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.pgen.0030131 | SREBP Controls Oxygen-Dependent Mobilization of Retrotransposons in Fission Yeast | Retrotransposons are mobile genetic elements that proliferate through an RNA intermediate. Transposons do not encode transcription factors and thus rely on host factors for mRNA expression and survival. Despite information regarding conditions under which elements are upregulated, much remains to be learned about the regulatory mechanisms or factors controlling retrotransposon expression. Here, we report that low oxygen activates the fission yeast Tf2 family of retrotransposons. Sre1, the yeast ortholog of the mammalian membrane-bound transcription factor sterol regulatory element binding protein (SREBP), directly induces the expression and mobilization of Tf2 retrotransposons under low oxygen. Sre1 binds to DNA sequences in the Tf2 long terminal repeat that functions as an oxygen-dependent promoter. We find that Tf2 solo long terminal repeats throughout the genome direct oxygen-dependent expression of adjacent coding and noncoding sequences, providing a potential mechanism for the generation of oxygen-dependent gene expression.
| Transposons are present at high copy number in diverse organisms ranging from single-celled bacteria to complex mammals and plants. Transposons are mobile genetic elements that can replicate and move to new locations within the genome. An ongoing debate exists regarding whether transposons are merely genetic parasites or whether they confer a benefit to the host organism. Previous studies have demonstrated that mobilization of transposable elements is induced in response to different cell stresses. Here, we describe the direct mechanism by which the fission yeast Tf2 class of retrotransposons is physiologically activated in response to changes in environmental oxygen. Tf2 transcription and mobilization are dramatically induced under low oxygen by the yeast ortholog of mammalian SREBP, a transcription factor that controls cholesterol homeostasis. Our studies demonstrate that Tf2 elements direct oxygen-dependent transcription of adjacent sequences, and a genome-wide survey identified several genes whose expression is under Tf2 control. These findings suggest that mobilization of Tf2 to new locations in the genome could reengineer the cell's transcriptional network with potentially beneficial consequences to the host.
| Transposable elements are mobile DNA sequences that are present in most eukaryotic organisms and occupy a large fraction of sequenced genomes; for example, human (∼50%) [1], C. elegans (12%) [2], and plants (10%–80%) [3]. Once viewed as simple mutagens, transposable elements are increasingly seen as playing a significant role in evolution by affecting genome size, structure, and host gene expression [4]. Retrotransposons are genetic elements that propagate through reverse transcription of their RNA and integration of the resulting cDNA into another genomic locus [4]. Retrotransposons resemble retroviruses in both gene structure and replication mechanism, but lack a viral envelope protein required for cell–cell infectivity. Long terminal repeat (LTR) retrotransposons have terminally redundant ends flanking an internal coding region that codes for viral particle coat, reverse transcriptase, integrase, and protease. The LTR is the functional promoter for the transposon and can be divided into three regions U3, R, and U5. U3 contains cis-acting sequences involved in transcriptional regulation, R encodes the transcription initiation site, and U5 carries the transcriptional termination signal, which plays a role in the 3′ LTR [5]. In addition to full-length retrotransposons, genomes contain many solitary LTRs and fragments of retrotransposons. Solo LTRs are footprints of previously intact retrotransposons and can be generated by intra-element homologous recombination between LTRs. Thus, solo LTRs can function as promoters, influencing transcription of adjacent genes. Several examples of genes transcribed by a solo LTR from a human endogenous retrovirus have been found: apolipoprotein-C1, endothelin-B receptor [6], β-globin [7], and Mid1 [8].
Transposable elements can also negatively impact the genome by causing mutations and affecting host viability, necessitating that a balance exist between element expansion and host mutagenesis [9]. To control this balance, mechanisms exist that limit expression of retrotransposons including RNAi, heterochromatization, cosuppression, dependence on host factors, and regulated element transcription [3,10–13]. Regulated transcriptional control provides a mechanism for restricting element expression to defined cellular conditions. Retrotransposon transcription and subsequent transposition are upregulated by different environmental conditions and stresses. Ionizing radiation, DNA damage, mating pheromone, and nutrient limitation activate S. cerevisiae Ty elements [10,14,15]; heat and sodium azide induce Drosophila copia elements [16]; and wounding, biotic elicitors, and pathogen attack activate Tnt1 in the Solanaceae plant family [17]. One well-understood regulatory mechanism involves the transcriptional activators Ste12 and Tec1, which regulate S. cerevisiae Ty1 elements in response to mating pheromones, nitrogen starvation, and invasive growth signals [18–20]. Transcription regulatory sequences for Ty1 exist both in the 5′ LTR and the element open reading frames.
Here, we report the physiological induction of the fission yeast Tf2 transposon family by low oxygen and describe the mechanism of this regulated expression. Schizosaccharomyces pombe contains two families of retrotransposons, Tf1 and Tf2; however, the laboratory-adapted strain has lost Tf1 [21]. Tf2 transposons are 4.9 kb in length and are flanked by 349-bp LTRs. When overexpressed from a heterologous promoter, Tf2 retrotransposons show a low transposition frequency, preferring to recombine with existing transposons [22,23]. While much is known about the mechanism of Tf retrotransposition in fission yeast, little is known about the regulation of element transcription. Genome-wide analyses indicate that neither heterochromatization nor RNAi-mediated silencing plays a major role in regulation of Tf2 element expression [24,25]. We previously identified Tf2 transposons as targets of the oxygen-dependent transcription factor Sre1 [26]. Sre1 is the fission yeast ortholog of sterol regulatory element binding protein (SREBP) that controls lipid homeostasis in mammalian cells [27–29]. Under low oxygen, Sre1 is proteolytically processed, enters the nucleus, and activates genes required for adaptation to low oxygen growth [26]. Here, we demonstrate that Sre1 directly activates Tf2 transcription and cDNA mobilization under low oxygen by binding to sequences in the Tf2 LTR. Sre1 does not activate only a single element but rather the family as a whole. Interestingly, Tf2 solo LTRs function as oxygen-dependent promoters that are capable of directing low oxygen transcription of adjacent DNA sequences. Taken together, these studies describe a detailed mechanism for the transcriptional regulation of retrotransposons by low oxygen, reveal a new environmental condition for element mobilization, and provide an example of LTR-mediated transcriptional control of host gene expression.
We identified Tf2 transposons as targets of the oxygen-dependent transcription factor Sre1 in a global analysis of low-oxygen gene expression. [26]. To confirm that Tf2 transcription was induced under low oxygen and required Sre1, we analyzed expression of Tf2 mRNA and Tf2 encoded proteins from cells grown in the presence or absence of oxygen (Figure 1A, upper panel). Tf2 mRNA increased after shifting to low oxygen, reaching a level 59-fold higher than in the presence of oxygen after 8 h. Expression of Tf2 reverse transcriptase and integrase protein was similarly induced (Figure 1A, lower panels). Consistent with our microarray data [26], Tf2 mRNA and protein expression were not induced by low oxygen in a sre1Δ strain (Figure 1B). Thus, Sre1 is required for the low oxygen induction of Tf2 retrotransposon mRNA and protein.
Fission yeast contains 13 full-length Tf2 transposons whose coding sequences are 99% identical at the DNA level and thus cannot be distinguished by hybridization [30]. To analyze the expression of individual transposons, we designed a strategy to tag each transposon with ura4+ (Figure 2A). We generated 13 different strains (Tf2–1 to Tf2–13), each carrying a single, tagged Tf2 element. These 13 strains and an untagged wild-type strain were grown in the presence or absence of oxygen for 8 h and processed for northern analysis using a strand-specific ura4+ probe. Expression of 12 out of 13 transposons increased under low oxygen (Figure 2B, upper panel). Notably, Tf2–11 was not induced despite low oxygen expression of other Tf2 elements in this strain (Figure 2B, lanes 21–22). Hereafter, we refer to the 12 coregulated elements collectively as Tf2. To test if low oxygen induction required Sre1, we deleted sre1+ from five Tf2-ura4+ tagged strains (Tf2–3, 4, 7, 10, and 11). Low oxygen induction of Tf2 transposons required Sre1 (Figure 2C), while deletion of sre1+ had no effect on Tf2–11 expression. These data demonstrate that with the exception of Tf2–11, Sre1 controls low oxygen expression of Tf2 transposons.
Tf2 transcription initiates in the 5′ LTR [31]. To test if the Tf2 LTR is sufficient to promote oxygen-dependent gene expression, we assayed expression of a lacZ reporter driven by either Tf2 LTR or Tf2–11 LTR in the presence or absence of oxygen (Figure 3A). Expression of Tf2 LTR-lacZ was induced more than 100-fold in the absence of oxygen and this induction required sre1+. Cells carrying Tf2–11 LTR-lacZ showed a background level of β-galactosidase activity that was not regulated by oxygen or sre1+. A search of the Tf2 5′ LTR revealed a DNA sequence (5′-ATCGTACCAT-3′) located 443 bp upstream of the Tf2 ORF in 12 elements that fits the consensus for a Sre1 regulatory element (SRE) [26]. Importantly, this sequence was different in Tf2–11 (5′-ATCGTAGATA-3′) and did not match the SRE consensus (Figure S1). In vitro DNA binding assays confirmed that Sre1 bound to the Tf2 SRE sequence, but not to the sequence present in Tf2–11 (Figure 3B). Furthermore, chromatin immunoprecipitation experiments demonstrated that Sre1 bound to Tf2 LTR in vivo under low oxygen conditions, but not to Tf2–11 LTR (Figure 3C). Thus, Tf2–11 contains a natural mutation in the Sre1 DNA binding sequence that prevents low oxygen regulation of this transposon [30]. Collectively, these data demonstrate that Tf2 LTR functions as an oxygen-dependent promoter that is directly regulated by Sre1.
Given that low oxygen induced expression of Tf2 mRNA and protein, we next tested whether induction of Tf2 by Sre1 resulted in increased element transposition. Previous studies examined fission yeast Tf1 or Tf2 mobilization when these elements were highly overexpressed in the presence of oxygen from a heterologous promoter [23,32,33]. Interestingly unlike Tf1, the majority of Tf2 mobilization events (>70%) did not require the Tf2 integrase and thus occurred by cDNA recombination [23]. Here, we measured the ability of an endogenous Tf2 element to mobilize in response to changes in environmental oxygen. To monitor transposition, we inserted an intron-containing neomycin resistance gene into the Tf2–12 3′ UTR in the opposite orientation to the Tf2–12 ORF (Figure S2). This neomycin resistance gene was interrupted by an artificial intron (neoAI) in the antisense direction that is spliced out of the Tf2–12 mRNA [34]. In this way, cells become G418 resistant only when Tf2–12 has mobilized and inserted into the genome via a cDNA intermediate.
To measure transposition, we cultured Tf2–12-neoAI cells in the presence or absence of oxygen for 8 h and then plated cells on selective medium containing G418 to determine the frequency of transposition. The basal aerobic frequency of Tf2–12-neoAI transposition (2.5 × 10−8/cell) increased 18-fold (44 × 10−8/cell) under low oxygen (Figure 4A). As expected, this oxygen-dependent increase in transposition frequency required Sre1. First, cells lacking sre1+ showed no increase in transposition under low oxygen. Second, deletion of the Sre1 DNA binding sequence from the Tf2 5′ LTR resulted in the loss of oxygen-dependent induction of transposition (Tf2 ΔSRE, Figure 4A). Importantly, this assay monitored only one of the 12 oxygen-responsive Tf2 elements in the fission yeast genome [30]. Thus, we expect the actual transposition frequency under low oxygen to be >10-fold higher (∼5 × 10−6/cell).
Southern blot analysis for the neo gene in nine independent G418-resistant colonies derived from a low oxygen culture revealed that each strain contained at least one novel Tf2–12-neo insertion not present in the parent strain (Figure 4B). Using a combination of Southern blotting, PCR-based screening, and ligation-mediated PCR, we determined the location of the spliced Tf2–12-neo cassette in 20 randomly selected low oxygen clones (22 total insertion events). All of the insertion events appeared to result from homologous recombination of Tf2 cDNA. By identifying sequences downstream of the Tf2–12-neo 3′ LTR, we determined that 12 insertion events occurred upstream of an existing Tf2 resulting in tandem Tf2 elements. Since S. pombe contains two tandem transposons (Tf2–8 and Tf2–7), these 12 events represent either a replacement of Tf2–8 or a new insertion upstream of an existing Tf2 element. In the remaining ten events, the Tf2–12-neo cDNA replaced an existing Tf2 element: Tf2–1 (two events), Tf2–7 (one event), Tf2–9 (one event), Tf2–10 (one event), Tf2–11 (one event), and Tf2–12 (four events). Replacements may occur preferentially at Tf2–12 due to the additional homology that exists between Tf2–12-neoAI and the Tf2–12-neo cDNA as compared to other Tf2 elements.
To test directly whether recombination was required for Tf2 mobilization, we determined the transposition frequency of Tf2–12-neo in cells lacking rhp51+, which is required for homologous recombination in S. pombe [35]. The Tf2–12-neo transposition frequency under low oxygen decreased 29-fold in rhp51Δ strain, and mobilization remained oxygen-dependent (Figure 4A). This decrease in transposition frequency indicated that Tf2 mobilization requires homologous recombination and was consistent with our mapping of these Tf2–12-neo cDNA insertions (n = 6) to existing Tf2 loci. In addition, the Tf2-neo insertions lacked new target site duplications flanking the elements, which are characteristic of integrase-mediated insertion events. Based on the high amino acid sequence identity among the 12 copies of Tf2 (99%), we infer that other Tf2 elements mobilize by recombination. A similar mechanism of cDNA mobilization has been observed for Ty elements in S. cerevisiae [36,37]. Mobilization of Tf2 by cDNA recombination to existing elements has been termed “integration site recycling” and may serve as a mechanism to protect the host cell genome while allowing Tf2 elements to evolve [23]. Collectively, these data demonstrate that Sre1 induces mobilization of Tf2 retrotransposons by homologous recombination in response to low oxygen.
The presence of 13 Tf2 elements and 35 Tf2 solo LTRs in the S. pombe genome suggests that a small fraction of mobilization events occur at new positions in the genome [30]. Unlike most characterized retrotransposons [38], fission yeast Tf elements preferentially insert upstream (∼100–400 bp) of RNA polymerase II transcribed genes [30,32,33]. Together with our data, this insertion site bias for RNA polymerase II promoters suggested that Sre1 may control low oxygen expression of genes adjacent to solo LTRs. To test this, we took advantage of the fact that solo LTRs can be formed by homologous recombination. We generated a new Tf2 solo LTR from the ura4+-tagged Tf2–6 element by counterselecting for expression of ura4+ on medium containing 5-fluoroorotic acid (Figure 5A) [39]. Next, we tested the ability of Tf2–6 solo LTR to promote transcription of adjacent sequences using quantitative RT-PCR and primers adjacent to the LTR. We detected Sre1-dependent, oxygen-dependent transcription downstream of Tf2–6 solo LTR, demonstrating that solo LTRs can direct transcription of non-Tf2 sequences (Figure 5B).
To test the promoter capabilities of preexisting solo LTRs in the fission yeast genome, we identified 25 solo LTRs that resembled Tf2 LTR and contained an intact SRE [30]. Using primers to adjacent noncoding downstream sequences, we detected transcripts from 20 solo LTRs by real-time reverse-transcriptase PCR (RT-PCR), and expression of 16 out of 20 transcripts was oxygen dependent (Figure 5C). Transcripts from four solo LTRs were not regulated by oxygen, possibly due to local chromatin effects, stability of LTR-specific transcripts, or the position of the amplifying primers. These results demonstrate that solo LTRs scattered across the S. pombe genome are functional oxygen-dependent promoters.
To investigate the genome-wide impact of Tf2 LTRs on low oxygen gene expression, we designed an experiment to identify RNA transcripts containing Tf2 LTR sequences using a S. pombe genomic tiling array. Briefly, we amplified cDNA from wild-type cells grown under low oxygen using a Tf2 LTR-specific forward primer and a random reverse primer (Figure S3; Materials and Methods). The amplified DNA was labeled and used to probe the Affymetrix S. pombe genome tiling array. The labeled probes should identify regions of the genome encoded in Tf2 LTR-containing RNAs. To eliminate artifacts due to nonspecific LTR primer binding, two tiling array experiments were performed using two different Tf2 LTR forward primers. The sequences presented here were identified in both experiments. As expected, we identified each of the Tf2 transposons, validating our methodology. Tf2–11 was also identified, presumably due to its basal expression (Figure 2C) or to cross-hybridization with probes from other Tf2 elements.
Importantly, the tiling experiment also identified four additional open reading frames (SPCC11E10.07c, SPAC1B3.08, SPAC823.14, and SPAC2E1P3.02c) using our cut-off criteria (p < 0.05), and each of these four genes were positioned downstream of a Tf2 LTR. Using RT-PCR and gene-specific primers, we confirmed that each gene was encoded in an oxygen-dependent transcript that originated from a Tf2 LTR. Northern analysis for the first gene SPCC11E10.07c, which codes for the alpha subunit of the translation initiation factor eIF2B, detected a novel oxygen-dependent transcript (Figure 5D, arrow upper panel). This low oxygen transcript represented 19% of the total message and originated from the upstream solo Tf2 LTR as confirmed by RT-PCR (Figure 5D, lower panel). The solo LTR for SPCC11E10.07c corresponds to PCRC_038 in Figure 5C, which showed a 55-fold increase in expression under low oxygen. Interestingly, GCN3, the S. cerevisiae homolog of SPCC11E10.07c, has a Ty1 solo LTR positioned 499 bp upstream.
Northern blot analysis for the second gene SPAC1B3.08 showed a pattern similar to SPCC11E10.07c with an oxygen-dependent upper transcript representing 16% of the total (unpublished data). The corresponding upstream solo LTR for SPAC1B3.08 is PARW_093 in Figure 5C. For the third gene SPAC823.14, we confirmed an LTR-derived low oxygen transcript originating from the upstream solo LTR PALW_049 by RT-PCR (Figure 5C). However, we were unable to detect a novel low oxygen transcript for SPAC823.14 by northern blotting, possibly because of the lower sensitivity of northern analysis. Finally, amt3+/SPAC2E1P3.02c, which encodes ammonium transporter 3, is positioned downstream of the Tf2–3 element and not a solo-LTR. Northern analysis revealed a longer, major transcript under low oxygen, which accounts for 78% of the total low oxygen transcript (Figure 5D, arrow upper panel). RT-PCR analysis confirmed that this upper transcript originated in the Tf2–3 LTR. Deletion of the solo LTR PCRC_038 and the Tf2–3 element resulted in the loss of the upper oxygen-dependent transcripts for SPCC11E10.07c and amt3+, respectively, consistent with transcription initiating within the Tf2 LTR (unpublished data). The functional significance of these oxygen-dependent, LTR-derived transcripts remains to be determined. In addition to the examples mentioned above for open reading frames, we detected LTR-derived transcripts from many of the solo LTRs examined in Figure 5C. However, these transcripts had low signal intensity and did not make our cut-off, possibly due to increased turnover of these noncoding RNAs. Together, these data establish Tf2 LTRs as promoters that can direct oxygen-dependent transcription of adjacent genes, demonstrating the ability of Tf2 transposons to regulate the S. pombe transcriptome.
Retrotransposons in different organisms have been shown to respond to a variety of environmental signals and stresses [10,40]. We report that the hypoxic transcription factor Sre1 directly controls the low oxygen induction of Tf2 retrotransposon expression and mobilization. In a genome-wide transcriptional analysis of environmental stress responses, Tf2 transposons were shown to be upregulated by heat and peroxide stress (∼4-fold), but not by heavy metal, osmotic stress, or a DNA alkylating agent [41]. We observed a similar induction of Tf2 transcription by heat stress, but this upregulation did not require Sre1, suggesting that other factors may regulate Tf2 expression. Peroxide stress also induced Tf2 transcription in a Sre1-dependent manner, consistent with the fact that hydrogen peroxide activates Sre1 (unpublished data). In addition, Tf2 was not upregulated by treatment of cells with the endoplasmic reticulum stress inducer tunicamycin. Thus, Sre1-dependent induction of Tf2 is a specific response, as Tf2 transcription is not broadly affected by environmental stress.
Accumulating evidence implicates transposable elements as regulators of gene expression in eukaryotes as diverse as plants and humans [42,43]. Transposable elements contribute to genomic evolution by donating regulatory elements, providing alternative promoters, or causing mutations by inserting into genes [44]. Our results now provide evidence for regulation of endogenous gene expression by transposons in fission yeast. Here, we report the regulation of Tf2 retrotransposons by oxygen and demonstrate that Tf2 LTRs direct low oxygen transcription of adjacent coding sequences. Given that the Tf family of transposons insert upstream of RNA polymerase II promoters, we speculate that Tf2 insertions may provide a mechanism for generating new oxygen-dependent gene expression.
Yeast strains, media, and standard procedures including northern blotting, western blotting, and β-galactosidase assays have been described previously [26,27]. Table S1 contains sequences of oligonucleotides used.
To tag individual Tf2 elements, the 1.8-kb ura4+ cassette was inserted upstream of the Tf2 ORF by homologous recombination using standard techniques [45]. The location of ura4+ insertion was confirmed by PCR using unique forward primers designed upstream of each transposon and a common reverse primer in the ura4+ cassette.
To generate the Tf2–6 solo LTR, the ura4+ tagged Tf2–6 strain was plated on Edinburgh minimal medium containing 1 mg/ml 5-fluoroorotic acid at a density of 106 cells per plate. The 5-fluoroorotic acid–resistant colonies were streaked for singles and the absence of Tf2–6 was confirmed by sequencing the PCR product across the transposon. This strain is referred to as Tf2–6 solo LTR.
The neoAI tagging of Tf2 and the transposition assay were modified from established protocols [23]. The Tf2–12 neoAI was tagged by homologous recombination following transformation of a linear DNA fragment assembled from the following DNA fragments: 670 bp of Tf2–12 (bp 3871–4534), neoAI cassette [23], 3′ UTR and 3′ LTR of Tf2–12 (bp 4535–4900), 1.8-kb HindIII ura4+ cassette from pREP4x [46], and the 500 bp downstream of Tf2–12 on the chromosome. These fragments were assembled in pBluescript (Stratagene, http://www.stratagene.com) and the linear fragment used for transformation was released with ApaI and SacI. Transformants obtained on selective medium lacking uracil were screened by PCR and confirmed by Southern blotting [47]. These clones were used as parents for transposition assays. A diagram of the tagged locus is shown in Figure S2.
The strain Tf2 ΔSRE contains Tf2–12 neoAI, in which the Sre1 DNA binding site (SRE) has been deleted from the 5′ LTR of Tf2–12. Tf2 ΔSRE was made using the Cre-loxP method for marker rescue [48]. Using this technique, a strain was created that contains a deletion of the 10-bp SRE sequence (ATCGTACCAT) in the Tf2–12 5′ LTR (Figure S1) and an insertion upstream of the Tf2–12 5′ LTR consisting of the 34-bp loxP sequence (5′-ATAACTTCGTATAGCATACATTATACGAAGTTAT-3′) and 19 bp of plasmid sequence (5′-CGAAGTTGAATTCCTGCAG-3′). This strain was then transformed with the ApaI-SacI cassette described above to introduce neoAI in the 3′ UTR of Tf2–12, resulting in the strain Tf2 ΔSRE.
Tf2–12-neoAI yeast cells were cultured in the presence or absence of oxygen for 8 h. Yeast (1 × 107 cells) were plated on rich medium (YES) containing 100 μg/ml G418 [27]. After 16 h, cells were replica plated to a second YES+G418 plate, and G418-resistance was confirmed by retesting individual colonies. Four independently derived Tf2–12-neoAI strains were used and the data were pooled. For the Tf2 ΔSRE experiment, two independent strains were used. A total of ∼5 × 108 cells/genotype was scored for mobilization both +/− oxygen. Genomic DNA from these independent G418 resistant clones were digested with HindIII and processed for Southern blotting. To identify the site of integration, a combination of inverse PCR and vectorette PCR was performed. Briefly, for inverse PCR, DNA was digested with restriction enzyme TaqI, self-ligated, and amplified using two divergent oligos positioned in the neoAI gene cassette. Vectorette PCR protocol was a modification of previously described methods [49]. Genomic DNA digested with EcoRI or SpeI was used for ligation with vectorette oligos. The PCR products obtained from both of these techniques were cloned using TopoTA (Invitrogen, http://www.invitrogen.com) vector and sequenced.
Assay was performed as described previously with minor modifications [26]. Binding of Sre1 to Tf2 LTR sequence was quantified by real-time PCR using Tf2 specific LTR primers. Fraction bound was calculated using the formula 2(Ctinput − Ctpulldown) for each treatment. All the values obtained for aerobic samples with anti-Sre1 antibodies were set to one and the other values were normalized accordingly.
cDNA was made from DNAse-treated RNA using Superscript II (Invitrogen). The cDNA used for detecting solo LTR derived transcripts was made with random hexamers. This was then used for real-time PCR amplification using a Bio-Rad iQ Cycler (http://www.bio-rad.com) and SyBr-Green mix from ABgene (http://www.abgene.com). cDNA preparations made without reverse transcriptase were used to control for signal from any contaminating DNA. Oligos to hcs1+ were used to normalize between reactions (hcs1+ transcript levels were unchanged among the tested conditions)[27]. Forward oligos used to detect solo LTR transcripts were positioned ∼20 bp downstream of LTR and yielded ∼100-bp products. To confirm the tiling array candidate transcripts, oligo dT-primed cDNA served as template for PCR with transcript or LTR-specific primers.
To detect transcripts containing LTR sequence, cDNA was synthesized using mRNA isolated from yeast cells grown in the absence of oxygen for 8 h and an oligonucleotide containing random hexamers with a 5′ adapter sequence (Figure S3 and Table S1). Following purification, the first-strand cDNA was copied to double-stranded DNA using a LTR-specific oligo and four cycles of amplification. This DNA was amplified for eight cycles using the LTR-specific oligo and the adapter oligo with Platinum Taq (Invitrogen). The resulting amplicons were labeled using Affymetrix cDNA labeling kit (http://www.affymetrix.com) and hybridized to a GeneChip S. pombe Tiling 1.0FR Array using manufacturer's protocols (Affymetrix). This high-density tiling array contains 25-mer oligos for both DNA strands that overlap by 5 bp, giving approximately 20-bp resolution. The data were analyzed using Partek GS software (Partek, http://www.partek.com/). Positive transcripts were identified using a test region of 100 bp, a signal intensity >3.8, and p-value <0.05, for continuous signals present over >200-bp regions. Two independent experiments were performed using different, nonoverlapping LTR-specific oligos to control for nonspecific amplification. |
10.1371/journal.pbio.1001897 | Inhibition of Plasmepsin V Activity Demonstrates Its Essential Role in Protein Export, PfEMP1 Display, and Survival of Malaria Parasites | The malaria parasite Plasmodium falciparum exports several hundred proteins into the infected erythrocyte that are involved in cellular remodeling and severe virulence. The export mechanism involves the Plasmodium export element (PEXEL), which is a cleavage site for the parasite protease, Plasmepsin V (PMV). The PMV gene is refractory to deletion, suggesting it is essential, but definitive proof is lacking. Here, we generated a PEXEL-mimetic inhibitor that potently blocks the activity of PMV isolated from P. falciparum and Plasmodium vivax. Assessment of PMV activity in P. falciparum revealed PEXEL cleavage occurs cotranslationaly, similar to signal peptidase. Treatment of P. falciparum–infected erythrocytes with the inhibitor caused dose-dependent inhibition of PEXEL processing as well as protein export, including impaired display of the major virulence adhesin, PfEMP1, on the erythrocyte surface, and cytoadherence. The inhibitor killed parasites at the trophozoite stage and knockdown of PMV enhanced sensitivity to the inhibitor, while overexpression of PMV increased resistance. This provides the first direct evidence that PMV activity is essential for protein export in Plasmodium spp. and for parasite survival in human erythrocytes and validates PMV as an antimalarial drug target.
| To survive within human red blood cells, malaria parasites must export a catalog of proteins that remodel the host cell and its surface. This enables parasites to acquire nutrients from outside the cell and to modify the cell surface in order to evade host defenses. Protein export involves proteolytic cleavage of the Plasmodium Export Element (PEXEL) by the aspartyl protease Plasmepsin V. We report here the development of a small molecule inhibitor that closely mimics the natural PEXEL substrate and blocks the activity of Plasmepsin V from the malarial parasites Plasmodium falciparum and Plasmodium vivax. The inhibitor impairs export and cellular remodeling and kills P. falciparum at the ring-trophozoite transition, providing direct evidence that Plasmepsin V activity is essential for export of PEXEL proteins and parasite survival within the host. These findings validate Plasmepsin V as a highly conserved antimalarial drug target.
| Each year malaria parasites cause several hundred million infections and over 650,000 deaths [1]. Plasmodium falciparum causes the most lethal malaria and is endemic in Africa [2]. Plasmodium vivax causes most malarial deaths outside Africa and is associated with liver-stage hypnozoites [3]. Although chloroquine and artemisinin have been effective antimalarials, their decreasing efficacy [4],[5] emphasizes the need for therapies against novel targets shared by both Plasmodium spp.
Malaria parasites develop in erythrocytes within a parasitophorous vacuole and export over 450 proteins to the cell (reviewed in [6],[7]). Export utilizes an N-terminal motif called the Plasmodium export element (PEXEL; RxLxE/Q/D) [8] or Vacuolar transport signal (VTS) [9]. Exported proteins are cleaved in the PEXEL after Leu (RxL↓) in the endoplasmic reticulum (ER) [10], which requires the conserved Arg and Leu residues [11]. PEXEL cleavage is performed by the aspartyl protease Plasmepsin V (PMV) [12],[13]. PEXEL-containing proteins and PMV are conserved in all Plasmodium spp. [8],[14]–[16]. Repeated attempts to disrupt the PMV gene have failed, suggesting it is essential [12],[13],[16], but direct and decisive proof is still lacking. A functional survey of P. falciparum exported proteins indicated that 25% or more are essential for parasite survival in human erythrocytes [17]. The current P. falciparum PEXEL exportome is predicted to be 463 proteins [18]; thus, possibly 100 or more exported parasite proteins are required for development in erythrocytes.
Some exported proteins lack a PEXEL, for example, skeleton binding protein 1 (SBP1) and the major virulence adhesin family known as P. falciparum erythrocyte membrane protein 1 (PfEMP1). PfEMP1 is expressed on the erythrocyte surface and mediates cytoadherence to microvascular endothelia, causing severe malaria [19]. PfEMP1 is thought not to be cleaved by PMV [18], but its transport to, and expression on, the erythrocyte surface requires exported PEXEL and PEXEL-negative proteins (reviewed in [7],[20]).
Aspartyl proteases can be inhibited by transition-state isosteres in which the scissile bond is replaced by a noncleavable moiety. Examples include statine (Sta)-containing inhibitors and several are now in clinical use [21].
Here, we developed a transition-state inhibitor that potently blocks PMV from P. falciparum and P. vivax. The inhibitor demonstrates that PMV activity is essential for protein export, PfEMP1 surface display, cytoadherence, and parasite survival in human erythrocytes.
The PMV gene is present in all Plasmodium spp.; however, only the P. falciparum enzyme (PfPMV; Pf3D7_1323500) has been characterized. A multiple alignment of PfPMV with putative P. vivax PMV (PvPMV; PVX_116695) indicated that they share 82.2% similarity, 54.7% identity (Figure S1). Both proteins are predicted to contain a signal peptide, an aspartyl protease domain with DTG and DSG residues defining the catalytic dyad, and a C-terminal transmembrane domain (Figure 1A). Due to four insertions, PfPMVHA is predicted to be approximately 7.5 kDa larger than PvPMVHA (Figure 1A); however, following signal peptide removal, PfPMVHA is predicted to be 8 kDa larger than PvPMVHA.
To determine whether PvPMV is an ortholog of PfPMV, we expressed it in P. falciparum fused to 3× hemagglutinin (HA) tags (Figure 1A). As a positive control, we expressed PfPMV fused to 3× HA tags (Figure 1A) [13]. Expression of PfPMVHA and PvPMVHA was confirmed by immunoblot using anti-HA antibodies (Figure 1B). PvPMVHA was ∼8 kDa smaller than PfPMVHA, as predicted (Figure 1B).
PfPMV was previously localized to the ER using a mouse anti-PfPMV antibody that colocalizes with BiP [16] and ERC [13]. To further study PMV, we developed a rabbit antibody that was specific for PfPMV (Figure 1C, compare lanes 1 and 2) and colocalizes with the ER signal from the mouse PfPMV antibody (Figure 1D, top) but does not cross-react with PvPMVHA (Figure S2A). Using anti-HA antibodies, a strong perinuclear signal was observed in parasites expressing PvPMVHA or PfPMVHA (Figure 1D, middle panels, red). Both proteins colocalized with rabbit anti-PfPMV antibodies, indicating the location was the ER (Figure 1D). PvPMVHA also colocalized with ERC (Figure 1D, bottom), as shown previously for PfPMVHA [13].
To investigate PvPMVHA activity, we affinity purified it, as well as the PfPMVHA control, using anti-HA agarose, as previously described [13],[18]. Immunoblot with anti-HA and anti-PfPMV antibodies showed that the purified proteins were species-specific (Figure S2A). The proteins were incubated with a fluorogenic peptide of nine amino acids that contained the PEXEL sequence from knob-associated histidine-rich protein (KAHRP), and efficient processing was observed by both enzymes (Figure 1E). Cleavage of KAHRP by PfPMVHA was previously shown to occur after Leu (RTL↓) by mass spectrometry [13], and this position was also confirmed for PvPMVHA (Figure S2B–E). Km values of 9.7 (±3.0) and 11.7 (±1.8) µM were calculated for PvPMVHA and PfPMVHA, respectively (Figure S2F). In contrast, no processing was observed when the PEXEL Arg and Leu residues were mutated to Ala (ATAAQ), consistent with the substrate specificity of PfPMV (Figure 1E). To verify that processing was due to HA-tagged PMV rather than other co-precipitated proteases, we expressed in P. falciparum a mutant of PvPMVHA or PfPMVHA [13], where one or both catalytic Asp residues were mutated to Ala (see Figure 1A,B). Following affinity purification, the mutant enzymes were incubated with KAHRP PEXEL peptides, but no processing was observed (Figure 1E; blue, white), confirming that the PEXEL-dependent cleavage activity observed for each protease was attributable only to HA-tagged PMV.
The substrate specificity of PfPMV is restricted, such that even minor changes to the PEXEL sequence markedly reduces cleavage efficiency—that is, Arg to Lys (R>K) or Leu to Ile (L>I) [18]. We assessed whether PvPMVHA shares this feature. Indeed, PvPMVHA poorly cleaved peptides possessing R>K or L>I mutations (Figure 1E; KTLAQ, RTIAQ). Collectively, these data show that PvPMV localizes to the ER and cleaves the PEXEL with the same restricted specificity as PfPMV.
While maintaining P. falciparum cultures overexpressing PfPMVmutHA, we noticed a delay in growth, suggesting a possible dominant negative phenotype, which has been reported previously with a different PfPMV catalytic mutant [12]. A flow cytometry-based growth assay revealed that parasites expressing PfPMVmutHA with WR99210 selection grew to a parasitemia 2.6-fold less than parasites expressing a similar episomal construct, encoding a mini PfEMP1 reporter fused to 3× HA tags (miniVarHA) with WR99210 selection (p<.0001; Figure 1F). This demonstrated that overexpression of inactive enzyme conveyed a growth disadvantage, providing evidence that PMV is important for parasite survival. Analysis of PMV protein levels in parasites overexpressing the PMVHA transgenes indicated that PvPMVHA and PvPMVmutHA had no effect on endogenous PfPMV levels; however, overexpression of inactive PfPMVmutHA caused a clear decrease in expression of the endogenous enzyme (Figure 1C, compare lanes 1 and 4, and Aldolase loading controls), indicating a negative feedback mechanism occurs in these parasites.
The conserved P3 Arg and P1 Leu residues in the PEXEL (see Figure 2A for a description of nomenclature) are crucial for PMV activity [12],[13]. We developed a homology model and designed compounds with a transition-state isostere that mimics the natural PEXEL substrate with the aim of inhibiting PMV. One mimetic, WEHI-916 (Figure 2B), consisted of Arg that could bind in the S3 pocket of PfPMV, Val that would position in the S2 pocket, and Leu-Statine (Leu-Sta), to engage the S1 pocket and inhibit both catalytic Asp residues of PMV (Figure 2C).
As control compounds, we synthesized analogs similar to 916 but that mimic noncleavable PEXEL mutant substrates, with the aim that they would be poor PMV inhibitors; the first replaced the P3 Arg with Lys (R>K; WEHI-024) and the second replaced the P1 Leu with Ile (L>I; WEHI-025; Figure 2B). These analogs were designed on the basis that mutations of the conserved PEXEL residues R>K or L>I almost completely inhibit cleavage by PMV (Figure 1E and [18]) and should therefore have lower affinity for PMV.
Each compound was incubated with PfPMVHA in the presence of KAHRP PEXEL peptides. WEHI-916 (henceforth 916) potently inhibited PEXEL cleavage by PfPMVHA with a 50% inhibitory concentration (IC50) of 20 nM (Figure 2D). In contrast, WEHI-024 and WEHI-025 (henceforth 024 and 025, respectively) had weak activity (IC50>100 µM and 1.11 µM, respectively; Figure 2D). 916 inhibited PvPMVHA with an IC50 of 24 nM, whereas 024 and 025 again had weak activity (Figure 2D).
To investigate potential off-target activity against human aspartyl proteases, the compounds were assessed against beta-secretase (BACE-1), for which PMV has distant relatedness [12], Cathepsin D, and two human cell lines; the compounds displayed poor activity (IC50>100 µM for BACE-1; 25 µM for Cathepsin D) and had negligible toxicity against human HEpG2 and fibroblast cell lines (Figure 2D). Collectively, this demonstrated that 916 potently inhibited PMV and had low off-target activity against BACE-1 and Cathepsin D, whereas the closely related analogs 024 and 025 were poorly active.
To assess whether 916 could inhibit PMV in P. falciparum–infected erythrocytes, parasites expressing the PEXEL protein P. falciparum erythrocyte membrane protein 3 (PfEMP3) fused to green fluorescent protein (GFP) [13] were treated with increasing concentrations of inhibitor, and PEXEL processing was evaluated by immunoblot. A dose-dependent increase in unprocessed PfEMP3-GFP was observed (black arrow, Figure 3A), which was the same size as uncleaved PEXEL R>A mutant PfEMP3-GFP (Figure 3A) [13]. The level of PEXEL-cleaved protein (blue arrow, Figure 3A) did not quantitatively reflect the degree of PMV inhibition, as inhibitor was added well after PEXEL processing and export of PfEMP3-GFP had initiated. A GFP-only band, representing degraded chimera in the food vacuole, was also observed at ∼26 kDa (Figure 3A). Together, this demonstrated that PEXEL processing was impaired by 916 treatment and that engagement of PMV occurred in P. falciparum–infected erythrocytes.
To understand the timing required for PMV inhibition in P. falciparum, parasites were treated with 916 for 1–5 h, and cleavage was evaluated by immunoblot. No effect was seen after 1 h; however, uncleaved PfEMP3-GFP increased between 2 and 5 h (Figure 3B), indicating 916 accessed the parasite ER slowly. Inhibition of PfEMP3-GFP cleavage by 916 was rescued following culture in inhibitor-free medium, to approximately 50% after 2 h (Figure S3A), indicating that cleavage inhibition was reversible or that additional active PMV was synthesized during the experiment.
We next assessed whether the control analogs 024 and 025, which were poor inhibitors of PMV in vitro, had an effect on PEXEL cleavage in parasites. Although a dose-dependent effect was again observed with 916, analogs 024 and 025 had no effect on PEXEL processing of PfEMP3-GFP or KAHRP-GFP, even at 50 µM (Figure 3C). In the case of KAHRP-GFP, 916 treatment caused accumulation of both uncleaved (black arrow) and signal peptide-cleaved (red arrow) protein, which were the same size as bands observed for PEXEL R>A mutant KAHRP-GFP (Figure 3C, right). These bands were shown previously to be uncleaved and signal peptide-cleaved KAHRP-GFP, respectively, by mass spectrometry [11].
As 916 treatment caused accumulation of both uncleaved and signal peptide-cleaved species of PEXEL proteins, the potential for off-target effects against signal peptidase was investigated using parasites expressing SERA5s-GFP. This protein contains a signal peptide but lacks a PEXEL and is efficiently secreted to the parasitophorous vacuole (Figure S3B). 916 treatment did not impair processing of the signal peptide from SERA5s-GFP (Figure 3D, see position of black arrow), indicating it cannot inhibit signal peptidase. Taken together, this shows that 916 can effectively inhibit PMV, but not signal peptidase, in P. falciparum–infected erythrocytes and that 024 and 025 have no affect on PMV or signal peptidase activity at concentrations up to 50 µM.
The rate of PEXEL protein synthesis, ER import, and processing by PMV in P. falciparum is unknown. We evaluated these processes by radiolabeling parasite proteins in culture for 0.5–15 min before immunoprecipitating PfEMP3-GFP with anti-GFP agarose, visualizing bands by autoradiography and quantifying them by densitometry. Labeled PfEMP3-GFP became evident 1 min after addition of label to the culture medium and increased exponentially throughout the experiment (Figures 3E and S3D). Uncleaved PfEMP3-GFP (black arrow) was faint and the major species was a doublet of approximately 33 kDa (red arrow; signal peptide-cleaved) and 29 kDa (blue arrow; PEXEL-cleaved; Figure 3E). This showed that signal peptidase cleaves PfEMP3 within seconds (<1 min) of protein synthesis (i.e., cotranslationaly), but this molecular species may be transient, as it was not detected by immunoblot of 916-treated parasites, or in PEXEL R>A mutant protein (Figure 3A–C). The radiolabeled bands on the 35S-membrane were confirmed to be GFP-specific by immunoblot (Figure S3C). PEXEL-cleaved PfEMP3-GFP (blue arrow) was also evident 1 min after addition of label to the culture medium and increased exponentially, indicating PMV cleavage was also rapid and likely cotranslational (Figures 3E and S3D). The proportion of PEXEL-cleaved protein increased slightly as signal peptide-cleaved protein decreased (Figure 3E), suggesting signal peptidase cleaves before PMV and that PMV may cleave after signal peptidase.
Addition of 916 to parasites for 5 h prior to radiolabeling caused accumulation of uncleaved PfEMP3-GFP in parasites (black arrow), which was evident 1–2 min postlabeling (Figure 3F). After a total of 15 min, the degree of PMV inhibition in parasites was quantified by densitometry, and a 13-fold decrease in PEXEL cleavage was observed compared to labeling without 916 (Figure 3G), indicating PMV was inhibited. The signal peptide-cleaved and PEXEL-cleaved species were only weakly detectable throughout the experiment and were not visible until 3–5 min postlabeling (Figure 3F), compared to a stronger signal within 1–2 min of labeling in the absence of inhibitor when the same quantity of parasites was used (Figure 3E). This indicated that 916 significantly blocked PMV in P. falciparum and caused a delay in protein synthesis, ER import, or N-terminal processing of PfEMP3-GFP.
Processing of uncleaved PfEMP3-GFP within the PEXEL was rescued after 15 min of culture in inhibitor-free medium (Figure S3E,F), indicating that PMV is able to process full-length PfEMP3-GFP.
Having demonstrated that 916 can directly engage PMV in P. falciparum–infected erythrocytes, the effect of the inhibitor on parasite viability was examined by treating early ring parasites for 72 h and assessing parasitemia by flow cytometry. 916 killed parasites with half maximal effective concentration (EC50) of 2.5–5 µM (Figure 4A). Analogs 024 and 025 had negligible effect on parasite viability at concentrations up to 20 µM, where 916 completely killed parasites; however, they had EC50 values of 66 µM and 30 µM, respectively, indicating they adversely affected parasite growth at high concentrations. Because PEXEL processing in parasites was unaffected by 024 and 025 treatment, even at 50 µM (Figure 3C), we conclude that the analogs impart toxicity at high concentrations independent of PMV.
To determine the stage of the parasite lifecycle that 916 exerted its toxic effects, ring parasites were treated with 15 µM 916 for increasing times through the 48 h cycle and then cultured in inhibitor-free medium to a total of 72 h to see if parasites could recover. Parasites grown in 916 for 1–20 h completely recovered and grew like DMSO-treated controls; however, treatment for >23 h adversely affected growth (Figure 4B), indicating the timing of killing began after 20 h of age, at the ring-trophozoite transition. Analogs 024 and 025 did not affect growth at any parasite stage at the concentration used (15 µM), whereas chloroquine and artemisinin killed parasites when added to rings for only 30 min before addition of inhibitor-free medium (Figure 4B, see 0 h). This indicated the controls either killed rapidly (artemisinin) or were retained inside parasites and killed later, as chloroquine is reported to kill trophozoites [22],[23]. Both profiles were clearly different than that observed for 916.
Toxicity by 916 after the ring-trophozoite transition was then investigated by adding compound to parasites at different time points through the 48 h cycle. At 48 h, inhibitor-free medium was added and parasitemia for all conditions was determined at 72 h. Toxicity decreased when 916 was added to parasites aged beyond 24 h and schizonts were resistant, indicating 916 did not affect merozoite egress or reinvasion (Figure 4C). As expected, the addition of 916 to rings or early trophozoites was lethal (Figure 4C). 024 and 025 did not have any effect on parasite growth at the concentration used, whereas all parasite stages were sensitive to chloroquine and artemisinin (Figure 4C).
Light microscopy of parasites following treatment of early ring stages with 916 revealed a normal ring-stage morphology after 16 h; however, treatment for 32 h revealed a blockage in the ring-trophozoite transition and the majority of parasites appeared pyknotic and abnormal (Figure 4D). As greater than 50% of parasites could not recover from this treatment condition (refer to Figure 4B), the majority of parasites with this appearance were dying or dead. Treatment with DMSO, 024, or 025 had no effect on development by 32 h at 15 µM (Figure 4D). The morphology of parasites treated with 916, 024, 025, and DMSO was distinctly different from that observed for E-64–treated parasites, which contained swollen food vacuoles from inhibition of haemoglobin breakdown [24] (Figure 4D, arrow). This indicated that parasite toxicity to 916 was unlikely due to off-target inhibition of those food vacuole proteases. Collectively, the toxicity profile seen in the above experiments defines the window of parasite death as between 20 and 30 h, consistent with perturbed protein export and erythrocyte remodeling [7].
To gain further insight into the effects of 916, 024, and 025 on parasites, we assessed global protein synthesis following drug treatment by radiolabeling parasite proteins. Treatment of trophozoites with inhibitor for 5 h prior to radiolabeling had no detectable effect on translation, even at 50 µM concentrations (Figure S4A), indicating the compounds are not direct inhibitors of the translation machinery. We next assessed protein synthesis following 23 h of drug treatment of ring stage parasites (aged 1–3 h old at the initiation of treatment). A minor reduction in translation was observed following 916 treatment, but not 024 or 025 treatment, even at 50 µM (Figure S4B). A small but concomitant decrease in the cytosolic protein, Aldolase, was also evident by immunoblot following 916 treatment but not 024 or 025 treatment (Figure S4B), suggesting that parasites were beginning to die from the 916 treatment (parasites ranged from 24–26 h old at this time point). Indeed, Giemsa smears following treatment identified a small proportion of pyknotic parasites in the population (results not shown). Treatment with Brefeldin A, which prevents retrograde trafficking and ER exit, for 23 h severely impaired translation and parasites appeared as dying rings (i.e., had not progressed to trophozoites). It is possible that PMV inhibition by 916 treatment has a similar but weaker effect to BFA, in that it causes the accumulation of uncleaved PEXEL precursors in the ER, perturbing ER transport, and that this negatively affects translation by an ER stress response [25]. An alternative possibility is that translation was decreased slightly as a result of parasites dying from 916-mediated impairment of erythrocyte remodeling. Either way, the profile for 916 was different to that seen for 024 and 025, even at 50 µM, indicating the latter analogs likely kill parasites via a different mechanism to 916.
Treatment of P. falciparum with 916 impaired PEXEL cleavage and killed parasites, strongly suggesting that PMV is essential. To investigate this phenotype further, conditional protein knockdown was attempted in P. falciparum using the RNA-degrading glmS ribozyme, which utilizes glucosamine (GlcN) as a cofactor [26]. DNA encoding 3× HA epitopes, a stop codon, and glmS was incorporated in frame at the 3′ of the PMV locus by homologous recombination (Figure S5A). Correct genomic integration was confirmed by immunoblot with anti-HA and anti-PfPMV antibodies (Figure 5A). To activate glmS, GlcN was titrated into the culture medium of trophozoites. From 75% to 90% PMV knockdown was achieved after 48 h using 4–6 mM GlcN, but higher concentrations adversely affected parasite HSP70 levels and were subsequently avoided (Figure S5B). Addition of 5 mM GlcN to trophozoites for 24 h reduced PMV levels in subsequent rings by approximately 80% and trophozoites by ∼90% but caused little knockdown in parasites expressing inactive glmS (M9) (Figure 5B) [26]. Protein export predominates in rings, when knockdown reached ∼80%; surprisingly, this substantial degree of knockdown did not significantly affect PEXEL processing or parasite growth rate (p = .6250; Figure 5C), indicating that the remaining PMV levels were sufficient to enable export and sustain parasite development. This demonstrates that PMV activity is potent in P. falciparum and that knockdown to approximately 20% of wild-type levels could not facilitate the characterization of PMV essentiality.
As 916 inhibited PMV in parasites (for example, by 13-fold, Figure 3G), the additive effect of PMV knockdown plus 916 treatment was investigated. Parasites expressing PMVHA-glmS were transfected with a construct encoding PfEMP3-GFP, and PEXEL processing was assessed by immunoblot. PEXEL processing of PfEMP3-GFP was barely affected by 48 h of PMV knockdown alone (Figure 5D, see “+GlcN,” 0 µM 916); however, addition of 916 to parasites for 5 h impaired PEXEL cleavage and this was significantly enhanced following knockdown of PMV (e.g., by 50% at 20 µM; Figure 5D). The quantity of PfEMP3-GFP expressed in the PMV knockdown (+GlcN) appeared slightly less than in parasites without knockdown (−GlcN), whereas the loading control HSP70 did not vary appreciably between conditions (Figure 5D).
Parasites expressing PMVHA-glmS were next assessed for toxicity to 916. The EC50 of 916 was reduced by 3.3-fold following PMV knockdown compared to no knockdown (Figure 5E). As a control, parasites expressing PMVHA-M9 were treated with 916 in the presence or absence of GlcN; the EC50 reduced by 1.4-fold in the presence of GlcN, indicating it had a minor effect. However, the enhancement of PEXEL cleavage inhibition and 3.3-fold sensitization of parasites to inhibitor following knockdown indicated that PMV is a direct target of 916 and that PMV inhibition is lethal to parasites.
We next investigated the possible effects of PMV overexpression on parasite sensitivity to 916. Although parasites expressing PfPMVHA do not overexpress enzyme, due to integration of the construct at the endogenous PMV locus [13], parasites expressing PvPMVHA from episomes also express wild-type levels of endogenous enzyme (Figure 1B,C) and therefore contain additional, active PMV in the ER. To control for the carriage of episomes and selection on WR99210, sensitivity to 916 was compared between parasites overexpressing a similar construct on episomes on WR99210 selection (encoding miniVarHA; see Figure 1F). The EC50 of 916 was 1.9-fold greater for parasites overexpressing PvPMVHA compared to parasites overexpressing the control construct, and 1.4-fold greater than wild-type 3D7 parasites without WR99210 selection, indicating that PMV overexpression had increased parasite resistance to 916 (Figure 5F).
Localization of parasite proteins in Maurer's clefts (MCs), which are parasite-induced membranous structures in the erythrocyte that facilitate protein trafficking, enables accurate quantification of export by immunofluorescence microscopy as the signal is concentrated in puncta [27]. To study export in P. falciparum, we investigated a novel PEXEL-containing protein with two transmembrane domains and unknown function, called Hyp8 (MAL13P1.61/PF3D7_1301700) [14],[28], that we hypothesized may localize to MCs. Transgenic parasites expressing Hyp8-GFP or Hyp8-HA were generated (Figure S6A). Immunoblotting revealed that Hyp8 is expressed in rings (Figure S6B), and immunofluorescence microscopy showed it is exported (Figure S6C) and colocalizes with SBP1 in MCs (Figure 6A). Immunoelectron microscopy confirmed that Hyp8 localizes in MCs (Figure 6A, right). Three independent attempts to delete the hyp8 gene were unsuccessful in this study, in addition to earlier reported attempts [17], suggesting that Hyp8 may be an essential exported protein.
The effect of 916 treatment on export in P. falciparum–infected erythrocytes was then examined. Early ring parasites were treated with inhibitor, and the subcellular Hyp8-GFP fluorescence was quantified by immunofluorescence microscopy (Figure 6B). 916 treatment caused a dose-dependent decrease of Hyp8-GFP in MCs (the GFP signal in puncta outside the EXP2-labelled parasitophorous vacuole membrane) compared to DMSO and 024 treatment (p<.0001; Figure 6C). A small but significant increase in nonexported GFP signal (the signal inside the EXP2 labeling) was observed as puncta of fluorescence internal to the parasitophorous vacuole membrane following 916 treatment (p<.0001; Figure 6D, see also arrows in B).
We next examined whether 916 treatment affected protein secretion in P. falciparum–infected erythrocytes by measuring the quantity of EXP2 signal at the parasitophorous vacuole membrane following treatment. There was no statistical difference in EXP2 signal between treatments (p = .0977; Figure 6E), indicating 916 specifically affected export but not secretion, under the conditions used.
As Hyp8 localizes to MCs, we quantified the number of GFP-positive MCs in infected cells following the drug treatments. The mean number of clefts was significantly reduced by 916 treatment, by up to 39%, compared to DMSO (p<.0001; Figure 6F). Collectively, these data demonstrated that 916 dramatically reduced the export of the PEXEL protein, Hyp8, and that MC development was impaired following treatment.
An important function of exported proteins in remodeling and virulence is to assemble the cytoadherence complex at the erythrocyte surface [17]. Because export of Hyp8 and MC formation was decreased following PMV inhibition, we investigated whether trafficking of PfEMP1 was also affected by quantifying its display on the surface of infected erythrocytes. Ring-stage CS2-GFP parasites [29] selected for expression of the PfEMP1 var2csa gene were treated with one of two sublethal doses of 916 (Figure S7A), and surface-expressed PfEMP1 was measured 24 h postinvasion using PAM1.4 antibodies [30] that specifically recognize VAR2CSA [31] by flow cytometry. PfEMP1 surface expression decreased in a dose-dependent manner, by up to 55%, following 916 treatment, but addition of DMSO and 025 had no effect (Figure 6G). Parasitemia across all treatment conditions was measured as GFP fluorescence by flow cytometry and was approximately equal at 24 h, confirming parasite viability (Figure S7B).
To evaluate whether decreased PfEMP1 surface expression affected cytoadherence of infected erythrocytes, static binding assays with purified CSA were performed [32],[33]. Adhesion to CSA was reduced by almost 50% following treatment with 916 compared to DMSO (p<.0001) and 025 had no effect (Figure 6H). Collectively, this experimentally validates PMV activity as essential for export of PEXEL-containing proteins, resulting in correct MC formation and PfEMP1 assembly at the erythrocyte surface and cytoadherence.
Protein export allows malaria parasites to remodel their cellular niche, and the protein machineries involved are obvious targets for the development of inhibitors. PMV acts by cleaving the PEXEL in the parasite ER and represents one such target. We developed a PEXEL-mimetic compound that potently inhibits the activity of PMV and, combined with protein knockdown or overexpression, used it to demonstrate the essentiality of PMV for parasite survival and its function for export.
The PMV inhibitor 916 mimics the transition-state of amide bond proteolysis for PEXEL substrates using statine. Our PfPMV structural model in complex with 916 outlines key interactions that are likely necessary for inhibitor binding: the guanidine side chain of Arg forms salt bridges with the acid of Glu179 and 215, and a π-stacking interaction with Tyr177 in the S3 pocket of PMV. This in part explains the necessity for Arg at P3 for PEXEL processing. The P1 Leu side chain is encased by hydrophobic residues in the S1 pocket formed in part by Ile116, Tyr177, and Val227, explaining the importance of Leu in PMV binding.
916 potently blocked PMV activity in vitro and reduced PEXEL cleavage, by up to 13-fold (Figure 3G), in cultured parasites, demonstrating the inhibitor directly engaged PMV in the ER. However, inhibition was time-dependent and incomplete at even 50 µM, indicating the inhibitor has suboptimal qualities. This may be due to a combination of poor diffusion across membranes, suboptimal final concentration in the ER, and the potent activity of PMV in parasites, revealed in this study by knockdown of PMV protein levels. Although further work is required to develop an inhibitor with enhanced properties, 916 has proven sufficiently active in parasites to examine PMV function and essentiality.
Previously, it has been shown that overexpression of a PfPMV D118A mutant produced a dominant-negative effect on parasite growth rate and protein export [12]. When we overexpressed an alternate PfPMV mutant (D118A, D365A, F370A) on episomes in P. falciparum, we observed a similar defect in parasite growth rate and subsequent down-regulation of endogenous PfPMV expression levels, suggesting a negative feedback effect. A similar negative feedback effect has been described for Toxoplasma gondii myosin A [34]. Collectively, these PMV dominant-negative mutants provide evidence that the enzyme is important for parasite survival.
The effects of 916 were amplified when PMV was knocked down and decreased when PMV was overexpressed, demonstrating that PMV is a target and that its inhibition is toxic to parasites. Addition of 916 to ring stages arrested their transition to trophozoites, between 20 and 30 h postinvasion, and parasites could not recover. Although this phenotype is consistent with death from impaired export and cellular remodeling, it is possible that ER stress due to accumulation of uncleaved PEXEL proteins in the organelle, and a decrease in translation, contributed [25]. The morphology of parasites treated with 916 was distinctly different to E-64–treated parasites, which contained swollen food vacuoles from inhibition of haemoglobin breakdown [24]. This indicates that parasite toxicity to 916 was unlikely due to off-target effects on food vacuole proteases. Although aspartyl protease inhibitors are known to kill P. falciparum, it is not entirely clear which aspartyl proteases (Plasmepsins) are essential (reviewed in [35]). In Plasmodium, there are 10 Plasmepsins; I–IV are important enzymes for haemoglobin degradation by P. falciparum, but the genes encoding each enzyme can be deleted from the genome, indicating they are not essential [36]. The death of P. falciparum following 916 treatment is therefore unlikely to be due to inhibition of these Plasmepsins. A survey of P. falciparum transcriptomes [37] suggests that, of the remaining five Plasmepsins (VI–X), only VII, IX, and X are expressed by asexual blood-stage parasites; however, VII and X are expressed at very low levels. Thus, Plasmepsin IX (PMIX) is considered the primary possible off-target in our study. However, we have shown previously that HA-tagged PMIX does not cleave the PEXEL motif [13], and although the enzyme itself possesses a PEXEL motif, its function and essentiality is currently unknown.
Analogs of 916 that mimic noncleavable PEXEL mutant sequences (R>K, L>I) were ineffective inhibitors of PMV in vitro and had no discernable effect on parasites at concentrations that 916 inhibited PEXEL cleavage in parasites and was lethal (i.e., <20 µM). However, at concentrations above 20 µM they were toxic to parasites and possessed EC50 values between 6- and 26-fold less potent than 916. At 50 µM, we saw no evidence of PMV or signal peptidase inhibition, export or secretion defects, or global effects on translation caused by 024 or 025. This suggests that they hit a target(s) distinct from 916.
The rapid rate of protein synthesis, ER import, signal peptide processing, and PEXEL cleavage in P. falciparum was determined for the first time. The rate was within immeasurable seconds after translation, consistent with both signal peptidase and PMV activity occurring cotranslationaly. The full-length PEXEL is thus only present in the proprotein very transiently, making its function in export even more remarkable. This underscores the importance of the remaining PEXEL residues (xE/Q/D) in export following processing [11]. The PEXEL has been suggested to function independent of PMV by binding PI3P in the ER, and ER-derived transport vesicles, via the PEXEL Arg [38]. The rate at which PEXEL processing occurred in our experiments is inconsistent with this hypothesis, as the PEXEL Arg is cleaved off during, or soon after, ER entry. It is also challenging to envisage how the PEXEL Arg could dock within the S3 pocket of PMV, where it is required for proteolytic cleavage, if it is bound to the ER membrane via an interaction with PI3P.
This work has characterized a novel PEXEL protein, Hyp8, which is exported in the early ring-stage to MCs. The function of Hyp8 is unknown, but the hyp8 gene was refractory to deletion and may be essential. 916 treatment dramatically impaired Hyp8 export, resulting in some accumulation in the parasite and possibly degradation. It is unknown whether the reduced export of Hyp8 directly contributed to parasite death; however, the phenotype demonstrates the importance of PMV in export of PEXEL-containing cargo. This is further supported by the decrease in MC numbers observed following 916 treatment; MC formation is known to require exported proteins (reviewed in [20]). The secretion of EXP2 to the parasitophorous vacuole membrane was unaltered by 916 treatment, demonstrating that the effects of 916 were specific to export. A clear defect in PfEMP1 surface exposure and cytoadherence was also observed following 916 treatment. PfEMP1 is unlikely to be a PMV substrate [18], but its trafficking through the erythrocyte and onto the surface requires at least six PEXEL-containing proteins [17]; thus, the demonstration that PMV activity is essential for PfEMP1 surface expression and cytoadherence is consistent with the current literature and validates the specificity of the inhibitor. Further, it directly demonstrates the importance of PMV at the first step in the export pathway for cellular remodeling that leads to virulence.
P. vivax is an important global pathogen that cannot be cultured in the laboratory, and novel therapeutic targets for this enigmatic parasite are urgently needed. This work has characterized PMV from P. vivax for the first time. PvPMVHA possesses the trafficking information to localize to the ER and has similar PEXEL cleavage activity and specificity to PfPMV. This indicates that PMV function is to cleave the PEXEL motif of exported proteins across Plasmodium spp., and future compounds that block PMV are likely to affect multiple Plasmodium spp. Protein export also occurs in gametocytes [28] and liver stages [39], and 916 may aid the characterization of PMV in these stages.
A putative PMV homolog, ASP5, is present in Toxoplasma and localizes to the Golgi [40]. Recent evidence suggests that some exported T. gondii proteins contain a PEXEL [41], and some are cleaved in a manner that requires the conserved PEXEL residues [42]. The PEXEL protease may therefore be conserved beyond the Plasmodium genus, and PMV and its homologs may therefore represent multistage, multispecies antiparasitic targets of the future.
P. falciparum 3D7 parasites expressing PfPMVHA, PfPMVmutHA, and PfEMP3-GFP were generated previously [13], as was KAHRP-GFP [18] and CS2-GFP [29]. DNA encoding PvPMV or PvPMVmut fused to 3× HA tags was synthesized (Epoch Biosciences) and cloned into pGlux.1 [11] with XhoI and PacI, removing GFP. DNA encoding miniVarHA [PfEMP1 NTS (residues 1–51) fused to SVL-TM-ATS (residues 2640–2734) of IT4 VAR2CSA] was synthesized (Epoch Biosciences) and cloned into pGlux.1 with XhoI and PacI. DNA encoding the signal peptide of SERA 5 (PFB0340c) (residues 1–25) or the entire hyp8 gene (MAL13P1.61) was amplified from P. falciparum gDNA and cloned in frame with GFP in pGlux.1 using XhoI and XmaI. For HA tagging Hyp8, the 3′ 800 bp of hyp8 was cloned into p1.2-SHA [13] (also called pHA3; [43]) using BglII and PstI. For tagging PfPMV with HA-glmS in P. falciparum NF54, the 3′ 1144 bp of PMV was cloned into pPTEX150-HA-glmS, which consisted of the glmS riboswitch from pGFP_glmS [26] cloned into pHA3 using BglII and PstI to replace the PTEX150 gene with PMV, generating pPMVHA-glmS. For tagging PfPMV with HA-M9, the M9 insert from pGFP_M9 [26] was cloned into pPMVHA-glmS to generate pPMVHA-M9. To express PfEMP3-GFP in P. falciparum NF54 harboring PMVHA-glmS, the dihydrofolate reductase selection cassette in pPfEMP3Glux.1 [13] was replaced with blasticidin deaminase using BamHI and HindIII prior to transfection. P. falciparum transfectants were selected with 5 nM WR99210 (Jacobus Pharmaceuticals) and/or 2 µg/ml Blasticidin S (Calbiochem) and grown in O+ human erythrocytes as described [18]. CS2-GFP parasites preferentially expressing the var2csa gene (PFL0030c/PF3D7_1200600) were selected every 2 wk by enriching for knob-positivity with gelatin [44] and panning for CSA-binding [45]. Rα-PfMV antibodies were generated by immunization of rabbits with recombinant PfPMV generated previously [13] and collecting serum during four boost immunizations. Affinity-purified polyclonal rabbit α-Hyp8 antibodies were generated by Genscript using the peptide N-55ETEQSTPAKPEPTE68-C.
PMV-agarose was prepared by adding α-HA-agarose (Sapphire Bioscience) to parasite lysates, prepared by sonication in 1% Triton X-100/PBS, for 1 h before extensive washing in same [12],[13]. PEXEL cleavage assays (20 µl total volume) consisted of 0.2 µl PMV-agarose in digest buffer (25 mM Tris, 25 mM MES, pH 6.4) with 1.5 µM FRET peptide substrate (DABCYL-RNKRTLAQKQ-E-EDANS, DABCYL-RNKATAAQKQ-E-EDANS, LifeTein; DABCYL-RNKKTLAQKQ-E-EDANS, DABCYL-RNKRTIAQKQ-E-EDANS; Mimotopes) ± inhibitor. Samples were excited at 340 nm and fluorescence emission measured at 492 nm using an Envision fluorescence plate reader (Perkin-Elmer) heated to 37°C for 150 min. Samples were shaken between measurements. For determination of the peptide cleavage position by PvPMVHA, the fluorogenic peptide (DABCYL-RNKRTLAQKQ-E-EDANS), representing the wild-type KAHRP PEXEL sequence, was incubated with and without PvPMVHA at 37°C for 48 h. Products of the incubation were detected by a molecular formula algorithm using an Agilent 6200 TOF/6500 series mass spectrometer.
Parasite growth assays were performed in 96-well plates by incubating highly synchronous ring-stage P. falciparum 3D7 or NF54 parasites with compounds solubilized in DMSO at the indicated concentrations for the indicated times. In the case of dose-response curves, medium was kept for the entire experiment; in the case of curves in Figure 4B,C, medium was replaced with inhibitor-free medium at 48 h postinfection. Parasitaemia was always determined at 72 h by flow cytometry. To knock down PMV, GlcN (Sigma) was added to trophozoites and drug curves initiated by adding compound at subsequent rings (24 h) for 24–72 h.
Trophozoites (30–34 h) expressing PfEMP3-GFP, KAHRP-GFP, or SERA5s-GFP were magnet-purified (Miltenyi Biotech), incubated with inhibitors in 400 µl total volume at 37°C for 1–5 h, treated with 0.09% saponin containing inhibitor, and washed pellets were solubilized in Laemmli's buffer, boiled for 3 min, and frozen at −20°C. Proteins were separated by SDS-PAGE, transferred to nitrocellulose and blocked in 10% skim milk/PBS-T, and probed with rat α-HA (Roche 3F10; 1∶1,000), mouse α-GFP (Roche; 1∶1,000), rabbit α-Aldolase (1∶1,000), rabbit α-HSP70 (1∶4,000), rabbit α-PfPMV (1∶1,000), or rabbit α-Hyp8 (1∶500) primary antibodies followed by horseradish peroxidase-conjugated secondary antibodies (Silenius) and detected by enhanced chemiluminescence (Amersham).
Whole parasite proteins were radiolabeled by culturing magnet-purified trophozoites (wild-type 3D7 or expressing PfEMP3-GFP) in Met/Cys-free medium for 30 min at 37°C before addition of 800 µCi/ml 35S-Met/Cys (Perkin/Elmer) to the medium for the indicated times. Pellets were snap frozen in ethanol/dry ice bath and stored at −80°C. For radiolabeling in the presence of PMV inhibitor, parasites were treated with 20 µM WEHI-916 for 5 h before labeling commenced. For pulse-chases, proteins were radiolabeled in the presence or absence of inhibitor, as above, before further culture in radiolabel-free, inhibitor-free complete medium for the indicated times at 37°C before snap freezing. Frozen samples were either solubilized in Laemmli's buffer (Figure S4) or solubilized in 1% Triton X-100/PBS with protease inhibitor cocktail (Roche) and PfEMP3-GFP species immunopurified with α-GFP agarose (MBL) at 4°C for 2 h (Figures 3 and S3), and proteins were resolved by SDS-PAGE, visualized by autoradiography (7-d exposures), and quantified using a GS-800 Calibrated Densitometer (Bio-Rad).
For immunofluorescence microscopy, smears were fixed in cold acetone∶methanol (90∶10) and probed with rabbit Rα-PfPMV (1∶750), mouse Mα-PfPMV (1∶25), rabbit α-EXP2 (1∶200), rat α-HA (Roche 3F10; 1∶50), mouse α-GFP (Roche; 1∶500), or rabbit α-Hyp8 (1∶200) antibodies followed by Alexa Fluor 488- or 594-conjugated secondary antibodies (Molecular Probes; 1∶1,000). DNA was stained with 4′-6-Diamidino-2-phenylindole (DAPI) at 0.2 µg/ml. Samples were viewed on a Deltavision Elite microscope and images collected with a Coolsnap HQ2 CCD camera through an Olympus 100× UPlanSApo NA1.4 objective with SoftWorx software. Images were assembled with ImageJ Fiji 1.47d and Adobe Photoshop CS6 v13.0 x64. Light and immunoelectron microscopy was performed as described in [46].
For quantification of events in cells infected with parasites expressing Hyp8-GFP, highly synchronous ring-stage parasites engineered to express Hyp8-GFP from the CRT promoter were obtained by incubation of erythrocytes with viable merozoites for 15 min [47] and treated with 20 or 50 µM 916 30 min postinvasion for 13 h (until Hyp8-GFP expression from the CRT promoter had occurred for 1 h). Smears were fixed in 90∶10 acetone∶methanol, labeled with anti-GFP and anti-EXP2 antibodies, and Z-stacks captured on a Deltavision Elite microscope using a 100× objective. Over 40 Z-stacks per condition were imaged using the same exposure settings to allow quantitative analysis between groups.
As 916 treatment has adverse affects on parasites after 24 h, a sublethal dosing regime was developed (see Figure S7) to maximize the inhibitor effect while ensuring parasites remained viable 24 h postinvasion when surface-exposed PfEMP1 was measured. Treatment with >15 µM for 23 h or with 50 µM for >12 h postinvasion prior to decreasing to 15 µM adversely affected parasite growth and was avoided.
To measure PfEMP1 display, highly synchronous ring-stage CS2-GFP parasites preferentially expressing VAR2CSA (see Plasmids, Parasites, and Antibody Production) were obtained by incubation of erythrocytes with viable merozoites for 15 min [47], and parasites were treated with 916, 025, or DMSO using the dosage regime above. At 24 h postinvasion (presence of inhibitor for no more than 23 h), erythrocytes were incubated with human monoclonal PAM1.4 serum [30] (1∶200) to label VAR2CSA followed by goat anti-human IgG Biotin-conjugated secondary antibodies (Invitrogen) (1∶200) and Alexa Fluor-633 Streptavidin-conjugated tertiary antibodies (Invitrogen) (1∶500) for 30 min each. Labeled cells were washed with 0.1% casein/PBS and analyzed on a FACSCalibur cytometer (Becton-Dickinson, USA). Fluorescence in channel FL1 was used to measure parasite-infected erythrocytes (GFP), and fluorescence in channel FL4 was used to measure bound PAM1.4 IgG antibodies (Alexa 633) for each sample. The geometric mean fluorescence of uninfected erythrocytes (treated with secondary and tertiary but not primary antibodies) was deducted from the geometric mean fluorescence of infected erythrocytes using >100,000 cells per condition. Experiments were conducted in duplicate. Analyses were performed using FlowJo 8.8.7 (Tree Star, USA).
Adhesion assays were performed as described previously [32],[33]. Briefly, CSA (Sigma) was spotted at 50 µg/ml in triplicate into petri dishes, incubated overnight at 4°C, and blocked in 1% casein/PBS for 2 h. Inhibitor-treated erythrocytes were added to CSA-coated dishes and incubated for 45 min at 37°C. Dishes were washed four times with 5 ml warm RPMI-HEPES, fixed in 2% paraformaldehyde for 2 h, stained with 10% Giemsa for 15 min, and the number of adherent erythrocytes per mm2 quantified by light microscopy counts. Assays were performed in triplicate.
This information is presented in Materials and Methods S1.
|
10.1371/journal.ppat.1003265 | Dengue Virus Co-opts UBR4 to Degrade STAT2 and Antagonize Type I Interferon Signaling | An estimated 50 million dengue virus (DENV) infections occur annually and more than forty percent of the human population is currently at risk of developing dengue fever (DF) or dengue hemorrhagic fever (DHF). Despite the prevalence and potential severity of DF and DHF, there are no approved vaccines or antiviral therapeutics available. An improved understanding of DENV immune evasion is pivotal for the rational development of anti-DENV therapeutics. Antagonism of type I interferon (IFN-I) signaling is a crucial mechanism of DENV immune evasion. DENV NS5 protein inhibits IFN-I signaling by mediating proteasome-dependent STAT2 degradation. Only proteolytically-processed NS5 can efficiently mediate STAT2 degradation, though both unprocessed and processed NS5 bind STAT2. Here we identify UBR4, a 600-kDa member of the N-recognin family, as an interacting partner of DENV NS5 that preferentially binds to processed NS5. Our results also demonstrate that DENV NS5 bridges STAT2 and UBR4. Furthermore, we show that UBR4 promotes DENV-mediated STAT2 degradation, and most importantly, that UBR4 is necessary for efficient viral replication in IFN-I competent cells. Our data underscore the importance of NS5-mediated STAT2 degradation in DENV replication and identify UBR4 as a host protein that is specifically exploited by DENV to inhibit IFN-I signaling via STAT2 degradation.
| Dengue virus (DENV) is the leading cause of mosquito-borne viral illness and death in humans. At present, there are no vaccines and no specific antiviral therapeutics to prevent or treat DENV infections. We previously described that the NS5 protein of DENV inhibits type I interferon signaling in virus-infected cells by mediating STAT2 degradation. This property allows DENV to overcome the antiviral effects of type I interferon, contributing to viral replication in the host. We have now obtained new insight into the mechanism by which DENV NS5 induces STAT2 degradation. NS5 bridges STAT2 with the cellular protein UBR4, a member of a family of predicted E3 ligases, resulting in UBR4-mediated STAT2 degradation. Elimination of UBR4 or mutations in NS5 that prevent its binding to UBR4 prevent NS5 from inducing STAT2 degradation. Importantly, UBR4 is required for optimal DENV replication in the presence of a competent type I interferon system. Our data demonstrate the requirement of a host factor, UBR4, for DENV to overcome the antiviral interferon response. This information might be important for the design of specific DENV inhibitors that prevent dengue virus from evading innate immunity.
| Approximately fifty million dengue virus (DENV) infections occur annually with more than forty percent of the human population at risk of developing dengue fever (DF) or dengue hemorrhagic fever (DHF). DF and its more severe form, DHF, are potentially fatal diseases caused by the four serotypes (1, 2, 3 and 4) of DENV [1]. As no approved vaccines or antiviral therapeutics are available for the prevention or treatment of DENV infections, it is imperative that the biology and immunology of DENV infections are better understood. An in depth comprehension of DENV-host interactions will accelerate our progress in developing DENV therapeutics.
DENV, along with other clinically relevant arboviruses such as West Nile virus (WNV), Japanese encephalitis virus (JEV) and yellow fever virus (YFV), belongs to the flavivirus genus of the Flaviviridae family. The flavivirus genome is a capped 11 kb genome that is translated into a single polyprotein, which is cleaved both by the viral protease (NS2B/3) and host proteases to yield three structural proteins (capsid [C], membrane [prM/M] and envelop [E]) and seven non-structural proteins (NS1, NS2A, NS2B, NS3, NS4A, NS4B and NS5) [2], [3]. The flavivirus structural proteins incorporate the viral genome into newly generated virions while the non-structural proteins replicate the viral genome and exploit the cellular machinery to subvert host immune responses. The approximately 900-amino-acid NS5 protein is the largest and most conserved flavivirus protein [4]. This multifunctional protein has RNA-dependent RNA polymerase (RdRp) activity as well as methyltransferase activity [5], [6], [7], [8], [9]. In addition, more recent studies have shown that NS5 is a potent interferon-signaling antagonist [10], [11], [12], [13], [14], [15], [16].
The significance of the interferon (IFN) response as an important component of host immunity is underscored by numerous examples of viruses that antagonize it [17], [18], [19], [20], [21], [22], [23], [24], [25]. Viruses express pathogen-associated molecular patterns (PAMPs) that trigger the production of type I IFN (IFNα/β or IFN-I) [26]. Binding of IFN-I to the cell-surface IFN-I receptor (IFNAR) initiates a signaling cascade that results in the activation and phosphorylation of the Janus kinases, Jak1 and Tyk2, and the transcription factors, STAT1 and STAT2. Phosphorylated STAT1 and STAT2 along with IRF9 form the heterotrimeric transcriptional complex, ISGF3 [27], [28], and induce the expression of antiviral IFN-stimulated genes (ISGs) [29], [30], [31], [32].
DENV encodes several antagonists of both IFN-I production and IFN-I signaling [13], [14], [33], [34], [35], [36], [37], [38]. The NS5 proteins of DENV and other flaviviruses have been shown to be potent inhibitors of IFN signaling. NS5 proteins of different flaviviruses may target different steps of the IFN signaling pathway. For example, WNV NS5 prevents the phosphorylation of the STAT proteins, while DENV NS5 binds human STAT2 and promotes its proteasomal degradation [13], [15].
Although STAT degradation is a common mechanism of viral IFN antagonism [19], [20], [39], [40], the requirements for DENV NS5-mediated STAT2 degradation are unique. DENV NS5-mediated STAT2 degradation requires NS5 to be proteolytically cleaved at its N terminus from a larger precursor protein [13]. N-terminal cleavage of NS5 normally occurs during DENV infection because the NS2B/3 protease cleaves at the junction located between NS4B and NS5 thereby releasing NS5 from the viral polyprotein [41]. Though both unprocessed and proteolytically-processed NS5 can bind STAT2, only processed NS5 can efficiently mediate STAT2 degradation [13]. Furthermore, the first ten amino acids of NS5 are dispensable for STAT2 binding but are indispensable for STAT2 degradation [13]. While the viral requirements for DENV-mediated STAT2 degradation are known, the cellular components were unspecified until now.
This study identifies the 600-kDa protein, UBR4, as a binding partner of DENV NS5. UBR4 is a member of the N-recognin family, which contains proven and predicted E3 ligases that recognize and degrade proteins containing destabilizing N termini [42]. UBR4 interacts preferentially with proteolytically-processed DENV NS5 but not with YFV NS5 or WNV NS5, highlighting the specificity of the DENV NS5/UBR4 interaction. Furthermore, we have identified two residues within the first 10 N-terminal amino acids of NS5, threonine 2 and glycine 3, that are required for NS5 binding to UBR4. These two residues are conserved across the four DENV serotypes but are not found in other flaviviruses. Finally, we show that UBR4 is required for DENV-mediated STAT2 degradation, and for efficient DENV replication in IFN-I competent cells. Our data confirm the importance of NS5-mediated STAT2 degradation for DENV replication, and identify UBR4 as a host protein that is specifically co-opted by DENV to inhibit IFN-I signaling via STAT2 degradation.
