+; sup-10(n983gf) males exhibit an intermediate phenotype (15.2 bends/min) between those of the more severely paralyzed sup-10(n983gf) males (4.7 bends/min) and the strongly suppressed sup-18(n1030); sup-10(n983gf) males (31.7 bends/min) (Table 2). This dose-dependent effect was observed for all lf alleles of sup-18 tested (Table 2). By contrast, the suppression of sup-10(n983gf) by sup-9(n1913), a channel null allele, was recessive.
Because the weak suppression of the locomotory defect of unc-93(e1500gf) mutants by sup-18(lf) mutations (Table 1) [30] makes a dosage analysis of sup-18(lf) suppression of unc-93(e1500gf) difficult, we examined weakly paralyzed unc-93(e1500gf); sup-10(n4025) males, which are more visibly suppressed by sup-18(n1030) (Table 2). We found that the locomotory rate of unc-93(e1500gf); sup-10(n4025) males heterozygous for sup-18(n1030) was similar to that of males wild-type for sup-18 (10.0 vs. 9.8, respectively) (Table 2). Similarly, sup-9(n1550gf)/+; sup-18(n1030)/+ males had only slightly improved locomotion compared to sup-9(n1550gf)/+ males (5.0 vs. 3.8, respectively) (Table 2). We conclude that the dose-dependent suppression of rubberband Unc mutants by sup-18 alleles is also gene-specific: the sup-10(n983gf) phenotype is much more sensitive to sup-18 levels than is that of the other rubberband mutants.
sup-18 had previously been mapped to the interval between daf-4 and unc-32 on LGIII [30]. Using three-point mapping we further localized sup-18 to the interval between ncl-1 and unc-36 (see Materials and Methods) (Figure 1A). Transgene rescue experiments with cosmids spanning the ncl-1-to-unc-36 interval and with smaller cosmid subclones identified a 4.5 kb minimal rescuing fragment from cosmid C02C2: as a transgene, this fragment restored the rubberband Unc phenotype to sup-18(n1010); sup-10(n983gf) mutants (Figure 1A). This rescuing fragment contained a single predicted gene, C02C2.5 [www.wormbase.org]. We screened a mixed-stage cDNA library [38] using the smallest cosmid subclone with sup-18 rescuing activity and obtained a single partial cDNA of this predicted gene. We defined the structure of this gene from RT-PCR and RACE experiments (see Materials and Methods) (Figure 1B).
sup-18 encodes a predicted protein of 325 amino acids. This protein is the only C. elegans ortholog of mammalian iodotyrosine deiodinase (IYD), which belongs to the NADH oxidase/flavin reductase superfamily (Fig. 1C) [7]–[10]. IYD catalyzes the recycling of iodide by deiodinating 3′-monoiodotyrosine and 3′, 5′-diiodotyrosine, the main byproducts in the process of thyroid hormone biogenesis [2]–[5], [7], [8]. The identity between SUP-18 and human IYD protein variant 2 (also named DEHAL1) [8] is 31% overall and 45% over the NADH oxidase/flavin reductase domain (Figure 1C). Like IYDs of Drosophila, mouse and human, SUP-18 has a hydrophobic region that precedes the NADH oxidase/flavin reductase domain and might serve as a transmembrane domain.
We identified molecular lesions in the sup-18 coding sequence of all 18 mutant strains analyzed (Table 3, Fig. 1C). The sup-18(n1033) mutation leads to the substitution of an isoleucine for the initiator methionine, which should cause any translational products to be nonfunctional. (The next three ATG sequences in the sup-18 cDNA are out of frame.) The sup-18(n1030) and sup-18(n1548) mutations cause premature stop codons that likely generate truncated protein products. Four mutations (n1038, n527, n463, n1539) cause a frameshift. Another four mutations (n1036, n1035, n1015, n1558) affect splice donor or acceptor sites. The remaining seven missense mutations (n1010, n1554, n1471, n1556, n1014, n1022, n528) disrupt residues within the NADH oxidase/flavin reductase domain.
To examine the expression pattern of sup-18, we introduced the coding sequence of gfp between codons 88 and 89 of a genomic clone of sup-18, generating a sup-18 translation fusion transgene (see Materials and Methods). Similar to transgenic animals expressing a Psup-10::gfp translational fusion transgene, Psup-18::gfp transgenic animals displayed GFP fluorescence in body-wall (Fig. 2A, D), defecation (Fig. 2B, E) and vulval muscles (Fig. 2C, F). In body-wall muscle cells (Fig. 2A, D), the SUP-10::GFP and SUP-18::GFP fusion proteins both localized to cell-surface regions aligned with dense bodies, the functional analogs to vertebrate Z-lines that connect the myofibril lattice to the cell membrane [39]. In addition to muscles, three neurons in the head of Psup-18::gfp transgenic animals also displayed GFP staining (I. de la Cruz and H. R. Horvitz, unpublished observations). We previously reported expression of a Psup-9::gfp reporter in the four SIA interneurons [28]. We stained the Psup-18::gfp transgenic animals with an anti-CEH-17 antibody, which labels the four SIA neurons and the ALA neuron [40], and found that the neurons expressing the SUP-18::GFP fusion protein were not the SIAs (I. de la Cruz and H. R. Horvitz, unpublished observations).
We generated a rabbit anti-SUP-18 antibody (see Materials and Methods). In immunostained animals, this antibody could detect overexpressed SUP-18 but failed to detect endogenous SUP-18, probably because of the low level of SUP-18 expression. We next generated transgenic animals co-expressing a Psup-10::gfp fusion transgene and sup-18 under control of a myo-3 promoter [41] and examined the subcellular expression of SUP-18 using the antibody and of SUP-10::GFP using GFP fluorescence. We found that SUP-10 and SUP-18 colocalize in subcellular structures, including the dense bodies in the body-wall muscles (Fig. 2G, H, I). Since GFP fusions to SUP-9 and UNC-93 localize similarly [28], this result suggests that SUP-18 colocalizes with a SUP-9/UNC-93/SUP-10 complex.
Mammalian IYD is a transmembrane protein [7], [8]. The presence of a possible transmembrane domain in the predicted SUP-18 protein sequence (Fig. 1C) suggests that SUP-18 is also a transmembrane protein. To distinguish whether the NADH oxidase/flavin reductase domain of SUP-18 resides intracellularly or extracellularly, we generated transgenic animals expressing different SUP-18::β-galactosidase fusion proteins and assayed β-galactosidase activity in vivo in fixed animals (Fig. 3A). When β-galactosidase is localized intracellularly it is enzymatically active, whereas extracellular localization results in loss of β-galactosidase activity [42], [43]. The use of β-galactosidase activity to elucidate the membrane topology of C. elegans proteins in vivo has been reported previously for the presenilin SEL-12 protein [44] and for the MEC-4 sodium channel subunit [45].
Fixed transgenic animals expressing β-galactosidase fused to either the C-terminal region of SUP-18 or immediately C-terminal to the putative transmembrane domain showed robust β-galactosidase activity (Fig. 3A). Introduction of a synthetic transmembrane domain [45] between SUP-18 and β-galactosidase in these chimeras eliminated β-galactosidase enzymatic activity, presumably because the membrane orientation of β-galactosidase had been reversed (Fig. 3A).
These results strongly suggest that SUP-18 is a transmembrane protein and that the NADH oxidase/flavin reductase domain of SUP-18 resides intracellularly. But they do not distinguish between a type-I transmembrane protein (single-pass transmembrane protein with the N-terminal domain located extracellularly) and a cytoplasmic protein that simply localizes at the cell surface, e.g., by interacting with another membrane protein or by linking to a GPI anchor [46]. To test if the putative transmembrane domain of SUP-18 can indeed behave as a transmembrane domain, we inserted a signal sequence at the N-terminus of SUP-18 (see Materials and Methods). While a fusion containing the presumptive extracellular domain of SUP-18 but lacking the putative transmembrane domain resided intracellularly as expected, the introduction of a signal sequence led to its secretion and loss of β-galactosidase enzymatic activity (Fig. 3A). By contrast, when either the SUP-18 putative transmembrane domain or the synthetic transmembrane domain [45] was added to this SUP-18::β-galactosidase fusion, the enzymatic activity was restored. These results indicate that the putative transmembrane domain of SUP-18 can indeed function as a transmembrane domain and suggest that SUP-18 is likely a type-I integral membrane protein, like IYD.
To establish an assay for in vivo SUP-18 activity, we expressed the sup-18 coding sequence under the control of the myo-3 promoter [41] in sup-18(n1033); sup-10(n983) mutant animals. While sup-10(n983gf) mutant animals are defective in locomotion, double mutants carrying the sup-18(n1033) null mutation had improved locomotory rates (Fig. 3B). Expression of Pmyo-3 gfp in sup-18(n1033); sup-10(n983gf) animals had little effect on their locomotory rate, whereas expression of a Pmyo-3 sup-18(+) transgene restored sup-10(n983gf) paralysis (Fig. 3B). By contrast, expression of two Pmyo-3 sup-18 mutant constructs containing either the n1554 missense mutation or the n1010 mutation (which affects a conserved amino acid in the NADH oxidase/flavin reductase domain; Fig. 1C and Table 3) did not restore the rubberband Unc phenotype to sup-18(n1033); sup-10(n983gf) mutants (Fig. 3B).
We found that the mouse IYD gene could not substitute for sup-18 in vivo in restoring the rubberband Unc phenotype of sup-18(n1033); sup-10(n983gf) animals (Figure 3B). We tagged mouse IYD with GFP at its C-terminus and found that C. elegans expressing the fusion protein displayed GFP fluorescence in body-wall muscle structures similar to that observed for the SUP-18::GFP fusion (I. de la Cruz and H. R. Horvitz, unpublished observations). These results suggest that mouse IYD had been expressed properly and that mouse IYD might be inactive or otherwise incapable of substituting for SUP-18 in C. elegans.
Interestingly, transgenic expression of the SUP-18 intracellular domain alone (amino acids 66–325) was sufficient to restore rubberband Unc paralysis to sup-18(n1033); sup-10(n983gf) animals, although the rescue was less robust than that conferred by full-length SUP-18 (Fig. 3B). This finding suggests that the extracellular and transmembrane domains of SUP-18 are not essential for its in vivo function and is consistent with the conclusion that the NADH oxidase/flavin reductase domain is intracellular.
The overexpression of sup-18(+) from a Pmyo-3 sup-18(+) transgene in sup-18(n1033): sup-10(n983gf) mutants not only restored the rubberband Unc phenotype but also apparently enhanced that phenotype beyond that of sup-10(n983gf) single mutants (Fig. 3B). This finding indicates a dose-dependent effect of sup-18(+) and is consistent with our gene-dosage observation that sup-18(lf)/+ can partially improve the locomotory rate of sup-10(n983gf) mutants (Table 2). Overexpression of sup-18(+) with the coinjection marker lin-15(+) in lin-15 mutant animals did not cause obvious differences in locomotion compared to animals injected with lin-15(+) alone (Table 4), indicating that overexpression of sup-18(+) itself did not slow locomotion.
We introduced the extrachromosomal arrays containing the transgenes from two independently-derived strains carrying sup-18(+) and the lin-15(+) coinjection marker into sup-10(n983gf) lin-15 double mutants by mating, so that each resulting strain would contain the same transgenes as the parental strain and therefore would overexpress sup-18(+) at equivalent levels. sup-18(+) overexpression caused a severe paralysis of sup-10(n983gf) lin-15 animals relative to control transgenic animals expressing lin-15 alone (0.1 and 0.0 vs. 5.7 and 5.4, bends/minute, respectively) (Table 4). sup-10(n983gf) mutants overexpressing sup-18(+) were smaller in size (Fig. 4A–D) and resembled severely paralyzed mutants carrying a sup-9(n1550gf) mutation (compare Figs. 4B and 4D).
We next tested if overexpression of sup-9(+), unc-93(+) or sup-10(n983gf) itself could enhance the sup-10(n983gf) defect as did overexpression of sup-18(+). Overexpression of these other genes under the control of the myo-3 promoter did not affect the locomotory rate of transgenic sup-10(n983gf) mutant animals compared to animals transgenic for lin-15 alone (Table 4). These results suggest that the activity of SUP-18 might be enhanced by increased expression, while increased expression of SUP-9, UNC-93 and SUP-10 does not increase the biological effects of these proteins.
We tested if overexpression of sup-18(+) could enhance the defects of unc-93(e1500gf) mutants and found no obvious difference in appearance compared to control animals overexpressing lin-15 alone (Fig. 4E, F). Because the locomotory rate of unc-93(e1500gf) mutants transgenic for either sup-18(+) or lin-15(+) transgenes was zero (Table 4) and an enhancement of locomotory defects could not be scored, we turned to a different aspect of the phenotype of rubberband mutants, a reduced brood size [29]. Consistent with the enhancement of locomotory defects, overexpression of sup-18(+) reduced the brood size of sup-10(n983gf) mutants by three-fold, from an average of 74 and 75 progeny for the two transgenic lines, to 17 and 27, respectively (Table 4). These low brood sizes are comparable to those of severely paralyzed sup-9(n1550gf); sup-18(n1030) mutants (Table 4). By contrast, the brood sizes of unc-93(e1500gf) mutants did not change in response to sup-18(+) overexpression (35 and 43 vs. 37 and 40, respectively). Thus, the effects of sup-18(+) overexpression on the locomotion and brood size of rubberband Unc mutants are gene-specific: the sup-10(n983gf) phenotype is more sensitive to increased sup-18 levels than is that of unc-93(e1500gf) mutants.
Like sup-18 mutations, the sup-9 allele n1435 strongly suppresses the locomotory defects of sup-10(n983gf) but not those of unc-93(e1500gf) mutants (Table 5) [29]. By contrast, null mutations in sup-9, such as sup-9(n1913), completely suppress the defects caused by gf mutations in both sup-10 and unc-93 (Table 5) [30], [31]. To determine if other sup-9 alleles exhibit similar gene-specific effects, we assayed 13 previously isolated sup-9 missense mutations [34], [35], [36], [39]. Four sup-9 mutations that had been isolated as sup-10(n983gf) suppressors and nine that had been isolated as unc-93(e1500gf) suppressors all strongly suppressed unc-93(e1500gf) and sup-10(n983gf) defects equally well (Table 5), confirming that sup-9(n1435) represents a rare class of sup-9 mutations.
The similarity of sup-18(lf) mutations and sup-9(n1435) in preferentially suppressing sup-10(n983gf) defects compared to those of unc-93(e1500gf) mutants suggests that sup-18(lf) mutations and the sup-9(n1435) mutation might act via the same mechanism. If so, n1435 might have no suppressive activity in the absence of sup-18. Indeed, the locomotory rate of the sup-9(n1435); unc-93(e1500gf) sup-18(n1030) triple mutant was similar to that of either the sup-9(n1435); unc-93(e1500gf) or the unc-93(e1500gf) sup-18(n1030) double mutant (Fig. 5A). This effect appears to be specific for sup-9(n1435), as a different weak sup-9 allele, n264, was enhanced by sup-18(n1030) (Fig. 5A). We also assayed the brood size of unc-93(e1500gf) mutants in the presence of either or both sup-18(n1030) and sup-9(n1435). For example, although the low brood size of unc-93(e1500gf) mutants was restored to wild-type levels by the null mutation sup-9(n1913) (Fig. 5B), sup-9(n1435) and sup-18(n1030) single mutations or sup-9(n1435); sup-18(n1030) double mutations only partially rescued the brood size of unc-93(e1500gf) mutants and the double mutations acted similarly to the sup-18(n1030) single mutation (Fig. 5B). As was the case for locomotion, for brood size sup-18(n1030) enhanced the effect of the weak loss-of-function allele, sup-9(n264) on unc-93(e1500gf) mutants (Fig. 5B). The lack of an additive effect of sup-18(n1030) and sup-9(n1435) in suppressing the locomotion and brood size defects of unc-93(e1500gf) mutants suggests that sup-9(n1435) and sup-18(n1030) mutations likely act through the same pathway.
To further examine this hypothesis, we tested for an additive effect between sup-18(n1030)/+ and sup-9(n1435)/+ in their suppression of the locomotory defects of sup-10(n983gf) mutants. (A test for an additive effect of sup-18(n1030) and sup-9(n1435) homozygous mutations would not be informative, as both mutations fully suppress the locomotory defect of sup-10(n983gf) mutants.) We found that sup-10(n983gf) males heterozygous for either sup-9(n1435)/+ or sup-18(n1030)/+ are partially suppressed for the locomotory defects (Fig. 5C). The sup-9(n1435)/+; sup-18(n1030)/+; sup-10(n983gf) male triple mutant moved only slightly better than sup-9(n1435)/+; sup-10(n983gf) mutants (23.7±0.6 vs. 21.6±0.9, mean ± SEM, respectively) (Fig. 5C), suggesting a very weak additive effect of sup-18(n1030)/+ and sup-9(n1435)/+. [This small effect might be caused by the presence in these animals of wild-type SUP-9 dimers at a fourth the wild-type level; this SUP-9 would respond to sup-18(n1030)/+ effects.] To verify the specificity of the interaction between sup-18(n1030) and sup-9(n1435), we tested sup-9(n264). sup-9(n264)/+ is as strong as sup-9(n1435)/+ in suppressing the locomotory defects of sup-10(n983gf) mutants. However, unlike sup-9(n1435)/+; sup-18(n1030)/+; sup-10(n983gf) mutants, sup-9(n264)/+; sup-18(n1030)/+; sup-10(n983gf) mutants moved better than sup-9(n264)/+; sup-10(n983gf) mutants (28.5±0.5 vs. 21.3±0.6 bends/minute, mean ± SEM, respectively) (Fig. 5C). This result is consistent with the finding that sup-18(n1030) and sup-9(n1435) lack an obviously additive effect in suppressing the locomotion and egg-laying defects of unc-93(e1500gf) mutants (Fig. 5A and B) and supports our conclusion that sup-9(n1435) and sup-18(lf) alleles act in the same pathway in affecting rubberband Unc mutants.
We determined the sup-9 coding sequences in sup-9(n1435) mutants and identified a C-to-T transition within codon 292, leading to a serine-to-phenylalanine substitution within the predicted intracellular C-terminal domain of SUP-9 (Fig. 6A). Although SUP-9 is 41%–47% identical in amino acid sequence over its entire region to several TASK-family two-pore domain K+ channels [28], the C-terminal cytoplasmic domain of SUP-9 is poorly conserved among these channels (Fig. 6A). However, the serine affected by the n1435 mutation is located in a small conserved stretch of amino acids with the sequence SxxSCxCY (Fig. 6A). We named this region the SC (Serine-Cysteine-rich)-box. The residues in the SC-box do not correspond to any reported motifs, including phosphorylation sites, as defined by the protein motif database PROSITE [47]. Variations of the SC-box are found in the human TASK-1 and TASK-3 channels and in two Drosophila two-pore domain K+ channels (Fig. 6A). We have not found an SC Box in other human two-pore domain K+ channels (I. de la Cruz and H. R. Horvitz, unpublished observations) or in TWK-4 (C40C9.1), a C. elegans two-pore domain K+ channel that is 41% identical to and the most closely related C. elegans channel to SUP-9 (Fig. 6A).
To determine if other residues in the SC-box of SUP-9 might function like the S292F substitution, we performed an in vivo mutagenesis study of the SC-box. We mutated residues S289, C293, C295 and Y296 to alanine individually and compared their effects in suppressing the egg-laying defects of the sup-10(n983gf) and unc-93(e1500gf) sup-18(n1030) double mutants. When assayed over a 3 hr period, both mutant strains laid fewer than three eggs, and a sup-9(n1913) null mutation drastically increased egg laying by both strains (Fig. 6B, C). As a control, overexpression of a sup-9(+) cDNA driven by the myo-3 promoter (Pmyo-3 sup-9(+)) in either sup-10(n983gf) or unc-93(e1500gf) sup-18(n1030) mutants did not increase egg-laying in each of three independent transgenic lines. By contrast, overexpression of a sup-9 cDNA containing the n1435 mutation (Pmyo-3 sup-9(n1435)) dominantly suppressed the egg-laying defects of sup-10(n983gf) mutants (Fig. 6B) but not those of unc-93(e1500gf) sup-18(n1030) animals (Fig. 6C). These results establish an in vivo assay for identifying mutations in sup-9 that preferentially suppress sup-10(n983gf) over unc-93(e1500gf) mutations.
A Pmyo-3 sup-9(S289A) and a Pmyo-3 sup-9(Y296A) transgene suppressed the defects of sup-10(n983gf) mutants but not of unc-93(e1500gf) sup-18(n1030) mutants, suggesting that the S289A and Y296A mutations act similarly to n1435 to mediate the gene-specific effects of sup-18(lf) mutations. By contrast, the cysteine-to-alanine mutations at residues 293 and 295 of SUP-9 suppressed both sup-10(n983gf) and unc-93(e1500gf) sup-18(n1030) mutants (Fig. 6B, C). We suggest that these mutations when overexpressed have a dominant-negative effect on the wild-type sup-9 allele.
To further understand how its C-terminal domain affects SUP-9 activity, we deleted in the sup-9 cDNA the region encoding the SUP-9 C-terminal cytoplasmic domain. We also replaced this region with the corresponding region of twk-4, which encodes a two-pore domain K+ channel without an SC-box, or of TASK-3, a mammalian homolog that contains an SC-box (Fig. 6). Deletion of the SUP-9 C-terminal domain caused suppression of both the sup-10(n983gf) and unc-93(e1500gf) sup-18(n1030) mutant phenotypes, suggesting that the truncated form of SUP-9 acts as a dominant-negative protein. Interestingly, both the sup-9::twk-4 and sup-9::TASK-3 fusion transgenes suppressed the sup-10(n983gf) egg-laying defect (Fig. 6B) but failed to suppress that of the unc-93(e1500gf) sup-18(n1030) mutants (Fig. 6C), suggesting that these fusion transgenes act similarly to sup-9(n1435) and affect rubberband Unc mutants in a gene-specific manner.
To identify more sup-9 mutations that act similarly to sup-9(n1435), we performed a genetic screen for mutations that semidominantly suppressed the sup-10(n983gf) rubberband phenotype (see Materials and Methods). We isolated eight mutations of sup-9 that define seven novel alleles (n3975 (n4265), n3976, n3977, n3935, n4259, n4262 and n4269) (Fig. 7A) and three additional mutations (n3942, n4253, n4254) that contained the same C-to-T transition and therefore caused the same S292F substitution as sup-9(n1435). As heterozygotes, five of the seven novel alleles (n3977, n3935, n4259, n4262, n4269) were stronger suppressors of sup-10(n983gf) mutants like sup-9(n1435)/+ (∼23 bends/minute), while the other two (n3975, n3976) were weaker (Fig. 7B). These mutations affect six different regions of SUP-9 (Fig. 7A), including the first (n3975) and second (n3977) transmembrane domains, the first pore domain (n3976), the beginning of the C-terminal cytoplasmic domain (n3935), the SC-box (n4259 and n4262), and a region C-terminal to the SC-box (n4269)
To determine if these novel sup-9 mutations conferred resistance to sup-18 activation or if they were simply dominant-negative lf mutations, we tested their responsiveness to changes in sup-18 levels in a similar manner to that used for testing sup-9(n1435) (Table 2 and Fig. 5). By comparing the locomotion of sup-9(mut)/+; sup-18(n1030)/+; sup-10(n983gf) mutants to that of sup-18(n1030)/+; sup-10(n983gf) mutants, we found that sup-9(n3935)/+, sup-9(n4259)/+, sup-9(n4262)/+ and sup-9(n4269)/+ caused a weak effect similar to that by sup-9(n1435)/+, while n3975/+, n3976/+ and n3977/+ caused a significant improvement in locomotory rate in response to a change in sup-18 levels (Fig. 7B). This result suggests that the channels generated by the three mutations n3975, n3976 and n3977 have impaired ability to generate K+ currents but retain regulation by SUP-18.
In addition to its sup-18 insensitivity, sup-9(n1435) was also a weak suppressor of the unc-93(e1500gf) locomotory defect, while the null mutation sup-9(n1913) completely suppressed the unc-93(gf) defect (Tables 1 and 5). Similarly, sup-9(n4259), sup-9(n4262) and sup-9(n4269) only weakly suppressed the locomotory defects of unc-93(e1500gf) animals (Fig. 7C), suggesting that these mutations belong to the class of sup-9 alleles defined by sup-9(n1435). However, sup-9(n3935) completely suppressed the locomotory defects of unc-93(e1500gf) animals (Fig. 7C), indicating that sup-9(n3935) was not only insensitive to sup-18 but also resistant to the activating effects of unc-93(e1500gf). Thus, mutations affecting different residues of SUP-9 confer differential channel sensitivity to its regulatory subunits.
Two-pore domain K+ channels are widely expressed and play important roles in regulating resting membrane potentials of cells [15], [17]. However, very little is known about protein factors with which these channels interact. We previously identified UNC-93 and SUP-10 as presumptive regulatory subunits of the SUP-9 two-pore domain K+ channel. We now suggest that SUP-18 also regulates the SUP-9/UNC-93/SUP-10 channel complex.
sup-18 encodes the C. elegans ortholog of mammalian iodotyrosine deiodinase (IYD), which belongs to the NADH oxidase/flavin reductase superfamily [7], [8]. By oxidizing NADH using flavin mononucleotide (FMN) as a cofactor, IYD catalyzes the recycling of iodide from monoiodotyrosine and diiodotyrosine, two major byproducts in the synthesis of thyroid hormones [7], [8]. Lack of IYD function can lead to congenital hypothyroidism [12], [13]. In C. elegans, no SUP-18 function besides regulating the SUP-9 channel has been identified. The enzymatic activity of SUP-18 remains to be defined.
Little is known about the metabolism and function of iodide in nematodes. The C. elegans genome contains two genes, ZK822.5 and F52H2.4, that encode homologs of the mammalian sodium/iodide symporter, which enriches iodide in the thyroid cells by active membrane transport [48]. The presence of both SUP-18 IYD and sodium/iodide symporter-like proteins suggests that iodide functions biologically in C. elegans. Although iodide appears not to be an essential trace element in the culture medium of C. elegans [49], it is possible that residual iodide in components of that medium can provide sufficient nutritional support for survival. C. elegans lacks homologs of mammalian iodothyronine deiodinase (I. de la Cruz, L. Ma and H. R. Horvitz, unpublished observations), enzymes that remove the iodine moieties from the precursor thyroxine (T4) and generate the more potent thyroid hormone 3, 5, 3′-triiodothyronine [50], which suggests that thyroid hormones might not be synthesized in C. elegans.
IYDs across metazoan species share a similar enzymatic activity in reductive deiodination of diiodotyrosine [51], and it seems likely that SUP-18 acts similarly in C. elegans. Like mammalian IYDs, SUP-18 contains a presumptive N-terminal transmembrane domain that is required for full activity. Interestingly, the SUP-18 intracellular region lacking the transmembrane domain could still partially activate the SUP-9 channel, suggesting that membrane association is not absolutely required for SUP-9 activation by SUP-18. Membrane association is important for mammalian IYD enzymatic activities [5], [52], [53].
The presence of a transmembrane domain suggests that SUP-18 IYD might interact with other transmembrane proteins. The genetic interactions we observe between sup-18 and the genes that encode the SUP-9/UNC-93/SUP-10 two-pore domain K+ channel complex support this hypothesis. Based on expression studies, we conclude that SUP-18 and SUP-10 localize to similar subcellular structures within muscle cells, further supporting the idea that SUP-18 and the channel complex interact physically. We found that transgenic expression of the SUP-18 intracellular domain could enhance the expression of the rubberband phenotype, suggesting that plasma membrane localization is not essential for SUP-18 function. Nonetheless, the expression of the full-length SUP-18 was more potent than the expression of the SUP-18 intracellular domain in rescuing the rubberband Unc phenotypes of sup-18(lf); sup-10(n983gf) mutants, suggesting that the presence of a transmembrane domain in SUP-18 IYD could enhance the activity of SUP-18 by targeting SUP-18 to the plasma membrane.
The crystal structure of mouse IYD reveals that eight residues contact the FMN cofactor: R96, R97, S98, R100, P123, S124, T235 and R275 [54]. Except T235, which is replaced by a serine in SUP-18, these residues are completely conserved (Figure 1C, yellow boxes). Furthermore, the sup-18(n1010) missense mutation leads to an S137N substitution at the position equivalent to the mouse S98 residue, likely disrupting the binding of FMN. This high degree of conservation at the cofactor binding site suggests that SUP-18 likely retains the ability to bind FMN and likely has a catalytic activity.
Three IYD missense mutations that cause hypothyroidism (R101W, I116T, and A220T) affect residues that are conserved in SUP-18 [12], [55] (Fig. 1C, red boxes). A fourth human mutation replaces F105 and I106 with a leucine [8]. The phenylalanine at position 105 is conserved in SUP-18 (Fig. 1C). The conservation of residues associated with IYD function supports the hypothesis that SUP-18 regulates the SUP-9 two-pore domain K+ channel complex via an enzymatic activity. The SUP-18 substrate remains to be elucidated.
That SUP-18 might function as a NADH oxidase/flavin reductase raises the intriguing possibility that SUP-18 might couple the metabolic state of muscle cells with membrane excitability. Mammalian Kvβ voltage-gated K+ channel regulatory subunits [56], which belong to the aldo-keto reductase superfamily [57], [58], have similarly been proposed to couple metabolic state with cell excitability based on indirect evidence. Kvβ2 has an NADP+ cofactor bound in its active site and a catalytic triad spaced appropriately to engage in enzymatic activity [58]. Although suggestive of an enzymatic activity, no substrate has been reported for Kvβ subunits. While Kvβ2 knockout mice have seizures and reduced lifespans, mice carrying a catalytic null mutation in Kvβ2 have a wild-type phenotype, suggesting that if an enzymatic activity for Kvβ2 exists, it is functionally dispensable in vivo [59]. By contrast, the predicted catalytic mutation sup-18(n1010) behaves like a null mutation in its inability to activate the SUP-9 channel, even though the SUP-18(n1010) protein is synthesized and localized normally to the cell surface of muscle cells (I. de la Cruz and H. R. Horvitz, unpublished observations). Five other sup-18 mutations affecting highly conserved residues in the NADH oxidase/flavin reductase domain also behave like null mutations, consistent with the hypothesis that SUP-18 enzymatic activity is essential for its function.
sup-18(lf) mutations strongly suppress sup-10(n983gf) mutants and weakly suppress unc-93(e1500gf) mutants. Certain specific mutations of sup-9, including n1435, n4259, n4262, and n4269, act similarly to sup-18(lf) and are strong suppressors of sup-10(n983gf) mutants and weak suppressors of unc-93(e1500gf) mutants. Together these sup-9 mutations and sup-18(lf) mutations represent a novel class of mutations that exhibit gene-specific suppression of the rubberband Unc mutants and are distinct from another class of gene-specific suppressors we identified previously, mutations in three splicing factor genes that strongly suppress unc-93(e1500gf) and sup-10(n983gf) but do not obviously suppress unc-93(n200gf) or sup-9(n1550gf) [60]–[62]. The difference between sup-18(lf) and sup-9(n1435, n4259, n4262, n4269) mutations and the splicing factor mutations in their patterns of suppressing the rubberband Unc mutants suggests that these two classes of suppressors function by distinct mechanisms.
SUP-9 is closely related to the subfamily of two-pore domain K+ channels that include human TASK-1 and TASK-3 [28]. TASK-1 is activated by multiple factors, including extracellular pH [22], [23], [63], inhalational anesthetics such as halothane [24] and oxygen [64]. TASK-1 is directly inhibited by sub-micromolar levels of the cannabinoid neurotransmitter anandamide [65] and by neuromodulators such as thyrotropin releasing hormone (TRH) [27]. A histidine residue in the first P-domain of TASK-1 modulates its sensitivity to pH [66], while a six amino acid stretch following its fourth transmembrane domain is required for both halothane activation and TRH suppression [24], [67]. Deletion of the TASK intracellular C-terminal domain, which contains the SC-box, does not change its basal activity or activation by halothane [24], [67], suggesting that the TASK-1 C-terminal domain and probably the SC-box represent an activation region that is required by some types of channel activator (e.g., human IYD) but not by others (e.g., halothane and pH). It remains to be determined whether IYD is involved in the inhibition of TASK-1 channel activity by TRH.
From our genetic analysis of the sup-9(n1435) mutation and site-directed mutagenesis of sup-9, we have defined the SC-box, a domain of SUP-9 required for SUP-10(n983gf)-specific activation. The importance of the SC-box in mediating this activation is supported by the results of a genetic screen in which we isolated additional sup-9 mutations (Fig. 7) that act like sup-9(n1435) and cause distinct amino acid changes in (n4259 (S292A), n4262 (S294A)) or near (n4269 (L303P)) the SC-box. Although conserved in the human TASK-1 and TASK-3 channels (Fig. 6A), no function has yet been assigned to the SC-box. Our analyses suggest that the SC-box and the C-terminal domain of SUP-9 likely mediate the functional interaction between SUP-9 and SUP-10/SUP-18 but are dispensable for interaction with UNC-93. We found that replacing the C-terminal domain of SUP-9 with the corresponding region of TWK-4 (which lacks an SC-box) or of TASK-3 (with an SC-box) makes the fusion channels behave like SUP-9(n1435), consistent with the model that the SC-box is required for SUP-9 activation by SUP-10(n983gf) and SUP-18 (based on the TWK-4 data) and suggests that SC-box-dependent activation requires one or more nearby residues in the C-terminal domain (based on the TASK-3 data). The unc-93(e1500gf) mutation results in a glycine-to-arginine substitution at amino acid 388 in one of the putative transmembrane domains [33], suggesting that the UNC-93(gf) protein activates SUP-9 through an interaction involving transmembrane domains, without a need for the SC-box or the rest of the cytoplasmic domain.
We describe three important properties of the unusual sup-9(n1435) mutation. First, SUP-9(n1435) channels cannot be activated by SUP-10(n983gf). Second, SUP-9(n1435) channels are insensitive to SUP-18 activity. Third, SUP-9(n1435) channels can be activated by UNC-93(e1500gf). The existence of a channel mutation that is insensitive to both SUP-18 and SUP-10(n983gf) suggests that these two inputs act through a common pathway. A mutant channel that can be activated by UNC-93(e1500gf) but not by SUP-10(n983gf) suggests that there is an independent pathway for SUP-9 activation by UNC-93.
We propose a model to explain the functional interactions between SUP-18 and SUP-9/UNC-93/SUP-10 (Fig. 8). In this model, SUP-10 and UNC-93 have an essential role in and are both required for activating SUP-9 channel, since the n1550 gf mutation in sup-9 is completely suppressed by sup-10(lf) and unc-93(lf) mutations [38]. SUP-18 activates SUP-9 only weakly and relies on SUP-10 for this activation (Fig. 8). SUP-10(n983gf) enhances the activity of SUP-18 and results in over-activation of SUP-9 by SUP-18. Our model is consistent with the genetic and molecular evidence described in this and previous studies [28]–[31], [33] and should provide a framework for understanding the interactions of SUP-18 and the SUP-9/UNC-93/SUP-10 channel complex. Our results do not distinguish whether SUP-18 regulates the SUP-9/UNC-93/SUP-10 complex via a direct physical interaction or indirectly through an unknown factor or factors.
In short, we identified SUP-18 IYD as a functional regulator of the SUP-9/UNC-93/SUP-10 two-pore domain K+ channel complex. We also defined an evolutionarily conserved serine-cysteine-rich domain, the SC-box, in the C-terminal region of SUP-9 and showed that this region is required for activation of the channel by SUP-18. Since IYD is likely to be an NADH oxidase/flavin reductase that uses the ubiquitous energy carrier molecule NADH as a coenzyme, our study suggests that IYD might couple cellular metabolic state to two-pore domain K+ channel activities. Future molecular analyses should reveal the mechanism underlying the interaction between the SUP-9 two-pore domain K+ channels and SUP-18 IYD.
C. elegans strains were cultured as described [49], except that E. coli strain HB101 was used instead of OP50 as a food source. Strains were grown at 20°C unless otherwise noted. The following mutations were used in this study:
LGII sup-9(n213, n233, n264 [31], n1016, n1025 [30], n1435, n1550gf [29], lr35, lr38, lr45, lr100, lr129, lr142, n1472, n1557, n1913 [28], n3935, n3942, n3975, n3976, n3977, n4253, n4254, n4259, n4262, n4265, n4269 (this study)).
LGIII unc-93(e1500gf, n200gf) [31], sma-3(e491) [49], mec-14(u55) [68], ncl-1(e1865) [69], unc-36(e251) [49]. sup-18(n463, n527, n528, n1010, n1014, n1015, n1022, n1030, n1033, n1035, n1036 [30], n1038, n1471, n1539, n1548, n1554, n1556, n1558 (this study)).
LGX sup-10(n183 [31], n1008, n983gf [30], n4025, n4026 (this study)), lin-15(n765ts) [70].
Since lf mutations in sup-10 completely suppress the paralysis of unc-93(e1500gf) mutants [31], we reasoned that partial lf mutations of sup-10 would partially suppress the unc-93(e1500gf) locomotory phenotype. To isolate such partial lf sup-10 mutations, we performed an EMS F2 genetic screen for partial suppressors of the locomotory defects of unc-93(e1500gf) mutants. From 17,500 haploid genomes screened, we isolated over 30 strong suppressors and seven weak suppressors. We assigned two of the seven weak suppressors, n4025 and n4026, to the sup-10 locus by complementation tests and three others to the unc-93 locus. All seven were saved for future analyses.
34 Sma non-Unc and 23 Unc non-Sma progeny were isolated from a sma-3 mec-14 ncl-1 unc-36/sup-18 parent. Scoring of the ncl-1 and sup-18 phenotypes identified the 57 recombination events to be distributed in the three relevant intervals as follows: sma-3 (30/57) ncl-1 (3/57) sup-18 (24/57) unc-36. A pool of cosmids C33C3, C08C3, C27D11, C02C2, C39F10 and C44C9 at 1 ng/µL each and a rol-6 marker [71] at 80 ng/µL were injected into sup-18(n1010); sup-10(n983gf) animals. Two Rol transgenic lines were obtained, one of which generated rubberband Unc animals. The four middle cosmids were injected separately, and C02C2 yielded 5/5 rescued lines, while transgenes containing cosmids C08C3 (0/7), C27D11 (0/5) or C39F10 (0/9) showed no rescue.
RT-PCR was performed on cDNA from the wild-type N2 strain using the primers 5′-TTGAAAACCCCTGTTAAATAC-3′ and 5′-CGAGTTTCTAATAAAAATAAACC-3′. PCR products were cloned into pBSKII (Stratagene), and their sequences determined. 5′ and 3′ RACE were performed using the corresponding kits (Gibco).
Genomic subclones of cosmid C02C2 were generated in pBSKII (Stratagene). The subclones, in the order shown in Figure 1, spanned the following sequences (Genebank Acc#L23649): EcoRV (9,790) - EcoRV (21,098); PstI (23,699) - PstI (32,833); PstI (23,699) - SacI (28,185); BstBI (24,448)-SacI (28,185); and HindIII (24,671) - HindIII (27,169).
All PCR amplifications used in plasmid constructions were performed using Pfu polymerase, and the sequences of their products were determined. The Pmyo-3 sup-18 vectors for ectopic expression of wild-type or mutant sup-18 alleles were generated by PCR amplification of the respective coding regions from sup-18 cDNAs using primers that introduced NheI and SacI sites at the 5′ and 3′ ends, respectively, and cloned into vector pPD95.86 (from A. Fire). Pmyo-3 sup-18(intra) was similarly constructed, except that the 5′ primer began at codon 66 of sup-18. The gfp-tagged version of this vector was created by PCR amplification of the gfp coding sequence from vector pPD95.77 (from A. Fire) and subcloned into Pmyo-3 sup-18(intra) just prior to the start codon of the sup-18 sequence.
Pmyo-3 mIYD (mouse IYD) was generated by PCR amplification of the coding region of the mouse cDNA (Gene Bank AK002363) with 5′ and 3′ primers containing NheI and EcoRV sites, respectively, and subcloning the PCR products into pPD95.86 at the NheI and SacI (blunted) sites. Pmyo-3 mIYD::gfp was generated by a similar strategy using a 5′ primer containing an NheI site and a 3′ primer that did not include the stop codon of mIYD but instead contained a BamHI site. The myo-3 promoter from pPD95.86 was subcloned into pPD95.77, such that upon subcloning of the mIYD PCR fragment into the NheI and BamHI sites of the vector the myo-3 promoter drove expression of the mIYD gene fused in-frame at its 3′ end to gfp.
The sup-18::gfp genomic fusion was constructed by introducing SphI sites at the ends of a gfp cassette by PCR amplification of plasmid pPD95.77 (from A. Fire) and subsequent subcloning into the single SphI site contained within a 9.1 kb PstI genomic sup-18 rescuing fragment. The resulting fusion contained 6.5 kb of promoter sequence, the entire sup-18 coding region with gfp inserted between the transmembrane and NADH oxidase/flavin reductase domains and 1.1 kb of 3′ UTR and downstream sequence.
The sup-10::gfp fusion used in colocalization studies was constructed by subcloning a 7.3 kb MfeI genomic fragment from cosmid C27G6 containing sup-10 into the EcoRI site of pBSKII. A 6.4 kb Pst I fragment was subcloned from this vector into pPD95.77, which contained 3.5 kb of promoter sequence and the sup-10 coding region. Using PCR, we introduced a SalI site immediately preceding the stop codon of sup-10 to create an in-frame fusion with the gfp coding sequence.
sup-18::β-galactosidase fusions were created by PCR amplification of 1869 bp of 5′ sup-18 promoter sequence and subcloning the product into the SphI and PstI sites of pPD34.110 (from A. Fire) to generate Psup-18 TM-β-Gal, which contains a synthetic transmembrane sequence [45] followed by the β-galactosidase coding sequence [72]. sup-18 genomic coding sequence spanning codons 1–42, 1–70 and 1–301 were PCR-amplified from the minimal rescuing fragment with 5′ and 3′ primers that contained PstI and BamHI sites, respectively, and subcloned into these sites in Psup-18 TM-β-Gal. The synthetic transmembrane domain was deleted from these plasmids by excising the KpnI fragment containing this domain. A signal sequence [73] was inserted into these vectors using standard PCR techniques.
The GST::sup-18(N) and MBP::sup-18(N) fusion genes used to generate and purify anti-SUP-18 antibodies were generated by PCR amplification of codons 1–258 of the sup-18 cDNA and subcloning the products into pGEX-2T (Pharmacia) and pMal-2c (NEB) vectors.
The full-length twk-4 cDNA was cloned by RT-PCR with primers 5′-CTCTGCTAGCAATGCATCAAATTGACGGAAAATCTGC-3′ and 5′-AGAGGATCCATATAGTTCAAGATCCACCAGATG-3′ from wild-type mixed-stage RNA. The sequence of the twk-4 cDNA obtained was in agreement with its predicted sequence (GenBank Acc#AF083646). The C-terminal cytoplasmic domain of sup-9 from the Pmyo-3 sup-9 vector (codons 257–329 of sup-9) was replaced by twk-4 codons (265–365) using standard PCR ligation techniques to generate Pmyo-3 sup-9::twk-4. Site-directed mutagenesis of the SC-box in the Pmyo-3 sup-9 vector was likewise performed.
Young adults were individually picked to plates with HB101 bacteria, and body-bends were counted for one minute using a dissecting microscope as described [74].
A GST::SUP-18(N) fusion protein was expressed in E. coli and the insoluble protein was purified by SDS-PAGE and used to immunize rabbits. Antisera were purified by binding to the MBP::SUP-18 protein immobilized on nitrocellulose strips and elution with 100 mM glycine-HCl (pH 2.5). This antibody could detect SUP-18 overexpressed in the body-wall muscles (Fig. 2H) but failed to detect endogenous SUP-18.
For immunofluorescence experiments, worms at mixed stages were fixed in 1% paraformaldehyde for 2 hrs at 4°C and permeabilized as described [75]. For colocalization studies, transgenic worms were stained with primary antibodies at 1∶200 dilution and a secondary goat-anti-rabbit antibody conjugated with Texas Red (Jackson Labs). Worms were viewed using confocal microscopy.
Germline transformation experiments were performed using standard methods [71]. Transgenic strains carrying the lin-15(n765ts) mutation contained the coinjection marker pL15EK(lin-15(+)) at 50 ng/µL [70], and transgenic animals were identified by their non-Muv phenotype at 22.5°C. The dominant rol-6 plasmid [71] was used at 100 ng/µl during cosmid rescue experiments, and transgenic animals were identified by their Rol phenotype. The dominant myo-3::gfp fusion vector pPD93.97 (from A. Fire) was used where indicated at 80 ng/µl, and transgenic animals were identified by GFP fluorescence. Experimental DNA was injected at 30–50 ng/µl.
One plausible genetic strategy for isolating sup-9 alleles similar to sup-9(n1435) would be to perform an F2 screen for suppressors of the sup-10(n983gf) locomotory defect and then test these suppressors for their effects on the locomotory defect of unc-93(e1500gf) mutants. Most sup-9 alleles isolated from such a screen would be typical lf alleles rather than rare alleles that would result in a SUP-9 protein specifically impaired in activation by SUP-10(gf) and SUP-18(+). We therefore opted for an alternative strategy based on the semidominance of the sup-9(n1435) mutation. While sup-9 null mutations, such as n1913, recessively suppress the locomotory defects of sup-10(n983gf) mutants, sup-9(n1435) caused a strong semidominant suppression (Fig. 5C). As two-pore domain K+ channels are homodimers [66], [76], this semidominance likely reflects the formation of nonfunctional heterodimers composed of n1435 and wild-type SUP-9 proteins. The strength of this semidominance (∼23 vs. ∼5 bends/minute for sup-9(n1435)/+; sup-10(n983gf) vs. sup-10(n983gf) mutants, respectively) formed the basis of an F1 screen for suppressors of the sup-10(n983gf) locomotory defect.
sup-10(n983gf) L4 hermaphrodites were mutagenized with EMS, and approximately 550,000 F1 progeny (1.1×106 genomes) were screened for improved locomotion on agar plates. From 89 candidate suppressors, 35 mutants retested in the next generation, representing at least 31 independent isolates. To quantify the semidominant character of these mutants (sup(new)), wild-type males were crossed with homozygous mutant hermaphrodites to generate sup(new)/+; sup-10(n983gf)/0 males, and their locomotory rate was scored. Because sup-10 is on the X chromosome, this strategy generates males hemizygous for sup-10(n983gf) while heterozygous for autosomal mutations, providing a convenient assay of semidominance. Four mutations completely suppressed the rubberband Unc phenotype of males, with locomotory rates very similar to that of wild-type animals (∼33 bends/minute). We reasoned that these four mutants were likely lf alleles of sup-10, as such animals would be hemizygous for sup-10. We confirmed this assignment by determining the sequences of the sup-10 locus and found mutations in all four strains (I. de la Cruz and H. R. Horvitz, unpublished observations). For the remaining strong mutants, we performed complementation tests with sup-9, sup-18 and unc-93 strains and identified 11 semidominant alleles of sup-9 (see Results).
|
10.1371/journal.pgen.0030031 | Genetic Complexity and Quantitative Trait Loci Mapping of Yeast Morphological Traits | Functional genomics relies on two essential parameters: the sensitivity of phenotypic measures and the power to detect genomic perturbations that cause phenotypic variations. In model organisms, two types of perturbations are widely used. Artificial mutations can be introduced in virtually any gene and allow the systematic analysis of gene function via mutants fitness. Alternatively, natural genetic variations can be associated to particular phenotypes via genetic mapping. However, the access to genome manipulation and breeding provided by model organisms is sometimes counterbalanced by phenotyping limitations. Here we investigated the natural genetic diversity of Saccharomyces cerevisiae cellular morphology using a very sensitive high-throughput imaging platform. We quantified 501 morphological parameters in over 50,000 yeast cells from a cross between two wild-type divergent backgrounds. Extensive morphological differences were found between these backgrounds. The genetic architecture of the traits was complex, with evidence of both epistasis and transgressive segregation. We mapped quantitative trait loci (QTL) for 67 traits and discovered 364 correlations between traits segregation and inheritance of gene expression levels. We validated one QTL by the replacement of a single base in the genome. This study illustrates the natural diversity and complexity of cellular traits among natural yeast strains and provides an ideal framework for a genetical genomics dissection of multiple traits. Our results did not overlap with results previously obtained from systematic deletion strains, showing that both approaches are necessary for the functional exploration of genomes.
| A familiar face or a dog breed is easily recognized because morphology of individuals differs according to their genetic backgrounds. For single-cell organisms, morphology reduces to the shape and size of cellular features. Microbiologists noticed that the shape of S. cerevisiae cells (baker's yeast) differs from one strain to another, but these differences were usually described qualitatively. We used a high-throughput imaging platform to study the morphology of yeast cells when they divide. Cells were stained with three fluorescent dyes so that their periphery, their DNA, and their actin could be recognized, and their images were analysed by a specialized software program. Numerous morphological differences were found between two distant strains of S. cerevisiae. By crossing these two strains, we performed quantitative genetics: several loci controlling morphological variations were found on the genome, and correlations were made between gene expression and morphology changes. Using bioinformatics, we showed that the results obtained do not overlap with previous results obtained from yeast cells in which specific genes are deleted. The study, therefore, illustrates how mutagenesis and the use of natural genetic variations provide complementary knowledge.
| Yeast genetics has long been powered by the ease of conducting genome manipulation and mutagenic screens. These experiments are usually performed on a restricted panel of laboratory strain backgrounds that serve as standards. Variability between backgrounds is often viewed as a problem that must be minimized by using nearly isogenic strains whenever possible. As an alternative approach, several recent studies have used wild-type strains from divergent backgrounds to identify regulators of specific phenotypes such as high-temperature growth [1], sporulation efficiency [2], drug response [3], telomere homeostasis [4], or global transcriptional regulations [5]. This approach, which relies on genome scans for quantitative trait loci (QTL), emerged after high-throughput genotyping was facilitated by oligonucleotide microarrays [6]. It offers an effective alternative to conventional yeast genetics by employing the natural genetic diversity of wild yeast strains [7,8], thereby furthering the study of natural genetic resources.
In addition, exploring strain-to-strain variation in yeast or other model systems is essential to our understanding of the regulation of complex traits [9]. For example, one yeast study described the complexity of a QTL containing three genes each contributing to phenotypic variation [1]. Another study mapped transcriptional regulators, estimating the proportion of cis- and trans-regulatory variations and providing evidence for “master” regulators [5]. Similar conclusions were later obtained from mouse and human [10,11]. Since then, QTL controlling gene expression (eQTL) are sometimes included in mapping designs (an approach sometimes referred to as “genetical genomics”) and can provide judicious prioritization of candidate genes [12]. Furthermore, multiple phenotypes are now being acquired from patients or agronomical organisms, and strategies for QTL mapping of multiple traits are being given increasing attention.
Functional annotations of genomes are highly dependent on the sensitivity and scale of phenotypic tests [13]. Consequently humans, as well as pet dogs or cats, provide an excellent example of a system for understanding physiological variations, in addition to studying disease mechanisms, because they benefit from receiving extensive medical care. However, the capacity to manipulate genomes of model organisms is essential, and batteries of phenotypic tests have been developed for most of them, including mouse [14] and yeast. One of the most sensitive methods used to detect phenotypic variation in yeast has been the high-throughput characterization of cellular morphology [15]. This method is based on triple fluorescent staining of fixed cells using concanavalin-A, 4′,6-diamidino-2-phenylindole (DAPI), and phalloidin to label the cell wall, DNA, and actin, respectively. Microscopy images are then automatically acquired and analysed to quantify simultaneously 501 parameters from hundreds of cells.
We have applied this phenotypic characterization to a cross between two wild-type divergent yeast strains, which had previously been used in the study of the segregation of gene expression traits [16]. As described below, we showed that yeast cellular morphology is a complex trait with evidence of both transgressive segregation and epistasis. Using previously published datasets, we identified QTL controlling numerous morphological traits and found correlations between morphological and gene expression traits. We applied a bioinformatic comparison of these results to the results obtained when phenotyping systematic deletion strains and showed that the two approaches are complementary.
We looked for cellular morphological differences between laboratory strain BY4716 (isogenic to S228c) and strain YEF1946, isogenic to RM11-1a, which is a haploid derivative of a wine strain (kindly provided by E. Foss). The original RM11-1a could not be used because of its clumpy nature [17], which was suppressed in YEF1946 by a single base substitution in the AMN1 gene. For simplicity, strains BY4716 and YEF1946 will, nonetheless, be referred to as BY and RM hereafter. Samples from nine independent cultures of each strain were characterized by triple-staining fluorescent microscopy and automated cell imaging as described previously [15]. At least 200 cells were analysed per culture to quantify 501 morphological parameters that were each considered as a quantitative trait (Table S1). For each trait, we tested the difference between the nine BY and the nine RM values using the Wilcoxon Mann-Whitney test. At p < 0.001, 143 traits showed significant difference. A permutation test determined that only one trait was expected to differ by chance at this p-value. The morphological differences between the two strains were found at different stages of the cell cycle and reflected various cellular aspects (Table 1). After nuclear division, buds from BY cells were bigger than those from RM cells (Figure 1A and 1B); they were also more elongated in BY (Figure 1C). No difference was seen in the direction of bud growth (Figure 1D). Mother cells from BY were more elongated (Figure 1E and 1F), bud necks were bigger in RM cells at early division (Figure 1G), and cell-wall thickness was more homogeneous in BY cells (Table 1). Surprisingly, a large majority of the differences corresponded to DNA staining patterns (106 of the 143 traits) (Figure 1H–1L; Table 1). However, DNA staining is covered by 272 of the 501 parameters estimated by CalMorph, which is a significant enrichment as compared to actin- or cell-wall–related parameters. In addition, using the dataset from Ohya et al. [15], we found that DNA-related traits had lower measurement errors than other traits (see Materials and Methods). Thus, our procedure seemed to have a better ability to detect differences in intracellular DNA distribution than differences in actin distribution or cell shape, possibly explaining the majority of DNA-related parameters in Table 1. In addition to phenotypic differences at specific stages of the cell cycle, we observed variation of the proportion of cells at particular stages. For example, the fraction of RM cells with small buds already containing DNA was high, whereas the progression of bud growth was similar in both strains. Similarly, the fraction of budded cells containing only one DNA region was small in RM. These observations could result from two phenomena: either premature nuclear division may occur in RM cells even when bud growth is inadequate, or the fraction of cells having passed nuclear division is overestimated in RM. In the former case, parameters reflecting actin organization should differ as well, because apical actin distribution begins in G1 and continues until beginning of nuclear division [18]. This was not the case, since we did not observe any change in the fractions of cells showing apical actin distribution over all stages (parameters A106 and A107). Alternatively, mitochondria were more abundant in RM cells when observed after MitoTrackerGreen labelling (unpublished data), and these organelles are known to move into the bud prior to the accomplishment of nuclear division [19]. It is, therefore, possible that mitochondria abundance or differences in their intracellular distribution bias the inventory of cells containing two nuclei.
In a previous study, the segregation of gene expression levels was studied across 112 F1 segregants from BY × RM [16]. We used these F1 strains to study the genetic segregation of morphological differences between BY and RM. Of the 112 segregants provided, five were flocculent and 45 were clumpy. These strains were not suitable for image analysis of isolated cells and were therefore discarded. The remaining 62 strains were cultured in triplicates and over 200 cells per culture were processed to quantify 501 morphological parameters. Each parameter was then treated as a quantitative phenotype. Median heritability among all 501 parameters was 49%, with 71, 121, 146, and 101 showing low (0%–25%), moderate (25%–50%), significant (50%–75%), or high (>75%) heritability. This suggested that experimental errors were low enough to study the genetic control of the majority of the traits (Figure 2A). We noted that 62 phenotypes had negative heritability values (no detectable genetic variance), which either means that their measurement errors are too high to detect genetic control, or that the genes controlling them do not harbour functional variations between the BY and RM backgrounds. Intriguingly, 16 of those traits belonged to the list of 143 traits differing between BY and RM (Table 1), suggesting epigenetic control. Consistently, many of these traits might be affected by mitochondrial DNA distribution (for example, the “mobility-of-nucleus-in-mother-cell”). Differences in mitochondria abundance or spatial repartitions between BY and RM can explain “non-heritable” differences, since mitochondria are not inherited via Mendelian segregation but undergo a complex fusion process through meiosis [20].
To estimate the complexity of the genetic control, we looked for cases of transgressive segregation or epistasis. We applied tests previously adapted for multiple traits [16] and found that about one-fourth and two-fifths of all phenotypes showed transgressive segregation and epistasis, respectively. We detected 34 phenotypes significantly transgressive at a False Discovery Rate (FDR) of 0.05, and 98 phenotypes significantly epistatic at FDR = 0.05 (see Materials and Methods; Table S2 and Table S3). As examples, the segregation patterns of two such traits, the short axis length of the mother cell and the long axis length of the bud, are shown in Figure 2B and Figure 2C, respectively.
Using a genetic map of 3,042 markers previously generated by Brem et al. [16], we sought to map QTL of morphological traits. Of the 501 total morphological traits, 254 were discarded because they showed low or moderate (<50%) heritability. Of the remaining 247 traits, 95 and 67 could be mapped (i.e., at least one locus controlling their variation could be mapped) at p < 9.04 × 10−5 and p < 3.43 × 10−5, respectively. A permutation test determined that these p-values corresponded to FDR = 0.10 and FDR = 0.05, respectively. A total of seven distinct loci controlling specific morphological features were identified at FDR = 0.05 (Table 2; Table S4). When several phenotypes were mapped to the same locus, they were usually different estimators of the same cellular features (for example, “long-axis-length-in-mother” and “distance-between-middle-point-of-neck-and-mother-center” are two measurements of the mother cell elongation). The phenotypes were mapped for various aspects of cellular morphology: the localization and the shape of the DNA regions within the mother cell or the bud, the heterogeneity of the DNA staining, and the size or shape of the mother cell (Figure 3; Table S4). Surprisingly, 12 traits informative of the size, shape, and position of the DNA staining were mapped to two unlinked loci located on Chromosomes XIV and XV, respectively.