DENV NS5 binds human and non-human primate STAT2 but cannot efficiently mediate STAT2 degradation unless it is expressed in the context of a precursor protein from which it is N-terminally cleaved [13]. When DENV NS5 is engineered to be expressed downstream of ubiquitin, cellular hydrolases cleave ubiquitin in a manner that mimics the cleavage of NS4B away from NS5 by the NS2B/3 protease during DENV infection [13]. To identify host proteins that are required for NS5-mediated degradation of STAT2, we generated a DENV2 NS5 construct consisting of RFP-ubiquitin fused to the NS5 N-terminus and a TAP (tandem-affinity purification) tag fused to the NS5 C-terminus. This NS5 construct was expressed in 293T cells, in the presence or absence of human STAT2-FLAG, and then purified using the TAP method. A high molecular weight protein band was consistently and specifically co-precipitated with NS5 both in the presence and absence of overexpressed STAT2 (Figure 1A). Trypsin digestion of this band yielded five peptides that were identified by mass spectrometry as sequences of the N-recognin, UBR4 (Table 1) [42]. To confirm the binding and specificity of the interaction between DENV2 NS5 and UBR4, HA-tagged DENV2 NS5, DENV1 NS5, and YFV NS5 were expressed in 293T cells and purified by immunoprecipitation with antibodies raised against the HA epitope. The NS5 protein of both DENV1 and DENV2 precipitated UBR4 from 293T cells but YFV NS5 was unable to precipitate UBR4 from these cells (Figure 1B). WNV NS5 was also unable to bind UBR4 (Figure 1C). In order to assess the specificity of DENV NS5 for UBR4, we also examined the ability of NS5 to bind another member of the N-recognin family, UBR5. DENV1 and DENV2 NS5 bound UBR4 but not UBR5, further highlighting the unique interaction between UBR4 and DENV NS5 (Figure 1B).
UBR4 was found to bind both processed (RFP-ubiquitin-NS5-TAP) (Figure 1A), as well as unprocessed NS5 (NS5-HA) (Figure 1B). Since cleavage of NS5 promotes STAT2 degradation, we tested whether proteolytic processing would also affect the binding efficiency of NS5 for UBR4. The E domain in Figure 2A refers to the E protein of DENV2. Inclusion of a protein (such as E or RFP) upstream of ubiquitin allows one to differentiate between cleaved and uncleaved NS5 [13]. HA-tagged unprocessed NS5 (NS5-HA) and processed NS5 (proNS5-HA) (Figure 2A) were expressed in 293T cells and the NS5 proteins were immunoprecipitated and tested for UBR4 binding. Although both constructs precipitated UBR4, proNS5-HA precipitated two-fold higher amounts of UBR4 than NS5-HA did (as quantified by densitometry) (Figure 2B).
Given that the first 10 amino acids of NS5 are dispensable for STAT2 binding but indispensable for STAT2 degradation [13], we asked whether this N terminal region of NS5 was also important for UBR4 binding. To test this, we expressed and immunoprecipitated processed HA-tagged DENV NS5 proteins containing a deletion of 10 or 306 residues at their N-termini, and assessed their ability to bind UBR4. Full length HA-tagged NS5 (NS5-HA) was able to precipitate UBR4 and STAT2, and its ability to precipitate UBR4, but not STAT2, was increased seven-fold (as quantified by densitometry) when DENV NS5 was proteolytically-processed (proNS5-HA) (Figure 2C). Proteolytically-processed NS5 lacking the first 10 amino acids (proΔ10NS5-HA) precipitated STAT2 but not UBR4, and proteolytically-processed NS5 lacking the first 306 amino acids (proΔ306NS5-HA) precipitated neither protein (Figure 2C). When protein levels were examined in the whole cell extracts (WCE), STAT2 was reduced in proNS5-HA-expressing cells and slightly reduced in NS5-HA-expressing cells compared with proΔ10NS5-HA- or proΔ306NS5-HA-expressing cells (Figure 2C), which is consistent with published reports [13]. Thus, only the NS5 proteins that bound UBR4 could mediate STAT2 degradation, and increased UBR4 binding by NS5 correlated with increased NS5-mediated STAT2 degradation. The interaction of NS5 with UBR4 and the requirement for the first 10 amino acids of NS5 in mediating this DENV NS5-UBR4 interaction was also observed by NS5-UBR4 colocalization using immunofluorescence analysis in Vero cells (Figure 2D). To further define which of the N-terminal residues of DENV NS5 are required for its interaction with UBR4, alanine scanning of the first 5 amino acids of DENV NS5 was conducted (Figure 2E). Immunoprecipitation experiments with these mutant proteins revealed that the threonine (T) at position 2 and the glycine (G) at position 3, which are conserved among the DENV serotypes but absent in other flaviviruses, were required for NS5/UBR4 interaction and NS5-mediated STAT2 degradation (Figure 2F).
The fact that NS5 mutants lacking residues T2 or G3 bound STAT2 but not UBR4 (Figure 2F) suggested that the interaction between NS5 and UBR4 was independent of STAT2. To confirm this result, the STAT2-deficient U6A cell line [43] was transfected with proNS5-HA or proΔ10NS5-HA. ProNS5-HA, but not proΔ10NS5-HA, precipitated UBR4 from U6A cells (Figure 3A). Unlike with human STAT2 (hSTAT2), NS5 does not bind and subsequently degrade mouse STAT2 (mSTAT2) [44]. The STAT2 proteins of mouse and human are divergent and share only 70% identity but the UBR4 proteins of mouse and human are 97% identical. When proNS5-HA was expressed in mouse cells (Hepa1.6), mouse UBR4 bound proNS5-HA but not proΔ10NS5-HA confirming that an interaction between NS5 and STAT2 is not required for NS5 to interact with UBR4 (Figure 3A). These data suggest that NS5 requires binding to both UBR4 and STAT2 to mediate STAT2 degradation.
NS5 binds the coiled-coil region located within the first half of hSTAT2 [44]. Though mSTAT2 and human STAT1 (hSTAT1) cannot bind NS5, chimeric proteins that replace the first 301 amino acids of mSTAT2 (h/mSTAT2) or the first 316 amino acids of hSTAT1 (hSTAT2/1) with those of hSTAT2 can bind NS5 [44]. We expressed and immunoprecipitated FLAG-tagged STAT proteins and STAT chimeric proteins in the presence or absence of proNS5-HA. When hSTAT2 was overexpressed, STAT2 degradation was not observed because STAT2 degradation was likely masked by the large amount of overexpressed STAT2 present (Figure 3B). However, we observed that while hSTAT1 and mSTAT2 did not bind UBR4, h/mSTAT2, hSTAT2/1 and hSTAT2 all bound UBR4 in the presence of proNS5-HA (Figure 3B). Since only those STAT molecules that could bind NS5 could also bind UBR4, and NS5 binds UBR4 in the absence of STAT2, we conclude that NS5 serves as a bridge molecule between STAT2 and UBR4.
These results were confirmed in the context of DENV infection of a transformed human cell line (293T) and primary untransformed human cells (monocyte-derived dendritic cells or MDDCs) (Figure 3C and 3D). Cells were infected with DENV2 at an MOI of 3 for 24 hours, and lysed for immunoprecipitation with STAT2 antibodies or control IgG. Although the majority of STAT2 was degraded during DENV infection, the remaining STAT2 co-immunoprecipitated UBR4 from DENV-infected cells but not from mock-infected cells (Figure 3C and 3D), which is consistent with NS5 binding to and bringing together STAT2 and UBR4 during DENV infection.
We next assessed the functional relevance of UBR4 in DENV-mediated STAT2 degradation. To test if UBR4 is required for DENV-mediated STAT2 degradation, UBR4 levels were stably reduced in 293T cells using small hairpin RNA (shRNA) directed against UBR4. Three stable UBR4-knockdown cell lines were generated using shRNA that targeted different sequences within UBR4. The cells were mock infected or infected with DENV2 at an MOI of 10, and lysed for western blot analysis at 4, 8, 12 and 24 hours post-infection. When cells expressing control non-targeting shRNA were infected with DENV2, STAT2 levels decreased by 4 hours post-infection. However, in the three independently-derived, UBR4-deficient 293T cell lines, STAT2 levels decreased at a slower rate (Figure 4A). Furthermore, NS5 levels were lower in the UBR4-knockdown cells than in the control cells suggesting that there was a DENV replication defect in UBR4-knockdown cells. The similar phenotype of the three UBR4-knockdown cell lines and their difference from the control cell line indicated that the effect of UBR4 knockdown on DENV-mediated STAT2 degradation was due to the decreased level of UBR4 and not to off-target effects of the shRNA. We next examined the functional relevance of UBR4 in mediating STAT2 degradation with the other three DENV serotypes (DENV1, 3 or 4). STAT2 levels were higher and NS5 levels were lower in shUBR4-expressing cells than in control cells (Figure 4B). Thus, UBR4 is required for efficient STAT2 degradation mediated by all four DENV serotypes.
The UBR4 gene is predicted to produce several splice variants encoding proteins of greater than 5000 amino acids. Since it is unclear which UBR4 isoform is required for DENV-mediated STAT2 degradation, we cloned a region of UBR4 (UBR4-NT) that is predicted to be present in all the large UBR4 isoforms and which also contains the UBR box, a 70-amino-acid zinc-finger-like domain required for recognition of N-end rule substrates [42]. The UBR box is located between amino acids 1662–1723 of the UBR4 reference sequence (NCBI Accession # Q5T4S7), and the UBR4-NT clone encodes amino acids 1–2233 of the reference sequence. Co-immunoprecipitation experiments revealed that proNS5-HA did not bind amino acids 1–2233 of UBR4 (Figure 4C) indicating that sequences in the C-terminal half of UBR4 are required to mediate its interaction with NS5. Also, expression of UBR4-NT had no effect on DENV-mediated STAT2 degradation (Figure 4D). The experiments in Figure 4 confirm that a functional UBR4-NS5-STAT2 complex is required for efficient STAT2 degradation and that multiple domains of UBR4 are required for this function.
The ability of DENV to degrade STAT2 determines how well it replicates in an IFN-I-competent cell [44]. Thus, a protein that is required for DENV-mediated STAT2 degradation should also enhance DENV replication in IFN-I-competent cells. To test if UBR4 is required for DENV replication, UBR4-knockdown 293T cells were infected with DENV2 at multiplicities of infection (MOI) of 0.1, 1 and 10, and measured for virus at 24 hours post-infection. DENV replicated to lower levels in UBR4-knockdown cells than in control cells (Figure 5A). The replication defect was most striking at a lower MOI and an approximately 10-fold decrease in virus levels was observed in shUBR4 cells with an MOI of 0.1 of DENV. In contrast, UBR4 depletion had no effect on the replication of YFV or encephalomyocarditis virus (EMCV), a positive-strand RNA virus belonging to the Picornaviridae family, indicating a specific requirement of UBR4 in DENV replication (Figure 5A). DENV1, 3 and 4 also replicated to lower levels in UBR4-knockdown cells than in control cells indicating that UBR4 is required for the efficient replication of all four DENV serotypes (Figure 5B).
Since UBR4 was required for DENV-mediated STAT2 degradation, we hypothesized that the DENV replication defect in UBR4-deficient cells was due to an inability of DENV to antagonize IFN-I signaling by degrading STAT2. If this is the case, lack of IFN-I should compensate for the requirement of UBR4 in DENV replication. To test this, we infected control and UBR4-knockdown Vero cells with DENV. Vero cells lack IFN-I genes and therefore cannot make IFN-I in response to viral infection [45]. DENV replicated to similar levels in UBR4-knockdown and control Vero cells (Figure 5C). Yet when Vero cells were infected with DENV and then exogenously treated with IFN-I 6 hours later, a DENV replication defect was observed in the UBR4-deficient Vero cells (Figure 5C). Protein levels of NS5, UBR4, and STAT2 in UBR4-knockdown Vero cells showed that UBR4 levels were indeed lower and that DENV-mediated STAT2 degradation was defective in UBR4-knockdown cells (Figure 5B).
Treating DENV-infected UBR4-knockdown 293T cells with a neutralizing anti-IFNAR antibody corroborated the effect of IFN-I on DENV replication in UBR4-deficient Vero cells (Figure 5D). We observed a significant increase in DENV replication in UBR4-knockdown 293T cells treated with the neutralizing anti-IFNAR antibody compared to UBR4-knockdown 293T cells treated with IgG control antibodies. This contrasts with what was observed in control 293T cells where DENV replication was unaffected by treatment with anti-IFNAR antibodies (Figure 5D).
IFN exerts its biological effect by upregulating interferon-stimulated genes (ISGs), which encode products that restrict viral replication. To examine the biological relevance of UBR4 in preventing the antiviral action of IFN-I during DENV infection, we examined the induction of ISG54 mRNA in UBR4-knockdown 293T cells. There was a significant induction of ISG54 mRNA in UBR4-knockdown cells during DENV infection compared to control cells (Figure 5E). These results indicate that UBR4 is required for preventing the antiviral action of IFN-I during DENV infection.
Dendritic cells are thought to be an important cell type in which DENV replicates in vivo [46], [47]. We reduced UBR4 levels in primary MDDCs from five donors using shRNA lentiviral constructs, and tested the effect of this decrease on DENV replication. When MDDCs from each donor were infected at an MOI of 3, approximately 35% of transfected cells were highly infected and showed viral glycoprotein (E) expression by FACS. At 12 hours post infection, as expected, the levels of UBR4 were decreased in the UBR4-knockdown cells compared to control cells (Figure 6A). When the levels of ISG15, RIG-I and ISG54 mRNA were analyzed in these cells, more ISGs were induced in four of the five donors (Figure 6B, 6C and 6D respectively). In addition, more DENV was present in control cells than in UBR4-knockdown cells at 48 hours post infection (Figure 6E). Thus, UBR4 is required for inhibiting ISG induction and increasing DENV replication in a primary cell type that is of importance in DENV infections.
The IFN-I response is one of the first lines of protection against DENV infection, and serves to curb viral replication and dissemination by generating an antiviral intracellular environment [48]. The potency of the type I IFN pathway is exemplified by the fact that DENV antagonizes both IFN synthesis and IFN signaling in order to ensure its replication and survival [13], [14], [33], [34], [35], [36], [37], [38]. DENV NS5 inhibits IFN-I signaling by mediating proteasome-dependent STAT2 degradation, and STAT2 degradation promotes DENV replication [13], [14], [44]. With this study, we report the discovery of a host factor, UBR4, that is essential for DENV-dependent STAT2 degradation. We describe the interaction of UBR4 with NS5 and show that this interaction is crucial for inhibiting type-I IFN signaling and promoting efficient DENV replication.
UBR4 associates with DENV NS5 but not with the closely related YFV NS5 or WNV NS5. UBR4 also binds preferentially to proteolytically-processed DENV NS5, which is the form of NS5 that efficiently mediates STAT2 degradation. Binding of UBR4 to DENV NS5 requires amino acids T2 and G3 of NS5, which are also critical for STAT2 degradation. These amino acids are conserved amongst the four DENV serotypes but are absent in other flaviviruses (Figure 2F). Though NS5 is the most highly conserved flavivirus protein, the high degree of specificity exhibited by UBR4 for DENV NS5 underscores the differences between the various flaviviral NS5 proteins.
In 293T cells and primary human dendritic cells, DENV replicates best when UBR4 levels are normal, but when UBR4 levels are reduced, DENV-mediated STAT2 degradation is reduced and DENV replication decreases as a consequence (Figures 5 and 6). In Vero cells, which do not produce IFN-I [45], UBR4 depletion does not affect DENV replication unless these cells are treated with exogenous IFN-I (Figure 5C). Furthermore, the DENV replication defect caused by UBR4 knockdown in 293T cells can be decreased by treating the cells with antibodies that block the IFN-I receptor and decrease IFN-I signaling (Figure 5D). The DENV replication defect seen in UBR4-knockdown 293Ts and MDDCs can be explained by an increase in ISG levels in DENV-infected UBR4-knockdown cells versus DENV-infected control cells (Figure 5E and Figure 6). Thus, in the absence of IFN-I, there is no need for DENV to antagonize IFN-I signaling and cellular levels of UBR4 are irrelevant for DENV replication. However, upon activation of the IFN-I signaling pathway, UBR4 becomes necessary for DENV replication. Reducing STAT2 levels is essential for DENV to preempt the establishment of a cellular antiviral state, thus ensuring its efficient replication.
Antagonism of IFN signaling is one of the factors responsible for the limited host tropism of DENV to human and nonhuman primates. DENV does not replicate to high levels or induce disease in IFN-competent mice [44], [49]. Our previous results indicated that the cellular machinery needed for DENV replication in murine cells is in place but is limited by the inability of NS5 to associate with murine STAT2 and inhibit murine IFN-I signaling [44]. Other blocks such as the type II IFN pathway also diminish DENV replication in mice, but the IFN-I signaling pathway restricts early replication [44], [50]. Here we show that DENV NS5 associates with murine UBR4 in murine cells. This is in keeping with our previous results [44], and suggests that the development of a genetically-modified mouse that expresses a functional human STAT2 in place of its murine counterpart should allow increased DENV replication. We predict, therefore, that DENV NS5 will mediate human STAT2 degradation in these mice by co-opting mouse UBR4. Such a mouse might provide the basis for the development of an immune-competent mouse model of DENV infections.
The 600 kDa large UBR4 is highly conserved and found in organisms as diverse as mammals, insects, plants and worms. It belongs to the N-recognin family, which contains proven and predicted E3 ligases that recognize and degrade proteins containing destabilizing N termini. The seven members of the UBR family, UBR1 to UBR7, encode a 70-amino-acid zinc finger motif known as the UBR box, which is necessary for substrate recognition [42]. The better-characterized members of the UBR family are UBR1, UBR2 and UBR5. UBR1 and UBR2 are RING domain-containing N-recognins, which recognize N-end rule substrates and target them for degradation [42]. UBR1 and UBR2 are also involved in N-end-rule-independent quality control protein degradation [51]. UBR5 is a HECT-domain containing E3 ligase that binds N-end rule substrates [42], but can also target non-N-end rule substrates like E6AP for degradation [52]. UBR4 contains neither a HECT nor a RING domain.
A dearth of UBR4 literature exists because of the difficulty that manipulating the UBR4 gene presents. The UBR4 gene contains 106 exons, and produces multiple splice variants that conceivably have different functions. UBR4 forms a chromatin scaffold when bound to retinoblastoma protein (Rb) in the nucleus, and it also influences cytoskeleton organization by binding clathrin in the cytoplasm [53]. Both of these are structural roles for which no N-end rule or other E3 ligase activities have been detected. A second virus, human papilloma virus, is known to exploit UBR4's role in cellular morphology to initiate anchorage-independent growth and cellular transformation [54], [55]. Although UBR4 is part of a family of UBR E3 ligases involved in the N-end rule pathway, the involvement of the N-end rule in the NS5-dependent degradation of STAT2 seems unlikely. Our group has previously demonstrated that the identity of NS5's first residue is not relevant for STAT2 degradation as long as the precursor is correctly processed [13]. In addition, we show that residues T2 and G3 of NS5 are critical for binding to UBR4 and for mediating STAT2 degradation, but they are considered to be stabilizing residues within the N-end rule. This does not exclude UBR4 from having E3 ligase activity that is independent of the N-end rule. Though it lacks an obvious catalytic domain such as the HECT or RING domains, UBR4 contains a cysteine-rich domain (CRD) that is unique to the UBR4 group. It is currently unknown if CRD functions as a ligase domain. Our experiments with the N-terminal region of UBR4 suggest that domains from the C terminus, which contain the CRD, are necessary for its function in DENV-mediated STAT2 degradation.
Finally, we propose two working models: one based on the hypothesized UBR4 E3 ligase catalytic activity, and another which postulates a scaffolding role for UBR4 based on its described interactions with clathrin and retinoblastoma protein (Figure 7). Efforts are currently being made to clone and express the predicted UBR4 isoforms so as to further evaluate the function of UBR4 in DENV-mediated STAT2 degradation, and to explore its potential as a target for rationally-designed DENV therapeutics.
293T, Hepa1.6, U6A, BHK and Vero cells were maintained in DMEM (Life Technologies) supplemented with 10% fetal bovine serum (Life Technologies) and 1% penicillin/streptomycin mix (Life Technologies). C6/36 cells were maintained in RPMI 1640 medium supplemented with 10% fetal bovine serum (Life Technologies). Hepa1.6 cells were kindly provided by Matthew Evans (Mount Sinai School of Medicine, New York, NY). U6A cells were a kind gift of George Stark (Lerner Research Institute, Cleveland, OH) and were previously described [43].
High-titer stocks of DENV1, DENV2, DENV3, DENV4, yellow fever virus (YFV-17D) and encephalomyocarditis virus (EMCV) were obtained by passage in C6/36 cells, BHK cells, and Vero cells, respectively.
pCAGGS-CTAP was a kind gift from Luis Martinez-Sobrido (University of Rochester). A gene cloned into pCAGGS-CTAP produces a fusion protein that is C-terminally tagged with a TAP tag: calmodulin binding protein followed by two tobacco etch virus (TEV) cleavage sites followed by a protein A tag. The sequences of the primers used for the construction of the RFP-ubiquitin-NS5 fragment that was cloned into pCAGGS-CTAP are available upon request. The primers sequences used for cloning UBR4-NT (1–2233 of UBR4, NCBI Accession Q5T4S7) into pCDNA6 are also available upon request. All other viral gene expression constructs were cloned into pCAGGS-HA and were described previously [13], [15]. The Flag-tagged STAT1, STAT2 and chimeric STAT constructs were previously described [44].
All cells were transfected using Lipofectamine 2000 (Invitrogen) according to the manufacturer's protocol. 293T cells were transfected at a ratio 1∶2 (µg plasmid DNA: µL Lipofectamine 2000) while Vero, Hepa1.6 and U6A cells were transfected at a ratio 1∶3.
Cells were lysed for tandem affinity purification (TAP) or immunoprecipitation two days post transfection. For tandem affinity purification, cells were lysed in TAP buffer (25% glycerol, 50 mM Tris HCL pH 8, 0.5% NP40, 200 mM NaCl, 1 mM β-mercaptoethanol, protease inhibitor cocktail (Roche). Lysates were spun at 15,000 g for 10 minutes and the supernatant was incubated with IgG beads (Roche) for 4 hours then washed with TAP buffer. The beads were then incubated with TEV buffer (TAP buffer containing 0.5 mM EDTA, 1 mM DTT units) and 50 units AcTEV enzyme (Invitrogen) overnight. The beads were spun at 15,000 g for 10 minutes then the supernatant was applied to calmodulin beads (Roche) in a calmodulin bead (CB) buffer (TAP buffer containing 4 mM CaCl2 and 2 mM imidazole) for 8 hours, then washed in CB buffer. The protein was eluted from the calmodulin beads by boiling for five minutes in Laemmli sample buffer (BioRad).
For immunoprecipitation, cells were lysed in TAP buffer then incubated for two hours with anti-FLAG or anti-HA beads (#F2426 and #E6779 respectively, Sigma-Aldrich) or for 4 hours with rabbit anti-STAT2 antibody (Santa Cruz) or mouse anti-GFP antibody (Abcam) followed by 2 hours of protein A-agarose (Roche) or protein G-agarose (Roche), respectively. The beads were washed with TAP buffer then the protein was eluted from the beads by boiling for five minutes in Laemmli sample buffer (BioRad).
Proteins lysates were boiled with Laemmli sample buffer and resolved on 4–15% or 7.5% gels (BioRad) and then transferred to PVDF membrane (Millipore) by standard methods. Membranes were blocked with 3% BSA in TBS-Tween (20 mM Tris-HCl, pH 7.4; 150 mM NaCl; 1% Tween) and then incubated with antibodies and subjected to western blot. Benchmark Protein Ladder (Invitrogen) was used to depict the size of protein bands. The primary antibodies used in this study were: rabbit anti-human STAT2 (sc-476, Santa Cruz), rabbit anti-mouse STAT2 (4597, Cell Signaling), rabbit anti-STAT1 (610120, BD Biosciences), mouse anti-β-tubulin (T0198, Sigma-Aldrich), mouse anti-HA (H9658, Sigma-Aldrich), mouse anti-Flag (F3165, Sigma-Aldrich), mouse anti-V5 (R960-25, Invitrogen), rabbit anti-UBR4 (ab86738, Abcam), HRP-linked anti-GAPDH (ab9385, Abcam), rabbit anti-UBR5 (ab70311, Abcam), and rabbit anti-NS5 [13]. The secondary antibodies used in this study were HRP-linked anti-mouse IgG (#NA931V, GE Healthcare) and HRP-linked anti-rabbit IgG (#NA934V, GE Healthcare). Where indicated, quantification of western blots was done by using Image J to compare the ratio of UBR4 (seen as two bands or one band based on the resolution of the tris-glycine gel used) to NS5.
To analyze the intracellular localization of endogenous UBR4 and DENV NS5, Vero cells that had been grown on glass cover slips were transfected with 1 µg of the indicated plasmids. After 24 hours post infection, cells were fixed and permeabilized for 30 minutes with ice cold methanol acetone (1∶1, v/v) and 0.5% NP-40, then washed with PBS. Following PBS washes, cells were blocked in blocking buffer (0.2% cold waterfish gelatin (Sigma-Aldrich, USA) and 0.5% BSA in PBS) for 1 hour at room temperature (RT), and stained with primary antibodies (anti-UBR4 at a 1∶100 dilution, and anti-HA at a 1∶1000 dilution) overnight at 4°C. The cells were washed in PBS and incubated with secondary antibodies to Alexa Fluor 488 and Alexa Fluor 555 (Invitrogen, USA) at 1∶500 dilution in blocking buffer for 1 hour at RT. Nuclear chromatin staining was performed by incubation in blocking solution containing 0.5 mg/ml 4′,6-diamidino-2-phenylindole, DAPI (Sigma-Aldrich). Cells were washed and coverslips mounted using Prolong antifade reagent (Invitrogen). Images were captured using a Leica SP5-DM confocal microscope at the Microscopy Shared Research Facility at Mount Sinai School of Medicine.
The 293T and Vero cell lines stably expressing non-silencing shRNA or shRNA against UBR4 were made by infecting cells with shRNA-encoding lentiviruses (according to the manufacturer's protocol) and selecting cells with puromycin (1 µg/ml for 293T cells and 5 µg/ml for Vero cells) for two weeks before DENV, YFV or EMCV infection.
The lentiviruses used to make 293T shUBR4 clones were purchased from Open Biosystems. Lentivirus 1 (Clone ID: V3LHS_318553; target sequence: CGCTTCGACTTCATGCTCT) targets nucleotides 11132–11150 of UBR4. Lentivirus 2 (Catalog #: V3LHS_318554; target sequence: CGGATCAGCTCCTATGTCA) targets nucleotides 3140–3158 of UBR4. Lentivirus 3 (Catalog #: V3LHS_318555; target sequence: AGGTTTTTGTCTACAATGA) targets nucleotides 2357–2375 of UBR4. The non-silencing control lentivirus was catalog number RHS4348. The Vero shUBR4 clone was made using lentivirus 1, while the Vero shControl was made using non-silencing control lentivirus.
Cells were infected at the indicated multiplicity of infection (MOI) and maintained in DMEM with 10% FBS. For exogenous IFN-I treatment of Vero cells, 1000 units/ml IFNβ (PBL Interferon Source) were added at 6 hours post infection. For western blotting, cells were lysed with TAP buffer at each time point and the lysates were clarified by centrifugation then boiled in Laemmli sample buffer. For virus titration, cells and media were frozen at each time point and clarified by centrifugation. DENV and YFV titers were measured by plaque assay on BHK-21 cells, and EMCV titers were measured by plaque assay on Vero cells.
Peripheral blood mononuclear cells were isolated from buffy coats of healthy human donors by Ficoll density gradient centrifugation (Histopaque, Sigma Aldrich) as previously described [33]. Buffy coats were obtained from the Mount Sinai Blood Donor Center and New York Blood Center. Briefly, CD14+ cells were purified using anti-human CD14 antibody-labeled magnetic beads and iron-based MiniMACS LS columns (Miltenyi Biotech). After elution from the columns, 2×105 cells were plated in 96-well plates and transduced with VSV-G pseudo-typed SIV VLPs (pSIV3+, an SIV gag-pol expression plasmid containing Vpx) and lentiviral control or UBR4-specific shRNA vectors for 3 hours by spinoculation in the presence of 2 µg/mL polybrene (Sigma) with sufficient viruses to transduce >95% of the cells. Subsequently, cells were washed, resuspended in DC medium (RPMI medium [Invitrogen], 10% fetal calf serum [HyClone], 100 U/ml penicillin, and 100 µg/ml streptomycin [Invitrogen]) supplemented with 500 U/ml human granulocyte-macrophage colony-stimulating (Peprotech), and 1,000 U/ml human interleukin-4 (hIL-4; Peprotech), and incubated for 5 days at 37°C. At 5 days post transduction, MDDCs were either mock infected or infected with DENV2 at an MOI of 3. At 12 hours post infection (hpi) cells were harvested for qPCR analysis, and at 48 hpi supernatants were collected for titration of virus levels by plaque assay on BHK-21 cells. Cells were also harvested for cytometry analysis.
Total RNA was isolated from samples using the RNeasy kit (Qiagen) and subjected to DNase digestion with Turbo DNase (Ambion). Reverse transcription was performed using the high capacity cDNA reverse transcription kit (Applied Biosystems). qPCR was performed in 384-well plates in triplicates using SYBR green I master mix (Roche) in a Roche LightCycler 480. Relative mRNA values were calculated using the ΔΔCt method using 18S rRNA as internal control and plotted as fold change by normalizing to mock-control samples.
UBR4 qPCR primers: Forward = GGTGTTCCAGAGGCTAGTGATC; Reverse = CCAACTGCTTCTGCGGTTCCTT
ISG15 qPCR primers: Forward = TCCTGGTGAGGAATAACAAGGG; Reverse = GTCAGCCAGAACAGGTCGTC
RIG-I qPCR primers: Forward = GGCATGTTACACAGCTGACG; Reverse = TGCAATATCCTCCACCACAA
ISG54 qPCR primers: Forward = ATGTGCAACCTACTGGCCTAT; Reverse = TGAGAGTCGGCCCATGTGATA
18S RNA qPCR primers: Forward = GTAACCCGTTGAACCCCATT; Reverse = CCATCCAATCGGTAGTAGCG
DENV-infected DCs were fixed and permeabilized with Cytofix and Cytoperm reagent (BD Pharmingen) according to the manufacturer's recommendations. Then, cells were stained with 4G2 (ATCC), a mouse monoclonal antibody specific for the E protein, as a primary antibody and a FITC-labeled anti-mouse antibody as a secondary antibody. Flow cytometry was performed using a FACScan flow cytometer (Becton Dickinson) and analyzed with FlowJo software.
|
10.1371/journal.pntd.0007414 | Persistent transmission of Plasmodium malariae and Plasmodium ovale species in an area of declining Plasmodium falciparum transmission in eastern Tanzania | A reduction in the global burden of malaria over the past two decades has encouraged efforts for regional malaria elimination. Despite the need to target all Plasmodium species, current focus is mainly directed towards Plasmodium falciparum, and to a lesser extent P. vivax. There is a substantial lack of data on both global and local transmission patterns of the neglected malaria parasites P. malariae and P. ovale spp. We used a species-specific real-time PCR assay targeting the Plasmodium 18s rRNA gene to evaluate temporal trends in the prevalence of all human malaria parasites over a 22-year period in a rural village in Tanzania.We tested 2897 blood samples collected in five cross-sectional surveys conducted between 1994 and 2016. Infections with P. falciparum, P. malariae, and P. ovale spp. were detected throughout the study period, while P. vivax was not detected. Between 1994 and 2010, we found a more than 90% reduction in the odds of infection with all detected species. The odds of P. falciparum infection was further reduced in 2016, while the odds of P. malariae and P. ovale spp. infection increased 2- and 6-fold, respectively, compared to 2010. In 2016, non-falciparum species occurred more often as mono-infections. The results demonstrate the persistent transmission of P. ovale spp., and to a lesser extent P. malariae despite a continued decline in P. falciparum transmission. This illustrates that the transmission patterns of the non-falciparum species do not necessarily follow those of P. falciparum, stressing the need for attention towards non-falciparum malaria in Africa. Malaria elimination will require a better understanding of the epidemiology of P. malariae and P. ovale spp. and improved tools for monitoring the transmission of all Plasmodium species, with a particular focus towards identifying asymptomatic carriers of infection and designing appropriate interventions to enhance malaria control.
| The reduction in the global burden of malaria has encouraged efforts for elimination. Attempts to control and monitor transmission have mainly focused on the predominant malaria parasites Plasmodium falciparum and P. vivax. However, eliminating malaria requires the elimination of all human malaria parasites and limited interest has been directed towards estimating the disease burden attributable to the neglected malaria parasites P. ovale spp. and P. malariae. The authors used molecular methods to analyse 2897 blood samples collected in five cross-sectional surveys over a period of 22 years, and described the transmission patterns of all human malaria parasites in a Tanzanian village. They demonstrate a persistent transmission of P. malariae and P. ovale spp. despite a substantial reduction in transmission of P. falciparum, highlighting the need for more attention towards non-falciparum malaria. The authors discuss the implications of these findings in the context of current efforts for regional malaria elimination.
| Since the turn of the millennium, there has been a substantial reduction in the global burden of malaria including a reduction in the clinical incidence of both Plasmodium falciparum and P. vivax malaria [1–3]. This reduction has largely been attributed to an increase in malaria control efforts using insecticide treated nets (ITNs), indoor residual spraying (IRS), improved diagnostics through the use of rapid diagnostic tests (RDTs), and better access to highly efficacious artemisinin based combination therapy (ACT) [2,4]. Several countries are approaching a hypoendemic or unstable transmission setting and in 2016 the World Health Organization (WHO) identified 21 countries in which malaria elimination was deemed feasible by the year 2020 [3].
The focus of malaria control programmes has historically mainly been directed towards limiting transmission of P. falciparum and to a lesser extent also of P. vivax. However, achieving malaria elimination requires the elimination of all malaria parasites infecting humans (i.e. P. falciparum, P. vivax, P. malariae, and P. ovale curtisi and wallikeri as well as the simian species, e.g. P. knowlesi in South East Asia) [3,5,6]. While P. malariae and P. ovale spp. are reported to be widely distributed throughout tropical Africa and other malaria endemic regions of the world, their epidemiology remains far less studied than that of P. falciparum and P. vivax and both global and local temporal trends in transmission intensity are largely unknown [7–10].
Although generally considered benign, P. malariae and P. ovale spp. have the potential to cause significant morbidity. Infection with P. malariae is an established cause of nephrotic syndrome, which can lead to progressive renal failure, particularly in adolescents or young adults [11,12] and has been associated with a high burden of anaemia [13]. Furthermore, P. ovale spp. has in recent years been recognised as a potential cause of severe malaria [14–16].
In malaria endemic areas of tropical Africa, the majority of clinical malaria attacks are attributed to P. falciparum [17]. This is partly due to an under-diagnosis of non-falciparum infections. Detection of infection and accurate discrimination of Plasmodium species using microscopy requires both highly skilled microscopists and good quality microscopes. It is particularly difficult in asymptomatic low density infections, or mixed species infections with P. falciparum, in which both P. malariae and P. ovale spp. frequently occur [7,18,19]. In addition, RDTs which are currently used as an important diagnostic tool in many settings, have shown poor performance for the detection of P. malariae and P. ovale spp. [20–23]. Given the potential to cause morbidity in combination with the under-diagnosis of non-falciparum malaria in many settings, it is likely that the global disease burden attributable to P. malariae and P. ovale spp. is largely underestimated.
Over the past decade, cross-sectional studies using PCR for parasite detection have generated evidence that the prevalence of both P. malariae and P. ovale spp. is greater than was previously reported [5,7]. These surveys usually find P. malariae to be more common than P. ovale spp. and have estimated the prevalence to range from 1 to 35% and 1 to 25%, respectively, depending on transmission setting [24–29]. Although a large number of longitudinal studies from sub-Saharan Africa have reported gradual reductions in the prevalence of P. falciparum, there have been few studies, and none using PCR, that investigate how the prevalence of P. malariae and P. ovale spp. change over time as the prevalence of P. falciparum decreases [2,30].
In this study, we used PCR to evaluate changes in the prevalence of P. falciparum and non-falciparum infection by analysis of samples collected in five cross-sectional surveys in a Tanzanian village over a period of 22 years. We assessed the temporal trends in prevalence of all human Plasmodium spp. in an area experiencing a substantial reduction in the prevalence and transmission of P. falciparum.
The Nyamisati Malaria Research Project was established in 1985 in Nyamisati, a rural fishing village located 150 km south of Dar es Salaam in the Rufiji river delta area in Kibiti District, Tanzania. Malaria transmission in the area is perennial with seasonal fluctuations. Within the project, the same research team conducted repeated cross-sectional surveys between 1986 and 2016 [31]. The surveys consisted of a physical examination including measurement of body temperature, as well as the collection of a venous blood sample and a blood smear. Each participating individual was assigned a unique individual identifier and demographic information (i.e. age, gender and household membership) was collected. The main intervention to reduce malaria transmission in the village was to provide rapid access to diagnosis and antimalarial treatment free of charge. Sulfadoxine-pyrimethamine (SP), alone or in combination with oral quinine, was the first-line treatment from the early 1990’s until ACTs became readily available in the village in 2009. In addition, ITNs were distributed after the surveys in 1993 (300 ITNs to pregnant women and young children) and in 1999 (900 ITNs). Additionally, long-lasting insecticidal nets (LLINs) were distributed after the survey in 2010. The estimated access to bed nets after the surveys was 45% in 1993–1994, 100% in 1999, and approximately 70% in 2010, assuming an average protection of 1.8 individuals per net [32]. Other vector control measures, e.g. indoor residual spraying, have not been used in the village. The study site, the research project, and temporal trends in the transmission of P. falciparum have been described in previous publications [31,33]. The present study is based on five cross-sectional surveys conducted at the start of the long rainy season (March-May) in 1994, 1995, 1999, 2010, and in 2016. All villagers were invited to participate in these surveys of the Nyamisati population. The final number of sampled individuals varied by survey year but are considered representative of the Nyamisati population, thus at each cross-section including a random selection of individuals with different levels of exposure. In the years when the cross-sectional survey sampling did not equally cover the entire age-range of the population, this was adjusted for in the statistical analyses. The present study included 2897 samples collected from 2005 unique individuals participating in the five cross-sectional surveys. A number of individuals (n = 544) participated in multiple surveys (range: 2–5) over the years, thus contributing 1435 of the total 2897 samples.
The project was approved by the Nyamisati village board and ethical approval was granted by the Ethical Review board of the National Institute for Medical Research in Tanzania, the Regional Ethical Committee at Karolinska Institutet (Dnr. 00–084), and the Regional Ethical Review Board in Stockholm, Sweden (Dnr. 2012/1151–32). In addition, ethical approval for the 2016 survey was granted by the Institutional Review Board at Muhimbili University of Health and Allied Sciences, a delegated activity of the Medical Research Coordinating Committee (MRCC), Tanzania. Oral informed consent was obtained from all study participants and/or their guardians and was documented in a research database. The use of oral consent was approved by the respective ethical review boards and was selected due to a low degree of literacy in the village.
Venous blood was collected in EDTA, separated, and stored frozen as plasma and packed cells. DNA was extracted from packed cells using Qiagen blood mini kit (Qiagen, Germantown, MD, USA) (1994–1999), a BioRobot M48 Robotic Workstation (Qiagen) (2010), or a magnetic bead separation method using a Hamilton Chemagic Star Robot (Hamilton, Bonadouz, Switzerland) (2016). Real-time PCR was used to qualitatively detect Plasmodium infection (P. falciparum, P. vivax, P. ovale spp., and P. malariae) in the ABI Taqman 7500 or QuantStudio™ 5 Real-Time PCR system (Applied Biosystems, Foster City, CA, USA), following a previously described protocol [34]. The master mix for a single reaction included species-specific probes and forward primers for all four Plasmodium species used in combination with a conserved reverse primer. The P. ovale- and P. malariae-probes (synthesized by BioSearch Technologies, Novato, CA, USA), and the P. vivax- and P. falciparum-probes (synthetized by Applied Biosystems) were each labelled with a distinct fluorophore, and, depending on the master mix, either ROX or Mustang Purple was used as the reference dye [35]. The reaction was performed in a final volume of 25 μl per well containing 5 μl DNA (corresponding to 5 μl of whole blood), 12.5 μl of either TaqMan universal master mix or TaqMan multiplex master mix (Applied Biosystems), 0.5 μl (10 μmol/L) of the P. falciparum-specific forward primer, 0.125 μl (10 μmol/L) of each of the other species-specific forward primers and 0.5 μl (10 μmol/L) of the reverse primer, 0.2 μl (10 μmol/L) of each species-specific probe, passive reference dye ROX or Mustang Purple and DNA/RNA-free water. The samples were run using a cut-off of 45 cycles to define positive samples, starting with 95 °C for 20 s, followed by the thermal cycles of 95 °C for 1 s and of 60 °C for 20 s. Standards, negative and species-specific positive controls were included on each plate. The assay was optimised to detect all species simultaneously, with a limit of detection of approximately 0.5 parasites per μl blood. The PCR method does not distinguish between the two sympatric species of P. ovale, i.e. P. ovale curtisi and P. ovale wallikeri, but we established that it could detect both of them using serial dilutions of positive controls (kindly provided by Colin Sutherland, LSHTM).
Data were analysed using R version 3.4.3 (Vienna, Austria. URL https://www.R-project.org) and Stata 14 (StataCorp., College Station, TX, USA). For the purpose of the analyses, a mixed species infection was defined as an infection with P. falciparum and P. malariae and/or P. ovale spp. A non-falciparum infection was defined as an infection with either P. malariae or P. ovale spp. or both. Fever at the time of survey was defined as an axillary body temperature of above 37.5 °C and/or a history of fever or “hot body” within 24 hours. Generalized estimating equation (GEE) regression models were used to estimate population-averaged effects while accounting for the statistical dependency of repeated observations from individuals participating in multiple surveys [36]. Multivariable logistic regression models were used to evaluate the prevalence of each of the Plasmodium spp. independently over time while adjusting for covariates, i.e. age, sex, and fever at the time of survey. A multinomial logistic regression model was used to jointly evaluate the relative risk ratio of P. falciparum mono-infections, mixed-species infections, and non-falciparum infections over time while adjusting for the above specified covariates. In all analyses, age was treated as a categorical variable with five categories (<5, 5–8, 9–12, 13–16 and >16 years). P-values <0.05 were considered significant.
The population characteristics at each of the five cross-sectional surveys are presented in Table 1. Among the total 2897 samples analysed, 1291 (44.5%) were positive for P. falciparum, 266 (9.2%) for P. malariae, and 136 (4.7%) for P. ovale spp. (Fig 1). No samples were positive for P. vivax. The observed overall parasite prevalence by PCR, including all species, was high during the 1990’s, ranging from 66.1 to 71.6%, but dropped to 19.1% in 2010 and to 17.9% in 2016. Plasmodium falciparum was most commonly detected, accounting for 76.3% of positive tests. Plasmodium malariae was the second most commonly detected species, found in 15.7% of positive tests, while P. ovale spp. were detected in 8.0% of positive tests (Fig 1).
The observed year-wise species-specific prevalence is presented in Fig 2A and stratified by age in Fig 2B. Logistic regression models were used to evaluate the temporal trends in the prevalence of each parasite species independently while adjusting for covariates (i.e. age, sex and fever at the time of survey). The temporal trends are presented as the model-estimated prevalence of infection (with all covariates at their mean values) as well as the corresponding adjusted odds ratios (OR). The logistic regression model estimated a slight decrease in the prevalence of P. falciparum during the 1990’s, from 73.9% in 1994 to 66.3% in 1999, but the prevalence was markedly reduced thereafter, reaching 17.4% in 2010. The adjusted OR for P. falciparum infection, comparing 1999 and 2010 to 1994 was 0.70 (95% CI 0.56–0.88; p = 0.003) and 0.07 (95% CI 0.06–0.10; p<0.001), respectively, i.e. corresponding to a 93% reduction in the odds of infection from 1994 to 2010 (Table 2). Compared to 2010, the prevalence of P. falciparum infection was further significantly reduced to 10.2% in 2016 (adjusted OR: 0.54; 95% CI 0.39–0.75; p<0.001).
The prevalence of P. malariae remained relatively stable during the 1990’s, with the model-estimated prevalence varying from 11.3% to 16.2%. This was followed by a reduction in the prevalence to 1.1% in 2010 corresponding to a significant 92% reduction in the odds of P. malariae infection from 1999 to 2010 (adjusted OR: 0.08; 95% CI 0.04–0.15; p<0.001) (Table 2). However, in contrast to the further reduction detected for P. falciparum, there was a significant increase in the prevalence of P. malariae infection to 2.4% in 2016 (adjusted OR: 2.24; 95% CI 1.01–4.97; p = 0.047) (Table 2).
Plasmodium ovale spp. were overall least frequently detected with the prevalence of infection declining gradually during the 1990’s, from 10.0% in 1994 to 4.4% in 1999 (adjusted OR: 0.42; 95% CI 0.26–0.67; p<0.001). Similar to P. falciparum and P. malariae, the prevalence of P. ovale spp. was further reduced to 0.6% in 2010 (adjusted OR: 0.13, 95% CI 0.05–0.36, p<0.001) corresponding to an estimated overall 94% reduction in the odds of infection between 1994 and 2010 (adjusted OR: 0.06, 95% CI: 0.02–0.15, p<0.001) (Table 2). However, similarly to P. malariae, and in contrast to P. falciparum, there was a subsequent significant increase in the infection prevalence of P. ovale spp. to 3.6% in 2016 (adjusted OR of 5.9; 95% CI 2.2–15.8, p<0.001) compared to 2010 (Table 2), making P. ovale spp. the second most common infection after P. falciparum in 2016.
For all species, the observed prevalence was overall highest among 5 to 16 year old children (Fig 2B). A shift of the peak prevalence towards older children was observed for P. falciparum infection in 2010, but was not as apparent for the other species.
Plasmodium falciparum mono-infections represented the majority of infections and accounted for overall 73.1% (95% CI 70.7–75.4%) of infections during the study period (Fig 1). Mixed species infections with P. falciparum and non-falciparum infections accounted for overall 21.6% (95% CI 19.4–23.9%) and 5.3% (95% CI 4.2–6.6%) of infections, respectively. In P. falciparum mixed species infections, the combination with P. malariae was most common, followed by P. ovale spp., and lastly by infections with all species (Table 3). A non-falciparum infection with both P. malariae and P. ovale spp. was detected only once throughout the study period (Table 3).
A multinomial logistic model was used to estimate the relative risk ratio of P. falciparum mono-infection, P. falciparum mixed infection and non-falciparum infection compared to being uninfected over time. The adjusted probability (adjusting for age, gender, and fever at time of survey) of being infected with either a P. falciparum mono-infection, mixed infection, or a non-falciparum infection declined significantly from 1994 to 2010 (Table 4, Fig 3A). With all covariates at their mean value, the model estimated a reduction in the prevalence of P. falciparum mono-infection from 52.0% to 16.8%, for mixed infections from 22.3% to 0.7%, and for non-falciparum infections from 2.1% to 1.0% (Fig 3A, Table 4). From 2010 to 2016, the model-predicted probability of P. falciparum mono-infections continued to decline while the probability of both mixed infections and non-falciparum infections increased significantly from 0.7% to 2.1% and 1.0% to 3.3%, respectively (Fig 3A and 3B, Table 4). In the beginning of the study period approximately 90% of P. malariae and P. ovale spp. infections were detected as mixed species infections with P. falciparum (Fig 3B). However, this changed over time towards a greater proportion of these infections occurring as mono-infections. In 2016, 60% of non-falciparum infections were found to occur as mono-infections (Fig 3B, Tables 3 and 4).
The number of symptomatic infections occurring at the time of the cross-sectional survey varied over the years and was greater for P. falciparum mixed and mono-infections compared to non-falciparum infections (S1 Table). The odds of presenting with fever at the time of survey (adjusted for age, sex, and survey year) was estimated to be approximately 4 to 5 times greater if harbouring a P. falciparum mono-infection (adjusted OR: 4.9, 95% CI 1.45–16.67, p = 0.011) or a P. falciparum mixed infection (adjusted OR: 3.84, 95% CI 1.08–13.57, p = 0.036) compared to a P. malariae and/or P. ovale spp. infection. There was no significant difference in the odds of presenting with fever at the time of survey between those infected with P. malariae and/or P. ovale spp. and those who were PCR negative (adjusted OR: 1.54, 95% CI: 0.45–5.19, p = 0.49).
In the present study we assessed the prevalence of Plasmodium spp. in five cross-sectional surveys over two decades in a Tanzanian village experiencing a substantial reduction in the prevalence of P. falciparum infection. We used real-time PCR to obtain a high sensitivity and specificity in detection of both mixed-species and non-falciparum infections. Plasmodium malariae and P. ovale spp., but no P. vivax, infections were detected throughout the study as both mixed and mono-infections. Although the prevalence of all species declined over time, the decline in P. ovale spp. prevalence was smaller leading to a relative increase in the number of infections being due to P. ovale spp. Furthermore, there was a shift of P. malariae and P. ovale spp. infections from occurring almost exclusively as mixed species infections with P. falciparum to occur more commonly as mono-infections. This illustrates that the transmission patterns of non-falciparum species do not necessarily follow those of P. falciparum. These findings emphasise the need to carefully monitor the prevalence and transmission trends of non-falciparum species of Plasmodium to improve our understanding of their epidemiology and to guide specific interventions aimed at achieving malaria control and elimination.
Previous studies of the Nyamisati cohort have examined the changing transmission intensity of P. falciparum between 1985 and 2010 [31,33]. Here, the expansion of the analysis to non-falciparum species revealed a parallel reduction in the odds of infection of 93% for P. falciparum and 94% for P. malariae, and P. ovale spp., between 1994 and 2010. We then observed a further 46% reduction in the odds of P. falciparum infection until 2016. In contrast, the odds of P. malariae and P. ovale spp. infection increased by 2-fold and 6-fold, respectively, from 2010 to 2016. The observed increase in the relative contribution of non-falciparum infections to the overall prevalence of infection is in line with reports from Burkina Faso where an increase in the prevalence of P. malariae infection was observed by microscopy as transmission of P. falciparum decreased [37]. However, the data is somewhat contrasted by the findings from Dielmo, Senegal, of a near elimination of P. malariae and P. ovale spp. between 1990 and 2010 when the prevalence of P. falciparum decreased [38]. Although this longitudinal study [38] used only microscopy for parasite detection, the near absence of P. malariae and P. ovale spp. in Dielmo has later been confirmed using PCR [39]. These differences between geographical sites highlight the need to obtain local estimates of the transmission patterns of all Plasmodium species.
The reduction in the prevalence of P. falciparum in Nyamisati between 1985 and 2010 has been attributed to the presence of a research and healthcare team who provided prompt access to diagnosis and treatment, more effective antimalarial treatment (i.e. ACTs), and vector control measures (ITNs were distributed after the surveys in 1993 and 1999) [31,33]. LLINs were distributed to all survey participants after the survey in 2010. This might have contributed to the further decline in P. falciparum prevalence observed after 2010. However, it does not appear to have affected the prevalence of P. malariae and P. ovale spp. to the same extent. In 2016, P. ovale spp. superseded P. malariae as the second most commonly detected species and its prevalence returned to levels similar to those in 1999, i.e. prior to any large-scale intervention with bed nets at the study site [31].
In Tanzania, all Plasmodium species appear to share the same primary malaria vectors [40]. Entomological data are not available from the study site but according to previous entomological studies in the Rufiji delta area, the important primary malaria vectors are members of the An. gambiae complex (e.g. An. gambiae ss, An. arabiensis and An. merus), all of which are highly anthropophilic and predominantly indoor-biting at night [41,42]. Accumulating evidence suggest that large-scale distribution of LLINs affects both the behaviour and composition of vector populations, making secondary vectors, which are prone to outdoor biting, more important for malaria transmission [40,43,44]. Specific changes in vector populations could in theory affect the transmission of each Plasmodium species differently, but whether P. malariae and P. ovale spp. are more or less efficiently transmitted by the secondary vectors compared to P. falciparum is currently unknown.
According to current WHO guidelines, primaquine treatment is recommended to prevent relapses of P. ovale spp. infections [45]. However, to our knowledge, primaquine has not been used in the village. The absence of relapse prevention, which is likely to be required in order to eliminate P. ovale spp., could theoretically contribute to a lower relative reduction in transmission of P. ovale spp. compared to the other species. With the available data, it is not possible to determine whether this could explain the observed transmission patterns in Nyamisati. The first-line antimalarial treatment used at the study site did not differ depending on Plasmodium species but changed during the study period from SP to ACT when ACT became readily available in the village in 2009 [31]. As for P. falciparum, ACTs are highly efficacious against asexual stages of both P. malariae and P. ovale spp. and the change of first-line anti-malarial is unlikely to have contributed to the smaller relative reduction in non-falciparum infections [46,47].
In sub-Saharan Africa, a majority of infections with P. malariae and P. ovale spp. are reported to occur as mixed species infections with P. falciparum [7,48–50]. Although a vast majority (approximately 90%) of non-falciparum infections occurred as mixed species infections during the early years of the study, this changed over time. At the end of the study period, approximately 31% of P. malariae and 57% of P. ovale spp. infections occurred as mono-infections. Furthermore, our data indicate that individuals harbouring non-falciparum infections are less likely to be symptomatic and thereby may be less likely to seek medical treatment. The observed shift has important implications for malaria control and monitoring of transmission intensity. It increases the importance of accurately identifying each species independently and highlights the need to detect and actively target asymptomatic carriers of infection in order to provide interventions that can reduce the transmission of non-falciparum malaria.
The present study is somewhat limited by the repeated cross-sectional design as well as the relatively long time-intervals between the surveys. However, a substantial number of individuals participated in multiple surveys, providing a longitudinal aspect of the study design. An even closer follow-up on the individual level may have provided a more detailed understanding of the epidemiology of non-falciparum infections. To account for annual variation in the start of the peak transmission season, all surveys were conducted during the beginning of the long rainy season (March-May, depending on year) [31]. During late 2015 and early 2016, the coastal regions of Tanzania were heavily affected by an El Niño Southern oscillation which lead to greater than average rainfall in the Rufiji area until mid-February 2016, as well as greater than average temperatures and humidity during the following months [51]. This likely increased both the Anopheles vector density and the rate of parasite development within the vector and thus the potential for malaria transmission [51].
Another limitation of the study is that the PCR-method used does not distinguish between the two recently described sympatric species of P. ovale (P. ovale curtisi and P. ovale wallikeri) [15,34]. Although the real-time PCR sensitively detects both, we were unable to examine whether both species are endemic in this area and how their relative frequencies might have changed over time.
Because molecular methods are still expensive and often difficult to implement in large scale for routine surveillance, detection of Plasmodium infection relies largely on the use of microscopy and/or RDTs that lack sensitivity for the detection P. malariae and P. ovale spp. [29,52,53]. Data from Kenya suggests that as much as 50% of P. malariae infections may occur as sub-microscopic infections [29]. In addition, current WHO guidelines regarding the selection and procuration of RDTs are based on the assumption that a vast majority of non-falciparum infections occur as mixed species infections [49,54]. The guidelines state that RDTs based only on the detection of P. falciparum histidine rich protein (HRP-2) are sufficient in most areas of sub-Saharan Africa [49,54,55]. The issue of using a P. falciparum HRP-2-only test, which by design cannot detect non-falciparum infections, has recently been recognised as a problem for diagnosis and surveillance in Senegal where P. malariae and P. ovale spp. have also been reported to occur more frequently as mono-infections [56].
Our findings highlight some of the key challenges that will need to be addressed if malaria elimination is to be achieved. The observed increase in the prevalence of P. malaria and P. ovale spp. that occurred while the prevalence of P. falciparum declined may support previously raised concerns that strategies designed for reducing transmission of P. falciparum may be less effective in reducing transmission of the non-falciparum species of Plasmodium [5,57,58]. For P. malariae and P. ovale spp., this is likely due to their species-specific ability to cause persistent asymptomatic infections in combination with a low effectiveness of current diagnostic and surveillance tools which contribute to their resilience to interventions [5]. In order to further limit malaria transmission, it is of utmost importance to be able to identify and target asymptomatic carriers of infection, not only for P. falciparum but also for P. malariae and P. ovale spp. where asymptomatic carriage appears to be even more common [24,38,59]. There is a pressing need for easy-to-implement, cost-effective tools for diagnosis and surveillance (e.g. species-specific RDTs) that can sensitively and accurately detect all species. This could be further improved by the development of reliable species-specific serological tools that can be used to monitor exposure [33,60].
In summary, we observed the maintenance of P. ovale spp., and to a lesser extent of P. malariae, infections despite a substantial and continuous reduction in the prevalence of P. falciparum over a period of 22-years. This demonstrates that the transmission patterns of non-falciparum species do not necessarily follow those of P. falciparum, stressing the need for attention towards P. malariae and P. ovale spp. transmission in Africa. Furthermore, the prevalence patterns observed by PCR highlight the need for field-applicable tools to detect non-falciparum infections. Malaria elimination will require a better understanding of the specific epidemiological features of P. malariae and P. ovale spp. as well as improved tools for efficient monitoring of all Plasmodium species, with a particular focus towards identifying asymptomatic carriers of infection and designing appropriate intervention strategies to reach the goals of elimination.
|
10.1371/journal.ppat.1004592 | A New Family of Secreted Toxins in Pathogenic Neisseria Species | The genus Neisseria includes both commensal and pathogenic species which are genetically closely related. However, only meningococcus and gonococcus are important human pathogens. Very few toxins are known to be secreted by pathogenic Neisseria species. Recently, toxins secreted via type V secretion system and belonging to the widespread family of contact-dependent inhibition (CDI) toxins have been described in numerous species including meningococcus. In this study, we analyzed loci containing the maf genes in N. meningitidis and N. gonorrhoeae and proposed a novel uniform nomenclature for maf genomic islands (MGIs). We demonstrated that mafB genes encode secreted polymorphic toxins and that genes immediately downstream of mafB encode a specific immunity protein (MafI). We focused on a MafB toxin found in meningococcal strain NEM8013 and characterized its EndoU ribonuclease activity. maf genes represent 2% of the genome of pathogenic Neisseria, and are virtually absent from non-pathogenic species, thus arguing for an important biological role. Indeed, we showed that overexpression of one of the four MafB toxins of strain NEM8013 provides an advantage in competition assays, suggesting a role of maf loci in niche adaptation.
| Many bacteria are able to secrete toxins targeted against neighboring cells. In order to protect themselves against their own toxin, they also express an “immunity” protein. In silico analysis of bacterial genomes predicts that numerous genes could encode potential new toxin-immunity systems. The recently described CDI system is involved in contact-dependent inhibition of growth and confers to its host strain a significant advantage in competitive ecosystems such as the gastro-intestinal tract. Indeed, an Escherichia coli CDI+ strain is able to outcompete CDI- strains and to become predominant. Here, we show that a large family of genes called “maf”, found in pathogenic Neisseria species, encodes a toxin-immunity system. We demonstrate that a toxin named MafBMGI-1NEM8013 inhibits the growth of E. coli by degrading RNA and show that the immunity protein MafIMGI-1NEM8013 is able to abolish the toxicity. MafB toxins exhibit highly variable toxic domains. This variability of secreted toxins could be important to compete against bacteria of different species sharing the same reservoir. Since a strain may contain numerous toxin-immunity systems that can all play a role in interbacterial competition, deciphering interactions between these systems will allow a better understanding of complex bacterial communities.
| The growing number of sequenced bacterial genomes has led to the computer-based prediction of numerous novel bacterial factors possibly involved in virulence. As a result, many novel putative bacterial toxins have been identified by sequence-homology criteria. However, very few of these bacterial proteins have been tested for their toxic activity. Using in silico analysis, Aravind and colleagues have recently described widespread genes encoding putative secreted multi-domain toxins grouped under the name of bacterial polymorphic toxin systems (or polymorphic toxin-immunity systems) [1]–[3]. In silico analysis identified over 150 distinct toxin domains in these systems including many putative peptidase, nuclease or deaminase domains. Immunity genes found immediately downstream of the toxin genes encode highly variable proteins that protect bacteria from their own toxins or from toxins secreted by neighboring cells [4]–[6]. Immunity genes are a characteristic of polymorphic toxin systems that distinguishes them from host-directed toxins (i.e. cholera toxin or pertussis toxin) [7]. The polymorphic toxin systems are typically encoded on hypervariable chromosomal islands with characteristics of horizontal gene transfer [1]. These systems are found in both Gram- negative and positive bacteria [8]. The dominant hypothesis is that polymorphic toxin systems are primarily involved in conflict between related bacterial strains. The N-terminal domain of the toxin is typically related to trafficking mode whereas the C-terminal domain carries the toxic activity [1]. In a defined family of polymorphic toxins, the N-terminal domains are similar, while the C-terminal domains are highly variable. Toxins are potentially secreted by Type II, V, VI, or VII (ESX) secretion systems [1]. Toxins are encoded in loci that also contain standalone cassettes and immunity genes. Cassettes encoding alternative C-termini could promote diversity of toxic activities through genetic recombination [1], [9], [10].
The recently described contact-dependent growth inhibition (CDI) systems are a subgroup of polymorphic toxin systems [1], [11]. Toxins encoded by CDI systems are large filamentous proteins that exhibit RHS (rearrangement hotspot) or filamentous haemagglutinin repeats in their central region [4], [6], [8]. Rhs proteins are likely to be exported through the Type VI secretion machinery [6], whereas toxins with filamentous haemagglutinin repeats are exported through the Type 5 Secretion System (T5SS) [8], [12]. The first CDI toxin secreted by a T5SS was reported in Escherichia coli EC93 (CdiAEC93) [11]. E. coli EC93 was found to inhibit the growth of other E. coli strains (i.e. E. coli K-12) in co-culture experiments. The growth inhibition mediated by E. coli EC93 required a direct contact between toxic and target cells. In CDI systems, the toxin CdiA is secreted by an outer membrane transporter named CdiB. CdiA and CdiB are part of a two-partner secretion protein family (type Vb). Subsequently, several studies have demonstrated that CDI systems are present in many species including Neisseria meningitidis [4], [5], [13]. Moreover, it has been recently demonstrated that the two-partner system TpsAB is indeed a functional CDI system in N. meningitidis strain B16B6 [14].
In addition to non-pathogenic commensal species, the genus Neisseria includes two human pathogens: N. gonorrhoeae (the gonococcus) and N. meningitidis (the meningococcus). N. gonorrhoeae colonizes the uro-genital tract and is a common cause of sexually transmitted infections [15]. N. meningitidis is commonly found in the nasopharynx of healthy individuals, where it can cross the mucosal epithelium and cause sepsis or meningitis [16]. For yet unknown reasons, some meningococcal strains belonging to a limited number of clonal complexes, known as hyper invasive clonal complexes, are much more likely to cause disease than others [17].
Comparison of the genomes of related bacteria that exhibit distinct pathogenic phenotypes can identify relevant genetic variations linked to virulence. The availability of complete genome sequences for several strains of both pathogenic and non-pathogenic species of Neisseria genus enabled their in silico comparison [18]–[23]. Genes involved in adherence to epithelial cells, in capsule biosynthesis, or in iron uptake are well known to be crucial for pathogenicity [16], [21], [24]. Nevertheless, their presence is not sufficient to explain the invasiveness of pathogenic strains compared to non-pathogenic strains. Thus, to date, genomic comparisons between pathogenic and non-pathogenic Neisseria species have failed to identify genes sufficient and necessary to cause disease [18], [21], [25]. The accessory genome, which is composed of genes found only in some strains, confers strain-specific traits and is commonly acquired through horizontal transfer [18]. The accessory genome may be linked to virulence as illustrated by pathogenicity islands (PAI) that are present in pathogenic strains, and absent in non-pathogenic strains of one species. PAIs are genomic islands (GIs) encoding virulence factors such as toxins, adhesins or invasins. Identification of GIs is primarily based on a different G+C content from the rest of the genome and on their association with insertion sequence (IS) elements or tRNA genes at their boundaries [26], [27].
There are several identified islands and prophages in meningococci and gonococci [23], [28]–[34]. It has been recently suggested that an island composed of 22 genes in the N. meningitidis isolate 053442 genome (NMCC_0592 to NMCC_0613) and called IHT-G (Island of Horizontally Transferred DNA-G) could be a “meningococcal pathogenicity island-like region” [20]. This island, which is adjacent to a tRNA-Pro gene, contains genes belonging to the multiple adhesin family (maf). Maf proteins were first described in the gonococcal strain MS11 as ligands interacting with a specific glycolipid (GgO4) [35]. Indeed, the heterologous expression of the neisserial protein in E. coli allows bacterial adhesion to GgO4 [35]. Since multiple genes in the gonococcus chromosome encode these proteins, they were subsequently termed “MafA adhesins”. The gene immediately downstream of mafA, the function of which was unknown, was termed mafB because both genes are organized in a putative operon.
In this study, we analyzed loci containing maf genes in several strains of N. meningitidis and N. gonorrhoeae. We propose here a novel uniform nomenclature of these loci. We demonstrated experimentally that mafB genes encode polymorphic toxins and that genes immediately downstream of mafB encode a specific immunity protein (MafI). Furthermore, we demonstrated that overexpression of one of the four MafB toxins of strain NEM8013 provides an advantage in competition assays.
Analysis of the amino acid sequences of several MafB-CT regions using the CDD server revealed homologies with putative or known toxic domains. For instance, proteins encoded by mafBMGI-1NEM8013 (NMV_0410) contains a domain belonging to the RNase EndoU-fold, mafBMGI-1FA19 (NGEG_01276) contains a domain belonging to a nucleotide deaminase superfamily and mafBMGI-5FA1090 (NGO1392) contains a domain belonging to the DNase HNH/EndoVII-fold. In this study, we decided to focus on the four putative MafB toxins encoded in meningococcal strain NEM8013 which are MafBMGI-1NEM8013, MafB2MGI-2NEM8013, MafBMGI-3NEM8013 and MafB1MGI-2NEM8013 (formerly MafB1, MafB2, MafB3 and MafB-related respectively).
In this study, we have demonstrated that mafB encodes a functional polymorphic toxin and mafI, the downstream gene, a specific immunity protein. Focusing on MafBMGI-1NEM8013, we were able to characterize its EndoU ribonuclease activity. Toxins carrying EndoU activities are predicted to be widespread among diverse polymorphic toxin systems, however no EndoU activity carrying toxins had been previously experimentally confirmed in bacteria.
Polymorphic toxins encompass numerous families distributed in all bacterial lineages. Nevertheless, very few of these systems have been experimentally characterized except the Cdi and the Rhs families in Gram-negative bacteria [6], [8] and the PF04740 family from Gram-positive bacteria [48]. CdiA and Rhs toxins are the best characterized and can be found in many species including E. coli, Yersinia pestis, Dickeya dadantii or Burkholderia pseudomallei [8]. CdiA and Rhs toxins are large filamentous proteins (over 1000 amino acids) with multiple repetitive elements. In contrast, MafB toxins are predicted to have a globular structure. Organization of the loci encoding CdiA and Rhs toxins shares similarities with MGIs organization. In particular, the presence of genes encoding CT cassettes/immunity modules downstream of the toxin gene is a common feature [8]. Interestingly, mafB genes are restricted to the genus Neisseria, which is very unusual in polymorphic toxin systems [1]. A mafA gene, encoding a surface exposed outer membrane lipoprotein, is frequently found immediately upstream of mafB. mafA is also specific of the Neisseria genus. It has been shown that gonococcal MafA binds to glycolipids but its biological function remains elusive [35]. Unlike CDI two-partner secretion system in which CdiB mediates the secretion of CdiA [49], the mafA gene located 5′ of mafB shares no homology with known secretion systems. Furthermore, the sequence of MafA offers no other clues to its function.
As observed for maf genes, there are several loci bearing haemagglutinin related genes in meningococcal genomes. Haemagglutinin related genes are termed tpsA and tpsB genes. tpsA encodes a secreted filamentous haemagglutinin and tpsB encodes its dedicated transporter. Of note, haemagglutinin related genes are only present as pseudogenes in gonococcal genomes [50]. Analysis of the sequences of the TpsA proteins encoded in N. meningitidis genomes revealed the presence of three distinct groups [51]. Group 1 is found in all meningococcal genomes whereas tpsA genes of group 2 and 3 are overrepresented in disease isolates compared to carriage isolates [51]. In meningococcus, the number of tpsA genes range from 1 to 5 [52]. Genomes of strains FAM18, B16B6 and Z2491 contain only one tpsA gene of group 1 system while strain MC58 contains five tpsA genes (NMB0493, NMB0497, NMB1214, NMB1768 and NMB1779) belonging to the three different groups [51].
The tps locus of meningococcal strain FAM18 is referred as a cdi locus in the comparative genomic study conducted by Poole et al. in 2011 [4]. Indeed, the FAM18 tpsA gene (NMC0444 also termed cdiA by Poole et al.) encodes a protein that exhibits a filamentous haemagglutinin family N-terminal domain and a Pre-toxin domain with a VENN motif, which is located before the putative C-terminal toxin domain. Since this study, several recent works have been published on the meningococcal tps loci. It has been shown that meningococcal strain B16B6 had an advantage in competition assay against a B16B6 deletion mutant lacking the cognate immunity gene of tpsA [14]. The tpsA gene of strain B16B6 (HQ420265) encodes a protein with >99% sequence identity with that of FAM18 [14] confirming the prediction of Poole et al. [4] that meningococcal TpsA proteins from group 1 constitute functional CDI systems. It remains to be investigated whether TpsA proteins from group 2 and 3 are also able to mediate growth inhibition.
The six secretion pathways identified in Gram-negative bacteria can be classified in two categories, either the pathways transporting proteins in a single-step across both inner and outer membranes (i.e. type I, III, IV and VI) or the two-step secretion pathways (i.e. type II and V), where proteins are first targeted to a machinery that recognizes their N-terminal signal peptide [53]–[55]. Type V secretion systems encompass auto-transporter proteins (type Va) and two-partner secretion systems (type Vb) [54]. Cdi toxins belong to the type Vb subclass. Only type I, Va and Vb secretion pathways are found in N. meningitidis, whereas only type IV and Va secretion pathways are found in N. gonorrhoeae [56]. Thus, the only common secretion pathway shared by pathogenic Neisseria species is autotransporters pathway. Autotransporters are single polypeptides that consist of a surface-exposed variable N-terminal domain (“passenger domain”) that can be released from the cell surface by a proteolytic cleavage and a C-terminal domain (“translocator domain”) folded into a β-barrel structure in the outer membrane. The β-barrel of translocator domains is in most cases composed of 14 β-strands [57]. The IgA protease of N. gonorrhoeae was the first described autotransporter [58]. Since MafB toxins possess an N-terminal signal peptide and are found in N. meningitidis and N. gonorrhoeae, they are likely to be secreted by a two-step secretion pathway found in both species and not yet identified. Indeed, according to their domains organization and the lack of β-barrel structure prediction (using TBBpred server [59] and HMM-TM server [60]), MafB toxins do not belong to the autotransporters family. Besides, secreted proteins known to be targeted to a TpsB transporter of two-partners systems exhibit a TPS secretion domain adjacent to the signal peptide sequence [61]. There is no such TPS domain in the N-terminal region of MafBs that could target the toxin to a TpsB transporter. Thus, in contrast to Cdi toxins, MafB toxins are unlikely to be secreted via type V secretion systems. Moreover, in contrast to Rhs toxins that can be secreted through type VI secretion system, there is no type VI secretion system in pathogenic Neisseria species.
In addition to these six types of secretion systems, Gram-negative bacteria, including Neisseria pathogenic species, constitutively produce outer membrane vesicles (OMV) during their normal growth [62], [63]. OMVs are mainly composed of outer membrane and periplasmic components. OMVs enable the secretion of virulence factors to the surrounding environment or directly to neighboring bacteria [64], [65] or to eukaryotic cells [66], [67]. For instance, OMVs derived from P. aeruginosa is able to kill other bacterial species [64] by the release of murein hydrolases capable of degrading the peptidoglycan of other species. P. aeruginosa is also able to deliver multiple virulence factors directly into the host cell cytoplasm by fusion of OMV with host cell membrane lipid raft [67]. Thus, OMVs can interact both with competing bacteria and with host cell to promote bacterial colonization of the host or pathogenesis. Pathogenic Neisseria are well known for their release of OMVs [56], [62]. Natural and engineered OMVs have recently gained interest for use as vaccine or adjuvants. The OMV vaccine strategy has been successfully used during several clonal outbreaks of serogroup B meningococcal strains in Cuba, Norway, and New Zealand [68]. As a consequence, several recent proteomic studies analyzed the Neisseria OMVs content. A proteomic study of naturally released OMVs isolated from four gonococcal strains (FA1090, F62, MS11, and 1291) revealed the presence of MafA and MafBMGI-2 (corresponding to ORFs NGO0225, NGNG_00563, NGFG_00362 and NGAG_00430) in OMVs of 4 strains [43]. In addition, supplemental data published by Zielke et al. in the same proteomic study suggest that four other MafB toxins produced by these gonococcal strains could also be present in OMVs. Since MafA has been previously described as an adhesin able to bind cellular glycolipids [35], the presence of MafA in OMVs could mediate attachment of OMVs to eukaryotic cells. This suggests that OMV-mediated release could be a mean for delivery of MafB toxins to neighboring bacteria or to eukaryotic cells. The presence of MafB toxins in OMVs and its potential implications for Neisseria pathogenesis have to be further explored.
A hallmark of MGIs is the presence of numerous mafB-CT genes. Genetic recombination, resulting from the replacement of the 5′ end of a full-length mafB gene by an alternative CT cassette is supported by genome comparison both for MGI encoding class 1 or class 3 MafBs. For example in MGI-1s, the mafB-CT cassette found in MGI-1NEM8013 (NVM_0408) encodes the same CT region that mafBMGI-1FA1090 (Fig. 2A). In MGI-3s, the mafB-CT cassette found in MGI-3NEM8013 (NVM_2314) encodes the same CT region that mafBMGI-3FA1090 (Fig. 2C). Genetic recombination of polymorphic toxins has been recently demonstrated for CdiA in N. meningitidis [14] and for Rhs in Salmonella enterica serovar Typhimurium [10].
In this study, we demonstrated a functional role for one of the four MafB toxins of NEM8013. Indeed, we showed that an unencapsulated derivative of NEM8013 was outcompeted by a strain overexpressing MafB1MGI-2NEM8013. Several studies have evidenced that meningococci frequently become acapsulated in the nasopharynx as a result of phase variation [69], [70] or by down-regulation of the genes involved in capsule biosynthesis [71], [72]. Thus, both capsulated and unencapsulated strains are likely to compete with each other for colonization of the nasopharyngeal niche. The function of the three other mafB genes of this strain remains unknown. Given the diversity of the CT extremities of MafBs, it is also plausible that they exhibit various biological targets such as other bacterial species or eukaryotic cells. Thus, biological function of maf genes remains a challenging question. Since these genes represent 2% of the genome of pathogenic Neisseria, but are virtually absent from non-pathogenic species, it is likely that they play important biological roles, including in pathogenesis. Regulation of the expression of maf genes could give clues on their physiological role. It has been recently demonstrated by Fagnocchi et colleagues that NadR is a regulator of maf operons [73]. The NadR regulon in meningococcal MC58 strain comprises nadA (encoding an adhesin), and the operons NMB0375-374 (encoding MafA and MafB in MGI-1MC58) and NMB0652-654 (encoding MafA, MafB and MafI in MGI-2MC58). In the presence of human saliva (or the small metabolite 4HPA that is secreted in human saliva) Fagnocchi et colleagues showed that maf genes are repressed, while nadA is induced in a NadR-dependent manner. This coordinate regulation indicates an adaptation of the bacteria in response to the signal molecules present in saliva, and suggests a role in colonization of the port-of-entry.
It has been suggested that, in addition to growth-inhibiting function, Rhs and Cdi might have a broader role in interbacterial communication. CdiA- or Rhs-CTs could serve as signal molecules when translocated in neighboring cells protected by the cognate immunity protein, in a manner similar to quorum sensing. Furthermore, recent studies showed that CDI plays a role in biofilm formation in Burkholderia thailandensis [13], [74], [75]. Indeed, CDI systems might be a mechanism allowing bacteria to discern kin versus non-kin within a complex population (e.g. in a polymicrobial biofilm). Only the cells expressing the same set of toxins and of immunity proteins will be able to live in close proximity. However, it is difficult to predict the evolution of a complex community since a strain may contain numerous toxin systems (such as Maf, Rhs, Cdi, bacteriocins or recently described type VI secretion systems effectors) that can all play a role in interbacterial competition. Deciphering interactions between these systems is a challenging question for future studies.
We used 6 fully sequenced and annotated genomes of N. meningitidis (Z2491, MC58, FAM18 and NEM8013) and N. gonorrhoeae (FA1090 and NCCP11945) present in the MicroScope database with curated annotations from the NeMeSys project [19], [76]–[80]. We also used N. gonorrhoeae MS11 genome sequence available from the Broad Institute, N. cinerea ATCC 14685 genome sequenced by Washington University and the following meningococcal strains: H44/76 [81], M04-240196 [82], M01-240355 [82], G2136 [82], M6190 [82], NZ-05/33 [82], WUE2594 [83] and 053442 [20]. We used BLASTp with default parameters to search for orthologs of the Maf proteins in other Neisserial strains present in the NCBI non-redundant protein database.
To identify the different classes of MafB proteins, we downloaded 150 sequences stored on the PFAM database server [84] that have been used to build the DUF1020 (PF06255) family. The alignment was performed with ClustalW (http://www.genome.jp/tools/clustalw/) [85], [86] using default parameters and a rooted phylogenic tree (UPGMA) was generated.
Multiple alignments were also performed using MUSCLE [87] or Clustal Omega [88] with default parameters and shaded using the BoxShade server (http://www.ch.embnet.org/software/BOX_form.html). The LipoP 1.0 server [40] and SignalP server 4.1 [36] was used to predict the presence of a signal sequence with default options for Gram-negative bacteria. PRED-TAT [37], TatP 1.0 [38] or TATFIND 1.4 [39] servers were used to identify putative tat (twin arginine translocation) signal. β-barrel structure prediction were performed with TBBpred server [59] and HMM-TM server [60] and conserved domains were identified with the Conserved Domain Database (CDD) [44] of the NCBI server. Pairwise genome comparisons were visualized using Easyfig [89].
Accession numbers of proteins mentioned in the text with corresponding locus tags are listed in the supplementary information (S1 Table).
All strains used in this study can be found in S2 Table.
Meningococci NEM8013 and N. cinerea were grown at 37°C in a moist atmosphere containing 5% CO2 on GCB (« Gonococcal Broth »; Difco) agar plates containing Kellog's supplements and appropriate antibiotics (100 µg/ml kanamycin, 6 µg/ml chloramphenicol and/or 4.5 µg/ml erythromycin for NEM8013 or 9 µg/ml erythromycin for N. cinerea). E. coli TOP10 (Life technologies) or BL21(DE3) (Life technologies) were grown at 37°C in liquid or solid Luria-Bertani (LB) medium (Difco), which contained appropriate antibiotics (50 µg/ml ticarcillin, 10 µg/ml chloramphenicol and/or 50 µg/ml kanamycin).
All vectors and primers used in this study can be found in S3 Table and S4 Table.
Full-length mafB genes including the putative signal peptide sequence or only the 5′ or 3′ regions of mafB genes from NEM8013 were cloned in pBAD33. Briefly, PCR products of mafB1MGI-2NEM8013, mafBMGI-1NEM8013, the 5′ region of mafBMGI-1NEM8013 (the first 1020 nucleotides), the 3′ region of mafBMGI-1NEM8013 (the last 477 nucleotides) and the 3′ region of mafBMGI-3NEM8013 (the last 339 nucleotides) were digested by SacI and XbaI, whereas PCR products of mafB2MGI-2NEM8013, mafBMGI-3NEM8013 and the 5′ region of mafBMGI-3NEM8013 (the first 1107 nucleotides) were digested by SmaI and XbaI. The digested PCR products were ligated to pBAD33 under the control of arabinose-inducible PBAD promoter. mafB2MGI-2NEM8013 and mafB1MGI-2NEM8013 amplified without their putative signal peptide sequence (without the first 93 nucleotides for mafB2MGI-2NEM8013 and the first 120 nucleotides for mafB1MGI-2NEM8013) were also cloned in pET22 using BamHI and XhoI to obtain proteins containing PelB peptide signal. mafIMGI-1NEM8013 was cloned in pET28 using NcoI and XhoI and mafIMGI-3NEM8013 was cloned in pET15 using XhoI and BamHI. pBAD33-mafB or pET22-mafB constructs were transformed in E. coli BL21(DE3) with or without pET-mafI constructs to perform toxicity assays.
All vectors constructed were verified by PCR and sequencing.
In order to product and purify MafBMGI-1NEM8013, the operon mafBIMGI-1NEM8013 (NMV_0410-NMV_0409) without the sequence of the signal peptide of MafBMGI-1NEM8013 was cloned downstream of an IPTG inducible promoter in pET15 using XhoI and BamHI. The resulting MafB protein harbors hexahistidine N-terminal tag.
In order to product and purify MafIMGI-1NEM8013, NMV_0409 was cloned downstream of an IPTG inducible promoter in pET28 using NcoI and XhoI. The resulting MafI protein harbors hexahistidine C-terminal tag.
In order to assess the presence of MafB in the supernatant of E. coli, the operon mafBIMGI-1NEM8013 (NMV_0410-NMV_0409) was cloned downstream of an IPTG inducible promoter in pET28 using NcoI and XhoI.
To assess the potential co-purification of MafBMGI-1NEM8013 and MafIMGI-1NEM8013, mafBMGI-1NEM8013 and mafIMGI-1NEM8013 were cloned in two different multiple cloning site (MCS) under two IPTG inducible promoters in pcolaDUET. mafBMGI-1NEM8013 was cloned in MCS2 using BglII and KpnI and mafIMGI-1NEM8013 was cloned in MCS1 using BamHI and HindIII.
All vectors constructed were verified by PCR and sequencing.
PCR reactions were used to amplify 400 bp upstream mafA (NEICIv1_50108) and 770 bp downstream mafI (NEICIv1_50110). The resulting products were digested by EcoRI/BamHI and BamHI/HindIII respectively and cloned into pUC19 digested by EcoRI/HindIII. A kanamycin resistance cassette apha-3 was then inserted in the BamHI site. The final construct containing kanamycin resistance cassette apha-3 flanked by homologous regions for recombination was amplified by PCR using primers with DNA uptake sequence. Amplicon was introduced in N. cinerea by transformation and transformants were verified by PCR and sequencing.
mafB genes, mafI genes or mafB-mafI operons of NEM8013 were cloned in pGCC4 under the control of an IPTG inducible promoter using PacI and ScaI. pGCC4-mafB or pGCC4-mafB-mafI were transformed in NEM8013, pGCC4-mafI1MGI-2NEM8013 was transformed in NEM8013ctrA [80] to insert mafB, mafI or mafB-mafI in the meningococcal intergenic region between lctP and aspC. NEM8013 strains harbouring IPTG inducible mafB-mafI operons have been used in competition assays. pGCC4-mafBIMGI-1NEM8013 was also transformed in NEM8013pilQ- [90] to assess the role of PilQ on MafB secretion.
In order to express a FLAG-tagged MafBMGI-1NEM8013 protein in NEM8013, the 5′ region of mafBMGI-1NEM8013, with or without the sequence of its signal peptide, was amplified with a reverse primer encoding the FLAG epitope (DYKDDDDK). The resulting PCR product was cloned in pGCC4 using PacI and ScaI. pGCC4-mafBMGI-1NEM8013FLAG or pGCC4-mafBMGI-1NEM8013FLAG-SP was transformed in NEM8013. pGCC4-mafBMGI-1NEM8013FLAG was also transformed in N. cinerea and in N. cinerea ΔmafABI to assess the role of MafA on MafB secretion.
All strains constructed were verified by PCR and sequencing.
Toxin activity was assessed by growing E. coli BL21(DE3) carrying pBAD33-mafBMGI-1NEM8013, pBAD33-mafB1MGI-2NEM8013, pBAD33-mafB2MGI-2NEM8013 or pBAD33-mafBMGI-3NEM8013 at 37°C on LB agar plates with or without 0.2% L-arabinose to induce gene expression. Growth curves were performed in LB broth at 37°C (200 rpm) and toxin expression was induced either by addition of L-arabinose to a final concentration of 0.2% (pBAD33-mafB) or by 1 mM IPTG (pET22- mafB).
To assess the protective role of mafI, E. coli BL21(DE3) were co-transformed with pBAD33-mafB and either cognate or non-cognate pET-mafI. Growth curves were performed with 0.01 mM IPTG to induce the antitoxin production throughout the experiment whereas L-arabinose (0.2%) was added 2 h post-inoculation to induce the toxin production. Growth was monitored every hour with OD 600 nm. Viability was assessed in parallel of the growth curves by spotting 5 microliters of bacterial cultures onto LB agar plates containing D-glucose (0.2%) before and after the induction of toxin expression.
Toxin MafBMGI-1NEM8013 and cognate immunity protein MafIMGI-1NEM8013 were expressed in E. coli BL21 (DE3) using plasmid pET15 resulting in the production of N-terminal hexahistidine-tagged toxin and untagged immunity protein. Protein expression was induced for 2 h at 37°C with 1 mM IPTG. Proteins were purified using NiNTA metal affinity resin (Qiagen) in denaturing conditions. Bacterial pellets were lysed by sonication in lysis buffer (100 mM NaH2PO4, 10 mM TrisHCl, 8 M urea, pH 8) and centrifuged for 20 mn at 10 000 g. Supernatants were incubated with Ni-NTA resin and loaded onto columns. The resin was washed with denaturing wash buffer (100 mM NaH2PO4, 10 mM TrisHCl, 8 M urea, pH 6.3) to first remove untagged immunity protein and then, MafBMGI-1NEM8013 was eluted using the same buffers by lowering pH (pH 5.9 and pH 4.5). Renaturation of MafBMGI-1NEM8013 was achieved by serial dialysis against buffers containing decreasing concentrations of urea.
Immunity MafIMGI-1NEM8013 was expressed in E. coli BL21(DE3) using plasmid pET28 resulting in the production of a C-terminal hexahistidine-tagged immunity protein. Protein expression was induced for 1 h at 37°C with 1 mM IPTG. Proteins were purified using NiNTA resin in native conditions. Bacterial pellets were lysed by sonication in lysis buffer (50 mM NaH2PO4, 300 mM NaCl, 10 mM imidazole, pH 8) and centrifuged for 20 mn at 10 000 g. Supernatants were loaded on NiNTA resin columns. The resin was washed (50 mM NaH2PO4, 300 mM NaCl, 20 mM and 50 mM imidazole, pH 8) and MafIMGI-1NEM8013 was eluted using a buffer containing 250 mM imidazole (pH 8).
MafBMGI-1NEM8013 and MafIMGI-1NEM8013 were co-expressed in E. coli BL21(DE3) using pcolaDUET. Protein expression was induced for 2 h at 37°C with 1 mM IPTG. Proteins were purified using NiNTA resin in native conditions as described above for MafIMGI-1NEM8013 alone. Bound complexes were eluted with native elution buffer containing 250 mM imidazole (pH 8) for SDS-PAGE analysis and immunoblotting.
E. coli Top10 carrying pBAD33 or pBAD33-mafBMGI-1NEM8013 were grown in LB containing chloramphenicol at 37°C, 200 rpm, until OD600 reached 0.2, then L-Arabinose was added to a final concentration of 0.2%. Total RNA from E. coli was isolated before induction (T0) and 30 min after addition of L-arabinose using TRIzol reagent (Life Technologies) according to manufacturer's instructions. Total RNAs were analyzed by denaturing gel electrophoresis (5% polyacrylamide/8 M urea) and visualized by staining with ethidium bromide.
Purified MafBMGI-1NEM8013-His6 alone (3 µM) and/or MafIMGI-1NEM8013-His6 (100 µM) were incubated with 4 µg of total RNA isolated from different sources with TRIzol reagent method (N. meningitidis NEM8013, E. coli TOP10 and human epithelial cells FaDu). Each reaction was performed for 30 min at 37°C in Tris-EDTA buffer and run on native 1% agarose gels containing ethidium bromide. To assess the role of divalent cations, buffers containing Mg2+ (10 mM Tris-HCl, 2.5 mM MgCl2) or Mn2+ (25 mM HEPES pH 7.4, 50 mM NaCl, 5 mM MnCl2, 1 mM DTT) have been used instead of Tris-EDTA buffer.
To assess the ability of MafBMGI-1NEM8013 to cleave synthetic RNA in vitro, purified MafBMGI-1NEM8013-His6 alone (3 µM) and/or MafIMGI-1NEM8013-His6 (100 µM) were incubated with 3 µM of a synthetic oligoribonucleotide in Tris-EDTA buffer for 15 min at 37°C. Two synthetic oligoribonucleotides (synthetized by Integrated DNA Technologies) were used: 5′-CCUGGUUUUUAAGGAGUGUCGCCAGAGUGCCGCGAAUGAAAAA -3′ (mRNA U) and 5′- CCAGGAAAAAAAGGAGAGACGCCAGAGAGCCGCGAAAGAAAAA-3′ (mRNA 0). The reactions were stopped by the addition of an equal volume of Gel Loading Buffer II (95% formamide, 18 mM EDTA, 0.025% SDS; Ambion) and incubation for 5 min at 95°C. The reaction products were separated by electrophoresis in 14% polyacrylamide/8 M urea and were visualized by ethidium bromide staining.
Preparation of protein samples, SDS-PAGE separation, transfer to membranes and immunoblotting were performed using standard molecular biology techniques. Proteins were quantified using NanoDrop, following manufacturer's instructions.
We raised polyclonal antibodies in rabbits against purified recombinant protein MafIMGI-1NEM8013 and against a synthetic peptide (KNSNIHEKNYGRD) of the COOH-terminal region of MafBMGI-1NEM8013 protein (Proteogenix). We used rabbit polyclonal anti-FLAG antibody directed against DYKDDDDK epitope (Cell Signaling Technology) and mouse monoclonal antibody directed against NADP-dependent glutamate dehydrogenase. Bound primary antibodies were detected by goat Anti-Rabbit or Anti-Mouse HRP-linked antibodies (Cell Signaling Technology) using ECL Plus detection reagents (Pierce).
Overnight cultures of N. meningitidis grown on GCB agar plates were used to inoculate RPMI 1640 medium (PAA) containing Kellog's supplements and 1 mM IPTG. Overnight cultures of BL21(DE3) grown on LB agar plates were used to inoculate LB medium containing appropriate antibiotic. When OD 600 reached 0.2, IPTG was added to a final concentration of 1 mM. When OD 600 reached 0.5, the cells were harvested by centrifugation (3 000× g for 30 min), and supernatants were passed through a 0.22-µm pore size filter unit. Supernatant proteins were concentrated by ultrafiltration (Amicon, Ultra-15, 3 kDa cutoff) according to the manufacturer's instructions. The pellets and the concentrated supernatants were used for immunoblotting analysis.
Overnight cultures of putative target cells and putative inhibitory cells grown on GCB agar plates were used to inoculate Ham's F12 (PAA) containing Kellog's supplements and 1 mM IPTG. Putative target cells were an unencapsulated derivative of NEM8013 with a transposon insertion in the ctrA gene [80] or this unencapsulated derivative overexpressing MafI1MGI-2NEM8013. Putative inhibitory cells were NEM8013 overexpressing each of the four mafB toxins and their cognate mafI immunity genes. When OD 600 reached 0.8, cultures were mixed at an inhibitor to target cell ratio of 10 to 1 (∼8×109 inhibitor cells and ∼8×108 target cells) and centrifuged at 4 000 rpm for 5 min. 10 µl of the mixed culture pellet were spotted on a membrane filter (pore size, 0.45 µm) placed on GCB agar plate containing 1 mM IPTG and incubated overnight. 10 µl of mixed cultures were also used for determination of the inhibitor/target initial ratio by plating on GCB with appropriate antibiotics. Filters recovered after overnight incubation were used to perform viable counts and the competitive index (CI) was calculated as the inhibitor/target ratio in the output divided by the initial inhibitor/target ratio. The data from three independent experiments were examined for significance using a two-tailed Student's t-test. A p-value p<0.05 was considered significant.
|
10.1371/journal.pbio.1000088 | A Cell Cycle Timer for Asymmetric Spindle Positioning | The displacement of the mitotic spindle to one side of a cell is important for many cells to divide unequally. While recent progress has begun to unveil some of the molecular mechanisms of mitotic spindle displacement, far less is known about how spindle displacement is precisely timed. A conserved mitotic progression mechanism is known to time events in dividing cells, although this has never been linked to spindle displacement. This mechanism involves the anaphase-promoting complex (APC), its activator Cdc20/Fizzy, its degradation target cyclin, and cyclin-dependent kinase (CDK). Here we show that these components comprise a previously unrecognized timer for spindle displacement. In the Caenorhabditis elegans zygote, mitotic spindle displacement begins at a precise time, soon after chromosomes congress to the metaphase plate. We found that reducing the function of the proteasome, the APC, or Cdc20/Fizzy delayed spindle displacement. Conversely, inactivating CDK in prometaphase caused the spindle to displace early. The consequence of experimentally unlinking spindle displacement from this timing mechanism was the premature displacement of incompletely assembled components of the mitotic spindle. We conclude that in this system, asymmetric positioning of the mitotic spindle is normally delayed for a short time until the APC inactivates CDK, and that this delay ensures that the spindle does not begin to move until it is fully assembled. To our knowledge, this is the first demonstration that mitotic progression times spindle displacement in the asymmetric division of an animal cell. We speculate that this link between the cell cycle and asymmetric cell division might be evolutionarily conserved, because the mitotic spindle is displaced at a similar stage of mitosis during asymmetric cell divisions in diverse systems.
| Throughout animal development, and in stem cells, many cell divisions are asymmetric. The one-cell-stage C. elegans embryo divides asymmetrically, as a result of a displacement of the mitotic spindle to one side of the cell. As in other cell divisions, a mitotic progression machinery ensures that all chromosomes are associated with the metaphase plate before anaphase begins. This machinery involves the anaphase-promoting complex and its activator Cdc20/Fizzy, which target proteins for destruction by the proteasome; the cyclin that is targeted for degradation by the proteasome; and a cyclin-dependent kinase. We have asked whether the same machinery has a second function, delaying movement of the spindle to an asymmetric position until spindle assembly is complete. To address this question, we used genetic, reverse genetic, and pharmacological techniques to disrupt the function of elements of the mitotic progression machinery. We find that the mitotic progression machinery does indeed time spindle positioning, acting to delay spindle displacement until spindle assembly completes. This demonstrates a previously unrecognized link between the mitotic progression machinery and asymmetric spindle positioning in an animal cell.
| Asymmetric cell divisions often involve the asymmetric segregation of cell fate determinants as well as asymmetry in daughter cell size. Such asymmetry in daughter cell size typically results from the displacement of the mother cell's mitotic spindle to an asymmetric position within the cell before cytokinesis. Asymmetry in size of cells alone is likely to be important to partition determinants precisely [1], to allow large stem cells to divide repeatedly without becoming depleted of cytoplasm [2], and to permit meiosis in oocytes to produce small polar bodies and large eggs [3]. Asymmetric spindle positioning has been recognized for over a century [4,5], yet the mechanisms involved are only beginning to be elucidated [6].
The mitotic spindle of the one-cell stage C. elegans embryo (Figure 1A) is moved to an asymmetric position by an inequality in microtubule pulling forces on the two sides of the spindle: The posterior cortex exerts greater net pulling forces on microtubules than the anterior cortex [7–9]. Several molecules that are critical for such pulling forces have been identified. Pulling forces depend on components of a protein complex that includes cortical G alpha protein, the mitotic spindle component LIN-5, and the posteriorly enriched activators of G protein signaling GPR-1/2. This complex recruits the microtubule motor dynein, which is also essential for strong pulling forces. There are a number of models for how these proteins, and analogous proteins in other model systems, result in stronger pulling forces on one side of a cell [6]. The transient enrichment of GPR-1/2 in the posterior cortex might increase the number of molecular links between dynein-associated microtubules and the cortex in the posterior. The G alpha, GPR-1/2, LIN-5 complex might also locally activate dynein motors, and/or promote handover of microtubules to dynein. Whether dynein functions here as a motor or only as a link to depolymerizing microtubules is not yet known. Other mechanisms may contribute to the inequality of pulling forces, including a PAR-dependent asymmetry in microtubule dynamics [10,11] and the local antagonism of G alpha-GPR-1/2 signaling by the DEP domain protein LET-99 [12]. Forces are temporally modulated during mitosis [9], although no temporal regulators have been reported to date.
We found previously that even before the spindle begins to move asymmetrically, two different kinds of forces hold the spindle in place. Pulling forces exist on the posterior side of the spindle, and these are balanced by a microtubule-based tether on the anterior side early in mitosis [9]. In the absence of the tether, the posteriorly directed pulling forces are sufficiently strong to move the spindle prematurely [9]. Pulling forces increase on both sides of the spindle near the time that spindle displacement begins [7,9]. We speculate, therefore, that a switch may exist to modulate forces and move the spindle at a precise time. Such a switch could be temporally regulated in a number of ways. Spindle displacement might be timed by links to the mitotic cell cycle. For example, the cell cycle machinery might affect the activity or localization of one or more of the cortical proteins discussed above, perhaps through the activity of one of the mitotic cyclin-cyclin-dependent kinase (CDK) complexes. Alternatively, mitotic progression and a progression of events at the cortex might run independently and in parallel. For instance, an intracellular signaling cascade at the cell cortex might affect the activity or localization of specific proteins that regulate spindle positioning. Such signaling could in principle result in active force generators being first activated beyond a threshold at a specific time after a developmental event such as fertilization, pronuclear meeting, or centration of the pronuclei.
We set out to test the hypothesis that mitotic progression regulates spindle displacement in an asymmetric cell division. C. elegans is an ideal system to explore this issue because it allows one to combine disruption of the mitotic progression machinery with microscopic imaging and precise quantification of chromosome and spindle dynamics. However, there has been a longstanding obstacle to studying this problem in C. elegans: Mitotic progression proteins are essential for successful completion of oocyte meiosis, such that disrupting these proteins' functions typically results in meiotic arrest before first mitosis [13–15]. This has precluded a number of straightforward tests of the hypothesis that mitotic progression regulates spindle displacement. Here, we report a set of genetic, reverse genetic, and pharmacological experiments, each designed to circumvent this obstacle for specific steps of a mitotic progression pathway. Our experiments tested whether mitotic progression pathway components are required for timely spindle displacement in embryos by disrupting the functions of individual components as well as multiple components for epistasis analysis. Cells with fluorescently tagged spindle components were used to analyze in detail spindle displacement and mitotic progression, and to analyze the consequences of misregulating the timing of spindle displacement. Our results demonstrate that anaphase-promoting complex (APC)-dependent negative regulation of CDK, long known to function as a mitotic progression timer, serves a second role as a timer for spindle displacement. We show that this timing mechanism ensures that spindle assembly is completed before spindle components begin to move away from the center of the cell.
There is disagreement about precisely when the mitotic spindle begins to shift to an asymmetric position in the C. elegans zygote. Some have stated that this begins at variable times before metaphase [16], during metaphase [17], or in late metaphase or anaphase [16,18–20]. Measurements have demonstrated that the spindle begins to shift well before anaphase begins [9,21].
We wished to determine precisely when the spindle begins to shift relative to the time when chromosome congression is completed, because we have observed in recordings that spindle displacement appears to closely follow completion of congression [9]. We therefore developed a method to quantitatively analyze the degree of chromosome congression and the position of the mitotic spindle simultaneously in individual embryos (Figures 1B–1D and S1). We first generated multiple-plane recordings through entire mitotic spindles in histone H2B:green fluorescent protein (GFP) and gamma-tubulin:GFP-expressing embryos [21] and tracked the movements of the chromosomes and both centrosomes. The degree of chromosome congression was quantified by measuring, at each timepoint, the ratio of histone-GFP fluorescence intensity in a small region at the center of the chromosome mass to histone-GFP fluorescence intensity just outside of this region. We calculated the center of the chromosome mass by a method that makes this ratio especially sensitive to single chromosomes unassociated with an otherwise complete metaphase plate (see Materials and Methods). This ratio peaks in metaphase (Figures 1D and S1). We found that spindle displacement toward the posterior began soon after chromosome congression was completed, early in metaphase (Figure 1B–1E and Video S1). This time is consistent with earlier measurements [9,21], and it additionally identifies the time when chromosome congression completes as the nearest recognizable mitotic event. The period from completion of chromosome congression to the beginning of anaphase chromosome separation lasted an average of 66.9 ± 8.8 (mean ± standard deviation) s. The spindle began to shift early during this period, and at a consistent time from embryo to embryo, starting 10.8 ± 11.3 s after we detected the completion of congression (Figure 1E).
The precise timing of spindle displacement relative to a mitotic event is consistent with our hypothesis that spindle displacement might be regulated by mitotic progression pathways. Mitotic progression depends in part on the degradation of specific proteins by the proteasome at the transition from metaphase to anaphase [22]. To determine directly if precise spindle displacement timing depends on proteasome activity, we disrupted the function of the proteasome by two methods that can circumvent a known requirement for the proteasome in meiosis [23]. First, to rapidly disrupt proteasome activity after meiosis was completed, we introduced a pharmacological inhibitor of the 20S proteasome's peptidase activity, clasto-lactacystin β-lactone (c-LβL) [24], to laser-permeabilized histone H2B:GFP and gamma-tubulin:GFP-expressing embryos. Second, we used rpt-6 RNAi for specific targeting of a proteasome component in histone H2B:GFP and gamma-tubulin:GFP-expressing embryos. RPT-6 is a component of the 19S proteasome subunit, and its disruption has been shown to delay mitotic timing in the early embryo without disrupting meiosis [25]. We confirmed that disrupting proteasome activity by these methods did not prevent spindle formation or chromosome congression in alpha-tubulin:GFP-expressing embryos and in histone H2B:GFP and gamma-tubulin:GFP-expressing embryos (Figure 2A and 2B). As expected, both rpt-6 RNAi and c-LβL caused delays in anaphase onset (Figure 2B and 2C). In support of our hypothesis, we found that both of these treatments also delayed spindle displacement toward the posterior (Figure 2B and 2C; Video S2), suggesting that proteasome function is required for timely spindle displacement. This effect was not reported in our earlier experiments using c-LβL [9], but our previous observations were made by Nomarski imaging of spindles alone, and the timepoints examined earlier would not have revealed a short delay. Because proteasome disruption is likely to have a wide spectrum of direct and indirect effects in cells, we next considered whether components of the mitotic progression machinery with more specific roles regulate the timing of spindle displacement.
The proteasome has a large number of targets, a subset of which are tagged for degradation by the APC, a multi-subunit E3 ubiquitin ligase that regulates the timing of anaphase through specific targets [22]. To test whether the APC temporally regulates spindle displacement, we disrupted the functions of C. elegans homologs of two key components of the APC. Because the APC is required for progression through meiosis in C. elegans, we used methods that can allow meiotic progression and then disrupt mitosis. First, we used a fast-acting temperature-sensitive allele of mat-3, the C. elegans homolog of APC8/CDC23 [14]. We crossed a histone H2B:GFP transgene into mat-3(or180ts) and shifted embryos to the restrictive temperature only after meiosis, just prior to mitosis. This produced a delay in anaphase onset, as expected. We found that this also delayed spindle displacement (Figures 2C and S2). Second, we used carefully timed dsRNA injections into histone H2B:GFP and gamma-tubulin:GFP hermaphrodites to attempt partial depletion of MAT-1, the C. elegans homolog of APC3/CDC27. We found a time window after injection of mat-1 dsRNA when embryos progressed through meiosis successfully, as judged by the presence of only two pronuclei in the embryo before mitosis began, and anaphase onset was delayed in mitosis. We found that spindle displacement was delayed as well (Figures 2C and S2). The length of the intervals between pronuclear meeting and nuclear envelope breakdown (NEBD) and between anaphase and cytokinesis were unaffected, suggesting that the effect we see is not an indirect effect on cell cycle timing more generally (Figure 2, legend). On the basis of these results, and on results below that show suppression of the spindle displacement delay by prematurely inactivating a known APC target, we conclude that the APC is required for timely spindle displacement.
If mitotic progression components serve as a bona fide timer for the onset of spindle displacement, rather than just being required for the efficient execution of spindle displacement, then the converse effect on timing should be possible: Premature inactivation of a critical APC target should result in premature spindle displacement. Cyclin B is an important target of the APC in mitotic progression, and degradation of cyclin B inactivates CDK [26]. Although it has not yet been possible to visualize directly the inactivation of CDK in C. elegans embryonic mitoses, these events are essentially universal in animal cell mitoses, and degradation of GFP-tagged cyclin B has been observed in C. elegans during meiosis [26–28]. We considered a number of methods to alter the timing of CDK inactivation in mitosis. Loss of maternal CDK in C. elegans mutants or by RNAi results in meiotic defects before first mitosis [29], and introducing a nondegradeable form of cyclin B would be expected to do the same. To inhibit CDK activity at specific times, we laser-permeabilized embryos to a highly specific pharmacological inactivator of CDK, the anticancer drug flavopiridol [30,31]. First, we performed two sets of functional tests of flavopiridol's efficacy on C. elegans embryos. (1) We determined whether flavopiridol treatment of alpha-tubulin:GFP embryos before NEBD could result in phenotypes consistent with CDK inactivation. We applied the drug to one-cell stage embryos prior to NEBD by laser-permeabilization of embryos at this stage. Because a cyclin-CDK complex promotes entry into mitosis, inhibition of CDK activity at this early stage should block mitotic entry [26]. Consistent with this, we found that NEBD failed to occur, and most microtubules were found unassociated with centrosomes (Figure 3A). (2) We next treated embryos with flavopiridol after NEBD. Flavopiridol treatment at this time had no apparent effect on microtubules or the mitotic spindle (Figure 3A), suggesting as expected that the dramatic effect of earlier treatment on microtubules was an indirect effect of blocking mitotic entry. Although the time window from NEBD until the normal time of anaphase is short, about 3 min, we found that flavopiridol treatment after NEBD succeeded in causing premature anaphase onset (Figure 3B). On the basis of the effects of flavopiridol on mitotic entry and anaphase timing, we conclude that flavopiridol is likely to be an effective inhibitor of CDK activity in C. elegans embryos, as it is in diverse animal and protozoan systems [30–32].
Flavopiridol treatment after NEBD resulted in anaphase bridges in some embryos, although most embryos succeeded in separating chromosomes completely (10/14 cases). The ability of chromosomes to separate early in many cases suggests that CDK inactivation is likely to promote chromosome separation in C. elegans by regulating separase activity, as in certain other systems [33]. Importantly, we found that flavopiridol treatment also caused the spindle to shift earlier than it would normally do so (Figure 3B and Video S3). We found that slightly earlier flavopiridol treatment, just as NEBD began, produced similar but more dramatic results, including earlier anaphase onset, more frequent anaphase bridges (8/11 cases), and earlier spindle displacement—about 60–90 s earlier (Figure 3). The length of the interval from NEBD to spindle displacement was less than half as long in these embryos as in untreated embryos (Figure 3A), yet the length of the interval between anaphase and cytokinesis was not shortened, suggesting that the effect we see is not an indirect effect of speeding up development or cell cycle timing more generally (Figure 3, legend). We treated embryos with a different CDK inhibitor, the purine analog olomoucine II [34], and found that this too resulted in earlier spindle displacement and earlier anaphase (Figure 3B). Since we found that targeting the APC can delay spindle displacement, and targeting CDK can produce the converse effect, we conclude that APC-dependent regulation of CDK activity is likely to serve as a bona fide timer for spindle displacement. To further test this hypothesis, we next asked whether a known substrate determinant used by the APC to target cyclin is involved, and we determined whether the delay caused by inactivating the APC depends on active CDK.
A key activator of the APC in mitosis is the tryptophan-aspartic acid repeat protein Cdc20/Fizzy. Cdc20/Fizzy is inhibited by a set of kinetochore proteins, the spindle checkpoint proteins, until all kinetochores are attached to the spindle. Upon checkpoint inactivation, Cdc20/Fizzy is able to bind to and activate the APC, serving as a substrate determinant for the APC to target both cyclin B and securin for degradation [35]. We were interested in determining whether the same substrate determinant is relevant to the APC's function in timing spindle displacement.
We considered several approaches for testing the roles of Cdc20/Fizzy and its regulators, the spindle checkpoint proteins, in timely spindle displacement. Experiments that compromise spindle assembly have been valuable for studying regulation of anaphase timing in various systems [36], but cannot be used to study spindle displacement because experimentally compromised spindles are known to move aberrantly under early-acting cortical forces [9]. We therefore considered targeting spindle checkpoint genes in embryos with intact spindles. The spindle checkpoint was first defined by the yeast (mitotic arrest deficient) MAD genes [37] and (budding uninhibited by benomyl) BUB genes [38]. Both sets of genes were named for their ability to bypass mitotic arrest caused by the microtubule-destabilizing drug benomyl, and neither of the screens that identified these genes required them to have essential functions in normal yeast. Indeed, null alleles of some MAD and BUB genes grow at wild-type rates [37–39]. This has indicated that some of the core functions performed by the checkpoint proteins are nonessential functions, required only when spindle assembly is compromised. For this reason, experiments targeting checkpoint proteins alone, in the absence of spindle damage, sometimes do not reveal the roles of checkpoint proteins in regulating anaphase timing.
We found that loss of MDF-2/Mad2 or MDF-3/Mad3 can delay spindle displacement (Figure S3A). However, it appears unlikely that the delays we observed can be explained solely by these proteins' well-known roles upstream of the APC and CDK, since by flavopiridol treatment of an mdf-3 mutant, much of the delay appears to be CDK-independent (Figure S3B), and because anaphase timing was not similarly affected in the absence of spindle damage (Figure S3A) [40]. We predicted, therefore, that the effect we observed on the timing of spindle displacement might reflect a dominant and potentially nonphysiological result of releasing active FZY-1/Cdc20 at a time when it would never normally be active. Indeed, we found that a gain-of-function allele of fzy-1 that behaves genetically as a constitutively active allele [41] produced the same results, delaying spindle displacement in a manner that appeared partially CDK-independent by flavopiridol treatment of the gain-of-function fzy-1 mutant (Figure S3). There is precedent for Cdc20/Fizzy functioning independently of the APC in other systems, although the mechanism by which it does so is not understood by us or by others [42].
Given these results, we decided to pursue the function of FZY-1 by disrupting its function, rather than by gain of function or by disrupting its negative regulators. Because the C. elegans Cdc20/Fizzy protein FZY-1 is required for meiotic progression before mitosis, we attempted to bypass its requirement in meiosis by partial depletion of FZY-1 using timed dsRNA injections. At 10 h postinjection, we found that fzy-1 dsRNA succeeded in consistently and significantly delaying anaphase onset at first mitosis without causing meiotic arrest or failure (12/12 embryos). This treatment also significantly delayed spindle displacement (Figures 2C and S2). Together with results above, this suggests that FZY-1, the APC, and proteasome activity are required for timely spindle displacement. We next considered whether these components function in timing spindle displacement by the same pathway they rely on in timing anaphase.
Our data suggest that negative regulation of CDK by the APC times spindle displacement. However, the APC targets other proteins for degradation in addition to cyclin B [26]. To determine whether the proteasome and the APC time spindle displacement primarily through negative regulation of CDK, we determined whether flavopiridol treatment could rescue the delay caused by disrupting the proteasome and the APC. First, we used c-LβL to disrupt proteasome function in one-cell stage embryos as before, and we then added flavopiridol after NEBD. Flavopiridol rescued most of the c-LβL-induced anaphase delay, and we found that flavopiridol also rescued most of the spindle displacement delay (Figure 4A). These results suggest that although c-LβL might have off-target effects, and although proteasome disruption should affect many processes, the effects of this drug on delaying anaphase and spindle displacement are primarily CDK-dependent. Second, we disrupted mat-1 by RNAi as before and treated these embryos with flavopiridol. We found that the delays in both anaphase onset and spindle displacement were completely rescued (Figure 4A). We conclude that the proteasome and the APC time spindle displacement in the one-cell C. elegans embryo primarily through their roles in inactivating CDK. Taken together, our results suggest that the timing of mitotic spindle displacement in this system is regulated by the well-studied pathway involving the APC and proteasome activity acting upstream of, and inhibiting, CDK activity (Figure 4B). Since each experiment that affected the timing of anaphase similarly affected the timing of spindle displacement, but spindle displacement preceded anaphase in each such experiment, we conclude that CDK inactivation has a more rapid effect on spindle displacement than on anaphase. An alternative explanation, however, is that anaphase timing may depend on timely spindle displacement in this system. We test this below.
In certain systems, spindle position is monitored by the cell cycle machinery, with aberrant spindle positioning causing a delay in the cell cycle [43]. For example, spindle misorientation in rat epithelial cells causes a short delay in anaphase [44]. Given this and our results above, we hypothesized that precise anaphase timing might depend on spindle displacement in the zygote. Although spindle displacement occurs at a precise time in the C. elegans zygote, to our knowledge, there has been no direct test of the hypothesis that anaphase may be delayed in the absence of spindle displacement. We tested this by genetically preventing spindle displacement and monitoring anaphase timing. Spindle displacement was prevented by disrupting the function of the G protein regulators GPR-1/2 by RNAi in histone H2B:GFP and gamma-tubulin:GFP-expressing embryos. We similarly disrupted the function of PAR-2, a ring finger protein that functions in polarity establishment and maintenance before spindle displacement [45]. The effectiveness of RNAi was confirmed by examining embryos for a failure in spindle displacement. We found that anaphase occurred in each background at a time that was statistically indistinguishable from that in normal embryos (Figure 5 and Video S4). We conclude that spindle displacement is not required for accurate anaphase timing in C. elegans. The results suggest instead that CDK activity affects spindle displacement and anaphase timing independently (Figure 4B).
To determine the biological function of the mechanism that times spindle displacement, we examined the consequence of unlinking spindle displacement from normal temporal regulation of CDK activity by the APC. Spindle displacement normally begins just seconds after chromosome congression. Therefore, premature spindle displacement in flavopiridol- or olomoucine II-treated embryos should produce a potentially important developmental consequence—shifting the spindle before all chromosomes become properly attached and aligned at the metaphase plate. Indeed, we found that upon treatment with either drug, as the spindle began to shift, spindles were only partially assembled. In some cases, one or more chromosomes were not yet associated with the metaphase plate as premature spindle displacement occurred (Figure 6A), and we found that even after spindle displacement began, chromosome movement could still be directed away from the metaphase plate, toward one of the spindle poles (Figure 6B). These results suggest that not all chromosomes have received a full complement of force-producing microtubule attachments from both spindle poles by the time of premature spindle displacement. As expected given these results, we found that flavopiridol treatment after NEBD resulted in chromosomes being spread on average over a significantly wider area than normal at the onset of premature spindle displacement (Figure 6C and Videos S5 and S6). In many cases, chromosomes never completed congression (6/8 cases upon treating at NEBD, 0/21 in untreated embryos, p < 0.05). These results suggest that normal temporal regulation of spindle displacement serves to ensure that spindles are fully assembled, with chromosomes aligned at the metaphase plate and with a full complement of force-producing microtubule attachments from each spindle pole, before spindle displacement begins.
We have shown here that the timing of spindle displacement is precisely regulated in the C. elegans zygote. We combined quantitative measurements of events in mitosis with genetic, reverse genetic, and pharmacological techniques to test whether a link exists between the roles of the APC and CDK in mitotic progression and mitotic spindle displacement. Our results demonstrate that negative regulation of CDK activity by the proteasome, FZY-1, and the APC serves as a previously unrecognized timer for mitotic spindle displacement. This regulation of spindle displacement by CDK activity results in an approximately 60–90-s delay in spindle displacement—a delay that may be important, because it allows spindle assembly to complete before cortical forces pull the spindle away from the center of the cell (Figure 6D, model).
How might CDK inactivation impinge on the mechanism of spindle displacement? The mechanism by which mitotic spindles become positioned asymmetrically is a topic of intense interest. In a number of animal systems, proteins involved in spindle displacement have been identified, including a set that acts downstream of general polarity regulators such as the PAR proteins [6]. Myristoylated, membrane-anchored G alpha proteins have been found to associate with G protein regulators, which serve as key links to mitotic spindle proteins—to the spindle protein NuMA in mammals, a NuMA-related protein Mud in flies, and another spindle-associated protein, LIN-5, in C. elegans [12,46–53]. Two key findings—that cortical G protein regulators can be asymmetrically positioned in worms and flies, and that these and LIN-5 associate with dynein complex components—have prompted a number of models for how an asymmetric distribution of G protein regulators might position the spindle in worms and flies [6,20,54]. How these components result in asymmetric forces on the spindle beginning in early metaphase remains a fascinating and incompletely understood issue. Given the results we report here, it will be important to determine if any of the proteins that are required for spindle displacement are temporally regulated by CDK-dependent phosphorylation. GPR-1/2, LIN-5, and LET-99 have distributions that are dynamic through mitosis, and these dynamic distributions might be temporally regulated directly or indirectly by CDK activity, although no such link is yet known [12,47,49,51,55]. Interestingly, work in Xenopus extracts has shown that NuMA is dephosphorylated and released from spindle poles through CDK inactivation [56].
It is also possible that additional players in asymmetric division have yet to be identified, and that one or more of these could be relevant CDK targets. A microtubule motor that functions in the C. elegans zygote, the MKLP kinesin-like protein ZEN-4, is temporally regulated by CDK phosphorylation, although ZEN-4 is not known to play an essential or redundant role in spindle displacement [57]. Separase, an APC- and CDK-regulated protease involved in separating chromosomes, has multiple targets, including some not involved in chromosome separation [58], for example in disengaging duplicated centrioles and in C. elegans cortical granule exocytosis [59,60]. Although there is no current evidence that separase functions in spindle displacement, it is at least conceivable that separase targets might include one or more proteins involved in positioning the mitotic spindle. In yeast, CDK and cyclin B proteins directly or indirectly regulate asymmetries in the distribution of dynein and the cytoskeletal-associated protein Kar9 [61–63]. C. elegans does not have a Kar9 homolog, and dynein distribution appears symmetric in the C. elegans zygote [64], although dynein is present on so many structures in the C. elegans zygote that if a subtle but biologically relevant asymmetry in cortical dynein levels exists, it might be difficult to detect.
It is not yet clear what acts upstream of CDC20/Fizzy to time spindle displacement. Spindle-assembly checkpoint components are reasonable candidates, given that they regulate CDC20/Fizzy in various systems, and given that we have implicated fzy-1 and downstream proteins in spindle displacement at a time just after the spindle-assembly checkpoint is normally satisfied [36]. Early-stage animal embryos have long been said to lack many cell cycle checkpoints [65], but multiple exceptions have been found in which checkpoints can cause delays in early embryonic cell cycles, delays that are typically much shorter than in nonembryonic cell cycles [40,66,67]. Spindle-assembly checkpoint proteins have been shown to function in the early C. elegans embryo by experiments in which spindle integrity was compromised in wild-type and mutant backgrounds [40]. As discussed, it can be difficult to discern the complex roles of spindle-assembly checkpoint proteins in the absence of damage to the spindle itself. We have made several attempts to cause more subtle insults to spindle integrity. Microtubule attachments to kinetochores appear to be important structural components of the spindle in C. elegans, as the spindle is pulled apart early in kinetochore-null mutants [21]. We tried to create situations where only some but not all chromosomes are unattached, by several methods: by partial RNAi depletion of kinetochore components, by introducing extra chromatin fragments by DNA injection, or by introducing only one set of kinetochore-deficient chromosomes through the sperm (EKMC and BG, unpublished data). As might have been expected, none of these methods caused a delay in anaphase without also compromising spindle integrity. In the absence of positive evidence that FZY-1 is responding to the spindle-assembly checkpoint, the possibility that FZY-1 is regulated in another way is a viable alternative. Other possible regulators of FZY-1 include homologs of yeast Emi1, the mammalian centrosomal protein RASSF1A, and the mammalian sperm protein speriolin, all of which can regulate CDC20/Fzy in other systems [68–70]. Despite this limitation in identifying FZY-1 regulators in spindle displacement, the consequence of misregulating spindle displacement timing is clear—spindles can move to an asymmetric position before spindle assembly is completed (Figure 6).
Given our finding that unlinking spindle displacement from the timing mechanism we describe results in displacement of partially formed spindles, it seems plausible that a link between mitotic progression and spindle displacement evolved to ensure that spindle assembly is complete before the spindle shifts. Delaying spindle displacement until spindle assembly completes may have been evolutionarily advantageous because it might help prevent a low frequency of chromosome loss during spindle displacement. Chromosome loss is an important contributor to tumorigenesis [36]. Mutations in several APC components have been implicated in human cancers, and failures of multiple functions of the APC have been suggested as the basis for these mutations in carcinogenesis [71]. We speculate that asymmetric divisions, for example in stem cells, may have an added burden to prevent chromosome loss as the spindle moves to one side of a cell, and that such movements as well as premature anaphase may contribute to chromosome loss in some cancers.
Our finding that the APC and CDK activity time spindle displacement identifies a new link between cell biology and development. Chromosomes first adopt an asymmetric position during M-phase of asymmetric divisions in diverse systems, including Drosophila, C. elegans, annelids, and mammals. Examples include neuroblast divisions in Drosophila [72] and leech [73], sensory organ precursor divisions in Drosophila [74], early embryonic cell divisions in leech [75] and Tubifex [76,77], and meiosis in mouse oocytes [78]. It is perhaps striking that in all of these, chromosomes are first positioned asymmetrically soon after chromosome congression, during metaphase or anaphase [72–78]. It will be interesting to learn whether mitotic progression is an evolutionarily conserved temporal regulator of spindle positioning in asymmetric cell divisions.
Published strains used in this study include the following: TH32 (unc-119(ed3) III; ruIs32[unc-119(+) pie-1::GFP::histoneH2B]; ddIs6 [unc-119(+) pie-1::GFP::TBG-1]), AZ212 (unc-119(ed3) III; ruIs32[unc-119(+) pie-1::GFP::histoneH2B]; ddIs6) and OD3 (ltIs24[pAZ132; pie-1::GFP::TBA-2 + unc-119(+)], a gift from Paul Maddox), cultured at 20 °C. For imaging of the checkpoint and APC alleles, strains of mdf-2(av16), mdf-3(av20), fzy-1(av15), and mat-3(or180) (gifts from Andy Golden) were crossed into TH32 or AZ212. Checkpoint alleles were cultured at 24 °C, and mat-3(or180) was cultured at 15 °C and moved to 25 °C 1 min prior to experiments and recorded at 25 °C.
mat-1 and fzy-1 functions were disrupted by injecting dsRNA as described previously [79], and imaging embryos at multiple time points after injection to identify a time when embryos reached first mitosis without meiotic defects, but had a delay in anaphase timing. rpt-6 function was disrupted by feeding bacteria expressing dsRNA, and par-2 was disrupted by injecting dsRNA as described previously [79,80]. gpr-1/2 were targeted by dsRNA to gpr-2 alone. As the DNA coding sequences for GPR-1 and GPR-2 are nearly identical [47,49,53], this should disrupt the functions of both genes.
To permeabilize embryos for drug treatment, embryos were mounted in a drug on poly-L-lysine-coated and washed coverslips, with clay feet used as spacers, coated in small pieces of charcoal, and sealed with valap (equal parts petroleum jelly, lanolin, and paraffin). Charcoal pieces attached to the eggshell were targeted with a 2-mW pulsed laser (model VSL-337; Laser Science Inc.) containing Coumarin 440 dye in a lasing chamber (Photonic Instruments), to produce small holes in the eggshell. Embryos were treated with the following drugs: 20 μM c-LβL (Calbiochem), 200 μM flavopiridol (NCI) for experiments prior to the entry into mitosis, and 400 μM flavopiridol or 2 mM olomoucine II (Sigma) for experiments during mitosis. As each drug was stored in DMSO, controls were carried out in egg buffer and the appropriate amount of DMSO for each drug. Such high concentrations of drugs were used because laser permeabilization of C. elegans embryos generally results in only a small hole in the eggshell. To treat embryos with flavopiridol or olomoucine II during mitosis, slides were mounted in egg buffer and sealed on only two sides. At a specific time, the drug was added to an unsealed side, while egg buffer was wicked from the other side. For the experiment in which flavopiridol was used to rescue the effects of c-LβL, embryos were permeabilized in c-LβL. During mitosis, a combination of both drugs was washed into the chamber to avoid washing out the c-LβL when washing in the flavopiridol.
Embryos (other than drug-treated embryos) were mounted as described previously [9]. Time-lapse images were acquired using a CSU10 Yokogawa spinning-disk confocal system (McBain) mounted on an inverted microscope (Eclipse TE2000; Nikon). The embryos were illuminated at 488 nm with a 50-mW air-cooled Argon laser (Laser Physics). Digital images were acquired by a 16-bit cooled CCD camera (Orca ER; Hamamatsu), and the acquisition system was controlled by MetaMorph software (Universal Imaging Corporation). For quantifying the duration of events in mitosis, images were acquired with 650-ms exposure at 3-s intervals. Images for multiplane z-series were acquired at 5-s intervals with 400-ms exposure time, in five steps of 1.25 μm each. All images were acquired using 100× Plan Apochromat VC NA1.4 or 60× Plan Apochromat NA1.4 objectives, and 2 × 2 binning in the camera. Images were analyzed using MetaMorph software and Microsoft Excel, and processed in Photoshop (Adobe Systems).
To quantitatively assess the degree of chromosome congression as spindle displacement began, we measured fluorescence intensity, using MetaMorph, from histone H2B:GFP; gamma-tubulin:GFP embryos along the length of a rectangular box running from the anterior to the posterior end of the embryo through the width of the chromatin in the plane of view, and through a projection of the entire spindle in all of the z-planes recorded. Fluorescence intensities were exported to Microsoft Excel, and further analysis was carried out in Microsoft Excel. Chromatin position was identified as the peak position of a 13-pixel-wide running average of fluorescence intensity values (or 5-pixel wide for one timepoint at anaphase to better resolve anaphase separation of chromatin), and two peaks were identified similarly after anaphase. The pixel size used was 0.14 μm. The degree of compactness of the chromatin before, during, and after metaphase is reported as the fluorescence signal ratio at the center of the spindle versus that just outside (Figure S1), obtained by collecting average pixel value along a 13-pixel-wide region at the center of the chromatin position (defined here as the peak value a 31-pixel-wide running average) and average pixel value for two 16-pixel-wide regions on either side of the center region, subtracting from each the background level of fluorescence, defined as the minimum pixel intensity value of a 158-pixel-wide region in the center of the embryo. These region widths were selected to ensure that the values produced were sensitive to individual chromosomes out of the metaphase plates observed in several recordings. To quantify the progress of chromosome congression as in Figure 6, the width of the area in which chromosomes reside in the spindle was calculated as a percentage of the spindle pole-pole distance.
To analyze the timing of NEBD in embryos, we measured the fluorescence intensity of the histone H2B:GFP signal. NEBD was defined as the time when the fluorescence intensity (defined as pixel intensity within a 20 × 20 pixel square positioned in an area of the nucleus free of a chromosome minus a 20 × 20 pixel square of background within the embryo) dropped to 50% the initial measurement. Chromosome congression (Figure 1) was defined as the time when the chromosome mass resided within 15% of the distance between spindle poles. The beginning of spindle displacement was defined as the time when the chromosomes moved to 52% embryo length and did not return past this mark. Anaphase onset was defined as the time when the single chromosome mass first became resolvable as two masses.
Kymographs (Figure 1) were created using Metamorph software, using an 80-pixel-tall line that spanned the embryo's length, calculating average intensities at each time frame.
We used two-tailed t-test p-values to determine significance in all experiments. For experiments represented in Figure 2, the p-values and n-values were the following: For treatments in which the proteasome was disrupted, anaphase onset was delayed in both rpt-6(RNAi) (p = 3.7 × 10−6 compared to wild-type embryos grown at 20 °C) and c-LβL treated embryos (p = 5.1 × 10−4 compared to DMSO controls). Spindle positioning was also delayed in both rpt-6(RNAi) (p = 3.2 × 10−6) and c-LβL treated embryos (p = 0.03). Disruption of the APC delayed both anaphase onset timing (mat-3(or180) p = 1.3 × 10−15; compared to wild-type embryos quickly shifted to 25 °C), (mat-1(RNAi) p = 7.5 × 10−14 compared to wild-type embryos grown at 20 °C), and spindle position timing (mat-3(or180) p = 0.01; mat-1(RNAi) p = 5.5 × 10−5). RNAi targeting fzy-1 delayed both anaphase onset (p = 1.6 × 10−26 compared to wild-type embryos grown at 20 °C) and spindle displacement (p = 6.2 × 10−4).
For experiments represented in Figure 3, the p-values and n-values were the following: After flavopiridol treatment after NEBD, anaphase onset (p = 3.0 × 10−11) and spindle displacement (p = 0.02) occurred earlier than in DMSO-treated wild-type. For flavopiridol treatment at the start of NEBD, anaphase onset and spindle displacement occurred earlier than wild-type (p = 3.8 × 10−6, 9.9 × 10−3, respectively), and also earlier than flavopiridol treatment after NEBD (p = 2.2 × 10−3, 3.5 × 10−3, respectively). For olomoucine II treatment after NEBD, anaphase onset (p = 1.0 × 10−6) and spindle displacement (p = 5.9 × 10−3) occurred earlier than in DMSO-treated wild-type.
For experiments represented in Figure 4, the p-values and n-values were the following: Flavopiridol treatment rescued most of the delay of anaphase onset (p = 1.3 × 10−5) and spindle displacement (p = 9.6 × 10−3) induced by the proteasome inhibitor c-LβL. The delay was not completely rescued compared to flavopiridol treatment alone (p = 4.2 × 10−7 for anaphase; p = 0.023 for spindle displacement). Flavopiridol treatment rescued the delay induced by mat-1(RNAi), for both anaphase onset (p = 1.5 × 10−7) and spindle displacement (p = 1.3 × 10−3). The rescue timing was not statistically distinguishable from flavopiridol treatment alone (p = 0.078 for anaphase onset; p = 0.60 for spindle positioning).
For chromosome congression measurement at the time of spindle displacement in Figure 6, metaphase plates in embryos treated with flavopiridol after NEBD (n = 7) were not as compact as WT embryos (n = 21) (p = 7.0 × 10−3).
|
10.1371/journal.pbio.1001440 | The Highwire Ubiquitin Ligase Promotes Axonal Degeneration by Tuning Levels of Nmnat Protein | Axonal degeneration is a hallmark of many neuropathies, neurodegenerative diseases, and injuries. Here, using a Drosophila injury model, we have identified a highly conserved E3 ubiquitin ligase, Highwire (Hiw), as an important regulator of axonal and synaptic degeneration. Mutations in hiw strongly inhibit Wallerian degeneration in multiple neuron types and developmental stages. This new phenotype is mediated by a new downstream target of Hiw: the NAD+ biosynthetic enzyme nicotinamide mononucleotide adenyltransferase (Nmnat), which acts in parallel to a previously known target of Hiw, the Wallenda dileucine zipper kinase (Wnd/DLK) MAPKKK. Hiw promotes a rapid disappearance of Nmnat protein in the distal stump after injury. An increased level of Nmnat protein in hiw mutants is both required and sufficient to inhibit degeneration. Ectopically expressed mouse Nmnat2 is also subject to regulation by Hiw in distal axons and synapses. These findings implicate an important role for endogenous Nmnat and its regulation, via a conserved mechanism, in the initiation of axonal degeneration. Through independent regulation of Wnd/DLK, whose function is required for proximal axons to regenerate, Hiw plays a central role in coordinating both regenerative and degenerative responses to axonal injury.
| Axons degenerate after injury and during neurodegenerative diseases, but we are still searching for the cellular mechanism responsible for this degeneration. Here, using a nerve crush injury assay in the fruit fly Drosophila, we have identified a role for a conserved molecule named Highwire (Hiw) in the initiation of axonal degeneration. Hiw is an E3 ubiquitin ligase thought to regulate the levels of specific downstream proteins by targeting their destruction. We show that Hiw promotes axonal degeneration by regulating two independent downstream targets: the Wallenda (Wnd) kinase, and the NAD+ biosynthetic enzyme nicotinamide mononucleotide adenyltransferase (Nmnat). Interestingly, Nmnat has previously been implicated in a protective role in neurons. Our findings indicate that Nmnat protein is down-regulated in axons by Hiw and that this regulation plays a critical role in the degeneration of axons and synapses. The other target, the Wnd kinase, was previously known for its role in promoting new axonal growth after injury. We propose that Hiw coordinates multiple responses to regenerate damaged neuronal circuits after injury: degeneration of the distal axon via Nmnat, and new growth of the proximal axon via Wnd.
| Axon degeneration can be induced by a variety of insults, including injury. When an axon is transected from the cell body, the distal axon “stump” degenerates through a regulated self-destruction process called Wallerian degeneration [1]. This process appears to be actively regulated in axons; however, the endogenous cellular machinery that regulates and executes this degeneration process is poorly understood.
Previous studies have implicated a role for the ubiquitin proteasome system (UPS) in Wallerian degeneration, since inhibition of UPS leads to a delay in the early stages of degeneration [2],[3]. One explanation for this result is that the UPS mediates bulk protein degradation via the combined action of many ubiquitin ligases. However an alternative model is that one or several specific E3 ligases target the destruction of key inhibitors of the degeneration process. Here, using an in vivo assay for Wallerian degeneration in Drosophila, we identify an essential role for a specific E3 ubiquitin ligase in promoting Wallerian degeneration.
The ligase, known as Highwire (Hiw) in Drosophila, Phr1 in mice, is well known from studies in multiple model organisms for its conserved functions in regulating axonal and synaptic morphology during development [4]–[12]. We found that mutations in hiw strongly inhibit the initiation of Wallerian degeneration in multiple neuronal types and developmental stages. Until recently [13],[14], such a strong loss-of-function phenotype has not been reported for this process.
Mutations in hiw also inhibit synaptic retraction caused by cytoskeletal mutations [15]. However the finding that Hiw promotes axonal degeneration was originally perplexing, since a known target of Hiw, the Wallenda (Wnd) MAP kinase kinase kinase (also known as dileucine zipper kinase [DLK]) [16],[17], was found to promote Wallerian degeneration in mouse DRG and Drosophila olfactory neurons [18]. In hiw mutants Wnd levels are increased [9],[16],[17], however degeneration is inhibited. A partial explanation for these opposing results is that Wnd plays a protective role in some neuronal types [19],[20]. However this alone could not account for the essential role of Hiw in Wallerian degeneration of all neuron types. These findings pointed to the existence of additional targets for Hiw.
Recent studies in vertebrate cultured neurons have suggested the NAD+ synthase enzyme nicotinamide mononucleotide adenyltransferase 2 (Nmnat2) as an attractive target of post-translational regulation in axons [21]. Nmnat2 is transported in axons, where it has a short protein half-life, and neurons depleted for Nmnat2 undergo axonal degeneration [21]. Moreover, many gain-of-function studies suggest that increasing the activity of an Nmnat enzyme in axons can effectively delay Wallerian degeneration [22],[23]. The most classic example of this comes from studies of the Wallerian degeneration Slow (WldS) gain-of-function mutation in the Nmnat1 locus, which causes a greater than 10-fold delay in the degeneration of injured axons [24]. However, despite the plethora of studies examining the effect of overexpressing Nmnat enzymes [23], very little is known about the role of the endogenous Nmnat enzymes in axons and how their activity may be regulated.
In contrast to the three isoforms in vertebrates, the Drosophila genome contains a single nmnat gene, for which two splice forms are annotated. nmnat is an essential gene, whose depletion in neurons causes neurodegeneration [25]–[27]. Here we find that Hiw and ubiquitination negatively regulate the levels of axonal Nmnat in vivo. Moreover endogenous Nmnat is required, in parallel to Wnd, for mutations in hiw to inhibit degeneration. By down-regulating the levels of Nmnat protein, Hiw promotes the initiation of Wallerian degeneration in axons and synapses. Moreover, through co-regulation of the Wnd/DLK kinase, whose function is required for proximal axons to initiate new axonal growth [28]–[32], Hiw coordinates both regenerative and degenerative responses to axonal injury.
We used a previously described nerve crush assay [20],[30] to study the degeneration of motoneuron and sensory neuron axons within segmental nerves in third instar Drosophila larvae. To quantify the degeneration of motoneuron axons, we used the m12-Gal4 driver to label only a subset of motoneurons with UAS-mCD8-GFP (Figure 1A, 1B, and Materials and Methods). In wild-type (WT) animals, these axons are completely fragmented within 24 h after injury (Figure 1A) [20].
Hiw is a large, highly conserved protein thought to function as an E3 ubiquitin ligase [17],[33]. Previous studies have suggested that Hiw regulates the ability of axons to regenerate after injury [28],[30]. Here we investigated whether Hiw plays a role in degeneration after injury.
In both hiw null (hiwΔN) and hypomorph (hiwND8) mutant animals, axonal degeneration was strongly inhibited. Even 48 h after injury (which is the latest time that can be visualized before pupation) the distal stump of injured axons remained intact in hiw mutants (Figure 1A and 1B). The protection from degeneration was also recapitulated in neurons that expressed the dominant negative mutation, hiw-ΔRING (Figure 1B), but not in adjacent neurons that did not express Gal4. These results strongly suggest that Hiw performs a cell-autonomous function in promoting axonal degeneration after injury. Similarly, we found that overexpression of the de-ubiquitinating enzyme UBP2 [34] delayed degeneration of Drosophila motoneuron axons and neuromuscular junctions (NMJs) (Figure 1B and 1D).
The hiw mutation also inhibited degeneration of the NMJ (Figure 1C). In wild-type animals, pre-synaptic proteins, such as the MAP1B homologue Futsch, disappeared completely from all NMJ boutons within 24 h after injury while the axonal membrane, detected with anti-HRP antibodies, fragmented into individual spheres (Figure 1C). In hiw mutants, all markers of NMJ structure remained intact (Figures 1C, 1D, and S1). Expression of hiw cDNA in motoneurons restored their ability to degenerate after injury (Figure 1D).
To test whether the distal stump of hiw mutants remained functional, NMJ synapses at muscle 6 were subjected to a standard electrophysiology recording paradigm either before or after injury (Figure 1E–1H). At 24 h after injury, wild-type NMJs were completely silent: no evoked excitatory junction potentials (EJPs) were observed (Figure 1H), and only one single spontaneous miniature event (mEJP) was observed in all ten recordings (Figure 1F). In contrast, at 24 h after injury, recordings in hiw mutant NMJs showed robust spontaneous mEJPs and evoked EJPs, resembling uninjured hiw NMJs [8]. Hence axons and synapses are functionally intact and resilient to degeneration in hiw mutants.
We then tested whether Hiw promotes axonal degeneration in other neuron types (Figure 2). The sensory neuron axons in larval segmental nerves were also injured in the nerve crush assay, and their distal axons also degenerated in a Hiw-dependent manner (Figure 2A). We then tested the role of Hiw in degeneration of adult neurons, which can be studied over a longer window of time. In wild-type animals, the distal stumps of olfactory neuron axons in the antennal lobe degenerated within 1 d after their cell bodies were removed by antennal lobe transection [2],[35]. In contrast, in hiw null mutants, olfactory neuron axons remained in the antennal lobe even 20 d after cell body removal (Figure 2B and 2C), which is comparable with the extent of protection by the WldS gain-of-function mutation [2],[35]. These dramatic phenotypes in multiple neuron types suggest that Hiw plays a fundamental role in the initiation of axonal degeneration after injury.
To understand the mechanism for Hiw in Wallerian degeneration we first considered a previously characterized target of Hiw regulation, the Wnd/DLK kinase. A previous study in mouse DRG and Drosophila olfactory neurons found that degeneration is delayed in wnd(dlk) mutants [18]. However, in larval motorneurons, we found the opposite result, since mutations in hiw lead to increased levels of Wnd kinase in axons [16], and overexpression of wnd in motoneuron axons can delay Wallerian degeneration [20]. Consistent with Wnd playing a protective role against degeneration downstream of Hiw, the protection from degeneration in hiw mutants was suppressed in hiw; wnd double mutants, although the suppression was only partial (Figure 3). In contrast, the synaptic overgrowth and overbranching phenotype in hiw mutants was completely suppressed in the hiw;wnd double mutants [16]. We also noticed that while hiw mutations inhibited degeneration in multiple neuron types, overexpression of wnd did not protect olfactory neuron and sensory neuron axons from degeneration [20]. Hence the degeneration phenotype for hiw mutants could not be accounted for by Wnd alone. This suggested the existence of additional downstream effectors of Hiw during axonal degeneration.
A well-known and intensively studied negative regulator of Wallerian degeneration is Nmnat [23]. An increased activity of this enzyme, first discovered in the WldS mutation, can strongly inhibit degeneration after injury [36]. This gain-of-function phenotype for nmnat bears a striking resemblance to the hiw loss-of-function phenotype in its ability to delay the onset of Wallerian degeneration.
There is only one nmnat gene in Drosophila and it has been shown to be required for neural integrity [25]–[27]. To disrupt expression of this essential gene post-embryonically, we used the Gal4/UAS system to express double-stranded RNA [37] targeting nmnat, (UAS-nmnat-RNAi), in neurons. Immunostaining with an anti-Nmnat antibody [25] indicated that the knockdown of Nmnat was effective (Figure S2A); however, it was unlikely to be complete, since neuronal clones that are homozygous mutant for Nmnat undergo spontaneous degeneration in uninjured animals [25],[26]. In contrast, RNAi-mediated knockdown of nmnat in larva motoneurons did not affect the development or stability of axons and synapses (Figure S2B), and only modestly affected the time course of degeneration after injury (Figure 4B). However knockdown of nmnat strongly suppressed the hiw protective phenotype, both in axons (Figure 4A and 4B) and NMJ synapses (Figure 4C and 4D). Similarly, reduction of Nmnat also suppressed the protection from degeneration caused by overexpression of UBP2 (Figure S3). These results suggest that Nmnat function is an important component of Hiw's role in the degeneration process. Interestingly the NMJ synaptic overgrowth phenotype of the hiw mutants was not suppressed by RNAi knockdown of nmnat (Figure 4C and 4E). This implies that Hiw regulates synaptic morphology independently of Nmnat function, or at least through a mechanism that is less sensitive to Nmnat function than degeneration. In contrast, Wnd is required for synaptic overgrowth in hiw mutants, and data presented below suggest that Nmnat and Wnd function independently.
To further probe the relationship between Wnd and Nmnat, we conducted genetic epistasis analysis. Overexpression (O/E) of either wnd or nmnat cDNA can delay Wallerian degeneration in Drosophila motoneurons (Figure 5A–5D), so we tested whether the phenotype of O/E nmnat required wnd, and vice versa, whether the phenotype of O/E wnd required nmnat.
We found that disruption of wnd function had no effect upon the protection from degeneration by O/E nmnat (Figure 5A and 5B). For the converse experiment, we tested whether knockdown of nmnat by expression of UAS-nmnant-RNAi affected the protection by O/E wnd (Figure 5C and 5D). While this method for disrupting Nmnat suppressed the hiw degeneration phenotype (Figure 4), it had no effect upon the O/E wnd phenotype (Figure 5C and 5D). These observations suggest that Nmnat and Wnd protect axons from degeneration through independent mechanisms.
We then tested whether knockdown of nmnat and wnd by RNA interference had additive effects in suppressing the hiw degeneration phenotype (Figure 5E and 5F). Since nmnat-RNAi rescues the hiw phenotype very strongly on its own at 24 h after injury, we assayed earlier time points, 12 and 18 h after injury, for additive effects with wnd-RNAi. Expression of wnd-RNAi alone in the hiw mutant background caused 42% of the NMJs to degenerate (including complete degeneration and partial degeneration) within 18 h of injury, while expression of nmnat-RNAi alone caused 59% of the hiw mutant NMJs to degenerate at this time point. Combined knockdown of both nmnat and wnd led to a nearly complete suppression of the hiw degeneration phenotype, with 92% of the NMJs degenerating (Figure 5E and 5F). Together, these results suggest that Wnd and Nmnat function independently downstream of Hiw in the Wallerian degeneration process (Figure 5G).
Hiw and its homologues are known to function within an E3 ubiquitin ligase complex [17],[33],[38]–[41]. An attractive hypothesis is that Hiw promotes ubiquitination and protein turnover of endogenous Nmnat protein. Consistent with this hypothesis, we found that knockdown of nmnat suppressed the protection from degeneration caused by overexpression of the de-ubiquitinating enzyme UBP2 (Figure S3). We therefore asked whether mutation in hiw leads to an increase in the levels of Nmnat protein. Most strikingly, we noticed an appearance of Nmnat protein in the synapse and neurite-rich neuropil of hiw mutants, which was not detectable in a wild-type background (Figure 6A and 6B). We also observed complex changes in the distribution of Nmnat in neuronal nuclei and glia (Figure S2).
To test whether Hiw regulates Nmnat in neurons via a post-transcriptional mechanism, we drove expression of transgenic HA-tagged nmnat cDNA in neurons via an ectopic Gal4/UAS promoter. In hiw mutants, the total level of HA-Nmnat protein, as detected on Western blots, increased in both larval brains (3.1±0.6-fold) and adult heads (5.2±1.1-fold) (Figure 6C). By immunocytochemistry, the HA-Nmnat protein (which represents a splice form that lacks the nuclear localization sequence) could readily be detected in motoneuron cell bodies (Figure 6D and 6G) and axons within segmental nerves (Figure 6E and 6H), but is barely detectable at NMJ synapses (Figure 6F and 6I). In hiw mutants, the levels of HA-Nmnat increased in all compartments, however the 5-fold increase quantified at NMJ synapses was most striking (Figure 6G–6I). The increase in Nmnat protein levels remained in the hiw;wnd double mutant background (Figure 6E–6I), adding further support to the model that Hiw regulates Nmnat protein independently of Wnd.
The hiw mutation led to an increase in the levels of transgenic Nmnat, which was expressed via the ectopic Gal4/UAS promoter. We confirmed that the hiw mutation did not increase expression from the different Gal4 drivers used (ppk-Gal4, OK6-Gal4, and BG380Gal4, unpublished data). Hence the regulation of Nmnat by Hiw takes place post-transcriptionally. To test whether Nmnat is regulated by ubiquitination, we overexpressed the yeast ubiquitin protease UBP2 in neurons, which can counteract the function of ubiquitin ligases [34],[42]. We found that co-expression of UBP2 in neurons with the HA-nmnat transgene caused an increase in the levels of HA-Nmnat protein (Figure 7A and 7C), resembling the hiw mutant (Figure 6). This suggests that the levels of Drosophila Nmnat are controlled by ubiquitination.
We next tested whether the action of the Hiw E3 ubiquitin ligase is sufficient to modify Nmnat protein level in axons and synapses. Co-overexpression of hiw cDNA (O/E hiw) with HA-nmnat caused a strong decrease in HA-Nmnat protein in motoneuron axons (Figure 7B and 7C). Because Nmnat protein was difficult to detect at the NMJ (Figure 6F), we also examined the nerve terminals of class IV sensory neurons, whose concentrated location in the ventral nerve cord was easier to visualize. O/E hiw caused a reduction in HA-Nmnat protein in sensory axon terminals (Figure 7D and 7E). In contrast, co-expression of the dominant negative hiw-ΔRING mutation caused an increased level of HA-Nmnat in the sensory axon terminals (Figure 7D and 7E). Further evidence that Hiw function is sufficient to down-regulate Nmnat comes from studies in S2R+ cells, which do not express Hiw endogenously. Co-expression of Hiw, but not of Hiw-ΔRING, led to down-regulation of HA-Nmnat protein (Figures 7F and S4A). These findings suggest that Hiw plays a direct role in regulating the levels of Nmnat protein.
Curiously, we were unable to obtain evidence that Hiw down-regulates Nmnat via the UPS. Inhibition of the proteasome by addition of MG132, using several different concentrations and periods of time that affect known targets to the UPS (Materials and Methods) [43],[44], did not affect the down-regulation of Nmnat by Hiw in S2R+ cells (Figure S4A). To inhibit the proteasome in vivo we co-expressed dominant-negative proteasome subunit mutations, DTS5 and DTS7, which in previous studies had been shown to lead to allow targets of the UPS to accumulate [45]–[47]. This led to only minor (7%) changes in the levels of HA-Nmnat in sensory neuron terminals (Figure S4B). Surprisingly, inhibition of the proteasome had a much greater effect upon HA-Nmnat levels in a hiw null mutant than in a wild-type background (Figure S4C). This observation does not favor a simple model that Hiw regulates Nmnat via the UPS. Instead, the data suggest that additional ubiquitin ligases may regulate Nmnat, and that the regulation of Nmnat may be more sensitive to the UPS when hiw is absent.
While the above data indicate that ubiquitination is important for the regulation of Nmnat, the detailed mechanism by which Hiw regulates Nmnat remains to be determined. The mechanism may involve a direct interaction, since co-immunoprecipitation experiments indicate that Nmnat can robustly interact with the enzyme dead Hiw-ΔRING protein in S2R+ cells (Figure 7G).
A recent study using vertebrate cultured neurons suggested that the disappearance of Nmnat2, which has a short half-life, from the distal stump of axons may serve as a trigger for the Wallerian degeneration process [21]. This leads to an attractive hypothesis that Hiw promotes the disappearance of Nmnat protein from the distal stump. Supporting this model, we observed that HA-Nmnat levels become significantly reduced in axons (Figure S5A) and synapses (Figure 8) distal to the injury site. In contrast, HA-Nmnat levels increase in the proximal stump after injury (Figure S5A), consistent with the model that a cytoplasmic form of this enzyme is transported in axons from the cell body [21]. Within 4 h after injury, the majority of HA-Nmnat in sensory axon terminals had disappeared (Figure 8). By comparison, a significant amount of green fluorescent protein (GFP)-Hiw remained at this time point (Figure S5B).
When hiw was mutant, the levels of HA-Nmnat in the distal stump did not decrease significantly below its starting level, even 24 h after injury (Figure 8A and 8B). Expression of UBP2 had a similar effect upon HA-Nmnat in the distal stump after injury (Figure 8A and 8B). These findings indicate that Hiw and the ubiquitination are required for the disappearance of Nmnat protein in the distal stump.
Vertebrates utilize three distinct Nmnat enzymes, which localize to distinct subcellular locations. We tested whether Hiw was capable of influencing the levels of ectopically expressed mouse Nmnat1, which localizes to nuclei, mouse Nmnat2, which co-localizes with golgi and late endosome markers, or mouse Nmnat3, which localizes to mitochondria [48]–[50], by crossing UAS-mNmnat1::myc, UAS-mNmnat2::myc, and UAS-mNmnat3::myc transgenes [51],[52] into the hiw mutant background. Intriguingly, mutations in hiw resulted in increased levels of mNmant2-myc protein within axons and synaptic terminals of class IV sensory neurons (Figure 9). This finding implies that mNmant2-myc protein can be transported to distal axons and synapses, and that mouse Nmnat2 shares a conserved protein feature with Drosophila Nmnat that allows it to be regulated by Hiw. In contrast, loss of hiw had no effect upon the levels of mNmnat1 or mNmnat3. We interpret that the distinct subcellular localization of mNmnat2 may make this protein more susceptible to regulation by Hiw, and that that a conserved mechanism, involving Hiw homologues, may regulate Nmnat2 in vertebrate neurons.
Since the discovery of the dramatic inhibition of degeneration by the WldS mutation, many studies have focused upon the action of the NAD+ biosynthetic enzyme isoforms, Nmnat1, Nmnat2, and Nmnat3, which in some circumstances can confer protection against axonal degeneration (reviewed in [22],[23]). Most of these studies involve gain-of-function overexpression experiments; it has been difficult to address the role of endogenous Nmnat enzymes in this process. Recent observations indicate that endogenous Nmnat activity plays an essential role in neuronal survival, and its depletion leads to neurodegeneration [21],[25]–[27]. In addition, recent studies in vertebrate neurons suggest that the cytoplasmic isoform, Nmnat2, has a short half-life in neurons [21]. An attractive model proposes that Nmnat2 is rapidly turned over in axons, and that its loss in the distal stump of an axon, which has become disconnected from its cell body, leads to the initiation of Wallerian degeneration [21].
Some aspects of this model are supported by our current in vivo characterization in Drosophila. We have identified Hiw, a highly conserved protein with features of an E3 ubiquitin ligase, as an important regulator of Wallerian degeneration. Hiw's role in this process involves the Nmnat protein, whose levels in axons and synapses are regulated post-transcriptionally by Hiw function. In hiw mutants, Wallerian degeneration is strongly inhibited, and the increased level of Nmnat protein in hiw mutants is both required and sufficient to inhibit degeneration.
While the localization of endogenous Hiw in Drosophila is not known, homologues in mice and Caenorhabditis elegans have been detected in axons and at synapses [9],[53], so it is in the appropriate location to target the destruction of Nmnat in distal axons (Figure 8C). However, it remains to be determined whether the down-regulation of Nmnat in the distal stump per se is the trigger for Wallerian degeneration. When HA-Nmnat was overexpressed, axons were protected from degeneration long after the rapid disappearance of detectable protein in the distal stump. It is possible that even very low levels of Nmnat protein are sufficient to protect from degeneration. It is also formally possible that the basal levels of Nmnat before injury, rather than the disappearance of Nmnat after injury, is an important determinant of degeneration. We also acknowledge that axonal degeneration likely involves additional steps downstream or in parallel to the regulation of Nmnat by Hiw. While overexpression of Hiw can induce a reduction in HA-Nmnat levels (Figure 7), we were unable to observe an enhanced rate of degeneration when Hiw was overexpressed.
Studies almost a decade ago suggested a role for the UPS in the initiation of Wallerian degeneration [3]. It is tempting to propose that this role is manifested by the regulation of Nmnat by Hiw. However our observations caution against a simple interpretation that Hiw regulates Nmnat via the UPS, since Hiw can promote disappearance of Nmnat protein in cells in a manner unaffected by proteasome inhibitors (Figure S4A). Moreover, in vivo, inhibition of the proteasome had only a minor effect upon Nmnat levels in a wild-type background (Figure S4B and S4C). However in hiw mutants, Nmnat levels were very sensitive to the function of the proteasome (Figure S4C). We interpret that additional ubiquitin ligases and the UPS may regulate Nmnat independently of Hiw.
Regardless of the role of the proteasome, our observations suggest that ubiquitin plays an important role in Nmnat regulation. Overexpression of the yeast de-ubiquitinating protease UBP2 leads to increased levels of Nmnat protein and inhibition of Wallerian degeneration, in a manner that requires endogenous Nmnat (Figure S3). Future studies of the mechanism by which Hiw regulates Nmnat will therefore consider potential proteasome-independent roles of ubiquitination. Of note, in yeast UBP2 has been shown to preferentially disassemble polyubiquitin chains linked at Lys63 [54], which have been found to perform non-proteolytic functions in DNA repair pathways [55], kinase activation [56], and receptor endocytosis [57],[58]. We should also consider the possibility that Hiw regulates Nmnat indirectly: since we have thus far been unable to detect any ubiquitinated Nmnat species, it is possible that an intermediate, yet unknown, regulator of Nmnat may be the actual substrate of ubiquitination. Nevertheless, co-immunoprecipitation studies from S2R+ cells indicate that Hiw and Nmnat have the capacity to interact (Figure 7G).
The mechanism and cellular location of Nmnat's protective action is a highly debated subject. Observations in the literature point to both NAD+-dependent and NAD+-independent models for the strong protection by the WldS mutation [23]. The location of its protective action may be the mitochondria, since mitochondrially localized Nmnat can protect axons from degeneration [51],[52],[59]. However golgi/endosomal localized Nmnat2 can also be protective [21],[27],[60],[61]. Our findings suggest that mutation of hiw leads to an increase in the pool of endogenous Nmnat that functionally impacts degeneration.
While the site of endogenous Nmnat function during axonal degeneration remains to be identified, we found that the levels of ectopically expressed mouse Nmnat2 were specifically increased in the hiw mutant background. In contrast, the levels of nuclearly localized mNmnat1 or mitochondrially localized mNmnat3 were unaffected by Hiw. Since Nmnat2 has a short half-life in vertebrate neurons [21], it is intriguing to propose that it is regulated by Hiw orthologs via an analogous mechanism.
Since Nmnat2 does not appear to localize to mitochondria, does this favor a non-mitochondrial activity, such as function as a chaperone [62],[63], for the protective action? It remains challenging to determine the exact location of protection, since the most apparent changes in Nmnat protein may not necessarily be the functionally relevant changes.
A previously characterized target of Hiw regulation is the Wnd MAP kinase kinase kinase [16],[17]. This axonal kinase is also capable of inhibiting Wallerian degeneration in motoneurons [20]. The protective action of Wnd requires a downstream signaling cascade and changes in gene expression mediated by the Fos transcription factor [20]. Loss of nmnat does not affect this signaling cascade (unpublished data) nor does it affect the protective action of Wnd (Figure 5C and 5D). Conversely, loss of wnd does not affect the protection caused by overexpressing nmnat (Figure 5A and 5B). Importantly, the regulation of Nmnat by Hiw does not appear to require Wnd function, and Wnd and Nmnat can protect axons independently of each other. These findings favor the model that Wnd and Nmnat are both regulated by Hiw and influence axonal degeneration through independent mechanisms.
The Wnd kinase plays additional roles in neurons, which can be genetically separated from Nmnat function. These include regulation of synaptic growth: a dramatic synaptic overgrowth phenotype in hiw mutants is fully suppressed by mutation of wnd, but is not at all affected by knockdown of nmnat (Figure 4E). Wnd/DLK also promotes axonal sprouting in response to axonal injury [30], which is also unaffected by nmnat knockdown (unpublished data). It is therefore clear that by regulating both Wnd and Nmnat, Hiw regulates multiple independent pathways in neurons.
It is intriguing that the actions of both Wnd and Nmnat promote cellular responses to axonal injury. Axonal regeneration requires an initiation of a growth program within the axon, which depends upon the function of Wnd and its homologues [28]–[32]. Equally important is a clearance of the distal stump to make room for the regenerating axon [64]–[66]. Since both Wnd and Nmnat are transported in axons [21],[30], Figure 8C proposes a model in which Hiw function in the distal axon terminal could simultaneously promote destruction of Nmnat in the distal stump, and accumulation of Wnd in the proximal stump. The latter is observed after injury [30], and is required to promote new axonal growth. The actual location in which Hiw regulates Nmnat remains to be determined. As an upstream regulator of both sprouting in the proximal stump and degeneration of the distal stump, Hiw may play a central role in regulating the ability of a neuron to regenerate its connection after injury.
Importantly, the protective action of Nmnat may not be limited to Wallerian degeneration. The WldS mutation can protect neurons from degeneration in a wide variety of paradigms, from models of neurodegenerative disease, diabetic neuropathy, excitotoxity, and loss of myelination [22],[23]. These findings suggest that action and regulation of Nmnat function is broadly important for neuronal function and maintenance. As a critical regulator of Nmnat, the Hiw ubiquitin ligase and its vertebrate homologues deserve further scrutiny for potential roles in human health and disease.
The following strains were used in this study: Canton-S (wild-type), hiwND8 [8], hiwΔN, UAS-hiw and UAS-hiw-ΔRING from [67], OK6-Gal4 [68], BG380-Gal4 [69] m12-Gal4 (P(GAL4)5053A) [70], ppk-Gal4 [71], Or47b-Gal4 [72], UAS-UBP2 [41], UAS-DTS5, and UAS-DTS7 from [45], wnd1, wnd3, and UAS-wnd from [16]. UAS-HA::nmnat [25], UAS-WldS [2], UAS-mNmnat1::myc, UAS-mNmnat2::myc, and UAS-mNmnat3::myc [51],[52], and UAS-Dcr2 were gifts from Grace Zhai, Liqun Luo, Marc Freeman, and Stephan Thor. UAS-wnd-RNAi (Construct ID 13786) and UAS-nmnat-RNAi (construct ID 32255) were acquired from the Vienna RNAi center [37].