In the study from Brem et al. [16], all 62 F1 strains were grown in the same conditions as here, and their expression profiles were determined on DNA microarrays. Using this dataset, we searched for genetic correlations between morphological trait values and gene expression levels by computing the absolute value of the Pearson correlation coefficient (|R|). Among 247 traits showing heritability greater than 50%, 104 and 29 could be correlated to the expression level of at least one gene at |R| > 0.565 and |R| > 0.62, respectively. A permutation test determined that these correlation values corresponded to FDR = 0.1 and FDR = 0.05, respectively. At FDR = 0.1, a total of 364 correlations involved the expression levels of 103 genes (Figure 4; Table S5). We found several cases where annotations of the genes involved were consistent with the correlated phenotypes. For example, expression levels of FLO11 (YIR019C) and ECM34 (YHL043W, involved in cell wall regulations) were correlated to brightness differences of the cell wall. To look more systematically for such consistencies, we clustered hierarchically the 104 traits and 103 genes involved, and examined territories of the correlation map containing several gene/traits correlations (Figure 4). We found four such territories where gene ontology (GO) annotations were indicative of a cellular process or component correlated to traits. Expression levels of PET117 (YER058W), SAL1 (YNL083W), SCO1 (YBR037C), YNR036C, DBF20 (YPR111w), and YHR080C were correlated to nine traits measuring the position of DNA in the mother cell and in the bud after nuclear division. This suggested a link between DNA positioning and the GO terms “protein metabolism” (4/6 genes, p = 0.02) and “mitochondrion” (5/6 genes, p = 0.00032). Since DNA positioning estimation can be affected by mitochondrial DNA staining, it is very likely that these variations in mitochondrial activities between BY and RM cells are associated with different regional distributions of mitochondrial DNA. Expression of 22 genes, including TOP2 (YNL088W) and MSH2 (YOL090W), were correlated to 18 traits also describing DNA positioning in mother cells and buds after nuclear division. These correlations linked these traits to many GO terms reflecting DNA metabolism, including “DNA-dependent DNA replication” (4/22 genes, p = 0.00014) and “DNA repair” (3/22 genes, p = 0.012). This association suggests that differences in nuclear DNA metabolism drive differences in DNA positioning. Expression levels of SEH1 (YGL100W), YHR200W, HTA1 (YDR225W), GRD19 (YOR357C), and HHT2 (YNL031C) were correlated to ten traits describing the position of DNA during cellular division, the size of the bud neck, and the position of DNA in the mother cell after nuclear division. These traits were therefore associated to GO terms “chromatin assembly or disassembly” (2/5 genes, p = 0.002). This suggests that differences in DNA intracellular distribution probably result from differences in chromatin dynamics throughout nuclear division. Finally, expression levels of 23 genes, many of which being involved in pheromone response, were correlated to 39 traits describing the mother cell shape (GO terms “conjugation” (17/23 genes, p < 10−25) and “response to pheromone” (15/23 genes, p < 10−24). This association strongly suggested that cell shape differences resulted from differences in the activation of the pheromone response pathway between the two genetic backgrounds. A direct explanation of this is given in the next section.
We then sought to identify polymorphisms responsible for morphological differences. We focused on 16 cell-elongation traits linked to a locus on Chromosome VIII that contained the GPA1 (YHR005C) gene (Table 2). A previous study showed that a single polymorphism in GPA1, S469I, was responsible for constitutive residual activation of pheromone response genes in BY [17]. Response to pheromone includes elongation of cells that prepare for mating (“shmoo” phenotype). This polymorphism was, therefore, an excellent candidate to explain the morphological QTL. To test the GPA1-S469I polymorphism for cell-elongation differences, we measured morphological phenotypes of BY-gpa1I469S, an engineered strain isogenic to BY except that it carried the RM allele of GPA1 [21]. Of the 16 traits linking to GPA1, nine differed significantly between the nine replicates of BY and RM (p < 0.05), and seven of these differed accordingly between BY and BY-gpa1I469S (p < 0.05) (Table 3). Values for one trait, the long-axis length in the mother cell, are shown in Figure 5. The results demonstrated that the GPA1-I469S polymorphism was responsible for cell-elongation differences. The fact that several traits linked to the GPA1 locus but were not significantly different between BY and RM (or between BY and BY-gpa1I469S) can be due to a reduced statistical power when comparing strains (2 × 9 values), as compared to linkage test (62 values), or to additional undetected QTL with opposing effects.
In a separate study, the cellular morphological alterations due to single gene disruptions were investigated in the context of the BY genetic background [15]. For all 4,718 nonessential genes of the genome, a haploid strain deleted for that gene was analysed using the same staining protocol and imaging platform as we used here, giving quantification of the same 501 morphological traits. We sought to compare the results from this systematic-deletion strategy to the results obtained in our quantitative genetics approach, addressing the following question: When a morphological trait T is mapped to a QTL, is there a gene in the neighbourhood of this QTL of which the deletion affects T? If the answer is yes, a polymorphism in this gene could explain the QTL mapping result. We found a single case where this overlap between the two datasets was observed, and we estimated that finding one such case by chance only was highly expected (see Materials and Methods). Therefore, the majority of QTL results cannot be “explained” merely by genetic variation occurring in genes previously identified from the deletion set.
We found extensive cellular morphological variations between two wild-type S. cerevisiae backgrounds. The appearance of yeast is therefore polymorphic, as is the appearance of two unrelated human beings or animals. Out of curiosity, we asked colleagues to distinguish BY4716 from YEF1946 liquid cultures under the microscope (light microscopy on living cells, 40× magnification). From their responses, it was obvious that no difference would have been characterized without the fluorescence staining and automated quantification that we used. We found significant QTL for only 27 of the 143 traits differing between the two backgrounds. Differences in actin distribution in buds, bud shape and size, early bud growth, thickness of the cell wall, or neck width, as well as most traits related to DNA distribution prior to nuclear division could not be correlated to genomic loci. This can result from the small heritability of some of the phenotypes, as mentioned above for parameters that can be affected by the abundance and distribution of mitochondria. It can also result from the high complexity of the genetic control, whereby many loci contribute to the phenotype, but their small individual effect prevents their detection. Conversely, 40 of the 67 traits for which QTL were found were not in the list of the significant differences between parental strains. This was the case for phenotype CCV115_C describing the variation of mother-cell axis ratio across the sample (mapped to Chromosome X), and phenotype DCV194_C describing signal heretogeneities in mother cell DNA (mapped to Chromosome XV 930000). There are several possible explanations for this: some of these linkage results might be false positives (which we would expect to be possible for one or two but certainly not 40 phenotypes), alleles acting in the opposite direction might shade their effects in the parental strains (which would be consistent with the extent of transgressive segregation), or simply because statistical power was sometimes higher during mapping (differences among 62 values in segregants instead of 18 in the parents).
Notably, only 12 out of the 67 traits for which significant linkage was found were mapped to more than one place of the genome. Considering the large extent of transgressive segregation and epistasis, it is very likely that other loci contribute to morphological variations, and detecting only one or two per trait is probably a statistical power limitation (<100 segregants).
We did not test additional wild backgrounds, but one could very well apply a similar protocol to many divergent strains and compare their morphological distance to their genetic or transcriptomic divergence. Such a study may indicate whether morphological differences co-evolved with genetic or regulatory divergence, or if they were driven by other differences such as environmental conditions.
One of the challenging aims of modern and future quantitative genetics is the simultaneous dissection of multiple traits. For example, clinical phenotypes are now collected systematically from cohorts of individuals, and molecular phenotypes such as biochemical dosage or gene expression profiling can be included in mapping strategies [14,22]. We show here that yeast cellular morphology represents a large set of quantitative phenotypes with complex inheritance. Although we applied a very basic mapping strategy involving single markers and single traits, the dataset presented here can provide a model framework for development and evaluation of mapping methodologies optimised for multiple traits [12,23]. In particular, we showed that many morphological phenotypes could be correlated to the inheritance of gene expression levels. For example, the brightness difference of the cell wall (highest minus lowest concanavalin-A signal along the wall of one cell) was correlated to the expression of FLO11, which is involved in cell-surface variation within yeast cultures [24]. In several cases, the gene expression trait in question was previously mapped to an eQTL [16], whereas the morphological trait remained unmapped. This probably results from the higher statistical power of Brem et al. who used 112 segregants instead of 62. This situation is similar to a clinical case where a disease trait is correlated to an expression signature on the basis of few clinical samples, but where genetic variations controlling this signature are mapped from a large-scale study contributed by many human donors. In this context, eQTL are candidate QTL for the correlated trait. We found 224 cases where a morphological trait was significantly correlated to a gene expression level for which Brem et al. identified an eQTL. We re-examined the nominal linkage p-values between these loci and the corresponding morphological trait: 92 cases showed p < 0.002, which is an acceptable threshold to account for multiplicity when considering only 224 candidate loci instead of the entire genome. Although most of these cases reflected genetic linkages already identified in our genome scan, they suggested eight additional loci. This indirect mapping of QTL via the use of eQTL seemed therefore promising. However, we tried but failed to validate candidate genes at one of these loci by engineering and characterizing strains where alleles were replaced. We consider that further investigation of these eight loci needs to be done before declaring them as valid QTL. The approach is nonetheless likely to be helpful in studies combining eQTL and phenotypic mapping.
Interestingly, several traits that we mapped were not direct measures of cellular features but coefficients of variation of such measures across the sample. In several cases, the measure itself (e.g., its mean) was not linked to the QTL controlling its coefficient of variation. For example, parameter C13_C, which measured the fitness of the mother cell for ellipse, was not mapped and was not correlated to the GPA1 genotype (p = 0.15), whereas its coefficient of variation was mapped to GPA1, and this mapping was further validated by the gpa1-I469S strain (p = 0.01). This argues that the gpa1I469S polymorphism does not affect the shape of all cells but rather the fraction of cells that are elliptic. These observations imply that genetic variation can affect cell-to-cell variability of cellular traits, without necessarily influencing the mean trait itself. This is particularly important when considering genetic susceptibility to common human or animal phenotypes. If genotypes affect the distribution of a phenotype among “identical” cells of a tissue, such genotypes are likely to modulate the phenotype penetrance. It is therefore tempting to propose a nondeterministic view of genetic predisposition, whereby incomplete penetrance does not only result from environmental exposures but also from levels of cell-to-cell heterogeneity.
We showed that the natural polymorphisms affecting yeast morphological traits do not preferentially reside in genes in which deletions affect these traits. This comparison of two alternative genomic approaches (systematic mutagenesis versus QTL mapping) was possible because both studies were performed on the same platform. We conclude that these two approaches are complementary for the functional study of genomes of model organisms. This is important since both approaches are widely used and heavily funded [25,26]. There are at least four possible explanations for this complementarity between the two approaches. First, unlike QTL mapping, the deletion approach does not interrogate essential genes. Second, deletion mutations may have dramatic phenotypic consequences as compared to natural polymorphisms. For example deletion of RAD50 (YNL250W) results in accumulation of large budded cells arrested or delayed in G2/M because of the failure to repair damaged DNA, but one can imagine that natural sequence variations in the gene might provoke more subtle alterations in the DNA repair system. Third, RM polymorphisms might provide gain-of-function alleles as compared to BY. In this case, the effects are likely to differ from the consequences of a null mutation. Finally, the QTL approach is far from exhaustive because it only interrogates genes containing functional polymorphisms between the backgrounds considered. We also note that in our case, although genome annotation was essential to characterize the effect of GPA1 alleles, data from systematic mutagenesis could not provide candidate QTL. The power of the candidate-gene approach for QTL mapping has been debated [27], and a previous study on yeast-sporulation efficiency illustrated how functional annotations of some genes poorly explained their QTL effect [2]. It is, therefore, essential to mix information from multiple sources to define candidates and to maintain efforts on genome scans that are free of hypotheses.
Strains used were BY4716 [28] and YEF1946, which is isogenic to RM11-1a [16]. BY-gpa1I469S [21] and BYxRM segregants [16] were kindly provided by L. Kruglyak together with their genotypes and transcriptome data.
Microscopic observation and data processing were essentially similar as previously described [15]. Briefly, cultures were grown to 1 × 107 cells/ml in synthetic C medium at 30 °C. Cells were fixed with 3.7% formaldehyde and stained with rhodamine-phalloidine, fluorescein isothiocyanate-concanavalin A, and DAPI. Cells were observed using an Axioplan 2 microscope with 100× Plan-neofluar objective lens (Carl Zeiss, http://www.zeiss.com). Digital images were acquired with CoolSNAP cooled-CCD camera (Roper Scientific, http://www.roperscientific.com) and MetaMorph Software (Molecular Devices, http://www.moleculardevices.com). Images were processed with CalMorph [15] to generate quantitative parameters (or traits) of yeast morphology. A minimum of 200 cells was analysed by culture. For natural variation of S. cerevisiae, nine cultures for either BY4716 or YEF1946 strain were used. For QTL mapping, three cultures for each segregant were used.
For every trait, its coefficient of variation across the 126 replicated cultures of the wild-type strain described in Ohya et al. [15] was computed. The 272 DNA-related traits had a mean coefficient of variation of 15% while the mean of all the others was of 29%, and the difference was significant (p = 0.015, Wilcoxon Mann-Whitney).
All statistical analyses were made using purpose-developed C codes or using R software (http://www.r-project.org). Heritability was measured as (VarS − VarE)/VarS, where VarS is the variance among segregants, and VarE is an estimate of the environmental variance calculated on the parental replicates. Since three independent phenotypic values were available for each segregant, the variance among segregants was computed three times on independent series. VarS was estimated by the average of these three variances.
For each of the 501 traits, transgressive segregation was tested as in [16], except that the procedure was applied three times, one for each independent series of segregant values. The statistic used was the number j of segregants showing a phenotype at least 2σ higher than the mean phenotype of the highest parent, or at least 2σ lower than the mean phenotype of the lowest parent, where σ was the pooled standard deviation of parental replicates. To infer significance, for each trait all parental and segregant values were pooled together and were reassigned to null parents and null segregants at random from this pool. The total number of such null traits with j greater than a given threshold j0 was the expected number of false positives, Nfalse(j0). FDR was the Nfalse(j0)/Nactual(j0) ratio, where Nactual(j0) was the number of real traits showing j > j0. The estimated total number of transgressive traits, Nactual(j0) − Nfalse(j0), was highest for j0 = 6 and was 157, 128, and 124 in the three replicated series of segregants data, respectively. This corresponded to one-fourth of the total number of traits. At j0 = 18, FDR was <0.05 and 34 traits were significant in all three series (Table S2).
Similarly, on each series of independent segregant data, we used the test described in Brem et al. [16] to assign p-values of epistasis to each trait. Highest Nactual − Nfalse values were 200, 207, and 221 traits in the three series, respectively, corresponding to two-fifths of the total number of traits. At p < 0.007, FDR was <0.05, and 98 traits were significant in all three series (Table S3).
For each trait with heritability greater than 50%, the genome was scanned for QTL as follows. Triplicates were averaged to give a single phenotypic value for each segregant. Marker–trait association was tested using the Wilcoxon Mann-Whitney test, and significance was assessed by permutation test as described previously [5]. At a given nominal p-value threshold, the FDR was computed as the ratio between the expected number of false positive counts (number of traits that could be mapped after permuting segregant index, averaged across 100 permutations) and the number of traits with detected linkage.
Since all tests were nonparametric, we worked on all 501 traits without distinguishing the set of 254 traits that were previously determined to fit normality (called “reliable” in Ohya et al. 2005) [15].
To test for association between expression levels (5,740 genes) and traits (247 traits with heritability >50%), Pearson correlation coefficients R were computed for each pair of messenger level and morphological trait. For a given R0 cutoff, the number of traits showing |R| > R0, with at least one expression level, was computed after permuting the segregant indexes. This number was averaged across 100 permutations. FDR was then determined as the ratio between this number of expected false positives and the number of traits correlated to expression levels prior to permutation. To mine for traits/GO-terms associations, we clustered the 103 genes and 104 traits correlated at FDR = 0.1 and visually examined the correlation map for regions enriched in genes/traits correlations (Figure 4). Gene lists of each of the four territories described in text were piped in the GO Term Finder (http://db.yeastgenome.org/cgi-bin/GO/goTermFinder) to infer significance of GO term enrichments.
We segmented the genome into 20-kb bins and assigned each of the 67 traits mapped at FDR = 0.05 to the bin showing highest genetic linkage (as in Table 2). Traits linking to both Chromosomes XIV and XV were assigned to both places by choosing the best bin on each chromosome. This way, 13 bins were linked to traits, with the number of traits per bin ranging from one to 21. For each bin, we considered all genes located within 20 kb of the center of the bin and asked whether their deletions affected one of the trait linked to the bin (at p < 0.0001 in the Ohya et al. 2005 dataset). This was the case for only one of the 13 bins. This search involved a high number (2,863) of gene/trait combinations, and finding one positive bin could therefore result from chance only. To test this, we reassigned the 13 bins to random places on the genome and re-examined it to determine if deletions of genes in their vicinity could explain one of their linked traits. We ran this test five times and obtained hits for zero, two, zero, three, and two bins at the respective runs. Thus, obtaining one positive bin in the actual data was not statistically significant. |
10.1371/journal.ppat.1006775 | Morphological switch to a resistant subpopulation in response to viral infection in the bloom-forming coccolithophore Emiliania huxleyi | Recognizing the life cycle of an organism is key to understanding its biology and ecological impact. Emiliania huxleyi is a cosmopolitan marine microalga, which displays a poorly understood biphasic sexual life cycle comprised of a calcified diploid phase and a morphologically distinct biflagellate haploid phase. Diploid cells (2N) form large-scale blooms in the oceans, which are routinely terminated by specific lytic viruses (EhV). In contrast, haploid cells (1N) are resistant to EhV. Further evidence indicates that 1N cells may be produced during viral infection. A shift in morphology, driven by meiosis, could therefore constitute a mechanism for E. huxleyi cells to escape from EhV during blooms. This process has been metaphorically coined the ‘Cheshire Cat’ (CC) strategy. We tested this model in two E. huxleyi strains using a detailed assessment of morphological and ploidy-level variations as well as expression of gene markers for meiosis and the flagellate phenotype. We showed that following the CC model, production of resistant cells was triggered during infection. This led to the rise of a new subpopulation of cells in the two strains that morphologically resembled haploid cells and were resistant to EhV. However, ploidy-level analyses indicated that the new resistant cells were diploid or aneuploid. Thus, the CC strategy in E. huxleyi appears to be a life-phase switch mechanism involving morphological remodeling that is decoupled from meiosis. Our results highlight the adaptive significance of morphological plasticity mediating complex host–virus interactions in marine phytoplankton.
| This study assesses the interplay between the globally distributed microalga Emiliania huxleyi and its specific lytic viruses, EhV, which drive the termination of vast oceanic blooms. E. huxleyi is characterized by a biphasic life cycle that alternates between morphologically dissimilar diploid and haploid cells. Here, we show that during viral infection, the bloom-forming diploid cells that are sensitive to EhV can produce virus-resistant cells. These latter cells are morphologically similar to the haploid phase but have diploid or aneuploid genomes. Therefore, a mechanism that mediates morphological remodeling appears to be activated during viral infection, enabling E. huxleyi to escape EhV. These results provide novel insights into morphological plasticity and viral resistance in marine phytoplankton, while highlighting the complexity of host–virus interactions in the oceanic microbial realm.
| The life cycle of an organism represents a multitude of cellular stages connected by reproductive processes. These complex chains of events have been selected over a long evolutionary history and represent a key feature underlying species ecology [1,2]. Thus, unraveling all cellular stages and the factors driving life-phase transitions will enhance our understanding of species' functional roles and their adaptive responses to environmental variations. With the exception of a few model organisms and human parasites, however, little is known about the life cycle of microbial eukaryotes, which paradoxically represent the vast majority of extant eukaryotic diversity [3]. This knowledge gap is exemplified by marine phytoplankton, which comprise a highly diverse assemblage of phototrophic species that make an important contribution to the base of the marine food web and global biogeochemical processes [2,4].
The coccolithophore Emiliania huxleyi (Lohmann) Hay and Mohler (Prymnesiophyceae) is a globally distributed marine microalga that forms large-scale blooms in high-latitude oceans with a significant ecological and biogeochemical impact [5,6]. E. huxleyi displays a dimorphic haplodiplontic life cycle [7,8]. Diploid cells are typically covered with calcareous scales (coccoliths) and dominate natural blooms [9]. However, some strains lack coccoliths after extended periods in culture and are denoted ‘naked’. Haploid cells are also devoided of coccoliths, but are biflagellate and their cell membrane is coated with thin organic scales. Therefore, 1N cells have been denoted scale-bearing swarmers or ‘S-cells’ [7]. Both 2N and 1N cells can grow independently by mitosis and likely interconnect through sex and meiosis, although sexual reproduction has never been observed in E. huxleyi.
The evolutionary stability of haplodiplontic life-cycle strategies is often interpreted as adaptation to fluctuating environments, where each life phase is better fit to different niches (Valero 1994, Hughes and Otto 1999). Here differences between niches is meant in a broad sense, and may include both abiotic (e.g. seasonal) and biotic (e.g. predators, pathogens) factors. In agreement with this view, it has been shown that while 2N E. huxleyi cells are sensitive to specific E. huxleyi viruses (EhV) that drive the termination of natural blooms [10–13], 1N cells are resistant to EhV [14]. Moreover, the same studies showed that biflagellate cells can emerge during viral infections, suggesting the occurrence of meiosis and the production of resistant cells in response to EhV. This process was metaphorically coined the ‘Cheshire Cat’ (CC) escape strategy, whereby a life-phase switch provides an escape mechanism from EhV. This could alleviate the viral pressure on host cells and potentially select for maintenance of a biphasic life-cycle strategy over evolutionary timescales [14]. This hypothesis has recently received support from exploration of the gene repertoire of E. huxleyi genotypes isolated from nutrient-rich areas, where blooms and EhV infections regularly seem to retain a biphasic sexual life cycle. In contrast, E. huxleyi genotypes isolated from low-productivity areas, where blooms do not develop and EhV are undetected, tend to lack the flagellar genes that are typically expressed in 1N cells. Arguably, these results suggest that in populations experiencing low viral pressure and low environmental variability, life cycling is not advantageous, and E. huxleyi cells may lose the ability to undergo sexual reproduction and produce 1N cells [15].
Although life-cycle transitions during viral infection play a pivotal role in E. huxleyi, the cellular mechanism underlying the CC strategy remains unclear. Here, we used morphological, ploidy and gene-expression analyses of meiosis- and life-phase-specific gene markers to test whether virus-resistant cells are produced during infection or are instead selected from a background subpopulation after elimination of the numerically dominant calcified diploid cells. In parallel, we also examined the fate of an E. huxleyi strain that seems to be unable to form biflagellate cells and that is plausibly impaired in CC capabilities. We further investigated whether life-phase transitions are induced by diffusible chemical cues (infochemicals) accumulated during infection. Collectively, our study provides novel insights into complex host–virus interactions and morphological differentiation in unicellular eukaryotes.
To investigate the molecular mechanisms underlying the CC strategy, we monitored the interplay between lytic virus EhV-201 [16] and two E. huxleyi strains: RCC 1216, a 2N calcified strain able to undergo sexual transitions and form biflagellate 1N cells [8]; and CCMP 2090, a 2N noncalcified strain, lacking essential flagellar genes and for which the production of 1N cells has never been recorded [15]. In the presence of EhV, both strains lysed to nearly undetectable levels. However, within variable time frames, a minor subpopulation of cells emerged and resumed growth in the presence of high EhV densities (Fig 1). During viral infection of RCC 1216 (Fig 1A–1I), there was a transient rise in noncalcified cells (low side-scattering subpopulation by flow cytometry) between 2 and 4 days postinfection (dpi). These cells comprised up to 35% of the total E. huxleyi population. Since at this stage, virtually all E. huxleyi cells were positive for the cell-death marker SYTOX-Green, this noncalcified population was essentially composed of dying cells which shed their coccoliths due to EhV infection (S1A Fig). However, at 7 dpi, we detected by light microscopy the presence of motile noncalcified biflagellate cells, either individually or in small motile groups of 3–6 cells. At this stage, we estimated that the motile fraction of cells represented ~0.05% of the maximal cell abundance at 2 dpi. Subsequent electron microscopy analyses revealed that the motile cells have thin organic scales with radiating patterns of fibrils (Fig 1G–1I), as is typical for E. huxleyi 1N cells [7] (Fig 1F). In the parallel assessement of strain CCMP 2090, we also detected the recovery of a new subpopulation, but it evolved over longer time scales of ~35 dpi as compared to 7 dpi in RCC 1216 (Fig 1J). The new emerging cells lacked flagella but had thin organic scales like 1N cells, as detected by electron microscopy (Fig 1P–1R). We termed these cells derived from CCMP 2090 nonmotile scaled cells (nonmotile-S cells).
To assess whether the formation of biflagellate and nonmotile-S cells was driven by meiosis, we used qRT-PCR to monitor the expression of a core set of meiosis-associated genes (i.e., two SPO11 variant genes, DMC1, HOP1, MER3, MND1, MSH5 [17,18], see S1 Table and S2 Fig) together with a set of genes reported to be specific to 1N cells [15,19]. The latter, herein globally termed S-cell genes (S1 Table), included four flagellum-associated genes (FLAG 4, 5, 8 and 11), two phototropins (PHOTO1, PHOTO2), one MYB transcription factor and one histone H2A. RCC 1216 was examined at high temporal resolution to provide a comparative basis to our previous observations [14].
Coordinated upregulation of all S-cell gene markers (102- to 104-fold) was detected in both E. huxleyi strains over the course of EhV infection (Fig 2, S2 Table). Detailed analyses of RCC 1216 revealed upregulation of FLAG11 and PHOTO1 within 24 h of infection, followed by the upregulation of the remaining S-cell genes at 2 dpi. All S-cell gene markers remained above control levels until the end along with biflagellate cells growth. The expression levels of meiotic markers was lower than that of S-cell markers (often below noninfected control cultures) and markedly irregular over time (up/downregulation). The only clear exceptions were HOP1 and MER3 that showed a nearly 2-fold increase in expression at 2 dpi relative to control cells, concomitant with cell-growth arrest and onset of the lytic phase. The same general trend was also detected for CCMP 2090, where S-cell genes were markedly upregulated during infection, whereas meiotic genes showed lower or no variability relative to control cells. We note that FLAG5 and PHOTO1 genes were not detected in CCMP 2090, likely due to their absence from the genome of this strain [15].
In addition, we used the same set of S-cell and meiosis markers to assess whether the production of biflagellate cells in RCC 1216 could be triggered in response to diffusible signals (infochemicals) produced during infection (S3A Fig). Therefore, diploid RCC 1216 cultures were exposed at 50% vol/vol to virus-free lysates (VFL), a conditioning medium derived from infected cultures. VFL was harvested at 4 h, 24 h, 48 h and 72 h postinfection. We also used UV-inactivated EhV virions (virus-to-host ratio = 5) to examine the cellular response to potential virus-borne elicitors. However, gene-expression analyses did not reveal any noticeable gene upregulation after 4 h or 24 h of exposure to VFL as compared to typical EhV infections under any of these conditions (S3B Fig). Furthermore, the emergence of biflagellate cells could not be detected by light microscopy during the weeks following each treatment.
To further assess whether meiosis is occurring during infection, we examined the variations in relative genome size (RGS) of infected cells as compared to control cells (Fig 3). This was done by measuring the nuclear DNA content of cells by flow cytometry (S4 Fig) relative to the 1N strain RCC 1217 that was derived from RCC 1216 [8]. RCC 1216 cells possessed an average RGS of ~1.8xN, similar to previous reports [8], whereas diploid CCMP 2090 cells possessed an average RGS of ~1.3xN. In both RCC 1216 and CCMP 2090, the nuclear DNA content remained nearly invariable during the first 4 dpi (Fig 3A and 3C). However, RGS levels during cell recovery revealed higher values than the original parental diploid cells. Biflagellate cell populations displayed ~2xN RGS levels, representing an average ~10% increase relative to RCC 1216 (Fig 3A), whereas nonmotile-S cell populations displayed ~1.9xN RGS levels, representing an average ~60% increase relative to CCMP 2090 (Fig 3C).
To validate these results, we isolated representative single biflagellate and nonmotile-S cells. RGS levels of five independent biflagellate clones (LC4A, LC4F, LC4G, LC4I, LC4J) were consistently 2xN (Fig 3B), as detected for the total recovering populations. In contrast, the four nonmotile-S cell clones analyzed displayed variable RGS levels ranging from 1.8xN to 2.2xN, representing increments of ~40% to ~70% relative to CCMP 2090 (Fig 3D). Furthermore, both biflagellate and nonmotile-S cells were invariably larger (spherical cell volume) than the parental cell lines, by >50% and >25%, respectively (Fig 3B and 3D). Confocal microscopy observations did not reveal any irregular structural changes in the nuclei and cells were characterized by single nuclei, like the parental cell lines (S5 Fig).
In a complementary approach, we used the microsatellite marker P02F11 [20] to assess the ploidy level of resistant cells. This analysis revealed that all of the biflagellate clones are heterozygous, displaying doubled allele bands for two loci like the 2N RCC 1216 cells, and in contrast to the single bands detected in 1N RCC 1217 cells (S6 Fig). This indicates that the biflagellate cells have two chromosomal copies, as expected in 2N organisms.