The segmental nerves of third instar larvae were visualized through the cuticle under a standard dissection stereomicroscope. While larvae were anesthetized with CO2 gas, the segmental nerves were pinched tightly through the cuticle for 5 s with Dumostar number 5 forceps. After successful injury, the posterior half of the larva was paralyzed. Larvae were then transferred to a grape plate and kept alive for varying periods of time at 25°C. Also see [30].
Larvae were dissected in PBS and fixed in 4% paraformaldehyde or formaldehyde in PBS for 25 min for the following antibodies used: ms anti-Futsch (1∶100), guinea pig (gp) anti-NMNAT [25], (gift from Hugo Bellen and Grace Zhai, 1∶1,000), rat anti-HA (Roche, 1∶100), rat anti-elav (1∶50), or fixed in Bouin's fixative for 15 min for the following antibodies: ms anti-Brp (1∶200), Rb anti-GluRIII (1∶1,000 [73]), Rb anti-DVLGUT (1∶10,000, [74]). Rat anti-elav (7E8A10) and ms anti-Brp (NC82) were obtained from Developmental Studies Hybridoma Bank, University of Iowa. The conjugated secondary antibodies are used and diluted as follows: Cy3-Gt anti-HRP and Cy5-Gt anti-HRP (from Jackson labs) at 1∶200, A488-Rb anti-GFP (from Molecular Probes) at 1∶1,000. For secondary antibodies Cy3 and Alexa-488 conjugated Goat anti-rb/mouse/rat/gp (from Invitrogen) were used at 1∶1,000. All antibodies were diluted in PBS-0.3%Triton with 5% normal goat serum.
Confocal images were collected at room temperature on an Improvision spinning disk confocal system, consisting of a Yokagawa Nipkow CSU10 scanner, and a Hamamatsu C9100-50 EMCCD camera, mounted on a Zeiss Axio Observer with 25× (0.8 NA) multi and 40× (1.3NA), 63× (1.5NA), and 100× (1.46 NA) oil objectives. Similar settings were used to collect all compared genotypes and conditions. Volocity software (Perkin Elmer) was used for all measurements of average and total intensities.
For measurement of Nmnat intensity in the neuropil, the neuropil area was selected based on co-staining for the synaptic marker Brp. Objects meeting intensity criteria of >0.8 standard deviations above the mean were selected within a 140-µm long region of the ventral nerve cord and then summed for total intensity. The average intensity of the HA-Nmnat staining in muscle 4 NMJs was measured within the synaptic area defined by HRP staining after subtraction of background intensity for each image. The average intensity of the HA-Nmnat staining in motoneuron axons and sensory nerve terminus was measured with a similar protocol. Likewise for neuronal nuclei, the average intensity for Nmnat staining was measured in neuronal nuclei defined by staining for the neuronal marker Elav. Numbers are shown normalized to the average intensity of the control for each figure.
To quantify axonal degeneration, we scored (while blind to genotype) the fragmentation of m12-Gal4, UAS-mCD8-GFP labeled axons within segmental nerves according to one of five categories between 0 and 100% (with 100% meaning completely degenerated) as described in [20]. All measurements indicate the average from >100 axons.
To quantify the degeneration of the NMJ, NMJs were stained for the MAP1B homologue Futsch and axonal membrane marker HRP, and were scored into one of three categories: (1) complete degeneration, defined by a complete loss of Futsch staining from the NMJ and fragmentation of the axonal membrane, (2) partial degeneration, defined by a partial loss of Futsch staining from the NMJ and partial membrane fragmentation, and (3) no degeneration, in which there was no fragmentation of the membrane or Futsch, similar to uninjured control animals. All quantifications shown represent the average scores from multiple NMJs from >six animals quantified in duplicate by two independent observers who were blind to the genotype.
Degeneration of ORN axons was quantified following the previously defined method [2],[35] by calculating the percentage of brains for each genotype in which contralateral axon projections could still be detected.
For all the statistical analysis, Student's t test was applied.
Intracellular recordings were made from muscle 6 in segments A3 and A4 of third-instar male larvae. The larvae were visualized with oblique illumination on an Olympus BX51W1 fixed stage upright microscope with a 10× water immersion objective. Sharp electrodes (15–25 MΩ), made of borosilicate glass (outer diameter 1.2 mm) were filled with 3 M KCl. The signal was amplified with a Geneclamp 500B (Molecular Devices), digitized with a Digidata 1322A interface (Molecular Devices), and stored on a PC with pClamp 10.2 (Molecular Devices). Recordings were performed in HL3 Stewart saline [75] that contained (in mM) 70 NaCl, 5 KCl, 20 MgCl2, 10 HCO3, 5 trehalose, 115 sucrose, 5 HEPES, 1 CaCl2,, the pH was adjusted to 7.2. For all genotypes the resting membrane potentials and input resistances were similar, with average resting potentials of −73±4 and input resistances of 6±0.2 MΩ. To elicit evoked EJPs, the nerve was drawn into a tight-fitting suction electrode and stimulated with brief (1 ms) depolarizing pulses controlled with Digidata interface. The stimulus amplitude was set to 125% of the amplitude necessary to activate the higher threshold of the two excitatory axons that innervate the muscle. For injured wild-type larvae (in which nerve stimulation did not produce evoked synaptic responses) the stimulus amplitude was set to double the amplitude used for un-injured larvae. However evoked responses were not observed, even at the largest stimulus amplitude that the equipment could produce. For analysis of evoked responses, 100 events per cell recorded at 0.2 Hz were measured using the cursor feature in Clampfit 10.2 (Molecular Devices) and then averaged. For analysis of spontaneous miniature EJPs, at least 50 consecutive events were measured per cell using MiniAnal (Synaptosoft). mEJP frequency was calculated from the first 30 s of recording time.
S2R+ cells were cultured in Schneider's medium (Gibco) which contains 10% (v/v) FBS (Gibco) and 1% penicillin-streptomycin (Invitrogen). For plasmid transfection, cells were transfected using FuGENE 6 (Promega) following the manufacturer's instructions. Copper sulfate solution (0.5 mM) was added 6 h after transfection to induce plasmid expression. Cell lysates were collected after 24 h. Plasmids used for transfection were pMT-Gal4 [76], pUAST-eGFP [77], pUAST-GFP-Hiw [67], pUAST-HiwΔRING [67], and pUAST-HA-Nmnat [25].
To inhibit the UPS, cells were treated with MG132 (InSolution, Calbiochem) or DMSO as control using several different conditions: 25 µM for 6 h, 5 µM for 20 h, and 5 µm for 36 h. All of these conditions led to an increase in the levels of polyubiquitinated proteins, detected by Western blots probed with FK1 antibodies.
The following antibodies were used for Western blotting: rb anti-Hiw (ref, 1∶2,000), rat anti-HA (Flourochem, 1∶2,500), ms anti-β-tubulin (1E7) and ms anti-β-catenin (armadillo, N27A1) from Developmental Studies Hybridoma bank (University of Iowa), ms anti-polyubiquitin, (FK1, Enzo Life Sciences, 1∶1,000), and ms anti-ubiquitin (P4D1, Cell Signaling, 1∶1,000). Westerns were probed with IRDye 800CW and 680RD conjugated secondary antibodies (LiCor biosciences, 1∶10,000) and imaged for quantitative analysis via a LiCor Odyssey imaging system.
S2R+ cells were transfected with either pUAST-HiwΔRING or pUAST-HiwΔRING and pUAST-HA-Nmnat. Cells from 6-cm dishes were harvested in 500-µl ice-cold lysis buffer (20 mM HEPES [pH 7.5]), 200 mM KCl, 0.05% Triton X-100, 2.5 mM EDTA, 5 mM DTT, 5% glycerol and Complete proteinase inhibitor [Promega]). 1.5 mg Protein G conjugated Dynabeads (Invitrogen) were used to capture 10 µl mouse monoclonal anti-HA antibody (HA-7 ascites fluid, Sigma) at room temperature for 30 min, and were then incubated with cell lysates for 30 min at room temperature. The immunoprecipitates were then washed three times with ice-cold lysis buffer and subjected to Western blotting analysis.
|
10.1371/journal.pgen.1000567 | Two Chromatin Remodeling Activities Cooperate during Activation of Hormone Responsive Promoters | Steroid hormones regulate gene expression by interaction of their receptors with hormone responsive elements (HREs) and recruitment of kinases, chromatin remodeling complexes, and coregulators to their target promoters. Here we show that in breast cancer cells the BAF, but not the closely related PBAF complex, is required for progesterone induction of several target genes including MMTV, where it catalyzes localized displacement of histones H2A and H2B and subsequent NF1 binding. PCAF is also needed for induction of progesterone target genes and acetylates histone H3 at K14, an epigenetic mark that interacts with the BAF subunits by anchoring the complex to chromatin. In the absence of PCAF, full loading of target promoters with hormone receptors and BAF is precluded, and induction is compromised. Thus, activation of hormone-responsive promoters requires cooperation of at least two chromatin remodeling activities, BAF and PCAF.
| In order to adapt its gene expression program to the needs of the environment, the cell must access the information stored in the DNA sequence that is tightly packaged into chromatin in the cell nucleus. How the cell manages to do it in a selective maner is still unclear. Here we show that, in breast cancer cells treated with the ovarian hormone progesterone, the hormone receptor recruits to the regulated genes two chromatin remodeling complexes that cooperate in opening the chromatin structure. One of the complexes puts a mark in a chromatin protein that anchors the other complex, enabling full gene activation. The present discovery highlights the importance of the concerted order of events for access to genomic information during activation of gene expression and reveals the intricacies of hormonal gene regulation.
| Regulation of eukaryotic gene expression implies mechanisms that permit transcription factors to gain access to chromatin packaged DNA sequences. The basic unit of chromatin, the nucleosome, consists of an octamer formed of two copies of each of the four core histones (H2A, H2B, H3, and H4) around which 147 bp DNA is wrapped in 1.65 left-handed superhelical turns [1]. Modulation of the structure and dynamics of nucleosomes is an important regulatory mechanism in all DNA-based processes and is catalyzed by chromatin remodeling complexes. Such complexes can either modify histone residues or use the energy of ATP hydrolysis to alter the relationship between histones and DNA [2],[3].
The yeast SWI/SNF complex, the first ATP-dependent chromatin remodeling complex to be identified, is a 2-MDa complex of 11 subunits that regulates gene expression by catalyzing octamer transfer, nucleosome sliding, dinucleosome formation, and H2A/H2B displacement [4]–[6]. RSC, (Remodel the Structure of Chromatin) [7], is a closely related yeast chromatin remodeling complex of 15 subunits, that shares two identical and at least four homologous subunits with the ySWI/SNF complex [8]. There are two human SWI/SNF-like complexes both containing ATPase subunits similar to yeast Swi2/Snf2, hBRM (human Brahma) or BRG1 (Brahma-Related Gene 1), as well as a series of other subunits, some of which differ in various cell types [8]. The hSWI/SNF-α complexes, also called BAF for BRG1/hBRM-Associated Factor, contain either hBRM or BRG1 as ATPase and is orthologue to yeast SWI/SNF. The hSWI/SNF-β complex, also called PBAF (Polybromo-associated BAF) contains only hBRG1 and is orthologue to yeast RSC complex [4],[9]. The BAF and PBAF complexes share many subunits but have also subtype specific subunits: BAF250 and hBRM are only found in BAF, whrereas BAF180 and BAF 200 are only found in PBAF [4],[10]. BAF57 has been reported as a common subunit for BAF and PBAF complexes [4],[9].
Histone acetyltransferases (HATs) and histone deacetylases (HDACs) represent other group of chromatin remodeling complexes that regulate the level of acetylation on the N-terminal tails of core histone proteins and other protein substrates [11],[12]. The HATs are divided into five families, including the GCN5-related N-acetyltransferases (GNATs) with GCN5 and PCAF as the best characterized members; the MYST ('MOZ, Ybf2/Sas3, Sas2 and Tip60)-related HATs; p300/CREB-binding protein (CBP) HATs; the general transcription factor HATs including the TFIID subunit TBP-associated factor-1 (TAF1); and the nuclear hormone-related HATs SRC1 and ACTR (SRC3) [13],[14]. Recombinant PCAF, like full-length GCN5, acetylates either free histones or nucleosomes [15], primarily on lysine -14 of histone H3 [16]. The role of PCAF in transcription has been investigated in multiple studies, and its requirement as a HAT and coactivator has been described in several processes including nuclear receptor mediated events [17],[18], but the precise mechanism of action has not yet been elucidated.
The functional relationship between different chromatin remodeling enzymatic activities is of great interest. A remarkable interdependence has been described during transcriptional activation in S. cerevisiae between the SWI/SNF complex and the histone acetylation complex SAGA [19]. Bromodomains of the RSC complex have been shown to recognize acetylated H3K14 [20]. HAT activity stabilizes SWI/SNF binding to promoter nucleosomes providing a mechanistic basis for the ordered recruitment of these complexes [21]. In mammalian cells, transcriptional activation by nuclear receptors requires multiple cofactors including CBP/p300, SWI/SNF and Mediator. The ordered recruitment of these cofactors to the promoters depends not only on the direct interactions between nuclear receptors and cofactors but also on cofactor-cofactor interaction and on histone modifications [22].
Gene regulation by steroid hormones is mediated by intracellular receptors, which upon hormone binding can activate signal transduction cascades and interact in the nucleus with other transcription factors and/or with dedicated DNA sequences, called hormone responsive elements (HREs) [23]. When ligand-receptor complexes interact with HREs they alter the transcriptional state of the associated genes via the recruitment of chromatin remodeling complexes and other coregulators to the target promoters [24],[25]. The mouse mammary tumor virus (MMTV) long terminal-repeat region encompasses a promoter including within 140 bp five degenerated hormone responsive elements (HREs) and a binding site for nuclear factor 1 (NF1). In chromatin the MMTV promoter is organized into positioned nucleosomes [26], with a nucleosome located over the five HREs and the NF1 binding site [27]. On this promoter nucleosome, the binding site for NF1 is not accessible and only two of the five HREs, the strong palindromic HRE1 and the weak half palindrome HRE4, can be bound by hormone receptors, while the central HREs, in particular the palindromic HRE2 and the half palindrome HRE3, are not accessible for receptor binding [28]. Following hormone induction in vivo all HREs and the binding site for NF1 are occupied simultaneously on the surface of a nucleosome-like structure and a functional synergism is observed between progesterone receptor (PR) and NF1 [27].
Already 5 minutes after addition of hormone to breast cancer cells with an integrated copy of an MMTV-luc reporter, the cytoplasmic signaling cascade Src/Ras/Erk is activated via an interaction of PR with the estrogen receptor, which activates Src [29]. As a consequence of Erk activation, PR is phosphorylated, Msk1 is activated, and a ternary complex PR-Erk-Msk1 is recruited to nucleosome B [30]. Msk1 phosphorylates H3 at serine 10, a modification which is accompanied by acetylation at H3 lysine 14, displacement of HP1γ, and recruitment of RNA polymerase II [30]. Inhibition of Erk or Msk activation blocks H3S10 phosphorylation, H3K14 acetylation, and hormonal induction of MMTV and other progesterone target genes [30].
There have been several reports indicating a role for SWI/SNF, Brg1 and Brm in glucocorticoid regulation of MMTV transcription [31]–[33], but the situation with progesterone is less clear. The MMTV promoter assembled in minichromosomes with positioned nucleosomes [34] is activated by PR in a process involving a NURF-dependent mutual synergism between PR and NF1 [35]. Progesterone treatment of a breast cancer cell line carrying a single integrated copy of a MMTV transgene leads to recruitment of PR, SWI/SNF, and SNF2h-related complexes to the MMTV promoter, accompanied by selective displacement of histones H2A and H2B from nucleosome B [6],[25]. Recently, the acidic N terminus of the Swi3p subunit of yeast SWI/SNF was identified as a novel H2A-H2B-binding domain required for ATP-dependent H2A/H2B dimer displacement [36]. Moreover, a characteristic and sharp DNase I hypersensitive site appears near the symmetry axis of the nucleosome encompassing the HREs [27]. The same hypersensitive site can be induced in the absence of hormone by treatment with low concentrations of histone deacetylases inhibitors, such as sodium butyrate or trichostatin A [37], suggesting that it can be initiated by a moderate increased in histone acetylation. However, the relationship between histone acetylation and ATP-dependent nucleosome remodeling following hormone treatment is not clear. To approach this question we have performed knockdown and ChIP experiments targeting human SWI/SNF complexes and histone acetyl transferases (HATs) in breast cancer cells.
The requirement of Brm and Brg1 for MMTV promoter activation was assessed in T47D-MTVL cells transfected with specific siRNAs. When siRNA against Brg1 was used, the levels of Brm protein increased (Figure 1A, left panels) and viceversa (Figure 1A, middle panels). A significant reduction (80% and 55% for Brm and Brg1, respectively) of both ATPases was only observed when the two siRNAs were combined (Figure 1A, right panels). The levels of the two PR isoforms, PRB and PRA, were unaffected by Brg1 or Brm siRNAs (Figure 1A, third row). In cells transfected with control siRNA an eight-fold increase in MMTV-luc transcription was observed 8 hours after hormone treatment (Figure 1B). In cells transfected with siRNAs against either Brg1 or Brm, the induction was reduced by 30% and 20% respectively, whereas transfection of both siRNA simultaneously reduced the induction of the MMTV promoter by 60%, confirming the essential role of hSWI/SNF on MMTV promoter activation (Figure 1B). Progestin induction of MYC and FOS, two other progesterone target genes, was also impaired by the combined siRNAs to a level similar to that observed with MMTV-luc (Figure 1C). However, not all hormone-dependent genes required SWI/SNF, as the combined siRNAs did not inhibit the progestin induction of the cyclin D1 gene (Figure 1C).
To identify the nature of the Brm/Brg1-containing complex in T47D-MTVL cells, we performed western blotting against subunits of BAF and PBAF, the two distinctive hSWI/SNF complexes. Subunits of both complexes exist in T47D-MTVL cells (Figure S1), indicating that our cell line is capable of forming both BAF and PBAF. The specific subunits that distinguish the two complexes are Brm and BAF250 for BAF, and BAF180 and BAF200 for PBAF. Knocking down BAF57 and BAF250 levels by approximately 50% resulted in a similar reduction in the hormonal induction of the MMTV promoter without changes in Brg1 and Brm protein levels (Figure 1D and 1E and data not shown). Co-immunoprecipitation experiments demonstrated that in T47D-MTVL cells BAF57 forms a complex with the core subunits BAF155, BAF170, BAF 180 and the ATPases Brg1 and Brm (Figure 1F, lane 2 and data not shown). We conclude that BAF and PBAF complexes in the breast cancer cell lines used for these studies contain BAF57.
To further identify the specific complex recruited to the MMTV promoter, we performed ChIP experiments using antibodies recognizing specific components of the BAF and PBAF complexes. Simultaneously with the recruitment of PR to the MMTV promoter after hormone treatment, we observed recruitment of Brg1, Brm, BAF250 and BAF57 (Figure 1G). In contrast, BAF180 and BAF200, two of the specific subunits of PBAF, were not recruited to the MMTV promoter after hormone treatment (Figure 1G, third and fourth rows from the bottom), while they were recruited to the RARβ2 promoter after retinoic acid treatment in U937 cells (Figure S2A). Moreover, siRNA mediated downregulation of BAF180 in T47D-MTVL cells did no influenced hormonal induction of the MMTV promoter (Figure S2B). Thus, BAF complexes containing either one of the two related ATPases together with BAF250 and BAF57 are recruited to the MMTV promoter along with PR and likely mediate promoter transactivation.
BAF57 has been shown to directly interact with the androgen and estrogen receptors [38],[39]. We used co-immunoprecipitation experiments to test whether BAF57 forms a complex with PR in cultured cells. In the absence of hormone, a certain proportion of BAF57 already coprecipitated with PR probably due to the large proportion of PR molecules already present in the nucleus in the uninduced state; however 30 minutes after hormone addition the extent of coprecipitation was increased (Figure 2A, lanes 4 and 5). In contrast, no complex of PR with the PBAF specific subunit, BAF180 was observed independently of the addition of the hormone (Figure 2B). As a positive control for this experiment we used BAF250, a known BAF specific subunit [40]. BAF250 as BAF57 also showed a hormone-dependent interaction with PR (Figure 2B, lanes 2 and 3).
The PR-BAF57 interaction is likely direct, as activated ligand-bound PR interacts with GST-BAF57 in pull-down studies (Figure 2D, lane 2). Deletion mutants lacking either the C-terminal domain or the N-terminal domain, or containing only the central NHRLI domain (Figure 2C, left panel), could not form a complex with PR in GST-pulldown experiments (Figure 2D top row, lanes 3–5).
The fact that BAF57 is present in both BAF and PBAF complexes but only BAF is recruited to the target promoters indicate that either the PR-BAF57 interaction is not sufficient for recruitment of the complex or BAF57 can only interact with PR in the context of the BAF complex but not when integrated in the PBAF complex.
As the central ATPases of the BAF complex, Brg1 and Brm, both contain bromodomains that recognize acetylated lysines [41],[42] and have been shown to bind acetylated histone H3 and H4 tails [43],[44], we wondered about possible roles of histone acetyltransferases (HATs) in hormone induction. Using siRNAs we knocked down PCAF and/or SRC1, which are known to be recruited to the MMTV promoter upon induction [6],[30] (Figure 3A). We do not have an explanation for the enhancement of the unspecific band (marked with an asterisk in Figure 3A, first row, lanes 2 and 4). When siPCAF was used, there was a 50% reduction of the MMTV induction (Figure 3A, right panel) paralleling the decrease in PCAF protein (Figure 3A, left panel), whereas siSRC1 caused only a 15% decrease (Figure 3A, right panel). The combination of the two siRNAs decreased induction by 62% (Figure 3A, right panel), indicating that PCAF is the more relevant HAT in hormonal activation of MMTV. PCAF depletion also decreased induction of other progesterone target genes such as MYC and FOS (Figure 3B), which were shown to depend on BAF.
The need for the catalytic activity of PCAF in regulating MMTV promoter was studied in T47D-MTVL cells transfected with wild type PCAF or an enzymatic deficient mutant (PCAFΔHAT). While transfection of wild type PCAF further increased the level of activity obtained after hormone addition, transfection of PCAFΔHAT did not further stimulate the MMTV promoter (Figure 3C). Although PCAF and PCAFΔHAT are expressed at similar levels (data not shown), PCAFΔHAT does not act as a dominant negative mutant possibly because its levels are not sufficient to compete with endogenous PCAF in transient transfection. Similar results were obtained when MYC and FOS genes were analysed (data not shown). Thus, the catalytic activity of PCAF is required for transcriptional activation of those progestin target genes, which induction depends on BAF. Regarding other acetyl transferases, no decrease in hormone-dependent activation of the MMTV was observed after transfection with specific siRNAs against GCN5 (data not shown).
We next tested the influence of histone acetylation on BAF recruitment to the MMTV promoter. In control cells hormone treatment induced increased acetylation of H3K14 accompanied by recruitment of BAF57 and BAF250, whereas knockdown of PCAF and SRC1 abrogated K14 acetylation and markedly reduced BAF57 and BAF250 loading on the promoter (Figure 3D, compare lanes 1–2 vs 3–4; quantification by real time PCR is shown in the right panel). To test whether H3K14 acetylation and BAF binding take place on the same promoter, we performed sequential ChIP (re-ChIP) experiments in the MMTV, FOS and the MYC promoters. Hormone treatment induced increased acetylation of H3K14 accompanied by recruitment of BAF250 to the corresponding promoters (Figure 3E, lanes 5 and 6), indicating that the H3K14 acetylation and BAF recruitment occur in the same genomic region. In contrast, control re-ChIPs performed with irrelevant IgG as first antibody showed no amplification product (Figure 3E, lanes 3 and 4).
To test whether the association of H3K14 acetylation and BAF is a general phenomenon, we performed co-immunoprecipitation of total hormone-treated chromatin followed by western blot. This seems to be the case, as immunoprecipitation with antibody against acetylated H3K14 co-precipitated BAF155 (Figure 3F) and Brg1 (data not shown).
We next used pull-down experiments with T47D-MTVL nuclear extracts and biotinylated peptides of H3 and H4 tails containing various modifications to test whether H3K14 acetylation influences the binding affinity of the BAF complex for histone tails. Histone H1 was used as loading control (Figure 4A, lower row). We detected binding to histone tails containing H3K14ac of nearly all subunits of the BAF and PBAF complexes, namely Brg1, Brm, BAF170, BAF155, BAF57, BAF180 and SNF5 (Figure 4A, lanes 4, 5, 7, and 8). Similar results were also obtained using HeLa nuclear extracts (data not shown). Moreover, binding of BAF and PBAF complexes to H3 peptides depends solely on H3K14ac and it is not affected by acetylation, phosphorylation or methylation of adjacent residues. Peptides carrying additional modifications, such as K4me3, K9ac, K9me3, or S10p, exhibited similar affinity for BAF subunits as those carrying only K14ac (Figure 4A lanes 5, 7, 8 and data not shown). H3 peptides acetylated at K9 and pan-acetylated (K5, 8, 12, 16) H4 peptides did not exhibit preferential BAF affinity (Figure 4A lanes 3, 10 and 11). Although H4K8 acetylation takes place very early after progesterone addition to T47D-MTVL cells (Figure 4B, upper panel), we found here no binding of the BAF complex to acetylated H4 peptides. The PBAF specific subunit, BAF180 showed the same behaviour as the BAF subunits when incubated with the modified peptides (Figure 4A, third row from the top). It is important to note, that we detected a faint and consistent binding of PR to all tested H3 and H4 peptides without any preference for particular epigenetic modification, indicating that PR is not acting as a bridging factor for the binding of BAF complex to the K14 acetylated peptides (Figure 4A, second row from the bottom). A similar behaviour was observed for PCAF without any preference for the tested modifications (data not shown).
Next, we quantify the interaction of the BAF subunits to H3 acetylated peptides by stable isotope labeling by amino acids in cell culture (SILAC) [45]. Nuclear extracts derived from human HeLaS3 cells grown in light or heavy medium were incubated with immobilized histone peptides in the nonacetylated (H3) and acetylated form, respectively (Figure S3A). We identified BAF180, BAF200, Brg1 and Brm as selective interactors for H3K9ac, H3K14ac and H3K9acK14ac. The SILAC ratio (Heavy/Light, that is modified/unmodified) for Brm was 1.8 for H3K9ac and 7.9 for H3K14ac (Figure 4C and Figure S3), indicating that BAF complex is attached preferentially to K14 acetylated peptides and confirming the results obtained with peptide pull downs (Figure 4A). Higher SILAC ratios with H3K14ac peptide were also obtained for Brg1 and for the PBAF specific subunits, BAF180 and BAF200 (Figure 4C). An additive effect on binding was observed when the double modification H3K9acK14ac was used (Figure 4C). However, we did not find an increase in H3K9ac on the MMTV promoter following hormone induction (Figure 4B, lower panel). These results highlight the role of K14 acetylation as the main modification responsible for BAF and PBAF binding.
Knocking down Brg1 and Brm reduces the interaction of BAF57 and BAF155 with H3K14ac peptides without affecting the intracellular levels of their complex with BAF170 (Figure S4 and data not shown), indicating that the bromodomains of the ATPases contribute to the anchoring of the BAF complex at H314Kac containing sites.
Next we used knockdown experiments to test whether BAF is involved in NF1 binding to the MMTV promoter, a critical step in hormonal induction [35]. The NF1 family of transcription factors in vertebrates is composed of four different genes: NF1A, NF1B, NF1C and NF1X [46], of which NF1C most abundantly expressed in mammary gland and is involved in reciprocal and sequential synergism with hormone receptors [47],[48]. In control cells, NF1C bound to the MMTV promoter in response to hormone, while binding was diminished upon Brg1 and Brm knockdown (Figure 4D, first row, compare lanes 1–2 vs 3–4; quantification by real time PCR is shown in the right panel), suggesting that BAF complexes are necessary for NF-1 to access the promoter in response to hormone.
To investigate how BAF action facilitates NF1 binding we investigated its involvement in the localized displacement of histones H2A and H2B dimers, which is observed after hormone induction [6]. In cells transfected with control siRNA, recruitment of PR 30 minutes after hormone addition (Figure 4D, second row, lanes 1–2) is accompanied by displacement of histone H2A (Figure 4D, third row, lanes 1–2). Depletion of Brm and Brg1 diminished PR binding and abolished H2A displacement (Figure 4D, lanes 3–4; quantification by real time PCR is shown in the right panel). Thus, BAF complexes are required for PR binding and H2A/H2B histone displacement, a requisite for NF1 binding. It is likely that the residual PR bound in the absence of BAF corresponds to receptor interacting with the exposed HRE1 [27].
The results of this study contribute to a better understanding of the molecular mechanisms of promoter activation by progesterone. We have shown previously that PR interacts with an exposed HRE on the surface of a nucleosome on the MMTV promoter [27] and recruits Brg1/Brm-containing complexes [6]. Here we demonstrate the need of the hSWI/SNFα complex, known as BAF, for activation of MMTV and other progesterone target genes. We show that BAF is recruited at least in part via an interaction between PR and the BAF57 subunit of the complex. In a previous report we have described that acetylation of H3K14 in response to progestins [30]. Now we show that H3K14ac, likely generated by PCAF, anchors the BAF complex, suggesting a mechanism for the cooperation between two types of chromatin remodeling activities. The recruited BAF catalyzes the ATP dependent displacement of histones H2A/H2B needed for NF1 to gain access to the promoter site. Thus synergism between the two transcription factors PR and NF1 is mediated by a cooperation between two chromatin remodeling machines, BAF and PCAF.
Activation of several hormone sensitive promoters exhibits SWI/SNF-dependence reflecting the requirement of chromatin remodeling [40], [49]–[51]. There is evidence that glucocorticoid receptor recruits a Brg1-containing complex to promoters via protein-protein interaction with BAF250 [40], whereas androgen receptor (AR) and ERα can directly bind BAF57 [39],[51]. We have described a progesterone dependent recruitment of Brg1 and Brm to the MMTV promoter in breast cancer cells and have shown that yeast SWI/SNF can displace H2A/H2B dimers from MMTV recombinant nucleosomes [6], but the nature of the complex recruited in intact cells and its function in gene activation was not known.
The human SWI/SNF complexes can have either Brm or Brg1 as ATPase subunits [52]. Although the two ATPases exhibit certain differences in their biological activities [53],[54], they can partly compensate for each other in mouse cells [55]. We found that in T47D-MTVL cells depletion of Brm increases Brg1 levels and viceversa. This could be due to the existence of an excess of both free ATPases in equilibrium with the complex bound forms. If one of the ATPase is depleted the other will be incorporated to a higher extent in the SWI/SNF complex thus becoming more resistant to proteolytic degradation. A similar finding has been reported in mice lacking Brm [56].
The BAF complex is recruited to the MMTV promoter within minutes after progestin treatment, likely through a direct interaction with the activated PR. The fact that PR was unable to form a complex with PBAF indicates that the receptor can discriminate between these two related machineries and promotes BAF recruitment to progestin target promoters. Although BAF57 is present in BAF and PBAF complexes it is possible that hormone-dependent interaction with PR is only possible in the context of the BAF complex. Similar results have been previously reported for AR and ERα [38],[39],[51]. Depletion of BAF250 has a similar effect on MMTV induction as depletion of both Brg1 and Brm, confirming the importance of the BAF complex. BAF dependence is observed with other progesterone responsive genes, such as FOS and MYC, indicating that chromatin remodeling by BAF plays a more general role in progesterone gene regulation. However, as exemplified by the cyclin D1 gene, not all hormone responsive genes required BAF for induction, indicating the existence of alternative pathways for hormonal gene regulation. It is likely that other ATP-dependent remodeling complexes participate in regulation of MMTV and other progesterone regulated genes. In minichromosomes reconstituted in Drosophila embryo extracts NURF mediates synergism between recombinant PR and NF1 in a SWI/SNF independent manner [35]. The difference with our present results may reflect the different nature of chromatin in breast cancer cells and Drosophila embryo, such as lack of linker histones and transcriptional activity. We know that hSnf2H is recruited to the MMTV promoter 30 min after progesterone treatment of T47D-MTVL cells but its role in hormonal induction remains to be established [30].
The role of histone acetylation on MMTV activity has been controversial. We have observed that low concentration of HDAC inhibitors leading to moderate increase in acetylation, enhance MMTV transcription [37], whereas high concentrations completely abolish transcription [37],[57]. In previous reports, we have demonstrated the co-recruitment of PCAF and SRC1 to the MMTV promoter following progestin treatment [6],[30]. The siRNA knockdown results presented here demonstrate that PCAF is one of the main HATs acting on the MMTV promoter, that its levels correlated with the extent of H3K14 acetylation, and that acetylation is necessary and important for progesterone activation of the MMTV promoter.
There have also been controversies over the nature of the histone modifications influencing binding of the SWI/SNF complex [58]. It was first claimed that acetylation of H4K8 recruits the SWI/SNF complex via Brg1, whereas acetylation of H3K9 and H3K14 is only important for the recruitment of TFIID general transcription factors [43]. Two independent recent reports demonstrated that the bromodomain-containing ATPase Brg1 has the highest binding affinity to acetylated H3K14 peptide or doubly acetylated H3K9/K14 peptide, whereas binding to acetylated H4K8 peptide is insignificant [59],[60]. Although H4 acetylation takes place very early after progesterone addition to T47D-MTVL cells (Figure 4B), we found here no binding of the BAF complex to acetylated H4 peptides (Figure 4A). In contrast, we show that components of the BAF complex bind to peptides containing acetylated H3K14, alone or in combination with other histone modifications. Furthermore, we quantify the interaction of components of BAF and PBAF to H3 acetylated peptides by SILAC. Our results indicate that compared to H3K9ac, H3K14ac is the major contributor to anchoring of these chromatin-remodeling complexes to nucleosomes. Several subunits of BAF and PBAF contain bromodomains that can bind acetylated residues. Obvious candidates are Brg1 and Brm, but other subunits, such as BAF180, could cooperate on the recognition of H3K14ac by PBAF. Despite the fact that both BAF and PBAF can bind acetylated K14, only BAF is recruited to the MMTV promoter, indicating that H3K14ac participates in anchoring or retention of the recruited complex but is not sufficient for recruitment.
The colocalization of BAF and H3K14ac (Figure 3F) could reflect a general role of this histone mark to BAF anchoring in chromatin, but other explanations are possible. For instance, a common transcription factor could recruit BAF and HATs to the same chromatin regions. Alternatively, components of the BAF complex could be important for HATs recruitment. A combination of these hypotheses is also possible.
Following hormone treatment, PR is phosphorylated, forms a complex with activated Erk and Msk1 and this ternary complex binds the exposed HRE1 on the surface of the MMTV promoter nucleosome leading to modification of the H3 tail and displacement of a repressive complex [30]. PCAF is also recruited to the promoter and acetylation of H3K14 is observed 5 min after hormone treatment [30]. The activated ternary complex pPR-pErk-pMsk1 recruits BAF that initiates nucleosome remodeling, since inhibiting Erk or Msk1 activation blocks BAF recruitment and the subsequent steps. We have not addressed the initial kinetics of acetylation and BAF recruitment in sufficient detail and cannot propose a precise order of events. Moreover, we cannot exclude that PCAF interacts with BAF and that both complexes are recruited together.
In T47D-MTVL cells the BAF mediated chromatin remodeling event is a localized displacement of dimers of H2A/H2B from promoter nucleosome B, a catalytic activity exhibited by the purified yeast SWI/SNF complex on recombinant MMTV nucleosomes [6]. An earlier study with recombinant nucleosomes and SWI/SNF complexes from yeast suggest that both BAF and PBAF complexes would have the capacity to displace histones H2A/H2B as they share the relevant subunit, Swi3p [36]. This remodeling event is a prerequisite for enabling access of NF1 to its binding site in the MMTV promoter following hormone treatment, as demonstrated by the lack of H2A displacement and NF1 binding in BAF-depleted cells. We have previously shown that NF1 binding stabilizes the open conformation of the nucleosome allowing binding of further PR molecules to the internal HREs and subsequent activation of transcription [35]. A model representing our present view of the initial steps in progesterone activation of the MMTV promoter is shown in Figure 5. The model underlines the feed-forward cyclical nature of the activation process and explains why interfering with any of the initial steps has consequences for binding of all factors involved, PR, NF1, BAF and PCAF.
Briefly, after hormone induction activated PR binds first to the exposed HRE1 in a process that does not require chromatin remodeling, as this site is accessible [28]. Along with the activated PR, kinases, BAF and PCAF are recruited to the promoter chromatin. Histone H3 S10 phosphorylation and K14 acetylation promote displacement of an HPIγ-containing repressive complex [30] and BAF anchoring, respectively. Now, BAF can displace histones H2A/B and thus facilitate NF1 binding. Bound NF1 stabilize the open conformation of the remodeled nucleosome exposes the previously hidden HREs allowing binding of further PR-BAF-kinases complexes and PCAF. Binding of PR to these previously hidden HREs is critical for the MMTV activation.
Finally, it is worth noting that we have not mentioned in these studies the possible role of linker histones. Our previous results [61] and those of other groups [62]–[64], suggest that changes histone H1 stoichiometry and phosphorylation by CyclinA/Cdk2 take place at different time points during the hormonal induction and are important for transcriptional activation. Future studies will be required to clarify the relationship of these changes to those reported in this study.
T47D-MTVL breast cancer cells carrying one stably integrated copy of the luciferase reporter gene driven by the MMTV promoter [27] were routinely grown in RPMI 1640 medium supplemented with 10% FBS, 2 mM L-glutamine, 100 U/ml penicillin and 100 µg/ml streptomycin. For the experiments, cells were plated in RPMI medium without phenol red supplemented with 10% dextran-coated charcoal treated FBS (DCC/FBS) and 48 h later medium was replaced by fresh medium without serum. After 24 h in serum-free conditions, cells were incubated with R5020 (10 nM) or vehicle (ethanol) for different times at 37°C.
Cells were cultured into 6-well plates at a density of 4×105 cells/well and treated as indicated above. Transient transfections were performed using Lipofectamine 2000 (Invitrogen). cDNAs expressing PCAF and its deacetylase defective mutant (PCAF ΔHAT) were kindly provided by Tony Kouzarides (Cambridge, UK).
ChIP assays were performed as described [65] using the NF1 specific antibody (gift from Dr Naoko Tanese), the H2A antibody (gift from Stefan Dimitrov), anti-BAF180 (gift from Dr Weidong Wang), anti-PR (H190) and anti-Brg1 (H88), both from Santa Cruz, anti-SMARCA2/BRM (ab15597) from Abcam, anti-BAF57/SMARCE1 from Bethyl and anti-acetyl(Lys14)-Histone H3 and anti-BAF250 from Upstate. Quantification of chromatin immunoprecipitation was performed by real time PCR using Roche Lightcycler (Roche). The fold enrichment of target sequence in the immunoprecipitated (IP) compared to input (Ref) fractions was calculated using the comparative Ct (the number of cycles required to reach a threshold concentration) method with the equation 2Ct(IP)−Ct(Ref). Each of these values were corrected by the human b-globin gene and referred as relative abundance over time zero. Primers sequences are available on request. For re-ChIP assays, immunoprecipitations were sequentially washed as previously described [66]. Complexes were eluted with 10 mM DTT at 37°C for 30 min, diluted 50 times with dilution buffer, and immunoprecipitated with the indicated antibodies. The antibodies used for ChIPs assays are listed in Figure S5.
All siRNAs were transfected into the T47D-MTVL cells using Lipofectamine 2000 (Invitrogen). After 48 h the medium was replaced by fresh medium without serum. After one day in serum-free conditions, cells were incubated with R5020 (10 nM) or vehicle (ethanol) for different times at 37°C. The down-regulation of Brg1, Brm, NF1, BAF57, BAF250, PCAF and SRC1 expression was determined by Western blotting. The siRNAs used are listed in Figure S5.
Total RNA was prepared and cDNA generated as previously described [30]. Quantification of LUC and GAPDH gene products was performed by real time PCR. Each value calculated using the standard curve method was corrected by the human GAPDH and expressed as relative RNA abundance over time zero. Primer sequences are available on request.
Cells were lysed and cell extracts (2 mg) were incubated with protein G/A agarose beads previously coupled with 6 µg of the corresponding antibodies or an unspecific control antibody. The immunoprecipitated proteins (IP) were eluted by boiling in SDS sample buffer. Inputs and IPs were analyzed by western blot using BRG1, BRM, BAF155, BAF170 and H3K14ac specific antibodies.
Nuclear extracts from T47DMTVL breast cancer cells were prepared as described [67]. Peptide pull down assays were performed as described previously [68], with the exception of using 100 µg of nuclear extract during incubation of peptide-bound beads. Synthetic biotinylated H3 and H4 peptides were either purchased from Upstate or were kind gifts from M. Vermeulen. For Western immunoblotting, antibodies against BRG1, BAF170, BAF155, PR (Santa Cruz), BRM (Abcam), BAF57, SNF5 (Bethyl) and H1 (Upstate 05-457) were used.
After trypsin digestion of gel slices peptides were extracted, desalted using stage tips [69] and analyzed using a nano-HPLC Agilent 1100 nanoflow system connected online to an LTQ-Orbitrap mass spectrometer (Thermo Fisher, Bremen). The mass spectrometer was operated in the data-dependent mode to automatically switch between MS and MS2. The instrument was operated with ‘lock mass option’ as recently described [70]. Survey spectra were acquired with 60,000 resolution in the orbitrap, while acquiring up to five tandem mass spectra in the LTQ part of the instrument. The raw data files were analyzed with an in-house developed quantitative proteomics software MaxQuant, version 1.0.12.5 [71], in combination with the Mascot search engine (Matrix Science). The data was searched against a decoy human IPI database 3.37 including common contaminants. False discovery rates, both at the peptide and protein level were set to 1%. Minimal peptide length was set to 6 amino acids. False positive rates for peptides are calculated as described in [72].
|
10.1371/journal.pntd.0006577 | The role of social determinants of health in the risk and prevention of group A streptococcal infection, acute rheumatic fever and rheumatic heart disease: A systematic review | Rheumatic heart disease (RHD) poses a major disease burden among disadvantaged populations globally. It results from acute rheumatic fever (ARF), a complication of Group A Streptococcal (GAS) infection. These conditions are acknowledged as diseases of poverty, however the role of specific social and environmental factors in GAS infection and progression to ARF/RHD is not well understood. The aim of this systematic review was to determine the association between social determinants of health and GAS infection, ARF and RHD, and the effect of interventions targeting these.
We conducted a systematic literature review using PubMed, the Cochrane Library and Embase. Observational and experimental studies that measured: crowding, dwelling characteristics, education, employment, income, nutrition, or socioeconomic status and the relationship with GAS infection, ARF or RHD were included. Findings for each factor were assessed against the Bradford Hill criteria for evidence of causation. Study quality was assessed using a standardised tool.
1,164 publications were identified. 90 met inclusion criteria, comprising 91 individual studies. 49 (50.5%) were poor quality in relation to the specific study question. The proportion of studies reporting significant associations between socioeconomic determinants and risk of GAS infection was 57.1%, and with ARF/RHD was 50%. Crowding was the most assessed factor (14 studies with GAS infection, 36 studies with ARF/RHD) followed by socioeconomic status (6 and 36 respectively). The majority of studies assessing crowding, dwelling characteristics, education and employment status of parents or cases, and nutrition, reported a positive association with risk of GAS infection, ARF or RHD. Crowding and socioeconomic status satisfactorily met the criteria of a causal association. There was substantial heterogeneity across all key study aspects.
The extensive literature examining the role of social determinants in GAS infection, ARF and RHD risk lacks quality. Most were observational, not interventional. Crowding as a cause of GAS infection and ARF/RHD presents a practical target for prevention actions.
| Rates of rheumatic heart disease (RHD) are high in disadvantaged populations globally. It results from acute rheumatic fever (ARF), a complication of Group A Streptococcal (GAS) infection. These are described as diseases of poverty, but exactly what components of poverty promote them has been unclear. The aim of this review was to find what specific social and environmental factors are associated with GAS infection, ARF and RHD, and if actions targeting these can reduce disease rates. We did a search of published literature and found 90 relevant articles. Many supported an association between GAS infection, ARF or RHD and crowding, dwelling characteristics, low education level and employment status, poor nutrition and low social class. There was enough evidence to show that crowding and socioeconomic disadvantage increase the risk of GAS infection and ARF/RHD. However, most studies were of fair to poor quality in their ability to answer the research question, and there was little interventional research. This may relate to challenges inherent in intervening to change social determinants of health, but may also suggest lesser research attention to health issues affecting disadvantaged populations. The association between crowding and disease risk strongly supports initiatives to reduce crowding. This should become a key target for ARF and RHD prevention.
| Rheumatic heart disease (RHD) is an important cause of cardiac morbidity and mortality in disadvantaged populations globally [1–4]. It results from acute rheumatic fever (ARF), which itself occurs as an abnormal immunological response to Group A Streptococcal (GAS) infection of the throat (streptococcal pharyngitis) and possibly streptococcal skin infection[5] in susceptible hosts [6]. GAS pharyngitis is spread through direct person-to-person transmission via saliva or nasal secretions [7]. Generally very few people will develop ARF after GAS exposure, but it may be as high as 5 to 6% in certain groups with greater susceptibility and heavy GAS exposure [8]. Recurrences of ARF cause progressive valvular damage, with between 50 and 75% of cases progressing to RHD [8]. RHD is a chronic and debilitating condition characterised by complications such as arrhythmias and heart failure [6]. ARF and RHD most often affect children and young adults [2]. The true burden of the disease is expected to be far higher than the benchmark estimates, but even at conservative calculations, it is equivalent to approximately one quarter of the global DALY burden of cancer [3].
Globally, ARF and RHD are almost exclusively seen in developing nations or among disadvantaged populations within developed nations [2]. Among populations safeguarded by high standards of living, RHD rates are virtually zero. This dramatic contrast highlights the influence of environmental, economic, social and behavioural conditions on risk of GAS infection and progression to ARF and RHD. Despite the role of social determinants of health in disease genesis, key RHD control programs and guidelines[4, 9, 10] do not specifically address primordial prevention. Primordial prevention aspires to establish and maintain conditions to minimize hazards to health. It consists of actions and measures that inhibit the emergence and establishment of environmental, economic, social and behavioural conditions, cultural patterns of living known to increase the risk of disease[11]–that is, strategies that aim to eliminate exposure to risk factors in the first place. Current strategies focus on primary prevention (early detection and treatment of GAS infections), secondary prevention (delivering intramuscular penicillin every four weeks) and tertiary prevention (medical and surgical management of heart failure) [12]. These cornerstones of RHD control are heavily reliant on health care services being available and accessible, a sufficient level of health literacy and appropriate health seeking behaviour among the general population, and ongoing commitment from cases to receive their injections. However, in settings of poverty or marginalisation, these elements cannot be guaranteed. Furthermore, medical treatment and case management can reduce morbidity and mortality, but they will not change the underlying risk to vulnerable populations [13].
Therefore primordial prevention should be part of a comprehensive strategy to eliminate RHD as a public health problem–a global goal [14]. But while ARF and RHD are generally referred to as ‘diseases of poverty’,[1–4] there is uncertainty about what specific aspects of poverty create the conditions that cause RHD. In order to implement evidence-based preventive strategies at the primordial level, more needs to be understood about the relative contribution of the individual components of poverty, such as household crowding, educational attainment, employment, income, nutrition and overall socioeconomic status to RHD risk. Additionally, evidence for any public health interventions targeting these primordial-level factors needs evaluating.
To provide the evidence base for primordial-level preventative interventions to control RHD, we undertook a systematic review with two aims: to determine the association between social determinants of health and GAS infection, ARF and RHD, and to determine the effect on GAS infection, ARF and RHD of interventions targeting these determinants.
We conducted a systematic literature review on the association between socioeconomic and environmental factors and GAS infection, ARF or RHD, and on the impact of interventions targeting these factors.
The search was conducted between August and October 2016 and eligible articles were identified by searching three databases: PubMed, Cochrane Library and Embase. MeSH terms, key words and Emtree terms (Embase thesaurus headings) searches were conducted. See S1 Table for full details of the search strategy.
All titles and available abstracts were screened by one author (PC). Articles were eligible for inclusion in the analysis of interventions if they were in English, reported on an intervention that encompassed: health promotion, education, or behaviour change targeting social determinants of GAS infection (including impetigo), ARF, or RHD; the provision of hygiene hardware, aids or household infrastructure; or household crowding reduction. Interventions could be at an individual, household, health centre, school or community level. Inclusion criteria for observational studies were that they must have reported on at least one socioeconomic or environmental variable and its relationship with GAS infection including impetigo), ARF or RHD incidence or prevalence, measured objectively at an individual, ecologic or population level with a comparison group (either study controls or use of population data e.g. census). Studies were excluded: if they were not in English; assessed pharmacological interventions only; were of very poor quality; or provided only a subjective appraisal of or no description of methods for ascertaining socioeconomic and environmental factors. No restrictions were set with regards to date of publication. Factors specified in the search strategy included: crowding (household or other settings), income, dwelling characteristics, education level, occupation/employment, social class, and nutrition. These were chosen through a scoping scan of selected literature sources informed by the authors’ prior knowledge in this area. There were no limitations on participant age, setting, geographic location or publication date of the studies. References of selected studies were searched for further articles not covered in the primary search strategy.
The primary outcome was reduction in GAS infection, ARF, or RHD from an intervention targeting a socioeconomic or environmental factor. The secondary outcome was evidence of a causal relationship or association of GAS infection, ARF or RHD with specific social determinants of health.
The full text of the articles were reviewed by the same author (PC) and data were extracted to a template that included information on article details (title, year, first author), study type, study methodology, participants and setting, outcomes, and additional notes including salient points from the discussion or authors' conclusions. Study quality and risk of bias pertaining to measurement of determinants and outcomes of interest to this review were assessed for each study using the National Institute of Health Study Quality Assessment Tool,[15] with the appropriate template used based on the study type (S1 Text). Ecologic studies were judged on additional criteria as described in Dufault and Klar (2011) [16]. Study quality as relevant to this review was rated as: very poor (subsequently excluded), poor, fair, fair to good and good.
Descriptive analysis involved recording the frequency of specific factors measured across the studies and whether a statistically significant association with GAS infection, ARF or RHD was identified, the direction of the association (considered positive if greater degrees of the factor of interest were associated with higher disease rates), mixed, negative, possible or not found. A p value of <0.05 was used to define a statistically significant association. ‘Possible’ designated a reported positive association where a test of significance was not provided, nor means to calculate one. ‘Mixed’ designated a combination of positive, negative or absence of association for different measures of the same variable. Results for GAS infection were presented separately and ARF/RHD to reflect the different epidemiology and natural history of the conditions.
The findings were synthesised in the context of study quality and strengths and limitations of findings. Systematic reviews (n = 4) were reviewed for quality and findings separately.
The evidence of the link between each of the main factors explored and GAS infection, ARF and RHD was assessed against the Bradford Hill criteria, a set of nine guiding principles for interpreting causal relationships between environmental influences and disease. The criteria are: strength of association, consistency, temporality, biological gradient, plausibility, coherence, experiment, specificity and analogy [17].
The studies were too heterogeneous in exposure and outcome measures for meta-analysis.
In some instances where statistical testing was not performed in the original study, but enough data were presented to allow analysis to be undertaken (i.e. a 2 X 2 table with denominators), simple tests of significance (chi-squared, odds ratio (OR) with 95% confidence intervals, or relative risk (RR)) were calculated using the Tables for Epidemiologists function in Stata 13.1 (Statacorp, Texas 2013). These additional analyses are indicated in S2–S8 Tables.
Statistical analysis was conducted to check for correlation of study type, quality of study (fair or greater), outcome of interest (GAS infection, ARF or RHD) and positive outcomes using chi-squared and Fisher’s exact tests of significance. Correlation between year of publication and study quality was analysed using binomial variables of published in previous 20 years (>1996) and study quality equal to or greater than fair. Analysis was carried out using Stata 13.1 (Statacorp, Texas 2013).
This review adheres to the Preferred Reporting Items for Systematic Reviews and Meta-Analyses (PRISMA) checklist (S1 Checklist) [18].
The search strategy identified 1,164 articles in PubMed, 273 articles in Embase and 84 articles in the Cochrane Library (S1 Fig). An additional 21 study articles were recovered from manual reference searches of the selected articles. Of these 114 were selected on the basis of title and abstract. The full text was unavailable for eight articles. The full texts of the remaining articles were reviewed and of these one article was excluded due to being very poor quality as were two studies from an article [19] of three studies were excluded. Hence 90 articles met the inclusion criteria comprising 91 individual studies included in the general analysis, as one article contained two separate studies (Table 1) [20].
Several articles were based on the same study population. Two articles by McDonald et al.(2006) [21] (2008) [22] shared the same participants and a third, a subset of these [5]. The articles authored by Adanja et al.(1988) [23] (1991) [24] and Vlajinac et al. (1989) [25] (1991) [26] utilised the same participants, as did two studies by Zaman(1998a) [27] (1998b) [28]. Four systematic reviews were analysed separately [2, 29–31].
33 studies reported ARF as the outcome, 27 reported RHD, 10 reported both, and 21 reported GAS infection. There were 16 case control studies, [23–28, 32–41] nine case series, [42–50] 12 cohort studies, [5, 21, 51–61] one control trial, [62] 28 cross section studies, [19, 63–89] 22 ecologic studies, [20, 90–105] and three randomised controlled trials (RCT) [106–108]. Studies were conducted in 27 countries with two studies including data from multiple countries.
The quality of the studies in design or ability to answer the research question was rated as poor in 45 (50.5%). Only three studies graded as being of good quality. Studies that were case series were most likely to be of poor quality. Two of the three RCTs, seven of 16 (43.8%) case control, and five of 12 (41.7%) cohort studies were of at least fair quality. Studies published after 1996 (25 of 37) were more likely to be of fair or better quality than those published earlier (6 of 54) (OR 16.7, 95% CI 5.59 to 49.71). Study quality was not associated with outcomes.
12 studies (57.1%) reported a positive relationship between at least one socioeconomic factor and risk of GAS infection, and 35 studies (50.0%) reported a positive relationship between a socioeconomic factor and risk of either ARF or RHD.
The likelihood of positive findings was not associated with publication after 1996 or study quality. 100% of the case control studies reported at least one positive association of a socioeconomic factor with disease rates compared to 64.9% of other studies (Fisher’s exact p = 0.002).
50 studies assessed the association between crowding and GAS infection (14 studies), ARF (16 studies), RHD (15 studies) or a combination (5 studies) (Tables 1 and 2; S2 Table). The most common measures of crowding were: persons per household, room, bedroom or bed; number of children or siblings; dwelling space; and sleeping space per person. 12 of 21(57.1%) studies noted a positive association between crowding and the different outcomes of GAS infection, 9 of 16 (56.3%) with ARF, and9 of 15 with RHD (60.0%) and combinations thereof (60.0%), though only 14 found consistent associations across all measures of crowding (42.9% among studies of GAS infection and 30.6% among ARF/RHD). A further 10 studies reported a possible association between crowding and GAS infection, ARF or RHD risk but did not provide a test of significance nor means to calculate one.
Of the 11 studies that assessed crowding after adjusting for one or more independent variables such as household income, seven noted a residual association between crowding and the outcome of interest. Two case control and three ecologic studies demonstrated a continuous effect gradient between crowding and risk of GAS infection, [38] ARF, [41, 90, 109] and RHD [96].
The remaining 11 (22%) studies that examined an association between crowding and disease rates did not demonstrate any association between crowding and disease risk.
Dwelling characteristics and facilities were assessed across a variety of measures such as general housing condition, construction type, specific characteristics (e.g. dampness, ventilation), or facilities (e.g. water, electricity, toilets) in relationship to GAS infection in seven studies and ARF/RHD in 19 (Tables 1 & 2; S3 Table). Four of seven studies reported positive or mixed associations between dwelling characteristics or facilities with increased GAS infection rates, eight of nine with ARF, two of eight with RHD, and both of the studies that assessed ARF and RHD together. General poor condition or standard of housing was associated with increased risk of ARF or RHD in five of nine studies and type of housing construction or material was associated with GAS infection, ARF or RHD risk in six of nine. Home dampness was associated with ARF in five of seven studies (three of these analysed the same population). There were no clear trends among other specific housing characteristics and facilities (e.g. electricity, kitchen facilities, light, potable water, sewerage, and ventilation) and GAS infection or ARF/RHD.
Three diverse RCTs assessed interventions aimed to induce or reduce acquisition of GAS or its sequelae through common household fomites [106, 108] or hygiene practices [107]. Of these only Luby et al’s (2005) RCT demonstrated an effect on GAS related disease (impetigo). Over 4,600 children were randomised to receive hand washing promotion and antibacterial soap, hand washing promotion and plain soap, or no intervention. The mean impetigo incidence was 36% and 24% lower among the antibacterial and plain soap groups compared to the control group respectively.
The education level of mothers, fathers and cases themselves were explored across 17 studies, with 16 studies reporting ARF/RHD as the outcome (Tables 1 & 2; S4 Table). Low maternal education or literacy was positively associated with ARF incidence in two of three studies, though these two studies shared the same population [23, 25]. Two studies [26, 46] reported results of multivariable analyses and found that maternal education remained associated with ARF after adjusting for other variables. Low maternal literacy was associated with both ARF and RHD in another study [35] and RHD prevalence in one [56] of two studies, [33] but this relationship did not hold in multivariate analysis. There were no clear trends for education levels of fathers, parents combined, or of cases with any of the outcomes.
The relationship between employment and ARF/RHD was explored in 15 studies. No studies examined relationship between employment and risk of GAS infection (Tables 1 & 2; S5 Table). Maternal employment was considered a marker of social or economic disadvantage rather than advantage in four studies. In these studies, maternal employment was positively association with ARF in two of two studies, [23, 41] and with both ARF and RHD in another study, [35] but was not associated with RHD risk in Mirabel et al’s study (2015) [56]. In contrast, a Fijian based case control study measured maternal employment as a marker of socioeconomic advantage and found maternal unemployment was associated with RHD prevalence [33]. There were no associations in any of the five studies that assessed paternal employment [33, 35, 41, 54, 65]. Employment status (unemployment/low class occupation) of the case was found to be associated with RHD in three [40, 61, 76] of four studies [72].
Findings of the three studies that reported GAS infection risk and low income were diverse (Table 1; S6 Table). For ARF/RHD, six of 18 studies reported a positive relationship with low income, while nine demonstrated no association (Table 2). In three studies that undertook multivariate analyses, a positive association was retained in one, [90] but lost in two [36, 40].
The method of assessing income (strata, a binomial measure or comparison of mean income) did not affect the likelihood of reporting an association.
15 studies (covering 12 separate study populations) assessed the relationship between nutrition and ARF or RHD using either dietary intake or anthropometric measures (Tables 1 & 2; S8 Table). Four (26.7%) reported only significant associations between nutritional impairment and ARF or RHD, though tended to only report one simple measure (e.g. low weight or BMI) [26, 74]. A further eight (53.3%) studies demonstrated mixed results.
The relative social position of cases compared to non-cases or to the general population was based on geographical, economic, occupational and social factors, or was based on the ownership of specific assets. Socioeconomic status (SES) or social class was assessed in six studies reporting GAS infection as an outcome, and 36 studies reporting ARF/RHD, with three (50%) and 16 (44.4%) respectively showing a definite association (Tables 1 & 2; S8 Table). A further 10 studies showed a possible association between lower social class with ARF/RHD (i.e. studies that did not provide a statistical test of the apparent relationship between ARF and RHD); nearly all had ecologic designs. 10 studies were able to demonstrate a clear gradient of social class and GAS infection,[20] ARF[87, 90] or RHD[63, 67, 71, 77, 98, 110] risk, with several others suggestive of a gradient.
Four systematic reviews of risk factors for GAS infection, ARF or RHD were identified [2, 29–31]. Only one used a systematic search method for social and environmental risk factors for GAS infection, ARF or RHD [29] and none identified the number of studies found for inclusion in the present review. All four reviews explored SES or poverty in some way. In addition, Kerdemelidis et al (2010) [29] and Steer et al (2002) [30] reviewed crowding, nutrition, and housing factors. The quality of the four studies ranged from poor to fair according to our criteria for measurement of determinants and outcomes of interest to this review, and no review produced firm conclusions.
Each socioeconomic or environmental factor was assessed against the Bradford Hill criteria to establish whether a causal relationship with RHD and its antecedents was supported by the evidence contained in this review (Table 3). Crowding provided a sufficient weight of evidence across the criteria to support a causal relationship, as did socioeconomic status to a lesser extent.
The strengths of the observed relationships of crowding and outcomes were modest; odds ratio or relative risk calculations for binary outcomes (i.e. crowded vs. not crowded whether at a bed, bedroom or housing level) produced around a twofold likelihood or risk of GAS infection, ARF and RHD. In Jaine et al’s ecologic study of ARF cases across New Zealand, after adjusting for average income and number of children aged 5–14 years, the authors noted that a 1% increase in the proportion of households defined as being crowded conferred a 6.5% increase in the expected ARF count at a census area unit level [90].
Regarding consistency, 29 of the 50 studies that assessed crowding reported statistical evidence of an association with GAS infection, ARF or RHD risk. Further studies demonstrated an apparent relationship but did not conduct any significance testing.
The necessary criteria of temporality between cause and effect was met.
While an individual may experience housing instability or change, the characteristics relevant to crowding, e.g. amount of bedrooms a family can afford, is not likely to change greatly during an individual’s childhood [23, 98]. Therefore a point in time capture of data (as in a cross-sectional study) would be likely to represent the living conditions that a case was subject to prior to the development of the condition. As previously described 5 studies demonstrated a gradient in the relationship between crowding and GAS infection, ARF or RHD [38, 41, 90, 96, 109].
The biological mechanism by which crowding exerts its effect on ARF and RHD is via its relationship to GAS acquisition. Crowding fosters intimate contact and consequent GAS transmission directly by human-to-human contact and via droplet spread [123]. Higher GAS infection rates increase the chance that from any one event ARF sequelae will develop. The role of household crowding as the chief driver of GAS transmission was demonstrated in Levine et al’s large cohort study (1966) [38]. Here, GAS infections were not randomly distributed throughout the cohort but clustered within discrete family units as evidenced by high serological concordance between positive family members. The cause and effect interpretation of crowding and GAS acquisition coheres with our knowledge of the natural history and biology of these conditions.
There were no experimental studies assessing crowding and GAS infection, ARF or RHD risk. The closest such evidence comes from findings of an US Air Base Streptococcal Laboratory [50]. This group reported higher acquisition rates of GAS in new army recruits relating to the vicinity of their bed to a known GAS carrier and the amount of carriers within each barracks. In this study, rates of GAS acquisition in new (unexposed) recruits was documented in relation to their distance from the untreated colonised index case, showing a gradient effect; closer proximity was associated with higher risk, giving the study an almost quasi-experimental design. However, this study failed to test whether an intervention against crowding (i.e. actively moving beds further away) was an effective means of decreasing GAS transmission.
Crowding does not exhibit specificity for GAS infection and its sequelae. It is an established health risk for transmissible diseases, especially of those with epidemic potential where outbreaks are more frequent and more severe when the population density is high [124]. An analogy to the droplet transmission of GAS and its predilection among children are offered by the firm observations of the link between crowding and meningococcal infection[111–113] and respiratory syncytial virus [114, 115, 125].
Socioeconomic status also compared favourably against the Bradford Hill criteria, however the causal relationship is exerted through the influence of intermediary factors (e.g. crowding, income, education), which individually do not carry the same weight of evidence. A gradient of higher disease risk with lower socioeconomic status was demonstrated across ten studies.
This systematic review identified 91 studies spanning 80 years that have assessed the relationship of social and environmental factors of crowding, dwelling characteristics and facilities, education, employment, income, nutrition and socioeconomic status with GAS infection, ARF, and RHD. Nearly all studies were observational rather than intervention studies. Crowding was the most frequently assessed factor followed by socioeconomic status. The majority of studies that assessed a measure of crowding and risk of GAS infection, ARF, or RHD reported a positive association with crowding; as did those examining dwelling characteristics, education levels and employment status of parents or cases, and nutrition. However, there was considerable heterogeneity in measures used, study settings and outcome ascertainment.
We noted a lack of well-designed research and interventions aimed at unravelling poverty as a mechanism for ARF and RHD, which is at odds with the widespread acceptance of ARF and RHD as diseases of poverty [2]. This paradox is exemplified in the hopeful remarks of Perry et al (1937)—the earliest study included in this review- that the noted association between crowding and ARF be ‘the starting-point and not the end of research…and that [further] research…may be fruitful in elucidating the aetiology of rheumatic heart disease and in discovering means for its prevention’; [96] yet 80 years later there remains a paucity of evidence of preventative actions at the primordial level.
This issue is well described among the neglected tropical diseases, [13, 122, 126, 127] a diverse group of communicable diseases that cause significant burden of suffering and economic impacts among poor and marginalised populations living in tropical and subtropical regions [128]. Poverty creates the milieu for these diseases to flourish; the low resource settings then exacerbate difficulties inherent in conducting high quality observational or interventional research. Finally, the neglect of the social, economic, political and physical contexts in which affected populations live, leaves the root causes unchanged [13, 122].
Given the study designs and limited quality of papers, we used the Bradford Hill criteria, a set of guiding principles for interpreting links between environmental influences and disease, as a pragmatic framework to consider the findings of this systematic review. These criteria are not definitive rules, rather they provide an analytical framework to consider whether cause and effect is the reasonable inference [129]. The weight of evidence in this systematic review supports a causative relationship between crowding and promotion of GAS transmission, and its rheumatic sequelae. Particular strengths were that evidence was collected across many study types including prospective and retrospective cohort studies and covered diverse population groups globally; features that enhance confidence in casual interpretation [129]. While the presence of a biological gradient may not rule out confounding (as a confounder may also exert a dose-response effect), it provides compelling evidence of a causative nature of this association. Furthermore, there is firm biologic plausibility, since GAS infection is transmitted by close contact and the respiratory route [130]. It can be inferred that crowding plays a critical mediating role between poverty and RHD prevalence, supported by those studies which included multivariable analyses [35, 40, 74, 90, 97, 105].
Overall socioeconomic status also effectively met the Bradford Hill criteria for causation of the outcomes of interest. Since this overarching category is collinear with (and either determined by or a determinant of) the other measures examined, it is not possible to tease out specifics relating to how each criterion was met. In general, it is well understood that rising socioeconomic status, prior to availability of penicillin, was associated with a steady decline in death rates from ARF in industrialised nations [131]. It is likely the combination of these adverse factors that creates the environment that drives ARF and RHD risk among socioeconomically disadvantaged populations. This compounding effect also explains why these factors individually do not necessarily exert the same risk.
The other factors explored in this review had insufficient evidence to suggest causal links. Dwelling characteristics frequently displayed an association with GAS infection, ARF, or RHD risk, but measures were so heterogeneous and context specific that generalizability is impossible. Further, authors consistently omitted a proposed mechanism to explain their findings. The role of fomites in GAS transmission was not supported [106, 108]. Soap and hand-washing had an impressive effect on reducing impetigo, but only one study explored this intervention [107]. Findings relating to education and employment were inconsistent. Specific nutritional interventions were suggested and tested, but lacked scale and consistency. Studies that demonstrated the association of low income and risk of GAS infection and associated diseases did not explore which economic deficits confer the risk (e.g. unaffordability of health care, poor diet etc.). General explorations of income as a social determinant of heath cite that low income exerts a risk to health through material deprivation (medical care, nutrition, housing, and sanitation) and social participation (education, employment) [120]. The association of income in this instance with RHD and its antecedents is likely a consequence of these intermediary factors.
This systematic review is the most extensive review to date targeting observational and experimental studies in the area of social determinants of health and GAS infection, ARF and RHD. A limitation was the exclusion of non-English articles; nevertheless, a wide variety of countries were represented. A further limitation was that all stages of article appraisal were undertaken by one reviewer. However, a strict methodological process was followed utilising assessment tools designed for each study type allowing greater specificity in appraisal and uniformity in the reviewing process.
The poor quality of this body of evidence is the most critical limitation in guiding preventative actions; however, it would be erroneous to reject all findings. Rather it should be considered how probable or not it is that this diverse collection of studies dispersed in time, person and place could all be flawed and biased in the same way. Confounding is another important consideration in this review. Simple cause-effect relationships within social determinants of health are not readily apparent [13, 132]- rather they are complex and are characterized by multiple determinants, multiple outcomes, and multiple potential interactions [133]. That few included studies undertook multivariate analysis is a limitation of this review, which is why it was important to highlight their results explicitly in this analysis.
Many studies presented outcomes at a group or ecological level rather than an individual level. However, assessing social determinants of health at the group level is also important because people do not live in isolation; a population level approach ensures that a wide variety of contributing settings and activities are not inadvertently overlooked [134]. Also, these studies designs can reflect the interventions addressing the social determinants of health that are aimed at the population level.
Finally, reporting bias is an important consideration in systematic reviews. The factors extracted for this review were frequently not the primary outcome of studies, and so their positive, neutral or negative findings would be less likely to influence whether a paper was published.
Rather than re-describe the problem, the aim of this research was to guide solutions. In ARF and RHD where treatment is logistically intensive and painful, [135] and vaccines or mass drug administration are not currently available options, the case for action on the social determinants that drive ARF and RHD risk is unquestionably convincing. Several candidate vaccines are in development, [136] but even if found to be safe and effective, not all at-risk populations would be able to readily access a new vaccine.
Based on these findings, we recommend that ARF and RHD control programs should address household crowding—particularly in high-income countries where funding and resourcing is more feasible. Structural crowding (inadequate living space including number of bedrooms) must be addressed in collaboration with designers and providers of public housing in partnerships that recognise housing needs to support good health.
Functional crowding (people sharing living spaces for safety, warmth or social cohesion, especially in traditional societies) is more difficult to address, requiring in-depth cultural understanding. For example, in Australian Aboriginal societies, rights and obligations around accommodating extra people in a house must be respected in interventions to reduce household crowding [137]. Where close living is important culturally, ways to live safely in larger households, focusing on ensuring adequate health literacy and washing facilities, need to be implemented.
Site-specific tailoring of interventions are needed: publically-funded interventions in a cold climate high RHD-burden setting for instance include the provision of household insulation and heating to reduce functional bedroom crowding [138]. Conversely, in hot climates, constructing community swimming pools is an effective intervention to decrease GAS skin infections, [139] though pools must be adequately managed so as to not introduce other health risks. Further practical interventions to mitigate the health risks arising from crowding include: community development projects to improve health literacy pertaining to infection transmission; creating community demand for sanitation and hygiene; [140] and effective community consultation about factors to motivate change in behaviour.
Interventions to tackle socioeconomic status may be considered beyond the reach of medical research and health service delivery, but this is not so. Research and service delivery initiatives in high RHD-burden settings should ensure that they provide opportunities for community engagement and employment; and funding bodies should ensure that initiatives supporting strengthening of socioeconomic status are valued as being critical in disease prevention.
RHD control programs, where they exist, should aim to routinely collect objective metrics on social and environmental factors to further inform advocacy, tailor it to local needs and add to the evidence base. Outcomes of advocacy should be interventions targeting social determinants such as crowding reduction and hygiene hardware improvement; and these should be accompanied by rigorous evaluation and sharing of findings.
Findings from this systematic review will be able to inform guidelines and policies regarding primordial-level preventative interventions against GAS infection, ARF and RHD. The wide body of evidence exploring links between certain social and environmental factors and these conditions is limited by poor quality and a lack of interventional studies. Historically, this has hampered the ability of control programs and guidelines to legitimately target these factors. However, when scrutinized against the Bradford Hill criteria assessing the evidence of a causal relationship, the link between overall socioeconomic status and crowding with ARF and RHD risk can be satisfactorily viewed as one of cause and effect. This clear and powerful message should be reflected in ARF and RHD control efforts. A critical role for ARF and RHD control programs and registers is to routinely collect and analyse data on these social determinants alongside clinical markers of case management to inform future advocacy and interventions.
|
10.1371/journal.pcbi.1002759 | Music in Our Ears: The Biological Bases of Musical Timbre Perception | Timbre is the attribute of sound that allows humans and other animals to distinguish among different sound sources. Studies based on psychophysical judgments of musical timbre, ecological analyses of sound's physical characteristics as well as machine learning approaches have all suggested that timbre is a multifaceted attribute that invokes both spectral and temporal sound features. Here, we explored the neural underpinnings of musical timbre. We used a neuro-computational framework based on spectro-temporal receptive fields, recorded from over a thousand neurons in the mammalian primary auditory cortex as well as from simulated cortical neurons, augmented with a nonlinear classifier. The model was able to perform robust instrument classification irrespective of pitch and playing style, with an accuracy of 98.7%. Using the same front end, the model was also able to reproduce perceptual distance judgments between timbres as perceived by human listeners. The study demonstrates that joint spectro-temporal features, such as those observed in the mammalian primary auditory cortex, are critical to provide the rich-enough representation necessary to account for perceptual judgments of timbre by human listeners, as well as recognition of musical instruments.
| Music is a complex acoustic experience that we often take for granted. Whether sitting at a symphony hall or enjoying a melody over earphones, we have no difficulty identifying the instruments playing, following various beats, or simply distinguishing a flute from an oboe. Our brains rely on a number of sound attributes to analyze the music in our ears. These attributes can be straightforward like loudness or quite complex like the identity of the instrument. A major contributor to our ability to recognize instruments is what is formally called ‘timbre’. Of all perceptual attributes of music, timbre remains the most mysterious and least amenable to a simple mathematical abstraction. In this work, we examine the neural underpinnings of musical timbre in an attempt to both define its perceptual space and explore the processes underlying timbre-based recognition. We propose a scheme based on responses observed at the level of mammalian primary auditory cortex and show that it can accurately predict sound source recognition and perceptual timbre judgments by human listeners. The analyses presented here strongly suggest that rich representations such as those observed in auditory cortex are critical in mediating timbre percepts.
| A fundamental role of auditory perception is to infer the likely source of a sound; for instance to identify an animal in a dark forest, or to recognize a familiar voice on the phone. Timbre, often referred to as the color of sound, is believed to play a key role in this recognition process [1]. Though timbre is an intuitive concept, its formal definition is less so. The ANSI definition of timbre describes it as that attribute that allows us to distinguish between sounds having the same perceptual duration, loudness, and pitch, such as two different musical instruments playing exactly the same note [2]. In other words, it is neither duration, nor loudness, nor pitch; but is likely “everything else”.
As has been often been pointed out, this definition by the negative does not state what are the perceptual dimensions underlying timbre perception. Spectrum is obviously a strong candidate: physical objects produce sounds with a spectral profile that reflects their particular sets of vibration modes and resonances [3]. Measures of spectral shape have thus been proposed as basic dimensions of timbre (e.g., formant position for voiced sounds in speech, sharpness, and brightness) [4], [5]. But timbre is not only spectrum, as changes of amplitude over time, the so-called temporal envelope, also have strong perceptual effects [6], [7]. To identify the most salient timbre dimensions, statistical techniques such as multidimensional scaling have been used: perceptual differences between sound samples were collected and the underlying dimensionality of the timbre space inferred [8], [9]. These studies suggest a combination of spectral and temporal dimensions to explain the perceptual distance judgments, but the precise nature of these dimensions varies across studies and sound sets [10], [11]. Importantly, almost all timbre dimensions that have been proposed to date on the basis of psychophysical studies [12] are either purely spectral or purely temporal. The only spectro-temporal aspect of sound that has been considered in this context is related to the asynchrony of partials around the onset of a sound (8,9), but the salience of this spectro-temporal dimension was found to be weak and context-dependent [13].
Technological approaches, not concerned with biology nor human perception, have explored much richer feature representations that span both spectral, temporal, and spectro-temporal dimensions. The motivation for these engineering techniques is an accurate recognition of specific sounds or acoustic events in a variety of applications (e.g. automatic speech recognition; voice detection; music information retrieval; target tracking in multisensor networks and surveillance systems; medical diagnosis, etc.). Myriad spectral features have been proposed for audio content analysis, ranging from simple summary statistics of spectral shape (e.g. spectral amplitude, peak, centroid, flatness) to more elaborate descriptions of spectral information such as Mel-Frequency Cepstral Coefficients (MFCC) and Linear or Perceptual Predictive Coding (LPC or PLP) [14]–[16]. Such metrics have often been augmented with temporal information, which was found to improve the robustness of content identification [17], [18]. Common modeling of temporal dynamics also ranged from simple summary statistics such as onsets, attack time, velocity, acceleration and higher-order moments to more sophisticated statistical temporal modeling using Hidden Markov Models, Artificial Neural Networks, Adaptive Resonance Theory models, Liquid State Machine systems and Self-Organizing Maps [19], [20]. Overall, the choice of features was very dependent on the task at hand, the complexity of the dataset, and the desired performance level and robustness of the system.
Complementing perceptual and technological approaches, brain-imaging techniques have been used to explore the neural underpinnings of timbre perception. Correlates of musical timbre dimensions suggested by multidimensional scaling studies have been observed using event-related potentials [21]. Other studies have attempted to identify the neural substrates of natural sound recognition, by looking for brain areas that would be selective to specific sound categories, such as voice-specific regions in secondary cortical areas [22], [23] and other sound categories such as tools [24] or musical instruments [25]. A hierarchical model consistent with these findings has been proposed in which selectivity to different sound categories is refined as one climbs the processing chain [26]. An alternative, more distributed scheme has also been suggested [27], [28], which includes the contribution of low-level cues to the large perceptual differences between these high-level sound categories.
A common issue for the psychophysical, technological, and neurophysiological investigations of timbre is that the generality of the results is mitigated by the particular characteristics of the sound set used. For multi-dimensional scaling behavioral studies, by construction, the dimensions found will be the most salient within the sound set; but they may not capture other dimensions which could nevertheless be crucial for the recognition of sounds outside the set. For engineering studies, dimensions may be designed arbitrarily as long as they afford good performance in a specific task. For the imaging studies, there is no suggestion yet as to which low-level acoustic features may be used to construct the various selectivity for high-level categories while preserving invariance within a category. Furthermore, there is a major gap between these studies and what is known from electrophysiological recordings in animal models. Decades of work have established that auditory cortical responses display rich and complex spectro-temporal receptive fields, even within primary areas [29], [30]. This seems at odds with the limited set of spectral or temporal dimensions that are classically used to characterize timbre in perceptual studies.
To bridge this gap, we investigate how cortical processing of spectro-temporal modulations can subserve both sound source recognition of musical instruments and perceptual timbre judgments. Specifically, cortical receptive fields and computational models derived from them are shown to be suited to classify a sound source from its evoked neural activity, across a wide range of instruments, pitches and playing styles, and also to predict accurately human judgments of timbre similarities
Responses in primary auditory cortex (A1) exhibit rich selectivity that extends beyond the tonotopy observed in the auditory nerve. A1 neurons are not only tuned to the spectral energy at a given frequency, but also to the specifics of the local spectral shape such as its bandwidth [31], spectral symmetry [32], and temporal dynamics [33] (Figure 1). Put together, one can view the resulting representation of sound in A1 as a multidimensional mapping that spans at least three dimensions: (1) Best frequencies that span the entire auditory range; (2) Spectral shapes (including bandwidth and symmetry) that span a wide range from very broad (2–3 octaves) to narrowly tuned (<0.25 octaves); and (3) Dynamics that range from very slow to relatively fast (1–30 Hz).
This rich cortical mapping may reflect an elegant strategy for extracting acoustic cues that subserve the perception of various acoustic attributes (pitch, loudness, location, and timbre) as well as the recognition of complex sound objects, such as different musical instruments. This hypothesis was tested here by employing a database of spectro-temporal receptive fields (STRFs) recorded from 1110 single units in primary auditory cortex of 15 awake non-behaving ferrets. These receptive fields are linear descriptors of the selectivity of each cortical neuron to the spectral and temporal modulations evident in the cochlear “spectrogram-like” representation of complex acoustic signals that emerges in the auditory periphery. Such STRFs (with a variety of nonlinear refinements) have been shown to capture and predict well cortical responses to a variety of complex sounds like speech, music, and modulated noise [34]–[38].
To test the efficacy of STRFs in generating a representation of sound that can distinguish among a variety of complex categories, sounds from a large database of musical instruments were mapped onto cortical responses using the physiological STRFs described above. The time-frequency spectrogram for each note was convolved with each STRF in our neurophysiology database to yield a firing rate that is then integrated over time. This initial mapping was then reduced in dimensionality using singular value decomposition to a compact eigen-space; then augmented with a nonlinear statistical analysis using support vector machine (SVM) with Gaussian kernels [39] (see METHODS for details). Briefly, support vector machines are classifiers that learn to separate, in our specific case, the patterns of cortical responses induced by the different instruments. The use of Gaussian kernels is a standard technique that allows to map the data from its original space (where data may not be linearly separable) onto a new representational space that is linearly separable. Ultimately, the analysis constructed a set of hyperplanes that outline the boundaries between different instruments. The identity of a new sample was then defined based on its configuration in this expanded space relative to the set of learned hyperplanes (Figure 2).
Based on the configuration above and a 10% cross-validation technique, the model trained using the physiological cortical receptive fields achieved a classification accuracy of 87.22%±0.81 (the number following the mean accuracy represents standard deviation, see Table 1). Remarkably, this result was obtained with a large database of 11 instruments playing between 30 and 90 different pitches with 3 to 19 playing styles (depending on the instrument), 3 style dynamics (mezzo, forte and piano), and 3 manufacturers for each instrument (an average of 1980 notes/instrument). This high classification accuracy was a strong indicator that neural processing at the level of primary auditory cortex could not only provide a basis for distinguishing between different instruments, but also had a robust invariant representation of instruments over a wide range of pitches and playing styles.
Despite the encouraging results obtained using cortical receptive fields, the classification based on neurophysiological recordings was hampered by various shortcomings including recording noise and other experimental constraints. Also, the limited selection of receptive fields (being from ferrets) tended to under-represent parameter ranges relevant to humans such as lower frequencies, narrow bandwidths (limited to a maximum resolution of 1.2 octaves), and coarse sampling of STRF dynamics.
To circumvent these biases, we employed a model that mimics the basic transformations along the auditory pathway up to the level of A1. Effectively, the model mapped the one-dimensional acoustic waveform onto a multidimensional feature space. Importantly, the model allowed us to sample the cortical space more uniformly than physiological data available to us, in line with findings in the literature [29], [30], [40].
The model operates by first mapping the acoustic signal into an auditory spectrogram. This initial transformation highlights the time varying spectral energies of different instruments which is at the core of most acoustic correlates and machine learning analyses of musical timbre [5], [11], [13], [41], [42]. For instance, temporal features in a musical note include fast dynamics that reflect the quality of the sound (scratchy, whispered, or purely voiced), as well as slower modulations that carry nuances of musical timbre such as attack and decay times, subtle fluctuations of pitch (vibrato) or amplitude (shimmer). Some of these characteristics can be readily seen in the auditory spectrograms, but many are only implicitly represented. For example, Figure 3A contrasts the auditory spectrogram of a piano vs. violin note. For violin, the temporal cross-section reflects the soft onset and sustained nature of bowing and typical vibrato fluctuations; the spectral slice captures the harmonic structure of the musical note with the overall envelope reflecting the resonances of the violin body. By contrast, the temporal and spectral modulations of a piano (playing the same note) are quite different. Temporally, the onset of piano rises and falls much faster, and its spectral envelope is much smoother.
The cortical stage of the auditory model further analyzes the spectral and temporal modulations of the spectrogram along multiple spectral and temporal resolutions. The model projects the auditory spectrogram onto a 4-dimensional space, representing time, tonotopic frequency, spectral modulations (or scales) and temporal modulations (or rates). The four dimensions of the cortical output can be interpreted in various ways. In one view, the cortical model output is a parallel repeated representation of the auditory spectrogram viewed at different resolutions. A different view is one of a bank of spectral and temporal modulation filters with different tuning (from narrowband to broadband spectrally, and slow to fast modulations temporally). In such view, the cortical representation is a display of spectro-temporal modulations of each channel as they evolve over time. Ultimately each filter acts as a model cortical neuron whose output reflects the tuning of that neuronal site. The model employed here had 30,976 filters (128freq×22 rates×11 scales), hence allowing us to obtain a full uniform coverage of the cortical space and bypassing the limitations of neurophysiological data. Note that we are not suggesting that ∼30 K neurons are needed for timbre classification, as the feature space is reduced in further stages of the model (see below). We have not performed an analysis of the number of neurons needed for such task. Nonetheless, a large and uniform sampling of the space seemed desirable.
By collapsing the cortical display over frequency and averaging over time, one would obtain a two-dimensional display that preserves the “global” distribution of modulations over the remaining two dimensions of scale and rates. This “scale-rate” view is shown in Figure 3B for the same piano and violin notes in Figure 3A as well as others. Each instrument here is played at two distinct pitches with two different playing styles. The panels provide estimates of the overall distribution of spectro-temporal modulation of each sound. The left panel highlights the fact that the violin vibrato concentrates its peak energy near 6 Hz (across all pitches and styles); which matches the speed of pulsating pitch change caused by the rhythmic rate of 6 pulses per second chosen for the vibrato of this violin note. By contrast, the rapid onset of piano distributes its energy across a wider range of temporal modulations. Similarly, the unique pattern of peaks and valleys in spectral envelopes of each instrument produces a broad distribution along the spectral modulation axis, with the violin's sharper spectral peaks activating higher spectral modulations while the piano's smoother profile activates broad bandwidths. Each instrument, therefore, produces a correspondingly unique spectro-temporal activation pattern that could potentially be used to recognize it or distinguish it from others.
Several computational models were compared in the same classification task analysis of the database of musical instruments as described earlier with real neurophysiological data. Results comparing all models are summarized in Table 1. For what we refer to as the full model, we used the 4-D cortical model. The analysis started with a linear mapping through the model receptive fields, followed by dimensionality reduction and statistical classification using support vector machines with re-optimized Gaussian kernels (see Methods). Tests used a 10% cross-validation method. The cortical model yielded an excellent classification accuracy of 98.7%±0.2.
We also explored the use of linear support vector machine, by bypassing the use of the Gaussian kernel. We performed a classification of instruments using the cortical responses obtained from the model receptive fields and a linear SVM. After optimization of the decision boundaries, we obtained an accuracy of 96.2%±0.5. This result supports our initial assessment that the cortical space does indeed capture most of the subtleties that are unique to a common instrument but distinct between different classes. It is mostly the richness of the representation that underlies the classification performance: only a small improvement in accuracy is observed by adding the non-linear warping in the full model.
In order to better understand the contribution of the cortical analysis beyond the time-frequency representation, we explored reduced versions of the full model. First we performed the timbre classification task using the auditory spectrogram as input. The feature spectra were obtained by processing the time waveform of each note through the cochlear-like filterbank front-end and averaging the auditory spectrograms over time, yielding a one-dimensional spectral profile for each note. These were then processed through the same statistical SVM model, with Gaussian functions optimized for this new representation using the exact same methods as used for cortical features. The classification accuracy for the spectral slices with SVM optimization attained a good but limited accuracy of 79.1%±0.7. It is expected that a purely spectral model would not be able to classify all instruments. Whereas basic instrument classes differing by their physical characteristics (wind, percussion, strings) may have the potential to produce different spectral shapes, preserved in the spectral vector, more subtle differences in the temporal domain should prove difficult to recognize on this basis (see Figure 4). We shall revisit this issue of contribution and interactions between spectral and temporal features later (see Control Experiments section).
We performed a post-hoc analysis of the decision space based on cortical features in an attempt to get a better understanding of the configuration of the decision hyperplanes between different instrument classes. The analysis treated the support vectors (i.e. samples of each instrument that fall right on the boundary that distinguishes it from another instrument) for each instrument as samples from an underlying high-dimensional probability density function. Then, a measure of similarity between pairs of probability functions (symmetric Kullback–Leibler (KL) divergence [43]) was employed to provide a sense of distance between each instrument pair in the decision space. Because of the size and variability in the timbre decision space, we pooled the comparisons by instrument class (winds, strings and percussions). We also focused our analysis on the reduced dimensions of the cortical space; called ‘eigen’-rate, ‘eigen’-scale and ‘eigen’-frequencies; obtained by projecting the equivalent dimensions in the cortical tensor (rate, scale and frequency, respectively) into a reduced dimensional space using singular-value decomposition (see METHODS). The analysis revealed a number of observations (see Figure 5). For instance, wind and percussion classes were the most different (occupy distant regions in the decision space), followed by strings and percussions then strings and winds (average KL distances were 0.58, 0.41, 0.35, respectively). This observation was consistent with the subjective judgments of human listeners presented next (see off-diagonal entries in Figure 6B). All 3 pair comparisons were statistically significantly different from each other (Wilcoxon ranksum test, p<10−5 for all 3 pairs). Secondly, the analysis revealed that the 2 first ‘eigen’-rates captured most of the difference between the instrument classes (statistical significance in comparing the first 2 eigenrates with the others; Wilcoxon ranksum test, p = 0.0046). In contrast, all ‘eigen’-scales were variable across classes (Kruskal-Wallis test, p = 0.9185 indicating that all ‘eigen’-scales contributed equally in distinguishing the broad classes). A similar analysis indicated that the first four ‘eigen’-frequencies were also significantly different from the remaining features (Wilcoxon ranksum test, p<10−5). One way to interpret these observations is that the first two principal orientations along the rate axis captured most of the differences that distinguish winds, strings and percussions. This seems consistent with the large differences in temporal envelope shape for these instruments classes, which can be represented by a few rates. By contrast, the scale dimension (which captures mostly spectral shape, symmetry and bandwidth) was required in its entirety to draw a boundary between these classes, suggesting that unlike the coarser temporal characteristics, differentiating among instruments entails detailed spectral distinctions of a subtle nature.
Spectral features have been extensively used for tasks of musical timbre classification of isolated notes, solo performances or even multi-instrument recordings. Features such as Cepstral Coefficients or Linear Prediction of the spectrum resonances yielded performance in the range of 77% to 90% when applied to databases similar to the one used in the present study [44]–[46].
There is wide agreement in the literature that inclusion of simple temporal features, such as zero-crossing rate, or more complex ones such as trajectory estimation of spectral envelopes, is often desirable and results in improvement of the system performance. Tests on the RWC database with both spectral and temporal features reported an accuracy of 79.7% using 19 instruments [47] or 94.9% using 5 instruments [42]. Tests of spectrotemporal features on other music databases has often yielded a range of performances between 70–95% [48]–[51].
Whereas a detailed comparisons with our results is beyond the scope of this paper, we can still note that, if anything, the recognition rates we report for the full auditory model are generally in the range or above those reported by state-of-the-art signal processing techniques.
Given the ability of the cortical model to capture the diversity of musical timbre across a wide range of instruments in a classification task, we next explored how well the cortical representation (from both real and model neurons) does in capturing human perceptual judgments of distance in the musical timbre space. To this end, we used human judgments of musical timbre distances using a psychoacoustic comparison paradigm.
Human listeners were asked to rate the similarity between musical instruments. We used three different notes (A3, D4 and G#4) in three different experiments. Similarity matrices for all three notes yielded reasonably balanced average ratings across subjects, instrument pair order (e.g. piano/violin vs. violin/piano) and pitches, in agreement with other studies [52] (Figure 6A). Therefore, we combined the matrices across notes and listeners into an upper half matrix shown in Figure 6B, and used it for all subsequent analyses. For comparison with previous studies, we also ran a multidimensional scaling (MDS) analysis [53] on this average timbre similarity rating and confirmed that the general configuration of the perceptual space was consistent with previous studies (Figure 6C) [8]. Also for comparison, we tested acoustical dimensions suggested in those studies. The first dimension of our space correlated strongly with the logarithm of attack-time (Pearson's correlation coefficient: ρ = 0.97, p<10−3), and the second dimension correlated reasonably well with the center of mass of the auditory spectrogram, also known as spectral centroid (Pearson's correlation coefficient: ρ = 0.62, p = 0.04).
The perceptual results obtained above, reflecting subjective timbre distances between different instruments, summarizes an elaborate set of judgments that potentially reveal other facets of timbre perception than the listeners' ability to recognize instruments. We then explored whether the cortical representation could account for these judgments. Specifically, we asked whether the cortical analysis maps musical notes onto a feature space where instruments like violin and cello are distinct, yet closer to each other than a violin and a trumpet. We used the same 11 instruments and 3 pitches (A3, D4 and G#4) employed in the psychoacoustics experiment above and mapped them onto a cortical representation using both neurophysiological and model STRFs. Each note was then vectorized into a feature data-point and mapped via Gaussian kernels. These kernels are similar to the radial basis functions used in the previous section, and aimed at mapping the data from its original cortical space to a linearly separable space. Unlike the generic SVM used in the classification of musical timbre, the kernel parameters here were optimized based on the human scores following a similarity-based objective function. The task here was not merely to classify instruments into distinct classes, but rather to map the cortical features according to a complex set of rules. Using this learnt mapping, a confusion matrix was constructed based on the instrument distances, which was then compared with the human confusion matrix using a Pearson's correlation metric. We performed a comparison with the physiological as well as model STRFs. The simulated confusion matrices are shown in Figure 7A–B.
The success or otherwise of the different models was estimated by correlating the human dissimilarity matrix to that generated by the model. No attempt was made at producing MDS analyses of the model output, as meaningfully comparing MDS spaces is not a trivial problem [52]. Physiological STRFs yielded a correlation coefficient of 0.73, while model STRFs yielded a correlation of 0.94 (Table 2).
In order to disentangle the contribution of the “input” cortical features versus the “back-end” machine learning in capturing human behavioral data, we recomputed confusion matrices using alternative representations such as the auditory spectrogram and various marginals of the cortical distributions. In all these control experiments, the Gaussian kernels were re-optimized separately to fit the data representation being explored.
We first investigated the performance using auditory spectrum features with optimized Gaussian kernels. The spectrogram representation yielded a similarity matrix that captures the main trends in human distance judgments, with a correlation coefficient of 0.74 (Figure 7C, leftmost panel). Similar experiments using a traditional spectrum (based on Fourier analysis of the signal) yield a correlation of 0.69.
Next, we examined the effectiveness of the model cortical features by reducing them to various marginal versions with fewer dimensions as follows. First, we performed an analysis of the spectral and temporal modulations as a separable cascade of two operations. Specifically, we analyzed the spectral profile of the auditory spectrogram (scales) independently from the temporal dynamics (rates) and stack the two resulting feature vectors together. This analysis differed from the full cortical analysis that assumes an inseparable analysis of spectro-temporal features. An inseparable function is one that cannot be factorized into a function of time and a function of frequency; i.e. a matrix of rank greater than 1 (see Methods). By construction, a separable function consists of temporal cross sections that are scaled versions of the same essential temporal function. A consequence of such constraint is that a separable function cannot capture orientation in time-frequency space (e.g. FM sweeps). In contrast, the full cortical analysis estimates modulations along both time and frequency axes in addition to an integrated view of the two axes including orientation information The analysis based on the separable model achieved a correlation coefficient of 0.83 (Table 2).
Second, we further reduced the separable spectro-temporal space by analyzing the modulation content along both time and frequency without maintaining the distribution along the tonotopic axis. This was achieved by simply integrating the modulation features along the spectral axis thus exploring the global characteristic of modulation regardless of tonotopy (Figure 7C, rightmost panel). This representation is somewhat akin to what would result from a 2-dimensional Fourier analysis of the auditory spectrogram. This experiment yielded a correlation coefficient of 0.70 (Table 2), supporting the value of an explicit tonotopic axis in capturing subtle difference between instruments.
Next, we addressed the concern that the mere number of features included in the full cortical model enough to explain the observed performance. We therefore undersampled the full cortical model by employing only 6 scale filters; 10 rate filters and 64 frequency filters by coarsely sampling the range of spectro-temporal modulations. This mapping resulted in a total number of dimensions of 3840; to be comparable to the 4224 dimensions obtained from the separable model. We then performed the dimensionality reduction to 420 dimensions, similar to that used for the separable analysis discussed above. The correlation obtained was 0.86; which is better than that of the separable spectro-temporal model (see Figure 8). This result supports our main claim that the coverage provided by the cortical space allows extracting specific details in the musical notes that highlight information about the physical properties of each instrument; hence enabling classification and recognition.
Finally, we examined the value of the kernel-learning compared to using a simple Euclidian L2 distance at various stages of the model (e.g. peripheral model, cortical stage, reduced cortical model using tensor singular value decomposition). Table 2 summarizes the results of this analysis along various stages of the model shown in Figure 2. The analysis revealed that the kernel-based mapping does provide noticeable improvement to the predictive power of the model but cannot –by itself– explain the results since the same technique applied directly on the spectrum only yielded a correlation of 0.74.
This study demonstrates that perception of musical timbre could be effectively based on neural activations patterns that sounds evoke at the level of primary auditory cortex. Using neurophysiological recordings in mammalian auditory cortex as well as a simplified model of cortical processing, it is possible to accurately replicate human perceptual similarity judgments and classification performance among sounds from a large number of musical instruments. Of course, showing that the information is available at the level of primary auditory cortex does not imply that all neural correlates of sound identification will be found at this level. Nevertheless, it suggests that the spectro-temporal transforms as observed at this stage are critical for timbre perception. Moreover, our analysis highlights the ability of the cortical mapping to capture timbre properties of musical notes and instrument-specific characteristics regardless of pitch and playing style. Unlike static or reduced views of timbre that emphasize three or four parameters extracted from the acoustic waveform, the cortical analysis provides a dynamic view of the spectro-temporal modulations in the signal as they vary over time. A close examination of the contribution of different auditory features and processing stages to the timbre percepts highlights three key points.
First, neither the traditional spectrum nor its variants (e.g. average auditory spectrum [54]) are well-suited to account for timbre perception in full. According to our simulations, these representations encode the relevant spectral and temporal acoustic features too implicitly to lend themselves for exploitation by classifiers and other machine learning techniques. In some sense, this conclusion is expected given the multidimensional nature of the timbre percept compared to the dense two-dimensional spectrogram; and is in agreement with other findings from the literature [19].
Second, when considering more elaborate spectro-temporal cortical representations, it appears that the full representation accounts best for human performance. The match worsens if instead marginals are used by collapsing the cortical representation onto one or more dimensions to extract the purely spectral or temporal axes or scale-rate map (Figure 3, Tables 1 and 2). This is the case even if all dimensions are used separately, suggesting that there are joint spectro-temporal features that are key to a full accounting of timbre. While the role of both purely spectral and temporal cues in musical timbre is quite established [12], our analysis emphasizes the crucial contribution of a joint spectro-temporal representation. For example, FM modulations typical of vibrato in string instruments are joint features that cannot be easily captured by the marginal spectral or temporal representations. Interestingly, acoustical analyses and fMRI data in monkeys suggest that the spectro-temporal processing scheme used here may be able to differentiate between broad sound categories (such as monkey calls vs. bird calls vs. human voice), with corresponding neural correlates when listening to those sounds [55].
Third, a nonlinear decision boundary in the SVM classifier is essential to attain the highest possible match between the cortical representation and human perception. Linear metrics such as L2 are less optimal, indicating that the linear cortical representation may not be sufficiently versatile to capture the nuances of various timbres. The inadequacy of the linear cortical mapping has previously been described when analyzing neural responses to complex sounds such as speech at the level of auditory cortex [35], [36], [38]. In these cases, it is necessary to postulate the existence of nonlinearities such as divisive normalization or synaptic depression that follows a linear spectro-temporal analysis so as to account fully for the observed responses. In the current study, the exact nature of the nonlinearity remains unclear as it is implicitly subsumed in the Gaussian kernels and subsequent decisions.
In summary, this study leads to the general conclusion that timbre percepts can be effectively explained by the joint spectro-temporal analysis performed at the level of mammalian auditory cortex. However, unlike the small number of spectral or temporal dimensions that have been traditionally considered in the timbre literature, we cannot highlight a simple set of neural dimensions subserving timbre perception. Instead, the model suggests that subtle perceptual distinctions exhibited by human listeners are based on ‘opportunistic’ acoustic dimensions [56] that are selected and enhanced, when required, on the rich baseline provided by the cortical spectro-temporal representation.
All behavioral recordings of timbre similarity judgments with human listeners were approved by the local ethics committee of the Université Paris Descartes. All procedures for recordings of single unit neural activity in ferrets were in accordance with the Institutional Animal Care and Use Committee at the University of Maryland, College Park and the Guidelines of the National Institutes of Health for use of animals in biomedical research.
The cortical model is comprised of two main stages: an early stage mimicking peripheral processing up to the level of the midbrain, and a central stage capturing processing in primary auditory cortex (A1). Full details about the model can be found in [54], [58]; but are described briefly here.
The processing of the acoustic signal in the cochlea is modeled as a bank of 128 constant-Q asymmetric bandpass filters equally spaced on the logarithmic frequency scale spanning 5.3 octaves. The cochlear output is then transduced into inner hair cells potentials via a high pass and low pass operation. The resulting auditory nerve signals undergo further spectral sharpening via a lateral inhibitory network. Finally, a midbrain model resulting in additional loss in phase locking is performed using short term integration with time constant 4 ms resulting in a time frequency representation called as the auditory spectrogram.
The central stage further analyzes the spectro-temporal content of the auditory spectrogram using a bank of modulation selective filters centered at each frequency along the tonotopic axis, modeling neurophysiological receptive fields. This step corresponds to a 2D affine wavelet transform, with a spectro-temporal mother wavelet, define as Gabor-shaped in frequency and exponential in time. Each filter is tuned (Q = 1) to a specific rate ( in Hz) of temporal modulations and a specific scale of spectral modulations ( in cycles/octave), and a directional orientation (+ for upward and − for downward).
For input spectrogram the response of each STRF in the model is given by:(1)where denotes convolution in time and frequency and and are the characteristic phases of the STRF's which determine the degree of asymmetry in the time and frequency axes respectively. The model filters filters can be decomposed in each quadrant (upward + or downward −) into into corresponding to rate and scale filters respectively. Details of the design of the filter functions can be found in [58]. The present study uses 11 spectral filters with characteristic scales [0.25, 0.35, 0.50, 0.71, 1.00, 1.41, 2.00, 2.83, 4.00, 5.66, 8.00] (cycles/octave) and 11 temporal filters with characteristic rates [4.0, 5.7, 8.0, 11.3, 16.0, 22.6, 32.0, 45.3, 64.0, 90.5, 128.0] (Hz), each with upward and downward directionality. All outputs are integrated over the time duration of each note. In order to simplify the analysis, we limit our computations to the magnitude of the cortical output (i.e. responses corresponding to zero-phase filters).
Finally, dimensionality reduction is performed using tensor singular-value decomposition [59]. This technique unfolds the cortical tensor along each dimension (frequency, rate and scale axes) and applies singular value decomposition on the unfolded matrix. We choose 5 eigenscales, 4 eigenrates and 21 eignefrequencies resulting in 420 features with the highest eigenvalues, preserving 99.9% of the variance in the original data. The motivation for this cutoff choice is presented later.
Data used here was collected in the context of a number of studies [60]–[62] and full details of the experimental paradigm are described in these publications. Briefly, extracellular recordings were performed in 15 awake non-behaving domestic ferrets (Mustela putorius) with surgically implanted headposts. Tungsten electrodes (3–8 MΩ) were used to record neural responses from single and multi-units at different depths. All data was processed off-line and sorted to extract single-unit activity.
Spectro-Temporal Receptive fields (STRF) were characterized using TORC (Temporally-Orthogonal Ripple Combination) stimuli [63], consisting of superimposed ripple noises with rates between 4–24 (Hz) and scales between 0 (flat) and 1.4 peaks/octave. Each stimulus was 3 sec with inter-stimulus intervals of 1–1.2 sec, and a full set of 30 TORCs was typically repeated 6–15 times. All sounds were computer-generated and delivered to the animal's ear through inserted earphones calibrated in-situ. TORC amplitude is fixed between 55–75 dB SPL.
STRFs were derived using standard reverse correlation techniques, and a signal-to-noise ratio (SNR) for each STRF was measured using a bootstrap technique (see [63] for details). Only STRFs with SNR≥2 were included in the current study, resulting in a database of 1110 STRFs (average 74 STRFs/animal). Note because of the experimental paradigm, STRFs spanned a 5-octave range with low frequencies 125, 250 or 500 Hz. In the current study, all STRFs were aligned to match the frequency range of musical note spectrograms. Since all our spectrograms start at 180 Hz and cover 5.3 octaves, we scaled and shifted the STRF's to fit this range.
The neurophysiological STRFs were employed to perform the timbre analysis by convolving each note's auditory spectrogram z(t,f) with each STRF in the database as in Equation (2).(2)The resulting firing rate vector was then integrated over time yielding an average response across the tonotopic axis. The output from all STRFs were then stacked together, resulting in a 142080 (128 frequency channels ×1110 STRFs) dimensional vector. We reduced this vector using singular value decomposition and mapped it onto 420 dimensions, which preserve 99.9% of the data variance in agreement with dimensionality used for model STRFs.
In order to test the cortical representation's ability to discriminate between different musical instruments, we augmented the basic auditory model with a statistical clustering model based on support vector machines (SVM) [39]. Support vector machines are classifiers that learn a set of hyperplanes (or decision boundaries) in order to maximally separate the patterns of cortical responses caused by the different instruments.
Each cortical pattern was projected via Gaussian kernel to a new dimensional space. The use of kernels is a standard technique used with support vector machines, aiming to map the data from its original space (where data may not be linearly separable) onto a new representational space that is linearly separable. This mapping of data to a new (more linear space) through a the use of a kernel or transform is commonly referred to as the “kernel trick” [39]. In essence, kernel functions aim to determine the relative position or similarity between pairs of points in the data. Because the data may lie in a space that is not linearly separable (not possible to use simple lines or planes to separate the different classes), it is desirable to map the data points onto a different space where this linear separability is possible. However, instead of simply projecting the data points themselves onto a high-dimensional feature space which would increase complexity as a function of dimensionality, the “kernel trick” avoids this direct mapping. Instead, it provides a method for mapping the data into an inner product space without explicitly computing the mapping of the observations directly. In other words, it computes the inner product between the data points in the new space without computing the mapping explicitly.
The kernel used here is given by(3)where and are the feature vectors of 2 sound samples. The parameter for the Gaussian kernel and the cost parameter for the SVM algorithm were optimized on a subset of the training data.
A classifier is trained for every pair of classes and . Each of these classifiers then gives a label for a test sample. Note that or . We count the number of labels . The test sample is then assigned to the class with maximum count given by . The parameter in Equation (3) was chosen by doing a grid search over a large parameter span in order to optimize the classifier performance in correctly distinguishing different instruments. This tuning was done by training and testing on a subset of the training data. For model testing, we performed a standard k-fold cross validation procedure with k = 10 (90% training, 10% testing). The dataset was divided into 10 parts. We then left out one part at a time and trained on the remaining 9 parts. The results reported are the average performance over all 10 iterations. A single Gaussian parameter was optimized for all the pair-wise classifiers across all the 10-fold cross validation experiments.
In order to better understand the mapping of the different notes in the high-dimensional space used to classify them, we performed a closer analysis of the support vectors for each instrument pair i and j. Support vectors are the samples from each class that fall exactly on the margin between class i and class j, and therefore are likely to be more confusable between the classes. Since we are operating in the ‘classifier space’, each of the support vectors is defined in a reduced dimensional hyperspace consisting of 5 eigen-scales, 4 eigen-rates, and 21 eigen-frequencies as explained above (a total of 420 dimensions). The collection of all support vectors for each class i can be pulled together to estimate a high-dimensional probability density function. The density function estimate was derived using a histogram method by partitioning the sample space along each dimension into 100 bins, counting how many samples fall into each bin and dividing the counts by the total number of samples. We label the probability distribution for the d-th dimension (d = 1,..,420) . We then computed the symmetric KL divergence, [43], between the support vectors for classes and from the classifier as shown in Equation (4). The KL divergence is simply a measure of difference between pairs of probability distributions, is defined is next:(4)The bins with zero probability were disregarded from the computation of the KL divergence. An alternative method that smoothed the probability distribution over the zero bins was also tested and yielded virtually comparable results. Overall, this analysis is meant to inform about the layout of the timbre decision space. We analyzed the significance of the results between the broad timbre classes (winds, percussions and strings) by pooling individual comparisons between instruments within each group (See Figure 5).
We tested the auditory model's ability to predict human listeners' judgment of musical timbre distances. Just like the timbre classification task, we used the cortical model augmented with Gaussian Kernels. In order to optimize the model to the test data, we employed a variation of the Gaussian kernel that performs an optimized feature embedding on every data dimension. The kernel is defined as follows:(5)where N is the number of dimensions of the features x and y. 's are parameters for the kernel that need to be optimized. We define an objective function that optimizes the correlation between the human perceptual distances and the distances in the embedded space.(6)where is the average profile for the ith instrument over all notes; D(i,j) is the average perceived distance between the ith and jth instrument based on psychoacoustic results and are the average distances from the kernel and the psychoacoustic experiment respectively. represents the variance of the kernel distances over all samples (all instrument pairs). Similarly is the variance of the human perceived distances. We used a gradient ascent algorithm to learn which optimize the objective function.
The correlation analysis employed the same dataset used for the human psychophysical experiment described above. Each note was 0.25 s in duration with sampling rate 44.1 kHz and underwent the same preprocessing as mentioned earlier. The absolute value of the model output was derived for each note and averaged over duration following a similar procedure as the timbre classification described above. The cortical features obtained for the three notes (A3, D4, G#4) were averaged for each instrument i to obtain . Similarly the perceived human distances between instrument i and j were obtained by averaging the (i,j)th and (j,i)th entry in the human distance matrix over all the 3 notes to obtain D(i,j).
Finally, the human and model similarity matrices were compared using the Pearson's correlation metric. In order to avoid overestimating the correlation between the two matrices (the two symmetric values appearing twice in the correlation), we correlated only the upper triangle of each matrix.
|
10.1371/journal.ppat.1007406 | The Citrobacter rodentium type III secretion system effector EspO affects mucosal damage repair and antimicrobial responses | Infection with Citrobacter rodentium triggers robust tissue damage repair responses, manifested by secretion of IL-22, in the absence of which mice succumbed to the infection. Of the main hallmarks of C. rodentium infection are colonic crypt hyperplasia (CCH) and dysbiosis. In order to colonize the host and compete with the gut microbiota, C. rodentium employs a type III secretion system (T3SS) that injects effectors into colonic intestinal epithelial cells (IECs). Once injected, the effectors subvert processes involved in innate immune responses, cellular metabolism and oxygenation of the mucosa. Importantly, the identity of the effector/s triggering the tissue repair response is/are unknown. Here we report that the effector EspO ,an orthologue of OspE found in Shigella spp, affects proliferation of IECs 8 and 14 days post C. rodentium infection as well as secretion of IL-22 from colonic explants. While we observed no differences in the recruitment of group 3 innate lymphoid cells (ILC3s) and T cells, which are the main sources of IL-22 at the early and late stages of C. rodentium infection respectively, infection with ΔespO was characterized by diminished recruitment of sub-mucosal neutrophils, which coincided with lower abundance of Mmp9 and chemokines (e.g. S100a8/9) in IECs. Moreover, mice infected with ΔespO triggered significantly lesser nutritional immunity (e.g. calprotectin, Lcn2) and expression of antimicrobial peptides (Reg3β, Reg3γ) compared to mice infected with WT C. rodentium. This overlapped with a decrease in STAT3 phosphorylation in IECs. Importantly, while the reduced CCH and abundance of antimicrobial proteins during ΔespO infection did not affect C. rodentium colonization or the composition of commensal Proteobacteria, they had a subtle consequence on Firmicutes subpopulations. EspO is the first bacterial virulence factor that affects neutrophil recruitment and secretion of IL-22, as well as expression of antimicrobial and nutritional immunity proteins in IECs.
| Citrobacter rodentium is a gold standard model to study pathogen-host-microbiome interactions. Two of the hallmarks of C. rodentium infection are colonic damage repair responses and colitis; symptoms that are shared with inflammatory bowel diseases in humans. The processes leading to tissue damage repair responses and the implicated bacterial virulence factors are still elusive. In this paper, we show that the C. rodentium type III secretion system effector EspO plays a major role in triggering damage healing responses, recruitment of neutrophils to the colonic villi, secretion of IL-22 from colonic explants and expression of IL-22 regulated genes in intestinal epithelial cells. This paper is the first to report a bacterial virulence factor that impacts on both intestinal epithelial cell proliferation and immune responses.
| Citrobacter rodentium is an extracellular, mouse specific, intestinal pathogen used to model mechanisms of virulence employed by the human pathogens enteropathogenic and enterohemorrhagic Escherichia coli (EPEC and EHEC) and inflammatory bowel diseases [1]. In C57BL/6 mice, shedding of C. rodentium peaks around 8 days post infection (DPI) before being cleared, first via IgG opsonization of bacteria expressing virulence factors and phagocytosis by neutrophils and then through competition by the endogenous microbiota [2]. Infection with C. rodentium elicits robust tissue repair responses, which are characterized by production of IL-22 and cell proliferation leading to colonic crypt hyperplasia (CCH) [3,4], as well as colitis. Although a number of host pathways involved in CCH have been identified [5,6], the C. rodentium virulence factor/s implicated in eliciting the tissue repair response remain elusive.
Both innate and adaptive immune responses are vital for C. rodentium elimination [1]. C. rodentium and its virulence factors are detected by pathogen recognition receptors (PRRs) such as toll-like receptors (TLR)-2 [7] and TLR-4 [8] and activate both the non-canonical (caspase-11) [9] and canonical (e.g. NLRP3) [10] inflammasome pathways in epithelial and myeloid cells. C. rodentium infection triggers expression of pro-inflammatory cytokines, e.g. TNF-α, Cxcl-1 (KC), IL-6 and IL-23, which activate innate lymphoid cells (ILCs) and induce differentiation of naïve T helper (Th) cells into Th1, Th17 or Th22 effector cells secreting interferon-γ (IFN-γ), IL-17A and IL-22, respectively [1,11]. IL-22 triggers production of Reg family antimicrobial peptides including Reg3β and Reg3γ in intestinal epithelial cells (IECs) and plays a critical role in maintaining the epithelial barrier and controlling the bacterial burden [12,13]. At an early stage of the infection (4 DPI), ILC3 are the major source of IL-22 [14,15] whereas CD4+ T cells secrete IL-22 at a later stage (after 9 DPI) [13]. Importantly, Lee et al. have recently reported that CD11b+ Ly6C+ Ly6G+ neutrophils are also a main source of secreted colonic IL-22 in response to C. rodentium infection [16].
C. rodentium colonizes the apical surface of IECs while forming attaching and effacing (A/E) lesions, which are characterized by intimate bacterial interactions with the brush border microvilli [17]. Key to the C. rodentium infection strategy is the injection of multiple effectors into IECs via a type III secretion system (T3SS). Following translocation, the effectors take control of cell signaling for the benefit of the adherent pathogen, including mitochondrial functions (Map, EspF) [18], apoptosis (NleB, NleH), necroptosis (EspL), cellular trafficking (EspG), phagocytosis (EspJ, EspH, EspF), non-canonical and canonical inflammasome pathways (EspI/NleA, NleF) and innate immune responses (NleC, NleD, NleE, Tir) [19,20]. C. rodentium also encodes the T3SS effector EspO, an orthologue of OspE found in Shigella spp [21]. Importantly, EHEC O157:H7 encodes two EspO paralogs [22]. Recently, we reported that infection of mice with a C. rodentium mutant lacking the effector EspO (ΔespO) results in significantly higher bacterial load from 14 DPI compared to infection with wild type (WT) C. rodentium, which concur with reduced levels of colonic CD4+ T cells and C. rodentium-specific serum IgG antibodies [23]. In this paper, we report that EspO plays a role in triggering nutritional immunity, cell proliferation, mucosal innate immune responses, phosphorylation of STAT3 and expression of antimicrobial peptides.
Infection of C57BL/6 mice revealed that at 8 and 14 DPI, C. rodentium ΔespO triggered reduced CCH (48% and 40% reduction, respectively) compare to WT; this phenotype was fully complemented with a plasmid encoding EspO (pespO) (Fig 1A and 1B). By 21 DPI, no difference in the level of CCH was recorded between mice infected with WT, ΔespO-pespO or ΔespO (Fig 1B).
Ki-67 is a marker of cell proliferation expressed in all phases of active cell cycle but absent in resting cells [24]. Consistently, significant reduction (72% reduction) in Ki-67 staining was seen in mice infected with ΔespO compared to mice infected with WT or the complemented strain at 8 DPI (Fig 1C and 1D). Importantly, no difference in body weight was observed between mice infected with WT, ΔespO and ΔespO-pespO at 8 DPI and the uninfected control mice (Fig 1E).
As ΔespO cause a significantly reduced cell proliferation, a marker of tissue damage repair, we determined the outcome of infection of the highly susceptible Rag2-/- il2rg-/- mice (n = 5), which lack NK, ILCs, T and B cells. This revealed that while both WT and ΔespO triggered a similar decline in body weight (Fig 1F), there was a small delay in mortality in mice infected with the ΔespO (Fig 1G), which is consistent with the mutant causing less colonic damage.
A previous study has recently shown that C. rodentium utilizes the T3SS to induce CCH as a means to oxygenate the mucosa and to drive bacterial expansion via CydAB-mediated aerobic respiration [25]. Enumeration of colony-forming units (CFUs) per gram of feces revealed that despite the significant difference in CCH there was no difference in bacterial shedding between WT, ΔespO and ΔespO-pespO up to 8 DPI (Fig 1H). As C. rodentium T3SS effectors execute their function by altering cell signaling in IECs, we aimed to investigate the intracellular processes affected by EspO. To this end, we compared the proteomes of IECs purified from mice infected with WT, ΔespO and ΔespO-pespO at 8 DPI. Assessing purity by flow cytometry revealed that the IEC preparations were enriched by over 90% [26]. For the proteomic analysis, we only considered changes to protein abundance between WT and ΔespO if they were fully restored upon ΔespO complementation.
Associated with the IEC proteomes were 773 C. rodentium proteins (S1 Table), including structural (e.g. EscN, EspA,B,D), chaperone (CesF, CesT) and T3SS effector (Tir, EspF, EspH, EspM2, EspI/NleA) proteins, intimin, the global regulator of virulence RegA [27] and the essential periplasmic serine protease HtrA [28]. Consistent with the similar bacterial shedding (Fig 1C), the proteomes of infected IECs revealed similar abundances of C. rodentium proteins across the challenged groups (Fig 1I).
As key to the C. rodentium infection strategy is formation of A/E lesions, which has a major impact on the shape and function of IECs, we first compared the abundance of brush border proteins in infected IECs. This revealed that Eps8, Villin, Plastin, Ezrin and Espin, as well as actin binding proteins Profilin, Gelsolin, Cobl and Spectrin (and their associated proteins) were in similarly lower abundance in the infected groups compared to the uninfected control mice (Figs 2A, 2B, S1A and S1B). These findings were confirmed by observations made in transmission electron microscopy (TEM) showing typical A/E lesions in colons infected with either the WT or ΔespO (Fig 2C).
We have recently reported that 1,447 proteins, mostly associated with metabolic processes (e.g. ATP production in the mitochondria, lipid biogenesis), were in lower abundance in C. rodentium-infected IECs. In contrast, C. rodentium infection induced the creatine and cholesterol biogenesis pathways, as well as cholesterol efflux [18]. As cell proliferation is dependent on metabolic activity and ATP consumption, we compared the abundance of the metabolic enzymes between WT- and ΔespO-infected IECs. This revealed a similar profile in the infected IECs, with decreased abundance of proteins in the TCA cycle, oxidative phosphorylation (Fig 3A and 3B) and lipid metabolism (Fig 3C), and increased abundance of proteins in the cholesterol, creatine and nucleic acid metabolism (Fig 3C).
Taken together, these results suggest that cellular processes involved in effacement of the brush border microvilli, disruption of the mitochondria and cell metabolisms are not related to CCH. Moreover, the data show that CCH is dispensable for proliferation of C. rodentium in vivo.
In order to determine which biological processes are affected by EspO, we searched for proteins with differential abundance in WT- and ΔespO-infected IECs. This revealed 206 EspO specific proteins with altered abundance compared to WT and ΔespO-pespO (Fig 4A; S2 Table). Among the EspO specific proteins, 41 were grouped under the GO term defense and reactive oxygen species (ROS) responses. Expression of the Nox2 system (e.g. Ncf1, Ncf2, Ncf4, Cyba) was induced in IECs infected with WT, but to a significantly lesser extent in IECs infected with ΔespO (Fig 4B). Nox2 activity is under the regulation of the calprotectin (a heterodimeric complex made of S100a8 and S100a9), which is translocated to the plasma membrane upon activation [29]. Whereas both S100a8 and S100a9 were in higher abundance in IECs infected with WT and ΔespO-pespO, they were in lower abundance in the ΔespO-infected IECs (Fig 4B). As ROS and calprotectin have an antimicrobial activity, we extended the comparison to other antimicrobial proteins (AMPs) expressed in IECs infected with WT, ΔespO and ΔespO-pespO 8 DPI and 14 DPI. This revealed elevated abundance of 16 AMPs, including the antimicrobial peptides Reg3β and Reg3γ, lysozyme (Lyz2) and proteins involved in nutritional immunity including lactotransferin (Ltf, involved in binding and transport of iron), Lipocalin-2 (Lcn2, targeting the bacterial ferric-siderophore enterobactin), as well as calprotectin (which sequesters Mn and Zn), in WT- and ΔespO-pespO-infected mice 8 DPI; compared to WT, these AMPs were found in significantly lower abundance following infection with ΔespO (Fig 4A). The abundance of these proteins, with the exception of Camp, Ctsg and Reg3β, decreased at 14 DPI, with no significant difference between WT and ΔespO. Importantly, several of the identified AMPs were thought to be expressed only by immune cells (Lyz2, S100a8, S100a9, Mpo, Ctsg). However, recent studies have shown that these AMPs are expressed by other cells types upon cytokines stimulation: Lyz2 is expressed by crypt and Paneth cells [30], while S100a8 and S1009 are produced by epithelial cells [31]. In addition, others AMPs, associated with neutrophil extracellular traps (NET), could be found on the surface of IECs (e.g. Mpo, Ctsg) [32].
In order to validate the proteomics data, we quantified the levels of the Reg3β and Reg3ϒ mRNA in IECs by qPCR and fecal Lcn2 by ELISA. We used Dmbt1 (a glycoprotein of the scavenger receptor cysteine-rich family targeting bacteria and reducing their adhesion to the cells) and Indoleamine 2,3-dioxygenase 1 (Ido1, mediates tryptophan depletion and increases antimicrobial metabolites including kynurenine and 3-hydroxy-kynurenine [33]) as controls. Whereas the mRNA levels of Reg3β and Reg3ϒ increased significantly following infection with both WT and ΔespO compared to control mice, the increase seen in the ΔespO was significantly lower compared to WT (Fig 4C and 4D). A similar trend was detected for the abundance of fecal Lcn2, quantified by ELISA (Fig 4E). In contrast, the level of Dmbt1 and Ido1 mRNA increased significantly after infection, with no difference between WT and ΔespO (Fig 4F and 4G). Taken together, these data show that EspO affects expression of antimicrobial peptides and nutritional immunity proteins in IECs and suggest the changes observed at the proteome may be partially regulated at the transcriptional level.
The transcription factor STAT3 commonly regulates expression of genes encoding AMPs, proteins involved in ROS production [34] and cell proliferation [35]. To determine if STAT3 may be differentially activated in IECs infected with WT compared to ΔespO, STAT3 phosphorylation on Tyr 705 was accessed by western blotting. Total STAT3 and GAPDH were used as loading controls. Whereas, little STAT3 phosphorylation was observed in uninfected IECs, robust phosphorylation was detected in IECs infected with WT. Importantly, lower level of STAT3 phosphorylation was observed in IECs infected with ΔespO (Fig 5A and 5B). No difference in the levels of total GAPDH was detected in the different mice whereas a small increased of total STAT3 is detected during infection. In a control experiment we determined if EspO itself can induce phosphorylation of STAT3. For this, HeLa cells were infected with WT or ΔespO for 3 h and the level of STAT3 phosphorylation was measured by WB. IL-6 was used as a positive control. While IL-6 induced strong STAT3 phosphorylation, no phosphorylation was observed in cell infected with either the WT or ΔespO (Fig 5C), suggesting that STAT3 phosphorylation is not a direct cell response to infection.
Secreted by immune cells, IL-22, which is a key cytokine needed to control C. rodentium infection [12], triggers STAT3 phosphorylation and expression of AMPs [36]. IL-22 also plays a role in cell proliferation [37]. We therefore tested if IL-22 secretion into supernatants of colonic explants was reduced in explants that were previously infected with either C. rodentium wild-type or ΔespO. This revealed that levels of IL-22 were 57% lower in mice infected with ΔespO (Fig 5D). In order to determine if the EspO influences expression of other cytokines, expression of Cxcl-1 produced by IECs and Ifn-γ produced by immune cells (e.g. natural killers, macrophages and T cells) was analyzed by qPCR in IECs and whole colonic tissue respectively. Whereas Cxcl-1 (Fig 5E) and Ifn-γ (Fig 5F) were induced during C. rodentium infection, no difference was observed between WT and ΔespO 8 DPI, suggesting that EspO selectively alters IL-22 related inflammation.
ILC3, at the early stage of the infection, and Th22 cells, at the later stage of the infection, are the major sources of IL-22. Moreover, neutrophils have also been shown to be an important source of IL-22 during C. rodentium infection [16]. To identify the immune cell types responsible for IL-22 secretion, colonic immune cells populations were analyzed 8 DPI by FACS. Unexpectedly, no difference was observed in the total number of colonic ILC3 and T cells (Figs 5G, 5I and S2), or in the number of IL-22 producing ILC3 and T cells (Fig 5H and 5J). Furthermore, no difference in other cell types producing IL-22 upon ex-vivo stimulation was seen between uninfected and infected colons (Fig 5K). This suggests that EspO did not influence the abundance of IL-22-producing cells, but instead affected the expression or release of IL-22 or the positioning of IL-22-producing immune cell with respect to the colonic mucosal surface. To test this hypothesis, IL-22 mRNA level was analyzed by qPCR and whole colonic tissue. Whereas IL-22 expression was strongly induced during C. rodentium infection, no difference was observed between WT and ΔespO 8 DPI (Fig 5L), suggesting that EspO alters either the release of IL-22 by immune cells or theirs localization in the tissue.
The proteomics analysis predicted that compared with WT infection, ΔespO induces reduced chemotaxis (activation score: -2.987; p-value 3.26 x 10−4), cell movement of granulocytes (activation score: -2.8; p-value 2.04 x 10−7) and immune response of neutrophils (activation score: -2.543; p-value 3.30 x 10−5) (Fig 6A). This is consistent with the predicted lower ROS production as well as the lower abundance of S100a8 and S100a9, which promote chemotaxis [38]. In addition, Mmp9, which digests extracellular matrix and opens tight junction allowing neutrophils transmigration [39] as well as Icam1, a neutrophil ligand which promotes their adhesion to the epithelial cells [40], were in higher abundance in IECs infected with WT and the ΔespO-pespO but to a lesser extent in ΔespO (Fig 6A).
Recently, neutrophils have been shown to be a main source of secreted IL-22 in the colon during C. rodentium infection [16] and colitis [31]. While we previously reported no global difference in total neutrophil numbers within inflamed tissue during infection with WT or ΔespO [23], supporting our IL-22 expression data (Fig 5K), neutrophil distribution within the inflamed colon was not assessed and could be altered after infection with ΔespO. To test this, colonic sections from mice infected with WT, ΔespO and ΔespO-pespO were stained with antibodies against Ly6G, C. rodentium and Hoechst stain (nuclei). Uninfected sections were used as control. While the number of granulocytes infiltrated into the tissue increased during infection (Fig 6B and 6C), the number was significantly lower in mice infected with ΔespO (68% reduction), suggesting that EspO affected neutrophils transmigration toward the site of bacterial attachment. These results suggest that by signaling inside IECs, EspO impacts on neutrophils chemotaxis to the site of C. rodentium colonization.
As CCH, neutrophil recruitment and AMPs affect the composition of the gut microbiota, we hypothesized the deletion of espO, which affects these parameters, will impact on the nature of dysbiosis induced by C. rodentium. To test this, we compared the composition of tissue-associated microbiota between mice infected with WT and ΔespO 8DPI using 16S RNA sequencing. Whereas C. rodentium infection induced a dysbiosis with a decreased of Bacteroidetes, Firmicutes and Tenericutes and a proliferation of Proteobacteria, no difference was observed at the Phylum level between mice infected with WT or ΔespO (Fig 7A). As expansion of C. rodentium in the colon has been linked to CCH [25], we tested if infection with ΔespO affects at the abundance of other Enterobacteriaceae (Fig 7B). Consistent with the similar level of shedding, the level of Citrobacter/Enterobacter were similar in mice infected with either WT or ΔespO (Fig 7B). Moreover, the abundance of other genera was similar in the infected mice, suggesting that proliferation of Enterobacteriaceae is independent of CCH and is not affected by the abundance of AMPs. As Reg3γ specifically kills Gram-positive bacteria [41], we analyzed the composition of tissue associated Firmicutes. Whereas, the abundance of most of the genera was similar in mice infected with either WT or ΔespO, the abundance of Aerococcus, Enterococcus and Anaerofuctis differ between the two infections, with the level in the ΔespO infected mice similar to that seen in the uninfected control mice (Fig 7C). This is consistent with a previous report showing that Reg3-/- mice have similar numbers of mucosa-associated bacteria belonging to the Gram-negative Bacteroidetes phylum and an increase of some Firmicutes (Eubacterium) [42].
Taken together, these results suggest that by signaling inside IECs, EspO impacts on chemotaxis of neutrophils to the site of C. rodentium adhesion, secretion of IL-22, phosphorylation of STAT3, cell proliferation and expression of AMPs, which leads to a specific alteration in the abundance of particular Firmicutes genera.
One of the main hallmarks of infection with C. rodentium is induction of tissue damage repair responses, i.e. elaboration of IL-22, secretion of AMPs and proliferation of transit amplifying cells [43]. In this study, we identified for the first time a T3SS effector, EspO, which once injected into infected IECs induces both processes. This raises the following question: what benefit C. rodentium gains from triggering tissue damage?
EspO is one of the smallest C. rodentium effectors (10 kDa), yet its deletion has profound effects on how the gut responds to infection. Induction of CCH is a complex event, involving a large number of biological pathways that are triggered directly by C. rodentium and by the host in response to the infection. Wnt signaling has been previously implicated in C. rodentium-induced CCH [44]. Roy et al. have shown that in Swiss–Webster mice, C. rodentium infection induces accumulation of β-catenin during the cell proliferation phase [5]. Our proteomics analysis did not reveal any hallmarks of the Wnt signaling pathway in infected C57BL/6 mice; the abundance of β-catenin was similar to the uninfected mice and none of the known Wnt target proteins were affected. Therefore, our data suggest that EspO delivered into IECs at the site of infection affects cell proliferation in a Wnt-independent fashion.
A triple C. rodentium mutant (ΔespH ΔcesF Δmap) has been recently reported to colonize the colonic mucosa at a significantly lower level than WT, which was mirrored by reduced CCH and Ki-67 staining. It was proposed that C. rodentium induces CCH as a means for oxygenation of the mucosal surface, which sustains the preference of the pathogen for aerobic metabolism and promotes pathogenesis [25]. However, the cause and effect relationship between the reduced colonization and lower CCH remained unresolved. Here, we show that that deletion of espO also caused significant reduction in CCH, yet colonization of the mucosal surface was unaffected, suggesting an independence of these two processes. Indeed, we have recently shown that C. rodentium infection diverts ATP production from the mitochondria to glycolysis, which was associated with production of phosphocreatine. In this study, we found that while colonizing at the same levels, WT and ΔespO trigger similar changes to metabolism in IECs, including disruption of the mitochondria, suggesting that C. rodentium is able to extract oxygen independently of CCH.
Induction of CCH seems to be influenced by inflammatory responses to infection. Indeed, deletion of aquaporin-3, an IECs’ basolateral water channel which mediates uptake of H2O2, has been shown to attenuate ROS responses, reduce CCH and impair C. rodentium clearance [45], resembling the ΔespO phonotype. The abundance of proteins linked to the production of ROS was lower in IECs infected with ΔespO. This suggests the existence of a strong link between CCH, ROS and bacterial survival. Moreover, mice infected with C. rodentium have an increased level of serotonin [46]. Interestingly, rectal injection of serotonin (5-hydroxytryptamine) in rat induced expression of Nox2, production of ROS, neutrophils recruitment and an increase of colonic wall thickness similar with C. rodentium infection in mice [47]. The exact relation between ROS production, CCH and neutrophils recruitment is still unclear, but it is likely linked to the secretion of chemokines [48].
To date, little is known about the interaction between neutrophils and C. rodentium. One reason for this is that these cells are short lived and mostly transcriptionally inactive upon their arrival to the site of infection. While the purity of our IECs was greater than 90%, our proteomes contained proteins reported to be neutrophils specific. This is mainly due to incorrect protein annotation, as most studies have examined protein contents in resting cells and not in the context of an inflamed tissue; nonetheless, we cannot exclude the possibility that during the purification process we co-purified proteins from neutrophil NETs. While NETs have been shown to be induced by E. coli, it is yet unknown if they play a role in controlling C. rodentium infection.
Neutrophils migration from the microcirculation to the tissues is a multistep process, which remains largely uncharacterized. It involves the Tlr4—Myd88 axis and requires chemokines (e.g. Ccl-3, Cxcl-1), matrix metalloproteinase (MMP) activation and lipids (lipid leukotriene B4). While the overall number of colonic neutrophils was similar between mice infected with WT or ΔespO, their tissue penetration was altered. We found similar level of Cxcl-1 expression in IECs, suggesting that EspO does not affect this step. Mice deficient in IECs’ Myd88 are unable to control C. rodentium infection, as they cannot induce epithelial repair response that maintains the protective barrier, production of neutrophil chemokines and an efficient adaptive immune response. While these mice succumb to C. rodentium infection, they do not show signs of CCH. Importantly, bone marrow transplantation from WT mice into Myd88-/- mice restored CCH [6]. In addition, the T3SS effector Tir, in a process dependent on Y451 and Y471, has been shown to modulate secretion of Cxcl-1 and recruitment of neutrophils 14 DPI [26]. Importantly, while exhibiting a similar level of shedding, mice infected with C. rodentium expressing Tir Y451A/Y471A presented reduced CCH 14 DPI. More recently, Cxcl-5 has been shown to mediate neutrophil infiltration during cancer or infection however, it requires further processing post secretion from epithelial cells. Mmp2 and Mmp9 have been shown to cleave Cxcl-5 and enhances it chemotactic activity [49]. Whereas we did not detect a significant increase of the abundance of Mmp2, the abundance Mmp9 increase 4-fold in IECs infected with WT, but only 2 fold in IECs infected with ΔespO. It is possible that by modulating the secretion of Mmp9, EspO indirectly modulates Cxcl-5 processing and neutrophil infiltration. Taken together, these results reinforce the link between neutrophils and CCH. However, the mechanism by which neutrophils contribute towards CCH remains unknown. Recent studies have shown that IL-22 can be secreted by neutrophils [16,31], however infection of IL-22 deficient mice with C. rodentium results in greater CCH [12].
Expression of colonic IL-22 is induced under inflammatory conditions such as infection and IBD. Indeed, many of the IL-22-regulated proteins belong to the IBD susceptibility genes [50]. Mucosal IL-22 has a dual role in mediating mucosal healing (through activation of activation of STAT3 and pro-proliferative genes) and combating pathogens (via expression of AMPs). IL-22 is an essential cytokine in the fight against C. rodentium infection [12]. IL-22 is produced by ILC3 prior to 8 DPI and by Th-17/22 T-cells after 8 DPI; importantly, the abundance of various IL-22 producing cells (e.g. ILCs, Th22) was similar between uninfected, WT- and ΔespO-infected mice 8 DPI. The abundance of neutrophils recruited to the site of infection was the only difference we observed between WT and ΔespO at 8 DPI. The putative role of neutrophils as a source of IL-22 is supported by a recent report showing that depletion of neutrophils with anti-Gr-1 neutralizing antibody during C. rodentium infection resulted in significant reduction in IL-22 productions. The reduced number of neutrophils following infection with ΔespO was consistent with lower abundance of Mmp9 and was mirrored by a global decrease in expression of genes encoding AMPs, with the exception of Dmbt1 and Iod1. However, while calprotectin, Reg3β and γ, Lcn2 and Lyz2 are regulated by IL-22, Dmbt1 is mainly induced by IL-27 [51] and expression of Ido1is triggered by interferon gamma [52], suggesting that EspO selectively modulates innate immune responses in IECs. It is important to note that even though the abundance of the IL-22 regulated antimicrobial proteins is reduced following infection with ΔespO, the residual levels are sufficient to mediate bacterial clearance (albeit delayed), unlike il-22 KO mice which succumbed to C. rodentium infection [12]. As ΔespO is shed for a longer period of time and triggers a novel immune response compare to WT (e.g. reduced abundance of T cells, neutrophils, IL-22 and C. rodentium-specific IgG) [23], the selective pressure for keeping EspO is not apparent. Our data show that by triggering damage repair responses EspO impacts on expression of antimicrobial peptides and the availability of trace minerals (e.g. Fe, Mg, Zn), which are required for survival by all living organism. While C. rodentium can resist nutritional immune responses and toxicity of antimicrobial peptides, these host responses affect the composition of the gut microbiota. Indeed, infection with ΔespO, which attenuates (yet not abolish) nutritional immunity, is associated with specific dysbiosis, particularly affecting the abundance of Firmicutes. Similarly, infection of mice lacking Mmp9, which is found in reduced abundance in ΔespO, with WT C. rodentium also resulted in increased abundance of Firmicutes, as well as Lactobacilli [53].
Although counterintuitive, the concept of bacterial T3SS effectors triggering inflammation is not new; indeed, multiple pathogens use this strategy as a means to disrupt both the epithelium and the microbiota in order to promote colonization [54]. T3SS effectors form a complex, yet robust, network that can resist sever perturbation (e.g. deletions). Alongside essential effectors (e.g. Tir, EspZ), C. rodentium encodes multiple accessory effectors, each making a refining contribution to the infection process. EspO is one such effector; its importance is emphasized by the fact that EHEC O157 contains two copies of the gene [22]. While we still do not know how EspO affects signaling (reflected by changing the abundance of 206 host proteins) in IECs, our data highlights a novel infection strategy involving activation of tissue healing responses as a means to trigger an advantageous nutritional immunity.
Wild type C. rodentium ICC169 (56) and ICC169 ΔespO (ICC1333) were grown at 37ºC in Luria–Bertani (LB) with necessary antibiotics. espO was amplified from ICC169 genomic DNA using primers GCTGGATCCTAGAAGAAGGAGATATACCATGCCATTGTCAATAAGAAA and GCTGTCGACTCAGGATTTATTTGAGTTATTAATCTCGGTC and was cloned in pACYC184 plasmid to generate pICC1379. The recombinant plasmid was confirmed by PCR and DNA sequencing (GATC Biotech).
All animal experiments complied with the Animals Scientific Procedures Act 1986 and UK Home Office guidelines and were approved by the Animal Welfare and Ethical Review Body (AWERB) at Imperial College London. The mouse experiments were performed under project licence PPL 70–8413.
Mouse experiments were designed in agreement with the ARRIVE guidelines [55] for the reporting and execution of animal experiments, including sample randomization and blinding.
Pathogen-free female C57BL/6 or Rag2-/- il2rg-/- mice (18 to 20 g) were inoculated by oral gavage with 200 μl of C. rodentium suspension (~5 x109 colony forming units (cfu)). Uninfected mice were mock treated with PBS (200 μl). The number of viable bacteria used was determined by retrospective plating. Number of viable bacteria per gram of stool was similarly determined by plating onto LB agar.
Terminal colon (0.5 cm) was fixed in 10% neutral buffered formalin and paraffin-embedded. Paraffin-embedded sections were then treated as previously described22. Anti-intimin (a gift from Professor Fairbrother, Montreal University), E-cadherin (BD Biosciences) and Ki67 (Thermo Scientific) were used as primary antibodies followed by secondary antibodies from Jackson ImmunoResearch. H&E stained tissues were evaluated blindly for CCH by measuring the length of well-oriented crypts from each section from all of the mice. Similarly, Ki-67 staining was assessed microscopically by measuring the distance from the bottom of the crypt to the last stained nuclei. Ki-67staining was expressed as a ratio over the total length of the crypt. Tissues were imaged with an Axio, images were acquired using an Axio camera, and computer-processed using AxioVision (Carl Zeiss MicroImaging GmbH, Germany).
For neutrophils staining, indirect immunofluorescence was performed on cryo-sections as previously described [56]. Chicken anti-intimin and rat anti-Ly-6G (RB6-8C5; Santa Cruz) were used as primary antibody follow by secondary antibodies from Jackson ImmunoResearch. Images were acquired using an AxioCam MRm camera and processed using AxioVision (Carl Zeiss MicroImaging GmbH, Germany).
Murine colonic tissues were fixed in 2.5% (vol/vol) glutaraldehyde/PBS and processed for electron microscopy. Samples for transmission electron microscopy were observed using a Phillips 201 transmission electron microscope at an accelerating voltage of 60 kV (Philips, United Kingdom).
IEC have been extracted as previously described [18]. Cell pellets were either kept frozen for proteomic analysis, Western blotting or RNA extraction.
HeLa cells (ATCC) were maintained in low glucose Dulbecco's Modified Eagle Medium (DMEM) supplemented with Heat-inactivated fetal calf serum (10% vol/vol) (FCS, Gibco), 2mM GlutaMAX (Invitrogen), and 0.1 mM nonessential amino acids at 37°C under 5% CO2 atmosphere. C. rodentium was cultured in Luria Broth at 37°C, 200 rpm with appropriate antibiotics for 8 h and then subculture (1/500) in DMEM with low glucose and grown overnight at 37°C without agitation in 5% CO2 incubator. After 3 h of starvation in DMEM only, cells were infected for 3 h. HeLa cells were incubated with IL-6 (Biovision, 50ng/ml) for 30 min. prior to analyzing cell extracts by Western blotting, using Hax-1 as a loading control.
IECs or HeLa cells were lysed in 50 mM Tris pH 7.4, 150 mM NaCl, 2 mM EDTA, 1% NP-40 and 1% SDS. Following gel electrophoresis and transfer, membranes were washed with PBS 0.1% Tween, blocked in TBS (0.1% Tween, 3% BSA, 0.5% gelatin) and probed with specific antibodies overnight. Blots were then incubated with secondary antibody (Jackson ImmunoResearch), followed by EZ-ECL assay, according to the manufacturer's instructions (Geneflow). Chemiluminescences were detecting using a Chemidoc (Biorad). Polyclonal anti-GADPH (Abcam), anti-Hax1 (Genetex), anti-Stat3 and monoclonal anti- phopho-Stat3 (Cell Signaling) were used to detect the different proteins.
Enterocytes mRNAs were isolated using a RNeasy minikit according to the manufacturer's instructions (Qiagen). Samples were treated with RQ1 DNase I and reverse transcription was perform using RQ1 DNase I according to the manufacturer's instructions (Promega). Targeted genes were amplified with specific primer pairs listed in S3 Table, using a 7300 Applied Biosystems instrument under standard cycle conditions for PowerUp SYBR Green Master Mix (Thermo Fisher). Changes in gene expression levels were analyzed relative to the controls (uninfected samples), with GAPDH as a standard, using the ΔΔCT method.
Last centimeter of distal part of the colon has been excised, weighted and washed thoroughly with RPMI medium with 100ug/ml of streptomycin and 100U/ml penicillin. Tissues have been then culture in RPMI (10%FCS, P/S, L-Glu) for 2h and placed in fresh media (0.1 ml / 10 mg of tissue). After 24h, the supernatant has been collected and centrifuged 15 min at 15000 rpm. IL-22 was then measured using Mouse IL-22 DuoSet ELISA (R&D Systems) according to manufacturer's instructions.
Fresh stool pellets were resuspended in PBS-0.1% Tween20 at a w/v ratio of 100 mg of stool per 1 ml PBS. Samples were left shaking for 20 min, centrifuged at maximum speed for 10 min before freezing the supernatant. Fecal LCN-2 was then measured using Mouse Lipocalin-2/NGAL DuoSet ELISA (R&D Systems) according to manufacturer's instructions.
IEC pellets isolated from WT, ΔespO and ΔespO-pespO C. rodentium infected and uninfected mice were analyzed as previously described [18]. Only unique peptides were used for quantification, considering protein groups for peptide uniqueness. Peptides with average reported S/N>3 were used for protein quantification. The IEC obtained from uninfected mice were used as controls for log2 ratio calculations. Differential expression p-values were computed based on a single-sample t-test. Specificity thresholds used to characterize the EspO-specific IECs proteins were define as p-value < 0.05 (Student's t-distribution) and log2 ratio > 0.59 or < -0.59 (equivalent to 1.5 fold change) compare to protein abundance following WT and ΔespO-pespO infections. Specifically regulated protein were uploaded in Ingenuity Pathway Analysis (IPA) (Qiagen) platform. Trends of activation/inhibition states of the enriched functions and regulators were inferred by the calculation of a z-score (-2 < z-score > 2).
Colons were collected from mice and DNA was isolated using PowerSoil DNA Isolation Kit (MO BIO Laboratories). For 16S amplicon pyrosequencing, PCR amplification was performed spanning the V3and V4 region using the primers 515F/806R of the 16S rRNA gene and subsequently sequenced using 500bp paired-end sequencing (Illumina MiSeq). Reads were then processed using the QIIME (quantitative insights into microbial ecology) analysis pipeline with USEARCH against the Greengenes database.
Cells were isolated from LI LP as previously described (53) using a digestion solution containing 25 μg/mL liberase TL (Roche) and 25 μg/mL Dnase1 (Sigma Aldrich). For cytokine ex-vivo stimulation, 1–5 x 106 cells were incubated at 37°C with IL-1β (R&D Systems; 100 ng/mL), IL-23 (R&D Systems; 100 ng/mL), PMA (Sigma-Aldrich; 50 ng/ml), Ionomycin (Sigma-Aldrich; 2.5μg/ml) and BD GolgiPlug (BD Biosciences) in 10% FCS DMEM (Gibco) for 3h.
Single-cell suspensions were stained with Flexible Viability Dye eFluor 506 (eBioscience) and blocked with FcR Blocking Reagent (Miltenyi) for 15 minutes followed by 30 minutes of surface antigens staining with a combination of fluorescently conjugated monoclonal antibodies (from BD Biosciences, eBioscience and Biolegend) on ice. For experiments involving intranuclear transcription factor staining, cells were fixed, permeabilized and stained using Fix & Perm Buffer Kit according to the manufacturer’s instructions (BD Biosciences). For intracellular cytokine staining, cells were fixed, permeabilized and stained using Fix/Perm kit according to the manufacturer’s instructions (BD Biosciences). All the samples were acquired on a custom-configuration LSR Fortessa (BD Biosciences) and the data were analyzed on FlowJo10 software (TreeStar).
GraphPad Prism software was used for all statistical calculations. Statistical test used was Mann-Whitney compared to controls (or as indicated in the figure). p-values < 0.05 were considered significant. For the microbiota, p-values were FDR corrected using Benjamini and Hochberg method.
|
10.1371/journal.ppat.1004359 | The Effects of Vaccination and Immunity on Bacterial Infection Dynamics In Vivo | Salmonella enterica infections are a significant global health issue, and development of vaccines against these bacteria requires an improved understanding of how vaccination affects the growth and spread of the bacteria within the host. We have combined in vivo tracking of molecularly tagged bacterial subpopulations with mathematical modelling to gain a novel insight into how different classes of vaccines and branches of the immune response protect against secondary Salmonella enterica infections of the mouse. We have found that a live Salmonella vaccine significantly reduced bacteraemia during a secondary challenge and restrained inter-organ spread of the bacteria in the systemic organs. Further, fitting mechanistic models to the data indicated that live vaccine immunisation enhanced both the bacterial killing in the very early stages of the infection and bacteriostatic control over the first day post-challenge. T-cell immunity induced by this vaccine is not necessary for the enhanced bacteriostasis but is required for subsequent bactericidal clearance of Salmonella in the blood and tissues. Conversely, a non-living vaccine while able to enhance initial blood clearance and killing of virulent secondary challenge bacteria, was unable to alter the subsequent bacterial growth rate in the systemic organs, did not prevent the resurgence of extensive bacteraemia and failed to control the spread of the bacteria in the body.
| The bacterium Salmonella enterica causes gastroenteritis and the severe systemic diseases typhoid, paratyphoid fever and non-typhoidal septicaemia (NTS). Treatment of systemic disease with antibiotics is becoming increasingly difficult due to the acquisition of resistance. Licensed vaccines are available for the prevention of typhoid, but not paratyphoid fever or NTS. Vaccines can be either living (attenuated strains) or non-living (e.g. inactivated whole cells or surface polysaccharides) and these different classes potentially activate different components of the host immune system. Improvements in vaccine design require a better understanding of how different vaccine types differ in their ability to control a subsequent infection. We have improved a previously developed experimental system and mathematical model to investigate how these different vaccine types act. We show that the inactivated vaccine can only control bacterial numbers by a transient increase in bactericidal activity whereas the living vaccine is superior as it can induce an immune response that rapidly kills, then restrains the growth and spread of infecting bacteria.
| Salmonella enterica causes systemic diseases (typhoid and paratyphoid fever) [1], food-borne gastroenteritis and non-typhoidal septicaemia (NTS) [2]–[4] in humans and in many other animal species world-wide. Current measures to control S. enterica infections are sub-optimal. The emergence of multi-drug resistant strains has reduced the usefulness of many antibiotics [5]–[6]. Prevention of infection of food-production animals by implementation of biosecurity or hygiene measures is expensive and is undermined by increased free-range production.
Vaccination remains the most feasible means to counteract S. enterica infections. There is an urgent need for improved vaccines against typhoid fever and there are currently no licensed paratyphoid or NTS vaccines [7]. To attain a high level of protective immunity against systemic infections with virulent strains of Salmonella in susceptible hosts it is necessary to induce both antibody responses and T-helper type 1 (TH1) cell-mediated immunity [8]. This is due to the fact that intracellular control requires TH1 immunity whereas antibodies can only target the bacteria in the extracellular compartment (reviewed in [9]–[10]).
New generations of live attenuated vaccines have been constructed in the last two decades and are currently being evaluated in field trials. These vaccines mimic the course of natural infection and are more protective than previous ones, but we do not understand the mechanisms responsible for this [11]–[12]. There is also a recent trend towards the development of non-living vaccines against S. enterica enteric diseases for humans and other animals. Current non-living vaccines are based on inactivated whole cells and surface polysaccharides (e.g. Vi polysaccharide and Vi conjugate vaccines for humans) [13]–[14] and subunit protein-based vaccines are being considered. However, non-living S. enterica vaccines vary greatly in their protective ability [15]–[19]. Vaccine design and selection is still largely an empirical process. This is due to our insufficient understanding of how vaccine-induced immune responses impact precisely on the dynamics of a secondary infection in terms of bacterial division, killing, spread and persistence in the tissues.
Interactions between infectious agents and their hosts occur in diverse environments and over a range of scales: from initial contact at the single cell level; spread throughout various compartments of the host; and between hosts at a population level. Intervention strategies to control infections can interfere with the host-pathogen relationship at all these levels so understanding the dynamics of infections at all scales is important. Mathematical approaches have been extensively used to model infection dynamics on the population level but until relatively recently within-host dynamics have been measured rather crudely, typically by monitoring total pathogen loads in a host or its organs. These measures cannot disentangle the relative contributions of pathogen replication and death to overall growth. For example an unchanging total pathogen load could be due to both replication and killing occurring in balance, or to a lack of replication and no killing. Attempts have been made to measure bacterial pathogen division rates within hosts and cells by techniques such as using non-replicating elements introduced into the bacteria, or dilution of a fluorescent marker that is not expressed within the cell in vivo [20]–[23] but these are often confounded by uncertainties concerning the dynamics of these elements in complex in vivo milieu and the potential effects on the phenotype of the pathogen being investigated.
We have established and used a research approach based on the tracking of bacterial subpopulations at multiple body sites to study infections of mice and to capture the parameters that govern the intra-organ and inter-organ infection processes. The system employs simultaneous infections with individually identifiable wild-type isogenic tagged strains (WITS) the relative proportions of which can be determined by quantitative real-time PCR (qPCR) [24]–[25] or other molecular techniques. One of the systems to which this technology has been applied is the systemic phase of experimental Salmonella infections in mice, a well-established model for invasive disease [25]. In this system, observation of both total bacterial loads and the changes in WITS population structure enabled inference to be drawn on the rates of bacterial replication, killing and spread between organs. This approach allowed us to capture the key dynamic traits of the primary infection process in unimmunised animals and to explore the impact of innate immunity on the infection. In this previous study, we determined that the infection process is multiphasic, with an initial phase of rapid bacterial replication and killing of part of the inoculum by the innate host response mediated by reactive oxygen species (ROS), followed by a phase of growth and intra-organ spread with the stochastic selection of individual bacterial subpopulations. Later the infection moves to a phase in which bacteraemia and the mixing of different subpopulations of bacteria in the organs occurs [25].
In the present study we refined our framework and used it to understand to what extent and at what stages of the infection different vaccine types and different branches of the vaccine-induced host immune response restrain intracellular division and enhance bacterial killing, whether there are changes in the patterns of local or systemic spread in vaccinated/immune animals, whether immunity acts simultaneously or equally on the global bacterial population, and whether there are intra- or inter-organ differences and heterogeneous traits in the control of individual bacterial subpopulations.
Vaccination with either live or killed Salmonella is known to increase the rate of blood clearance of an intravenous challenge, result in a bias of segregation of bacteria towards the liver rather than the spleen, and to enhance the reduction in bacterial loads in the spleen and liver that occurs within the first hours [26]–[30]. To investigate the contributions made by bacterial killing and growth to these changes in total bacterial numbers we performed two separate experiments vaccinating mice with either live attenuated Salmonella Typhimurium (STm) SL3261 (Live Vaccine group: LV) or acetone-killed STm SL1344 (Killed Vaccine group: KV). Naive control and vaccinated animals (n = 9 or 10, see Table S1) were then challenged with ∼300 colony forming units (CFU) of STm WITS and net bacterial numbers and WITS proportions were monitored over a period of 72 h post-challenge (timepoints 30 min, 24, 48, 72 h). The challenge dose was chosen as the minimum to yield reliable colonisation of both the liver and spleen of vaccinated animals as determined in pilot experiments described in the Methods section.
Vaccination with either type of vaccine resulted in accelerated clearance of the WITS challenge inoculum from the blood with no circulating bacteria detected at 30 min post-challenge (Figure 1). In contrast the majority of naive animals still harboured bacteria in the blood at this time. By 6 hr post-infection (p.i.) the majority of vaccinated and non-vaccinated animals had cleared the challenge dose from the blood and at 24 hr no bacteraemia was observed in any of the mice tested.
At 24 h post-challenge, animals vaccinated with either preparation had a lower bacterial load in the organs compared to naive controls but the kinetics of this reduction differed somewhat between vaccine types (Figure 1): at 6 h p.i. bacterial loads in both organs of mice immunised with the LV were lower than in the controls whereas for the KV loads were only significantly lower in the spleen at this timepoint; by 24 h p.i. bacterial loads in both vaccinated groups were lower than in the controls. Subsequently between 24 and 72 h after challenge, mice immunised with the KV and the unimmunised control mice both allowed a rapid increase in bacterial numbers (∼10-fold per day), with no statistically significant difference between these groups, and a resurgence of bacteraemia from 48 h p.i. In clear contrast, net bacterial growth was much slower in the tissues of mice vaccinated with live bacteria and bacteraemia was observed in only one of nine mice tested and even then at a very low level; Salmonella were absent from the spleens in 4/9 animals by 72 h p.i.
Next we analysed the population structure of WITS in the samples to determine when bacterial death and inter-organ spread occurred. The challenge inoculum contained equal numbers of each of 8 WITS. We first considered the number of distinct WITS per mouse that were present in both organs (liver plus spleen), in one organ only, or in neither of the organs (Figure 2). In the absence of bacterial killing we would expect all individual WITS to be represented in one or both organs. In contrast, if bacterial killing was significant then the number of strains present in only one organ, or absent from both would increase. Based on the disappearance of WITS from either organ, we predicted that bacterial death was higher during the first 24 h in animals vaccinated with either vaccine compared to naive mice. In the naive and KV groups, between 48 and 72 h p.i., we observed increases in the numbers of individual WITS that were simultaneously present in both spleen and liver (Figure 2). This was suggestive of bacterial transfer between organs and coincided temporally with detection of bacteraemia in these mice from the 48 hr time-point (Figure 1). The number of WITS that were present or absent in the organs of mice immunised with the LV remained approximately constant between 24 h and 72 h post-challenge indicating that neither transfer between organs nor substantial bacterial death had occurred in this group of animals.
In addition to determining the presence and absence of WITS from the organs, we measured the proportion of each WITS present in the spleen or liver (Figure 3). For the unvaccinated animals at 30 min each WITS could be found in both the spleen and liver with a proportion of approximately 1/8, i.e. the population structure was similar to that in the inoculum (which contained equal numbers of the eight WITS). By 6 h p.i. and until the 24 h time-point, the population structures (i.e. presence, absence and proportions of each WITS) in each organ of naive mice had diverged (consistent with stochastic killing of bacteria within the organs). By 48 h, coincident with the onset of bacteraemia, the hepatic and splenic WITS population structures became more similar – presumably due to transfer of bacteria between the organs via the blood. Between 48 and 72 h p.i. the liver and spleen populations became nearly homogenous within each individual unvaccinated mouse (i.e. a given WITS was present in the same proportion in the spleen and liver) (Figure 3; Table 1).
For animals vaccinated with either LV or KV, there was a bias in segregation of bacteria towards the liver, shown by compression of the points around the vertical in Figure 3, and enhanced bactericidal activity resulted in a number of WITS being absent from one or both organs at 6 hr post-challenge. From 48 h the WITS population structures in the KV and LV groups diverged: the KV immunised animals showed a similar pattern to the unvaccinated controls with highly correlated WITS organ populations in all animals by 72 h (Figure 3; Table 1), and with an increase in the number of WITS simultaneously present in both the liver and spleen (Figure 2); in contrast, for the LV group there was no increase in the co-occurrence of WITS in both organs (Figure 2) and highly correlated WITS organ populations were only observed in a minority of animals (Figure 3; Table 1), therefore indicating that in most of these animals significant inter-organ spread had not occurred up to 72 h post-challenge. The bias in bacterial numbers towards the liver that was observed earlier in the infection was still present in the LV group at this late timepoint, presumably as a consequence of the restraint of both bacterial growth and spread in this group resulting in populations similar to the early timepoints. In contrast for the KV group this bias had disappeared, likely a consequence of uncontrolled spread and growth of bacteria in these animals.