To provide a functional characterization of biflagellate and nonmotile-S cells, we conducted growth assays of all isolated clones and examined their susceptibility to viral infection, in comparison to the parental strain RCC 1216, RCC 1217 and CCMP 2090 (Fig 4). Both biflagellate and nonmotile-S cells exhibited significantly lower growth rates and carrying capacities (average cell density during stationary phase) than each parental cell line. Fitness reduction in the five biflagellate cell lines was diagnosed by an average ~15% decline in growth rate (t-test, P < 0.05) and ~55% decline in carrying capacity (t-test, P < 0.01) relative to RCC 1216 (Fig 4B). In nonmotile-S cells, fitness decline was more severe, with growth rates declining ~30% (t-test, P < 0.01) and carrying capacities declining ~75% (t-test, P < 0.01) relative to CCMP 2090 (Fig 4E). Importantly, both biflagellate and nonmotile-S clones were resistant to EhV infection, i.e., their growth in the presence of EhV was similar to that under control conditions and the production of new viral particles was not detectable by flow cytometry (Fig 4C and 4F).
The CC strategy originally described E. huxleyi's ability to escape from EhV by alternating from a 2N virus-sensitive phase to a morphologically distinct 1N virus-resilient phase [14]. Here, we reassessed the CC strategy to further understand the mechanisms mediating the life-cycle shift in E. huxleyi in response to EhV. For RCC 1216 (2N), we detected the emergence of biflagellate cells that were resistant to viral infection following the lysis of calcified 2N cells (Fig 1), recapitulating earlier observations [14]. In contrast, the response of 2N CCMP 2090 cells to EhV was unprecedented. These cells, which are unable to form motile cells and may lack the ability to undergo sexual transitions [15], produced nonmotile-S cells during viral infection that bore organic scales similar to 1N cells. This result indicated that E. huxleyi cells lacking flagellar genes [15] can still undergo life-phase transitions and may retain a sexual life cycle. Additional laboratory and field work, possibly making use of population genomic approaches [21], is required to further understand the complexity of E. huxleyi's life cycle in its natural habitats, including in oligotrophic systems where cells displaying genomic erosion of flagellar genes seem to predominate.
Originally, the CC model stated that virus-resistant cells (haploid) are produced through meiosis in response to EhV infection [9,14]. However, there is some doubt as to whether instead of a sexual transition, the CC strategy might involve a selection process, where low background levels of 1N cells take over after lysis of the virus-susceptible 2N cells. Our detection of marked and rapid overexpression of S-cell genes during infection (Fig 2) provides molecular support for the original view that a life-phase transition is triggered during infection. However, based on our population-level analyses, the fraction of the population undergoing this life-phase transition remains unclear. Further single-cell approaches are required to quantify this rare subpopulation during infection [22]. Moreover, we did not detect any phenotypic response to diffusible cues accumulated during infection, suggesting that a direct host–virus interaction may be required to trigger the production of resistant cells (S3 Fig).
Although a life-phase transition seemed to be triggered during infection in the two tested E. huxleyi strains, the increment in RGS levels in cells recovered after infection indicated that meiosis is probably not involved in the process. In the case of biflagellate cells, both RGS measurements and microsatellite analyses indicated that these cells are 2N (Fig 3, S6 Fig). However, we did note a mismatch in RGS between biflagellate cells and the parental cell line RCC 1216, the first being 2xN and the second 1.8xN. The nature of this ~10% discrepancy requires clarification through further cytogenetic and genomic analyses but it might have resulted from a bias in our ploidy-level analyses by flow cytometry as a result of differential condensation levels of the DNA and staining efficiencies in the two cell types, as has been documented in other systems [23]. This being the case, the biflagellate cells are regular 2N cells, produced via a phenotypic-switch mechanism that is independent of the sexual cycle. Such decoupling between life-phase phenotype and ploidy level resembles apomictic life cycles observed in haplodiplontic plants and algae [24,25], which can be triggered under stress conditions [26]. Apomixis can involve either apospory with the formation of a 2N gametophyte (typically the 1N phenotype) without meiosis, or apogamy with the formation of a 1N sporophyte (typically the 2N phenotype) without syngamy. Some evidence suggests that these types of processes can also occur in other noncalcified prymnesiophytes related to E. huxleyi [27–29]. Thus, it is plausible that phenotype remodeling through apospory, putatively mediated through genetic or epigenetic mechanisms [30–32], is at the basis of the formation of biflagellate diploid cells in response to EhV-mediated stress.
In the case of nonmotile-S cells, the explanation is less straightforward because these cells exhibit a variable range of aneuploid genomes considerably larger than CCMP 2090 (Fig 3). Thus, it is possible that major genomic rearrangements, including the duplication of chromosomal parts and disruption of other sections, led to differential regulation of gene expression and subsequent production of modified phenotypes. This could have occurred during viral infection [33,34] or be a host specific response to EhV. Aneuploidization involving chromosomal duplication (or partial duplication) has well-documented roles in adaptation conferring for example fitness advantages under a variety of abiotic (e.g. temperature, nutrients) and biotic stress conditions [35–37], including resistance to viruses as recently reported in a marine picoeukaryote [38]. However, given that nonmotile-S cells also exhibited morphological phenotypes resembling haploid cells (i.e., presence of organic scales), we argue that a similar aposporic mechanism was also involved in the production of these cells in response to EhV.
Given that meiosis did not appear to underlie the production of biflagellate and nonmotile-S cells, the role of meiotic gene transiently detected during infection (Fig 2) is unclear. Meiosis is a necessary part of sexual reproduction and a core set of genes involved in DNA double-strand break formation and crossover regulation seem to be conserved and to be fairly specific across eukaryotic lineages [17,18,39]. However, it has been shown that some meiosis-related genes can also play a role in other recombination mechanisms [39–41]. During EhV infection, while host cells undergo major cellular and metabolic reprogramming, pathways related to DNA repair are upregulated [42,43]. Moreover, we could also detect the expression of meiotic genes under control conditions, implying an alternate role for these genes in other cellular processes that could have been further enhanced during infection.
Cell-growth assessment of recovered biflagellate and nonmotile-S clones confirmed a stable resistance phenotype to EhV-201 (Fig 4). During a parallel test, we found that both cell types were also resistance to diverse EhV strains (EhV-86, EhV-163, EhV-ice 01 [16,44]), which suggests the existence of a generic, albeit unknown resistance mechanism against EhV that is common to all cells expressing a phenotype resembling the haploid cell. Further analyses of the processes of EhV adsorption onto host cells and the role of the organic scales or cell-surface properties of resistant cell lines may provide new insights into the mechanisms of viral resistance.
Concomitantly, both biflagellate and nonmotile-S cells showed compromised growth fitness as compared to the parental cell lines under control conditions, i.e., decreased growth rate and carrying capacity (Fig 4). Tradeoff costs are often detected in bacterial and eukaryotic cells after the acquisition of resistance to viruses [45–50]. Here, it is possible that a fitness decline resulted from the increased cell volumes and consequent decline in cell surface-to-volume ratio, which often leads to decreasing nutrient-uptake rates and growth [51]. In nonmotile-S cells that showed higher levels of fitness decline, a decrease in nutrient-uptake rates may have been associated with additional physiological costs for DNA biosynthesis or other genomic destabilization following aneuploidization [35,52].
Recent analyses of control 1N RCC 1217 cells detected minute amounts of viral glycosphingolipids and EhV transcripts, suggesting a possible mode of persistence of EhV within haploid cells [53,54]. To determine whether this might also be the case in biflagellate and nonmotile-S strains, we extracted both DNA and RNA from various clones and screened for EhV-specific gene markers. However, all of the results were negative, indicating that none of the resistant strains carry any form of EhV.
The suite of isolated biflagellate and nonmotile-S cell cultures will provide a powerful tool for future cellular and comparative omics analyses to dissect the cellular mechanism enabling morphological remodeling and viral resistance in E. huxleyi.
Our results provide novel evidence for a CC model in which E. huxleyi cells' ability to escape viral attack through life-phase change is decoupled from the sexual cycle; this stands in contrast to the original CC scheme [14] (Fig 5). This process seems to be triggered in a small fraction of cells during infection by EhV and to enable the production of diploid (or aneuploid) cells that display phenotypes resembling haploid cells and that are resistant to EhV. The morphological and genome-size properties of both biflagellate and nonmotile-S cells have been stable in culture for the last 1.5 years of isolation. However, it is plausible that these cells constitute an intermediary state produced under stressful conditions and that they are capable of reverting back to the original calcified nonmotile state, or instead undergoing meiosis to resume life-cycle progression (Fig 5). This extended capability of decoupling phenotype from ploidy level may improve the adaptability of these microbial cells to the highly fluctuating stressful conditions at sea and enhance survival rates during the interplay with EhV during bloom events.
Replicate cultures of E. huxleyi RCC1216 (calcifying, diploid; previously referred to as strain TQ26 [8]) and CCMP 2090 (noncalcified, diploid; equivalent to strain CCMP 1516 for which genomic information is available [55]) were grown in seawater-based K/2 medium [56] and infected with the lytic viral strain EhV-201 [16] at a virus-to-host ratio of 0.2 (initial 105 cell mL-1). Noninfected cultures were used as controls. The haploid E. huxleyi strains RCC 1217 (isolated from RCC 1216 after sporadic diploid-to-haploid transitions, [8]) was grown under identical conditions and used for comparative assays.
Cells and EhV were enumerated by flow cytometry (Eclipse, iCyt equipped with 488-nm solid-state air-cooled laser and standard filter setup) [57]. Algal cells were differentiated based on chlorophyll autofluorescence and side-scatter signatures, enabling the segregation of calcified from noncalcified cells (higher and lower side scatter, respectively) [14]. Algal cell death was determined with 1 μM SYTOX Green (Invitrogen) by flow cytometry [58]. The diameter of the cells was assessed with a Multisizer 4 Coulter counter (Beckman Coulter). Relative average nuclear DNA content (RGS) was monitored by flow cytometry (LSR, BD Biosciences) using extracted cell nuclei labeled with the fluorochrome SYBR Green [8] and the 1N RCC 1217 as an internal standard for data normalization.
Light microscopy was performed using a differential-phase contrast setup at x100 magnification (Olympus, Japan). Electron transmission microscopy preparation was performed as described in Schatz and Shemi [59]. E. huxleyi cell nuclei and chloroplasts were observed with a confocal microscope (Eclipse Ti-E Inverted microscope, Nikon, Japan) using cells fixed with 1% formaldehyde for 2 h at 4°C and stained with 5 μg mL-1 SYTO13 Green (Molecular Probes) for 10 min.
Genomic DNA was extracted from cell pellets (~106 cells) from RCC 1216, RCC 1217 and all biflagellate clones isolated in this study using a standard phenol–chloroform extraction method. The microsatellite marker P02F11 was amplified by PCR [20] and the products separated by electrophoresis using a Criterion TGX Any kD precast gels with Tris-borate buffer at 50 V for ~3.5 h. The size of the amplified products was determined using a standardized 100 bp DNA ladder (Promega).
Meiotic genes (S1 Table) were manually defined and aligned as described in Feldmesser et al. [60]. Briefly, the choice of target genes was based on a list of core meiosis genes from other protists (e.g. [17,18]) and Arabidopsis previously published and deposited at the National Center for Biotechnology Information (NCBI) and the Joint Genome Institute (JGI). All hits were analyzed for transcript evidence (ESTs) or gene models. A gene model was then built manually based on the existing transcripts, models, and BLAST results. Whenever transcriptome information was available [42] it was used to improve the manual gene models. After the definition and translation of E. huxleyi genes, the protein sequence was used for searches against the protein collections of NCBI. Multiple sequence alignments and phylogenetic analyses were performed using maximum likelihood with Mega (version 7.0.16) (S2 Fig). The manually curated genes from E. huxleyi were deposited in GenBank (KY224381–KY224389) and are detailed in S1 Table. S-cell genes were all derived from previous studies as listed in S1 Table.
250-mL cultures collected by centrifugation (8000g, 4°C, 10 min) at each time point. RNA was isolated with the RNeasy Plant Mini Kit (Qiagen) according to the manufacturer’s instructions. Following DNAse treatment (Turbo DNAse, Ambion), the RNA was reverse-transcribed to cDNA with the ThermoScript RT-PCR system (Invitrogen). Transcript abundance was determined with the Platinum SYBR Green qPCR SuperMix-UDG with ROX (Invitrogen). Primers are listed in S1 Table. All of the reactions were performed on StepOnePlus real-time PCR Systems (Applied Biosystems) as follows: 50°C for 2 min, 95°C for 2 min, 40 cycles of 95°C for 15 s, 60°C for 30 s. Relative gene expression of each gene was calculated using the 2-ΔΔCt method [61] against control uninfected samples per time point.
To test whether the RCC 1216 and RCC 1217 strains and the new biflagellate clonal strains contained intracellular EhV, we subjected 300 ng of host DNA or RNA to qPCR for several viral genes: viral major capsid protein [16] and viral serine palmitoyl transferase [22].
The effect of chemical signals that might trigger the production of virus-immune cells was tested using conditioned medium derived from infection. Briefly, E. huxleyi RCC 1216 cultures (10 L) were infected with EhV (virus-to-host ratio of 0.2), and 200-mL subsamples were collected at 4 h, 24 h, 48 h and 72 h postinfection and sequentially filtered through a 0.45-μm filter and 300 kDa tangential-flow filtration system (PALL) to remove cells and EhV particles. The filtrate (VFL, see scheme in S3 Fig) from each time point was then added (50:50 vol/vol) to exponentially growing RCC 1216 cultures at 5 x 105 cell mL-1. Controls were diluted in the same proportion with fresh K/2 medium. Samples (200 mL) for microscopy and qRT-PCR analyses were collected at 4 h and 24 h after exposure to VFL. In addition, we tested the effect of EhV-derived components. A 50-fold concentrate of purified EhV virions [59] was exposed to 4000 μJ of UV light (UV Stratagene) to inhibit viral activity and added to cultures at a virus-to-host ratio of 5. EhV particles not exposed to UV were used as a positive control. Samples (200 mL) for microscopy and qRT-PCR analyses were collected at 4 h, 24 h and 48 h. In all of the assays, cells and viral enumeration as well as gene-expression analyses were assessed according to the procedures described above.
|
10.1371/journal.pcbi.0030038 | Spatial Acuity and Prey Detection in Weakly Electric Fish | It is well-known that weakly electric fish can exhibit extreme temporal acuity at the behavioral level, discriminating time intervals in the submicrosecond range. However, relatively little is known about the spatial acuity of the electrosense. Here we use a recently developed model of the electric field generated by Apteronotus leptorhynchus to study spatial acuity and small signal extraction. We show that the quality of sensory information available on the lateral body surface is highest for objects close to the fish's midbody, suggesting that spatial acuity should be highest at this location. Overall, however, this information is relatively blurry and the electrosense exhibits relatively poor acuity. Despite this apparent limitation, weakly electric fish are able to extract the minute signals generated by small prey, even in the presence of large background signals. In fact, we show that the fish's poor spatial acuity may actually enhance prey detection under some conditions. This occurs because the electric image produced by a spatially dense background is relatively “blurred” or spatially uniform. Hence, the small spatially localized prey signal “pops out” when fish motion is simulated. This shows explicitly how the back-and-forth swimming, characteristic of these fish, can be used to generate motion cues that, as in other animals, assist in the extraction of sensory information when signal-to-noise ratios are low. Our study also reveals the importance of the structure of complex electrosensory backgrounds. Whereas large-object spacing is favorable for discriminating the individual elements of a scene, small spacing can increase the fish's ability to resolve a single target object against this background.
| Extracting and characterizing small signals in a noisy background is a universal problem in sensory processing. In human audition, this is referred to as the cocktail party problem. Weakly electric knifefish face a similar difficulty. Objects in their environment produce distortions in a self-generated electric field that are used for navigation and prey capture in the dark. While we know prey signals are small (microvolt range), and other environmental signals can be many times larger, we know very little about prey detection in a natural electrosensory landscape. To better understand this problem, we present an analysis of small object discrimination and detection using a recently developed model of the fish's electric field. We show that the electric sense is extremely blurry: two prey must be about five diameters apart to produce distinct signals. But this blurriness can be an asset when prey must be detected in a background of large distracters. We show that the commonly observed “knife-like” scanning behaviour of these fish causes a prey signal to “pop-out” from the blurry background signal. Our study is the first to our knowledge to describe specific motion-generated electrosensory cues, and it provides a novel example of how self-motion can be used to enhance sensory processing.
| Weakly electric fish are commonly found in the freshwater systems of South America and Africa [1,2]. These nocturnal fish use a unique sensory modality, called the “electrosense,” to help them navigate, communicate, and find prey in the absence of strong visual cues [3]. The electrosense involves a specialized electric organ that emits an electric discharge resulting in a dipole-like electric field in the surrounding water [4]. The transdermal potential (the so-called “electric image”) is continuously monitored via electroreceptors found in the skin layer. Changes in the spatial properties of the electric image can provide cues that help the fish determine the location, size, and electrical properties of nearby objects [5–10].
Recent studies have shed new light on the weakly electric fish's perceptual world. In the context of distance perception, the amplitude and width of an electric image were shown to be analogous to visual contrast and blur [11]. The electric image produced by an object can also be distorted by nearby objects; consequently, conductive objects can act as electrosensory “mirrors” [12]. In contrast with the visual sense, however, the electrosense has no focusing mechanism and is limited to the near-field, so it is generally considered a “rough” sensory modality [13–16]. In fact, the range of active electrolocation in weakly electric fish is likely only about one body length [7], and considerably less for small prey-like objects [17]. Within this range, much is known about the fish's temporal acuity [18,19], but relatively little is known about the fish's ability to resolve multiple nearby objects.
Here, we consider the notion of “electro-acuity,” analogous to the notion of visual acuity found in the visuo–sensory lexicon, to investigate the quality of electrosensory information in the spatial domain. A common measure of acuity in other sensory systems is the just-noticeable difference, or the minimum difference between two stimuli such that they are perceptually distinct [20]. In the present context, we consider an analogous measure to describe the quality of electrosensory input available for a discrimination task. We define this measure as the minimum spatial separation of two objects (Smin), such that two distinct peaks remain in the electric image on the fish's skin (Figure 1). Using a 2-D finite element method model of A. leptorhynchus' electric field [9], we show that Smin is smallest in the fish's midbody and decreases for objects placed farther away from the fish. This suggests an interesting contrast with the “electrosensory fovea” in the head region [10,17], where the highest density of electroreceptors is found [21]. Overall, we found that electroacuity is poor relative to visual acuity in humans, but is comparable with that of the human somatosensory system.
Despite the apparent low quality of electrosensory signals, weakly electric fish are able to detect small prey [7,17]. Although there is no direct evidence, it is reasonable to assume that they do so even in the presence of noisy background signals [7]. In a related task (object tracking), background noise has been shown to degrade performance [22,23]. Single-cell recordings in midbrain neurons have further revealed that some low-frequency background signals can interfere with directional selectivity [24]. It is thus believed that some of the natural behaviors exhibited by the fish play a central role in signal extraction. In particular, simulations have suggested that tail-bending could improve object detection by increasing the electric image's amplitude [13,14].
It has also been suggested that the back-and-forth swimming, or scanning motion, observed in these fish could be used to generate specific electrolocation cues [25–28], although this has not yet been demonstrated. Indeed, to elucidate the nature of these motion-related cues, we have simulated this scanning motion and show that, under some conditions, this behavior could assist in extracting small prey-like signals from large background ones. We show that the component of the electric image produced by a sufficiently dense background does not change during scanning, whereas the one produced by the prey object, albeit miniscule in comparison, does. This process is similar to motion-related cues and active sensing techniques seen in other contexts [28,29].
In the following analyses, we use our previously described finite-element model of the electric field produced by A. leptorhynchus (see Materials and Methods and [9,30]). Figure 1A shows the simulated dipole-like potential map for this fish in the presence of two prey-like objects. Such objects do not greatly perturb the fish's natural field due to their small size and conductivity (which is similar to that of the water). Figure 1B shows overlays of electric images due to single objects at different locations (i.e., each image is computed separately). Such images show characteristic shapes but vary systematically in amplitude and width with rostral–caudal and lateral location [5,9,10]. Figure 1C shows images produced by object pairs for three different interobject distances (shown in inset). Prey-like objects that are located too close together (green trace) produce a single peak in the electric image (similar to the images in Figure 1B), while objects separated by a larger distance produce two distinct peaks (red trace). The blue trace illustrates the electric image in which two peaks are just barely distinguishable; we define the associated interobject distance as Smin. Thus, Smin, measured in these noiseless conditions, delineates a limit to electroacuity. A smaller Smin suggests better electroacuity (i.e., increased spatial resolution). For this specific prey-like object and rostro–caudal location, the Smin is 14 mm. This suggests that, at this lateral distance, these two objects must be separated by at least 14 mm, a distance approximately five times their diameter, to be distinguished.
Electroacuity varies for different lateral and rostro–caudal object locations (Figure 2, see insets). Figure 2A and 2C shows the effects of object size and conductivity, respectively, on electroacuity for different lateral positions (rostro–caudal position fixed near the fish's midpoint, 0.11 m). Smin increases (electroacuity decreases) for objects that are placed farther away from the fish, regardless of object size or conductivity. When objects are far from the fish, Smin is roughly independent of object size (Figure 2A). At the closest location possible for the largest object (blue curve), Smin is smaller than for the other object sizes. This is a consequence of the relative sharpening of the image for close large objects (see Figure 1B). The sharpness of an image can be quantified by the reciprocal of its normalized width (width divided by amplitude). Image sharpness decreases (normalized width increases) with lateral distance and, in general, is independent of object size [5]. However, object size becomes a factor for locations close to the skin (see largest object in Figure 2A and 2B), as larger objects produce relatively sharper images in these cases [9]. Note also that there is a slight inflection at a lateral distance of 0.016 m (Figure 2A and 2C) due to the spatial heterogeneity of the electric field (higher density of field lines near the zero potential line, which curves rostrally as seen in Figure 1A).
Figure 2B and 2D shows the effects of object size and conductivity, respectively, on electroacuity for different rostro–caudal positions (lateral object center-to-skin distance fixed at 0.012 m). In general, Smin is smaller for larger objects, all along the length of the fish. The largest objects (2 cm) can actually be distinguished in the artificial condition of overlapping (i.e., the two objects are fused into a single composite peanut-shaped object), suggesting a mechanism for shape discrimination under some conditions. The position x = 0.11 m suggests a point of optimal acuity along the side of the fish. The two peaks in the image can be distinguished more easily for objects in this region because this is the rostro–caudal location where electric images are sharpest [9,10], so that there is minimal interaction between the individual images produced by each object. Object conductivity has comparatively little effect on the Smin in both lateral and rostro–caudal directions (Figure 2C and 2D). Overall, Smin varies much more in the lateral direction than in the rostro–caudal direction (compare Figure 2A–2C and 2B–2D) due to the relatively large changes in image sharpness as lateral object distance increases [5,8].
The effect of water conductivity on electroacuity was also studied for a specific location (x = 0.11 m, y = 0.015 m). For the range of water conductivity values found in the rivers in which A. leptorhynchus live (between 0.00085 and 0.01135 S/m [2]), Smin changes only slightly. As an overall trend, Smin decreased as water conductivity diminished (from 15.5 to 12.5 mm as water conductivity decreased from 0.05 to 0.0005 S/m).
As a first step toward understanding electroacuity in a more natural context, the electric images produced by differently sized arrays of background objects (with “plant-like” conductivity) were studied systematically. In Figure 3A, the red trace shows the electric image produced by a single such object located 0.11 m caudally from the tip of the fish's head (red object in inset located close to the fish's midpoint). The orange trace shows the electric image produced by three objects: the central one (red) plus one (orange) added 0.03 m on each side. In a similar progression, electric images are shown for up to 11 objects. With larger numbers of aligned objects, the electric images converge. Thus, for an array of seven objects (approximately a fish body length), the image is almost the same as with 11 objects. The electric images are each marked by a singular peak because the interobject distance is too small (at this lateral distance of 0.05 m) to resolve different peaks, i.e., object separation is less than Smin. The small bumps at approximately 0.03 m and 0.2 m are due to abrupt changes in fish geometry near the head and tail, respectively, and are not due to individual objects within the background array. Similar results were also observed for object arrays positioned closer to the fish, where different peaks were observed in the electric image, as well as for solid bars of increasing widths (unpublished data). Figure 3B shows the effect of changing the object spacing in similar arrays. At the largest spacing (red), the image is dominated by the contribution from the central object. For arrays that are more spatially dense (green, blue), the contributions of individual objects are blurred, resulting in an image with a broad peak.
These object arrays provide a simplified model of the background signals comprising a natural electrosensory landscape. To better understand how weakly electric fish are able to detect miniscule prey in the presence of large-background signals, we calculated the electric image produced by a small Daphnia-like prey object against a large-background array of objects (Figure 4). Even though the prey is located just 0.008 m from the fish's skin (compared with the 0.05 m lateral position of the background), the electric image with the prey and background is not much different than the one obtained with the background alone (largest deviation between the two images is about 4%; compare Figure 4A and 4B). The interesting feature, however, is that the overall image shape is similar regardless of the fish's position during a simulated scanning movement (even though the background was simulated as a discrete set of objects). This can be understood in terms of electroacuity: the background objects are too close together to be distinguished and thus form a blurred image. It is critical to note that during the scan, however, the small blip created by the prey does change location within the electric image (Figure 4B; note that the images do not overlap perfectly). Next, we demonstrate this point explicitly by considering the time-varying image during a simulated scanning movement.
The consequence of the relative differences between background and prey during a scanning movement is that the small prey signals can be extracted by looking at the time-varying transdermal potential at specific locations along the fish's body. Figure 5 illustrates the temporal profile of the transdermal potential at two distinct body locations under different conditions. The signal measured at Location A (see inset) reveals a clear prey-dependent component (Figure 5A, compare green and blue traces). Note also that this prey signal (in the presence of the background) is very similar to that for the prey-alone condition (Figure 5A, compare blue and red traces). When the interobject distance in the background becomes too large, as in Figure 5B, the background image is no longer blurred and individual object characteristics appear, thereby masking the prey-specific signal. This effect can be even more pronounced when the objects are randomly spaced over the same area (Figure 5C). Figure 5D–5F shows a similar result for a different body location (note that the prey-specific signal occurs slightly later in time at this location, due to the scanning direction).
Figure 5A and 5B suggests that as the objects within the background are increasingly separated, the prey will be less distinguishable. We confirm these observations in terms of a signal-to-noise ratio (SNR) of prey signal versus background (see Materials and Methods). The SNR decreases with increasing interobject separation in the background (Figure 6; left axis, blue trace). For reference, we can compare this situation with the expected discriminability of two individual objects (see Materials and Methods), where the electric image components due to each object become increasingly distinct as the objects are moved apart (Figure 1B; Figure 5C: right axis, green trace). This applies to the case of two prey-like objects in the absence of background, as in Figure 1A and 1C and Figure 2, as well as to the case of two background-like objects. In a more natural context, the blurriness of the electrosense interestingly has the effect of enhancing sensory performance. And indeed, this should apply to a wide range of electrosensory landscapes, as blurriness will be unaffected by small changes in object conductivity (Figure 2C and 2D).
The extraction of small environmental signals is a problem faced by all sensory systems. The mechanisms by which this problem is solved have been studied extensively, not only in the human senses, but also in sensory modalities unique to other species [28,31]. Indeed, the electrosensory system exhibits many parallels with other senses, including human vision and audition [11,32], but we know relatively little about small-signal extraction and the spatial resolution of this modality. Here, we have considered these aspects of electrosensory processing in terms of the primary sensory input as a first step toward understanding acuity and object detection at the behavioral level.
Many recent studies have contributed to our understanding of electrosensory scene analysis [9,26,27,33,34]. In particular, Rother et al. [12] have shown that the electric image due to two objects is the result of complex interactions between the effects of each object. To extend these studies in the context of object discrimination, we have introduced the notion of electroacuity. This measure, comparable to the notion of visual acuity, has helped us quantify the “sharpness” of the electrosense in the spatial domain. Studies have suggested that this was a rather “rough” sensory modality [7,14], and our findings, in terms of the sensory input, confirm this quantitatively. For example, we found that two prey-like objects located within the range of natural prey detection (which is typically less than 20 mm, [17]), must be separated by 9 mm for the electric image to show features of both objects (Figure 2). We characterize this limit by the quantity Smin, analogous to the psychophysical notion of the just noticeable difference and the Rayleigh criterion in optics (see Materials and Methods). Electroacuity is much lower than human visual acuity [35]. In contrast, the electrosense fares much better when compared with tactile two-point discrimination in humans, where thresholds are as high as 50 mm in some body locations [36,37].
The magnitude of Smin will increase with the disparity in both the image amplitudes and widths for the two objects. It will also be influenced by nonlinear effects between image amplitude and image width for close pairs of objects (which our simulations implicitly capture), but we have not systematically investigated them here (but see [12]). That said, to a reasonable approximation, Smin is proportional to the normalized width of the image due to each of the objects (see Materials and Methods).