The enhanced reduction in total bacterial numbers, the fluctuations in the WITS population structure and the disappearance of WITS from the spleen and liver indicated that the overall dynamics of bacterial division and death are different in naive mice and in vaccinated animals in the early stages of the infection. Bacterial dynamics were described using a stochastic model that keeps track of the number of copies of a single WITS in the blood, liver and spleen simultaneously [9], [25]. The parameters of the model (inoculum size, rates of bacterial replication, killing and transfer from the blood to the organs) were all estimated by fitting the model to the data from each experimental group at 0.5, 6 and 24 h post-inoculation, using maximum likelihood. In order to allow for the possibility of early killing or inactivation of bacteria before they start to colonise the organs, we estimated an effective inoculum size consistent with the total number of bacteria 30 min p.i. in each experimental group. By comparing these values with the average inoculum doses actually used we obtained an estimate of the fraction of bacteria eliminated in the very early stage of infection (i.e. within 30 min of inoculation): this fraction was highest (44%) in the LV group (Table 2). Biologically this could be a consequence of bacterial killing and/or entry into a non-replicative state [21], [31]–[32]. In our model the remaining bacteria settle into the liver and spleen where they undergo a process that involves replication and killing. Comparing the model estimates for these processes between groups (Table 2) we see that for both vaccine types, the model captures the enhanced blood clearance and increase in the proportion of bacteria going to the liver that was observed in the data.
Estimates of the intra-organ replication and killing rates (Figure 4) were very similar in naive mice and in mice immunised with the LV in the first 6 h, whereas higher rates of bacterial replication and killing in the liver were estimated in mice immunized with the KV during this initial period of the secondary infection, resulting in a more rapid net reduction in bacterial numbers. All groups exhibited a large reduction in both the killing and replication rates in the liver after 6 h, with replication rates marginally higher than killing rates. The estimated killing and replication rates were lower in the spleen than in the liver for the first 6 h of the secondary infection, but became similar in both organs between 6 and 24 h. In all cases, the fitted models predicted distributions of bacterial loads and WITS abundancies similar to the data (see Supporting Information Text S1)
Thus, based on variations in the population structure of WITS, our model predicts that the reduction in the total numbers of viable bacteria that is seen in the first 24 h in both groups of immunised mice is largely due to enhanced bacterial killing in the first few hours after challenge. More specifically, the models predict that in mice immunised with the LV there was a substantial reduction (by 44%) in the number of viable bacteria compared to naive mice. This is predicted to occur before the bacteria can be detected in the spleen and liver. The models predict that once the bacteria are in the spleen and liver of the LV-immunised mice they are subject to enhanced bactericidal activity in the spleen and normal bactericidal activity in the liver as compared to naïve mice. For the KV, in contrast, we did not observe any loss of viable bacteria by 30 min p.i., there was then more intense bactericidal activity (combined with faster bacterial replication) in the liver than any other group. Between 6 and 24 h p.i. bacterial replication and killing rates decreased markedly in animals immunized with either vaccine type showing that the control of bacterial numbers after 6 h proceeds mainly by bacteriostatic mechanisms.
Taken together the model estimates, measures of WITS co-occurrence, and correlation between the organs show that immunisation with KV enhances blood clearance, and increases the killing rates in the organs (up to 24 h). Subsequently bacteriostatic effects predominate, but these are not enhanced by the killed vaccine over what is seen in naive animals, as shown by similar increases over time in bacterial numbers in naive and KV mice. In contrast, the immune response induced by the LV rapidly inactivated (within 30 min) a large fraction of the challenge dose, enhanced clearance of the bacteria from the blood and their transfer into the organs and subsequently exerted a stronger bacteriostatic effect which restrained bacterial growth more than in naive or KV animals.
Animals immunised with a live vaccine can control the growth of a secondary challenge in the spleen and liver. CD4+ and CD8+ T-cells are known to play a key role in immunity conferred by live vaccination [8], [33]. We therefore wished to determine how and at which time T-cell dependent immune functions impact on the dynamics of a secondary challenge in terms of control of bacterial division, enhancement of death and restraint of spread between systemic sites. LV-immunised mice were treated with either anti-CD4 plus anti-CD8 antibodies, or with control immunoglobulins two days before and after challenge with ∼300 CFU of WITS. Pilot experiments showed that differences in bacterial loads between T-cell positive and T-cell negative animals became significant only by 72 h post-challenge therefore we extended the time-course of the experiments to capture these differences; groups of mice (n = 5 or 10, see Table S1) were sacrificed at 30 min, 6, 24, 48, 72, 96, 120 and 144 h and total bacterial numbers and WITS population structures were assessed in the spleen, liver and blood.
Depletion of T-cells had no effect on the bacterial loads of the challenge organisms up to 48 h, but subsequently the groups diverged with the LV immunised T-cell depleted animals having higher loads and from 96 h bacterial numbers in this group were increasing at a rate similar to that seen in naive animal (Figure 5). Subsequently bacterial loads in T-cell depleted mice were consistently higher than in LV-immunised control animals and a resurgence of bacteraemia was detected in the T-cell depleted mice. Bacterial counts in the control mice immunised with the LV increased more slowly than in LV-immunised T-cell depleted mice.
Observation of WITS co-occurrence in the spleens and livers (Figure 6) showed similar overall bactericidal activity (absence of individual WITS from both organs) up to 96 h post-challenge in both control and T-cell depleted mice. Fitting the mathematical model to this dataset showed that while estimated killing and replication rates over the first 6 h were somewhat higher than in the previously conducted LV experiment, there was little difference between the T-cell depleted and control (T-cell positive) mice (see Supporting Information Text S1). By 120 h, the number of distinct WITS in the organs decreased in LV immunised control mice, indicating killing of bacteria had started to occur in LV. Conversely there was no evidence of bacterial killing in T-cell depleted animals and the number of WITS simultaneously present in both organs increased markedly from 72 h post-challenge which was also indicative of inter-organ spread. The relative proportions of individual WITS in this group were similar in the spleens and livers of T-cell depleted mice (Figure 7; Table 3) confirming that inter-organ spread was significant in these animals. A resurgence of bacteraemia was observed at later time-points in vaccinated T-cell depleted animals, this could not have been due to a defect in production of antibody following T-cell depletion as there was no difference in circulating Ig levels between T+ and T− animals post-secondary challenge (Figure S1). Unfortunately it was not possible at this time to quantify these processes using our mathematical model due to the increased complexity of the system at later time points.
We also observed highly correlated WITS organ populations in the majority of LV-immunised T-cell positive mice later in the infection, but only from 120 h (Table 3) indicating that in these mice bacteria were able to spread between organs despite the very low-grade bacteraemia. By considering mice individually we see that highly correlated populations were observed in those animals with higher bacterial loads and that only one WITS made up the majority of the population in these mice (Figure S2). T-cell depleted animals also showed a limited number of WITS predominating at 144 h although more individual WITS were present in each sample presumably as bacterial killing was not significant in these animals. KV-immunised mice showed the same effect by 72 h post-challenge, as did naive mice albeit to a more limited extent.
Taken together these data show that neither the LV-induced initial rapid bacterial kill nor the subsequent bacteriostatic mechanisms that predominate early in the infection were T-cell dependent and that T-cell dependent bacterial killing became observable by 120 h p.i.
Here we show that previous immunisation with a live attenuated Salmonella vaccine results in the rapid kill of a great proportion of the challenge inoculum and additionally enhanced bacteriostasis until the robust LV-induced T-cell response controls the infection by bactericidal mechanisms. LV also controls secondary bacteraemia to extremely low levels and delays and reduces inter-organ spread. Immunisation with a whole-cell killed vaccine enhanced blood clearance and only transiently (within the first 24 h after challenge) increased the rates of bacterial killing within the organs. Later in the infection, the KV neither controlled the net growth of the virulent challenge in the spleen and liver nor was able to induce clearance of the bacteria from the tissues. The KV did not control the rapid resurgence of bacteraemia and did not control inter-organ spread despite its ability to induce anti-Salmonella antibodies.
Mixed infections with individually identifiable wild-type isogenic tagged strains (WITS) have recently been used to gather information on the population dynamics of bacteria within hosts [24]–[25], [34]. Use of multiple isogenic strains facilitates more refined measurement of subpopulations than using a more limited number of strains selectable with different antibiotics [35]–[37] as well as limiting the possibility of phenotypic variation between competing strains with different genotypes.
In this study we used WITS to study the in vivo population dynamics of S. Typhimurium during secondary challenge following vaccination. In the early phase of the challenge (up to 24 h) we estimated bacterial replication and killing rates by mathematical modelling. For later stages in the infection we could infer the times at which inter-organ spread and bacterial killing became observable by tracking the presence or absence of WITS in the organs and calculating the correlation between the abundances of WITS in the organs. We chose the parenteral route to investigate the secondary challenge dynamics in mice immunised with either a killed vaccine or a live attenuated strain. This route enabled us to study specifically the dynamics that underpin control of the bacteria in the systemic compartment. Systemic control of the infection - in other words the suppression of bacterial growth, restraint of dissemination in the spleen, liver and blood, and clearance of the bacteria from the tissues - is an absolute requirement for both host survival in systemic Salmonella infections and for the successful elimination of the infection. We used secondary challenge inoculum sizes of ∼300 CFU as the lowest dose that consistently prevented rapid clearance in immunised animals. This dose is similar to a recent estimate of a rate of migration of 298 CFU/day of STm from the gastrointestinal tract into the cecal lymph node following oral infection of mice [34].
We used a host-pathogen combination where a virulent strain was injected as the challenge organism into innately susceptible mice. This stringent combination always results in lethal infections in naïve animals. Bacterial numbers initially decline following challenge due to reactive oxygen radical mediated killing which exceeds division, but subsequently killing becomes negligible and intracellular replication results in a constant and exponential increase in bacterial numbers in the spleen and liver until death of the animal [25]. In this stringent model we showed that the LV induces an immune response that controls a secondary infection in four phases: rapid clearance of bacteria from the blood into the organs coupled with an initial very rapid (within 30 min) inactivation of a large fraction of the challenge dose; a period (∼6 h) of relatively rapid bacterial replication and killing essentially equivalent to that in naive animals; a T-cell independent bacteriostatic phase that lasts for about three days and restrains bacterial growth more than in naive controls; and a subsequent phase in which T-cell dependent bacterial killing becomes significant. Assessment of net bacterial numbers alone would not have distinguished the time of transition between these latter two phases as total bacterial loads were relatively stable over the time period under investigation. The very rapid initial inactivation of the challenge could be a consequence of bacterial killing, or entry into a non-replicative state [21], [31]–[32] that cannot be recovered from following direct plating of organ homogenates. Any early bacterial killing is unlikely to be the result of serum bactericidal activity as mouse serum lacks such activity against Salmonella due to a reduced ability to deposit C3 [38]–[39]. We confirmed this experimentally by assessing the survival of S. Typhimurium SL1344 in serum collected from mice immunised with the LV (Figure S3). Entry into a non-replicative state is also unlikely to account for the large drop in viable bacterial numbers seen as non-replicating salmonellae have not been observed at high levels in the spleen, and resume growth upon culturing [40].
We showed that the whole-cell killed vaccine did not induce the very rapid inactivation of the challenge dose seen in mice immunised with the LV, but did result in a decrease in total bacterial loads as compared to naive controls by 6 h post-challenge. This was primarily due to rapid blood clearance leading to more time being available for bactericidal mechanisms to exert their effect coupled with an increase in killing rate in the liver. While it has been known for many years that living and non-living vaccines can reduce total bacterial loads within the first hours of a re-infection [26]–[30] it is now clear that these reductions proceed with markedly different kinetics. After the first 24 h for animals immunised with the KV there was no additional restraint of bacterial growth, enhancement of killing or control of bacteraemia and the infection proceeded as in naive animals despite the presence of circulating antibodies. We also saw a decrease in the number of WITS absent simultaneously from both livers and spleens at later time-points in KV-immunised animals. This could have been due to re-emergence of bacteria from sites that we did not sample in this study, for example the bone marrow, during the haematogenous spread phase of the infection. This phenomenon was not observed in LV-immunised animals.
Vaccination with live attenuated Salmonella is known to induce higher levels of IgG2a (IgG2c in C57BL/6 mice) as compared to a killed vaccine [16], which we also confirmed in our model (Figure S4). Opsonisation of Salmonella with immune serum increases both the uptake of bacteria by macrophages and bacterial killing via the production of reactive oxygen intermediates (ROI) by the phagocyte NADPH oxidase (phox) [41]. phox also possesses bacteriostatic functions [42], which we showed were important in naive animals in vivo after its initial bactericidal effect [25]. Macrophage uptake of IgG-opsonised Salmonella is primarily mediated by FcγRI which binds IgG2c [41]. It is therefore possible that both the LV-induced immediate bacterial killing and extended bacteriostasis are due in part to enhanced production of anti-Salmonella IgG2c. Reactive nitrogen intermediates (RNI), produced by iNOS, are also known to have bacteriostatic activity both in vitro and in vivo [43]–[44] and while opsonisation has not been reported to increase RNI production by macrophages in vitro [41] it is possible that such an effect does occur in vivo.
In this study we observed that the bacterial population structures in naive mice, and in KV and LV-immunised mice eventually become homogeneous between the livers and spleens indicating that inter-organ mixing of WITS occurs during the infection process. However, mixing occurs earlier in naive mice, in mice immunised with the KV and in LV-immunised T-cell depleted animals compared to control LV-immunised wild-type mice. This inter-organ mixing coincided with bacteria becoming detectable in the blood suggesting that haematogenous spread is responsible for transfer of bacteria between organs. In the majority of the LV-immunised mice inter-organ mixing was seen in the absence of detectable bacteraemia which leads us to speculate that in these animals very few bacteria are released into the blood at any time due to a more efficient control of bacterial release from the infection foci.
In the KV-vaccinated mice, as the secondary infection progressed a small number of individual WITS predominated within the overall population structure. This phenomenon was also observed in naive control mice in this study although to a lesser extent than was previously seen when naive animals were challenged with a lower dose (∼90 CFU) [45]. Given that we do not observe bacterial killing during the net growth phase of WITS in the tissues in these mice (from 24 h p.i. onwards), the increased prevalence of some WITS is likely to be due to a relatively faster expansion of some WITS populations over others within each organ. Salmonella adapt in vivo to enter a state in which net intra-organ growth is faster [10] and individual foci of infection arise from distinct clonal populations [46]. We speculate that the expansion of a limited number of WITS populations is a consequence of a loss of control of the infection at a small number of foci (each of which contains a single WITS) resulting in release of bacteria that then establish secondary foci with rapidly growing bacteria [9], [25], [47]–[48].
A similar prevalence of a low number of distinct WITS was also seen in the later stages of the secondary infection in mice immunised with LV suggesting that loss of focal control at a limited number of sites also occurs in these animals. However in these mice we observed a T-cell dependent bactericidal activity of the immune system from 120 h onwards and therefore it is possible that bacterial killing is also responsible for the stochastic prevalence of some bacterial subpopulations over others.
Overall our study indicates the superiority of a live attenuated vaccine over a whole-cell killed one in controlling the infection process by suppressing bacterial growth, exerting bacterial killing and achieving clearance of the inoculum from the tissues. The latter function is ascribable to the T-cell response induced by the LV since it can be abrogated by depletion of T-cells before challenge. This work indicates that a KV is inadequate to control a secondary infection in a very stringent host pathogen combination. New generations of Salmonella vaccines such as polysaccharide and subunit vaccines for typhoid and NTS are currently being considered and empirically tested. These preparations rely on the induction of antibodies for their protective activity so the optimisation of these vaccines and their delivery would benefit from improvements in their ability to induce T-cell mediated immunity.
S. enterica serovar Typhimurium (STm) WITS strains 1, 2, 11, 13, 17, 19, 20 and 21 which have been described previously [25] were made by inserting 40 bp signature tags and a kanamycin resistance cassette between the malXY pseudogenes of STm JH3016 [49], a gfp+ derivative of wild-type virulent SL1344. They are phenotypically wild-type for growth in broth and infectivity for mice. The live attenuated STm SL3261 aroA [50] strain was used for all immunizations. Bacteria for infection were grown for 16 h at 37°C in L-broth (LB) without aeration and diluted in phosphate-buffered saline (PBS) prior to inoculation. Enumeration of bacteria was by plating dilutions on LB agar plates.
All experiments involving animals were conducted under project licences approved by the University of Cambridge Animal Welfare and Ethical Review Body, granted by the United Kingdom Home Office (licence numbers PPL 80/2135 and PPL 80/2572), and performed in observance of licensed procedures under the United Kingdom Animals (Scientific Procedures) Act 1986.
Female age-matched C57BL/6 mice were purchase from Harlan Laboratories and used when over 9 weeks of age. Live bacteria for parenteral immunization were prepared from a 16 hr static culture of STm SL3261, diluted 1/100 in PBS and administered by i.v. injection into the tail vein in 200 µl aliquots (∼106 CFU/mouse). Actual inoculum dose was determined by plating dilutions. We confirmed that this immunisation regime elicited anti-Salmonella antibodies (Figure S1) and elicited a TH1-type memory response (Figure S5). Challenge with WITS was performed three months after immunisation with the LV by which time the primary infection was cleared, as determined by sampling livers and spleens of mice in pilot experiments (n = 2), and demonstrated previously [51].
Acetone killed wild-type bacteria for immunization were prepared as previously described [16]. Briefly, bacteria from a 100 ml culture of STm SL1344 grown for 16 h at 37°C with aeration in tryptic soy broth (Oxoid) were harvested by centrifugation at 3220 g, resuspended in 100 ml PBS and viable counts determined by plating. Following three sequential washes in acetone, bacteria were harvested, acetone completely removed by evaporation, cells resuspended in PBS to an equivalent concentration of 5×1010 CFU/ml and stored in aliquots at −20°C or below. Immunization was by administration of two doses of 108 cells given subcutaneously in the back, 3-weeks apart. The subcutaneous route was chosen as in our experience this results in a strong antibody response with the killed vaccine whereas i.v. immunisation does not. Regardless of route the KV is known to be unable to induce protective T-cell immunity against virulent challenge. We confirmed that this immunisation regime elicited anti-Salmonella antibody, although there was a somewhat lower production of IgG as compared to immunisation with the LV, due to a decrease in levels of IgG2c (Figure S4). Challenge with WITS was performed 6 weeks post-primary immunisation.
For T-cell depletion experiments 500 µg each of rat IgG2b anti-mouse CD4 and anti-mouse CD8 (prepared by Harlan Bioproducts from hybridomas YTS 191.1 and YTS 169.4.2 respectively. Hybridoma lines were a kind gift of Prof. Anne Cooke, University of Cambridge) in PBS were injected into the tail vein at days −2 and +2 with respect to the challenge [33], [52]–[53]. Control animals received normal rat globulins (mpBiomedical). In this model the development of immunity post-vaccination proceeds normally as the mice are wild-type; removal of T-cells only occurs around the time of challenge. This regime was confirmed by FACS to deplete the mice of both CD4+ and CD8+ splenocytes (Figure S6).
For infections with WITS, each strain was individually grown statically for 16 h in L-broth and aliquots mixed to obtain a stock with each strain in an equal amount. Bacteria from 1 ml of this stock were harvested by centrifugation for subsequent qPCR. Dilutions of the stock were then made in PBS (typically 1 in 5×105) to achieve the desired final cell density for i.v. injection of 200 µl doses into the tail vein. Actual cell density of the inoculums were determined by plating triplicate 200 µl aliquots, cell density in the individual WITS cultures was also determined. Experimental group sizes are shown in Table S1.
To enable mathematical modelling of subpopulation structures we attempted to determine a single challenge dose that would result both in consistent colonisation of the organs in immunised animals, and a frequency of WITS absence in naive animals that would enable paramaterisation of our previously developed model; in naive animals a dose of ∼90 CFU is appropriate [25]. We conducted pilot experiments where animals immunised with the LV or naive controls were challenged with 90, 300, 900 or 9000 CFU total WITS and we determined the WITS population structure in spleens and livers at 6 h post-infection (Table S2) when bacterial counts are at a minimum in naive animals [25]. At the lowest dose there was excessive WITS loss in the spleen of immunised animals while at the other doses there was insufficient loss of WITS from naive animals. We therefore selected a challenge dose of ∼300 CFU which consistently resulted in colonisation of the spleen in immunised animals and modified the mathematical models to account for variations in the relative proportions of WITS between animals, rather than merely for the presence or absence of subpopulations.
Blood was obtained from the tail vein in heparin-coated tubes, mice were humanely killed by cervical dislocation, spleens and livers removed and individually homogenised in a Stomacher80 (Seward) with 5 ml distilled water. If required, dilutions were made to enable enumeration by pour plating 100 µl aliquots in 6-well plates. Entire blood samples or tissue homogenates in 1 ml aliquots were inoculated onto the surface of 90 mm agar plates. Following overnight incubation at 37°C, colonies were enumerated and total bacteria harvested from the plates by washing with 2 ml PBS. Bacteria were thoroughly mixed by vortexing, harvested by centrifugation and stored at −80°C prior to DNA extraction.
DNA was prepared from aliquots of bacterial samples (typically ∼5×109 CFU) using a DNeasy Blood and Tissue kit (QIAGEN). DNA concentration was determined using a NanoDrop 1000 spectrophotometer (Thermo Scientific). Approximately 106 total genome copies were analysed for the relative proportions of each WITS by qPCR on a Rotor-Gene Q (QIAGEN). Duplicate reactions were performed for each sample with primer pairs specific for each WITS in separate 20 µl reactions (primers in Table S3). Each reaction contained 10 µl QuantiTect SYBR Green (QIAGEN), 1 µM each primer, 4 µl sample and DNase-free water to 20 µl. Thermal cycling was 95°C for 15 min; 35 cycles of 94°C for 15 s, 61°C for 30 s, and 72°C for 20 s. The copy number of each WITS genome in the sample was determined by reference to standard curves for each primer pair. Standard curves were generated for each batch of PCR reagents by performing qPCRs in duplicate on 4 separate dilution series of known concentrations of WITS genomic DNA.
We used a branching process model that kept track of the distribution of the number of bacteria of a single WITS in three compartments: blood, liver and spleen. The model is outlined here, a full description is in Supporting Information Text S1. While we had determined the average number of CFU in the inoculum by plating, total bacterial loads in the LV immunized animals at 30 min were markedly lower, presumably due to the rapid activity of bactericidal mechanisms. In order to account for this observation we assumed that at time t = 0, the bacteria followed a Poisson distribution with unknown mean ν in the blood, whence they migrate into the liver (at rate cL) and the spleen (at rate cS). Upon entering the liver, bacteria replicate at rate rL and are killed at rate kL. Likewise, those in the spleen replicate at rate rS and are killed at rate kS. For the sake of parsimony, we assume that these rates remain constant for the first six hours, and we estimate them by fitting the model to the data from first two time points (0.5 h and 6 h post inoculation). We then allow the replication and killing rates to change at t = 6 h and assume they remain constant until t = 24 h; we estimate these new values by fitting the model to the data from the third time point (24 h p.i.). For each experimental treatment, we thus estimate 11 parameters (ν, cL, cS, kL1, kS1, rL1, rS1, kL2, kS2, rL2, rS2) from up to 720 data points (3 time points ×9 or 10 mice ×3 compartments ×8 strains), where subscripts 1 and 2 refer to the two time intervals considered (0–6 h and 6–24 h respectively) We assumed that all strains and all mice were independent and shared identical parameters within each experimental group. Our model did not allow movement of bacteria from the liver or the spleen back into the blood, as data strongly suggested that this does not happen until after 24 h [25].
Inference was done by maximum likelihood; the complete mathematical formulation is presented in Supporting Information Text S1. Basically, for a given set of model parameters, we computed the joint probability distribution of the number of copies of a single WITS in the blood, liver and spleen during the first 24 h by solving the master equations of the branching process. In order to compute the likelihood of the model, we also needed to determine the probability of observing a set of data (CFU and proportions of the 8 WITS in the blood, liver and spleen from a given mouse) given the unobserved numbers of copies of each WITS that were actually present in the blood and organs of that mouse (which correspond to the variables in our stochastic model); in other words, we had to estimate the noise generated by the experimental procedure. This was achieved by an additional calibration experiment: we plated known numbers of each WITS, harvested the colonies and processed known combinations of the eight WITS by qPCR. We then performed regressions of the observed proportions of the WITS by qPCR against the known numbers of colonies. We compared six models by AIC (Akaike's Information Criterion) for the distribution of ω (the product of the total number of colonies harvested by the proportion of a single WITS reported by qPCR) given n (the actual number of colonies of this particular WITS on the plate). The most parsimonious was a log-normal distribution where the log-standard-deviation decreases exponentially with n, namely: log(ω)∼N(log(n),0.267e−0.0148n). See Supporting Information Text S1 for further detail. All the analyses were performed in R version 3.0 [54], and likelihood estimation was done using the R library Powell [55].
|
10.1371/journal.pcbi.1002438 | Dynamic Effective Connectivity of Inter-Areal Brain Circuits | Anatomic connections between brain areas affect information flow between neuronal circuits and the synchronization of neuronal activity. However, such structural connectivity does not coincide with effective connectivity (or, more precisely, causal connectivity), related to the elusive question “Which areas cause the present activity of which others?”. Effective connectivity is directed and depends flexibly on contexts and tasks. Here we show that dynamic effective connectivity can emerge from transitions in the collective organization of coherent neural activity. Integrating simulation and semi-analytic approaches, we study mesoscale network motifs of interacting cortical areas, modeled as large random networks of spiking neurons or as simple rate units. Through a causal analysis of time-series of model neural activity, we show that different dynamical states generated by a same structural connectivity motif correspond to distinct effective connectivity motifs. Such effective motifs can display a dominant directionality, due to spontaneous symmetry breaking and effective entrainment between local brain rhythms, although all connections in the considered structural motifs are reciprocal. We show then that transitions between effective connectivity configurations (like, for instance, reversal in the direction of inter-areal interactions) can be triggered reliably by brief perturbation inputs, properly timed with respect to an ongoing local oscillation, without the need for plastic synaptic changes. Finally, we analyze how the information encoded in spiking patterns of a local neuronal population is propagated across a fixed structural connectivity motif, demonstrating that changes in the active effective connectivity regulate both the efficiency and the directionality of information transfer. Previous studies stressed the role played by coherent oscillations in establishing efficient communication between distant areas. Going beyond these early proposals, we advance here that dynamic interactions between brain rhythms provide as well the basis for the self-organized control of this “communication-through-coherence”, making thus possible a fast “on-demand” reconfiguration of global information routing modalities.
| The circuits of the brain must perform a daunting amount of functions. But how can “brain states” be flexibly controlled, given that anatomic inter-areal connections can be considered as fixed, on timescales relevant for behavior? We hypothesize that, thanks to the nonlinear interaction between brain rhythms, even a simple circuit involving few brain areas can originate a multitude of effective circuits, associated with alternative functions selectable “on demand”. A distinction is usually made between structural connectivity, which describes actual synaptic connections, and effective connectivity, quantifying, beyond correlation, directed inter-areal causal influences. In our study, we measure effective connectivity based on time-series of neural activity generated by model inter-areal circuits. We find that “causality follows dynamics”. We show indeed that different effective networks correspond to different dynamical states associated to a same structural network (in particular, different phase-locking patterns between local neuronal oscillations). We then find that “information follows causality” (and thus, again, dynamics). We demonstrate that different effective networks give rise to alternative modalities of information routing between brain areas wired together in a fixed structural network. In particular, we show that the self-organization of interacting “analog” rate oscillations control the flow of “digital-like” information encoded in complex spiking patterns.
| In Arcimboldo's (1527–1593) paintings, whimsical portraits emerge out of arrangements of flowers and vegetables. Only directing attention to details, the illusion of seeing a face is suppressed (Figure 1A–B). Our brain is indeed hardwired to detect facial features and a complex network of brain areas is devoted to face perception [1]. The capacity to detect faces in an Arcimboldo canvas may be lost when lesions impair the connectivity between these areas [2]. It is not conceivable, however, that, in a healthy subject, shifts between alternate perceptions are obtained by actual “plugging and unplugging” of synapses, as in a manual telephone switchboard.
Brain functions –from vision [3] or motor preparation [4] up to memory [5], attention [6]–[8] or awareness [9]– as well as their complex coordination [10] require the control of inter-areal interactions on time-scales faster than synaptic changes [11], [12]. In particular, strength and direction of causal influences between areas, described by the so-called effective connectivity [13]–[15], must be reconfigurable even when the underlying structural (i.e. anatomic) connectivity is fixed. The ability to quickly reshape effective connectivity –interpreted, in the context of the present study, as “causal connectivity” [16] or “directed functional connectivity” (see Discussion)– is a chief requirement for performance in a changing environment. Yet it is an open problem to understand which circuit mechanisms allow for achieving this ability. How can manifold effective connectivities –corresponding to different patterns of inter-areal interactions, or brain states [17]– result from a fixed structural connectivity? And how can effective connectivity be controlled without resorting to structural plasticity, leading to a flexible “on demand” selection of function?
Several experimental and theoretical studies have suggested that multi-stability of neural circuits might underlie the switching between different perceptions or behaviors [18]–[22]. In this view, transitions between many possible attractors of the neural dynamics would occur under the combined influence of structured “brain noise” [23] and of the bias exerted by sensory or cognitive driving [24]–[26]. Recent reports have more specifically highlighted how dynamic multi-stability can give rise to transitions between different oscillatory states of brain dynamics [27], [28]. This is particularly relevant in this context, because long-range oscillatory coherence [12], [29] –in particular in the gamma band of frequency (30–100 Hz) [29]–[32]– is believed to play a central role in inter-areal communication.
Ongoing local oscillatory activity modulates rhythmically neuronal excitability [33]. As a consequence, according to the influential communication-through-coherence hypothesis [31], neuronal groups oscillating in a suitable phase coherence relation –such to align their respective “communication windows”– are likely to interact more efficiently than neuronal groups which are not synchronized. However, despite accumulating experimental evidence of communication-through-coherence mechanisms [34]–[38] and of their involvement in selective attention and top-down modulation [30], [39], [40], a complete understanding of how inter-areal phase coherence can be flexibly regulated at the circuit level is still missing. In this study we go beyond earlier contributions, by showing that the self-organization properties of interacting brain rhythms lead spontaneously to the emergence of mechanisms for the robust and reliable control of inter-areal phase-relations and information routing.
Through large-scale simulations of networks of spiking neurons and rigorous analysis of mean-field rate models, we model the oscillatory dynamics of generic brain circuits involving a small number of interacting areas (structural connectivity motifs at the mesoscopic scale). Following [41], we extract then the effective connectivity associated to this simulated neural activity. In the framework of this study, we use a data driven rather than a model driven approach to effective connectivity [16] (see also Discussion section), and we quantify causal influences in an operational sense, based on a statistical analysis of multivariate time-series of synthetic “LFP” signals. Our causality measure of choice is Transfer Entropy (TE) [42], [43]. TE is based on information theory [44] (and therefore more general than causality measures based on regression [45], [46]), is “model-agnostic” and in principle capable of capturing arbitrary linear and nonlinear inter-areal interactions.
Through our analyses, we first confirm the intuition that “causality follows dynamics”. Indeed we show that our causal analysis based on TE is able to capture the complex multi-stable dynamics of the simulated neural activity. As a result, different effective connectivity motifs stem out of different dynamical states of the underlying structural connectivity motif (more specifically, different phase-locking patterns of coherent gamma oscillations). Transitions between these effective connectivity motifs correspond to switchings between alternative dynamic attractors.
We show then that transitions can be reliably induced through brief transient perturbations properly timed with respect to the ongoing rhythms, due to the non-linear phase-response properties [47] of oscillating neuronal populations. Based on dynamics, this neurally-plausible mechanism for brain-state switching is metabolically more efficient than coordinated plastic changes of a large number of synapses, and is faster than neuromodulation [48].
Finally, we find that “information follows causality” (and, thus, again, dynamics). As a matter of fact, effective connectivity is measured in terms of time-series of “LFP-like” signals reflecting collective activity of population of neurons, while the information encoded in neuronal representations is carried by spiking activity. Therefore an effective connectivity analysis –even when based on TE– does not provide an actual description of information transmission in the sense of neural information processing and complementary analyses are required to investigate this aspect. Based on a general information theoretical perspective, which does not require specifying details of the used encoding [44], we consider information encoded in spiking patterns [49]–[53], rather than in modulations of the population firing rate. As a matter of fact, the spiking of individual neurons can be very irregular even when the collective rate oscillations are regular [54]–[57]. Therefore, even local rhythms in which the firing rate is modulated in a very stereotyped way, might correspond to irregular (highly entropic) sequences of codewords encoding information in a digital-like fashion (e.g. by the firing –“1”– or missed firing –“0”– of specific spikes at a given cycle [58]). In such a framework, oscillations would not directly represent information, but would rather act as a carrier of “data-packets” associated to spike patterns of synchronously active cell assemblies. By quantifying through a Mutual Information (MI) analysis the maximum amount of information encoded potentially in the spiking activity of a local area and by evaluating how much of this information is actually transferred to distant interconnected areas, we demonstrate that different effective connectivity configurations correspond to different modalities of information routing. Therefore, the pathways along which information propagates can be reconfigured within the time of a few reference oscillation cycles, by switching to a different effective connectivity motif.
Our results provide thus novel theoretical support to the hypothesis that dynamic effective connectivity stems from the self-organization of brain rhythmic activity. Going beyond previous proposals, which stressed the importance of oscillations for feature binding [59] or for efficient inter-areal “communication-through-coherence”, we advance that the complex dynamics of interacting brain rhythms allow to implement reconfigurable routing of information in a self-organized manner and in a way reminiscent of a clocked device (in which digital-like spike pattern codewords are exchanged at each cycle of an analog rate oscillation).
In order to model the neuronal activity of interacting areas, we use two different approaches, previously introduced in [60]. First, each area is modeled as a large network of thousands of excitatory and inhibitory spiking neurons, driven by uncorrelated noise representing background cortical input (network model). Recurrent synaptic connections are random and sparse. In these networks, local interactions are excitatory and inhibitory. A scheme of the network model for a local area is depicted in Figure 2A (left). In agreement with experimental evidence that the recruitment of local interneuronal networks is necessary for obtaining coherent gamma cortical activity in vitro and in vivo [61], [62], the model develops synchronous oscillations () when inhibition is strong, i.e. for a sufficiently large probability of inhibitory connection [54]–[57], [63]. These fast oscillations are clearly visible in the average membrane potential (denoted in the following as “LFP”), an example trace of which is represented in Figure 2A (bottom right). Despite the regularity of these collective rhythms, the ongoing neural activity is only sparsely synchronized. The spiking of individual neurons is indeed very irregular [54], [56] and neurons do not fire an action potential at every oscillation cycle, as visible from the example spike trains represented in Figure 2A (top right). Structural network motifs involving areas are constructed by allowing excitatory neurons to establish in addition long-range connections toward excitatory or inhibitory neurons in a distant target area (see a schematic representation of an structural connectivity motif in Figure 2C). The strength of inter-areal coupling is regulated by varying the probability of establishing an excitatory connection.
In a second analytically more tractable approach, each area is described by a mean-field firing rate variable (rate model). The firing rate of a local population of neurons obeys the non-linear dynamical equation (4) (see Methods). All incorporated interactions are delayed, accounting for axonal propagation and synaptic integration. Local interactions are dominantly inhibitory (with coupling strength and delay ). Driving is provided by a constant external current. A cartoon of the rate model for a local area is depicted in Figure 2B (left). As in the network model, the firing rates undergo fast oscillations for strong inhibition (, [60]). An example firing rate trace is shown in Figure 2B (right). In order to build structural networks involving areas, different mean-field units are coupled together reciprocally by excitatory long range interactions with strength and delay (see a schematic representation of an structural motif in Figure 2D). Remarkably, the rate model and the network model display matching dynamical states [60] (see also later, Figures 3, 4 and 5). More details on the network and the rate models are given in the Methods section and in the Supporting Text S1.
For simplicity, we study fully connected structural motifs involving a few areas (). Note however that our approach might be extended to other structural motifs [64] or even to larger-scale networks with more specific topologies [41], [65].
In the simple structural motifs we consider, delays and strengths of local excitation and inhibition are homogeneous across different areas. Long-range inter-areal connections are as well isotropic, i.e. strengths and delays of inter-areal interactions are the same in all directions. Delay and strength of local and long-range connections can be changed parametrically, but only in a matching way for homologous connections, in such a way that the overall topology of the structural motif is left unchanged. As previously shown in [60], different dynamical states –characterized by oscillations with different phase-locking relations and degrees of periodicity– can arise from these simple structural motif topologies. Changes in the strength of local inhibition, of long-range excitation or of delays of local and long-range connections can lead to phase transitions between qualitatively distinct dynamical states. Interestingly, however, within broad ranges of parameters, multi-stabilities between dynamical states with different phase-locking patterns take place even for completely fixed interaction strengths and delays.
We generate multivariate time-series of simulated “LFPs” in different dynamical states of our models and we calculate TEs for all the possible directed pairwise interactions. We show then that effective connectivities associated to different dynamical states are also different. The resulting effective connectivities can be depicted in diagrammatic form by drawing an arrow for each statistically significant causal interaction. The thickness of each arrow encodes the strength of the corresponding interaction. This graphical representation makes apparent, then, that effective connectivity motifs or, more briefly, effective motifs, with many different topologies emerge from structural motifs with a same fixed topology. Such effective motifs are organized into families. All the motifs within a same family correspond to dynamical states which are multi-stable for a given choice of parameters, while different families of motifs are obtained for different ranges of parameters leading to different ensembles of dynamical states.
We analyze in detail, in Figures 3, 4 and 5, three families of motifs arising for strong intra-areal inhibition and similarly small values of delays for local and long-range connections. We consider (panels A and B) and (panels C and D) structural motifs. Panels A and C show TEs for different directions of interaction, together with “LFPs” and example spike trains (from the network model), and rate traces (from matching dynamical states of the rate model). Panels B and D display motifs belonging to the corresponding effective motif families.
A first family of effective motifs occurs for weak inter-areal coupling. In this case, neuronal activity oscillates in a roughly periodic fashion (Figure 3A and C, left sub-panel). When local inhibition is strong, the local oscillations generated within different areas lock in an out-of-phase fashion. It is therefore possible to identify a leader area whose oscillations lead in phase over the oscillation of laggard areas [60]. In this family, causal interactions are statistically significant only for pairwise interactions proceeding from a phase-leading area to a phase-lagging area, as shown by the the box-plots of Figure 3A and C (right sub-panel, see Discussion and Methods for a discussion of the threshold used for statistical significancy). As commented more in detail in the Discussion section, the anisotropy of causal influences in leader-to-laggard and laggard-to-leader directions can be understood in terms of the communication-through-coherence theory. Indeed the longer latency from the oscillations of the laggard area to the oscillations of the leader area reduces the likelihood that rate fluctuations originated locally within a laggard area trigger correlated rate fluctuations within a leading area [35] (see also Discussion). Thus, out-of-phase lockings for weak inter-areal coupling give rise to a family of unidirectional driving effective motifs. In the case of , causality is significant only in one of two possible directions (Figure 3B), depending on which of the two areas assumes the role of leader. In the case of , it is possible to identify a “causal source” area and a “causal sink” area (see [66] for an analogous terminology), such that no direct or indirect causal interactions in a backward sense from the sink area to the source area are statistically significant. Therefore, the unidirectional driving effective motif family for contains six motifs (Figure 3D), corresponding to all the possible combinations of source and sink areas.
A second family of effective motifs occurs for intermediate inter-areal coupling. In this case, the periodicity of the “LFP” oscillations is disrupted by the emergence of large correlated fluctuations in oscillation cycle amplitudes and durations. As a result, the phase-locking between “LFPs” becomes only approximate, even if it continues to be out-of-phase on average. The rhythm of the laggard area is now more irregular than the rhythm in the leader area. Laggard oscillation amplitudes and durations in fact fluctuate chaotically (Figure 4A and C, left sub-panel). Fluctuations in cycle length do occasionally shorten the laggard-to-leader latencies, enhancing non-linearly and transiently the influence of laggard areas on the leader activity. Correspondingly, TEs in leader-to-laggard directions continue to be larger, but TEs in laggard-to-leader directions are now also statistically significant (Figure 4A and C, right sub-panel). The associated effective motifs are no more unidirectional, but continue to display a dominant direction or sense of rotation (Figure 4B and D). We refer to this family of effective motifs as to a family of leaky driving effective motifs (containing two motifs for and six motifs for ).
Finally, a third family of effective motifs occurs for stronger inter-areal coupling. In this case the rhythms of all the areas become equally irregular, characterized by an analogous level of fluctuations in cycle and duration amplitudes. During brief transients, leader areas can still be identified, but these transients do not lead to a stable dynamic behavior and different areas in the structural motif continually exchange their leadership role (Figure 5A and C, left sub-panel). As a result of the instability of phase-leadership relations, only average TEs can be evaluated, yielding to equally large TE values for all pairwise directed interactions (Figure 5A and C, right sub-panel). This results in a family containing a single mutual driving effective motif (Figure 5B and D).
Further increases of the inter-areal coupling strength do not restore stable phase-locking relations and, consequently, do not lead to additional families of effective motifs. Note however that the effective motif families explored in Figures 3, 4 and 5 are not the only one that can be generated by the considered fully symmetric structural motifs. Indeed other dynamical configurations exist. In particular, anti-phase locking (i.e. locking with phase-shifts of for and of for ) would become stable when assuming the same interaction delays and inter-areal coupling strengths of Figures 3, 4 and 5, but a weaker local inhibition. Assuming different interaction delays for local and long-range interactions, out-of-phase lockings continue to be very common, but in-phase and anti-phase locking can become stable even for strong local inhibition, within specific ranges of the ratio between local and long-range delays [60]. For , in the case of general delays, more complex combinations can arise as well, like, for instance, states in which two areas oscillate in-phase, while a third is out-of-phase. In-phase locking between areas gives rise to identical TEs for all possible directed interactions, resulting in effective motifs without a dominant directionality. Anti-phase lockings for give rise to relatively large inter-areal phase-shifts and, correspondingly, to weak inter-areal influences (at least in the case of weak inter-areal coupling), resulting in small TE levels which are not statistically significant (not shown). However, in the framework of this study, we focus exclusively on out-of-phase-locked dynamical states, because they are particularly relevant when trying to achieve a reconfigurable inter-areal routing of information (see later results and Discussion section).
To conclude, we remark that absolute values of TE depend on specific parameter choices (notably, on time-lag and signal quantization, see Methods). However, the relative strengths of TE in different directions –and, therefore, the resulting topology of the associated effective motifs– are rather robust against changes of these parameters. Robustness of causality estimation is analyzed more in detail in the Discussion section.
How can asymmetric causal influences emerge from a symmetric structural connectivity? A fundamental dynamical mechanism involved in this phenomenon is known as spontaneous symmetry breaking. As shown in [60], for the case of the structural motif, a phase transition occurs at a critical value of the strength of inter-areal inhibition. When local inhibition is stronger than this critical threshold, a phase-locked configuration in which the two areas oscillate in anti-phase loses its stability in favor of a pair of out-of-phase-locking configurations, which become concomitantly stable. The considered structural motif is symmetric, since it is left unchanged after a permutation of the two areas. However, while the anti-phase-locking configuration, stable for weak local inhibition, share this permutation symmetry with the full system, this is no more true for the out-of-phase-locking configurations, stable for strong local inhibition. Note, nevertheless, that the configuration in which leader and laggard area are inverted is also a stable equilibrium, i.e. the complete set of stable equilibria continue to be symmetric, even if individual stable equilibria are not (leading thus to multi-stability). In general, one speaks about spontaneous symmetry breaking whenever a system with specific symmetry properties assumes dynamic configurations whose degree of symmetry is reduced with respect to the full symmetry of the system. The occurrence of symmetry breaking is the signature of a phase transition (of the second order [67]), which leads to the stabilization of states with reduced symmetry.
The existence of a symmetry-breaking phase transition in the simple structural motifs we analyze here (for simplicity, we consider the case) can be proven analytically for the rate model, by deriving the function , which describes the temporal evolution of the phase-shift between two areas when they are weakly interacting [47]:(1)The function for the rate model is shown in the left panel of Figure 6B. Stable phase lockings are given by zeroes of with negative slope crossing and are surrounded by basins of attraction (i.e. sets of configurations leading to a same equilibrium), whose boundaries are unstable in- and anti-phase lockings (Figure 6A). For the network model, a function with an analogous interpretation and a similar shape, shown in the right panel of Figure 6B, can be extracted from simulations, based on a phase description of “LFP” time-series (see Methods and Supporting Figure S1A). The analogous distribution of the zero-crossings of and results in equivalent phase-locking behaviors for the rate and network models. Thus spontaneous symmetry breaking leads to multi-stability between alternative out-of-phase-lockings and to the emergence of unidirectional effective driving within a symmetric structural motif.
Because of multi-stability, transitions between effective motifs within a family can be triggered by transient perturbations, without need for structural changes. We theoretically determine conditions for such transitions to occur. The application of a pulse of current of small intensity advances or delays the phase of the ongoing local oscillation (see Supporting Figure S1B). This is true for rate oscillations of the mean-field rate model, but also for “LFP” oscillations reflecting rhythmic synchronization in the network model. In the latter case, the collective dynamics is perturbed by synchronously injecting pulse currents into all of the neurons within an area. The induced phase shift depends on the perturbation strength but also on the phase at which the perturbation is applied. For the network model, this can be measured directly from numeric simulations of a perturbed dynamics (see Methods and right panel of Figure 6D). For the rate model, the phase shift induced by an instantaneous phased perturbation can be described analytically in terms of the Phase Response Curve (PRC) [47] (see Figure 6D, left, and Supporting Text S1). After a pulse, the phase-shift between two areas is “kicked out” of the current equilibrium locking and assumes a new transient value (solid paths in Figure 6C), which, for weak perturbations and inter-areal coupling, reads:(2)where the approximate equality between square brackets holds for the mean-field rate model. If falls into the basin of attraction of a different phase-locking configuration than , the system will settle within few oscillation cycles into an effective connectivity motif with a different directionality (dashed green path in Figure 6C). Even relatively small perturbations can induce an actual transition, if applied in selected narrow phase intervals in which the induced grows to large values. For most application phases, however, even relatively large perturbations fail to modify the effective driving direction (dashed red path in Figure 6C), because the induced perturbation is vanishingly small over large phase intervals (Figure 6D). This is a robust property, shared by the two (radically different) models we consider here and –we hypothesize– by any local circuit generating fast oscillations through a mechanism based on delayed mutual inhibition. As a consequence, for a given perturbation intensity, a successful switching to a different effective motif occurs only if the perturbation is applied within a specific phase interval, that can be determined analytically from the knowledge of and of for the rate model, or semi-analytically from the knowledge of and (see Methods). Figure 6E–F reports the fraction of simulated phased pulses that induced a change of effective directionality as a function of the phase of application of the perturbation. The phase intervals for successful switching predicted by the theory are highlighted in green. We performed simulations of the rate (Figs. 6E–F, left column) and of the network (Figs. 6E–F, middle column) models, for unidirectional (Figs. 6E) and leaky driving (Figs. 6F) effective motifs. Although our theory assumes small inter-areal coupling and is rigorous only for the rate model, the match between simulations and predictions is very good for both models and families of motifs.
In Figs. 6E–F, we perturb the dynamics of the laggard area, but changes in directionality can also be achieved by perturbing the leader area (Supporting Figure S2). Note also that, in the network model, direction switchings can take place spontaneously, due to noisy background inputs. Such noise-induced transitions, however, occur typically on time-scales of the order of seconds, i.e. slow in terms of biologic function, because the phase range for successful switching induction is narrow.
A second non-linear dynamic mechanism underlying the sequence of effective motifs of Figures 3 and 4 is effective entrainment. In this phenomenon, the complex dynamics of neural activity seems intriguingly to be dictated by effective rather than by structural connectivity.
We consider as before a rate model of reciprocally connected areas (Figure 2D). In order to properly characterize effective entrainment, we review the concept of bifurcation diagram [68]. As shown in [60], when the inter-areal coupling is increased, rate oscillations become gradually more complex (cfr. Figure 7A), due to the onset of deterministic chaos (see also [69] for a similar mechanism in a more complex network model). For small , oscillations are simply periodic (e.g. ). Then, for intermediate (e.g. ), the peak amplitudes of the laggard area oscillation assume in alternation a small number of possible values (period doubling). Finally, for larger (e.g. ), the laggard peak amplitudes fluctuate in a random-like manner within a continuous range. This sequence of transitions can be visualized by plotting a dot for every observed value of the peak amplitudes of oscillation cycles, at different values of . The accumulation of these dots traces an intricate branched structure, which constitutes the bifurcation diagram (Figure 7B).
Bifurcation diagrams for the leader and for the laggard area are plotted in Figure 7B (top panel, in orange and green color, respectively). We compare these bifurcation diagrams with the analogous diagrams constructed in the case of two unidirectionally coupled oscillating areas. Qualitatively similar bifurcation sequences are associated to the dynamics of the laggard area (bidirectional coupling) and of the driven area (unidirectional coupling, Figure 7B, bottom panel, green color), for not too strong inter-areal couplings. In the case of unidirectional coupling, the peak amplitudes of the unperturbed driver area oscillations do not fluctuate at all. Therefore, the corresponding bifurcation diagram is given by a constant line (Figure 7B, bottom panel, orange color). In the case of bidirectional coupling, the peak amplitudes of the leader area oscillations undergo fluctuations, but only with a tiny variance. Thus, the corresponding bifurcation diagram has still the appearance of a line, although now “thick” and curved (zooming would reveal bifurcating branches). Note that, for unidirectional coupling, the structural connectivity is explicitly asymmetric. The periodic forcing exerted by the driving area is then known to entrain the driven area into chaos [70]. Such direct entrainment is the dynamical cause of chaos. On the other hand, for bidirectional coupling, the structural connectivity is symmetric. However, due to spontaneous symmetry breaking, the resulting effective connectivity is asymmetric and the system behaves as if the leader area was a driver area, entraining the laggard area into chaos being only negligibly affected by its back-reaction. Such effective entrainment can be seen as an emergent dynamical cause of chaos. Thus, the dynamics of a symmetric structural motif with asymmetric effective connectivity and of a structural motif with a matching asymmetric topology are equivalent.
For a sufficiently strong inter-areal coupling, symmetry in the dynamics of the bidirectional structural motif is suddenly restored [60], in correspondence with a transition to the mutual driving family of effective motifs (Figure 5). As a result, in absence of symmetry breaking, effective driving cannot anymore take place. Thus, for a too strong inter-areal coupling, the emergent anisotropy of effective connectivity is lost, and, with it, the possibility of a dynamic control of effective connectivity (at least via the previously discussed strategies).
Despite its name, Transfer Entropy is not directly related to a transfer of information in the sense of neuronal information processing. The TE from area to area measures indeed just the degree to which the knowledge of the past “LFP” of reduces the uncertainty about the future “LFP” of [43], [71]. As a matter of fact, however, the information stored in neural representations must be encoded in terms of spikes, independently from the neural code used. Therefore, it is important to understand to which extent an effective connectivity analysis based on “macroscopic” dynamics (i.e. TEs estimated from “LFPs”) can pretend to describe actual “microscopic” information transmission (i.e. at the level of spiking correlations).
In order to address this issue, we first introduce a framework in which to quantify the amount of information exchanged by interacting areas. In the case of our model, rate fluctuations could encode only a limited amount of information, since firing rate oscillations are rather stereotyped. On the other hand, a larger amount of information could be encoded based on spiking patterns, since the spiking activity of single neurons is very irregular and thus characterized by a large entropy [44], [58]. As illustrated by Figure 8A, a code can be built, in which a “1” or a “0” symbol denote respectively firing or missed firing of a spike by a specific neuron at a given oscillation cycle. Based on such an encoding, the neural activity of a group of neurons is mapped to digital-like streams, “clocked” by the ongoing network rhythm, in which a different “word” is broadcast at each oscillation cycle. Note that we do not intend to claim that such a code is actually used in the brain. Nevertheless, we introduce it as a theoretical construct grounding a rigorous analysis of information transmission.
We focus here on the fully symmetric structural motif of areas of Figure 2C. We modify the network model considered in the previous sections by embedding into it transmission lines (TLs), i.e. mono-directional fiber tracts dedicated to inter-areal communication (see Figure 8B). In more detail, selected sub-populations of source excitatory neurons within each area establish synaptic contacts with matching target excitatory or inhibitory cells in the other area, in a one-to-one cell arrangement. Synapses in a TL are strengthened with respect to usual synapses, by multiplying their peak conductance by a multiplier (see Methods section). Such multiplier is selected to be large, but not too much, in order not to affect the phase-relations between the collective oscillations of the two areas. Indeed, selecting a too large would lead to an in-phase-locking configuration in which collective dynamics is enslaved to the synchronous activity of source and target populations. As analyzed in the Supporting Figure S3, a suitable tuning of ensures that source-to-target neuron communication is facilitated as much as possible, without disrupting the overall effective connectivity (associated to the unperturbed phase-locking pattern). Note that such TL synapses are here introduced as a heuristic device, allowing to maximize the potential capacity of inter-areal communication channels [44]. However, due to the occurrence of consistent spike-timing relations in out-of-phase locked populations, it might be that spike-timing-dependent plasticity [72] lead to the gradual emergence of subsets of synapses with substantially enhanced weight [73], which would play a role in inter-circuit communication very similar to TL synapses.
The information transmission efficiency of each TL, for the case of different effective motifs, is then assessed by quantifying the Mutual Information (MI) [44], [58] between the “digitized” spike trains of pairs of source and target cells (see Methods). Since a source cell spikes on average every five or six oscillation cycles, the firing of a single neuron conveys of information per oscillation cycle. MI normalized by the source entropy H indicates how much of this information reaches the target cell, a normalized MI equal to unity denoting lossless transmission. As shown by Figure 8C–D, the communication efficiency of embedded TLs depends strongly on the active effective motif. In the case of unidirectional driving effective motifs (Figure 8C), communication is nearly optimal along the TL aligned with the effective connectivity. For the misaligned TL, however, no enhancement occurs with respect to control (i.e. pairs of connected cells not belonging to a TL). In the case of leaky driving effective motifs (Figure 8D), communication efficiency is boosted for both TLs, but more for the TL aligned with the dominant effective direction. For both families of effective motifs, despite the strong anisotropy, the communication efficiencies of the two embedded TLs can be “swapped” within one or two oscillation cycles, by reversing the active effective connectivity through a suitable transient perturbation (see Figure 6E–F). The considered structural motif acts therefore as a “diode” through which information can propagate efficiently only in one (dynamically reconfigurable) direction determined by effective connectivity.
We have shown that a simple structural motif of interacting brain areas can give rise to multiple effective motifs with different directionality and strengths of effective connectivity, organized into different families. Such effective motifs correspond to distinct dynamical states of the underlying structural motif. Beyond this, dynamic multi-stability makes the controlled switching between effective motifs within a same family possible without the need for any structural change.
On the contrary, transitions between effective motifs belonging to different families (e.g. a transition from a unidirectional to a leaky driving motif) cannot take place without changes in the strength of the delay of inter-areal couplings, even if the overall topology of the underlying structural motif does not need to be modified. Each specific effective motif topology (i.e. motif family) is robust within broad ranges of synaptic conductances and latencies, however if parameters are set to be close to critical transition lines separating different dynamical regimes, transitions between different families might be triggered by moderate and unspecific parameter changes. This could be a potential role for neuromodulation, known to affect the net efficacy of excitatory transmission and whose effect on neural circuits can be modeled by coordinated changes in synaptic conductances [74], [75].
Note that dynamical coordination of inter-areal interactions based on precisely-timed synchronous inputs would be compatible with experimental evidence of phase-coding [76]–[81], indicating a functional role for the timing of spikes relative to ongoing brain rhythms (stimulus-locked [82], [83] as well as stimulus-induced or spontaneous [84]). Note also that the time of firing is potentially controllable with elevated precision [85]–[87] and has been found to depend on the phase of LFPs in local as well as in distant brain areas [37].
In general, control protocols different from the one proposed here might be implemented in the brain. For instance, phased pulses might be used as well to stabilize effective connectivity in the presence of stronger noise. Interestingly, the time periods framed by cycles of an ongoing oscillation can be sliced into distinct functional windows in which the application of the same perturbation produces different effects.
Finally, in addition to “on demand” transitions, triggered by exogenous –sensory-driven– or endogenous –cognitive-driven– control signals, noise-driven switching between effective motifs might occur spontaneously, yielding complex patterns of activity during resting state [26], [88], [89].
As revealed by our discussion of spontaneous symmetry breaking and effective entrainment, an analysis based on TE provides a description of complex inter-areal interactions compliant with a dynamical systems perspective. It provides, thus, an intuitive representation of dynamical states that is in the same “space” as anatomical connectivity.
Note that it is currently debated whether TE should be considered as a measure of effective connectivity in strict sense [13], [15], or, rather, of yet another type of connectivity beyond functional connectivity (that could be dubbed causal connectivity [16], [66] or directed functional connectivity). Our position is that TE constitutes, at least in the context of the present study, a measure of effective connectivity in proper sense. Indeed, as indicated by the analysis of Figure 8C–D, the connectivity motifs inferred by TE correctly represent characteristic dynamic mechanisms, like spontaneous symmetry breaking or asymmetric chaos [60], enabling specifically associated modalities of inter-areal information transmission. Therefore, we can conclude that causality (as inferred by TE) follows dynamics (by representing the action of corresponding dynamic mechanisms).
TE constitutes thus a model-free approach (although, non “parameter-free”, cfr. forthcoming section and Figure 9) to the effective connectivity problem, suitable for exploratory data-driven analyses. In this sense it differs from regression-based methods like usual implementations of Granger Causality (GC) [45], [46] or from Dynamic Causal Modeling (DCM) [90], which are model-driven [15], [16], [91]. Strategies like DCM, in particular, assume prior knowledge about the inputs to the system and works by comparing the likelihood of different a priori hypotheses about interaction structures. Such an approach has the undeniable advantage of providing a direct description of actual mechanisms underlying effective connectivity changes (the stimulus-dependence of effective couplings is indeed modeled phenomenologically). However, it might be too restrictive (or arbitrary) when the required a-priori information is missing or highly uncertain. TE, on the contrary: does not require any hypothesis on the type of interaction; should be able to detect even purely non-linear interactions and should be robust against linear cross-talk between signals [92]. These features, together with the efficacy of TE for the causal analysis of synthetic time-series, advocate for a more widespread application of TE methods to real neural data [93]–[95] (at the moment limited by the need of very long time-series [92]).
Note that we do not intend to claim superiority of TE in some general sense. As a matter of fact TE is equivalent to GC, as far as the statistics of the considered signals are gaussian [71]. Furthermore, non-linear generalizations of GC and DCM [96]–[99] might be able to capture at a certain extent the complex self-organized dynamics of the neural activity models analyzed in the present study. However, a systematic comparison of the performance of different methods in capturing causal connectivity of realistic non-linear models of neural dynamics goes beyond the focus of the present study and is deferred to future research.
We finally would like to stress, to avoid any potential confusion, that the structural motifs analyzed in the present study are well distinct from causal graphical models of neural activity, in the statistical sense proper of DCMs [90], [100]. They constitute indeed actual mechanistic models of interacting populations of spiking neurons, with a highly non-linear dynamics driven by background noise. Connections in these models are model synapses, i.e. mere structural couplings, not phenomenological effective couplings. Thus, effective connectivity is not constrained a priori, as in DCMs, but is an emergent property of network dynamics, consistent with the existence of effective motif topologies different from the underlying structural topology.
The effective connectivity analyses presented in this study were conducted by evaluating TEs under specific parameter choices. However, absolute values of TE depend on parameters, like, notably, the resolution at which “LFP” signals are quantized and the time-lag at which we probe causal interactions. As discussed in detail in the Methods section, estimation of TE requires the sampling of joint distributions of “LFP” values in different areas at different times. Such distributions are sampled as histograms, based on discrete multi-dimensional binning. In practice, each “LFP” time-series is projected to a stream of symbols from a discrete alphabet, corresponding to different quantization levels of the continuous “LFP” signals [101]. The actual number of used bins is a free parameter, although some guiding criteria for its selection do exist [43]. Concerning time-lag , our TE analysis (conducted at the first Markov order [42], following [41], [94]) describes predictability of “LFPs” at time based on “LFPs” at time . The used time-lag is once again a free parameter. To deal with this arbitrariness in parameter choices, we explore systematically the dependence of TE estimations from the aforementioned parameters, by varying both and in a wide continuous range. Figure 9 summarizes the results of this analysis, for three different effective motifs.
Considering the dependency on time-lag , a periodic structure is clearly noticeable in the TE matrices reported in Figure 9. TE values tend to peak in precise bands of , related to latencies between the oscillations of different areas. The analysis of the unidirectional driving motif (Figure 9A), associated to leader-laggard periodic configurations, is particularly transparent (and has a high pedagogic value). Two characteristic time-lags can be defined: a “short” lag , given by the time-shift from oscillation peaks of the leader area to oscillation peaks of the laggard area ; and a “long” lag, , given by the time-shift from laggard to leader oscillation peaks (here, is an average oscillation period, common to both areas leader and laggard areas and ). TE in the direction from leader to laggard, , peaks for the first time at a time-lag (and then at lags , where is a positive integer). TE in the direction from laggard to leader, , peaks first at a time-lag (and then at lags ). If the “LFP” signals were deterministic and strictly periodic, the quantities and would be identical (and diverging for infinite precision [42]). However “LFP” signals are only periodic on average and have a stochastic component, due to the joint effect of random network connectivity and noisy background inputs. This stochastic component is responsible for small cycle-to-cycle fluctuations in the amplitude of “LFP” oscillation peaks. As discussed more in depth in a next subsection, the efficiency with which fluctuations in the output of a local area can induce (i.e., can “cause”) fluctuations of the output of a distant interconnected area depends on the instantaneous local excitability of this target area, which is undergoing a rhythmic modulation due to the ongoing collective oscillation [31], [33]. As a result, TE can reach different peak values in different directions (and, as a matter of fact, ).
Considering then the dependence on signal quantization, we observe that TE values tend to grow for increasing number of bins , i.e. for a finer resolution in tracking “LFP” amplitude variations. This can be once again understood in terms of the temporal structure of “LFP” signals. As just mentioned, dynamic correlations between small “LFP” amplitude fluctuations carry information relevant for causality estimation. This information would be completely lost by using a too small number of bins for TE evaluation, given that the largest contribution to the dynamic range of “LFP” signals is provided by its fairly stereotyped oscillatory component. Conversely, using a too large number of bins would lead to under-sampling artifacts (therefore, we do not consider the use of more than quantization bins).
By evaluating a threshold for statistical significance independently for each direction and combination of and , we find that, for weak inter-areal coupling, TE never goes above this threshold in the laggard-to-leader direction (Figure 9A). We are also unable to find any choice of and such that, for intermediate inter-areal coupling, TE in the laggard-to-leader direction becomes larger or equal than TE in leader-to-laggard direction (Figure 9B). Looking at matrices of the causal unbalancing (see Methods, and Figure 9, third column), we see indeed that, for weak and intermediate coupling strengths, effective connectivity is robustly asymmetric in the parameter regions in which causal interactions are statistically significant. Effective connectivity is on the contrary balanced for strong inter-areal coupling (Figure 9C).
We can thus summarize the previous statements by saying that absolute values of TE depend on the choices of and , but that the topology of the resulting effective motif does not (at least in the wide range considered for this robustness analysis).
Traditionally, studies about communication-through-coherence or long-range binding between distant cell assemblies have emphasized the importance of in-phase locking (see, e.g. [35], [102]). Although, as previously mentioned, in-phase locking (as well as anti-phase locking) can also arise in our models for different choices of coupling delays and inhibition strengths [60], we decided in the present study to focus on out-of-phase lockings. The case of spontaneous symmetry breaking is indeed particularly interesting, because it underlie the emergence of a dominant directionality in the causal influences between areas reciprocally coupled with comparable strengths. Furthermore, spontaneous symmetry breaking is responsible for the multi-stability between effective connectivity configurations, thus opening the way to a self-organized control of inter-areal interactions [11], [12].
In particular, our study confirms that the reorganization of oscillatory coherence might regulate the relative weight of bottom-up and top-down inter-areal influences [17], [30] or select different interaction modes within cortical networks involving areas of similar hierarchical level, as in the case of motor preparation or planning [4], [103] or language [104].
As a next step, we directly verified that “information follows causality”, since changes in effective connectivity are paralleled by reconfiguration of inter-areal communication modalities. Following [32], [35], we explain the anisotropic modulations of communication efficiency (see Figure 8) in terms of a communication-through-coherence mechanism. In fact, because of the out-of-phase locking between rhythms, spikes emitted by neurons in a phase-leading area reach neurons in a phase-lagging area at a favorable phase in which they are highly excitable. Conversely, spikes emitted by neurons in a phase-lagging area reach neurons in a phase-leading area when they are strongly hyperpolarized by a preceding peak of synchronous inhibition. This same mechanism underlie also the anisotropy of “LFP”-based TE, since “LFP” fluctuations are the manifestation (at least in our model) of local population firing rate fluctuations.
Therefore, by combining TE analyses of “LFP”-based effective connectivity with MI analyses of spike-based information transmission, we are able to establish a tight link between control of effective connectivity and control of communication-through-coherence, both of them being emergent manifestations of the self-organized dynamics of interacting brain rhythms.
To conclude, we also note that similar mechanisms might be used beyond the mesoscale level addressed here. Multi-stabilities of structural motifs might be preserved when such motifs are interlaced as modules of a network at the whole-brain level [64]. Likewise, dynamic control of information routing between neuronal clusters [73], [105] or even single cells might occur within more local microcircuits [106], [107].
The previous discussions suggest that oscillations, rather than playing a direct role in the representation of information, would be instrumental to the reconfigurable routing of information encoded in spiking activity. Original formulations of the communication-through-coherence hypothesis [31] suggested that oscillatory coherence facilitates the transmission of local fluctuations of firing rate to a distant site, thus assuming implicitly a rate-based encoding of information in neuronal activity. However, more complex coding mechanisms based on patterns of precisely timed spikes might be achievable by biologically-plausible neuronal circuits [85], [86].
As a matter of fact, our study reveals that the inherent advantages of “labelled-line” codes [51], [108] (in which the information about which local neuron is firing is preserved) –i.e., notably, an augmented information capacity with respect to “summed-population” codes– might be combined with the flexibility and the reliability offered by the communication-through-coherence framework. Indeed, as shown by the analyses of Figure 8, suitable inter-areal phase relations make possible the transmission of information encoded in detailed spiking correlations, rather than just in population firing rate fluctuations.
This is particularly interesting, since many cortical rhythms are only sparsely synchronized, with synchronous oscillations evident in LFP, Multi-Unit Activity or intracellular recordings but not in single unit spike trains [109]–[111]. Such sparse firing might possibly reflect population-coding of behaviorally-relevant information transcending rate-based representations [49]–[53]. Independently from the complexity of these hypothetic representations, our study shows that self-organized communication-through-coherence would have the sufficient potential to dynamically route the rich information that these representations might convey.
It is very plausible that flexible inter-areal coordination is achieved in the brain through dynamic self-organization [11] as in our models. However, qualitatively different mechanisms than symmetry breaking might contribute to the generation of dynamic effective connectivity in other regimes of activity. Despite sparse synchronization, the level of coherence in our model neuronal activity is larger than in many brain oscillations. However, our results might be generalized to activity regimes in which synchronization is weaker. Phase-relations have been shown to impact effective connectivity even in essentially asynchronous regimes [112]. It would be interesting to understand whether the dominant directionality of effective connectivity can be controlled when out-of-phase locking is only transient [12], [41].
Another open question is whether our theory can be extended to encompass the control of effective connectivity across multiple frequency bands [94]. This is an important question since top-down and bottom-up inter-areal communication might exploit different frequency channels, possibly due to different anatomic origins of ascending and descending cortico-cortical connections [113].
Finally, we are confident that our theory might inspire novel experiments, attempting to manipulate the directionality of inter-areal influences via local stimulation applied conditionally to the phase of ongoing brain rhythms. Precisely timed perturbing inputs could indeed potentially be applied using techniques like electric [114] or optogenetic [115] microstimulation, especially in closed-loop implementations with millisecond precision [116], [117].
Each area is represented by a random network of excitatory and inhibitory Wang-Buzsáki-type conductance-based neurons [118]. The Wang-Buzsáki model is described by a single compartment endowed with sodium and potassium currents. Note that results (not shown) of simulations performed with a more realistic ratio of excitatory and inhibitory neurons per population would lead to qualitatively similar results with small parameter adjustments (using, for instance, parameters as in [69]).
The membrane potential is given by:(3)where is the capacitance of the neuron, is a leakage current, is an external noisy driving current (due to background Poisson synaptic bombardment), and and are respectively a sodium and a potassium current, depending non linearly on voltage. The last input term is due to recurrent interactions with other neurons in the network. Excitatory synapses are of the AMPA-type and inhibitory synapses of the GABA-type and are modeled as time-dependent conductances. A complete description of the model and a list of all its parameters are given in the Supporting Text S1. “LFP” is defined as the average membrane potential over the cells in each area.
Short-range connections within a local area from population to population are established randomly with probability , where and can be either one of the type (excitatory) or . The excitatory populations are allowed also to establish connections toward populations and in remote areas (). Such long-range connections are established with a probability (). For simplicity, however, we assume that and that . For each of the considered dynamical states, probabilities of connection are provided in the corresponding figure caption.
First, a structural motif of interconnected random networks of spiking neurons is generated, as in the previous section. Then, on top of the existing excitatory long-range connections, additional stronger long-range connections are introduced in order to form directed transmission lines. In each area a source sub-population, made out of 400 excitatory neurons, and a non-overlapping target sub-population, made out of 200 excitatory and 200 inhibitory neurons, are selected randomly. Excitatory cells in the source populations get connected to cells in the target sub-populations of the other area via strong synapses. These connections are established in a one-to-one arrangement (e.g. each source cell establishes a TL-synapse with a single target cell that does not receive on its turn any other TL-synapse).
The peak conductance of TL-synapses is times stronger than the normal excitatory peak conductance . For the simulations with TL (Figure 8 of the main paper), we set respectively for the unidirectional and for the leaky driving effective motifs. Such unrealistically strong peak conductances, whose purpose is to optimize information transfer by enhancing spiking correlations, can be justified by supposing that each source neuron establishes multiple weaker synaptic contacts with the same target neuron. The multiplier is selected to be as large as possible without altering the original out-of-phase locking relations between the two populations (Figure S3A). Concretely, is tuned by raising it gradually until when a critical point is reached in which the populations lock in-phase (Figure S3C). Then, is set to be just below this critical point (Figure S3B).
Each area is represented by a single rate unit. The dynamical equations for the evolution of the average firing rate in an area are given by:(4)Here, if , and zero otherwise. A constant current represents a background input, stands for the strength of intra-areal inhibition, for the strength of inter-areal excitation and and are the delays of the local and long-range interactions, respectively. We consider in this study only fully symmetric structural motifs of mutually connected areas. For each of the considered dynamical states, the values of , , and are provided in the figure caption.
Given an oscillatory time-series of neuronal activity, generated indifferently by a rate or by a network model, a phase , for , is linearly interpolated over each oscillation cycle. Here denotes the start time of the oscillation cycle. Note that this definition does not require that the oscillation is periodic: this empiric phase “elastically” adapts to fluctuations in the duration of oscillation cycles (see Supporting Figure S1A).
The phase shift induced by a pulse perturbation (see Supporting Figure S1B) is described by the Phase Response Curve (PRC) (see Eq. (2) and [47]). For the rate model, the PRC can be evaluated analytically if certain general conditions on the relation between the oscillation period and the local inhibition delay are fulfilled [60]. Analytical expressions for the PRC of the rate model, as plotted in Figure 6D (left), are reported in the Supporting Text S1.
In the network model, it is possible to evaluate the phase-shift induced by a perturbation, by directly simulating the effects of this perturbation on the oscillatory dynamics. A perturbation consists of a pulse current of strength injected synchronously into all neurons of one area at a phase of the ongoing local oscillation. The induced phase-shift is estimated by comparing the phases of the perturbed and of the unperturbed oscillations, when a new equilibrium is reached after the application of the perturbation. In detail, since the “LFP” time-series are not strictly periodic and the phase relation is fixed only on average, the average time-lag between the perturbed and the unperturbed “LFPs” is measured by computing their crosscorrelogram over 50 oscillation cycles, starting from the 10-th cycle after the perturbation. This average time lag (readable from the position of the crosscorrelogram peak) is then translated into a phase-shift, by dividing it by the average period (estimated through autocorrelation analysis of the perturbed and unperturbed time-series over the same observation window). Vanishingly small perturbations do not induce long-lasting phase-shifts. Therefore, moderately large perturbation strengths have to be used. In this case, the dependence of on is sensibly non-linear. As a consequence, we evaluate directly the resulting for the used perturbation strength , plotted in Figure 6D (right). The qualitative shape of however does not depend strongly on . In particular, changes of affect the amplitude of the maximum phase-shift but not the perturbation phase for which it occurs. The curve is evaluated point-wise by applying perturbations at 100 different phases within a cycle. For each given phase, the perturbation is applied 100 times to 100 different cycles and the corresponding phase-shifts are averaged. Confidence intervals for are determined phase-by-phase by finding the 2.5-th and the 97.5-th percentile of the induced phase-shift distribution across these 100 trials.
For simplicity, we focus in the following on the case of areas, although our approach can be extended to larger motifs. The instantaneous phase-difference between two areas and is given by . For vanishing inter-areal coupling, the time evolution of is described by Eq. (1). The term is a functional of the phase response and of the limit cycle waveform of the uncoupled oscillating areas. For the rate model, is determined from analytic expressions of and of the rate oscillation limit cycle (note that the dependence on is replaced by a dependence on after phase-reduction) for . It can be expressed as , with:(5)The resulting expression is reported in the Supporting Text S1 and plotted in Figure 6B (left). Given Eq. (1), the phase shifts between the two areas and in stable phase-locked states correspond to top-down zero-crossings of the functional (i.e. zeroes with negative tangent slope, ).
For the network model, the waveform of “LFP” oscillations can be determined through simulations. Since not all oscillation cycles are identical, the limit cycle waveform is averaged over 100 different cycles –as for the determination of – to yield an average limit cycle . Then, it is possible to evaluate a functional , where:(6)The functional is plotted in Figure 6B (right) for the used perturbation strength . Although Eq. (1) does not exactly hold for the network model, the top-down zero-crossings of the functional (whose position only weakly depends on ) continue to provide an approximation of the phase shifts between the two areas and in stable phase-locked states. In particular it is possible to predict whether the stable lockings will be in-phase, anti-phase or out-of-phase.
Phase intervals in which the application of a pulsed perturbation leads to a change of effective connectivity directionality are determined theoretically as shown below. For and in a given phase-locking state, the phase of the leader area can be written as and the phase of the laggard area as . The application of a pulse perturbation of strength to the laggard area shifts the phase of the ongoing local oscillation to , where holds for the rate model in the case of small perturbations. If the achieved transient phase-shift between the two areas, , is falling into the basin of attraction of an alternative stable phase-locking (see Figure 6C), then a switching toward a different effective motif takes place. Considering the dynamics of the instantaneous phase-shift, determined by the functionals for the rate model and for the network model (see Figure 6B), switching will occur when:(7)Here, we consider perturbations which induce a phase advancement, because the positive part of both the PRC in the rate model and the empiric in the network model is larger than the negative part (see Figure 6D). For a fixed perturbation intensity , the condition (7) will be fulfilled only if when the phase of application of the perturbation falls within specific intervals, determined by the precise form of . These intervals are highlighted in green in Figure 6E and F. Analogous considerations can be done in order to determine the intervals for successful switching when perturbing the leader area (see Supporting Figure S2).
Let us consider first a structural motif with areas. Let and be the “LFP” time-series of the two areas and , and let quantize them into discrete levels (bins are equally sized). The continuous-valued “LFP” time-series are thus converted into strings of symbols and from a small alphabet [101]. Two transition probability matrices, and , where the lag is an arbitrary temporal scale on which causal interactions are probed, are then sampled as normalized multi-dimensional histograms over very long symbolic sequences. These probabilities are sampled separately for each specific fixed phase-locking configuration. Epochs in which the system switches to a different phase-locking configuration, as well as transients following state switchings are dropped. The evaluation of and is thus based on disconnected symbolic subsequences, including overall oscillation cycles. Then, following [42], the causal influence of area on area is defined as the Transfer Entropy:(8)where the sum runs over all the three indices , and of the transition matrices.
This quantity represents the Kullback-Leibler divergence [44] between the transition matrices and , analogous to a distance between probability distributions. Therefore, will vanish if and only if and coincide, i.e. if the transition probabilities between different “LFP” values of area do not depend on past “LFP” values of area . Conversely, this quantity will be strictly positive if these two transition matrices differ, i.e. if the past “LFP” values of area affect the evolution of the “LFP” in area .
We also measure the causal unbalancing [93]:(9)which is normalized in the range . A value close to zero denotes symmetric causal influences in the two directions, while large absolute values of signal the emergence of asymmetric effective connectivity motifs.
Considering now a structural motif with areas, equation (8) has to be modified in order to distinguish causal interactions which are direct (e.g. toward ) from interactions which are indirect (e.g. toward , but through ). A solution allowing to assess only direct causal influences is partialization [42], [71]. Indirect interactions from area to area involving a third intermediate area are filtered out by conditioning the transition matrices for the “LFP” activity of with resepect to the activity of the . Two conditional transition matrices, and , are then constructed and used to evaluate:(10)where the sum runs over all the four indices , , and . The effective connectivity in the panels C of Figures 3, 4 and 5 is computed using pTE according to equation (10).
Absolute values of depend strongly on the time-lag and on the number of discrete levels . Nevertheless, we find that relative strengths of causal influences are qualitatively unchanged over broad ranges of parameters, as displayed in the Supporting Figure S1. Furthermore the “plug-in” estimates of TE given by equations (8) and (10) suffer from finite-sampling biases, and a rigorous debiasing procedure is not yet known [43]. Therefore, for each value of and it is necessary to assess the significancy of the inferred causal interactions through comparison with suitably randomly resampled data [119]. To estimate the confidence intervals for the estimated TEs and the baseline for significancy we adopt a geometric bootstrap method [120], guaranteed to generate resampled time-series with similar auto- and cross-correlation properties up to a certain lag. This is important, since “LFP” time-series have a strong oscillatory component, whose correlation structure has to be maintained under resampling. Each resampled time-series consists of a concatenation of blocks sampled from the original time-series . Each has the same length as the original . Every upward crossing, i.e. every time at which crosses from below its time-averaged value , is a potential start-time for a block. The first element of each block is obtained by selecting randomly one of these potential start-times. Then, the block consists of the oscillation cycles following the chosen start-time, where the random integer follows a geometric distribution , with and an average block length of (we have taken oscillation cycles, longer than the mean correlation time for all the simulated “LFPs”). Randomly selected blocks are then concatenated up to the desired length.
When considering a structural motif involving more areas, the “LFP” time-series of each area can be resampled jointly or independently. When resampling jointly, matching starting points and block-lengths are selected for each block of the resampled time-series of each area, leading to resampled multivariate time-series in which the structure of causal influences should not be altered. The distribution of over jointly resampled “LFP” time-series describes then for each directed pair of areas and the strength of the corresponding effective connectivity link, as well as the fluctuations of this strength. Conversely, when resampling independently the time-series, start-points and block-lengths of the resampled blocks are chosen independently for each area. This second procedure leads by construction to causally independent time-series. Any residual between directed pairs of independently resampled “LFPs” indicates therefore systematic biases, rather than actual causal influences. For each directed pairs of areas and , significance of the corresponding causal interaction can be assessed by comparing the bootstrapped distributions of and of . This comparison is performed in Figures 3, 4 and 5 and in Supporting Figure S3D–E. Here, boxes indicate the median strength of for different directions and the corresponding confidence intervals, comprised between a lower extreme and and upper extreme , where and are respectively the first, the second and the third quartiles of the distribution of over jointly resampled time-series. Median values of and the corresponding confidence intervals, evaluated as before, are represented by horizontal dashed lines and a surrounding shaded band. When the distributions of and are not significantly different, a single baseline band is plotted. In this study, strengths and base-line for significancy of effective connectivity for each direction are validated based on, respectively, 500 jointly resampled and 500 independently resampled replicas.