Figure 2B shows that for locations in the rostral half of the fish, Smin changes relatively little. This interesting feature is primarily due to the uniformity of the field in this range: the current lines are nearly perpendicular to the fish body axis. The field uniformity is a result of the spatial filtering effects (smoothing) due to the tapered body shape [9,10,38]. This means that the spatial extent of an object's influence on this field (image sharpness) will be relatively constant. For locations closer to the midbody, the field lines are more concentrated (i.e., the field is not as uniform as for more rostral locations), so the influence of the object is more focused. The image amplitude also increases in this range of body locations (Figure 1B; Figure 5 of [9]), further contributing to a sharper image. However, as outlined in detail in Materials and Methods, although the image amplitude increases, then decreases, in the rostro-to-caudal direction [9], Smin is determined by image sharpness (normalized image width) and is much less sensitive to absolute amplitude (Figure 2B, compare red and green traces).
In terms of the quality of sensory input, our results reveal a point of optimal electroacuity located in the fish's midbody. This is in contrast to the notion that optimal discrimination should occur near the fish's head region, the electrosensory fovea, which has the highest density of electroreceptors [21]. However, determining acuity in the head region is a complex task due to a number of factors. For example, some enclosed environments can interact with this geometry and produce an electric “funneling” effect that increases the local field amplitude and enhances object discrimination [39,40]. Although these studies were performed on a different species of electric fish (pulse-type discharge) with a different electric organ morphology, a detailed investigation of the head region in A. leptorhynchus (the species we consider here) is still warranted. This will, however, require a more complicated 3-D model, so determining how the electric field, body geometry, and receptor density combine to determine electroacuity in the electrosensory fovea is not possible at this time. Nevertheless, on the lateral body surface, the combination of body geometry and current density are such that electric images are sharpest in the midbody [9], thus allowing the objects to be closer in that region before their electric images blur and form a single peak. This apparent tradeoff between more receptors rostrally and higher-quality images caudally may explain why prey detection occurs at approximately equal rates over all rostro–caudal locations [17].
An additional consideration, which again points to interesting future research, is that our current model does not account for the electric field dynamics that could in principle cause midbody acuity to vary over the electric organ discharge cycle. It is possible, for example, that the lowest Smin seen here in the midbody region may shift to other locations for other phases of the cycle, due to the spatial variation of the field in time [38].
In a strict sense, the values we obtain for Smin can be considered as an upper-bound limit on spatial acuity, since various noise sources would undoubtedly result in lower acuity at the behavioural level. However, there are additional cues available from the electric image, and potentially from other sensory modalities, which could help distinguish adjacent objects, and hence increase acuity. Specifically, the electric image produced by two objects is still wider than the image of one of the objects alone, even when their individual peaks are not discernable (see Figure 1C). Moreover, we have only considered two adjacent objects located in parallel with the fish's contour. Indeed, different criteria are required to measure the discrimination of objects that are situated one-behind-the-other (i.e., perpendicular to the fish's contour). Rother et al. [12] have studied such object locations, but not in the context of spatial acuity.
We have shown that electroacuity did not vary with object conductivity. This implies that the fish's ability to resolve two equally sized, equally conductive objects is the same, regardless of whether these objects are animate or inanimate. However, it is possible that the addition of environmental noise to the electric images would make one of these types of objects more “resolvable,” as the SNR would be greater for high-conductivity objects. Water conductivity, on the other hand, does (slightly) affect Smin. Our results are in accord with other findings, which state that object detection is best-achieved in low-conductivity water [17,41,42], confirming the notion that increased water conductivity acts as a type of electrosensory “fog.”
To resolve all of these issues, further behavioral experiments are required. Our current studies using a 2-D electric field model [9] have generated many hypotheses to test in such experiments. Despite the fact that the 2-D model very accurately reproduces many spatial aspects of the electric field [9], ultimately a more detailed 3-D model of the time-varying electric field will be necessary. Measuring electroacuity (behaviorally) in these fish could be accomplished by using a forced-choice experimental paradigm. In this task, the fish could be trained to choose between a single object and a pair of objects, with a reward given for the choice of the latter. An estimate of electroacuity could be obtained by tracking the accuracy of the choices as the interobject distance was decreased (see [33,43,44] for similar protocols).
Weakly electric fish are subject to a wide range of stimuli in natural electrosensory landscapes. Large conducting boundaries, such as rocks or the river bottom, constitute extensive background clutter [27]. The fish therefore has the challenging task of extracting small prey signals from enormous background ones. To investigate this scenario, we have modeled a plant-like background. We have shown that, as this background increases in width, the electric images produced on the fish's skin converge (i.e., the images are blurred). In fact, the image is not much different for background arrays ranging from 0.18 m to 0.3 m wide. In the presence of such a large-background image, the Smin for prey objects would be much larger than for the conditions we have considered thus far, and may in fact be defined only for much larger objects. In other words, as discussed in the following, the electric image component due to the background obscured that due to the two small prey-like objects.
Figure 4 clearly indicates that the effect of a prey is miniscule in the presence of a relatively large-background array. Even at different times during a simulated scanning behavior, the prey only affected the image due to the background by a few percent at most. This suggests that for any static “snapshot” the fish would not be able to extract the prey signal from the large-background signal. On the other hand, weakly electric fish are known to detect minuscule signals under some laboratory conditions [17,45], and presumably can do so in the wild while hunting. We suggest that movement is required in these situations. In fact, due to the blurring effect, the background component of the electric image does not change with fish scanning, whereas the prey component does (see Figure 4B). As a consequence, the small-prey signal is revealed during the scanning motion by looking at the transdermal potential at individual locations on the fish's body (Figure 5A and 5D). In contrast, when background objects are more separated, the prey signal remains confounded by the background (Figure 5B, 5C, 5E, and 5F).
The separation of small signals from background is a universal problem in sensory processing. In vision, the so-called figure-from-ground separation has been extensively studied; luminance and contrast differences between figure and ground provide information-rich cues for this task. In the absence of such cues, however, relative motion (due to figure, background, or observer motion) can provide information that is critical for effective figure-ground separation [29,46]. Motion of an auditory stimulus can also provide cues for sound-source localization in a noisy background [47,48]. Though the particular mechanisms involved in each sense may differ [47], both rely on coherent changes in stimulus parameters (spatial correlation in vision, systematic sweep of interaural time delays in audition). Similarly, we have shown that motion can also lead to small-signal detection in an electrosensory landscape under certain conditions. When the constituent objects of a complex scene are close enough to each other to result in a blurred (spatially uniform) image, a small spatially localized prey signal will pop out due to motion cues (and without motion the prey signal is masked by the large background). On the other hand, to evaluate the specific features of a scene, a greater spacing among constituent objects is required (see Figure 6).
It is important to note that we have only considered the information available to the electrosensory system and have not considered the potential for extracting this information. Information encoded by individual electroreceptor afferents will be pooled in the hindbrain electrosensory lateral line lobe (ELL). Here, the principle neurons, ELL pyramidal neurons, have receptive fields that vary systematically in size across three somatotopic maps. The largest of these receptive fields (lateral segment map) are about 2 cm in width along the body axis of the fish; the smallest receptive fields (centromedial segment map) are about 0.5 cm in width [26,49]. As previous studies have shown, the different maps may take on different roles depending on the type of information available [26,50]. In the context of this paper, the most focused images due to nearby prey objects may be preferentially encoded using pyramidal neurons of the centromedial segment (smaller receptive fields), and the more blurred images due to background objects may be encoded by neurons of the lateral segment (larger receptive fields).
In addition, there are mechanisms in the ELL (via feedback pathways) that can cancel out predictable or redundant stimuli [51,52]. In principle, when the background is spatially uniform (blurred), such feedback mechanisms could cancel out the large-image component due to the background and further enhance small signal extraction during scanning. Recent studies on the signal processing features of ELL neurons have shown that coherence to spatially global time-varying input is high-pass [53], suggesting again that responses to spatially dense backgrounds can be filtered out. Information encoded by ELL neurons is transmitted to higher-order neurons of the midbrain. Recent studies have described plasticity and motion sensitivity in these neurons [24,54], but further studies will be required to determine how these neurons contribute to the computations involved with prey detection and discrimination in complex landscapes.
It has been widely hypothesized that the stereotypical back-and-forth scanning behavior exhibited by weakly electric fish could be used to generate electrolocation cues [25,55,56]. In fact, cues obtained by self-motion are used by many different animals to extract relevant sensory features [28]. For example, primates move their fingers laterally to detect fine features in textured surfaces, which would otherwise go unnoticed [57]; rodents perform whisking behaviors [58]; and insects, such as mantids, can obtain information about an object's depth using a side-to-side “peering” movement (by means of motion parallax cues; [59]). Such examples have led to the reasonable notion that the exploratory behaviors exhibited by weakly electric fish, such as the aforementioned scanning, act similarly to provide relevant information from complex electrosensory scenes. Our study describes the nature of these motion-generated cues for the first time, and indeed shows that their effectiveness depends on context.
In particular, our results predict that weakly electric fish should exhibit the specific search behavior that is most suitable for signal extraction in a given context. The scanning behavior would be best suited for spatially dense or uniform backgrounds, whereas the fish might preferentially use tail-bending in cases where the background is sparse (as in Figure 5B, 5C, 5D, and 5F where the prey component is confounded with the background signal). In future studies, we aim to determine which behaviors are used most frequently by the fish to explore electrosensory landscapes with varying spatial characteristics.
The 2-D electric field of a 21-cm A. leptorhynchus was simulated using a finite-element–method model described previously in [9]. Briefly, the model reproduces the field measured at one phase of the quasisinusoidal electric organ discharge. It consists of three components: an electric organ (EO), a body compartment, and a thin skin layer. The EO current density and the conductivities of the three components were optimized using raw data provided by Christopher Assad [38]. The optimized EO current density is spatially structured; as compared with a simple dipole, it is skewed toward the tail. Such a profile in the EO current density, as well as the spatial filtering due to the tapered body shape, reproduces the asymmetric “multipole” electric field [9,10,27]. To distinguish this situation from that of a simple dipole, we sometimes refer to the fish's electric field as “dipole-like.” This model is a 2-D simplification that is static in time, and so, in principle, any results derived from it are qualitative. It is important to note, however, that the model provides a quantitatively accurate representation of the data measured in the horizontal plane [9], and thus should be very reliable. Of course, as we note in the Results and Discussion sections, there are some questions that will require a detailed time-varying 3-D model.
Electric images were calculated in one of two ways using custom MATLAB subroutines. In Figures 1–4, images are defined as the differences in transdermal potential, with and without objects present (this has become the standard definition of an electric image, [5]). In Figure 5, images are displayed as the raw transdermal potential, the natural electrosensory input. All images are shown only for the side of the fish body closest to the objects. Water conductivity was set to 0.023 S/m, as in [38]. The prey chosen, Daphnia magna, was modeled as a 3 mm–diameter disc with a conductivity of 0.0303 S/m, as in [15,17]. The background objects (2-cm discs) simulated throughout this paper were based on the conductivity of the aquatic plant Hygrophilia [22] (0.0005 S/m). The goal was not to mimic the plant's geometry accurately, but rather to get a general idea of the effects caused by varying backgrounds with plant-like conductivity and size.
To estimate the fish's ability to resolve two distinct objects (electroacuity), the minimal distance Smin was calculated. This measure is the interobject distance, which delimits an electric image with one peak from one with two peaks (for example, see Figure 1C). This quantity depends on a number of parameters such as the object's size, its rostro–caudal and lateral location, and the water conductivity. We can develop more intuition for how Smin behaves assuming that images of objects are idealized Gaussians. Consider two Gaussians along the x-axis, of similar standard deviation σ and amplitudes, but centered on μ1 and (−μ1 ), respectively. Assuming linear superposition, their sum along the x-axis will have one or two maxima, depending on the relation between the standard deviation and the mean, i.e., on the relative width. It can be shown that Smin in this case corresponds to (2σ). If the amplitudes of the Gaussians change in the same way, as they do when the object is closer to the fish, Smin remains the same; it will increase, however, if there is disparity in the amplitudes. Smin will also increase with increasing image width. Although this provides some insight on the behavior of Smin, it is important to note that linear superposition is not valid in general (for example, see Rother et al. [12]). Also, all of the images we show are computed using our model, which can accommodate arbitrary object combinations. In no cases do we assume linear superposition of images due to individual objects.
For a given pair of objects, the rostral object's center coordinates were chosen as the spatial location for which the Smin was determined. Therefore, this object was held stationary during a given Smin measurement. The caudal object was moved systematically in the caudal direction until two distinct peaks appeared in the electric image (object center-to-skin distance was kept constant). Using this technique, Smin measurements were accurate to within 0.5 or 1 mm, representing the chosen sampling (see error bars in Figure 2).
In the last part of the paper, where fish motion is simulated, a scanning speed of 0.1 m/s was chosen, which is in the range of measured weakly electric fish scanning velocities [45,56]. For quantifying the SNR between the two different transdermal potential time series (Figure 5, green and blue curves), i.e., the ones produced by the background alone (Φback) and by the background and prey (Φback+prey), a root-mean-squared difference measure was used (Equation 1):
where n represents the number of different fish locations that were simulated, i.e., samples of the transdermal potential at a given body location during a 1-s scan (we chose n = 21). A large SNR value means that the two time series are very distinct. We have also quantified the discriminability of two objects as they are separated (Equation 2). Here, we assumed that the separate (simulated) electric images generated by each object is a spatial Gaussian function (along one dimension; each of mean μi and width σi) and have computed the discriminability d′ [60,61]:
|
10.1371/journal.pcbi.1003627 | AprioriGWAS, a New Pattern Mining Strategy for Detecting Genetic Variants Associated with Disease through Interaction Effects | Identifying gene-gene interaction is a hot topic in genome wide association studies. Two fundamental challenges are: (1) how to smartly identify combinations of variants that may be associated with the trait from astronomical number of all possible combinations; and (2) how to test epistatic interaction when all potential combinations are available. We developed AprioriGWAS, which brings two innovations. (1) Based on Apriori, a successful method in field of Frequent Itemset Mining (FIM) in which a pattern growth strategy is leveraged to effectively and accurately reduce search space, AprioriGWAS can efficiently identify genetically associated genotype patterns. (2) To test the hypotheses of epistasis, we adopt a new conditional permutation procedure to obtain reliable statistical inference of Pearson's chi-square test for the contingency table generated by associated variants. By applying AprioriGWAS to age-related macular degeneration (AMD) data, we found that: (1) angiopoietin 1 (ANGPT1) and four retinal genes interact with Complement Factor H (CFH). (2) GO term “glycosaminoglycan biosynthetic process” was enriched in AMD interacting genes. The epistatic interactions newly found by AprioriGWAS on AMD data are likely true interactions, since genes interacting with CFH are retinal genes, and GO term enrichment also verified that interaction between glycosaminoglycans (GAGs) and CFH plays an important role in disease pathology of AMD. By applying AprioriGWAS on Bipolar disorder in WTCCC data, we found variants without marginal effect show significant interactions. For example, multiple-SNP genotype patterns inside gene GABRB2 and GRIA1 (AMPA subunit 1 receptor gene). AMPARs are found in many parts of the brain and are the most commonly found receptor in the nervous system. The GABRB2 mediates the fastest inhibitory synaptic transmission in the central nervous system. GRIA1 and GABRB2 are relevant to mental disorders supported by multiple evidences.
| Genes do not operate in vacuum. They interact with each other in many ways. Therefore, to figure out genetic causes of disease by case-control association studies, it is important to take interactions into account. There are two fundamental challenges in interaction-focused analysis. The first is the number of possible combinations of genetic variants easily goes to astronomic which is beyond current computational facility, which is referred as “the curse of dimensionality” in field of computer science. The other is, even if all potential combinations could be exhaustively checked, genuine signals are likely to be buried by false positives that are composed of single variant with large main effect and some other irrelevant variant. In this work, we propose AprioriGWAS that employees Apriori, an algorithm that pioneers the branch of “Frequent Itemset Mining” in computer science to cope with daunting numbers of combinations, and conditional permutation, to enable real signals standing out. By applying AprioriGWAS to age-related macular degeneration (AMD) data and bipolar disorder (BD) in WTCCC data, we found interesting interactions between sensible genes in terms of disease. Consequently, AprioriGWAS could be a good tool to find epistasis interaction from GWA data.
| Gene-gene interactions have been proposed as one potential explanation of the well-known problem of missing heritability [1], and a recent report [2] has quantitatively demonstrated that possibility. Researchers have long attempted to identify interactions, with methods ranging from evolutionary genetic studies [3], [4], systems biology studies of model microbes [5] and quantitative genetic studies of inbred model organisms, to linkage [6] and association studies in human populations [7]–[14]. Although the definitions of the term “epistasis” used by biologists (Batson 1909) [15] and statisticians (Fisher 1918) [16] are different, they have the same consequences regarding different distributions of genotype patterns among different phenotypes.
The main obstacle of interaction analysis is that the large number of multi-locus genotype combinations generated from large numbers of genetic variants is too high for current computational resources. This is in fact a well-known computational problem, known in the field of computer science as the ‘curse of dimensionality’ [17]. In this work we developed AprioriGWAS, a tool to address this problem. This tool is based on a successful algorithm in the field of computer science, Apriori [18].
Apriori was originally designed for supermarket data mining to assist shop owners in designing the layout of displayed products. Given customers' transactions, the algorithm can identify sets of items that frequently co-exist in transactions. For example, by knowing that customers usually buy milk and bread together, the shop owner can put them near each other in the store.
Before describing the algorithm, we briefly give definitions of a few key terms: item is defined as an individual product, for example, bought by a customer; itemset stands for a set of items purchased together; length of itemset is defined as the number of items in the itemset. The process of growing a short itemset to a longer itemset is referred to as pattern growth. Generally, the key insights of Apriori are that: (1) frequent itemset with many items can be gained by growing itemset of short length; and (2) since subsets of any frequent itemset should also be frequent during pattern growth, itemsets predicted not to have any effect can be dropped during pattern growth, thereby significantly reducing the search space. In the case of GWAS, the number of individual genotypes is analogous to the number of transactions in supermarket data. The genotype of a variant is an item, and genotype combinations of different variants are an itemset, here also called a genotype pattern. Instead of just finding frequent genotype patterns, we want to find genotype patterns with different frequencies in cases and controls. We call them differential genotype patterns. While Apriori originally works on one database to find the most frequent itemsets, we are interested in patterns with different frequencies in two databases (cases and controls). To assess whether a pattern should be retained during pattern growth, we make use of the proportion test [19] (Methods).
Interaction among variants is carried out after obtaining all differential genotype patterns. We test the possibility of interaction among variants involved in a differential genotype pattern by conducting Pearson's Chi-square test for the contingency tables composed of all genotype patterns found for variants and phenotypes (Methods). In this step, we try to distinguish whether a differential pattern is caused by variants with marginal effects or by interaction effect. The process of pattern growth helps to narrow down the number of variant combinations to be tested for interaction effect.
Using simulations following Marchini et al's procedure [11], we demonstrate that AprioriGWAS can approximately achieve the same coverage of associated patterns as an exhaustive search, but with far lower CPU time.
Determining all potential combinations that are statistically associated with disease does not automatically identify genuinely interacting genes. The daunting number of all combinations of variants heavily increases the load of multiple tests and mixes genuine signals with noise. As summarized by Anderson [20], in the regression model with two main effects terms and one interaction term, there is no exact permutation method for testing the significance of the interaction term. Buzkova et al [21] proposed a parametric bootstrap test for gene-gene and gene-environment interactions, which unfortunately is not practical for very large numbers of possible combinations of variants. Computer simulation [22] shows that whenever a trait is controlled by more than a single factor, it becomes possible for a neutral variant together with a major-effect variant as a pattern to be more strongly associated with the trait than with any of the causative factors [13]. These indirect associations are true associations for statistical purposes, and can be indistinguishable from medical causative associations [22]. To distinguish general association and interaction effects, we developed a new conditional permutation test to distinguish genuine interactions from the artifacts generated by the combination of a major-effect variant with a neutral variant (Methods). We demonstrate that our new approach has a magnitude lower false discovery rate (FDR) compared with regular permutation, while maintaining comparable power.
We applied AprioriGWAS to age-related macular degeneration (AMD [MIM 153800]), which has been deemed a good example of a small number of common variants explaining a large proportion of heritability [1]. Among the most significant patterns, we found six pairs of retinal genes interacting with each other. An exciting example is the interaction of a gene involved in an AMD treatment target, ANGPT1, with another important AMD gene, CFH. Overall, the potentially interacting genes were enriched in glycosaminoglycan biosynthetic process (). Many studies have shown that the interaction between glycosaminoglycans (GAGs) and CFH plays an important role in the disease pathology of AMD. We also applied AprioriGWAS to bipolar disorder; we found potential interactions inside individual gene (8 out of 18 genes are related with mental disorder) and interactions across gene or chromosomes. Further results will be presented in full later.
The remainder of this paper is organized as follows. In the next section we introduce the AprioriGWAS algorithm for mining possible interaction variants, as well as the conditional permutation approach for testing interactions. We then evaluate the performance of AprioriGWAS with simulated data and compare it with logistic regression implement in Epistasis function of PLINK. Lastly we demonstrate applications of AprioriGWAS to AMD and WTCCC bipolar data and exciting findings from both datasets.
Historically, the Apriori algorithm can be traced back to the seminal paper published by IBM Research in 1993 [18]. The concept of the main technique is that a subset of frequent itemset should also be frequent. Based on this concept, frequent itemset with more items may be found by stepwise growth of smaller frequent itemset, which saves substantial computational resources. Interested readers may refer to their original paper [18] for a professional description or to our own longer report [23] for illustrative descriptions. Here we briefly outline the main steps. Suppose one wants to mine frequent itemset with length no more than n. Apriori will usually scan dataset in n rounds (unless there is no new frequent itemset generated in a certain round before n, thereby forcing the algorithm to halt). In the first round, it will initiate the 1-itemsets that are frequent. In each subsequent round, it will take the frequent itemset generated in the last round as starting point and grow any itemset by adding one more item. Retention of the new itemset will be decided by firstly predicting how likely it will be and then, given a positive prediction, by checking the actually supporting transactions. Finally, the collection of all frequent itemset in all rounds will be reported.
In this paper, genotype patterns are defined as genotype combinations of different variants. We use integer numbers as ids of variants; then we can have, for instance, a pattern like 46AT_609GG_1099CC, denoting a pattern composed of a variant with id 46 and genotype AT combined, a variant with id 609 and genotype GG, and a variant with id 1099 and genotype CC. The key goal is to find genotype patterns that have a significant frequency difference in cases and controls (called differential patterns in this paper).
The algorithm of AprioriGWAS is divided into two steps. First, detecting differential genotype patterns by an Apriori-like strategy. Obviously, the same set of variants can lead to several differential genotype patterns. Second, testing interaction among a set of variants by testing association of all possible combinations of genotype patterns against case/control status. The first step helps to narrow down the combinations of variants need to be tested. Due to multiple test problems and potential association of single variants involved in the differential genotype pattern, we adopt a new conditional permutation in the second step to control the marginal effect of single variants for testing of variant interactions.
We simulated data by two-locus interaction models proposed by Marchini et al [11] (Methods), in which three types of interactions are generated. We then applied regular permutation and conditional permutation to control family-wise type I error.
The performances of regular permutation and conditional permutation test (Methods) are demonstrated in Figure 1A and 1B. We compared both power and FDR, using regular permutation and conditional permutations to adjust thresholds for type I error. Family-wise type I error was set to 0.05 for both methods. It is evident that the FDR was significantly reduced by the conditional permutation test, although some power is sacrificed compared with regular permutation.
To demonstrate that the nominal p-value of a contingency table for multi-variants could be in large part caused by individual variants with strong marginal effect, we took a real example from analyzed AMD data. Figure 2A shows two variants, each with no marginal effect, but in combination with strong marginal effect. Figure 2B shows two variants, one has strong marginal effect, and the other does not show any marginal effect. Although the nominal p-value of the contingency table is more significant than the pair of variants in Figure 2A, one can deduce that the low p-value from Figure 2B is in large part caused by the variants with strong marginal effect; in Figure 2A, on the other hand, there must be some interaction effect.
As mentioned, AprioriGWAS manages to dramatically speed up the search process by dropping the candidate genotype patterns unlikely to grow to differential pattern. Since it is based on prediction at an early stage in the search, it still theoretically runs the risk of mistakenly dropping sensible patterns. Here we quantitatively tested the percentage of mistakenly dropped differential patterns by comparing AprioriGWAS and exhaustive search (Method).
Figure 3 shows the comparison between searching for combinations of variants (with default parameters in AprioriGWAS) and exhaustive search. We found that 97% of all differential genotype patterns found by exhaustive search were covered by the results from AprioriGWAS. With such high coverage, the chance of losing possible interaction variants is minimized. There are a few points below 85%, reflecting that there is variation of power to cover all potential combinations. It is true that the overall coverage is subject to lots of parameters, like sample size and allele frequency. To minimize this variation, larger sample size is always desirable.
We compared the ability of AprioriGWAS to find interacting variants with traditional single locus genotypic test and exhaustive search in PLINK [28] (epistasis function). The epistasis function in PLINK for case control data is basically stepwise logistic regression. We chose to use the all combinations option. The power comparison is based on two levels: finding at least one casual variant, or finding both interacting variants (Figure 4).
For Level 1, detecting at least one causal variant, we found that the traditional single variant test had the highest power in Model 1, which has explicit marginal effects for both causal variants. AprioriGWAS performed similarly with the single loci test in Model 2, and had better power in Model 3 (Figure 5). This is natural, since Model 2 and 3, which contain no explicit marginal effects, are expected to be harder to detect without an interaction-based searching strategy.
For Level 2, detecting both interacting variants, it is evident that AprioriGWAS had the highest power in most cases of Model 2 and 3 (Figure 4). On the other hand, the performance of the epistasis function in PLINK, which exhaustively searches all combinations, was not as good in all cases. This is because: (1) stepwise logistic regression does not capture the interactions well, since the effects of the terms are added in a linear manner, whereas AprioriGWAS explicitly addresses detailed patterns; (2) in stepwise logistic regression the genuine interactions are buried by the noise of a too large number of combinations, whereas with the conditional permutation test used in AprioriGWAS, genuine interactions are able to stand out.
When comparing corresponding panels in Figure 4 and Figure 5, it is observed that for the single variant test the power of finding both interacting variants (i.e., Level 2) dropped significantly compared with the power of finding at least one causal variant (i.e., Level 1). By contrast, interaction based methods, i.e., both AprioriGWAS and PLINK epitasis, maintained similar power for both levels. This was not unexpected since the interaction-based strategies should be better able to find an epistasis effect.
We also simulated data that have more SNPs (1,000,000) and find that the relative power between three methods and interaction models remain similar although the absolute powers are all decreased. (Figure S1)
Figure 6 shows the power of AprioriGWAS and single variant test on three classical genetic models studied in model organisms. There are three powers for each genetic models: power for detecting at least one gene using single variant test, power for detecting both genes using single variant test, and power for detecting both genes using AprioriGWAS. Since PLINK is not scalable for such a dataset, we have not achieved power estimates for logistic regression. For the model “Duplicated Dominant”, AprioriGWAS outperforms single marker test for detecting single gene or both genes, whereas for models “Duplicated Recessive” and “Dominant & Recessive Interaction”, AprioriGWAS is more powerful for detecting both genes, but not for detecting single genes. It is notable that the power of detecting both genes in the model “Dominant & Recessive Interaction”, in which epistasis is functioning; single variant test has almost zero power (0.1%) while AprioriGWAS has around 50% power.
We compared the speed of our method with the epistasis function in PLINK. Figure 7 shows that the default threshold setting in AprioriGWAS was approximately a magnitude faster. Although retaining candidate genotype patterns in memory can help speed up the algorithm, its affordability is subject to the particular computational resources.
We took the strategy of writing candidate patterns on hard disk for each round of pattern extension. The genotype data used to be relatively small comparing with the patterns however is getting larger and larger empowered by new sequencing platforms. To solve this problem, we implemented AprioriGWAS using HDF5-based data format [29] which stores genotype data on disk and accesses them as though stored in main memory. Therefore, the memory usage is scalable to whatever size of potential dataset and the speed is not scarified. (See more on computational and memory complexity in section Discussion.)
We applied AprioriGWAS on published AMD data [26]. We identified 168 significant pairs of variants (family-wise type I error = 0.01), presented in Table S1. By checking published functional literals and gene annotations, as well as GO enrichment of the genotype patterns, we learned that the findings are well validated by existing functional studies and clinical applications.
Besides AMD data that were extensively analyzed by the community interested in gene-gene interactions, we also applied AprioriGWAS on Bipolar Disorder data from WTCCC [27] to further test whether it is scalable for larger dataset. The whole task was distributed onto 1,000 CPUs in a cluster and the average execution time for a single job is 56.8 hours. Only 4 Gb memories were employed during the computation, evidencing the great performance of HDF5-based implementations.
We have introduced AprioriGWAS, patterned after the Apriori algorithm in the bioinformatics field of frequent itemset mining (FIM), as a tool for detecting main and interaction effects of genetic variants in case-control association studies. One of its outstanding properties is that it can find variants whose disease association lives solely from their interaction without having (appreciable) main effects. We applied our approach to a published dataset on AMD and documented that AprioriGWAS furnishes sensible results. In fact, it found an AMD-associated variant (ANGPT1) not previously reported to be associated with AMD. We also identified interesting genes from WTCCC bipolar disorder data. One good point is that GO term enrichment analyses of all the genes identified, always show sensible terms for relevant disease. Our description of these findings is primarily intended to show the efficacy of our approach rather than to provide research findings about AMD and bipolar disorder.
Regardless the goal being interaction or single gene, statistical tests all suffer from the problem of false positives. Since the numbers of variants (and their combinations) are usually a few magnitudes larger than the sample size for most association studies, it will be common to see false positives. The current practice in the community is that researchers who would like to claim association or carry out experimental validations usually have to check whether the results are replicable in other independent dataset(s) Researchers who use AprioriGWAS can also use this to filter results before doing experimental validations. As an example, we use another independent dataset for AMD study [64] to check whether the results are replicable. Among the five interactions with CFH reported in this paper, we found that BBS9/CFH and CHRM2/CFH are replicated in the other dataset. However, we understand that these two datasets are very different: one is wet AMD and the other is dry AMD. One of them is more prevalent in Asia than the other. Therefore, our further analysis of data in [64] may not serve as perfect replication of the findings presented, although it suggests that BBS9 and CHRM2 may be of higher priority for further experimental validations.
The most commonly used multiple variants analysis is stepwise regression, in which variants are added to the regression equation one after another by some suitable criteria. But statistical analysis shows that the usual stepwise model selection methods are path dependent and therefore suboptimal [65]. Besides regression, some methods are based on discrete mathematics, like the Combinatorial Partition Method (CPM) [66] and its refined version, the Restricted Partition Method (RPM) [67]. However, RPM still requires a daunting number of tests when the number of variants is high. This is because its insight into reducing tests lies in its practice to combine close phenotypes, which consequently does not entirely solve the problem of too many combinations of genotypes. Another well-known method of counting potential combinations is multifactor dimensionality reduction (MDR). It collapses cells in a contingency table into two groups and conducts a test on them. Essentially however it reduces the dimensionality of testing, rather than reducing the dimensionality of the process of counting genotype patterns. Therefore, when the number of variants is large, it still suffers from the “curse of dimensionality” [17]. Bayesian methods leveraging MCMC, e.g, BEAM [53] or epiMODE [8], should theoretically suffer less from computational limitations, but they do not directly test detailed combinations of genotype patterns and thereby sacrifice the advantages of fine scale learning of gene-gene interactions. Another branch of frequently used methods is two-stage analysis [68], by which the investigator can utilize relatively “simple” or computationally efficient tests to choose qualified variants in the first stage analysis. Then, taking advantage of the relatively small number of variants, the investigator can adopt some advanced but computationally heavy test to identify interacting genes. However, due to a lack of strong prior knowledge, the true signals might have been removed from the first stage if the procedure was not well designed. As an example, interacting variants with no marginal effect may be filtered out if one uses tests based on marginal effects of single variants in the first stage. Nevertheless, with good design, this approach is still very promising and can be combined with all the approaches reviewed above; and it can naturally also be combined with the method proposed in this work.
Computation time and spatial complexities of the tool may be interesting to the reader. The number of transactions for original Apriori corresponds to sample size in GWAS; the number of items is equivalent to the number of variants and the itemsets. In contrast to supermarket data, GWAS data have a limited number of “transactions”, but a large number of “items” in two datasets, cases and controls. Both conditions make the problem more difficult. The time spent reading the data in each round of pattern growth is constant. In addition, the computational resources cost depends on how many combinations of genetic variants will be generated and tested. The more combinations are tested, the less likely it is that genuine patterns are missed, though of course more resources will be used. In AprioriGWAS, there are several parameters for the user to specify according to their computer resources and understanding of the disease model. The threshold for the proportion test and minimal support of concerned itemsets are parameters that affect candidate search space, algorithm speed, and power of detecting all distinct genotype patterns. When these parameters are set to zero, AprioriGWAS will exhaustively search all possible combinations. (Please refer to our Manual of AprioriGWAS for the tradeoffs and discussions on setting these parameters according to computational resources.)
Those familiar with Apriori may suggest that, given Apriori's ability to also mine association rules, one could also treat the case control label as items and directly adopt Apriori for case/control data. The result will then be a subset of variants that can imply the case/control labels. But searching frequent itemsets and then mining the association between genotype pattern and disease status is inefficient, since frequent genotype patterns are not necessarily associated with phenotype; on the other hand, genotype patterns strongly associated with phenotype may not necessarily be in high frequency, and such an association could be distributed in different patterns than the same variants combinations.
Instead of the conditional permutation proposed here, one could also consider Bonferroni correction. For n variants with search length of m, the total number of combinations is huge. Given the natural correlation of the combinations, it is clearly far more stringent than necessary. However, only correcting on the number of differential pattern tested produces a bias in the other direction, since the nominal value of the significance level of the chi-square test for the contingency table will be inflated by the selection procedure [69]. It is therefore always preferable to use a permutation test for the whole procedure. With regular permutation, one permutes the Case/Control label and then performs the whole test process. The smallest P-value of each permutation are ranked, allowing one to get the distribution of test statistics under “Null” from the permuted dataset. With regular permutation, no variant should have marginal effect, and the p-value of the contingency table for the combination of variants is under the null hypothesis of no variants having marginal effect.
However, regular permutation suffers from an inflated significance level for contingency tables containing variants with marginal effects. This is due to the fact that when a contingency table is composed of at least one variant with strong marginal effect, the p-value for that contingency table becomes extremely small compared with regular permutation results. The FDR is therefore very high, even close to 1.
To solve the problem of an inflated significance level by a contingency table composed of at least one variant, v, with strong marginal effect, we developed a conditional permutation procedure (Methods), which helps get the null distribution of the p-value of a contingency table composed of the variant and other variants. Simulation results show that, when we control the family-wise type I error by conditional permutation, we also keep FDR well controlled. Compared with INTERSNP [13], which lists only the top 50 variant combinations including the variant with marginal effect, conditional permutation in AprioriGWAS keeps FDR well controlled in a systematic way.
Another concern might be whether these differential genotype patterns are artifacts caused by linkage disequilibrium (LD). We believe this is not the case, since the LD should impact both cases and controls, and therefore the pattern created by LD will not be differential unless the LD structure is significantly different in cases and controls for particular genetic variants. If that is the case, then there must be some reason of selection to explain the deviation in the genotype pattern, and it is difficult to judge whether this is an artifact or something of interest. In addition, our conditional permutation also breaks LD between interacting variants.
Low-frequency or rare variation might impact the performance of the method, even when explicitly only testing for interactions among common variants. What matters is the extent of LD between causal rare variants and testing common variants. We haven't addressed this problem in the current method. It would be interesting to extend AprioriGWAS toward that direction. There may be non-trivial statistical challenges since the low-frequency or rare variants are usually less shared by the individuals therefore their combinations that form genotype patterns will be even less shared by individuals. For a given set of variants, we will have many patterns with little supports.
|
10.1371/journal.pbio.0050268 | Force-Induced Unfolding of Fibronectin in the Extracellular Matrix of Living Cells | Whether mechanically unfolded fibronectin (Fn) is present within native extracellular matrix fibrils is controversial. Fn extensibility under the influence of cell traction forces has been proposed to originate either from the force-induced lengthening of an initially compact, folded quaternary structure as is found in solution (quaternary structure model, where the dimeric arms of Fn cross each other), or from the force-induced unfolding of type III modules (unfolding model). Clarification of this issue is central to our understanding of the structural arrangement of Fn within fibrils, the mechanism of fibrillogenesis, and whether cryptic sites, which are exposed by partial protein unfolding, can be exposed by cell-derived force. In order to differentiate between these two models, two fluorescence resonance energy transfer schemes to label plasma Fn were applied, with sensitivity to either compact-to-extended conformation (arm separation) without loss of secondary structure or compact-to-unfolded conformation. Fluorescence resonance energy transfer studies revealed that a significant fraction of fibrillar Fn within a three-dimensional human fibroblast matrix is partially unfolded. Complete relaxation of Fn fibrils led to a refolding of Fn. The compactly folded quaternary structure with crossed Fn arms, however, was never detected within extracellular matrix fibrils. We conclude that the resting state of Fn fibrils does not contain Fn molecules with crossed-over arms, and that the several-fold extensibility of Fn fibrils involves the unfolding of type III modules. This could imply that Fn might play a significant role in mechanotransduction processes.
| Cells are embedded within an extracellular matrix that regulates many cellular processes, including stem cell differentiation and cancer progression. Yet the underlying molecular mechanisms that mediate these processes remain unknown. Within the extracellular matrix of cells, super-molecular assemblies of fibronectin are dynamically stretched many times beyond their resting length by cell traction forces. Whether mechanical forces generated by cells can mechanically unfold fibronectin has been controversial. Clarification of this issue is important since fibronectin displays a large number of molecular recognition sites whose display might be altered by protein unfolding. Here, we used spectroscopic approaches to visualize whether this extracellular matrix protein is unfolded in cell culture. We show that indeed cell traction forces straighten fibronectin and unfold its modules. Fluorescence resonance energy transfer reveals the extent to which the extracellular matrix unfolds and thus potentially regulates cell signaling processes.
| Fibronectin (Fn), a major component of the extracellular matrix (ECM) of developing tissues and healing wounds, is a large, dimeric protein consisting of more than 50 repeating subunits (for review see [1–6]). Fn displays a number of surface-exposed molecular recognition sites for cells, including integrin binding sites such as the RGD loop, PHSRN synergy site, and LDV sequence, and binding sites for other ECM components, including collagen, heparin, and fibrin. Together these binding sites provide Fn with a diverse array of scaffolding and cell recognition functions. In addition, a number of cryptic binding sites, sequences normally buried in the equilibrium fold of the protein, and surface-exposed binding sites have been proposed to be exposed or deactivated, respectively, as a result of force-dependent conformational change (as reviewed in [2]). It has therefore been hypothesized that in addition to other physical properties of the ECM such as substrate rigidity and matrix composition, matrix unfolding may alter outside-in cell signaling.
Soluble Fn in physiological buffer has a compact, folded quaternary structure (Figure 1A) stabilized through intermonomer ionic interactions between III2–3 of one arm and III12–14 of the opposing arm [7]. Low concentrations of chemical denaturants first destabilize these ionic interactions, leading to separation of the crossed-over arms (extended structure; Figure 1B), and increasing denaturant concentrations finally unfold Fn (Figure 1C). Erickson originally proposed that module unfolding constituted the mechanism for fibril extensibility by estimating the free energy of denaturation and extension of Fn type III (FnIII) modules in comparison to the force generated by single myosin or kinesin motor proteins [8]. Yet the conformation of Fn within fibrillar ECM is still debated [8–12]. Two structural models to explain the several-fold, force-induced extension of Fn within fibrils have been proposed:
The quaternary structure model proposes that Fn within fully relaxed fibrils assumes a compact structure with crossed-over Fn arms similar to that found in solution (Figure 1D; [12]). Under the influence of cell traction, tensile force would first separate the Fn arms, and finally align them along the force vector (Figure 1E) with still intact tertiary/secondary structures of individual modules (Figure 1F; [12]). This model postulates that the fibrils break before Fn starts unfolding.
The unfolding model for Fn elongation proposes that fully relaxed fibrils are composed of Fn in an already extended conformation where the dimeric arms are already separated (Figure 1F) and that fibril extension originates from the unfolding of FnIII modules (Figure 1G). Type I and II modules are stabilized by intramodular disulfide bonds, and therefore only FnIII modules can be completely unfolded by force. FnIII modules have been shown to unfold by passing through several intermediate states [13–18] .
While previous fluorescence resonance energy transfer (FRET) studies indicated that cell contractility is sufficient to unfold fibrillar Fn [9,10,19], the alternative quaternary structural model was proposed based on the following two observations. First, a single-molecule Fn–green fluorescent protein (GFP) study [12] showed that the mechanical stabilities of GFP and the FnIII module are similar over the range of pulling velocities tested (50 to 1,745 nm/s). Second, if Fn-GFP is assembled by cells into micro-sized ECM fibers, stretched Fn-GFP fibrils are reported to maintain a constant integrated level of fluorescence even after contraction to 1/3 or 1/4 their starting length [11,20]. Combining this information derived from single-molecule mechanics with spectroscopic data obtained from a densely packed Fn-GFP fiber, it was concluded that FnIII modules do not unfold in Fn fibrils under the influence of cell traction forces since GFP fluorescence was independent of the extent to which the Fn fibril was strained [11,20]. However, it is unclear whether Fn-GFP embedded in a densely packed fiber may have an altered mechanical stability compared to single-molecule Fn-GFP.
Determining whether Fn is indeed unfolded in ECM fibrils in vivo by cell contractile forces is essential to understand the molecular mechanism of Fn fibrillogenesis and whether exposure of the numerous molecular recognition and cryptic sites alters Fn function in a strain-dependent manner [1,2]. Since no experimental techniques were available to directly probe the loss of tertiary/secondary structure of Fn in cell culture, intramolecular FRET was used to gain conformational information [9,10,21–23]. Here, two Fn labeling schemes for FRET were utilized. Amine/cysteine FRET-labeled Fn (amine/cys Fn-DA) was produced by labeling plasma Fn on all four free cysteines within III7 and III15 (yellow modules in Figure 1) with Alexa 546 acceptors and on seven random amines with Alexa 488 donors. In the second labeling approach, only the free cysteines per Fn dimer were randomly labeled with two donors and two acceptors (cysteine/cysteine FRET-labeled Fn [cys/cys Fn-DA]). The Förster radius of this fluorophore pair is ∼6 nm (from Invitrogen); therefore, energy transfer is limited to within 12 nm of III7 and III15 (yellow fading spheres in Figure 1). Amine/cys Fn-DA provides sensitivity to the full conformational range since energy transfer can occur between the crossed arms (intermonomer FRET; Figure 1A and 1D) and along the arms (intramonomer FRET; Figure 1B, 1C, and 1E–1G). The crossover of Fn arms in the fully compact conformation brings the fluorophores attached to III7 of opposing arms into close proximity (see Figure 1A and 1D). Therefore, cys/cys Fn-DA is sensitive to intermonomer energy transfer in the compact conformation. However, energy transfer between opposing arms cannot occur when the arms are separated [24,25]. The spatially resolved ratio of acceptor to donor intensities (IA/ID) was quantified using both labeling schemes within the ECM of living fibroblasts cultured on glass.
Cys/cys Fn-DA with an average of 1.8 donors and 1.9 acceptors was generated by labeling the free cysteines in modules III7 and III15 of isolated human plasma Fn (unlabeled FN [Fn-u]) with a 15-fold molar excess each of Alexa 546 acceptors and Alexa 488 donors. Labeled Fn was verified to be both dimeric and not contaminated with Fn fragments (as confirmed by Coomassie blue staining of samples run in SDS-PAGE gels; Figure S1A). Given the large size of Fn, our approach utilizes multiple fluorophores on each Fn dimer such that the cumulative IA/ID is sensitive to large conformational changes. Unlike single donor/acceptor pair–labeled proteins, which can be used as nanoscopic rulers [26,27], donor–acceptor distances can thus not be calculated. Confocal microscopy was used to acquire donor, acceptor, and differential interference contrast (DIC) images, and ratiometric IA/ID images were generated by averaging, background subtraction, thresholding, and color-coding. Intensities were measured simultaneously on two photomultiplier tubes (PMTs) for each pixel in a given field of view, with 12-nm bandwidths over acceptor and donor emission peaks.
Energy transfer was first quantified for known conformations of cys/cys Fn-DA in solution with varying concentrations of guanidine hydrochloride (GdnHCl). FRET was independent of the concentration of Fn labeled with Alexa 488 and Alexa 546 (Fn-DA) in the range from 0.05 to 0.25 mg/ml, indicating that intermolecular FRET did not occur in solution (data not shown), and solution IA/ID measurements were found to be constant over a large range of measured intensities (Figure S1B). Up to 6 M GdnHCl did not impact the FRET efficiency of a similar fluorophore pair (Alexa 488/Alexa 594; [28]), suggesting denaturation measurements are not affected by GdnHCl.
Compactly folded dimeric cys/cys Fn-DA in phosphate-buffered saline (PBS) showed energy transfer (IA/ID = 0.74) that was significantly attenuated in the extended conformation in 1 M GdnHCl (IA/ID = 0.37) and reached a baseline of IA/ID = 0.34 in both 2 M and 4 M GdnHCl (Figure 2A). Monomeric Fn-DA (Figure S1A) reduced in 50 mM DL-dithiothreitol (DTT) [29] was used to eliminate intermonomer energy transfer. Reduced Fn-DA showed intermediate energy transfer in PBS (IA/ID = 0.48); however, IA/ID values for monomeric and dimeric cys/cys Fn-DA in 1, 2, and 4 M GdnHCl were identical. IA/ID does not fall to zero in 4 M GdnHCl although Fn-DA is almost completely unfolded (Figure 2B). This baseline ratio does not reflect residual conformational sensitivity, but instead reflects spectral crosstalk from direct excitation of the acceptor with 488-nm light and bleed emission from the donor into the acceptor detection window.
To test whether the above baseline IA/ID for monomeric cys/cys Fn-DA in PBS results from the reported artificial dimerization of Fn monomers stabilized through intermonomer ionic interactions [7,30], monomeric labeled Fn-DA was incubated with monomeric Fn-u, since artificial dimerization of one monomer of Fn-u with another monomer of Fn-DA should result in a loss of IA/ID relative to a solution of pure Fn-DA in PBS. Incubation of monomeric Fn-DA with increasing concentrations of Fn-u in PBS, but not in 1 M GdnHCl, where dimerization is inhibited by denaturant, resulted in a gradual decrease in the solution IA/ID (data not shown), suggesting dimerization of reduced monomers affected measurements in PBS. Thus, energy transfer with cys/cys Fn-DA only occurs between dimer arms, and cys/cys Fn-DA is well suited to probe for the overlap of Fn arms or the compact, folded quaternary structure of Fn in ECM fibrils.
Circular dichroism (CD) spectra of cys/cys Fn-DA at 228 nm—an indication of β-sheet content—was measured for Fn-DA to determine whether labeling of Fn with Alexa dyes prevented refolding of Fn after labeling (Figure 2B). The mean residue ellipticity of cyc/cys Fn-DA indicated that Fn-DA refolding was not impacted by fluorophore labeling within experimental error (Figure 2B): measurements in 0 M GdnHCl were not statistically different from measurements on Fn-u and were similar to previously published data [24,31,32].
Amine/cys Fn-DA was produced through site-specific labeling of Alexa 546 on III7 and III15 and random labeling of Alexa 488 on amines. The number of donors per Fn dimer must be large enough so that a randomly attached donor is present within ∼12 nm of III7 or III15—a requirement for intramonomer energy transfer. However, excess donors increase background fluorescence from donors that do not form FRET pairs and may cause donor fluorescence self-quenching [33]. Control experiments using donor-labeled Fn showed that only donor-labeled Fn with more than ∼10 donors per Fn dimer had an increased fluorescent yield in denaturant relative to PBS, which indicates fluorescence self-quenching (data not shown). For these reasons, we used amine/cys Fn-DA with 3.9 acceptors and 7.0 donors generated with 30- and 65-fold molar excesses of acceptors and donors, respectively.
The compact quaternary conformation of amine/cys Fn-DA in PBS revealed strong energy transfer (IA/ID = 0.85; Figure 2C). The extended conformation in 1 M GdnHCl had reduced energy transfer (IA/ID = 0.61), while partial unfolding of the extended structure in 2 M GdnHCl reduced IA/ID further to 0.56. Significant unfolding of amine/cys Fn-DA in 4 M GdnHCl led to a further reduction to 0.47, indicating amine/cys Fn-DA retains intramonomer energy transfer in 1 and 2 M GdnHCl and may be used to differentiate extended Fn (Figure 1F) from unfolded Fn (Figure 1G). Monomeric amine/cys Fn-DA showed reduced energy transfer in PBS (IA/ID = 0.65) and 1 (IA/ID = 0.55), 2 (IA/ID = 0.53), and 4 M GdnHCl (IA/ID = 0.44) relative to dimeric Fn-DA. Finally, CD spectra of amine/cys Fn-DA at 228 nm indicated that amine/cys Fn-DA refolds after labeling (Figure 2D), and Fn-DA in 4 M GdnHCl was unfolded (Figure 2D). Thus, amine/cys Fn-DA can be used to differentiate between extended and unfolded conformations of Fn.
In order to estimate the IA/ID value below which unfolding occurs, FRET from Fn-DA denatured in solution was compared to the corresponding CD spectra (Figure 2). While the first loss of secondary structure is seen in CD above 1 M GdnHCl and the protein is almost completely unfolded in 4 M GdnHCl, FRET differences remain between the monomeric and dimeric amine/cys Fn-DA solutions even up to 4 M GdnHCl. This is most likely due to occasional crossover of dimer arms during free diffusion in solution, an effect that is far less pronounced in cys/cys Fn-DA solutions because of the reduced number of fluorophores distributed along the Fn arms (the similar values for monomeric and dimeric cys/cys Fn-DA also indicate that the difference in monomeric and dimeric amine/cys Fn-DA in 1 M GdnHCl was not due to DTT but reflected fluorophore separation). Therefore, IA/ID for monomeric amine/cys Fn-DA in 1 M GdnHCl (IA/ID = 0.55) was used here to indicate the onset of module unfolding. The conclusions drawn in this study, however, are not altered in a major way if slight differences exist between the IA/ID values at which secondary structure is lost by chemical denaturation or mechanical unfolding. The data indicate that the unfolding point within fibrils might occur between the IA/ID value taken at 1 or 2 M GdnHCl since only a slight drop in IA/ID was noted between monomeric amine/cys Fn-DA in 1 (IA/ID = 0.55) and 2 M GdnHCl (IA/ID = 0.53).
To determine whether the compact quaternary and unfolded conformations of Fn are present in ECM, cys/cys or amine/cys Fn-DA was next added to a culture medium of human fibroblasts that readily incorporate exogenously added soluble Fn into their matrix fibrils [9,10,34]. At a seeding density of 20 × 103 cells/cm2, fibroblast ECM was three-dimensional after 24 h and ranged from ∼4 to 10 μm in thickness. Cells were either fixed in 3% formaldehyde and imaged under PBS [10] or imaged live under PBS. Data were similar between fixed and live samples (data not shown). Photobleaching of fluorophores was routinely tested by acquiring consecutive images of a random field of view and plotting histograms of the differences in acceptor (Figure S1C and S1D) and donor intensities (Figure S1E and S1F) on a pixel by pixel basis between consecutive images. Finally, the ratio of amine/cys or cys/cys Fn-DA to Fn-u was titrated over a wide range to determine the percentage of Fn-DA above which intermolecular FRET occurred. This analysis demonstrated that 10% amine/cys Fn-DA is not contaminated by intermolecular FRET (Figure S1G). However, greater than 10% amine/cys Fn-DA led to progressively increased mean IA/ID values in fibrils. In contrast, mean IA/ID measurements in ECM fibers using cys/cys Fn-DA were not dependent upon the ratio of cys/cys Fn-DA to Fn-u at ratios below 50% Fn-DA. This is consistent with the reduced number of fluorophores with this approach (∼4) relative to amine/cys Fn-DA (∼11).
To assess the extent to which IA/ID measurements represent conformational variability versus instrument noise, FRET histograms of ECM were compared to those taken from Fn-DA in solution (Figure 3A). The width of histograms derived from solution measurements could be reduced without affecting population means by using higher concentrations of Fn-DA, higher excitation power, or multiple acquisitions and subsequent averaging of each pixel, indicating that noise is the predominant contributor to the width of histograms from solution measurements. Noise contributes significantly to the variability of individual ratiometric measurements as signals in the numerator (IA) and denominator (ID) are acquired by separate PMTs (see Text S1 and Figure S2). Figure 3 and Video S1 show a histogram (Figure 3B) for all pixels of amine/cys Fn-DA–containing ECM in a fibroblast culture acquired 1 μm above the cell–glass interface (Figure 3C), with an overlay of IA/ID on the DIC image (Figure 3D). The 80-μm Fn fiber at the interface between two cells in the confluent culture transitions from an extended conformation at the upper left edge (Figure 3D; purple outline) through a region with unfolded Fn (Figure 3E; pink outline). While noise contributes significantly to the broadening of FRET data, the different colors seen in the ECM images in every field of view are not just due to instrument noise. Individual fibers with higher or lower mean IA/ID values can be distinguished, and their histograms are narrower than the histogram derived from the entire field of view (Figure 3D and 3E; each region >100 pixels), thereby demonstrating that different Fn conformations do coexist.
To test for the presence of the fully compact solution structure of Fn, cys/cys Fn-DA was incorporated into a 24-h fibroblast ECM. Figure S3A shows a ratiometric confocal slice 4 μm above the glass–ECM interface. Cell extraction with Triton X-100 (Figure S3B) or alternatively Rho kinase (ROCK) inhibition with Y-27632 (Figure S3C) were used to inhibit cellular contractile forces, both causing matrix relaxation [10,20]. Histograms for all pixels in three random fields of view each from three separate experiments are plotted in Figure S3D. Some regions of cyc/cys Fn-DA matrix were composed of Fn with some partial backfolding of the monomeric arms upon themselves since IA/ID values were higher than that measured for the extended structure but lower than that measured for the fully compact solution structure (see Discussion). These data confirm that the compact solution structure of Fn was never achieved in fibrillar Fn even after release of cell traction forces.
To test for the presence of unfolded Fn, amine/cys Fn-DA was next incorporated into a 24-h fibroblast matrix. Figure 4A shows a representative ratiometric image acquired 4 μm above the glass–ECM interface. On average, Fn was in the extended conformation in fibroblast culture (IA/ID = 0.59 ± 0.14); however, unfolded Fn fibers with mean IA/ID < 0.55 were present within all imaged confocal slices. These regions were not randomly dispersed, but instead whole fibrils encompassing at least 50 pixels were present (Figure 4A, white arrow). Cell extraction (Figure 4B) or inhibition of ROCK (Figure 4C) resulted in an increase in IA/ID and a decrease in the presence of unfolded pixels (IA/ID < 0.55). Histograms generated from three random fields of view, each from three separate experiments (Figure 4D), showed that loss of cell contractile force leads to refolding of Fn within matrix fibrils. Finally, only fewer than 1% of pixels had IA/ID values representative of the fully compact solution conformation of Fn. In addition, whole fibrils consisting of at least 50 pixels were never measured with mean IA/ID > 0.75 (data not shown). Since each pixel measurement represents an average over a population of molecules, multiple and distinct conformations may coexist within a single pixel. If a fraction of Fn-DA molecules within each pixel is in the fully compact conformation, they would have to be offset by another fraction of molecules with IA/ID significantly lower than the overall mean value. However, some histograms for all pixels within regions of interest containing individual ECM fibers had both width and median values similar to those measured for Fn-DA in 4 M GdnHCl (Figure 3B and 3F). If these fibers were in fact heterogeneous populations, a small portion of Fn-DA that was not unfolded would have to be offset by Fn-DA that had lower IA/ID than that measured in 4 M GdnHCl. This is unlikely since Fn-DA is almost completely unfolded in 4 M GdnHCl, and these distributions are indicative of a population of Fn molecules with conformational homogeneity. Finally, accumulated FRET between several donors and acceptors cannot differentiate between partial unfolding into structural intermediates and complete unfolding of FnIII modules. However, more than twice as many partially unfolded FnIII modules as completely unfolded modules would be needed to account for a given increase in end-to-end length. Partial unfolding into an intermediate conformation where two β-strands are separated would only increase the module end-to-end length by ∼1/3 of the increase due to complete unfolding [13,14,16,18].
Contractility inhibition studies showed that matrix refolding occurs after loss of cell generated forces (Figure 4); however, the relationship between FRET changes and changes in fibril length were unknown. We therefore designed a strain device (see also Text S1 and Figure S4) to stretch denuded Fn matrix fibrils attached to a stretchable substrate, which allowed us to calibrate the relationship between mechanical strain and FRET. Schematic diagrams of the strain device before and after strain application are shown in Figure 5A and 5B. Figure 5C and 5D show representative ratiometric images of denuded amine/cys Fn-DA matrix on unstretched poly(dimethylsiloxane) (PDMS) and on PDMS after a 1.7-fold increase in length (70% elongation, 28% transverse shortening), respectively. IA/ID values for individual amine/cys Fn-DA–containing fibrils were measured for up to six fibrils per field of view from 26 fields of view in six experiments, and the strain was calculated based on their angular orientation with respect to the strain axis. The individual fibril mean IA/ID values are plotted versus the calculated strain in Figure 5E. Relative strain of individual fibrils ranged from relaxation to 3/5 the starting length (−40% strain) to a 1.7-fold stretch (73% strain). Cell-free control fibrils measured prior to stretch or relaxation (IA/ID = 0.62 ± 0.03) were composed mainly of Fn in the extended conformation, but ranged from unfolded (IA/ID lower limit = 0.54) to extended with some backfolding of the monomeric arms (IA/ID upper limit = 0.69). Therefore, denuded Fn matrix assembled on PDMS was similar in conformation to denuded Fn matrix on Fn-coated glass. Mean IA/ID dropped to 0.59 ± 0.04 for fibrils stretched by more than 1.4-fold (>41% strain), while relaxation to less than 4/5 the starting length (−21% strain) resulted in a statistically significant increase in IA/ID to 0.65 ± 0.04. Thus, the greater than 2-fold length change required to traverse from −21% to 60% strain resulted in a shift in mean fibril IA/ID of only ∼0.07, a difference that is less than the total range of IA/ID (0.54 to 0.69) measured for control cell-free fibrils prior to stretch or relaxation. From these data we conclude that matrix fibrils in a typical field of view containing ∼15–20 confluent cells are heterogeneously stretched by a more than 2-fold difference in length and that ECM strain covers a wider range than was previously appreciated. Control experiments using cys/cys Fn-DA did not show a strain response, even if stretched by up to 75% (Figure S5), indicating that cys/cys Fn-DA is insensitive to differentiating between extended and unfolded Fn.