Note that geometric bootstrap can be applied to arbitrary signals, and does not depend on their strict periodicity. However it is precisely the strong periodic component of our signals that makes necessary the use of geometric bootstrap techniques. Indeed, conventional bootstrap, strongly disrupting signal periodicity, would lead to artificially low thresholds for statistical significance of TE (not shown).
We evaluate information transmission between pairs of mono-synaptically connected cells in different areas, linked by a TL-synapse (TL pairs) or by a normally weak long-range synapse (control pairs). Inspired by [58], spike trains are digitized into binary streams , where = 1 or 0 respectively when neuron fires or does not fire during the -th local oscillation cycle (cycle counting is performed independently for each area and includes all the oscillation cycles following a common reference initial time). Note that neurons fire very sparsely and, due to the elevated degree of synchrony in our model, only in narrow temporal intervals centered around the peaks of the ongoing “LFP” oscillations. In particular, they fire at maximum once per oscillation cycle. Thus, this oscillatory spiking activity is naturally quantized in time and binning [58] is not required. For each considered directed pair of cells ( source cell, target cell), based on very long duration spike trains, we sample normalized histograms for three probability distributions: , and . When sampling the joint probability distribution we have to distinguish two cases: (i) If the presynaptic cell belongs to a leader area, i.e. the oscillation of the source area leads in phase over the oscillation of the target area of the considered synapse, then ; (ii) Conversely, if the presynaptic cell belongs to a laggard area, i.e. the oscillation of the target area leads in phase over the oscillation of the source area of the considered synapse, then . This means that we seek for spiking correlations only in pairs of spiking (or missed spiking) events in which the “effect” follows temporally its potential “cause”, since physical information transmission cannot occur backward in time. As for the estimation of TE (see previous section), the probabilities , and are sampled separately for each specific phase-locking configuration of the ongoing “LFPs”. Epochs in which the system switches to a different phase-locking configuration, as well as transients following state switchings are dropped. The evaluation of these probabilities is thus based on disconnected spike train chunks, including overall oscillation cycles. Based on these probabilities, the Shannon entropy H of the spike train of the presynaptic neuron (measuring the information content in its activity) is evaluated as:(11)and MI between pre- and postsynaptic cells as:(12)MI is then normalized by the entropy of the pre-synaptic cell, in order to measure the relative efficiency of information transmission along each TL or control synapse.
Statistics are taken over 400 pairs of cells per synapse set, i.e. one set of strong synapses per embedded TL, plus one set of (control) weak synapses. The box-plots in Figure 8C–D report median efficiencies of information transmission efficiencies (for different active effective connectivities), as well as their confidence intervals, estimated non-parametrically from distribution quartiles, as discussed above for TE. Both MI and H are computed for (finite) spike trains of the largest available length . Following [58], [121], it is possible to correct these results for finite-size sampling bias (see Supporting Figure S4). MI and H are computed again, based on randomly selected shorter matching sections of the full length spike trains. The results of obtained for various shorter lengths are then plotted against the so-called inverse data fraction , where correspond then to estimations based on full length spike trains. Quadratic extrapolation to provides a debiased estimation of . Note that, in order to allow a more direct comparison with the non-debiased TE analysis, the results plotted in Figure 8C–D do not include any finite-size correction. As a matter of fact, as discussed in Supporting Figure S4, finite size bias induces a small quantitative overestimation of information transmission efficiency (from to ), that does not affect qualitatively any of the results presented here.
|
10.1371/journal.pbio.1001828 | Mismatch Between Birth Date and Vegetation Phenology Slows the Demography of Roe Deer | Marked impacts of climate change on biodiversity have frequently been demonstrated, including temperature-related shifts in phenology and life-history traits. One potential major impact of climate change is the modification of synchronization between the phenology of different trophic levels. High phenotypic plasticity in laying date has allowed many bird species to track the increasingly early springs resulting from recent environmental change, but although changes in the timing of reproduction have been well studied in birds, these questions have only recently been addressed in mammals. To track peak resource availability, large herbivores like roe deer, with a widespread distribution across Europe, should also modify their life-history schedule in response to changes in vegetation phenology over time. In this study, we analysed the influence of climate change on the timing of roe deer births and the consequences for population demography and individual fitness. Our study provides a rare quantification of the demographic costs associated with the failure of a species to modify its phenology in response to a changing world. Given these fitness costs, the lack of response of roe deer birth dates to match the increasingly earlier onset of spring is in stark contrast with the marked phenotypic responses to climate change reported in many other mammals. We suggest that the lack of phenotypic plasticity in birth timing in roe deer is linked to its inability to track environmental cues of variation in resource availability for the timing of parturition.
| Climate change can alter the synchronization of life cycles between organisms at different points in the food chain. If species do not respond to climate change, the timing of peak resource availability may fail to match the timing of peak energy expenditure. Many bird species have been able to advance their laying date to match a change in the timing of caterpillar abundance. Herbivores are similarly expected to track changes in the timing of vegetation growth. In this study, we combine statistical analysis with demographic modeling to analyze the influence of a climate-driven shift in the timing of the spring vegetation flush on the birth date and demography of roe deer. In recent years, climate change has generated a marked increase in local temperatures and a progressively earlier vegetation flush. Despite these changes, we observed no shift in timing of the birth date of roe deer over the 27-year study period. This failure to track environmental change resulted in a mismatch between vegetation flush and birth date, which in turn caused a decrease in survival of the young, and hence a reduction in roe deer fitness. Birth date was under strong directional selection, but was not strongly heritable, suggesting that any evolutionary response of birth date to climate change might be limited. We suggest that a plastic response in birth date did not occur because reproduction is triggered by day length rather than resource availability in roe deer.
| Marked impacts of climate change on biodiversity have frequently been demonstrated, including temperature-related shifts in phenology and life-history traits [1]. Species living at high altitudes or latitudes are particularly affected by climate change [2],[3], but widespread species inhabiting temperate areas are also responding [4]. Global temperatures have risen by 0.89°C since 1901 [5], and this has led to an advance in the timing of key life-history events by, on average, 2.8 d per decade [6]. Earlier springs have caused phenological modifications in most taxonomic groups [1],[7]. The phenology of vegetation, particularly trees, has advanced with time (by 3.3 d per decade [6]). A failure of species to track these changes may have important demographic consequences that, in turn, could impact conservation and management issues. Changes in the timing of reproduction have been well studied in birds [8]–[10], but have only recently been considered in mammals [3],[4],[11]. These studies suggest that a change in the timing of peak resource availability typically generates a change in median laying or breeding date [9],[12],[13]. This response ensures that individuals can synchronize their energetic demands for offspring production and provisioning [14] with the period when environmental conditions are the most favorable [15].
A key question about the consequences of global change is which species can respond and how [16]. Although most studied species have responded to match resource availability with energy requirements, the response is not always exact or immediate [17],[18]. Phenotypic plasticity may play a major role in the adjustment of reproductive timing to earlier springs [19]. In a population of great tits, those individuals that could vary their reproductive timing the most had higher fitness [20]. But for individuals to shift their reproductive cycle in order to track environmental modifications, they require a reliable environmental cue [21],[22]. The timing of breeding is influenced by photoperiod in many species of birds and mammals [23]–[25], a cue that is clearly unaffected by climate change. Nonetheless, some species, including the great tit and red deer, rely on temperature [26],[27] to minimize the mismatch between birth timing and the peak resource availability. If earlier breeding increases fitness, selection could also drive a micro-evolutionary change in terms of advanced reproductive timing [8] as long as birth date is heritable. The relative role of phenotypic plasticity and micro-evolutionary change remains largely unquantified [28], although Réale et al. [4] showed that the advance in birth timing in red squirrel was mostly due to phenotypic plasticity rather than micro-evolution. Although some species have advanced their birth timing in response to increasing temperature, some species have not [29] whereas others have delayed their reproductive phenology [1]. For instance, Columbian ground squirrels have delayed their breeding phenology by 0.47 d per year over a period of 20 y, leading to a reduction in fitness by a half between 1993 and 2003 [3]. On the other hand, birth timing in caribou advanced at a much slower rate than the vegetation flush over a period of 33 years, so that the mismatch between birth timing and peak resource availability increased, causing calf production to decline [30],[31].
Births are highly seasonal and synchronous in most large herbivores [32], including roe deer [33],[34], in which more than 90% occur within 1 mo [35]. Roe deer females are income breeders and selectively feed on highly digestible and nutritious young shoots, especially during early lactation when energetic demand peaks [14]. Synchrony between births and the peak availability of high-quality vegetation is expected to be crucial for successful recruitment. The reproductive cycle of roe deer is unique among ungulates, including a phase of embryonic diapause that appears not to vary in duration among females [36]. As in reindeer where reproductive timing may be driven by day length [37], both ovulation and conception dates appear to be under the control of photoperiod [38], which could explain the lack of variation in parturition date across years for a given female [33],[35].
Focusing on the mismatch between birth date and plant phenology, we investigate how climate change is currently affecting roe deer fitness. Our study on the intensively monitored roe deer population at Trois Fontaines, eastern France, spans 27 y, from 1985 to 2011. We tested the three following predictions: (i) As most mammals studied so far have shown a response to climate change, we expected that roe deer births should occur earlier in response to the advance in vegetation phenology. (ii) Because parturition date varies little across years for a given roe deer female [35], indicating limited phenotypic plasticity, but because it has been shown to be heritable in mammals [39],[40], and markedly influences early offspring survival [40], we expected any change in birth timing to be mainly the result of natural selection. And (iii) because such micro-evolutionary responses are often delayed, we expected that despite any advance in birth timing, the mismatch between peak energetic demand and peak resource availability should likely increase over time, leading to negative impacts on roe deer performance. Our study provides a unique quantification, to our knowledge, of the demographic costs associated with the failure of a species to modify its phenology in response to a warming world.
Analysis of local weather variables revealed a strong local impact of climate change that translated into increasingly earlier and warmer springs over time. Annual spring (April to June) temperature increased by 0.07°C per year (SE = 0.02, p = 0.001, Figure 1) at Trois Fontaines over the study period. Analysis of flowering date in the vineyards of the Champagne region indicated that annual timing of plant phenology in the region had advanced by 0.6 d per year (SE = 0.18, p = 0.002, Figure 1) over this period, so that the peak in availability of high-quality resources for roe deer was increasingly early from 1985 to 2011. Spring mean temperature and flowering date in the Champagne region were negatively correlated over this period (ρ = −0.89, p<0.001). We therefore used flowering date as a proxy of the vegetation flush and compared it to annual variation in the timing of roe deer births. Roe deer give birth about 1 mo before the onset of flowering in Champagne because they preferentially feed on leaves and young shoots that become available before flowering. The mismatch between median birth date and vegetation phenology was estimated from the difference between median birth date and annual flowering date in the Champagne vineyards. We standardized this measure (by subtracting the observed value of mismatch in the first year (1985) from this variable) to obtain a relative measure of mismatch ranging from 0 in 1985 to 36 d in 2011.
In contradiction with our first prediction, annual median birth date did not occur earlier over time. Both the mean and median birth dates of roe deer at Trois Fontaines remained remarkably stable among years (based on 1,095 birth dates, mean = 136.1, SE = 8.56; time trend, t = −0.82, p = 0.421; and median = 136, the 16th of May, t = −1.23, p = 0.232 for mean and median birth dates, respectively, Figure S4A). Neither the spring mean temperature nor flowering date in the Champagne region had a detectable influence on median birth date (Pearson's product moment, ρ = −0.07, p = 0.730 and ρ = 0.11, p = 0.575 for spring mean temperature and flowering date, respectively). Consequently, the mismatch between median birth date and vegetation phenology increased by 0.54 d per year (SE = 0.20, p = 0.011, Figure 1) between 1985 and 2011. We did not find any correlation between median birth date and other environmental drivers (Table S1), suggesting that roe deer females are unable to track these potential environmental cues.
We investigated whether birth date of roe deer fulfilled the three necessary conditions for evolutionary change to occur: variability, heritability, and a selection pressure [41]. First, variation in parturition date among roe deer females has been recently quantified and found to be consistently high within several populations [35], with long-lived and/or heavier females (i.e., high-quality individuals) giving birth earlier than low-quality females [42]. Second, we found no strong statistical support for heritability in parturition date when estimated from the parent–offspring relationship based on 28 daughter–mother pairs (β = 0.234, SE = 0.13, p = 0.094, h2 = 0.127). Third, we identified directional selection favoring early births, with a strong negative relationship between individual birth date and individual early survival from May 12th onwards (on a logit scale β = −0.06, SE = 0.01, p<0.001, Figure S1). Note that year was included as a categorical variable in this model to control for interannual variation in environmental conditions affecting early survival. A model including a threshold effect of individual birth date on individual early survival provided a better fit than a linear (ΔAIC = 6.12) or a quadratic (ΔAIC = 0.50) model. Thus, a fawn born before May 12th had, on average, a 50% chance of surviving to 8 mo of age, whereas a fawn born on May 31st had, on average, only a 24% chance of surviving to that age. Adding a term describing interactive effects between birth date and year did not improve the fit of the model (ΔAIC = 9.28), indicating that the response of individual early survival to individual birth date was consistent across cohorts. Taken together these results suggest that we should not expect a strong micro-evolutionary response of parturition date in roe deer.
At the population level, cohort-specific survival (measured as the proportion of fawns that survived to the onset of winter each year) was negatively correlated with our index of mismatch (arcsine-square root transformation: β = −0.009, SE = 0.003, p = 0.012). Adding a quadratic term (ΔAIC = 1.95) or a threshold effect (ΔAIC = 1.86) of mismatch did not improve model fit. Early cohort-specific survival of juveniles decreased by 40% with an increase in mismatch of 1 mo (Figure 2).
At the individual level, the mismatch was a better predictor of early survival (R2 = 0.037) than birth date (R2 = 0.025) (ΔAIC = 8.90, Table S2; note that year was not included in these models because interannual variation was integrated within the mismatch variable). Individual early survival was constant when the mismatch was 16 d or less, but then decreased linearly beyond 16 d of mismatch (Table S2, Figure 3). When birth occurred at least 1 mo (35 d) before flowering date in Champagne's vineyards, a fawn had an expected probability of 0.5 of surviving to 8 mo of age, whereas this probability was only 0.25 when birth occurred 2 wk prior to flowering (Figure 3).
We built an IPM describing the temporal dynamics of parturition date in our roe deer population to quantify the impact of this increasing mismatch on roe deer fitness. The distribution of parturition date in the population at time t+1 depends on the distribution of parturition date at time t and on the four relationships linking parturition date with survival, recruitment, transition between two successive parturition events, and inheritance of parturition date between mother and offspring (see Text S1, Tables S2, S3, S4, and Figure S2 for further details [43],[44]). Annual flowering date in the vineyards of the Champagne region was included in the IPM to model the local shift in plant phenology. As most roe deer females give birth to two fawns [45] and the sex ratio is close to 0.5 at birth at Trois Fontaines [46], the recruitment function linking the number of female offspring a mother has successfully weaned given its parturition date was modeled by individual early survival (Table S5, Figure 3).
The IPM predicted mean parturition date to occur on the 17th of May each year, with a very slight, but statistically significant, advance of just 0.27 d over the whole 27-y study period (β = −0.010, SE = 0.003, p = 0.006, Figure S4B). The model also predicted earlier parturition as females aged, with 2-y-old females giving birth, on average, 5 d later than older females (Figure S5). The estimated population growth rate, and so the mean fitness of females in the population, decreased by 6% on average over the study period (β = −0.003, SE = 0.001, p = 0.008, Figure 1), from 1.23 in 1985 to 1.06 in 2011. Marked variability in environmental conditions between successive years often leads to a decrease in the arithmetic mean population growth rate [47]. Consequently, we estimated the geometric mean of population growth rate for successive periods of 4 y. The geometric mean population growth rate also decreased over the study period (β = −0.009, SE = 0.002, p = 0.002). Thus, the IPM allowed us to demonstrate a clear impact of the mismatch between energy demand and peak resource availability on mean fitness, which declined in this population of roe deer over the entire study period. In accordance with these results, we also observed a decrease in mean annual female reproductive success over the study period (β = −0.027, SE = 0.005, p<0.001; note that these data are independent of those used to build the IPM, see Figure S3).
This study has demonstrated that mean fitness is currently decreasing in this roe deer population due to the lack of response in parturition date to the increasingly early availability of high-quality resources induced by climate change. Warming at Trois Fontaines over the last 27 y (0.46°C per decade) was more than threefold greater than the average global expectation from the 2013 IPCC report on climate change (0.12 [0.08–0.14] per decade since 1951, [5]). This local warming has led to an advance in spring plant phenology [48] demonstrated by the advance in flowering date in Champagne's vineyards. In contrast to most other studied mammals that have been able to track resource availability by advancing their birth timing [3],[4],[13],[49], the median birth date of roe deer remained constant over years. This generated an increased mismatch between mean birth date and phenology of the vegetation such that at the end of the study period fawns were born relatively later with respect to the peak in availability of high quality resources. Post and Forchhammer [30] were the first to describe a negative impact of a mismatch between resource availability and birth timing on calf production in Greenland caribou. In our study, climate change over recent decades has had a similarly negative impact on early survival (i.e., a “climatic debt” [16]), both at the individual and at the population levels. Furthermore, we were able to show that this mismatch between parturition and the availability of highly digestible forage led to a decline in mean fitness of 6% over the study period, and of 14% between 1985 and those years when the vegetation flush was particularly early (2007 and 2011). This link between plant phenology and roe deer population dynamics, mostly driven through recruitment, is the likely mechanism for the observed decrease in population growth rate over time [50].
Our study provides an illustration of the probable fitness costs for species which do not respond to climate change. Indeed, the increasing mismatch between the peak of roe deer births and the onset of the vegetation flush in recent years had a negative impact on both early survival and mean fitness. Previous studies have reported an impact of climate change on recruitment [30],[51],[52]. However, the influence of mismatch on fitness and population dynamics has received much less attention. IPMs allow the phenotypic consequences of climate change to be explored (see also [49]), which is not possible using classical statistical methods. In contrast to a recent study on birds [53], we found that population growth rate of roe deer was not buffered against phenological mismatch. Mean fitness was most strongly affected during years when plant phenology was particularly early, for example, in 2007 and 2011 (λ = 1.07 versus λ = 1.23 for the first year of the study). We can therefore predict that this increasing mismatch will further increase the energetic costs of breeding for females [54] as spring phenology continues to advance in the future.
The lack of response in roe deer birth date to climate change provides a stark contrast with the previous findings on most mammalian species studied to date, which have shown phenotypic responses to climate change [3],[4],[13],[49]. Despite the clear selection pressure that we demonstrated, which should favor earlier births over time, we showed that a strong evolutionary change is not expected in roe deer. Indeed, we found no strong statistical support for heritability of birth date, despite the fact that parent–offspring regressions are known to overestimate heritabilities [55]. However, as the number of mother–daughter pairs (N = 28) available to assess heritability of birth date was low, further work on a much larger sample size is required to explore this question. Nonetheless, both the classical statistical approach and the IPM provided similar results and clearly indicated no change in roe deer reproductive timing. Although roe deer did not exhibit an evolutionary response to climate change, why have they not responded plastically [4],[20]?
Roe deer females appear unable to track environmental cues such as temperature to time their birth event. Birth timing in mammals is mainly driven by the date of conception and gestation length. Ovulation and, thereby, conception date is mainly under the control of photoperiod in roe deer [38]. Gestation begins with a phase of embryonic diapause that probably originally evolved to increase gestation length [56], but we expected this historical selection pressure for delayed birth to be counterbalanced by selection for earlier birth date in response to climate change over recent times. However, diapause appears to be triggered by an intrinsic mechanism involving the mother or even the embryo itself, and the 5-mo duration appears not to vary among females [36]. In many species, adaptive phenotypic plasticity has generated a response to changes in phenology (great tit [57], red squirrel [4]). In red deer, a species related to roe deer that is able to track earlier plant phenology [13], gestation length decreased with increasing average temperature in March [58]. In contrast, in roe deer, we have shown that parturition timing is independent of changes in temperature and in the onset of the vegetation flush, suggesting this lack of phenotypic plasticity in birth timing is associated with an inability to track environmental cues of variation in resource availability for the timing of parturition.
Earlier plant phenology is likely the main cause of the observed decrease in early survival, and thereby in mean fitness, in this roe deer population. Although the roe deer population consistently displayed positive growth over the 27-y study period (i.e., λ consistently higher than 1), population growth rate (and therefore average individual fitness) decreased in a continuous fashion by 6% over this period. Moreover, temperatures are expected to increase further in the future, causing the phenology of vegetation to advance still further. We suggest that these combined effects could impose a brake on the demographic and geographical expansion of roe deer, a common and previously successful species across all Europe [59].
Trois Fontaines (48°43N, 2°61W) is an enclosed 1,360 ha forest located near Saint-Dizier, at the border of Marne and Haute-Marne counties in north-eastern France. In spring (from mid-March to mid-June), the number of rainy days averaged 31.5 (ranging between 18 and 52 d during the study period 1985–2011) and the temperature averaged 10.4°C (ranging between −6.06°C and 15.00°C). The forest is dominated by oak (Quercus sp.) and beech (Fagus sylvatica). Roe deer feed mainly on coppice and the understory is dominated by hornbeam (Carpinus betulus), ivy (Hedera helix), and bramble (Rubus sp.).
The roe deer population at Trois Fontaines has been intensively monitored for more than 35 y by the Office National de la Chasse et de la Faune Sauvage based on a detailed Capture-Mark-Recapture program. Roe deer are individually marked using numbered collars and ear-tags. A systematic search for newborn fawns was conducted every year from late April to mid-June between 1985 and 2010 [60]. In 2011, searches ended earlier and the last fawn was found on May 20th. Fawns were handled by experienced people, ear-tagged, and weighed. Their sex was recorded and their age estimated to the nearest day using umbilicus characteristics and behaviour at marking [61]. Birth dates were back-calculated using these estimated ages and the day of capture. The average age at marking was about 5 d, and all fawns were marked within 20 d. The identity of the mother for a given fawn was established, when possible, through direct observations of lactating behavior or by the identification of an escaping female in the vicinity of the fawn. From January to March, annual capture sessions took place, with capture of more than 50% of the roe deer population each year, providing reliable information on the fate of animals marked at birth. Individual early survival was defined as the probability of survival of a fawn from birth to the next winter (see [62] for further details). At Trois Fontaines, fawn mortality was most likely associated with shortage of high-quality food because this population is not subject to marked predation or hunting pressure. At the population level, mean cohort-specific early survival was estimated using Capture-Recapture analyses [63] from 1985 to 2011. Cohort-specific early survival could not be estimated with accuracy for 2012 (see [62]), so these data were not included here. Data are available from Table S6.
The local daily temperature was collected from the Météo-France weather station of Saint-Dizier located at less than 5 km from the study site. Spring temperature (April to June) was used to assess the magnitude of climate warming at the local scale. To measure annual changes in vegetation phenology, we used the mean flowering date of vineyards in the Champagne region collected by the Comité Interprofessionnel du Vin de Champagne and available on the website of the French Observatoire National sur les Effets du Réchauffement Climatique (http://www.developpement-durable.gouv.fr/-Impacts-et-adaptation-ONERC-.html). Flowering date is little influenced by human activity and reliably reflects the phenology of the vegetation of that year. Moreover, flowering date in Champagne is highly correlated with the local sum of degree-days (ρ = −0.61, p<0.001), which is often used to index vegetation growth [13],[64], with spring temperature (ρ = −0.83, p<0.001) and with annual mean temperature (ρ = −0.57, p<0.001). As a consequence, flowering date in Champagne reliably indexes the overall changes in plant phenology over years. When flowering date in Champagne was early in the year, we assumed that the availability of high-quality resources for roe deer provided by spring vegetation was also early. Consequently, we used the difference between birth date and flowering date in Champagne as a measure of the mismatch between birth date and peak resource availability.
To assess temporal trends in local temperature, vegetation phenology, and birth date, we fitted linear regressions with a Gaussian error. Subsequently, we examined the relationships of cohort-specific median birth date with mean spring temperature and with flowering date in Champagne to test whether birth date tracked climate change. To quantify the mismatch between median birth date and the vegetation flush at the population level, we subtracted the median birth date from the annual flowering date.
Available data were not detailed enough to build pedigrees, so we measured heritability using the weighted regression of the median parturition date of each daughter against the median parturition date of her mother [65]. To assess whether roe deer birth date was under selection, we analyzed the relationship between birth date and early survival at the level of the individual with a generalized linear model and a logit link. We included year in the model to control for interannual variation in environmental conditions. We tested for linear, quadratic, and threshold effects of birth date on individual early survival. Finally, we tested for an interaction between birth date and year to investigate whether the selection pressure was similar over time.
To assess whether roe deer exhibited a phenotypic response to climate change, both at the population and individual levels, we used mean cohort-specific early survival after an arcsine-square root transformation; we investigated the relationships between cohort-specific early survival and the mismatch, testing for linear, quadratic, and threshold effects of the mismatch. Each point of the regression was weighted (using the inverse of the variance of cohort-specific early survival) to account for uncertainty in the estimates of cohort-specific survival. Then, we investigated whether the mismatch was a better predictor of early survival than birth date at the individual level. We did not include year in this model as among-year variations were integrated within the mismatch variable. We tested for linear, quadratic, and threshold effects of birth date or the birth date–vegetation phenology mismatch on individual early survival with a generalized linear model and a logit link. We compared the relative fit of the different models using the Akaike Information Criterion (AIC).
To investigate the influence of plant phenology on mean fitness, we built an Integral Projection Model (IPM, [43],[44]) describing the dynamics of parturition date in the population. Selection and estimations for the models describing the four functions defining the IPM (survival, recruitment, transition, and inheritance) are detailed in the Supporting Information section. This IPM allowed us to investigate the influence of the timing of peak resource availability on the outputs of the model: the annual asymptotic population growth rate, in other words, annual mean fitness over the study period. As previous studies in this population have revealed no effect of density dependence on any of the demographic parameters, we did not include density in our demographic analysis (supplementary material of [50]).
|
10.1371/journal.ppat.1003165 | Structural and Functional Analysis of the CspB Protease Required for Clostridium Spore Germination | Spores are the major transmissive form of the nosocomial pathogen Clostridium difficile, a leading cause of healthcare-associated diarrhea worldwide. Successful transmission of C. difficile requires that its hardy, resistant spores germinate into vegetative cells in the gastrointestinal tract. A critical step during this process is the degradation of the spore cortex, a thick layer of peptidoglycan surrounding the spore core. In Clostridium sp., cortex degradation depends on the proteolytic activation of the cortex hydrolase, SleC. Previous studies have implicated Csps as being necessary for SleC cleavage during germination; however, their mechanism of action has remained poorly characterized. In this study, we demonstrate that CspB is a subtilisin-like serine protease whose activity is essential for efficient SleC cleavage and C. difficile spore germination. By solving the first crystal structure of a Csp family member, CspB, to 1.6 Å, we identify key structural domains within CspB. In contrast with all previously solved structures of prokaryotic subtilases, the CspB prodomain remains tightly bound to the wildtype subtilase domain and sterically occludes a catalytically competent active site. The structure, combined with biochemical and genetic analyses, reveals that Csp proteases contain a unique jellyroll domain insertion critical for stabilizing the protease in vitro and in C. difficile. Collectively, our study provides the first molecular insight into CspB activity and function. These studies may inform the development of inhibitors that can prevent clostridial spore germination and thus disease transmission.
| Clostridium difficile is the leading cause of health-care associated diarrhea worldwide. C. difficile infections begin when its spores transform into vegetative cells during a process called germination. In Clostridium sp., germination requires that the spore cortex, a thick, protective layer, be removed by the cortex hydrolase SleC. While previous studies have shown that SleC activity depends on a subtilisin-like protease, CspB, the mechanisms regulating CspB function have not been characterized. In this study, we solved the first crystal structure of the Csp family of proteases and identified its key functional regions. We determined that CspB carries a unique jellyroll domain required for stabilizing the protein both in vitro and in C. difficile and a prodomain required for proper folding of the protease. Unlike all other prokaryotic subtilisin-like proteases, the prodomain remains bound to CspB and inhibits its protease activity until the germination signal is sensed. Our study provides new insight into how germination is regulated in C. difficile and may inform the development of inhibitors that can prevent germination and thus C. difficile transmission.
| The Gram-positive, spore-forming obligate anaerobe Clostridium difficile is the leading cause of nosocomial diarrhea worldwide [1]–[3]. The symptoms of C. difficile-associated disease (CDAD) range from mild diarrhea to pseudomembranous colitis and even death. Although CDAD is primarily a toxin-mediated disease [3], [4], the high cost and difficulty in treating C. difficile infections largely arises from its ability to form endospores [5], [6]. Because spores are metabolically dormant and intrinsically resistant to harsh physical insults [3], [7]–[9], they allow C. difficile to resist antibiotic treatment and persist in healthcare-associated settings. Thus, spores are the primary vectors for transmission [10] and the cause of recurrent infections, the latter of which occurs in ∼25% of cases and can lead to severe CDAD [6], [11].
In order to initiate an infection, C. difficile spores ingested from the environment must germinate into toxin-producing vegetative cells in the intestinal tract [1], [3], [12]. Similar to other spore-forming bacteria, C. difficile spores germinate specifically in response to small molecules known as germinants [13], [14]. For C. difficile, these germinants are bile salts, which are abundant in the small intestine [15]–[17]. While germinants have been identified for a number of bacterial species, the molecular events that occur upon germinant sensing remain poorly characterized [13], [14], [18]. Shortly after germinant addition, cortex hydrolases become activated and degrade the spore cortex, a thick protective layer of modified peptidoglycan. Because the cortex maintains the spore in a dehydrated, metabolically dormant state, the removal of this physical constraint is essential for germination to proceed and metabolism to resume in the spore core [13], [14], [18]. Nevertheless, despite the importance of cortex hydrolysis, little is known about the molecular mechanisms that regulate cortex hydrolase activity.
In the Clostridia, the primary cortex hydrolase appears to be SleC, since disruption of sleC in both C. difficile [19] and the related foodborne pathogen C. perfringens [20] results in a severe germination defect. In C. perfringens, SleC undergoes several processing events. During sporulation, the N-terminal prepeptide is removed to produce pro-SleC, which consists of an N-terminal propeptide attached to the hydrolase domain [21]–[24]. During germination, the zymogen pro-SleC is cleaved at a conserved site to release the propeptide (Figure 1A); this event appears to activate its hydrolase activity [22], [25].
Biochemical analyses of C. perfringens germination exudates have shown that a fraction containing three serine proteases (CspA, CspB, and CspC) can proteolytically activate SleC hydrolase activity in vitro [25]. CspB alone appears sufficient to activate SleC, since the food-poisoning isolate SM101 encodes only the cspB gene, and disruption of this gene abrogates SleC cleavage and spore germination [26]. In the genome of C. difficile, three csp homologs are present in a bicistronic operon (cspBA-cspC, Figure S1), with cspB and cspA being present as a gene fusion [13]. Since disruption of the cspBA-cspC operon by transposon insertion results in a severe germination defect [27], cortex hydrolysis in C. difficile and C. perfringens would appear to be similarly regulated.
While studies have shown that SleC and CspB are key players during germination, the molecular mechanisms regulating their function are unknown. The sequence homology between Csp proteases (Csps) and the subtilase protease family [25] provides a starting point for understanding how Csps transduce the germination signal and activate SleC. Subtilases are serine proteases that contain a catalytic triad in the order of Asp, His and Ser [28], [29], and most subtilases are produced as pro-enzymes that autoproteolytically remove their prodomain [28], [30], [31]. While Csps purified from C. perfringens germination exudates are N-terminally processed [25], whether Csps are regulated in a manner analogous to other subtilases is unclear. Indeed, whether Csps actually have protease activity has not yet been directly demonstrated.
In this study, we investigated the protease activity of CspBA in C. difficile. By analyzing CspBA in sporulating C. difficile and purified spores, we demonstrate that CspBA is processed to CspB during spore assembly and that CspB undergoes autoprocessing. We also present the first crystal structure of the conserved Csp family of proteases at 1.6 Å resolution and define its key structural domains. These biochemical and mutational analyses reveal that, in contrast to previously characterized prokaryotic subtilases, wildtype CspB forms a stable complex with its prodomain. Similar to other subtilases [31], the prodomain acts as both an intramolecular chaperone and an inhibitor of CspB protease activity. These findings provide the first molecular insight into Csp function and may inform the development of strategies that can either prematurely activate C. difficile spore germination in the environment or prevent spore germination during disease transmission and recurrence.
The cspBA fusion gene is encoded in the genomes of only five clostridial species (Figure S2). C. difficile is unique among these species in that the CspA portion of CspBA (CD2247) lacks an intact catalytic triad (Figures 1A and S2). In order to determine whether CspBA is produced as a fusion protein, we raised antibodies against the CspB portion of CspBA and analyzed CspB production in both sporulating cells and purified spores by Western blotting. As a control, we constructed a targeted gene disruption [32] of the cspBA-cspC (cd2247-cd2246) operon (Figure S1). In sporulating C. difficile cells, the anti-CspB antibody detected two polypeptides of ∼130 kDa and ∼55 kDa (Figure 1B). The former corresponds to the predicted MW of CspBA of 125 kDa, while the latter corresponds to the size of Csp proteases detected in C. perfringens spores (∼60 kDa) [25]. Notably, the ∼55 kDa protein was enriched in purified spores, suggesting that interdomain cleavage of CspBA occurs during spore formation and that CspB may be preferentially incorporated into the developing spore. Although the mutant strains exhibited similar levels of sporulation (Figure 1C), CspB levels in sleC− mutants spores were consistently ∼3-fold lower than in wildtype spores (Figure 1B). Nevertheless, these results indicate that CspBA is processed to CspB during C. difficile spore assembly.
While the cspBAC locus was previously identified by transposon mutagenesis as being essential for C. difficile spore germination [27], the effect of CspBA on SleC cortex hydrolase processing was not tested. To determine whether loss of CspBA prevents SleC processing, we analyzed SleC cleavage in response to a bile salt germinant [16] by Western blotting. As predicted, disruption of cspBAC in C. difficile prevented SleC cleavage during germination (Figure 1B), and this defect could be complemented by ectopic expression of the cspBAC locus from a multicopy plasmid [33] (Figure S1). Thus, the cspBAC locus appears to regulate SleC activity in a manner similar to C. perfringens [20], [22], [25], [26].
In order to gain insight into the mechanism by which CspBA activates SleC during germination, we conducted structure-function analyses of the CspB domain of CspBA, since CspB is the only Csp protease encoded by C. difficile with an intact catalytic triad (Figure 1A). Based on its homology to subtilases [25] (Figure S3), we hypothesized that CspB is synthesized as a pro-enzyme that undergoes autoprocessing. To test this hypothesis, we recombinantly produced wildtype CspB difficile (residues 1–548 of CspBA) and CspB perfringens, along with mutants with the catalytic serine inactivated, and compared their apparent MW by SDS-PAGE. Whereas mutation of the catalytic serine caused both CspB difficile and CspB perfringens to run at their expected MWs of ∼60 kDa, the wildtype CspB proteins migrated with MWs of ∼55 kDa (Figure 2A). Thus, Csps autocatalytically remove their prodomain in a manner similar to other subtilases.
Using Edman degradation, we mapped the autoprocessing site of CspB difficile (1–548 aa) to Gln66 (data not shown). Alignment of this autocleavage site with previously mapped processing sites for CspA, CspB, and CspC of C. perfringens [25] revealed that Csps cleave at a similar position relative to their mature domains (Figure 2B and S4). Given the limited conservation in amino acid sequence around the Csp autoprocessing sites (Figures 2B and S4), we tested whether CspB recognizes specific amino acid residues upstream of the cleavage site. Mutation of the CspB perfringens P1 serine to a bulky, charged Arg did not affect autoprocessing (P1 refers to the residue N-terminal to the scissile bond based on the Schecter and Berger convention [34], Figure 2A); similarly, mutation of the P3-P1 residues to alanine did not affect CspB perfringens and CspB difficile autoprocessing (Figure 2) or the position of cleavage (data not shown). In contrast, deletion of the P3-P1 residues (ΔYTS) of CspB perfringens markedly reduced prodomain cleavage (Figure 2A), suggesting that the length of the prodomain affects substrate recognition or binding.
While these findings highlighted similarities between Csps and other subtilases, all CspB proteins capable of undergoing autoprocessing unexpectedly remained in complex with their prodomain following multiple rounds of purification (Figure 2A). In contrast, all previously characterized prokaryotic subtilases degrade their prodomain shortly after autoprocessing [31]. To gain insight into the interaction between the prodomain and subtilase domain, we determined the crystal structure of the CspB homolog from C. perfringens. CspB contains a subtilase domain that is similar to other subtilisin-like proteases [31], with the active site tucked within a conserved fold comprised of a six-stranded antiparallel β-sheet that is sandwiched between four conserved α-helices (Figure 3A). The catalytic triad of the active site of CspB superimposes directly with other active subtilisin-like proteases (Figure 3C), with an RMSD over the Cα atoms of only 0.11 Å between the catalytic triads of CspB and Tk-SP, the most structurally related enzyme from Thermococcus kodakaraensis as determined by the Dali server [35], [36].
In contrast with all previously solved prokaryotic subtilase structures, the autoprocessed prodomain stays bound to the wildtype, mature enzyme in our CspB structure. Notably, structures of prokaryotic subtilases in complex with their prodomain exist only for active site mutants [37]–[41]. The prodomain of CspB exhibits a similar structural organization to these subtilases, consisting of a 4-stranded antiparallel β-sheet and 3 α-helices (Figure 3A), with an additional β-strand extending into the catalytic cleft. The C-terminus of the CspB prodomain also extends directly into the oxyanion hole, with 19 hydrogen bonds stabilizing the intimate interaction between the C-terminal P6-P1 residues of the prodomain and the catalytic cleft (Figure S5 and Table S1 in Text S1). The prodomain-subtilase domain interface buries 1,472 Å2 of accessible surface area (Figure 3C).
A second major feature that distinguishes the structure of CspB from other subtilases is the interruption of the protease domain by an ∼130 aa insertion (Figure S3). This insertion assumes a β-barrel jellyroll fold, consisting of nine antiparallel β-strands that pack in a small hydrophobic core. The jellyroll domain interacts with both the prodomain and subtilase domain (Figure 3 and S5, Table S1 in Text S1). Although a similar jellyroll fold is present in the archaeal subtilisin Tk-SP [35] (Figures 3D and 4A, RMSD of 2.1 Å over 81 Cα atoms), the Tk-SP jellyroll domain is a C-terminal extension that interacts exclusively with the subtilase domain (Figure 3D). Nevertheless, both Tk-SP and CspB hold their jellyroll domains tightly in place with 22 and 19 bonds (primarily hydrogen bonds, Table S1 in Text S1), with a buried surface area between the jellyroll domain and subtilase domain of 1,115 Å2 and 1,018 Å2, respectively.
In Tk-SP, the jellyroll domain has been shown to stabilize enzyme activity at high temperatures (>90°C) [35]. To test whether the jellyroll domain might similarly stabilize CspB, we compared the susceptibility of wildtype CspB and a mutant lacking the jellyroll domain (CspB Δjelly) to limited proteolysis. In vitro structure-function analyses were done on CspB perfringens rather than C. difficile because the structure was solved for CspB perfringens. In the presence of increasing concentrations of chymotrypsin, wildtype CspB exhibited remarkable resistance to degradation even when chymotrypsin levels were approximately equimolar to CspB (0.5 mg/mL or 20 µM chymotrypsin, Figure 4B). While mutation of the catalytic serine had little effect on CspB degradation, deletion of the jellyroll domain sensitized the mutant to chymotrypsin digestion at 50 ng/mL (Figure 4B). Loss of the jellyroll domain also reduced the efficiency of CspB autoprocessing, since both uncleaved and mature CspB Δjelly were observed following purification. In contrast, only uncleaved CspB Δjelly was observed upon mutation of the catalytic serine (Figure 4B). Taken together, these results implicate the jellyroll domain in (1) positioning the prodomain to undergo autocleavage and (2) markedly restraining the conformational flexibility of CspB.
Having identified functions for the jellyroll domain, we next investigated the role of the prodomain in regulating CspB activity. For many subtilases, the prodomain acts as an intramolecular chaperone that catalyzes proper folding of the subtilase domain; once folding is complete, the mature enzyme autocatalytically separates the prodomain from its subtilase domain [30], [31]. In most subtilases, the prodomain acts as a temporary inhibitor until it is autoproteolytically removed [31], [42]. To determine the extent to which Csps follow this model of subtilase maturation, we examined the chaperone activity of the CspB prodomain. Similar to other subtilases, deletion of the prodomain dramatically reduced the solubility and yield of mature CspB, while co-expression of the prodomain in trans restored folding to the subtilase domain (Figure S6). The chaperone activity of the prodomain was highly specific for CspB perfringens, since co-expression of the prodomains of CspB difficile and CspC perfringens in trans only marginally restored folding to the subtilase domain (Figure S6).
Indeed, the CspB subtilase domain recognizes its prodomain with an extensive network of interactions, consisting of 27 hydrogen bonds and three salt bridges (Figure S5 and Table S1 in Text S1). The prodomain adopts a similar fold to the prodomains of related subtilisin-like proteases (Figure 5A), with the C-terminal region extending deep into the catalytic cleft (Figure 3A). The 94 Cα atoms of the prodomain align with an RMSD of 2.4 Å compared to the Tk-SP prodomain and 2.5 Å when compared to the mammalian proprotein convertase subtilisin kexin type 9 (PCSK9), respectively [43].
We compared the CspB prodomain to the PCSK9 prodomain because PCSK9 is the only other example of a wildtype subtilase that remains bound to its prodomain [43]–[45], whereas the prodomain of Tk-SP only stays bound if the catalytic serine of Tk-SP is mutated to cysteine [40], [41]. Since we did not observe any obvious structural differences to account for the difference in prodomain retention, we examined the free energy of dissociation of prodomains from their cognate subtilase domains using PDBe PISA, which is a computational server for examining interaction interfaces on proteins [46]. This analysis revealed that CspB and PCSK9 have the highest energy barriers to prodomain dissociation relative to other subtilases bound to their cognate prodomain or inhibitor (ΔG = −19.2 and −17.7 kcal/mol, respectively, Figure 5B). Interestingly, while most of the interactions holding the prodomain to the subtilase domain are not sequence specific (Table S1 in Text S1), with 15 bonds directed at backbone atoms, there are a few salt bridges that mediate specific recognition of the prodomain (Figure 5C). These salt bridges occur between Glu35/Glu59 of the prodomain and Arg231 of the subtilase domain and between Lys91 of the prodomain and Asp257 of the subtilase domain (Figure 5C). To determine the contribution of these salt bridges to CspB folding, we mutated each salt bridge residue and analyzed the effect on CspB solubility. Mutations of Glu35 to glutamine and Glu59 to alanine slightly reduced yields relative to wildtype, whereas mutation of Arg231 to glutamine strongly decreased recovery of CspB (Figure 5D), presumably because it disrupts both potential salt bridge interactions. Flipping the charges on Glu35 and Arg231 (E35R or R231E, respectively) also significantly reduced CspB yields, while swapping the Glu35-Arg231 salt bridge (E35R-R231E) failed to rescue CspB solubility. In contrast, flipping the charge on Lys91 to aspartate (K91D), which forms a salt bridge with subtilase domain residue Asp257, had little effect on K91D solubility relative to wildtype CspB (Figure 5C and 5D, Table S1 in Text S1). Taken together, these results highlight the importance of the Glu35-Arg231 and Glu59-Arg231 prodomain-subtilase domain salt bridges in promoting subtilase domain folding.
Having demonstrated the intramolecular chaperone activity of the prodomain, we next tested whether the prodomain functions as an inhibitor similar to other subtilases. Consistent with this hypothesis, the C-terminal P3-P1 residues of the prodomain bind the catalytic site in a manner analogous to an inhibitory peptide, fitting snugly within the catalytic cleft and presumably occluding access to the active site residues (Figure 6A). The S1 and S2 binding pockets perfectly accommodate the P1 serine and P2 threonine (P1 refers to the residue N-terminal to the cleavage site; S1 refers to the P1 substrate binding pocket). The bulky P3 tyrosine residue is wedged between Arg222 and Ser254 of the subtilase (Figure S5 and Table S1 in Text S1). The C-terminal P1-P3 prodomain residues form a total of 13 bonds to the S1–S3 regions of the subtilase domain. The P1 Ser96 forms seven hydrogen bonds with NE2 of catalytic His183, Ser252, Asn287, Thr493 and catalytic Ser494; P2 Thr95 forms four hydrogen bonds to different atoms of Arg222; and P3 Tyr94 forms hydrogen bonds to both the backbone amide and carbonyl of Ser254.
To test whether these residues block substrate access to the CspB active site, we used a small activity-based probe (FP-Rh, Figure 6B) to detect CspB catalytic activity. The fluorophosphonate electrophilic group of the probe reacts exclusively with catalytically competent serine hydrolases such as the subtilisins, which are a subfamily of the subtilases [47]. Nucleophilic attack by the catalytic serine results in the probe becoming covalently bound to the catalytic serine, while the rhodamine tag allows for detection of the covalently labeled enzyme by fluorescent gel scanning. Incubation of either wildtype or catalytically inactive S461A CspB with FP-Rh failed to produce detectable fluorescence, implying that the active site is inaccessible in the wildtype enzyme (Figure 6C). In contrast, mutation of the P3-P1 residues (YTS/AAA) produced a CspB variant that could be labeled on its catalytic serine, suggesting that the C-terminal prodomain residues act as gatekeepers to a catalytically competent active site. Accordingly, truncation of the C-terminal gatekeeper of the prodomain expressed in trans of residues YTS (P3-P1) or LYTS (P4-P1) permitted labeling of the CspB active site, whereas the full-length prodomain expressed in trans prevented labeling (Figure 6C). Taken together, these results indicate that the C-terminal YTS prodomain residues inhibit CspB activity.
Having identified key structural features of CspB perfringens in vitro, we next tested their functional significance in regulating CspBA activity in C. difficile. To this end, we cloned cspBAC complementation constructs in which the jellyroll domain was deleted (Δjelly, Figure 7A) or the active site serine was mutated (S461A). The cspBAC constructs were expressed from their native cspBA promoter on a multicopy plasmid (pMTL83151) [33]. Deletion of the jellyroll domain appeared to destabilize CspBA, since CspBA Δjelly levels were markedly reduced relative to wildtype and the cspBAC complementation strain and degradation products were apparent (Figure 7B). In contrast, mutation of the catalytic serine (S461A) did not affect CspBA levels relative to the cspBAC complementation strain, although CspBA S461A failed to undergo autoprocessing (Figure 7B). In purified spores, the predominant form of CspB was autoprocessed (m-CspB) in wildtype and cspBAC-complemented spores, whereas the predominant form of CspB in S461A spores was not autoprocessed (Figure 7C). Given that CspBA S461A was still processed at the CspB-CspA junction, an as-yet-unidentified protease apparently separates CspB from CspA.
To determine the role of CspBA autoprocessing in C. difficile spore germination, we examined the ability of S461A mutant spores to germinate in response to bile salts. Relative to wildtype and cspBAC-complemented spores, S461A mutant spores exhibited an ∼20-fold defect in germination and SleC cleavage (Figure 7C), while loss of the jellyroll domain (Δjelly) reduced spore germination by ∼70-fold (Figure 7C). Nevertheless, loss of CspBA and CspC production in the cspBAC− mutant produced a more severe phenotype than loss of the catalytic activity (S461A) or jellyroll domain (Δjelly) of CspBA. Taken together, these results indicate that CspB catalytic activity and its jellyroll domain are required for efficient C. difficile spore germination.
The observation that ∼5% of pro-SleC undergoes cleavage during germination of S461A mutant spores (Figure 7C) raised the question as to how SleC was being activated in the absence of CspB protease activity. One possibility is that a redundant protease cleaves SleC during germination of S461A mutant spores. Another possibility is that CspB activates a second protease that directly cleaves SleC. While this latter model is more complicated, it reflects how the subtilisin-like proprotein convertase PCSK9 indirectly regulates low-density lipoprotein receptor (LDLR) levels. Rather than enzymatically degrading LDLR, PCSK9 binds and targets LDLR to the lysosome [42], [48]. However, in order to bind LDLR, PCSK9 must undergo autoprocessing to form a non-covalent complex with its prodomain; only after autoprocessing can PCSK9 recognize LDLR [42], [48]. As a result, PCSK9 is the only other wildtype subtilisin-like protease that retains its prodomain in its crystal structure following autoprocessing [43]–[45].
If CspB activity is regulated similarly to PCSK9, CspB protease activity should be dispensable once autoprocessing has occurred. To test this hypothesis, we co-expressed the CspBA prodomain with a CspBA variant lacking its prodomain such that the CspBA produced is identical to wildtype CspBA after autoprocessing (Q66, Figure 8A). The prodomain was also co-expressed with a catalytically inactive CspBA variant lacking its prodomain (Q66/S461A, Figure 8A). As predicted, Q66 and Q66/S461A transcomplementation mutants produced CspBA variants that were indistinguishable in size from wildtype in sporulating cells (Figure 8B) and purified spores (Figure 8C), although more CspBA fusion protein was observed in the transcomplementation mutant spores relative to wild type (Figure 8C). Nevertheless, Q66/S461A mutant spores exhibited a 10-fold defect in both germination and SleC cleavage relative to wildtype and Q66 mutant spores. This result indicates that the catalytic activity of CspBA is required for efficient SleC cleavage downstream of CspBA autoprocessing.
Spore germination is essential for Clostridium sp. pathogens such as C. perfringens and C. difficile to initiate infection [12], [13]. A critical step during germination is the degradation of the thick, protective cortex layer surrounding the spore core by cortex hydrolases [13], [14], [18]. However, despite their functional importance, little is known about the molecular mechanisms that control cortex hydrolase activity. In this study, we provide the first molecular insight into cortex hydrolase regulation by solving the structure of CspB, a protease required for cortex hydrolase activation. Combined with our functional analyses of CspB in vitro and in vivo, the structure reveals that Csps are subtilisin-like proteases with two distinctive functional features: a central jellyroll domain and a retained prodomain.
The central β-barrel jellyroll domain of CspB interrupts the subtilase domain and wedges itself tightly between the subtilase domain and prodomain in three-dimensional space (Figure 3C). This unique position is likely critical for CspB function, since the jellyroll domain markedly restrains the conformational mobility of CspB through extensive and specific interactions at the subtilase-jellyroll domain interface (Figure 4B and S5). The rigidity conferred by the jellyroll domain presumably helps CspB survive the environmental extremes that spores can encounter, such as freeze-thaw cycles and boiling temperatures [9]. The jellyroll domain also facilitates CspB autoprocessing in vitro (Figure 4B), indicative of a role in helping CspB adopt the correct subtilase fold. Consistent with this proposal, deletion of the jellyroll domain in C. difficile markedly reduced CspBA levels relative to wild type (Figure 7B).
In these respects, the jellyroll domain is more functionally analogous to the β-barrel P-domains of kexin-like subtilisins than to the jellyroll domain of prokaryotic Tk-SP subtilisin. Like the CspB jellyroll domain, the P-domain of kexin-like proteases, such as the mammalian enzyme furin, is important for autoprocessing, folding, stability, and activity of the subtilase domain [31], [42], [49]–[51]. In contrast, the jellyroll domain of Tk-SP is dispensable for autoprocessing, protein folding and activity in vitro, despite being important for Tk-SP thermostability [35].
The retention of the CspB prodomain is another unique feature identified by our study. Unlike the majority of subtilisin-like proteases, the prodomain stays bound to the wildtype subtilase domain via a network of interactions that result in tighter prodomain binding relative to other subtilases (Figure 5B and S5). Prodomain binding to its cognate protease appears highly specific, since prodomain swapping does not result in efficient folding of CspB (Figure S6). This conclusion is consistent with the limited sequence conservation of prodomains across Csps (Figure S3); indeed, even the salt bridges critical for prodomain chaperone activity (Figure 5D) are not conserved. Despite the low level of sequence conservation, the position of prodomain autoprocessing is highly conserved (Figure 2B), and a small internal deletion of the prodomain disrupts autocleavage even though diverse residues are tolerated at the P1 position (Figure 2A).
Mechanistically, Csps exhibit less specificity in P1 substrate recognition than most subtilases [52]–[56]. Nevertheless, while residues around the prodomain cleavage site do not affect autocleavage efficiency, they do control active site accessibility after autoprocessing, excluding even a small, highly reactive, serine protease probe in vitro (Figure 6C). Taken together, Csps appear more functionally similar to the site-specific kexin-like protease subfamily than to the highly processive subtilisin subfamily [28], [31]. Similar to kexin-like proteases, Csps cleave their putative substrate, SleC, at a single site during germination [22] (Figure 1B) and remain more closely associated with their prodomain following autoprocessing [31], [42]. By contrast, subtilisin subfamily members such as Tk-SP function as major degradative enzymes that rapidly degrade their prodomain following autoprocessing [31].
While these observations provide new insight into the structure and function of Csp proteases, they raise a number of questions for future study. Does the prodomain remain associated with autoprocessed CspB in dormant spores as it does in vitro? If the prodomain stays bound to mature CspB in dormant spores, what happens to the prodomain during germination? Given that chymotrypsin cannot access numerous prodomain cleavage sites during extended incubation in vitro (Figure 4), a significant change in CspB conformational mobility would appear to be required for the prodomain to be degraded and its putative substrate SleC to gain access to the CspB substrate binding pocket.
Another question raised by our study is the role of CspC in regulating germination in C. difficile. Given that Δjelly, S461A, and Q66/S461A mutant spores exhibit germination defects that are >100-fold less severe than cspBAC− spores and that a major difference between these mutant spores is the absence of CspC in cspBAC− mutant spores (Figures 7 and 8), catalytically inactive CspC (Figure 1A) may play a role in SleC activation. Recent data suggests that CspC helps transduce the germination signal to CspB (J. Sorg, personal communication). In addition, it is unclear what fraction of pro-SleC must be proteolytically activated to induce successful spore germination. Approximately 5% of spores of the CspBA catalytic mutant S461A successfully germinate, which correlates with a small fraction of pro-SleC undergoing processing in the mutant strain (Figure 8). This result suggests that only a small fraction of SleC must be proteolytically activated in order to mediate spore germination in some cells; alternatively, a small fraction of S461A spores could efficiently cleave pro-SleC and thus germinate successfully.
While further experimentation is needed, the work presented here provides the first structure-function analyses of Csp proteases in vitro and in vivo and lays the groundwork for mechanistically addressing how the germination pathway senses and integrates the germination signal. Furthermore, this study may provide the structural basis for designing therapeutics that either block prodomain and/or jellyroll domain binding to the CspBA subtilase domain during spore formation or prematurely activate CspBA to induce cortex hydrolysis. These CspBA agonists or antagonists could prevent C. difficile transmission and disease recurrence.
Bacterial strains and plasmids used in this study are listed in Table S2 in Text S1. The C. difficile strains are isogenic with the erythromycin-sensitive strain JIR8094 [57], a derivative of the sequenced clinical isolate 630 [58]. C. difficile strains from freezer stocks were grown on BHIS agar plates [59] supplemented with and 0.1% sodium taurocholate (BioSynth International). To induce sporulation, C. difficile strains were grown on 70∶30 agar plates (63 g BactoPeptone, 3.5 g Protease Peptone, 11.1 g BHI, 1.5 g yeast extract, 1.1 g Tris base, 0.7 g NH4SO4 per liter). Media for C. difficile were supplemented with 10 µg thiamphenicol (Thi) mL−1, 50 µg kanamycin (Kan) mL−1, 8 µg mL−1 cefoxitin (TKC); 10 µg thiamphenicol mL−1; or 5 µg erythromycin mL−1 (Erm) as needed. C. difficile strains were maintained at 37°C in an anaerobic chamber (Coy Laboratory Products) with an atmosphere of 10% H2, 5% CO2, and 85% N2. E. coli strains were grown at 37°C in Luria-Bertrani (LB) broth. Antibiotics were used at 100 µg mL−1 carbenicillin for pET22b, 30 µg mL−1 kanamycin for pET28a, and 20 µg mL−1 chloramphenicol for pMTL83151 and pMTL84151 vectors in DH5α E. coli strains, and 100 µg mL−1 and 20 µg mL−1 in HB101 E. coli strains.
For details, see Text S1.
C. difficile strains were inoculated from frozen stocks onto BHIS plates containing 0.1% taurocholate. After 24 hr growth, a heavy streak of the strain was transferred to a 70∶30 plate and spread uniformly across the plate. Whereas <0.1% of cells are sporulating on BHIS plates [60], ∼25% of cells undergo sporulation at the timepoints analyzed in this study as determined by phase-contrast microscopy similar to Burns et al. [12] (Figure 1C). While the induction of sporulation occurs at different rates within the population, this assay allows us to produce relative high rates of sporulating cells. At the indicated timepoints, cells were scraped from the plate and resuspended in PBS. The cells were pelleted and then resuspended in PBS. For Western blot analysis, 50 µL of the cell resuspension was removed, and the sample was frozen at −80°C. The remainder of the sample was analyzed by phase contrast microscopy to assess the progression of sporulation.
For sporulating C. difficile samples, cell pellets harvested from the sporulation assay were subject to three cycles of freeze-thaw. On the final thaw, 100 µL of EBB buffer (9 M urea, 2 M thiourea, 4% w/v SDS, 10% v/v β-mercaptoethanol) was added, and the sample was boiled with occasional vortexing for 20 min. The lysate was pelleted for 5 min at 13,000×g and then resuspended; 7 µL of 4× loading buffer (40% v/v glycerol, 0.2 M Tris pH 6.8, 20% v/v β-mercaptoethanol, 12% SDS, 0.4 mg/mL bromophenol blue) was added. The sample was boiled again for a minimum of 5 min, pelleted at 13,000×g, and 15 µL was resolved on a 7.5% (for analysis of CspB in sporulating cells) or an 11 or 12% SDS-polyacrylamide gel (for analyses of purified spores).
For analyses of purified spores, ∼5×106 spores were pelleted at 15,000×g for 5 min. The spore pellet was resuspended in 50 µL EBB buffer, boiled for 20 min with periodic vortexing, pelleted at 13,000×g for 5 min, and resuspended to further solubilize proteins. Five µL of 4× loading buffer was added, and the sample was boiled for 5 min. After pelleting at 13,000×g, 10–15 µL of the sample was resolved on an 11% or 12% SDS-polyacrylamide gel.
All antibodies used in this study were raised in rabbits by CoCalico Biologicals, with the exception of the SleC antibody, which was raised in rabbits by Pacific Immunology. The antigens used were His6-tagged CspB(1–548 aa), full-length His6-tagged CspB perfringens, His6-tagged CspC, His6-tagged SleC and His6-tagged CD1433. Samples resolved by SDS-PAGE were transferred to Immobilon-FL PVDF membranes (Millipore). The membranes were blocked in 50∶50 PBS∶LiCOR blocking buffer (LiCOR) for 30 min, after which Tween 20 was added to 0.1% v/v, and polyclonal antisera was added at 1∶1,000 for all antibodies with the exception of the anti-SleC antibody, which was used at a 1∶5,000 dilution. After a minimum of 1 hr incubation with shaking, the membranes were washed a minimum of 3 times in PBS+0.01% v/v Tween. Anti-rabbit secondary antibodies conjugated to IR800 dye (LiCOR) were added at 1∶30,000 dilution in 50∶50 PBS∶LiCOR blocking buffer containing 0.1% v/v Tween and 0.1% v/v SDS then incubated with shaking for 1 hr. The membranes were washed a minimum of 3 times in PBS+0.1% v/v Tween before imaging on an Odyssey Clx scanner (LiCOR). Western blot quantitation was performed using the indicated loading controls and LiCOR ImageStudio software.
Sporulation was induced for 3–4 days on five 70∶30 plates. Spores and cell debris were scraped off the plate into 1 mL ice-cold sterile water and purified as previously described [59]. Briefly, the sample was subjected to 5 washes in ice-cold sterile water, followed by a HistoDenz gradient purification and 3–5 washes in ice-cold sterile water. Spores were stored at 4°C in water.
Purified spores were enumerated using disposable semen test counting chambers (InCyto C-Chip). Approximately 5×107 spores were resuspended in a total volume of 100 µL sterile H2O. The spores were heat activated at 60°C for 30 min, cooled for 2 min on ice, then 100 µL of 2× BHIS was added. 100 µL of the spores were removed to a tube containing 2 µL of 10% sodium taurocholate to induce germination. Both samples were incubated at 37°C for 20 min after which spores were serially diluted 10-fold into PBS. 10 µL of the dilutions was spotted onto either BHIS or BHIS+0.1% taurocholate agar plates in triplicate and incubated anaerobically at 37°C for ∼24 hr before assessing spore viability. Equivalent numbers of viable spores were recovered on untreated spores plated on BHIS+0.1% taurocholate plates and taurocholate-treated spores plated on BHIS or BHIS+0.1% taurocholate plates. Because spore clumping increased the variability in counting spores, CD1433 [61] was used as a loading control in some Western blot analyses.
For details see Text S1.
For details see Text S1.
Appropriate protein concentrations for crystallization were determined using Pre-Crystallization Test (Hampton Research, Aliso Viejo, CA). Hanging drop crystallization experiments were conducted with CspB (11 mg/mL) in 150 mM NaCl, 10 mM Tris-HCl pH 7.5 and Crystal Screen 2 (Hampton Research). Crystal trays were incubated at 12°C and initial crystal hits in 25% (v/v) ethylene glycol (Condition 3) were discovered within 24 hours. After refinement of crystallization conditions, crystals grew reproducibly to about 100*250*60 µm3 in 27–30% (v/v) ethylene glycol buffered to pH 5 with 50 mM sodium acetate. Crystals grew in space group P212121, with unit cell dimensions a = 73.87, b = 138.17, and c = 140.08 Å and two molecules in the asymmetric unit for an estimated 57% solvent content [62]–[64]. As crystallization conditions contained sufficient ethylene glycol to serve as a cryoprotectant, crystals were flash cooled in liquid nitrogen directly from the crystallization drop.
A complete 1.6 Å single-wavelength data set of a representative selenomethionyl-CspB crystal was collected at the selenium edge (0.9794 Å) at 100 K at the General Medical Sciences and Cancer Institutes Structural Biology Facility (GM/CA @ APS) beamline 23ID-B at the Advanced Photon Source, Argonne National Laboratory (Chicago, IL, Table S4 in Text S1).
Data were processed using Denzo and Scalepack [65]. Twelve selenium sites were expected, from 6 methionines in the protein sequence and two predicted molecules in the asymmetric unit, using the Matthews Coefficient program [62], [63] in the CCP4 Program Suite [64]. ShelXC/D/E, also part of the CCP4 Suite, was used to identify the selenium sites and gain initial phase information [64], [66], [67]. The 12 selenium sites and phase information were used in ShelX/E for density modification and generation of the initial phased map (Fig. S7) [66], [67].
The initial model was produced by Phenix.AutoBuild using input phases from ShelX/E [66], [68]. Manual building was performed into the original phased map to reduce model bias. Refinement of the structure was done with manual building and adjustment in COOT [69] and refinement of the latest iteration of the model using Phenix.Refine [68]. All protein and ligand (non-water) B-factors were refined anisotropically. Phenix.AutoBuild with simulated annealing was used after multiple rounds of refinement to gain density for some poorly-resolved loops in the structure, resulting in the placement of several previously missing residues [68]. Ten percent of reflections were set aside for Rfree calculation. Model was refined to an Rwork/Rfree of 0.15/0.18 and Ramachandran statistics were 97.9% in favored regions and 2.1% in allowed regions, with no Ramachandran outliers. 957 water molecules were placed by Phenix.Refine and checked with the Check Waters feature in COOT [68], [69].
Although CspB is a dimer in the asymmetric unit, gel filtration chromatography experiments (see Text S1) indicate that CspB is a monomer in solution. In monomer 1, five residues were not built due to disorder in the electron density map (residues 411–415); in monomer 2, three residues were not built due to disorder (residues 411–413). These residues are part of a small loop located between two strands of the jellyroll domain. Additionally, the first four residues (residues 1–4) of the prodomain in each monomer were disordered and not built. Electron density for the C-terminal His6-tag used for protein purification (see Text S1) was seen in the second monomer only; these residues were stabilized by a crystal-packing interface, thus enabling residues 566–573 to be built in this monomer. Although the presence of calcium in the model was expected because this metal is present in many subtilisin family members [31], elemental analysis did not detect Ca2+ in our enzyme preparation (Dartmouth Trace Elemental Analysis Lab, data not shown). Two putative Na+ and three Cl− atoms (confirmed by sodium iodide soaks) were placed in the model, in addition to ethylene glycol, a crystallization reagent.
Wildtype CspB and its mutant variants were diluted to 15 µM in 10 mM Tris pH 7.5 buffer in a total volume of 150 µL. Twenty-four microliters of the mixture were transferred into 8 well strip tubes. One microliter of chymotrypsin (Sigma, 25-fold concentrate relative to indicated concentration) was added, and the mixture was mixed then incubated for 60 min at 37°C. Chymotrypsin activity was quenched by the addition of 8 µL of 4× loading buffer. The samples were boiled for 3 min at 95°C and then 7 µL was resolved on a 15% SDS-PAGE gel and visualized by Coomassie staining.
E. coli cultures were grown as described for protein purification. One hour after IPTG induction, a 1 mL sample was removed, the OD600 measured, and the sample pelleted at 13,000×g for 2 min. Cells were lysed in 1× loading buffer (10 OD600/mL). To obtain cleared lysate samples, 30 µL of the supernatant produced upon high-speed centrifugation of sonicated lysates was added to 10 µL of 4× loading buffer. For eluate samples, 30 µL of the eluate was added to 10 µL of 4× loading buffer. All samples were boiled at 95°C for 5 min, pelleted at 13,000×g for 5 min, then 2.5 µL of induced and cleared lysate samples or 5 µL of eluate samples were resolved on a 12% SDS-PAGE gel and analyzed by Western blotting.
Wildtype CspB and its mutant variants were diluted to 10 µM in 10 mM Tris-HCl pH 7.5 buffer in a total volume of 155 µL. Twenty-five microliters were aliquoted in triplicate into strip tubes. 0.25 µL of 100 µM FP-Rh (fluorophosphanate-rhodamine probe) was added to CspB and incubated at RT for 10 min. Labeling was quenched by adding 8 µL 4× loading buffer to the sample and boiling at 95°C for 3 min. Six microliters of the labeling reaction was resolved on a 15% SDS-PAGE gel, and fluorescence was imaged using a Biorad PharosFX scanner.
Coordinates and structure factors have been deposited in the Protein Data Bank (www.rcsb.org) under the accession number 4I0W.
|
10.1371/journal.pntd.0004399 | Epidemiology and Clinical Burden of Malaria in the War-Torn Area, Orakzai Agency in Pakistan | Military conflict has been a major challenge in the detection and control of emerging infectious diseases such as malaria. It poses issues associated with enhancing emergence and transmission of infectious diseases by destroying infrastructure and collapsing healthcare systems. The Orakzai agency in Pakistan has witnessed a series of intense violence and destruction. Military conflicts and instability in Afghanistan have resulted in the migration of refugees into the area and possible introduction of many infectious disease epidemics. Due to the ongoing violence and Talibanization, it has been a challenge to conduct an epidemiological study.
All patients were sampled within the transmission season. After a detailed clinical investigation of patients, data were recorded. Baseline venous blood samples were taken for microscopy and nested polymerase chain reaction (nPCR) analysis. Plasmodium species were detected using nested PCR (nPCR) and amplification of the small subunit ribosomal ribonucleic acid (ssrRNA) genes using the primer pairs. We report a clinical assessment of the epidemic situation of malaria caused by Plasmodium vivax (86.5%) and Plasmodium falciparum (11.79%) infections with analysis of complications in patients such as decompensated shock (41%), anemia (8.98%), hypoglycaemia (7.3%), multiple convulsions (6.7%), hyperpyrexia (6.17%), jaundice (5%), and hyperparasitaemia (4.49%).
This overlooked distribution of P. vivax should be considered by malaria control strategy makers in the world and by the Government of Pakistan. In our study, children were the most susceptible population to malaria infection while they were the least expected to use satisfactory prevention strategies in such a war-torn deprived region. Local health authorities should initiate malaria awareness programs in schools and malaria-related education should be further promoted at the local level reaching out to both children and parents.
| The malaria epidemic and endemic in Pakistan is a present and ongoing threat to public health which could have an impact in the nearby regions as well. For the first time, we report a clinical assessment of malaria endemicity in the Orakzai Agency, which is Pakistan’s most neglected area due to Talibanization and war in Afghanistan. Febrile patient blood samples of the area were investigated to report the clinical assessment of malaria caused by Plasmodium vivax and P. falciparum infections. The nested polymerase chain reaction (nPCR) examination detected 154 (86%) and 21 (12%) P. vivax and P. falciparum infections, respectively. We found worsening hygiene conditions in FATA, likely caused by poor socioeconomics and the collapse of the public health infrastructure. Decompensated shock was a common and prominent clinical feature of malaria among all the clinical presentations caused by both P. vivax (53%) and P. falciparum (42.9%). Our results have significant implications for both public health and malaria control in FATA and Pakistan. Our findings illustrate higher prevalence of malaria in children compared to other age groups. Further research on sensible estimates of refugees is required, as well as resistance to anti-malarials.
| Disease emergence is influenced by both natural and human factors. Among human activities, military conflicts characterized by war and regional tribal and/or sectarian strife have had huge impact by destroying infrastructure and collapsing healthcare systems. The affected region and the people therein face diverse short-term and long-term consequences. Large population displacements present higher risks of infectious diseases, lack of resources, which is often the case within crowded refugee camps with sanitation issues, increased exposure of the population to disease vectors, and destruction of healthcare systems lead to negative consequences [1]. Moreover, if the situation were to prolong, discontinued public and healthcare professional education investment and lack of proper surveillance and control of the disease should produce more severe and chronic outcomes. These outcomes make the people in the affected and nearby regions become newly vulnerable to a variety of communicable diseases and directly associated with the prevalence and emergence of the infectious diseases [1, 2].
The prolonged Soviet war in Afghanistan between 1979 and 1995 annihilated the malaria vector control programs in the nation, which had been implemented in the 1960s and 1970s. The programs were so successful that Afghanistan had virtually eradicated the disease in the late 1970s [3]. The collapse of the program resulted in the re-emergence of malaria, and turned the region into a malaria endemic area. After the Soviet war, Afghanistan and the nearby Pakistani regions, specifically the Federally Administered Tribal Areas (FATA, a semi-autonomous territory running along the Pakistan–Afghanistan border) transformed into religious fundamentalism where tribe heads lost their authority and the rule of militant groups such as the Taliban started. Moreover, the subsequent 9/11 triggered occupation of the United States initiated a sizable population displacement of the refugees to FATA Pakistan. The border-crossing migration of Afghan refugees has overwhelmed the local public health system, and has caused an malaria epidemic [4]. Many epidemiology studies attributed the roughly 24%–36% increase in malaria prevalence in this region to the post-Soviet war influx of Afghan refugees into FATA [5].
Different researchers reported that permanent residents have low susceptibility and high immunity against malaria as compared to Afghan refugees [6]. The proportion of malarial cases due to P. vivax is increasing every year. According to the World Health Organization (WHO) report, approximately 75% (previously 64%) of the infections are transmitted through P. vivax, whereas 25% (previously 36%) are caused by P. falciparum, which are the two most prevailing species in Pakistan [7]. Currently, the P. vivax malaria is accounted in 70% of the malaria burden in Pakistan, and FATA being the most impoverished and extremely underdeveloped area in the Pakistan has the highest malaria burden due to the large Afghan refugees and IDPs [8]. Social development gages are appallingly low. There are only 41 hospitals for a population of 3.1 million.
The Orakzai Agency (Fig 1) has been one of the most neglected areas in FATA during the multi-decade-long military conflicts in the region. The Orakzai tribes have been linked to and have provided safe haven to the Taliban militant groups since late 2001. A large part of the Orakzai Agency soon fell under control of the Taliban. The Agency was once home to Hakimullah Mehsud, the Tehrik-i-Taliban Pakistan chief who led militant operations, targeting hundreds of NATO supply vehicles in 2008 and 2009. The regional medical centers and educational institutions were destroyed by repeated militant attacks [9]. This caused disruption to malaria management efforts as well as the increase of the malaria parasite reservoir in the region. For proper detection and control of malaria, obtaining and analyzing reliable epidemiology data is essential. To the best of our knowledge, however, there has been no report on malaria epidemics and clinical manifestations in Orakzai Agency. It has been a challenge to carry out such studies due to the severity of military conflicts.
We report here the first study of epidemiology and clinical burden of malaria in this war-torn and Talibanized region, Orakzai Agency in Pakistan (the sample collecting region is shown in Fig 1). Keeping in view all the state of affairs and distribution of malaria epidemics in nearby FATA regions, the current study aimed to investigate the epidemiology and clinical manifestations of P. vivax and P. falciparum malaria among male and female patients of different age groups. We also report the analytical, epidemiological and clinical differences between P. vivax and P. falciparum infections.
Orakzai Agency (FATA in Pakistan, Fig 1) is divided into the upper Orakzai and the lower Orakzai. Orakzai is a long neglected area with lack of basic health necessities, occasional engagement of armed insurgents, poor living conditions, limited access to vaccines, limited use of vector control measures, and unequal distribution of economic resources; most importantly, this is the case of approximately 60% of the FATA residents [10]. Furthermore, the medical centers and educational institutions have also been ruined by militant attacks. Healthcare-related non-government organization (NGO) activities are not permitted in FATA. In 2012, according to UNHCR, nearly 758,000 internally displaced persons (IDPs) fled from their homes as a result of security operations in the FATA [11]. More recently, in June 2014, Pakistani military initiated operation against militant groups in FATA resulted in approximately 450,000 IDPs displacement in the Bannu district [12]. Presently, FATA has the highest burden of different infectious diseases, due to the large Afghan refugees and IDPs. The population suffering from or at risk of contracting malaria significantly increased in the FATA as did the malaria parasite reservoir.
This retrospective case-control study was conducted at the District Headquarters Hospital (DHH) Kalaya in Orakzai Agency, Pakistan between April 2011 and December 2013. Patients presented with major clinical symptoms (fever, headache, cough, dyspnea, vomiting, diarrhea, abdominal pain, and convulsions) of malaria at different partially functional health care centers were referred to the DHH Kalaya. The major clinical symptoms of malaria were based on WHO criteria [8, 13]. All children and adults presenting to hospital were screened for study eligibility and were hospitalized. A total of 216 microscopy-confirmed patients aged 1–60 years were evaluated in the study. Demographic and clinical records were collected upon enrollment, and baseline venous blood samples were collected for further biochemical and molecular analyses. Control group was collected at the same site as uncomplicated cases and used for statistical analyses. To exclude the confounding effect of sex, age and locality, control population was matched by sex, age and locality. Pregnant women were not included in the study. The inclusion criteria for patients were as follows: (i) prostration (unable to sit), (ii) multiple seizures, (iii) impaired consciousness, (iv) multiple convulsions, (v) hyperpyrexia, (vi) anemia, (vii) decompensated shock, (viii) dark urine, (ix) jaundice, (x) hypoglycaemia, (xi) hyperparasitaemia, and (xii) respiratory problems [14]. In severe malaria, the level of impaired consciousness was assessed by computing the Glasgow Coma Scale (GCS) score (<11) in adults or Blantyre Coma Scale Score (<3) in children [14]. For further investigation, studied patients were divided into groups on the basis of their age and sex.
After a detailed clinical investigation of patients, a standardized case report template was designed to compile the complete clinical data of each patient. For children younger than five years, their parents or relatives were asked of their medical history.
Baseline venous blood samples were taken for microscopy and nested polymerase chain reaction (nPCR) analysis. The initial diagnosis of Plasmodium spp. infection was made by thick or thin smears. Two slides were made from each patient’s blood and both thick and thin films were prepared on the slides in the DHH Kalaya laboratory. Giemsa-stained thick blood smears of patients were examined using Giemsa stain and the parasitemia quantified independently by two skilled microscopists [15]. A thick smear was considered negative if no parasite was seen in at least 200 fields.
For molecular analysis, the parasite DNA was extracted from filter papers using the Qiagen DNA extraction kit (QIAGEN, Valencia, CA, USA), following to the manufacturer’s protocol. The Plasmodium species were detected using nested PCR (nPCR) and amplification of the small subunit ribosomal ribonucleic acid (ssrRNA) genes using the primer pair set A (5’-TTAAAATTGTTGCAGTTAAAACG-3’ and 3’-CCTGTTGTTGCCTTAAACTTC-5’) for the detection of P. vivax; primer pair set B (5’-CGCTTCTAGCTTAATCCACAT AACTGATAC-3’ and 3’-ACTTCCAAGCCGAAGCAAAGAAAGTCCTTA-5’)for the detection of P. falciparum; primer pair set C (5’-CTGTTCTTTGCATTCCTTATGC-3’ and 3’-GTATCTGATCGTCTTCACTCCC-5’) for the detection of P. ovale; and primer pair set D (5’-GTTAAGGGAGTGAAGACGA-3’ and 3’- TCAGAAACCCAAAGACTTTGATTTCTCAT-5’) for the detection of P. malariae. PCR reactions were carried out on a thermal cycler (Nyx Technik USA), beginning with 5 minutes at 94°C, followed by 25 cycles of 45 seconds at 94°C, 45 seconds at 58°C, and 5 minutes at 72°C for the first round; 30 cycles of 45 seconds at 94°C, 45 seconds at 65°C, and 2 minutes at 72°C was then performed for the second round. The final cycle was followed by an extension time of 5 minutes at 72°C. The amplified PCR products were analyzed by 2%–2.5% agarose gel electrophoresis stained with ethidium bromide and visualized on the Bio-Rad gel doc system (Bio-Rad Laboratories, Hercules, CA, USA). Due to the absence of well-equipped laboratory facilities in the DHH Kalaya, PCR and biochemical analyses were carried out at the Kohat University of Science and Technology and Kohat hospital.
Statistical analysis was carried out on SPSS version 19. Means, odds ratios with 95% CIs, and χ2 test of independence were calculated when applicable. Statistica (version 12) was used for box-and-whisker plots. In all the studied parameters, a p value of ≤0.05 was considered statistically significant. Patients with prior comorbid conditions were excluded from relevant subanalyses, for example, diabetes mellitus patients were excluded from hypoglycemia analysis. All analyses were repeated after excluding all patients with associated infections and comorbid illnesses.
Ethical approval for project activities was provided by the Kohat University of Science and Technology. Written informed consent was obtained from the patients and their parents/guardians before recruitment.
We collected 216 blood samples from febrile patients between April 2011 and December 2013 and screened for malaria. These febrile samples were collected from the DHH Kalaya in Orakzai Agency, Pakistan, where ordinary disease detection and control activities had been halted due to decade-long military conflicts. Many of the febrile patients were known to be displaced Afghan refugees, but such data was not recorded. A total of 216 blood samples were diagnosed positive by the microscopic examination. As baseline patient demographics are shown in Table 1, 178 of 216 patients were identified by nPCR to have contracted malaria (mean ± SD age 19.9 ± 11.9 years). Among these, we have found monoinfections of P. vivax and P. falciparum, as well as co-infections of both pathogens in diverse age groups (Table 1). In our study, 11 of 216 patients were presented with impaired consciousness (GCS <11 or BCS <3) during hospitalization, however they were excluded from the study because all 11 patients showed associated infections and comorbid illnesses such as pneumonia. Furthermore, all 11 patients were not identified by nPCR to have contracted malaria.