To test the primary assumption of the quaternary structure model (Figure 1D), which proposes that Fn assumes a compact (solution) conformation in the resting state of fibrils [12], fibroblast matrix composed of cys/cys Fn-DA (Figure 6A–6C) or amine/cys Fn-DA (Figure 6D–6G) was grown on Fn-u adsorbed onto PDMS sheets. Without covalent attachment of Fn-u to PDMS, two methods could be used to partially detach the denuded Fn matrix from the substrate. We found that 10%–20% strain resulted in detachment of large regions of de-cellularized ECM (Figure S6). To circumvent the straining step, fibroblast matrix was also assembled on Fn-u adsorbed to prestretched PDMS and partially detached by relaxation of the denuded matrix to from 4/5 (Figure 6A–6C; 3.7% transverse stretch) to 3/5 (Figure 6D–6G; 10% transverse stretch) the starting length. Figure 6A shows a ratiometric image just above the PDMS substrate, while Figure 6B shows an image from the same field of view but acquired 3 μm above the substrate. This relaxed Fn mat was loosely attached to the matrix shown in Figure 6A, but it randomly diffused in the field of view around its points of attachment. Both detachment schemes resulted in IA/ID means for fully relaxed fibrils containing cys/cys Fn-DA (Figure 6C) or amine/cys Fn-DA (Figure 6G) that varied from 0.48 to 0.58 and 0.68 to 0.75, respectively. The percentage of fully compact pixels never exceeded 2% using cys/cys Fn-DA, while at most 6% of pixels were fully compact using amine/cys Fn-DA. These data show that complete relaxation of cell-free fibers leads to a complete loss of unfolded Fn and a complete absence of fully compact Fn. Therefore, we conclude that extension of Fn fibers by cell traction forces occurs through both straightening of partially backfolded monomer arms (residual quaternary structure) (Figure 1E; see also Discussion) and unfolding of FnIII modules (Figure 1G). Moreover, unfolded conformations, extended molecules with intact tertiary structure, and extended conformations with some residual quaternary structure dynamically coexist within the ECM.
The ECM contains numerous super-molecular fibrillar assemblies of proteins that display a remarkable range of end-to-end extensions under physiologically relevant forces. Indirect evidence from single-molecule atomic force microscopy [17] and cryptic epitope exposure detected by monoclonal antibody binding [35] suggested that Fn fibers in the ECM are unfolded by cell traction forces; however, this possibility was recently debated [8–12,20]. We used two different FRET labeling schemes to test for the presence of the compact solution conformation in fibrils, resulting from dimer arm overlap, and the unfolded conformation of Fn within ECM (Figure 1). Intramolecular energy transfer using cys/cys Fn-DA was shown in solution measurements to be sensitive mainly to dimer arm overlap (Figure 2). In contrast to cys/cys Fn-DA, amine/cys Fn-DA is also sensitive to both the extended and unfolded conformations of Fn (Figure 2), thus confirming earlier findings [9,10,24]. FRET from both labeling schemes was used to demonstrate that the compact solution conformation of Fn is not present within fibroblast matrix in the presence of cell contractile forces, after inhibition of cell contractility (Figures 3, 4, and S3), or even in strain-free fully relaxed Fn fibrils (Figures 6 and S6). Partially unfolded Fn, however, is present in ECM fibers of live fibroblast cultures, and some FnIII modules refold after blocking cell contractility (Figure 4). Unfolded Fn is absent in fully relaxed fibers (Figures 6 and S6).
Since Fn contains more than 54 domains and is ∼130 nm long in the extended conformation without unfolding of its FnIII modules [36], a 3- to 4-fold change in end-to-end length could theoretically occur through either quaternary structural change (Figure 1D–1F) or secondary/tertiary structural change (Figure 1F and 1G). Erickson and collaborators showed that Fn-GFP chimeric proteins embedded within ECM fibrils retained constant levels of fluorescence after fibrils were cut [20], while single-molecule force measurements revealed that individual GFP and FnIII modules (Figure 7A) were similarly resistant to mechanical unfolding [12]. In consideration of this experimental evidence together with the FRET data presented here that FnIII modules are indeed unfolded within fibroblast ECM, we conclude that Fn-GFP embedded within tightly packed Fn fibrils has an increased mechanical stability compared to single-molecule Fn-GFP as used for the atomic force microscope studies for the following reasons. While FnIII modules are ellipsoidal in shape with a 2- to 2.5-nm cross-sectional diameter, GFP is a significantly bigger molecule. It forms a cylindrical 4-nm-long β-barrel that protrudes into the space surrounding the string of FnIII modules and might distort the otherwise parallel alignment of Fn molecules within fibrils. The lateral pressure imposed by the adjacent Fn molecules on the GFP barrel if mechanically strained is likely to reduce its freedom of motion, and by minimizing local distortions, its tilt angle with respect to the strain axis might be changed as well (Figure 7B). These and other factors might contribute to the possibility that GFP embedded in densely packed fibers submitted to tensile force unfolds along a different unfolding pathway. Although Fn-GFP fibers tested by Erickson and colleagues contracted up to 1/4 their starting length once cut [11,20], we have recently determined that manually deposited Fn fibrils can be extended 5- to 6-fold from their resting state before they begin to break (W. C. L., M. L. S., U. Ebneter, V. V., unpublished data). Accordingly, we cannot exclude that fibrillar GFP-Fn might still lose its fluorescence if fibrils are stretched more than 4-fold.
Although FRET data showed that the compact solution conformation of Fn does not exist in ECM fibers, cys/cys Fn-DA from fully relaxed fibrils indicates some minor quaternary structure that disappears in strained fibrils. Bending in hinge regions, for instance between modules I1–5 and III2 [37] or III12–14 [38], could permit backfolding of a monomer arm upon itself, thus bringing otherwise distant Fn segments into close proximity and increasing energy transfer. Although the FRET approach presented here is not capable of differentiating between these different structures, high-resolution cryo-scanning electron microscopic images of cell-derived Fn nanofibrils as small as 5 nm in diameter were either straight or contained nodules 10–15 nm in diameter [39,40]. Smooth nanofibrils were attached at each end, while fibers decorated with nodules had one free end and thus more likely represented fully relaxed fibrils. In addition, fibrils that were free at one end were as narrow as 5 nm in diameter, while fully relaxed fibers consisting of fully compact Fn should have a diameter of ∼35 nm. From these cryo-scanning electron microscopy data we speculate that a partially backfolded structure, as sketched in Figure 1E with artistic freedom, may represent the fully relaxed conformation of Fn in ECM. However, the structure of Fn in fully relaxed fibers as well as the locations and extent of crosslinking and amount of axial offset between adjacent Fn molecules in fibrils remain to be determined.
Finally, unfolding of FnIII modules occurs through rupture of backbone hydrogen bonds, leading to a peeling away of the β-strands (for review see [2]). Therefore, Fn extension is not a continuous response to tension, but rather an energy-dissipating process with major hysteresis between the unfolding and refolding trajectories [17]. The word “extensibility” was thus used here to describe Fn elongation in place of “elasticity.”
FRET is sensitive to unfolding of modules within less than 12 nm of III7 or III15. Therefore, partial or full elimination of FRET by tensile force indicates unfolding within modules III13–14, IIIV (present on one monomer of plasma Fn), III5–6, or III8–9, with possible minor contributions to energy transfer from FnIII modules at the Förster limit (III4, III10, and III12). Based on experiments done with the strain device, the conformational differences seen in a random field of view correspond to a roughly 2-fold difference in Fn end-to-end extension. Each completely unfolded FnIII module would lengthen from ∼3.2 nm to an extended contour length of ∼28.5 nm [17], and an extended Fn molecule estimated to be 130 nm in length [36] could double in length by complete unfolding of only ∼5–6 FnIII modules per Fn molecule.
Not only does Fn end-to-end extensibility contribute to the mechanical resiliency of tissues, but strain could furthermore alter Fn function through unfolding of FnIII modules into partially unfolded intermediate conformations with deactivated or newly exposed molecular recognition sites [1,2]. For instance, steered molecular dynamics simulations predicted that a major intermediate state exists for FnIII1 or III2 along the force-induced unfolding trajectory [14]. By peeling off two β-strands, force unmasks an analog of anastellin, a proteolytic fragment of module III1, that as an isolated peptide inhibits metastasis and angiogenesis but promotes fibrillogenesis [41,42]. Steered molecular dynamics simulations also showed that an increase in the distance between the loop-exposed integrin-binding RGD sequence on module III10 and the partner PHSRN synergy site on III9, which favors binding of the integrin α5β1 over αvβ3 [43,44], occurs prior to unfolding of the respective FnIII modules. However, these and other hypotheses that state that the ECM might be involved in mechanotransduction processes are only physiologically relevant if Fn fibrils partially unfold. In addition, molecular recognition sites might also be buried within the nodules and become exposed upon their disappearance. It is thus important to note that the conformational distribution across each field of view within a living cell culture was rather heterogeneous at the length scale of cells and always included a fraction of fibers or fibrillar sections that presented IA/ID values indicative of partially unfolded Fn. Living cells are thus surrounded by and potentially respond to a broad range of fibrillar Fn conformations. Future research is now needed to elucidate the hypothesized physiological importance of partially unfolded Fn. It also remains to be clarified whether module unfolding plays a role in the force-induced fibrillogenesis of Fn and thus only occurs heterogeneously within unfolded regions of the ECM [35,41,45–47].
Force has a pervasive influence on the behavior of tissues, cell aggregates, and individual cells. However, the molecular motifs that convert force into biochemical signals thus permitting force sensation are only now emerging. The altered functions of non-equilibrium structures of proteins resulting from mechanical tension are beginning to be appreciated as essential to cell function inside the cell cytoplasm, as was recently shown for the signaling molecule p130Cas [48], and in the cell membrane, for instance as a regulator of ion channel function [49]. Here, we show that the extracellular protein Fn is stretched by cell traction forces both from straightening of an initially nodular assembly with quaternary structure and unfolding of FnIII modules. Outside-in cell signaling might thus be regulated not only by the composition and rigidity of the matrix, but also by the extent to which it is unfolded. These results contribute to our understanding of Fn's remarkable mechanical properties and serve as an impetus for future elucidation of the biological importance of mechanically regulated cryptic epitope exposure on Fn.
Fn isolated from human plasma (see Text S1) was doubly labeled with Alexa 488 and Alexa 546 (Molecular Probes, http://probes.invitrogen.com/) as FRET donors and acceptors, respectively, using two different labeling schemes based on established protocols [24,25]. To produce cys/cys Fn-DA, Fn was site-specifically labeled exclusively on buried cysteines within modules III7 and III15 of each dimer arm. Isolated plasma Fn at ∼1 g/l in PBS was denatured in an equal volume of 8 M GdnHCl and incubated for 2 h with a 15-fold molar excess each of Alexa 546 maleimide and Alexa 488 maleimide. Fn-DA was separated from free dye by size exclusion chromatography (PD-10 Sephadex, Amersham, http://www.amersham.com/). Western blot analysis was used to determine the point at which Fn fragments contaminated the eluent (data not shown). Fn was also labeled to produce amine/cys Fn-DA with donors and acceptors located on random amines and cysteines, respectively. Fn at ∼1 g/l in PBS was denatured in an equal volume of 8 M GdnHCl and incubated for 1 h with a 30-fold excess of Alexa 546 maleimide. Acceptor-labeled Fn was separated from free dye by overnight dialysis (Slide-a-lyzer dialysis cassette, 10,000 MW cutoff; Pierce, http://www.piercenet.com/) with three changes of amine labeling buffer (PBS with 0.1 M NaHCO3 [pH 8.5]). Subsequently acceptor-labeled Fn was incubated with a 70-fold excess of Alexa 488 succinimidyl ester for 1 h. Fn-DA was separated from free dye using a PD-10 column equilibrated with PBS. The labeling ratio of donors to acceptors per Fn dimer was determined by measuring the absorbances of Fn-DA at 280, 496, and 556 nm and using published extinction coefficients for dyes and Fn. Fn-DA was stored with 10% glycerol at −20 °C until needed and used within 5 d of thawing and storage at 4 °C.
IA/ID was measured for monomeric or dimeric Fn-DA freely diffusing in solution in different concentrations of denaturant using the same microscopic setup. Small chambers were made by cleaning a 0.17-mm-thick coverslip, 1 × 6 cm PDMS spacers cut from 0.25-mm-thick sheets (Specialty Manufacturing, http://www.specmfg.com/), and a microscope slide in 2% PCC-54 (Sigma-Aldrich, http://www.sigmaaldrich.com/), 70% ethanol, and finally deionized water. Parts were dried, and PDMS strips were placed on the microscope slide to make small channels. The coverslip and the exposed glass between the PDMS spacers were coated with 4% BSA (Sigma-Aldrich) for 1 h to prevent Fn-DA binding to chamber surfaces. BSA-coated surfaces were rinsed with water and dried under filtered air. Finally, the coverslip was placed onto the PMDS strips to make parallel channels between the coverslip and microscope slide. Fn-DA at 0.1 to 0.2 g/l with or without 0 to 4 M GdnHCl or 50 mM DTT in PBS (Sigma-Aldrich) was then drawn into the chambers by capillary forces. Fn-DA in solution within the channels was imaged through the coverslip.
CD spectra were measured on a Jasco (http://www.jascoinc.com/) model J-715 spectropolarimeter with temperature control using Fn-DA at a concentration c that varied from 0.1 to 0.2 g/l. Mean residue ellipticity, [θ], presented in units of degrees × square centimeters/decimole, was calculated from the observed ellipticity, θobs, measured in a cell with path length l by
where a value of 108 was used for the mean residue molecular weight, MRW. Measurements of Fn-DA were made in duplicate, while measurements of Fn-u were made in triplicate at 20 °C.
Primary human dermal fibroblast cells derived from foreskins (PromoCell, http://www.promocell.com/) were maintained for less than eight passages in Fibroblast Growth Medium plus Supplement (PromoCell). For IA/ID measurements, the growth medium was switched to Dulbecco's modified Eagle's medium (DMEM) plus 10% newborn calf serum (NCS; Invitrogen, http://www.invitrogen.com/), since high serum was found to shorten the culture time necessary for production of a three-dimensional matrix (data not shown). Eight-well Lab-Tek chambers (Nalgene Nunc, http://www.nalgenunc.com/) were coated with 0.03 g/l Fn-u in PBS for 1 h prior to cell seeding at 20 × 103 cells/cm2. Cells were allowed to adhere for 30 min, after which the medium and unbound cells were removed and replaced with DMEM plus 10% NCS, 0.005 g/l Fn-DA, and 0.045 g/l Fn-u. Cells were incubated for 24 h before imaging. Samples to be fixed were first washed two times with warm PBS and then treated with 3% formaldehyde for 20 min. Samples were imaged under PBS.
A custom strain device was manufactured from stainless steel (Figure S3), and cells were grown on modified stretchable PDMS surfaces. PDMS sheets (0.25 mm thick), cut into 5 by 1.7 cm rectangles, and PDMS rings, 2 mm thick with 1.0-cm inner and 1.6-cm outer diameters, were cleaned in PCC-54, ethanol, and finally water. After drying, PDMS rings were placed in the middle of the PDMS sheets, and the PDMS was treated with air plasma (PDC-32G, Harrick Scientific, http://www.harricksci.com/) at 250 mbar for 30 s to render the surface hydrophilic. For noncovalent attachment of Fn-u to the surface, plasma-treated surfaces were used. For covalent attachment of Fn-u, amino groups were attached to the plasma-treated surface with 3-aminopropyltriethoxysilane vapor in an evacuated chamber for 1 h. Vapor-based silanization resulted in significantly lower surface fluorescence than solution-based silanization (data not shown), as has been previously reported [50]. After silanization, 0.125% glutaraldehyde (Sigma-Aldrich) was added and the sample incubated for 20 min. Finally, the surface was rinsed with water and dried prior to assembly of the strain device.
All metal strain device parts were autoclaved prior to assembly within a tissue culture hood. PDMS sheets were clamped into the strain device, and the culture surface within the PDMS ring was incubated with 0.03 g/l Fn-u for 1 h. Unreacted glutaraldehyde was quenched by incubation with DMEM plus 10% NCS for 20 min at 37 °C. Then 20 × 103 fibroblast cells/cm2 were seeded onto the PDMS surface and allowed to adhere for 30 min prior to removal of unbound cells and replacement with DMEM plus 10% NCS, 0.005 g/l Fn-DA, and 0.045 g/l Fn-u. After 24 h, cells were washed two times in 37 °C PBS and extracted, leaving behind the detergent-insoluble Fn matrix, with a 5-min incubation in 0.5% Triton X-100 plus 20 mM NH4OH in PBS (pH ∼9.7). The strain device with attached cell-free matrix was then imaged. Three separate fibril detachment experiments were performed with both Fn labeling approaches, while four and six separate stretching experiments were performed with amine/cys Fn-DA or cys/cys Fn-DA, respectively, on Fn-u that was covalently attached to the membrane. During strain application, it was found that mounting the PDMS ring on the PDMS membrane prior to treatment with air plasma most often resulted in the PDMS ring sliding on the membrane surface during stretching. Strain samples were discarded if the ring stuck to the membrane, thus leading to inhomogeneous strain application, or leaked culture medium.
All FRET images were acquired from living, fixed, or denuded acellular samples in LabTek chambers or in the strain device. All images were acquired with an Olympus (http://www.olympus-global.com/) FV1000 confocal microscope with an oil immersion 1.35NA 60× objective. Emitted light from the sample was split with a 50/50 beam splitter, and it was detected with two PMTs. Spectral information was determined using a diffraction grating and slit. Acceptor and donor intensities were detected using 12-nm bandwidths across the donor (514–526 nm) and acceptor (566–578 nm) emission peaks. Images were generally acquired at 512 × 512 pixel resolution for a 212 × 212 μm field of view with a 200-μm pinhole diameter. Acquisition parameters including laser transmissivity and pixel dwell time were adjusted to prevent photobleaching while maximizing detection sensitivity, and PMT voltages were constant at 600 V. Donor, acceptor, and DIC transmission images were taken from multiple regions of each sample. PMT dark current background values were acquired every 30 min.
All 16-bit images (4,096 relative intensity units) were processed with Matlab (The MathWorks, http://www.mathworks.com/). Images were averaged with 2 × 2 pixel sliding blocks, mean PMT dark current background values were subtracted from donor and acceptor images, and donor images were corrected for light attenuation from the 50/50 beam splitter with a multiplication factor of 1.09. A threshold mask of 100 relative intensity units was applied to both images, and the acceptor image was divided pixel by pixel by the donor for all pixels above threshold intensity values in both channels. Histograms were computed from all data pixels within each field of view using bin widths of 0.01 intensity ratio units, and IA/ID was color-coded within the range of 0.05 to 1.0. Photobleaching of samples was routinely tested by acquiring two or more consecutive images in both the donor and acceptor channels. Consecutive images were averaged and background subtracted. Next, intensity values from the second image acquisition were subtracted from intensity values from the first acquisition on a pixel by pixel basis, and a histogram was generated for the resultant differential intensity image. Data were not included if histograms showed a greater than 5% mean drop in either donor or acceptor differential intensity histograms (as in Figure S1C and S1D).
In some cases, individual fibril mean IA/ID values were calculated for those fibrils that met specific criteria using region of interest analysis. All regions of interest were chosen from donor images where IA/ID was not shown to limit experimenter bias. Fibrils had to be straight without intermediate branch points and have a length to width ratio of at least five, to increase the probability that the central axis of the fibril was a correct indication of the direction of strain. Fibrils had to be at least 15 μm in length and 50 pixels in size (∼8 μm2 surface area within the confocal slice) to increase accuracy of the mean intensity ratio estimate. Straight fibrils were considered only if they were connected on each end by larger Fn structures such as branch points or sheet-like aggregates to increase the probability that each end of the fibril was connected to the PDMS substrate. If more than six fibrils in a field of view met these criteria, the six largest fibrils were chosen. The region of interest was drawn around the straight section of fibrils, and the mean plus standard deviation of the intensity ratio; the fibril length, l1; the total pixel number; and the orientation angle of the fibril relative to the direction of strain application, θ1, were recorded. The macroscopic strain applied to the PDMS membrane, ε, was calculated by measuring the length between PDMS clamp points in the strain device before and after application of strain or relaxation. The transverse compression due to anisotropy of the strain field, εT, was measured macroscopically in each experiment by measuring the width of the stretchable membrane beneath the region of cell culture. Hence, the starting length of each individual fibril, l0, was calculated by
Finally, the estimated strain for each fibril, εfibril, was calculated by
|
10.1371/journal.pgen.1001388 | SHINE Transcription Factors Act Redundantly to Pattern the Archetypal Surface of Arabidopsis Flower Organs | Floral organs display tremendous variation in their exterior that is essential for organogenesis and the interaction with the environment. This diversity in surface characteristics is largely dependent on the composition and structure of their coating cuticular layer. To date, mechanisms of flower organ initiation and identity have been studied extensively, while little is known regarding the regulation of flower organs surface formation, cuticle composition, and its developmental significance. Using a synthetic microRNA approach to simultaneously silence the three SHINE (SHN) clade members, we revealed that these transcription factors act redundantly to shape the surface and morphology of Arabidopsis flowers. It appears that SHNs regulate floral organs' epidermal cell elongation and decoration with nanoridges, particularly in petals. Reduced activity of SHN transcription factors results in floral organs' fusion and earlier abscission that is accompanied by a decrease in cutin load and modified cell wall properties. SHN transcription factors possess target genes within four cutin- and suberin-associated protein families including, CYP86A cytochrome P450s, fatty acyl-CoA reductases, GSDL-motif lipases, and BODYGUARD1-like proteins. The results suggest that alongside controlling cuticular lipids metabolism, SHNs act to modify the epidermis cell wall through altering pectin metabolism and structural proteins. We also provide evidence that surface formation in petals and other floral organs during their growth and elongation or in abscission and dehiscence through SHNs is partially mediated by gibberellin and the DELLA signaling cascade. This study therefore demonstrates the need for a defined composition and structure of the cuticle and cell wall in order to form the archetypal features of floral organs surfaces and control their cell-to-cell separation processes. Furthermore, it will promote future investigation into the relation between the regulation of organ surface patterning and the broader control of flower development and biological functions.
| The cuticular layer that covers all aerial parts of plants plays a vital role not only in the interaction with environment but also in plant development and growth. Despite the recent significant achievements in the identification of structural genes involved in cuticle biosynthesis and secretion, little is known regarding the regulation of metabolic pathways generating cuticular constituents, more specifically wax and cutin. The Arabidopsis AP2-type transcription factor SHINE1/WAX INDUCER1 (SHN1/WIN1) was the first assigned regulator of a cuticle-related metabolic pathway; nevertheless, its mode of action and biological function remain uncertain due to redundancy with two additional clade members. Here, by co-silencing all three SHN clade members using an artificial microRNAs approach, we demonstrated that SHN transcription factors act redundantly in patterning reproductive organ surface, modulating processes associated with cell elongation, adhesion, and separation, which secure the proper function of these organs. It appears that SHN transcription factors act directly on downstream cutin and cell wall–modifying genes. These factors are likely part of the genetic network controlling floral organ development. Thus, SHN transcription factors link together cuticle assembly, cell wall remodeling, and flower development to form the archetypal surface of floral organs mediating plant reproduction through pollination and seed dispersal.
| In contrast to other plant cell layers, the epidermis develops a unique cell wall that not merely constitutes of cellulose, hemicelluloses, pectins, and proteins but also of a cuticular matrix, which is largely composed of cutin embedded and overlaid with waxes [1]. Cutin, an insoluble cuticular polymer, is largely composed of interesterified hydroxy and hydroxy epoxy fatty acids and is attached to the outer epidermal layer of cells by a pectinaceous layer [2]. As the epidermal cell grows, the cuticle merges gradually with the cell wall components [3]. Although the role of the epidermis layer in regulating organ growth has remained controversial [4]–[5], it is clear that it is vital for plant survival, development and the interaction with the environment [6]–[7]. Cutin and wax are synthesized exclusively in the epidermis [8] and a massive flux of lipids occurs from the sites of lipid synthesis in the plastid and the endoplasmic reticulum (ER) to the plant surface during cuticle deposition [9]. Significant progress has been made over the past decade in identifying genes involved in the biosynthesis and secretion of cuticular lipids [10]–[11] and in the metabolism and assembly of primary cell wall components [12]–[14]. Despite the close connection between the cell wall and the cuticular matrix, mutants and phenotypes in one of these processes were rarely examined for alteration in the other. Furthermore, to our knowledge, co-regulation of these two processes at the molecular genetic level was overlooked up to now.
Biosynthesis of plant cuticle components and their secretion to the extracellular matrix involve the coordinated induction of several metabolic pathways, in which transcription factors may play a key role [9], [15]. The Arabidopsis SHINE1/WAX INDUCER1 (SHN1/WIN1) AP2-domain protein was the first transcription factor reported to control metabolic pathways generating cuticular waxes [16]–[17]. A subsequent study [18] indicated that SHN1/WIN1 controls cuticle permeability by regulating the expression of cutin biosynthesis genes, particularly LACS2 (LONG CHAIN ACYL-COA SYNTHETASE 2). The induction of wax formation in leaves by over expression of individual SHINE clade genes was suggested to be a second step, possibly an indirect process following cutin biosynthesis [18]. Nevertheless, our current knowledge is limited with respect to the SHN1/WIN1 protein's mode of action and the involvement in particular developmental processes.
Arabidopsis SHN1/WIN1 transcription factor belongs to a small distinct clade of three proteins [16]. They all share two unique conserved motifs outside the AP2 domain, and all three proteins display the same shiny phenotype upon overexpression, suggesting their functional redundancy in cuticular lipid biosynthesis. Additional evidence for functional redundancy among the SHN clade members in cuticular lipid biosynthesis was provided by silencing SHN1/WIN1 [18]. In these plants, floral morphology was not altered and the subtle reduction in the levels of cutin detected in entire flower extracts was enhanced in isolated petals. Besides, their notable expression patterns in reproductive organs suggested that they are probably redundant in function. The expression of SHN1/WIN1 and SHN3 overlapped in various flower organs including in the abscission zones while SHN2 and SHN3 were both expressed in the silique dehiscence zones. Interestingly, expression of SHN2 was very specific to cell separation regions in the anthers and siliques. These expression profiles indicated that SHN transcription factors may also act in a combinatorial manner to secure reproductive organ development, protecting the exterior layers of the plants from environmental stresses. On the other hand, these three clade members differ in their spatial and temporal expression patterns, which suggests that each of them may play specific roles in various organs or under different conditions, and that the actual redundancy between the SHN factors is most probably in their target genes [16]. Further elucidation of the mode of SHN action, their target genes, and their precise connection to plant cuticle formation and plant development requires in-depth characterization of the SHN clade factors, which can be achieved by using double, possibly triple mutants to eliminate redundant activities [16]–[18]. In contrast to Arabidopsis, mutation in the barley SHN1/WIN1 ortholog (Nud) was sufficient to generate a severe morphological change in which the typically hulled caryopses developed into naked ones [19]. Nud was suggested to direct the deposition of a lipidic matter on the pericarp epidermis that adheres the hull to the caryopsis in a way similar to postgenital fusions displayed by numerous cuticular mutants [20]–[21].
In this study we have co-silenced the three SHN clade members in order to decipher their modes of action and resolve their biological roles. We revealed that SHN clade genes regulate the elongation and decoration (i.e. nanoridges formation) of reproductive organ epidermal cells, particularly in the petal surface. They also emerge as mediators of cell adhesion and separation during abscission and dehiscence. Additionally, the results suggest that beside their function in the cutin pathway, these transcription factors possess putative downstream target genes that are involved in cell wall configuration through pectin modifying enzymes and structural proteins. Thus, the study of SHN transcription factors provides novel insight to the transcriptional control that mediates the patterning of reproductive organs surfaces and their associated separation processes in between cell layers.
To circumvent the likely functional redundancy between the Arabidopsis SHN clade members we generated plants in which they were simultaneously silenced through an artificial microRNA approach (Figure S1A and Text S1). The presence of cleaved products and transcriptional downregulation of all three SHN genes was confirmed in the 35S:miR-SHN1/2/3 plants (Figure 1A and Figure S1B–S1C). No visual change was observed in these plants during vegetative growth and cuticle permeability of their rosette leaves was normal (Figure S1D–S1G). However, reproductive organs, particularly petals, were severely affected (Figure 1C–1D). This was evident already in buds that displayed postgenital fusions between petals and other floral organs at their tops (Figure 1H–1I). The expansion of petals and elongation of the carpels were restrained and they were curved and/or twisted (Figure 1I and Figure S1L–S1M). The changes in flower organ morphology also impinged on self-pollination and semi-sterility was occasionally detected (Figure 1B). Interestingly, mutant flower organs abscised earlier (Figure 1E and Figure S1J–S1K), and in some cases the abscised flower parts stayed attached to the top of the silique due to the postgenital organ fusion between them (Figure 1F–1G).
Microscopic observation of floral organs surfaces in the 35S:miR-SHN1/2/3 plants revealed extensive alterations to their archetypal epidermal cells (Figure 2 and Figure S2). Both abaxial and adaxial conical cells of petals appeared less elongated, more spherical and compact in addition to being separated with wider spaces as compared to the wild-type (WT) cells (Figure 2). Remarkably, nanoridges, typically displayed on WT petal epidermis [22]–[23], were either absent (adaxial) or significantly reduced (abaxial) in the 35S:miR-SHN1/2/3 petal cells (Figure 2A–2F). Altered epidermis cell size, shape and nanoridge decoration was also observed in surfaces of additional floral organs such as sepals, styles, filaments, nectaries, and pedicles (Figure S1N–S1Q and Figure S2). The observed phenotypes provided evidence that the SHN clade genes function redundantly in cell elongation, separation and nanoridge formation of reproductive organs. In contrast to the 35S:miR-SHN1/2/3 floral organs, silencing SHN1/WIN1 alone did not cause any visible morphological changes in floral organs, particularly in petal surfaces (Figure S3).
In order to unravel the molecular mechanism by which the SHN factors regulate the patterning of reproductive organ surfaces we compared the transcriptome of 35S:miR-SHN1/2/3 flower buds to the one of WT. A modest set of 38 differentially expressed genes was detected; 30 transcripts including SHN1 and SHN3 (SHN2 was not represented in the array) were downregulated while 8 others were upregulated in 35S:miR-SHN1/2/3 buds (Table 1).
Interestingly, one of the two main functional categories that dominated the differential genes represented six cell wall related genes (Table 1). Four of them corresponded to enzymes associated with pectin degradation or modification, including two pectate lyases (PLL14 and PLL23), a polygalacturonase (ADPG1) and a pectin methylesterase inhibitor (PMEI). Two additional genes putatively encode cell wall structural proteins: a hydroxyproline-rich glycoprotein (HRGP) and a glycine-rich protein (GRP). The second major category consisted of seven genes that putatively encode cuticular lipids (mainly cutin) related proteins, including 2 cytochrome P450s (CYP86A4 and CYP86A7) implicated in flower cutin biosynthesis [18], [23], three GDSL-motif lipase/hydrolases (RXF26, At2g42990, and At5g33370) that are highly similar to the reported cutin related lipase At2g04570 [18], and one hydrolase (BODYGUARD 3, BDG3), the closest homolog of BDG1, an epidermis-specific extracellular protein associated with cuticle formation [24]. Fatty Acyl-CoA Reductase 1 (FAR1), the seventh gene was associated with primary fatty alcohol production [25]; its additional and/or alternative function with relation to surface lipids will be discussed below.
Two downregulated genes encoded a potassium transporter (KUP5) and an ABC transporter (PGP13/MDR15); both are involved in cell growth [26]–[28]. Additional three downregulated genes encoded kinase and/or kinase like proteins, that are potentially involved in reporting sensing aspects of cell wall structure and function [29]. Differential expression of 24 genes including the three SHN genes was subsequently validated using realtime RT-PCR assays (Figure S4 and Text S1). Altogether, gene expression analysis results indicated that the phenotype observed in 35S:miR-SHN1/2/3 reproductive organs probably result from the altered expression of their target genes, particularly those related to cutin and cell wall remodeling and function.
Because plant organ fusion and separation have been reported to be associated with cuticle [19]–[20], [22], we subsequently examined the changes in cuticular lipids in leaf and flower tissues of the 35S:miR-SHN1/2/3 plants. While the amount of leaf cutin was not significantly changed (Figure S5A), the amount of flower cutin in the 35S:miR-SHN1/2/3 plants was reduced to 48.4% of the wild-type (Figure 3A). The changes in flower cutin loads reflected the changes in the cuticle permeability in flower tissues (Figure S1F–S1I). The substantial decrease of dioic acids (DFA, particularly C16, C18:2 and C18:1), ω-hydroxy fatty acids (ω-HFA, particularly C16 and C18:3), 9/10,16-dihydroxy hexadecanoic acid (C16-9/10,16-DHFA) and 9(10)-hydroxy-hexadecanedioic acid (C16-9/10-HDFA) largely contributed to the reduced flower cutin in the 35S:miR-SHN1/2/3 plants. Levels of cuticular waxes in either leaves or flowers were not significantly altered in the 35S:miR-SHN1/2/3 lines (Figure S5B–S5C).
The finding that co-silencing the three SHN genes affected the expression of pectin modifying genes prompted us to analyze the cell wall pectin composition in the seed mucilage and buds. GC-MS analysis did not reveal any significant compositional changes in seed mucilage and the bud cell wall pectic monosaccharides (Figure S5D–S5E and Text S1). We next used Fourier transform infrared (FTIR) spectroscopy to examine if petals of the 35S:miR-SHN1/2/3 plants exhibited structural changes in their cell walls. Principal component analysis (PCA) showed a clear separation of the petal FTIR spectra between 35S:miR-SHN1/2/3 petals and WT ones (Figure 3C). The difference spectrum (Figure 3B) generated by digitally subtracting the average 35S:miR-SHN1/2/3 spectrum from the average WT petals spectrum showed that WT petal cell wall had more acyl esters (1740 cm−1) [30]–[31], amide III proteins (1230 cm−1) [32], and non-cellulosic carbohydrates (1100 to 900 cm−1) [33]. In contrast, 35S:miR-SHN1/2/3 petal cell walls contained more salt-form of pectin (1430 and 1600 cm−1, respectively) [32], amide I and amide II proteins (1650 and 1550 cm−1, respectively) [32]–[33], and phenolic esters or aromatic lignins (1635 and 1510 cm−1) [32]–[33].
To localize the pectic polysaccharides in the cell walls, two novel rat monoclonal antibodies LM19 and LM20, which recognize pectic homogalacturonan (HG) epitopes [34], were used to hybridize transverse sections of inflorescence stems (pith parenchyma) and flowers. Similar to an earlier observation in tobacco plants [34], LM19 localized pectin to junctures (middle lamella) while LM20 localized pectin to the intercellular spaces (air spaces) in both WT and 35S:miR-SHN1/2/3 inflorescence stems (both antibodies appeared as green fluorescence) (Figure 3D). However, the florescence of LM19 in transverse sections of the 35S:miR-SHN1/2/3 samples became weaker and they were aggregated along the middle lamella line. Moreover, the florescence of LM20 in 35S:miR-SHN1/2/3 was enhanced not only in the air spaces but also in the middle lamella. In addition, the florescence of LM20 binding to air spaces become stronger in microtome sections of 35S:miR-SHN1/2/3 petals and developing seed coats, as compared to WT ones (Figure 3E). Because the binding of both LM19 and LM20 to pectin is sensitive to pectate lyase treatment and they bind preferably to HG [35], these results indicated alteration to HG distribution in the mutants. Therefore, silencing the SHN clade genes not only affected the cutin matrix of the cuticle but also the cell wall matrix of the cell.
Remarkably, in silico analysis (Table 1) showed that as SHN1/WIN1, 13 of the differentially expressed genes (12 downregulated and one up regulated in the 35S:miR-SHN1/2/3 plants) display a petal-specific expression pattern [35]. Moreover, all those 12 petal-specific downregulated genes, together with SHN1/WIN1, SHN3, and 3 more genes display decreased expression in senescing petals [36]. Furthermore, 9 of the differential genes in addition to SHN1/WIN1 are expressed in the stamen abscission zone (AZ) [37] while 2 genes and SHN1/WIN1 are enriched in the nectary [38], and 13 genes and SHN3 are differentially expressed in senescing siliques [36]. These results provided evidence that both the SHN factors and their putative targets are associated with reproductive organ development (i.e. petals and siliques) and possibly cell separation as well. The series of genes altered in the 35S:miR-SHN1/2/3 plants were also strongly co-expressed with the SHN factors (Figure S6 and Table S2), further indicating the functional link between the groups of genes we have identified in the array analysis.
In order to examine whether loss of function of the putative SHN clade proteins target genes results in alteration to petal surface we screened for T-DNA insertions in the entire set of 28 downregulated genes. Homozygous knockout lines could be identified for thirteen of them and their petals surface was examined using scanning electron microscopy (Figure S7). Petals of the At5g23970 (a putative acyltransferase) and At5g33370 (a putative GDSL-lipase) knockout plants exhibited collapsed conical cells, while those of At4g24140 (bodyguard3/bdg3), At5g03350 (a receptor like protein) and At1g01600 (cyp86a4) displayed abnormal abaxial nanoridges (Figure 4A–4D).
Some differential genes identified in microarray analysis belong to large multi-gene families as for example lipases and cytochrome P450s. This suggested that they might be functionally redundant with other family members. We therefore co-silenced the CYP86A4 with CYP86A7, and the GDSL-lipase At5g33370 with its closest homolog At3g04290, LTL1 [39], via the artificial microRNA method. Plants co-silenced for either one of these pairs of genes displayed severe floral organ fusion and alteration in the conical cell shape and/or epidermis cell decoration (Figure 4E–4H). These results from single knockouts and the co-silenced lines provided additional evidence for the functional link between the putative SHN proteins target genes and the patterning of the petal surface.
We subsequently examined the activation of promoters of genes that were differentially expressed in the 35S:miR-SHN1/2/3 plants by the SHN transcription factors using a dual luciferase assay system [40]. Promoter regions of 23 putative targets and the 3 SHN clade genes were examined. Thirteen out of 23 were significantly activated by at least one of the three SHN transcription factors (Figure 5). Promoter regions of seven genes were activated by all three factors including the ones of RXF26, CYP86A4, CYP86A7, BDG3, FAR1, GRP, and GRXC11. The promoters of PRX02 (a peroxidase), ARD3 (an acireductone dioxygenase), and At2g43620 (a chitinase) were only activated by SHN1/WIN1, SHN2, and SHN3, respectively. Interestingly, SHN1/WIN1 and SHN2 were able to activate each other's promoter, while SHN3 was able to activate all three SHN genes promoters. We included LACS2 promoter as a positive control [18], however, activation of this gene promoter by the SHN transcription factors was not detected in our assay. These results further confirmed the functional redundancy of SHN transcription factors in cuticle and cell wall metabolism by acting directly on common targets and by regulating each other and possibly their own transcription.
Gibberellins (GAs) are a class of plant hormones involved in the regulation of flower development in Arabidopsis. GA promotes the expression of floral homeotic genes APETALA3 (AP3), PISTILLATA (PI), and AGAMOUS (AG) by antagonizing the effects of DELLA proteins, thereby allowing continued flower development [41]. Publically available array data suggested that GA promotes the expression of SHN1/WIN1 while DELLA suppresses SHN1/WIN1 expression, which was examined in the ga1-3 and the ga1-3 gai-t6 rga-t2 rgl1-1 rgl2-1 (i.e. penta) [35]. Remarkably, in young flower buds, GA promotes the expression of thirteen of the putative SHN target genes identified in this study while it down regulates the expression of another four putative target genes, all of them in a DELLA dependent manner ([42], Figure S8A–S8B). In addition, GA regulates another two putative SHN target genes, AT4G27450 and AT1G27940, in a DELLA-independent way [42]. The results described above led us to suggest that GA might be involved in cuticle assembly during flower organ development via modulating the expression (directly or indirectly) of the SHN transcription factors and their downstream target genes.
To test this assumption, we examined the expression of SHN genes in different GA biosynthesis or signaling mutants (Figure 6A). Quantitative RT-PCR analysis showed that expression of SHN1/WIN1 is downregulated in the ga1-3 mutant that is defective in GA biosynthesis. It also showed that DELLA significantly suppressed SHN1/WIN1 expression, since the expression of SHN1/WIN1 in the double (rga-t2 rgl2-1; partial loss of DELLA signaling) and quadruple DELLA (gai-t6 rga-t2 rgl1-1 rgl2-1) mutants in the ga1-3 background was recovered to equal and even much higher levels than that of the wild type, respectively. Knockout of SPY4, another repressor of GA signaling, also enhanced SHN1/WIN1 expression as compared to the wild type. As compared to SHN1/WIN1, SHN2 showed the opposite expression pattern in the background of the various GA biosynthesis and signaling mutants. Expression of SHN2 was upregulated in the ga1-3 background while it was significantly downregulated in the penta and spy4 mutant backgrounds. Interestingly, neither GA biosynthesis nor the signaling mutants significantly altered SHN3 expression.
We also examined the expression of SHN clade genes in both the WT and ga1-3 flower buds in response to exogenous GA application (Figure 6B). Quantitative RT-PCR analysis showed that GA application to the ga1-3 mutant increased the levels of SHN1/WIN1 and decreased the levels of SHN2 expression as compared to ga1-3 alone, as does the endogenous GA (Figure 6A). The response of SHN3 might be different between endogenous and externally applied GA as its expression did not change significantly in the ga1-3 background alone while it was altered upon GA supplementation in either the WT or ga1-3 (Figure 6B).
Finally, we also carried out GC-MS analysis of the flower cuticular lipids of the GA biosynthesis and signaling mutants. While flower waxes were not significantly altered in the ga1-3 and penta mutant flowers, the total cutin load, particularly of the 9/10,16-dihydroxy hexadecanoic acid (C16-9/10,16-DHFA), the predominant monomer of the Arabidopsis flower cutin, was significantly different between WT and ga1-3 and between ga1-3 and the penta mutant (Figure S9). Nevertheless, SEM observation did not reveal any significant changes in the petal surface of the open flowers in the mutant plants (Figure S9). Since we applied exogenous GA to ga1-3 plants to induce flowering [43] prior to the SEM observation, this might explain the absence of a surface phenotype in mutant petal surface. All together, these results suggest that SHN transcription factors might play a key role in the GA-mediated flower organ development regulatory network.
Aerial plant organs display tremendous variation in their surface topography and composition of the cuticular layer covering their outer epidermis. This diversity in the exterior layer is essential for both organogenesis and the interaction with the environment. In flowers for instance, the typical surface of organs is vital for their function as it ensures their proper development by preventing postgenital fusions while at the same time mediating the interaction with insect pollinators [44]–[45]. Whereas many molecular components of pathways determining flower organ initiation and identity have been characterized to date [46], our knowledge regarding formation and function of their outer surface, namely the cuticle, is limited. Here, in-depth analysis of Arabidopsis plants in which the three SHN transcription factors were co-silenced revealed that these regulators play a prominent role in patterning floral organ surface by controlling metabolism of cuticular lipids and possibly the associated cell wall components.
The lack of any visual phenotype in floral organs of SHN1/WIN1 silenced plants ([18]; Figure S3), pointed to functional redundancy among the 3 SHN clade members. Even though expression of either one of the three SHN genes was not entirely reduced, the use of an artificial microRNA targeting the entire clade was sufficient to obtain several, striking, visual phenotypes that matched the previously described SHN genes expression patterns [16]. Floral organs were affected, likely as a result of altered cuticle composition, structure and consequently permeability. However, cuticle alteration might not be the only explanation to the defects observed in organ formation since they might also be a result of SHN genes effect on the process of epidermal cell differentiation and development. This was evidenced in the altered epidermal cells size and shape in petals and sepals of the 35S:miR-SHN1/2/3 plants. These strong epidermis phenotypes (in pavement cells, trichomes and stomata) observed previously in plants overexpressing either one of the three SHN genes support this proposal [16].
Down regulation of the SHN clade genes had an additional effect on floral organs as SEM and transmission electron microscopy (TEM) revealed changes in nanoridges that typically decorate surfaces of flower organs [44]. Formation of nanoridges in Arabidopsis flowers was recently associated with cutin, particularly with C16-9/10,16-DHFA, the major monomer of Arabidopsis petal cutin [22]–[23], that was also dramatically reduced in the 35S:miR-SHN1/2/3 plants. However, the absence of nanoridges on the surface of tomato fruit that also contains C16-9/10,16-DHFA as a major monomer, suggests additional factors including polymer structure and distribution that mediate nanoridge formation [23], [47].
Earlier work using promoter-reporter assays suggested that SHN transcription factors act not only in the interface between the plant and its environment but also at the interface between cells and cell layers [16]. Of particular interest was SHN2 that showed strict expression in the anther and silique dehiscence zones upon organ maturation. The proposed role of SHN transcription factors in the adhesion of cell layers was strongly corroborated by the recent finding that an SHN-like gene in barley (Nud) mediates the contact of the caryopsis surface to the inner side of the hull by forming a specialized lipid layer [19]. In this study we detected earlier abscission of floral organs in the silenced lines which corresponded well with SHN genes expression in the base of sepals, petals, stamens and siliques in the abscission region. Organ separation events including pod shatter, seed detachment from the maternal plant, pollen separation after meiosis, anther dehiscence and floral organ abscission, are thought to be associated with alterations to properties of the cell wall matrix, mainly pectins and wall proteins [1], [48]–[49]. The pectin degradation activity of polygalacturonases (PGs) has been linked with all separation events described above. Recently, three Arabidopsis PGs have been associated with cell separation during reproductive development [50]. One of these, ADPG1, displayed altered expression in the 35S:miR-SHN1/2/3 plants and its promoter was shown here to be activated by SHN1/WIN1 and SHN2. Thus, SHN action on organ adhesion/separation possibly combines modification to cuticular lipids (i.e. cutin) as well as pectins of the cell wall.
Array analysis revealed a concise set of genes that are putative downstream targets of the SHN transcription factors in flower buds, only two out of them (CYP86A4 and CYP86A7) overlapped with the previously reported group of 11 SHN1/WIN1 putative targets [18]. This could be explained by the fact that while Kannagara et al. (2007) detected genes that were upregulated following induction of SHN1/WIN1 in fully expanded leaves [18], we examined flower buds in which the SHN clade genes were co-silenced. Thus, genes from these two experiments most likely represent downstream targets in either leaves or flowers or both tissues. Together, these studies also demonstrated that wax load changes in the SHN overexpression lines were probably an indirect effect.
SHN transcription factors emerge as regulators of genes derived from four prominent families associated with the cuticle including two cytochrome P450s of the CYP86A clade (CYP86A4 and CYP86A7), BDG3, encoding one of the five BDG1-like proteins [24], three genes of the large family of GDSL-motif lipase/hydrolases [39] and one of the eight-member clade of fatty acyl-CoA reductases [25]. Apart from the latter, these genes or their family members have been reported to be involved in either cutin biosynthesis or polymer assembly in the extracellular matrix in plant reproductive organs [10], [23], [51]–[53]. FAR1 has been recently associated with formation of suberin, a polymer that is structurally related to cutin and is often deposited following cell to cell separation in aerial organs to form a protection layer that will shield against penetration of pathogens and dehiscence [25], [54]. Below ground, endodermal suberin is thought to regulate the apoplastic movement of water and solutes into the stele [55]–[56]. The SHN3 expression in roots ([16], Figure S10) and the endodermal expression of FAR1, BDG3, CYP86A4 and At1g16760 (Figure S10) suggested that the latter 4 genes are targets of SHN transcription factors both above and below ground. Hence, SHN transcription factors and their targets are not only involved in cutin assembly in reproductive organs but are likely to play a role in root suberin deposition. CYP86A4 was suggested to provide ω-hydroxylation activity that is complementary to CYP86A1 in the biosynthesis of suberin [57] and FAR1 was recently reported to be associated with generating primary fatty alcohols for suberin deposition [25]. However, the role of BDG3 and At1g16760 in root suberin remains to be determined.
Previous reports regarding the SHN clade members highlighted their role in regulating the biosynthesis of cuticular lipids for surface formation [16]–[18]. However, the results of the present study imply that activity of these factors goes beyond regulating a single metabolic pathway (i.e. cutin) for cuticle formation and they take part in the genetic program that mediates floral organ morphogenesis, more specifically in determining organ size and shape as well as the formation of specialized epidermis cell types (e.g. the petal conical cells). Related to this, gene expression changes detected in the 35S:miR:SHN1/2/3 flower buds strikingly resemble the ones implicated in the formation of the single epidermis cotton fiber cell during its elongation. These include altered expression of genes associated with cell wall loosening through modification of pectin [58], genes associated with the build-up of a higher turgor by increased accumulation of the major osmoticum such as soluble sugars, K+, and malate [27], redox-related genes [59]–[60], genes related to phytohormone biosynthesis and signaling cascades [61].
Flowering in Arabidopsis consists of three distinct phases: floral initiation, floral organ initiation and floral organ growth. Earlier studies on GA signaling revealed that GA promotes Arabidopsis petal, stamen, and anther development by opposing the function of the DELLA proteins [62] and that GA signaling is not required for floral organ specification but essential for the normal growth and development of these organs [63]. Different combinations of DELLA proteins are key to floral organ development (RGA, RGL1, RGL2), because individual DELLA proteins have different temporal and spatial expression patterns [62]. The unique temporal and spatial expression patterns of SHN clade genes in the flower tissues [16] and their distinct expression patterns in response to the alteration of the GA signaling reported here suggest that SHNs might be part of GA floral regulatory networks. In this context, GA might act as a positive regulator of SHN1/WIN1 in the regulation of floral organs development (i.e. elongation of petal, stamen, and anther) [37], [62] in the early stages of flower development. In addition, GA emerges as a negative regulator of SHN2 in modulating the cell separation processes related to silique and anther dehiscence, floral organ abscission in the later stages of flower development. Hence, GA might be involved in cuticle assembly during the expansion of petals and other floral organs. The growth and elongation of organs requires the interaction between the outer and inner cell layers, which is coordinated by hormonal signals [4]–[5]. GA has been shown to promote cutin synthesis during other growth related processes including the rapidly growing internodes of deep-water rice [64], in extending stems of peas [65], and in developing tomato fruit [66]. Similarly, in this study, GA application resulted in a significant increase in the cutin load of ga1-3 mutant flowers. Future studies positioning the SHN proteins in the wide genetic network that controls flower development will shed light on how cuticle and cell wall metabolism is coordinated with the processes of flowering and fertility.
All Arabidopsis plants used in miR-SHN1/2/3 experiment were in the Col-0 genetic background, while those used for DELLA or GA experiment were in Ler genetic background. Plants were grown on a soil mixture in a growth room at 20°C, 70% relative humidity, a 16/8-h light/dark cycle at a fluorescent light intensity of 100 µmol m−2s−1. All knock out lines were bought from either ABRC or NASC, while GA biosynthesis and signaling mutant were kind gifts from Hao Yu (National University of Singapore, Singapore) and David Weiss (The Hebrew University, Israel). Exogenous GA application was carried out as described [67] with minor modifications. 100 mM GA3 or ethanol containing water was fine sprayed daily for 6 days on 6-week-old plants, and the buds were collected for analysis.
For the 35S:miR-SHN1/2/3 construct, the designed artificial miR-SHN1/2/3 sequence was directly synthesized from BIO S&T (Bio S&T Inc., Montreal, Canada). After being sequenced, it was put into pART7 vector, and finally subcloned to pART27. Transformation to Agrobacterium tumefaciens strain GV3101 was done via electroporation and planta transformation was done via floral dipping as described [68]. Promoter sequences of the putative SHN target genes (approximately 2 kb upstream of the start codon) were cloned from WT genomic DNA, and coding sequences of the three members of SHN clade were cloned from WT flower cDNA, using yellow Taq DNA polymerase (Roboklon Gmbh, Berlin, Germany) with corresponding gene specific primer pairs (Table S1). Those promoters and TFs were cloned into pGreen II 0800-LUC vector and pBIN plus vector, respectively, and then transformed to Agrobacterium tumefaciens strain GV3101. All DNA sequence cloned were examined by direct sequencing.
Toluidine blue examination of cuticle permeability was performed as previously described [69]. For Rethinium red staining, the inflorescences of 7-week-old plants were fixed and embedded in LR White resin (London Resin Co., Basingstoke, UK) as described previously [70]. Sections were cut to a thickness of 0.5–1 mm using a diamond knife on an Ultracut microtome (Leica) and sections were collected on glass slides. The slides were stained with 0.1% Rethinium red for 5 min and washed with double distilled water, and then observed with Nikon ECLIPSE E800 microscope.
All electron microscopy works were done as previously described [22]. For scanning electron microscopy (SEM), flowers from 7-week-old plants were collected, fixed with glutaraldehyde using standard SEM protocol [71], dried using critical point drying (CPD), mounted on aluminum stubs and sputter-coated with gold. SEM was performed using an XL30 ESEM FEG microscope (FEI) at 5–10 kV. For TEM, flowers from 7-week-old plants were collected and processed using a standard protocol [72]. The Epon-embedded samples were sectioned (70 nm) using an ultramicrotome (Leica) and observed with a Technai T12 transmission electron microscope (FEI).
Total RNA was extracted from closed buds from 7-weeks-old WT and homozygous 35S:miRSHN1/2/3 T3 plants using RNeasy Plant Mini Kit (Qiagen) with an on column DNAse treatment. The subsequent microarray analysis and qRT-PCR analysis were performed as described previously [21]. For microarray analysis, the double-stranded cDNA was purified and served as a template in the subsequent in-vitro transcription reaction for complementary RNA (cRNA) amplification and biotin labeling. The biotinylated cRNA was cleaned, fragmented and hybridized to Affymetrix ATH1 Genome Array chips. Statistical analysis of microarray data was performed using the Partek® Genomics Suite (Partek Inc., St. Louis, Missouri) software. CEL files (containing raw expression measurements) were imported to Partek GS. The data was preprocessed and normalized using the RMA (Robust Multichip Average) algorithm [73]. The normalized data was processed by PCA (Principal Component Analysis) and hierarchical clustering to detect batch or other random effects. To identify differentially expressed genes one-way ANOVA analysis of variance was applied. Gene lists were created by filtering the genes based on: fold change, p<0.01, and signal above background in at least one microarray. Up-regulated genes were defined as those having a greater than or at least 1.5-fold linear intensity ratio while down-regulated genes were defined as those having a less than or at most −1.5-fold linear intensity ratio. The experiment was performed in duplicate, preparing two independent biological replicates from 5–6 plants each.
Waxes were extracted and analyzed as described [22]. For cutin analysis, soluble lipids were extracted from leaf and closed buds by dipping them in 10 ml of a methanol/chloroform (1∶1, v/v) mixture for 14 days (solvent changed daily). The tissues were dried, weighed (about 10–20 mg) and kept in N2 till analysis. The cutin was depolymerized and analyzed as described previously [22], [54].
Petals from 7-week-old flowers were collected (60 petals each sample, n = 8), cleared with chloroform and methanol (1∶1), and then air-dried overnight [74]. Samples were ground with solid crystalline KBr to fine powder and pressed to 1-mm tablelets. FTIR spectra were acquired in the absorbance mode at a resolution of 4 cm−1 with 32 co-added scans at wave number range 4000 to 250 cm−1 using a NICOLE1 380 FITR Spectrometer (Thermo Electron Corporation). Each spectrum was baseline corrected and spectral area normalized prior to generating average spectra and digital subtraction spectra. Primary component analysis was performed using Multiple Experiment Viewer.
Inflorescence stems transverse sections were prepared according to Willats et al [75]. Regions (0.5 cm long) of 7-week-old Arabidopsis stem (3th internodes from the bottom) were excised and sectioned by hand to a thickness of ∼100–300 µm. Sections were immediately placed in fixative consisting of 4% paraformaldehyde in 50 mM PIPES, 5 mM MgSO4, and 5 mM EGTA. Following 30 min of fixation, sections were washed in the PIPES buffer, and then in 1× PBS buffer. Petals and gynoecium transverse section were prepared as described [65] and In vitro immunocytochemistry was carried out as described by Verhertbruggen et al [34]. Sections were incubated for 1.5 h in 5-fold dilution of two new rat monoclonal antibody hybridoma supernatant (LM19 and LM20) diluted in 5% Milk/PBS, respectively. After being washed by gently rocking in PBS at least three times, sections were incubated with a 100-fold dilution of anti-rat IgG (whole molecule) linked to fluorescein isothiocyanate (FITC) in 5% Milk/PBS for 1.5 h in darkness. After washing in PBS for at least 3 times, sections were mounted in a glycerol∶PBS (vol∶vol, 1∶1) solution. Immunofluorescence was observed with Nikon ECLIPSE E800 microscope equipped with epifluorescence irradiation and DIC optics. Images were captured with a camera and NIS-Elements BR30 software.
Transient assay was carried out as described [40] with the exception that 150 µg/ml instead of acetosyringone was included in the infiltration media [76]. Luminescence was measured using Modulus Microplate Luminometer (Turner Biosystems, Sunnyvale, CA) by mixing 20 µl sample extract with 80 µl Luciferase assay reagent or Renillase assay reagent, respectively, and the data was collected as ratio. Background controls were run with only the transcription factor, promoter-LUC, and pBIN Plus empty vector, and pBIN Plus empty vector with promoter-LUC in the preliminary assay, and pBIN Plus empty vector with promoter-LUC was chosen later for background control in all experiments due to its relatively higher induction of Luciferase activity than other plasmid tested.
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.