The diagnosis was initially made by the microscopic examination and nPCR. The microscopic examination identified 175 patients (81%) to be infected by P. vivax, and 31 patients (14%) by P. falciparum, and 10 patients (5%) doubly infected by both P. falciparum and P. vivax (Fig 2). In contrast, the nPCR examination detected 154 (86%) P. vivax infections and 21 (12%) P. falciparum infections, respectively. This method detected 3 (2%) double infections. However, P. ovale and P. malariae infections were not identified in any of the investigated samples by both test methods. Noticeable discrepancies between microscopic (216 patients) and nPCR (178 patients) detections were observed (Fig 2). Previous comparative diagnosis studies demonstrated nPCR to produce sensitive and reliable diagnosis results better than other methods including microscopy [16, 17]. nPCR was acceptable to serve as the reference standard in malaria diagnosis. Therefore, we decided to rely on the nPCR diagnosis.
We observed greater prevalence of Plasmodium infection in males (70%). The further clinical and biochemical tests of patients with P. vivax and P. falciparum infections (Table 2) demonstrated the majority of the subjects (80%; n = 142 of 178) exhibiting severe malaria complications by World Health Organization criteria. As shown in Table 2, comparable and similar rates of various complications were observed in both P. vivax and P. falciparum patients. Among 121 febrile patients with severe P. vivax infection (Table 2), the frequency of complications was as follows: decompensated shock (n = 64; 53%; p = <0.001), hypoglycaemia (n = 12; 10%), anemia (n = 12; 10%), hyperpyrexia (n = 10; 8%), multiple convulsions (n = 10; 8%), hyperparasitaemia (n = 7; 6%), and jaundice (n = 6; 5%). On the other hand, the following frequency of complications was observed in febrile patients with severe P. falciparum infection (n = 21 of 178): decompensated shock (n = 9; 43%; p = <0.001), anemia (n = 4; 19%), jaundice (n = 3; 14%), multiple convulsions (n = 2; 9%), hypoglycaemia (n = 1; 5%), hyperpyrexia (n = 1; 5%), and hyperparasitaemia (n = 1; 5%). The frequency of complications among all patients who tested positive with malaria (n = 178) were as follows: decompensated shock (n = 73; 41%), anemia (n = 16; 9%), hypoglycaemia (n = 13; 7%), multiple convulsions (n = 12; 7%), hyperpyrexia (n = 11; 6%), jaundice (n = 9; 5%), hyperparasitaemia (n = 8; 5%), and mixed complications (n = 36; 20%) more than one criteria. The most common malarial complication caused by P. vivax and P. falciparum was decompensated shock (p = <0.001). Decompensated shock symptom had the highest odds ratio (OR) for being reported in patients affected by both malarial species (Table 2). Hypoglycemia, multiple convulsions and anemia had an OR in the similar range in case of P. vivax infected patients whereas a different pattern of OR was observed in P. falciparum infected patients (Table 2). Although all other statistical associations held, the strength of association varied.
The mean parasite count for P. vivax patients was (20241.9; p = 0.744), which is significantly greater than that of P. falciparum patients (11848; p = 0.744). The mean illness duration was 5.2 ± 2.0 days for P. falciparum male patients and 4.4 ± 0.7 days for females. Similarly, the mean illness duration for P. vivax male patients was 5.5 ± 1.6 days and 6.1 ± 1.3 days for female patients. The incidence of both vivax and falciparum malaria gradually increased between the ages of 1–20 years with increasing age (Fig 3). The prevalence of malaria reached its peak among late teenagers (the age of 15–20, see Fig 3). It was also obvious that P. vivax infections were most prevalent in children of the age group between 5 and 15 years old as shown in Fig 3A, while P. falciparum infections in children populations were less prevalent as compared with P. vivax (Fig 3B). However, the most noteworthy characteristic was the drastic decrease in malaria incidence in post-puberty males and females (21–60 years). This age-dependency was observed in both parasite species infections (P. falciparum and P. vivax). High childhood malaria parasite exposure resulted in children (1–15 years) bearing the brunt of the disease burden (Fig 3). This group turned out to be the most vulnerable.
The malaria epidemic and endemic in Pakistan is a present and ongoing threat to public health which could have an impact in the nearby regions as well. In 2008, 2.6 million malaria cases were reported countrywide with a death rate of 50,000 per year [8, 18, 19]. National and provincial malaria control programs are dependent on a network of facilities providing diagnosis data. However, the regional situation due to the constant military conflicts and Talibanization of the Orakzai Agency in FATA collapsed the regional healthcare systems, and has made any epidemiological study practically impossible albeit essential for disease control and management.
For the first time, we carried out a reliable epidemiological and clinical study with febrile patients. In this study, 86.5% of the malaria cases were attributed to P. vivax infections, and 11.8%, to P. falciparum (p = 0.258). This result agreed with previous studies performed in nearby areas in Pakistan, as it was shown that the prevalence of P. vivax changes from 70% (30% P. falciparum) in mid Pakistan province to 90% (10% P. falciparum) in areas near or inside Afghanistan including FATA [8]. Moreover, we could not find any P. ovale and P. malariae infections in our febrile patient samples. This result also agreed with the fact that these two species had been negligible in this region [7, 20, 21]. To ensure that the results represented unbiased reality of the malaria endemic and epidemic in Orakzai Agency, all samples were collected from patients during the transmission season as the P. vivax transmission season peaks between April and September, while the P. falciparum peaks between August and December [7, 21]. P. vivax is the most prevalent in hilly areas, while P. falciparum has the lowest prevalence rate [22]; similarly, the Orakzai Agency is covered with hills and two rivers, making this area suitable for malarial parasites breeding.
We observed a relationship between patient sex and malarial infections. The predominance of malarial infection in male patients between the ages of 10 and 20 years was observed in our studies as well as in earlier investigations in Pakistan [22, 23]. It is believed that males have more outdoor exposure then females, and hence have increased vector exposure because they have to do laborious work in agricultural fields but are not as well-covered as adult females. In fact, in the Orakzai Agency women are totally covered with Burqa: an enveloping outer garment worn by women. As a result, they faced increased Anopheles bites. Therefore, there has been the disproportionate number of male malaria patients. On the other hand, during this study, we were intrigued to see that poor socioeconomic conditions together with the lack of public health infrastructure in Orakzai Agency might also be the cause of the disproportionateness. In Talibanized regions, women were often banned from traveling to a hospital without an accompanying male. In addition, in extreme poverty, a family may not afford medical attention to a sick woman. Indeed, comparable rate of malaria infections in both genders in other Pakistani provinces was observed in previous studies [22, 23].
In this study, we also observed children being excessively susceptible to malaria. Approximately half of the vivax malaria patients (n = 54) were children (≤20 years old) identified as having severe illness. This is consistent with other studies of hospitalized children in the South Asia region and in India [24]. This unequal susceptibility may possibly be attributed to the poor yet-to-develop immunity against the parasites [14]. As they get older and more exposed to the parasites, however, they can progressively develop adequate immunity to malaria.
In our study, clinical and laboratory analysis identified 80% of malaria patients (n = 142) as having severe illness (Table 2). Anemia has been the typical consequence of malaria, and we found 9% of malaria patients (n = 16 out of 178; p = 0.061) with severe anemia. This was more prevalent in P. vivax infected males (10%; n = 12 out of 121), while anemia induced by P. falciparum is considered more recurrent and more severe than anemia induced by P. vivax. Similar findings of prevalent P. vivax-induced anemia have also been reported previously [25–27]. In addition, 10% of patients infected by P. vivax (n = 12) were found to be with severe hypoglycemia. This result agreed with a previous report which had identified 9% of P. vivax malaria patients as having severe hypoglycemia [28, 29].
In our study, it is interesting to note that the manifestations of severe malaria instigated by P. vivax infection were more intricate than that of P. falciparum malaria. P. vivax malaria has reportedly caused tertian ague malaria that rarely led to a severe form [29]. However, many recent studies have shown an elevated risk of morbidity and mortality in P. vivax malaria [30–32]. Jaundice is one of the common manifestations of vivax malaria. In Northwestern India, jaundice has been reported in up to 57% of hospitalized patients infected by P. vivax [33]. In this study, jaundice was observed much less frequently, only in 5% P. vivax-infected patients (n = 6). We also found a similar prevalence of hyperpyrexia (8%; n = 10 out of 121) among P. vivax-infected patients to that of the previous study [34]. Cerebral malaria in patients with P. vivax infection was observed with multiple convulsions accounting for 8% of severely ill subjects, but not as frequently (14%) as seen in Bikaner, India [24].
Interestingly, the parasite count of P. vivax was very high in our studies. The lofty level of parasitic density of P. vivax as well as that of P. falciparum in some patients may presumably reflect the immune tolerance resulting from repeated exposures. Our study also reports the frequency of decompensated shock (41%; n = 73 out of a total of 178 patients; p = <0.001). This rate is the highest among similar studies reported previously. We found that decompensated shock was a common and prominent clinical feature of malaria caused both by P. vivax (53%; n = 64; p = <0.001) and P. falciparum (42.9%; n = 9; p = <0.001). Our result corresponds to the findings which also reported that P. vivax malaria could cause decompensated shock in infected individuals [23, 27].
Nonetheless, our study has its own limitations including lack of meticulous refugee statistics. Our study failed to estimate the true number of imported malarial cases in Orakzai Agency. Due to the presence of the armed Taliban insurgents in the Upper Orakzai, we were unable to collect the disaggregated data of Orakzai Agency. However, this study is the first report on the epidemic situation and clinical analysis regarding this most neglected region. Furthermore, in our study, children were the most susceptible population to malaria infection whereas they were the least expected to use satisfactory prevention strategies in such a war-torn deprived region. Local health authorities should initiate malaria awareness programs in schools and malaria-related education should be further promoted at the local level reaching out to both children and parents. We conclude that this overlooked distribution of malaria should be considered by malaria control strategy makers in the world and by the Pakistani government.
|
10.1371/journal.pgen.1002058 | A Large-Scale Complex Haploinsufficiency-Based Genetic Interaction
Screen in Candida albicans: Analysis of the RAM Network during
Morphogenesis | The morphogenetic transition between yeast and filamentous forms of the human
fungal pathogen Candida albicans is regulated by a variety of
signaling pathways. How these pathways interact to orchestrate morphogenesis,
however, has not been as well characterized. To address this question and to
identify genes that interact with the Regulation of Ace2 and Morphogenesis (RAM)
pathway during filamentation, we report the first large-scale genetic
interaction screen in C. albicans. Our strategy for this screen
was based on the concept of complex haploinsufficiency (CHI). A heterozygous
mutant of CBK1
(cbk1Δ/CBK1), a key RAM pathway
protein kinase, was subjected to transposon-mediated, insertional mutagenesis.
The resulting double heterozygous mutants (6,528 independent strains) were
screened for decreased filamentation on Spider Medium (SM). From the 441 mutants
showing altered filamentation, 139 transposon insertion sites were sequenced,
yielding 41 unique CBK1-interacting genes. This gene set was
enriched in transcriptional targets of Ace2 and, strikingly, the cAMP-dependent
protein kinase A (PKA) pathway, suggesting an interaction between these two
pathways. Further analysis indicates that the RAM and PKA pathways co-regulate a
common set of genes during morphogenesis and that hyper-activation of the PKA
pathway may compensate for loss of RAM pathway function. Our data also indicate
that the PKA–regulated transcription factor Efg1 primarily localizes to
yeast phase cells while the RAM–pathway regulated transcription factor
Ace2 localizes to daughter nuclei of filamentous cells, suggesting that Efg1 and
Ace2 regulate a common set of genes at separate stages of morphogenesis. Taken
together, our observations indicate that CHI–based screening is a useful
approach to genetic interaction analysis in C. albicans and
support a model in which these two pathways regulate a common set of genes at
different stages of filamentation.
| Candida albicans is the most common cause of fungal infections
in humans. As a diploid yeast without a classical sexual cycle, many genetic
approaches developed for large-scale genetic interaction studies in the model
yeast Saccharomyces cerevisiae cannot be applied to C.
albicans. Genetic interaction studies have proven to be powerful
genetic tools for the analysis of complex biological processes. Here, we
demonstrate that libraries of C. albicans strains containing
heterozygous mutations in two different genes can be generated and used to study
genetic interactions in C. albicans on a large scale. Double
heterozygous mutants that show more severe phenotypes than strains with single
heterozygous mutations are indicative of genetic interactions through a
phenomenon referred to as complex haploinsufficiency (CHI). We applied this
approach to the study of the RAM (Regulation of Ace2 and Morphogenesis)
signaling network during the morphogenetic transition of C.
albicans from yeast to filamentous growth. Among the genes that
interacted with CBK1, the key signaling kinase of the RAM
pathway, were transcriptional targets of the RAM pathway and the protein kinase
A pathway. Further analysis supports a model in which these two pathways
co-regulate a common set of genes at different stages of filamentation.
| Candida albicans is a member of the resident flora of the
gastrointestinal tract and is the most common fungal pathogen in humans. The most
severe manifestations of candidiasis occur in immunocompromised patients and include
debilitating mucosal disease such as oropharyngeal candidiasis as well as
life-threatening disseminated infections of the bloodstream and major organ systems
[1]. Animal
studies have shown that the pathogenic potential of C. albicans is
associated with its ability to transition between three morphological states: yeast,
pseudohyphae, and hyphae [2], [3]. Further insights into the contributions of the different
morphotypes to pathogenesis have emerged from elegant studies with C.
albicans strains that allow the conditional induction of filamentation
in vivo
[4]. For example,
C. albicans genetically restricted to the yeast form by
constitutive expression of NRG1 are able to establish infection in
mice but no disease results until the expression of NRG1 is
repressed and the organism is able to form filaments.
The relationship between morphogenesis and virulence in C. albicans
is, however, not a simple one. Many mutants that are unable to undergo morphogenesis
also display other phenotypes. For example, many transcription factors that are
required for morphogenesis regulate a host of other genes and display pleiomorphic
phenotypes. The complicated nature of the relationship between morphogenesis has
been further highlighted by the elegant study recently reported by Noble et
al.
[5]. Noble
et al. generated a bar-coded collection of homozygous deletion
mutants and used it in a signature-tagged mutagenesis study of infectivity in a
mouse model [5].
Mutants with defects in morphogenesis were more likely to have decreased
infectivity; however, a significant portion of mutants with severe morphogenesis
defects retained the ability to cause infection. It is important to note that Noble
et al. assayed for infection and not for disease. Thus, their
results are not necessarily in conflict with studies discussed above that indicate
that morphogenesis is required for disease progression in animal models [4]. Furthermore,
their work serves to highlight the fact that additional studies will be required to
fully understand the complex relationship between morphogenesis and pathogenesis in
C. albicans.
Given the close association of morphogenesis with C. albicans
pathogenesis, the genetic and cell biologic analysis of this process has been the
subject of intensive study [6]. Consequently, many genes have been shown to affect
filamentation, and, correspondingly, a number of regulatory pathways have been shown
to play a role in the orchestration of the morphogenetic program in C.
albicans
[7]. The
PKA, CPH1, HOG1,
RIM101, CHK1, and CBK1/RAM
pathways are among those that regulate morphogenesis under a variety of conditions
[6], [7]. Although much
remains to be learned about how individual pathways and genes contribute to
morphogenesis, an important question that has not been extensively studied is how
these various pathways interact to regulate morphogenesis.
In the model yeast S. cerevisiae, relationships between regulatory
pathways can be readily characterized using recently developed systematic,
genome-wide genetic interaction strategies [8]–[10]. These approaches have yielded a
wealth of information regarding the mechanisms through which cells regulate complex
biological processes [11]. However, because C. albicans is diploid
and lacks a classical meiotic cycle, the mating-based genetic strategies used to
create genome-wide libraries of double mutant strains in S.
cerevisiae are not applicable. Consequently, genetic interaction
studies in C. albicans have been limited to gene-by-gene analyses.
Despite these limitations, such studies have proven quite informative and suggest
that large scale interaction studies could represent a powerful approach to studying
regulatory networks in C. albicans. For example, Braun et
al. carried out a thorough, systematic epistasis analysis of three
transcriptional regulators (EFG1, TUP1 and
CPH1) and showed that each played a distinct role in the
regulation of filamentation [12].
Recent advances in the genetic analysis of C. albicans have greatly
facilitated the development of innovative approaches to the study of this important
human pathogen [13]. Among these important developments is the application of
transposon-based mutagenesis strategies [14] to the creation and study of
large-scale libraries of heterozygous [15], and homozygous [17], [18]
C. albicans mutants. Similarly, large collections of homozygous
null [19] and
conditional mutants [20] have been created in a targeted manner and analyzed for a
variety of phenotypes including morphogenesis, virulence and drug susceptibility. To
our knowledge, one area that has not been explored is the development of approaches
to large-scale synthetic genetic interaction analysis in C.
albicans.
Here, we describe the first large-scale synthetic genetic interaction screen in
C. albicans. Our strategy builds on pioneering yeast genetics
approaches developed in both S. cerevisiae and C.
albicans and is based on the concept of complex haploinsufficiency
(CHI). CHI is a special case of a genetic phenomenon referred to as unlinked
non-complementation in the context of yeast genetics and as dominant enhancers or
dominant modifiers when applied to Drosophila
[21]. Unlinked
non-complementation occurs when a cross between two haploid strains containing
single recessive mutations located in separate loci results in a diploid strain
(complex heterozygote) that retains the phenotype of a parental strain. In yeast,
the construction of such mutants was used to great advantage in the genetic analysis
of cytoskeletal genes such as tubulin [22] and actin [23]. CHI, which is a
special case of unlinked non-complementation, occurs when strains containing
heterozygous mutations at two separate loci display a more severe phenotype than
strains that contain heterozygous mutations at the single loci alone [21]. In essence,
CHI can also be called synthetic haploinsufficiency. Recently, a genome-wide
CHI-based strategy was developed in S. cerevisiae and successfully
used to create a genetic interaction network for the essential gene,
ACT1
[21].
As described in the seminal work of Uhl et al.
[15], large scale
haploinsufficiency-based screening was first applied to C. albicans
in the transposon-mediated, insertional mutagenesis analysis of filamentation and,
thus, haploinsufficiency-based screening has excellent precedence in this system.
Whereas Uhl et al. carried out their haploinsufficiency screen
starting with a “wild type” strain [15], we reasoned that transposon
mutagenesis of a parental strain containing a heterozygous mutation at a locus of
interest would represent an expedient approach to the generation of a large library
of complex heterozygotes that could then be the basis of a CHI screen for genes that
interact with the parental mutant.
In principle, CHI-based screening has a number of attractive features. First, CHI
allows one to identify genes that function within the pathway affected by the
parental or query mutation including upstream and downstream components of the
pathway, transcriptional outputs of the pathway, and substrates of pathway enzymes.
Second, CHI-based screening can also identify genes or pathways that function in
parallel with the query pathway and, therefore, allow one to identify pathways that
co-regulate a given process. Third, CHI is ideal for the study of essential genes
because only heterozygous mutations are generated.
We developed a CHI-based screening strategy (Figure 1) and applied it to the identification of
genes that interact with the RAM signaling network during C.
albicans filamentation [24]–[27]. The RAM network has been
extensively studied in S. cerevisiae
[28] and is
required for a variety of cellular processes in both S. cerevisiae
and C. albicans including polarity, cell wall synthesis, cell
separation and filamentous growth. Cbk1 is the key serine/threonine protein kinase
[24], [27] that mediates
many of the functions of the RAM network through its regulation of the transcription
factor Ace2 [24],
[27]. RAM
pathway mutants in C. albicans show two distinct filamentation
phenotypes: CBK1 null mutants are unable to form filaments on
Spider Medium (SM) or serum-containing medium [24], [27] while ACE2 null
mutants are constitutively pseudohyphal and form true hyphae on serum [25]. Although our
understanding of the RAM network in C. albicans has increased in
recent years [24]–[27], many questions remain, including: how does it interact
with the many other regulatory pathways during morphogenesis and what genes and
proteins are regulated by Cbk1 and/or its downstream transcription factor Ace2?
Through this novel application of a CHI-based screening strategy, we have identified
RAM/Ace2 transcriptional targets and generated genetic evidence for an interaction
between the RAM and PKA pathways during morphogenesis. Follow-up studies of the
screening results further suggest that a balance between RAM and PKA-pathway
activity is required for cells to establish a normal distribution of morphotypes
during nutrient-induced filamentation. Taken together with previous work on these
two pathways, our observations support a model where PKA-regulated transcriptional
activity is most important in the transcription of RAM/PKA co-regulated genes early
in morphogenesis, while the RAM pathway is more important as daughter nuclei
accumulate within the hyphal structure.
An outline of the CHI-based screening strategy is presented in Figure 1. In preparation for
the CHI screen, we first constructed a transposon suitable for large-scale
insertional mutagenesis in C. albicans. To enable efficient
mutagenesis with limited transposition bias, we generated a donor plasmid
derived from the bacterial element Tn7. The
Tn7 system has been used extensively for in
vitro mutagenesis [29], [30] with low reported insertion site specificity [31]. For
purposes of this screen, the Tn7 element was modified to
contain a recyclable allele of the CaURA3 gene; specifically,
we inserted the URA3-dpl200 allele into Tn7
sequence encoded in the donor plasmid pGPS3. The URA3-dpl200
allele was designed by Wilson et al.
[32] to allow
recombinational excision of the URA3 gene under
counter-selection with 5-fluoro-orotic acid (5-FOA). Subsequently, we performed
in vitro mutagenesis of the genomic library pEMBLY23 (Materials and Methods) derived from
C.albicans strain WO-1. Non-specific Tn7
transposition was achieved using the TnsA, TnsB, and TnsC* proteins paired
with the TnsAB transposase and appropriate cofactors [29]. The genomic library was
mutagenized to yield an estimated 20,000 independent insertions. The resulting
insertional library was recovered in E. coli, and genomic DNA
inserts were released by enzyme digestion for introduction into the C.
albicans Ura- parental strain,
cbk1Δ/CBK1 (CAMM292, see Table S1
for strain table). By homologous recombination, the mutagenized genomic DNA
fragment will replace its native chromosomal locus, thereby generating a
heterozygous insertion mutant in the parent
cbk1Δ/CBK1 strain. DNA transformations
were performed nine times, yielding a total of 6528 independent Ura+
transformants. The C. albicans double heterozygotes were
isolated and screened for decreased filamentation as follows.
The cbk1Δ/CBK1 mutant was originally
studied in C. albicans by McNemar and Fonzi [24] and was
found to be haploinsufficient with respect to filamentation on a variety of
media. Uhl et al. also isolated a heterozygous
cbk1 insertion mutant in their haploinsufficiency screen
[15]. As
shown in Figure 2A,
cbk1Δ/CBK1 colonies show a decreased
area of central wrinkling and a more prominent ring of peripheral filamentation
on SM at 37°C. The haploinsufficiency of this parental strain was
advantageous for two reasons. First, it provided increased sensitivity in that
the strain was already deficient for filamentation. Second, it could also
improve specificity because weak phenotypes of non-interacting,
transposon-derived mutants would not be apparent due to masking by the
cbk1Δ/CBK1 phenotype.
As described above [24], [25], [27], RAM pathway mutants show two distinct phenotypes
depending on the conditions used to induce filamentation, but both phenotypes
are apparent on solid Spider Medium (SM). In order to identify mutations that
potentially interacted with both general functions of the pathway, we,
therefore, screened for decreased filamentation on SM at 37°C. All
subsequent experiments were conducted under these conditions unless otherwise
indicated.
The library of 6528 complex heterozygous mutants was spotted in 96-well format
and scored for decreased peripheral invasion and altered colony wrinkling
relative to a Ura+ derivative of the parental
cbk1Δ/CBK1 strain (11, Figure 2A). Clones showing
both phenotypes were re-tested using two independent colonies. A total of 441
complex heterozygous mutants with decreased peripheral invasion and altered
colony wrinkling were re-confirmed on both SM and SM containing uridine to
control for positional effects of the URA3 gene (Figure 2A). We specifically
selected mutants with decreased zones of peripheral
filamentation and less pronounced central wrinkling relative to
the parental strain (Figure 2A and
2B). All mutants showed some degree of peripheral filamentation. The
most common composite phenotype indicated a small zone of peripheral agar
invasion with a broad region of moderate wrinkling (Figure 2B).
The transposon insertion sites for approximately one-third of the mutants (139
strains) showing potential synthetic genetic interactions were identified using
a 3′-RACE/sequencing approach (see Materials
and Methods), yielding 42 unique transposon-derived mutations as
putative CBK1-interactors. Since 8 insertion sites were
identified in at least 5 separate clones (Figure 3A and 3B), the screen appeared to be
saturated to the limits of the library and the mutagenesis technique. Therefore,
we did not sequence the remaining two-thirds of the mutants and focused on
evaluating the initial set of 42 mutants. It is, however, important to note that
the screen itself is unlikely to be saturated for all possible
CBK1 interactors, as the library almost certainly did not
contain insertions in all predicted C. albicans genes.
The URA3 marker was recycled from the heterozygotes by 5-FOA
mediated recombinational excision [32]. Following phenotypic
re-testing to confirm that homozygosis was not responsible for curing the
URA3 marker, CBK1 was re-integrated at its
chromosomal position using plasmid pMM4 [24]. The phenotypes of 41 of
42 candidate CHI strains were modified by re-integration of
CBK1 (Figure
2A), indicating that the observed phenotypes were dependent on the
cbk1 mutation and were likely due to a synthetic genetic
interaction between cbk1Δ/CBK1 and the
transposon insertion. The high percentage of CBK1-dependent
phenotypes may be due to the fact that the parental cbk1Δ
heterozygote is itself haploinsufficient on SM and most non-interacting
insertion mutations that are themselves haploinsufficient do not have
sufficiently strong phenotypes to appreciably change the phenotype of the double
heterozygote relative to the parental strain. To confirm these interactions
further, a subset of ten complex heterozygous mutants was independently
constructed from CAMM-292 by single gene-replacement [33]. All ten double mutants
recapitulated phenotypes displayed by the transposon-derived mutants and showed
distinct phenotypes relative to strains with single deletions of the interacting
genes. Representative images from this analysis are shown in Figure 2B.
To further characterize the morphologies of the mutants, we determined the
proportion of yeast, pseudohyphae and hyphae after 3 hours induction in liquid
SM at 37°C. The interacting mutants consistently showed an increased
proportion of pseudohyphal cells relative to wild type and
cbk1Δ/CBK1 strains (Figure 2C). Similarly,
examination of cells scraped from SM plates showed that the filaments of double
mutants had constricted septal areas characteristic of pseudohyphae (Figure 2D). Importantly, all
of the mutants were indistinguishable from wild type and the parental strain
when serum was used as the inducer of filamentation (data not shown). Since
ace2Δ/Δ mutants also show decreased peripheral
invasion, decreased central wrinkling, increased levels of pseudohyphae, and
normal filamentation on serum (25), we conclude that the majority of the
CBK1-interacting genes isolated in the screen appear to
affect the Ace2-dependent functions of the RAM pathway.
Literature analysis of the set of CBK1-interactoring genes
revealed that approximately one-half are involved in glycolysis/respiration,
biosynthesis, or cell wall metabolism (Figure 3B), cell processes consistent with
established functions of the RAM pathway [24]–[28]. An
important interactor in terms of validating the screen is SSD1
because it is a likely Cbk1 substrate in S. cerevisiae
[34],
displaying well-characterized genetic interactions with CBK1 in
both S. cerevisiae
[35] and
C. albicans
[27]. Comparison
of our dataset with that generated by the haploinsufficiency screen of Uhl
et al. revealed no overlap [15]. As discussed above, we
suspect that this lack of overlap is also related to the fact that our parental
strain is haploinsufficient for filamentation and, thus, non-interacting
transposon-derived mutations causing simple haploinsufficiency were, in effect,
masked by the phenotype of the parental strain.
In principle, the Ace2-deficient phenotypes displayed by the double heterozygous
mutants could result from mutations that interfere with the activation of Ace2
or from mutations that affect a key transcriptional target of Ace2. We isolated
three mutants that could cause a CHI-interaction with CBK1
through the former mechanism. First, we isolated orthologs of two genes that
regulate mitotic exit in S. cerevisiae, CDH1
[36] and
SLK19
[37]. Ace2 is
well known to localize to the nuclei of daughter cells in both S.
cerevisiae
[38] and
C. albicans
[25], [39]. Since Cdh1
and Slk19 regulate mitotic exit, the point in the cell cycle when Ace2 localizes
to the nuclei [38], we suggest that disruption of mitotic exit through
the loss of these proteins may further decrease the overall activity of Ace2. In
addition, NSP1, a key component of the nuclear import
machinery, was isolated. Studies in S. cerevisiae
[40] have
indicated that decreased NSP1 gene dosage leads to inhibition
of nuclear import, and it seems plausible that a strain lacking an allele of
NSP1 could have decreased nuclear import of Ace2 which
would further decrease the overall Ace2-mediated transcriptional activity of the
cbk1Δ/CBK1 mutant.
The larger class of CBK1-interacting mutants that relate to Ace2
function is the set of genes that appear to be part of the transcriptional
output of the RAM pathway (Figure
3C and 3D). To identify such genes in our data set, we searched the
promoter regions of CBK1-interactors and found 22 genes that
contain a C. albicans Ace2-consensus binding sequence [MMCCASC, 26].
Of these genes, 11 have been shown to display decreased expression in
ace2Δ/Δ mutants during hyphal induction as reported
in a recent transcriptional profiling study [26]. To further confirm that
our screen identified genes regulated by Ace2, we examined the binding of Ace2
to the promoters of 5 CBK1-interactors with consensus binding
sites (ACT1, ADH1, ENO1,
HGT6, & RGD3) during both yeast and
hypha-phase growth by chromatin immunoprecipitation (ChIP). Consistent with ChIP
data for Ace2 reported by Wang et al.
[39], the
absolute enrichment was relatively low, most likely due to its cell cycle
regulation and our non-synchronous experiments (Figure 4). Nevertheless, all five promoters
were bound by Ace2 at levels comparable to those observed for the
well-established Ace2 target CHT3 and to those reported by Wang
et al. [39]
during yeast growth. In addition, three promoters were bound in hyphal phase
(Figure 4). Taken
together, the presence of Ace2 binding sites, the transcriptional profiling
data, and ChIP data support the notion that many of the
CBK1-interacting genes are transcriptional targets of Ace2.
Comparison of the set of CBK1-interactors with data from a
variety of transcriptional profiles of C. albicans
morphogenesis indicated that a substantial subset of
CBK1-interactors (14 interactors, 34%) are regulated by
the cAMP/PKA pathway through the transcription factor Efg1 [41]. Indeed, 10
CBK1-interactors contain consensus binding sites for both
Ace2 and Efg1 (Figure 3C and
3D), suggesting that these two transcription factors may regulate a
common set of genes. Further supporting this notion are previous studies
indicating that both Ace2 and Efg1 induce glycolytic genes and repress genes
involved in oxidative respiration [26], [41]. Indeed, we searched the
C. albicans genome and found that the promoters of 384
genes contain consensus binding sites for both Ace2 and Efg1 (Table S2).
Consistent with previous studies of the two pathways, the set of putatively
co-regulated genes is enriched for genes contributing to glycolysis,
biosynthesis, and cellular stress responses. Recently, Wang et
al. have also shown that the promoters of Ace2-regulated cell wall
and cell separation genes are bound by both Efg1 and Ace2 during morphogenesis
[39]. Taken
together our genetic data strongly support the notion that genes regulated by
the PKA pathway may also be important components of the transcriptional output
of the RAM pathway during morphogenesis.
In addition to transcriptional targets of the PKA pathway, three other
CBK1 interactors (MAF1,
SLF1 & ACT1) have connections to the
PKA pathway (Figure 3A).
MAF1 and SFL1 are both orthologs of
PKA-regulated transcriptional regulators in S. cerevisiae
[42], [43],
suggesting that proper PKA-mediated transcriptional control is important in the
absence of full RAM pathway activity. Further suggesting that the activity of
the PKA pathway is important in RAM pathway mutants, we isolated
ACT1 as a CBK1-interactor. Although
ACT1 is, of course, a crucial part of the cell
cytoskeleton, it also plays an important role in activation of the cAMP/PKA
pathway. The Sundstrom lab has shown that actin dynamics regulate PKA activity
[44]
and, recently, Zou et al. have elegantly demonstrated that
actin functions as part of a PKA sensor/activator complex during hyphal
development [45].
Indeed, decreased G-actin levels lead to decreased PKA pathway activity and, in
turn, decreased filamentation in C. albicans
[45]. As such,
one explanation for the interaction between ACT1 and
CBK1 is that the lowered ACT1 gene dosage
in the act1Δ/ACT1
cbk1Δ/CBK1 mutant exacerbates the
filamentation defects of decreased RAM pathway activity by concomitantly
limiting PKA activity. This explanation also implies that the PKA pathway may
compensate for decreased RAM pathway activity during morphogenesis.
To test the hypothesis that the RAM and PKA pathways regulate a common set of
genes during morphogenesis, we examined the expression of two
CBK1-interacting genes containing both Ace2 and Efg1
binding sites in ace2Δ/Δ and
efg1Δ/Δ mutants after 3 hours of hyphal induction with
SM. As shown in Figure 5A,
the expression of the transcripts increased in both strains relative to wild
type by quantitative RT-PCR. These observations suggest either that Efg1 and
Ace2 are functioning as transcriptional repressors or that compensatory
responses are occurring to maintain expression of these genes during
morphogenesis when one of the two pathways is disabled.
To test the latter hypothesis, total cell lysates of the RAM pathway mutant
ace2Δ/Δ were prepared and the level of PKA
enzymatic activity determined after 3 hours exposure to hypha-inducing
conditions (Figure 5B). At
this time point, PKA activity has reduced to low levels in wild type cells [45], but there is
clearly increased PKA activity in the ace2Δ/Δ mutant.
This suggests that the PKA pathway is hyperactive in RAM pathway mutants and is
consistent with the hypothesis that the PKA pathway may compensate for decreased
RAM pathway activity. To further test the interaction between the RAM and PKA
pathways, we deleted one allele of CBK1 in strains containing
homozygous null mutations in one of the catalytic subunits of the PKA enzyme
[46] to
yield the mutants cbk1Δ/CBK1
tpk1Δ/Δ and cbk1Δ/CBK1
tpk2Δ/Δ. The two triple mutants along with wild type and
the parental mutants were incubated in SM for 3 hours at 37°C to induce
filamentation. As shown in Figure
5C, deletion of TPK1 in the
cbk1Δ/CBK1 background decreases the
proportion of pseudohyphae formed by the
cbk1Δ/CBK1 mutant, while deletion of
TPK2 has no effect (data not shown), suggesting that the
increased proportion of pseudohyphae formed by
cbk1Δ/CBK1 is dependent on
TPK1. The phenotypic differences evident upon deleting the
two isoforms of PKA are consistent with previous data indicating that they have
distinct and redundant roles in filamentation [46].
Interestingly, cultures of cbk1Δ/CBK1
tpk1Δ/Δ in SM contained significant numbers of filaments that
showed characteristics of both pseudohyphae and hyphae (Figure 5D). This hybrid morphology was not
observed in cultures of wild type,
cbk1Δ/CBK1, or
tpk1Δ/Δ cells. Similar hyphae-pseudohyphae hybrid
morphologies were recently observed by Carlisle et al. in cells
expressing an intermediate level of UME6
[47],
suggesting that concurrent disruption of both RAM and PKA pathways interferes
with the ability of the cell to commit to one morphotype. These observations
also suggest that a balance between the activities of the PKA and RAM pathway is
required for normal morphogenesis.
Increased and/or dysregulated PKA pathway activity has been linked previously to
increased pseudohyphae formation. For example, Tebarth et al.
have shown that overexpression of EFG1 induces constitutive
pseudohyphae [48]. We, therefore, hypothesized that elevated PKA
activity might be responsible for the constitutively pseudohyphal phenotype
displayed by ace2Δ/Δ as well as the increased
proportion of pseudohyphae observed with
cbk1Δ/CBK1 heterozygotes showing CHI.
Three observations support this hypothesis. First, treatment of
ace2Δ/Δ cells with the substrate-based PKA
inhibitor MyrPKI [49], under non-inducing conditions, significantly
increased the number of yeast-like cells and decreased the number of mature
pseudohyphae (Figure 6A),
strongly supporting the notion that increased PKA activity is involved in the
constitutive pseudohyphal phenotype of ace2Δ/Δ. Second,
EFG1 expression is elevated in both RAM pathway mutants and
cbk1Δ heterozygotes relative to wild type over the time
course of hyphal induction (Figure
6B). Densitometric analysis of three replicates of the 180 min time
point indicates that the EFG1 levels are 2–4 fold higher
in each of the mutants relative to wild type (p<0.02,
Student's t test). To further confirm this elevation, we
compared the levels of EFG1 in wild type and the double
heterozygote cbk1Δ/CBK1
pgk1Δ/PGK1. Consistent with the
semi-quantitative data, EFG1 is elevated in
cbk1Δ/CBK1
pgk1Δ/PGK1 relative to wild type (4.8
log2, std. dev. 0.9,
p = 0.01, Student's
t test). Third, deletion of both alleles of
EFG1 in the
cbk1Δ/CBK1 background decreases
expression of ENO1 by a modest 1.5-fold and
PGK1 a more significant 8-fold relative to the parental
strain (Figure 6C),
indicating that at least a portion of the increased expression of putatively
co-regulated genes in RAM mutants is mediated by the PKA-Efg1 pathway. Taken
together, these experiments suggest that some of the
CBK1-interacting genes isolated in our screen are part of the
transcriptional output of both the PKA and RAM pathways and that decreased RAM
function in the CBK1 double heterozygotes leads to a
compensatory increase in PKA pathway activity which, in turn, manifests as a
phenotype of increased pseudohyphal growth due to increased
EFG1 levels [45].
Although our results strongly suggest that the RAM and PKA pathways interact
during morphogenesis and that the PKA pathway may be hyper-activated in the
absence of RAM activity, it remained to be determined how these pathways
interact during normal morphogenesis. As discussed above, one of the best
characterized functions of Ace2 in both S. cerevisiae and
C. albicans is as a daughter cell-specific transcription
factor [26],
[38],
[39]. Two
other laboratories [26], [39] have previously shown that in C.
albicans, Ace2 localizes to daughter nuclei in actively dividing
yeast-phase cells as well as in serum-induced filaments; our results confirm
those findings in SM (Figure
7A). We, therefore, hypothesized that the relative contributions of
Ace2 and Efg1 to gene regulation during the course of hyphal development may
correspond to the timing of their nuclear localization. To our knowledge, the
nuclear localization of Efg1 during filamentation had not been described
previously.
To test this hypothesis, we used indirect immunofluorescence to compare the
proportion of cells with nuclear Efg1 at the initiation of hyphal development to
the proportion in hyphal cell nuclei. As shown in Figure 7B, Efg1 is present in the nuclei of
50–60% (n = 100 cells) of cells prior to
shifting to SM. In contrast, Efg1 is detectable in only ∼10% of
hyphal nuclei. Correspondingly, Efg1 occupancy of the promoter regions of
ENO1 and PGK1 is also higher at the
initiation of hyphal development by ChIP analysis (Figure 7C). This suggests that Efg1 may be
more important at the onset of, or early in, the filamentous transition, while
Ace2 contributes to Efg1/Ace2 co-regulated gene transcription as daughter cell
nuclei accumulate within the hyphal structure.
Consistent with this model, ACE2 expression increases over the
3-hour time course of hyphal induction (Figure 7D); this finding is also consistent
with its role in gene expression within daughter cell nuclei. Interestingly, the
promoter region of ACE2 has five Efg1 consensus binding sites,
suggesting that the PKA pathway may contribute to the regulation of
ACE2 expression. However, treatment with the PKA inhibitor
MyrPKI reduced levels of ACE2 expression only modestly after 3
hours in SM (Figure 7E).
Although this observation supports a possible direct link between the PKA and
RAM pathways, it suggests that PKA-Efg1 is not the sole, or even dominant,
regulator of ACE2 expression.
As a whole, these data support a model in which Efg1 plays a more important role
at the initiation of hyphal development in SM, and Ace2 plays a more important
role once daughter nuclei accumulate within the hyphal structure. Since
EFG1 expression is maintained throughout the time course of
hyphal development (Figure
6B) and Efg1 is present in some hyphal nuclei (Figure 7A), it is unlikely that the
relationship between Ace2 and Efg1 represents an “either/or” type of
scenario. Instead, it seems more likely that a balance exists between the
relative contributions of the two transcription factors to gene expression and
that this balance varies during hyphal development.
Methods for the large-scale genetic analysis of Candida albicans
have advanced tremendously in recent years, leading to a number of important and
informative studies [14]–[20]. To our knowledge, however, no large-scale synthetic
genetic analyses have yet been reported. Here, we present the first such screen. Our
approach was based on a CHI strategy, and, like other large-scale genetic analyses
of C. albicans, we employed transponson-mediated insertional
mutagenesis to generate a large collection of double heterozygous mutants derived
from a parental strain containing a heterozygous null mutation of the RAM pathway
kinase CBK1. This library was then used to screen for genes that
interacted with CBK1 during SM-induced morphogenesis.
First and foremost, our data establishes that CHI-based genetic interaction screening
is a useful method for the genetic analysis of the obligate diploid yeast C.
albicans. A priori, CHI-based genetic screening of a signaling network
such as the RAM pathway would be expected to identify genes that interact with the
query gene through a variety of mechanisms. Inspection of our dataset confirms these
expectations in that it includes transcriptional targets of the RAM pathway (e.g.,
ENO1, PGK1), genes that likely affect the
function of pathway components (e.g., NSP1,
SLK19), and genes that function in parallel pathways (e.g.,
MAF1, SLF1). In the specific case of screening
a protein kinase mutant, it should also be possible to identify substrates of that
kinase. Although no bona fide substrate of Cbk1 has been confirmed in C.
albicans, our screen identified a very likely candidate in Ssd1. Ssd1
is a well characterized Cbk1 substrate in S. cerevisiae
[34] and has been
shown previously to interact genetically with CBK1 in both
S. cerevisiae
[35] and
C. albicans
[27]. A consensus
Cbk1 phosphorylation sequence has recently been identified in S.
cerevisiae
[34]. Supporting
the possibility that CaSsd1 is a substrate of
CaCbk1 is the presence of this consensus phosphorylation sequence.
Of the remaining CBK1-interactors, RGD3, an
uncharacterized potential Rho GTPase, and VPS13, a protein involved
in vacuolar protein sorting, also have sequences that match the consensus
phosphorylation sequence for ScCbk1 (data not shown). Studies
directed towards confirming these putative Cbk1 substrates are in progress.
The CBK1-derived double heterozygous mutants isolated in our screen
displayed phenotypes indicative of defects in the Ace2-dependent functions of the
RAM pathway in that they were only observed on SM [25]; mutations in genes affecting
Ace2-independent functions would be expected to display filamentation defects on
both SM and serum [27]. Since many of the interacting genes appear to be
transcriptional targets of Ace2, we propose that the effect of partially disabling
the RAM pathway by deletion of one allele of CBK1 is exacerbated by
further deletion of one allele of a gene regulated by the
CBK1-dependent transcription factor Ace2. The cumulative effect of
these two mutations results in phenotypes (increased proportion of pseudohyphae)
consistent with a further decrease in Ace2-mediated RAM transcriptional activity. By
this analysis, Ace2-transcriptional targets that display CHI interactions with
CBK1 would, therefore, appear to be particularly important
components of the transcriptional output of the RAM pathway during morphogenesis on
SM.
A particularly powerful feature of synthetic genetic analysis is the ability to
identify interactions between regulatory pathways and, in this regard, our CHI
screen of cbk1Δ/CBK1 was quite informative,
highlighting the interplay between the RAM and PKA pathways during morphogenesis.
Although no components of the PKA signaling pathway were identified as
CBK1-interactors, analysis of the dataset revealed that many of
the interactors were regulated by the PKA pathway. Indeed, the similar
transcriptional characteristics of the PKA-regulated transcription factor Efg1 and
Ace2 in C. albicans have been previously noted [50] and, while our
work was in progress, Wang et al. reported that Efg1 and Ace2 bound
to the promoters of C. albicans genes involved in cell separation
[39]. In
addition, the PKA and RAM pathways have been linked genetically in S.
cerevisiae through experiments showing that ectopic over-expression of
the PKA kinase subunit TPK1 suppresses growth and budding defects
of RAM pathway mutants in an Ace2-independent manner [51]. Our data
suggest that the PKA and RAM pathway interact in C. albicans with
respect to Ace2-dependent functions.
Consistent with this model, consensus binding sites for both Efg1 and Ace2 are
located in the promoter regions of a significant proportion of
CBK1-interactors. A genome-wide search identified 384 putative
Efg1/Ace2 co-regulated genes, suggesting that the two pathways interact to modulate
the expression of a substantial subset of genes. The interaction of these two
pathways is further supported by our isolation of two PKA-regulated transcriptional
modulators (MAF1 & SLF1) as
CBK1 interactors as well as by the synthetic genetic
interactions between CBK1 and TPK1 observed in our
follow-up studies.
The simplest manifestation of a model in which the PKA and RAM pathways co-regulate a
set of genes would be that deletion of either ACE2 or
EFG1 results in the decreased expression of co-regulated genes.
As shown in Figure 5A, this is
not the case as the expression of putatively co-regulated genes is increased in both
ace2Δ/Δ and efg1Δ/Δ mutants.
This suggested that the two pathways may compensate for one another when the other
pathway is disabled. Supporting this notion, the activity of the PKA pathway is
increased in RAM pathway mutants (Figure 5B), and EFG1 mediates a substantial portion of
the increased expression of co-regulated genes in the absence of full RAM pathway
activity (Figure 6B).
Accordingly, the level of EFG1 expression is also increased (Figure 6B) and, since
inappropriately high levels of EFG1 promote pseudohyphal growth
(48), this observation provides an explanation for the increased amounts of
pseudohyphae displayed by RAM pathway mutants.
We, therefore, propose that the increased PKA activity in RAM pathway mutants
represents a compensatory response that maintains expression of Ace2/Efg1
co-regulated genes in the absence of a fully functional RAM pathway. However,
constitutively elevated levels of PKA activity represent a dysregulated state and,
consequently, the expression levels of the genes are not returned to normal but are
elevated. Thus, it appears that a balance between the activity of the PKA and RAM
pathways is required to maintain properly regulated expression of co-regulated
genes. Maintaining a balance between the activities of the two pathways appears to
be required for normal hyphal development because: 1) loss of EFG1
leads to a failure to form filaments; 2) loss of ACE2 leads to the
accumulation of pseudohyphae; and 3) concurrent partial disruption of both pathways
leads to the formation of filaments with characteristics of both hyphae and
pseudohyphae (Figure 5D).
If, as our results suggest, a balance between PKA and RAM pathway-mediated
transcription is required for the cell to normally undergo filamentation, then how
is this balance established and maintained? Although further work will be required
to determine the molecular mechanism of this interaction, the cell cycle-regulated
nature of both Efg1 and Ace2 suggests that the pathways might be active at different
times during morphogenesis. Ace2, for example, localizes to the nuclei of daughter
cells in both yeast and filamentous C. albicans
[26], [39]. Efg1, on the
other hand, has been shown to be rapidly down-regulated soon after hyphal induction
in some conditions [48]. These considerations led us to propose that Efg1 may be
more important in the expression of co-regulated genes earlier in morphogenesis,
while Ace2 is the dominant regulator later in morphogenesis when daughter nuclei
appear within the filament.
Consistent with that model, we showed that more nuclei contain Efg1 at the initiation
of morphogenesis than later in the process. Ace2, on the other hand, is absent from
the vast majority of nuclei at the initiation of morphogenesis but is found in
daughter nuclei as they accumulate within the filament (Figure 7A). Consistent with its role later in
morphogenesis, overall expression of ACE2 also increases as the
cells are exposed to inducing condition for longer periods of time (Figure 7D). Since Efg1 remains
detectable in hyphal nuclei (Figure
7B), it is unlikely that Ace2 replaces Efg1 entirely but rather Ace2 may
become relatively more important as daughter nuclei accumulate within the filament
and undergo mitosis. Thus, it seems that a balance between the PKA and RAM pathways
exists and that this balance is important for smooth morphogenesis. A potential
illustration of the importance of this balance is provided by the morphologies
displayed by the tpk1Δ/Δ
cbk1Δ/CBK1 mutant in which single
filaments show characteristics of both hyphae and pseudohyphae.
This model is also consistent with the observations of Wang et al.,
who reported that Efg1 represses the expression of Ace2-regulated cell separation
genes during hyphal development [39]. They found that in wild type strains, the Ace2-regulated
expression of chitinase CHT3 occurs approximately 3 hours
post-hyphal induction, a point at which multiple septa and daughter nuclei have
formed within the hyphal filament. The 3-hour time point also corresponds to the
time when we observed high levels of ACE2 expression. In
EFG1 mutants, on the other hand, Wang et al.
found that CHT3 is inappropriately expressed within the first hour
of induction and is expressed at higher levels at 3 hours [39]. Our observations regarding the
timing of Efg1 nuclear localization correlate well with these expression data in
that Efg1 is present early when it suppresses Ace2-mediated CHT3
expression but is absent when CHT3 expression is induced. It is
important to note that Efg1 has previously been proposed to function as both a
transcriptional activator and repressor during hyphal morphogenesis [39], [41] and, taken
together with the observations of Wang et al., our data are
consistent with such a role.
At this point, further work will be required to understand the molecular mechanisms
by which the RAM and PKA pathway interact. As noted above, ACE2
does possess a number of Efg1 consensus binding sites within its promoter. This
suggests a possible feed-forward mechanism by which Efg1 activates the expression of
ACE2, which, in turn, takes over transcription of co-regulated
genes. However, chemical inhibition of the PKA pathway only modestly reduced
expression of ACE2 during hyphal induction (Figure 7E). Similarly,
efg1Δ/Δ mutants also exhibit very slight changes in
ACE2 expression (data not shown). Although there may be an
operative component of this feed-forward mechanism, it seems to be a relatively
minor contributor to the crosstalk between these pathways.
In summary, we have shown that CHI-based genetic interaction screening is a useful
approach for the analysis of complex phenotypes in C. albicans. The
application of this approach to the RAM pathway has provided insights into the
mechanisms by which the PKA and RAM signaling pathways function together during the
transition from yeast to filamentous cells in C. albicans.
All strains are derived from CAI4
(ura3Δ::imm434/ura3Δ::imm434).
CAMM-292
(ura3Δ::imm434/ura3Δ::imm434/cbk1-Δ1::hisG/CBK1)
[24] was
used as the parental strain for transposon mutagenesis. A complete list of
strains and genotypes is provided in Table S1. Yeast peptone dextrose supplemented
with 80 mg/L uridine, synthetic dextrose medium lacking uracil, and SM were
prepared using standard recipes [15], [52]. Induction of filamentation was carried out using SM
plates (37°C, 3D) or liquid SM (37°C, 3 h). All phenotypes were
confirmed on SM plates supplemented with uracil to control for possible
positional effects of URA3 expression. Proportions of yeast,
pseudohyphae and hyphae in liquid cultures were determined by light microscopy
using morphological scoring criteria described by Sudbery et
al.
[53].
C.albicans strain WO-1 pEMBLY23 genomic DNA library (NIH AIDS
Research & Reference Reagent Program) was mutagenized (9 independent
reactions) in vitro using the GPS3-Mutagenesis system from New
England Biolabs (Beverly, MA) and a donor plasmid (pGPS3) containing the
CaURA3-dpl200 cassette [16] inserted at the
Spe I restriction site. Mutagenized genomic fragments were
released by PvuII digestion and transformed into CAMM-292 using a lithium
acetate-protocol with heat shock at 44°C for 20 min [54]. The library is available
upon request from the Kumar laboratory ([email protected]).
Transposon insertion sites were amplified by 3′ RACE (rapid amplification
of cDNA ends) using primers complementary to the ends of the transposon
construct, cloned into a TA vector, and sequenced. Insertion sites were then
identified by BLASTN searches using the Candida Genome Database (www.candidagenome.org).
Ten double heterozygotes that showed CHI were independently constructed from the
Ura- parental strain cbk1Δ/CBK1 (CAMM292)
using fusion PCR methods to generate URA3-based knockout
cassettes [33].
The cassettes were used to transform CAMM292 to Ura prototrophy, and correct
integration was confirmed by PCR. Two independent isolates were evaluated for
all phenotypes.
Total RNA was isolated using the RiboPure Yeast Kit (Ambion, Austin, TX) and
reverse transcribed using the SuperScript III First Strand Synthesis Kit
(Invitrogen, Carlsbad, CA). Changes in transcript levels of target genes were
analyzed using the Platinum SYBR Green Mix (Invitrogen) and normalized to
ACT1 levels using the
2−ΔΔCt method [55]. ChIP
assays were performed as described previously [56] using Ura+ CAI4-dervatives
containing ACE2-TAP and EFG1-MYC alleles.
Protein kinase A activity was measured in total cell lysates using the PepTag
cAMP-dependent protein kinase assay kit (Promega, Madison WI) following a
protocol previously developed for C. albicans
[57]. Lysates
were prepared from wild type and ace2Δ/Δ cells that had
been exposed to SM for 3 h. Phosphorylation of the PepTag substrate was
determined by agarose gel electrophoresis; the unphosphorylated substrate
migrates in the opposite direction as the phosphorylated substrate. Images of
the gel were captured on a gel-doc imaging system and processed using Adobe
PhotoShop software. Identical contrast and levels were used for each image.
Light and fluorescence microscopy was performed using a Nikon ES80
epi-fluorescence microscope equipped with a CoolSnap CCD camera. Images were
collected using NIS-Elements Software and processed in PhotoShop. Indirect
immunofluorescence was performed as previously described using anti-Myc
(Invitrogen) primary- and TexasRed-conjugated (Molecular Probes) secondary-
antibodies [58]. DAPI and Calcofluor white staining was performed as
described [52].
|
10.1371/journal.pbio.0060292 | COPI Complex Is a Regulator of Lipid Homeostasis | Lipid droplets are ubiquitous triglyceride and sterol ester storage organelles required for energy storage homeostasis and biosynthesis. Although little is known about lipid droplet formation and regulation, it is clear that members of the PAT (perilipin, adipocyte differentiation related protein, tail interacting protein of 47 kDa) protein family coat the droplet surface and mediate interactions with lipases that remobilize the stored lipids. We identified key Drosophila candidate genes for lipid droplet regulation by RNA interference (RNAi) screening with an image segmentation-based optical read-out system, and show that these regulatory functions are conserved in the mouse. Those include the vesicle-mediated Coat Protein Complex I (COPI) transport complex, which is required for limiting lipid storage. We found that COPI components regulate the PAT protein composition at the lipid droplet surface, and promote the association of adipocyte triglyceride lipase (ATGL) with the lipid droplet surface to mediate lipolysis. Two compounds known to inhibit COPI function, Exo1 and Brefeldin A, phenocopy COPI knockdowns. Furthermore, RNAi inhibition of ATGL and simultaneous drug treatment indicate that COPI and ATGL function in the same pathway. These data indicate that the COPI complex is an evolutionarily conserved regulator of lipid homeostasis, and highlight an interaction between vesicle transport systems and lipid droplets.
| Fat cells, and cells in general, convert fatty acids into triglycerides that are stored in droplets for future use. Despite the enormous importance of lipid droplets in obesity and other disease processes, we know very little about how lipid reserves in droplets are formed and how those reserves are drawn down. We have used the model fruit fly Drosophila to identify candidate regulators of lipid storage and utilization, and have shown that many of these candidates have functions that are conserved in mammals. We focused our attention on a vesicle-trafficking pathway that we show is required for the modulation of the types of regulatory and enzymatic proteins found on the lipid droplet surface. Interfering with the function of this trafficking system with either RNA interference or small-molecule compounds alters lipid storage. The understanding of this new pathway, as well as the specific reagents we used, may ultimately lead to new therapeutics.
| Lipid homeostasis is critical in health and disease, but remains poorly understood (for review see [1]). Non-esterified free fatty acid (NEFA) is used for energy generation in beta-oxidation, membrane phospholipid synthesis, signaling, and in regulation of transcription factors such as the peroxisome proliferator-activated receptors (PPARs). Essentially all cells take up excess NEFA and convert it to energy-rich neutral lipids in the form of triglycerides (TG). TG is packaged into specialized organelles called lipid droplets. NEFA is regenerated from lipid droplet stores to meet metabolic and energy needs, and lipid droplets protect cells against lipotoxicity by sequestering excess NEFA. Lipid droplets are the main energy storage organelles and are thus central to our understanding of energy homeostasis. Despite their importance, we know very little about the ontogeny and regulation of these organelles.
Lipid droplets are believed to form in the ER membrane by incorporating a growing TG core between the leaflets of the bilayer, and ultimately are released surrounded by a phospholipid monolayer. Cytosolic lipid droplets possess a protein coat and grow by synthesis of TG at the lipid droplet surface [2] and by fusion with other lipid droplets [3]. Formation of nascent droplets and aggregation of existing droplets is likely to require a dynamic exchange of lipids and proteins from and to the droplet. Indeed, the range of proteins identified in lipid droplet proteomic studies suggests extensive trafficking between lipid droplets and other cellular compartments, including the endoplasmic reticulum (ER) [4–6]. Additionally, lipid droplet-associated proteins translocate between the cytosol and lipid droplets [7]. For example, tail interacting protein of 47 kDa (TIP47) associates with small, putative nascent, lipid droplets [8–10], but is not found on larger droplets, which are coated by other members of the perilipin, adipocyte differentiation related protein (ADRP), TIP47 (PAT) protein family. Intriguingly, TIP47 mediates mannose 6-phosphate receptor trafficking between the lysosome and Golgi [11], raising the possibility that trafficking is involved in lipid droplet ontogeny or fate. However, unlike the well-studied Golgi trafficking system, the routes to and from the lipid droplet are unknown.
Once lipid droplets are formed, stored TG is mobilized in a regulated manner. Triglyceride, diglyceride (DG), and monoglyceride lipases convert TG back into NEFA. Most of our knowledge concerning lipolysis is based on extensively studied adipocytes in which at least two lipolytic enzymes have been identified: adipocyte triglyceride lipase (ATGL) [12–14] and hormone sensitive lipase (HSL) [15]. Due to the hydrophobic properties of the lipid droplet TG core, lipases are likely to act at the surface of lipid droplets [16], where members of the PAT protein family regulate lipase access to the TG core. Mammalian genomes encode at least five PAT-proteins. Whereas perilipin is the dominant PAT protein in adipocytes, ADRP is the dominant PAT protein in nonadipose tissues in which it is tightly associated with the lipid droplet surface [17]. PAT members appear to have a hierarchical affinity for the lipid droplet surface. In nonmammalian genomes, there are fewer PAT proteins. For example, two PAT proteins termed lipid storage droplet 1 and 2 (LSD-1 and LSD-2) are found in Drosophila melanogaster [10]. The crucial role of PAT proteins is evolutionary conserved as the absence of perilipin in mice [18,19], or LSD-2 in flies [20,21] results in lean animals. Overexpression of LSD-2 results in obese flies [20]. These data indicate the conserved PAT proteins at the lipid droplet surface are important regulators of energy storage.
It seems likely that PAT proteins protect lipid from lipolysis, but the role of PAT proteins may not be limited to passive steric hindrance of lipase access to the TG core, as illustrated by perilipin. Unphosphorylated perilipin protects the lipid droplet from lipase activity. Following stimulation by protein kinase A (PKA), however, phospho-perilipin acts as a docking site for HSL [22,23], which translocates from the cytosol to the droplet surface [24]. Whereas phospho-perilipin promotes massive NEFA release from the droplet, this is not mediated exclusively by HSL, as mice lacking HSL function show a relatively mild phenotype marked by the accumulation of DG, thus demonstrating that HSL acts as a DG lipase in vivo [25]. The TG lipase functioning in HSL null mice is ATGL. In the current view of adipocyte lipolysis, ATGL is responsible for the first step in TG hydrolysis, liberating DG and NEFA, whereas HSL acts as a DG lipase. We know very little about how ATGL is targeted to the lipid droplet.
In contrast to the lean phenotype in animals lacking perilipin (mouse) or LSD-2 (fly), both mice and flies lacking ATGL are obese. In mice, the absence of ATGL results in excessive TG accumulation in liver and muscle [12,14]. Similarly, human patients suffering from neutral lipid storage disease carry mutations resulting in truncated ATGL isoforms [26]. ATGL function is evolutionary conserved, as flies lacking the Drosophila ATGL ortholog, Brummer, accumulate copious amounts of body fat [13]. The lipid droplet-associated protein Comparative Gene Identification-58 (CGI-58) acts as an ATGL colipase [27]. Mutations in the CGI-58 gene result in ectopic fat accumulation in patients suffering from Chanarin Dorfman Syndrome (CDS, [28]), supporting the idea that both ATGL and CGI-58 are required for mobilizing lipid stores in nonadipose tissue. Interestingly, CGI-58 physically interacts with perilipin as demonstrated by both coimmunoprecipitation and fluorescence resonance energy transfer (FRET) studies [22,29,30]. In addition, there are other lipases and probably many more cofactors encoded in the genome. Understanding which ones act at the lipid droplet surface and how their localization is regulated will be important.
Drosophila is a powerful model for pathway discovery due to well-developed genetics. Additionally, greater than 60% of the genes associated with human disease have clear orthologs in Drosophila [31]. Drosophila is highly relevant to lipid droplet study, as lipid droplets in Drosophila and mammals are associated with many of the same proteins [4–6,32–35]. Finally, the emerging model of lipid storage and endocrine regulation are similar in humans and Drosophila [36], suggesting that Drosophila will be a good genetic model for lipid storage and lipid storage diseases in humans. We therefore utilized genome-wide RNA interference (RNAi) screening in Drosophila tissue culture cells to identify and characterize novel regulators of lipid storage. We then tested for the function of these regulators in mouse lipid droplet regulation by directed RNAi studies. We identified 318 Drosophila genes required to limit lipid storage and 208 Drosophila genes required to promote lipid storage. These genes encode known regulators of lipid storage as well as genes not previously associated with lipid storage regulation.
Because the protein composition of the lipid droplet surface is so critical for lipid droplet function, and because very little is known about how lipid droplet decoration is regulated, we focused on the exciting finding that the retrograde vesicle-trafficking machinery, utilizing the Coat Protein Complex I (COPI) and COPI regulators, was required to utilize lipid stores. COPI subunit knockdown by RNAi, as well as COPI inhibition with compounds, resulted in increased lipid storage both in Drosophila and mouse tissue culture cells, demonstrating evolutionary conservation of our findings.
COPI and COPII vesicles are essential components of the trafficking machinery cycling between the ER and Golgi (reviewed in, e.g., [37]). COPI vesicles mediate cargo transport from the Golgi back to the ER, including escaped ER-resident proteins. The anterograde counterpart, COPII, mediates transport of proteins and lipids from the ER to the Golgi. Whereas interference with either COPI or COPII complexes disrupts Golgi function [38,39], only COPI was required for lipid droplet utilization, clearly demonstrating that COPI and not general Golgi function is required for TG utilization. Although we certainly do not rule out communication between the Golgi and lipid droplet, we suggest that there is a novel ER/lipid droplet trafficking system using a subset of the ER/Golgi transport machinery.
We found that the basis for lipid overstorage following COPI knockdown was a decreased lipolytic rate. Using our existing knowledge of the PAT family members and lipases in the regulation of lipolysis, we examined changes in protein composition at the lipid droplet surface. Interestingly, we found that interfering with the COPI pathway results in ectopic accumulation of TIP47 at the lipid droplet surface. Furthermore, ATGL at the lipid droplet surface was greatly reduced. Combining the effects of ATGL knockdown and compounds affecting COPI function did not elicit a stronger decrease in lipolysis, indicating that ATGL and COPI are both part of the same lipolytic pathway. Thus, our studies provide a functional link between COPI retrograde trafficking and the proteins at the lipid droplet surface. More generally, these results indicate that Drosophila RNAi screening is suited to detect uncharted pathways affecting NEFA regulation and to achieve a deeper understanding of cellular lipid droplet regulation.
Lipid droplets are well studied in mammalian cells, but Drosophila cells have not been extensively used in lipid droplet studies. Lipid droplets are ubiquitous organelles, and we found that Drosophila S2 and SL2 (unpublished data), as well as S3 and Kc167 cells (this study) accumulated TG in lipid droplets in the presence of excess NEFA. Kc167 cells, for example, stored little lipid when grown on standard media (Figure 1A), whereas in the presence of NEFA (400 μM oleic acid), they readily (within 12 h) accumulated TG packaged in droplets (Figure 1B), which we visualized with the lipid-specific dye BODIPY493/503 [40].
Treatment of Drosophila cells with double-stranded RNA (dsRNA) decreases, or “knocks down,” transcript levels for genes sharing the dsRNA sequence, a process known as RNAi [41]. To help determine whether Drosophila tissue culture is a good model for lipid droplet function, we used RNAi to target genes encoding known lipid droplet regulators. Flies or mice lacking ATGL store more TG than wild type (“overstorage”) [12–14], whereas those lacking diacylglycerol acyl transferase1 (Dgat1), a key enzyme in TG synthesis [42,43], store less lipid (“understorage”). Knockdown of bmm, which encodes Drosophila ATGL, increased lipid storage as expected (Figure 1C and 1D). Conversely, treating cells with dsRNA targeting midway (mdy), which encodes Drosophila Dgat1, decreased lipid storage (Figure 1E and 1F). Thus, Drosophila cells can be used to analyze gene functions necessary to increase as well as decrease lipid storage.
Although differences in lipid storage are often obvious, we were interested in generating a fully quantitative dataset to support future meta-analysis. To systematically identify and characterize the genes involved in lipid storage, we developed a microscopy-based quantification method based on image segmentation and measurement of nuclear to lipid droplet cross-sectional area (see Figure 2A–2D and Materials and Methods). This technique allowed us to detect lipid storage differences caused by the different feeding conditions and control dsRNA treatments (Figure 2E). We used this imaging method to perform a genome-wide RNAi screen with the well-characterized dsRNA library of the Harvard Drosophila RNAi Screening Center (DRSC). This collection covered more than 95% of the predicted Drosophila genes [44]. dsRNAs against bmm and mdy were included in each screening plate as controls. We also included wells with no dsRNA and with or without oleic acid as controls.
As a screening cell line, we used Kc167 cells, which showed the best balance of lipid droplet deposition, RNAi susceptibility characteristics, and adhesion during assay development (unpublished data). Following dsRNA treatment of oleic acid-fed cells and image analysis, ratiometric data were normalized within plates and across the entire screening collection using linear models, B-score, Z-score/median absolute deviation (MAD), and strictly standardized mean difference (SSMD) [45–48], all of which gave similar results. B-score normalization [46] across the entire screen marginally out-performed other methods (see Materials and Methods, Table S1). B-score results were used for all analyses reported here.
Rank-order analysis of the genome-wide screening results demonstrated that the majority of dsRNAs had no effect on lipid storage. However, two cohorts of dsRNAs resulted in lipid overstorage, as expected for genes required for promoting lipid utilization, or understorage, as expected for genes required for promoting lipid storage (Figure 2F). Thresholds for determining whether a particular dsRNA resulted in a phenotype were selected to balance false negatives and false positives based on the results for bmm, mdy, and no oleic acid controls. At B-scores ≥ 2.0 and ≤ −1.7, greater than 89% of wells treated with dsRNA targeting bmm or without oleic acid resulted in the correct overstorage or understorage call, respectively (Figure 2G). Using these cutoffs, we identified 208 candidate genes required for increasing lipid storage (understorage on knockdown, B-score ≤ −1.7, Tables S2 and S9) and 318 required for reducing lipid storage or lipid utilization (overstorage on knockdown, B-score ≥ 2.0, Tables S3 and S9). These data suggest that about 3% of the Drosophila genome is directly or indirectly involved in lipid storage. All data are available in the supplement (Table S4) and at http://lipofly.mpibpc.mpg.de/ and http://flyrnai.org.
The most critical test of screen performance is coherence as measured by the identification of multiple genes in a multisubunit complex or a known pathway [49]. Such coherent gene sets are also the best candidates for more detailed analysis. To categorize the dsRNA phenotypes according to molecular networks, we analyzed the identified genes using Gene Ontology (GO) [50] terms with the VLAD tool [51]. This analysis allows for the detection of statistically overrepresented GO terms among a set of genes and projects those enrichments onto the GO-term hierarchy. Genes with a possible function in lipid storage regulation as detected by the RNAi screen were tested against the complete Drosophila gene set for enrichment of GO terms associated with biological process, molecular function, and cellular component. Identified, enriched terms were structured in hierarchical networks (Figures 3–5; the results are also tabulated in Table S5). We also took advantage of data from a concurrent lipid storage screen using an independent dsRNA library and Drosophila S2 cells [52]. This allows us to develop a robust overview of lipid droplet storage.
Duplication of extensive RNAi screens using different libraries on different cell types provides a cross-validating function that is extremely useful in the analysis of comprehensive datasets. The overlap (25%, 57 genes) between the S2 cell screen (227 genes identified; Table S6) and our genome-wide study on Kc167 cells (526 genes identified) was highly significant (p < 1e−14, Wilcox test). More importantly, the GO term networks were quite similar and suggest that key pathways have been identified (Figure 3). For example, both screens show that interfering with translation factors and ribosomes result in lipid storage defects (GO:0022613, GO:0006412). Additionally, genes resulting in lipid storage defects are enriched for transcriptional regulators in both screens (GO:0010467) and trafficking (GO:0006911, GO:0006890). The only major differences between the screens were that genes involved in pre-mRNA processing were enriched in our Kc167 cell screen and genes involved in proteasome function were enriched in the S2 cell screen. However, five genes required for lipid storage in our study (suppressor of deltex, ubiquitin conjugating enzyme 2, ubiquitin activating enzyme 1, Roc1a, and Roc1b) are involved in ubiquitin-mediated proteolysis at the proteasome [53]. Thus, the screens are largely cross-validating.
Gene knockdowns resulting in understorage have a candidate wild-type function in promoting lipid storage. Whereas we identified gene functions linked to neutral lipid synthesis (Table S5), the most striking enrichments were for regulatory functions within the nucleus (Figure 4). GO terms associated with the nuclear functions transcription or transcript processing were particularly prominent (Figure 4). These data suggest that lipid storage requires a complex regulatory network.
In contrast, the candidate genes required for lipid utilization were enriched for cytoplasmic functions (Figure 5, Table S5). We found that lipid storage increased after treatment with dsRNAs targeting genes encoding lipid droplet-associated proteins (GO:0005811). In addition to GO term analysis, we directly compared the identified candidate lipid storage-modulating genes functions with genes encoding proteins of the recently described, but functionally uncharacterized, lipid droplet-associated mammalian [5,6,32,35] and Drosophila [4,33] subproteomes, only some of which have lipid droplet GO terms. These genes were far more likely to result in a lipid overstorage phenotype when subjected to knockdown in Drosophila (p > 1e−16, Wilcox test) than the reference genome-wide dsRNA targets. This suggests that many of the genes revealed by our RNAi experiments encode direct regulators of lipid storage. Gene functions involved in mitochondrial fatty acid beta-oxidation, which utilize NEFA as a substrate, as well as genes involved in protein synthesis, were also enriched. Indeed, knockdown of 12% of the Drosophila genes encoding translation-related functions (GO:0006412), including 32% of the genes encoding ribosomal subunits (GO:0033279), resulted in lipid overstorage (Figure 5, Table S5). It is possible that decreased ATP demand for protein synthesis and decreased ATP generation in mitochondria simply decrease the need for energy in the cells, resulting in increased lipid storage. Mitochondrial uncoupling and beta-oxidation pathways are areas of therapeutic interest for diabetes and other metabolic disorders [54–57].
One of the most striking results was the prevalence of cellular transport functions in general (GO:0006909, GO:0006890; and GO:0000022), and the COPI trafficking pathway mediating Golgi to ER transport in particular, among the genes resulting in a lipid overstorage phenotype on knockdown (Figure 5, Table S5). Nascent lipid droplets are thought to form at the ER and then enlarge and fuse to form larger droplets [8–10]. Thus, our result is somewhat surprising, as we expected that wild-type ER functions might be involved in promoting lipid storage rather than lipid utilization. Similarly, it is known that lipid droplets are transported as cargo on microtubules in Drosophila embryos and that such transport is required for fusion of lipid droplets in muscle cells [3,58]. There was a strong enrichment for genes involved in spindle microtubule elongation (Figure 5C) among the genes showing overstorage on knockdown. Again, whereas microtubule involvement in lipid storage is predicted, interfering with microtubule cargo transport might be expected to decrease lipid storage.
To validate a “gold set” of genes ready for extended follow-up, we selected genes for additional Drosophila treatments using original and secondary dsRNAs. At least two different nonoverlapping dsRNAs in our screen or in the Guo et al. screen [52] resulted in confirmed understorage or overstorage phenotypes for a subset of candidate genes (Table S7). Additionally, mouse orthologs of 127 Drosophila genes selected on the basis of lipid storage phenotypes in Kc167 cells (including orthologs of 54 genes that failed to pass our cutoff) were knocked down in two mouse cell lines using short interfering RNAs (siRNAs). We used a mouse fibroblast cell line (3T3-L1), in which lipid droplets have been extensively characterized, and a liver cell line, AML12, which was previously used as a model of ectopic fat deposition [59]. Retesting in mouse cells is a particularly stringent validation of the Drosophila dsRNA data as it simultaneously provides information about evolutionary conservation as well as obviating concerns about spurious off-target effects [49,60,61]. The 33 genes resulting in lipid storage defects when knocked down in both Drosophila and in mouse cells validate the involvement of many of the biological processes implicated by the primary screen (Table S7). For example, knockdown of the Ubiquinol cytochrome c reductase complex III subunit VII gene (Uqcrq; ortholog of the Drosophila CG7580 gene), which encodes a component of the mitochondrial respiration chain, results in greatly enlarged AML12 cells storing dramatically more lipid than control cells (Figure 6A–6C). Similarly, knockdown of Smarca4 (ortholog of the Drosophila brahma gene), which encodes a member of the SWI/SNF chromatin modifying complex [62], results in lipid overstorage (Figure 6D). Knockdowns of COPI complex members resulted in overstorage in Drosophila S2 and Kc167 cells, and in mouse 3T3-L1 and AML12 cells (Table S7). Although there is much to be gleaned from the screen, we focused our attention on the Golgi to ER trafficking COPI complex.
Overrepresentation of genes encoding ER/Golgi vesicle-associated proteins among the genes showing a lipid overstorage phenotype on knockdown suggests that vesicle trafficking proteins participate in lipid utilization. Most strikingly, six out of the seven genes encoding COPI subunits (Figure 7) that mediate retrograde transport from the Golgi to the ER, showed dramatically increased lipid storage following dsRNA treatment in the genome-wide RNAi screen (B-score = 4.6 to 11.1, false discovery rate [FDR]-corrected p = 1e−5 to 1e−34). Enrichment for members of such multisubunit complexes in RNAi screens has outstanding predictive value [49]. Our observed enrichment for essentially all the COPI-associated factors among the knockdowns resulting in lipid overstorage, strongly suggests that COPI is required for limiting lipid storage (FDR-corrected p < 1e−6). In addition, dsRNAs targeting ADP ribosylation factor at 79F (Arf79F) had the same effect as COPI knockdown. Arf79F encodes a small G protein homologous to mammalian Arf1, the key regulator of COPI vesicle formation at the Golgi [63]. Surprisingly, εCOP was the only COPI subunit repeatedly failing to produce a lipid storage phenotype following RNAi in both the S2 [52] and our Kc167 cell screens. Although this is a negative result, we suggest that this subunit is not involved in lipid storage regulation (see Discussion). Interestingly, none of the seven COPII members required for anterograde transport from the ER to the Golgi [37,38] showed a lipid accumulation phenotype following RNAi (Figure 7B; B-score = 0.0 to 1.4, FDR-p = 0.99 to 0.78), strongly suggesting that lipid overstorage due to COPI knockdown is not a general consequence of disrupted trafficking between the ER and Golgi.
In organisms, cells are exposed to differing NEFA levels due to feeding and fasting. Therefore, to test for the function of the COPI complex in physiological conditions without elevated NEFA, we also performed new RNAi experiments with or without supplementing the media with oleic acid (Figure 8A–8G; additional data not shown). Even in the absence of oleic acid, knockdowns of all the members of the COPI complex that promoted lipid droplet deposition under fed conditions also promoted accumulation without feeding (Figure 8A–8G; additional data not shown). Thus, the lipid storage phenotype was also independent of the nutritional status of the cells.
To further investigate whether the observed lipid storage phenotype after the loss of COPI-subunit function is due to a specific pathway or a more general effect of interference with Golgi and ER integrity, we also tested additional dsRNAs targeting transcripts encoding the COPII-associated proteins CG10882, Sar1, Sec23, Sec31, and PLD (Figure 8G). Furthermore, the Drosophila genome encodes five Arf proteins [64], which we also reinvestigated in additional RNAi experiments. Arf79F encodes ARF1, which is required for COPI function, but Arf51F, Arf72A, Arf84F, and Arf102F are not known to be required for COPI-mediated transport [65]. Only Arf79F resulted in a mutant lipid droplet phenotype upon RNAi knockdown (Figure 8G). These experiments demonstrated that the lipid overstorage phenotype is specific to COPI loss of function and raise the possibility that the lipid overstorage phenotype is Golgi independent.
Although multiple dsRNAs verified the phenotypic effect of COPI knockdown, we sought to further validate those results with an independent technique, to rule out effects based on the RNAi treatment, or the prolonged incubation time (4 d) due to the knockdown procedure. Therefore, we also tested pharmacologically for COPI involvement in lipid storage. We treated Drosophila S3 cells for 18 h with 24 different concentrations of Exo1, a selective inhibitor of Arf1 activity [66], and determined the dose response (Figure 8H). Lipid droplets were stained with the same dye as for the RNAi experiments. As in the RNAi experiments, we used internal controls, including cells with no oleic acid feeding, cells treated with the compound solvent, and cells treated with Triacsin C, a known inhibitor of TG synthesis [67]. Dose-response curves for Exo1 were determined by enumerating cells that showed an increase in lipid staining (or relative to the enumerated cells based on a cytosolic counterstain; inset in Figure 8H) and expressing this as per cent activity relative to cells incubated in oleic acid and full inhibited by Triacsin C. Thus, increased activity indicates lipid overstorage. Treatment of Drosophila S3 cells with Exo1 resulted in a dose-dependent (half maximal effective concentration [EC50] = 5 μM) increase in lipid storage that was greater than 10-fold. Thus, multiple dsRNAs targeting COPI and Arf79F mRNAs as well as Exo1, a compound targeting Arf79F (Arf1 in mammals), resulted in the same phenotype. These data strongly indicate that COPI is required to limit lipid storage in droplets in Drosophila.
To explore the function of COPI in lipid droplet cell biology in greater detail, we performed additional experiments in the mouse 3T3-L1 and AML12 cells. As positive and negative controls, we used irrelevant “ALLStars negative control” siRNAs, or siRNAs targeting transcripts encoding known lipid droplet regulators, and compared the resulting cellular phenotypes to the results of parallel siRNA treatments targeting transcripts encoding COPI components. As in the Drosophila experiments, we required that at least two siRNAs resulted in the same phenotype.
Like AML12 cells, 3T3-L1 cells also stored little lipid in the absence of exogenous NEFA (Figure 9A and 9G), whereas small, clustered lipid droplets appeared upon addition of oleic acid (Figure 9B and 9H). Depletion of both ADRP and TIP47 by RNAi resulted in fewer and much larger lipid droplets (Figure 9C and 9I [68]) relative to wild type (we used double knockdowns for these controls because single knockdowns resulted in a minimal phenotype [68]). Conversely, knockdown of Atgl (bmm in Drosophila) transcripts resulted in increased lipid storage (Figure 9D and 9J), but no differences in the appearance of the lipid droplets. Targeting the genes encoding α, β, β′, γ, δ, or ζ COPI subunits by siRNAs resulted in increased lipid storage (Figure 9E, 9F, and 9K–9P). As in the Drosophila knockdown experiments, εCOP knockdown failed to increase lipid storage (unpublished data). We also failed to observe a phenotype following knockdown of either of two genes, sec24 and Pld1, encoding COPII components (unpublished data). Thus, the Drosophila and mouse RNAi experiments unambiguously indicate that COPI subunits (with the exception of εCOP) have evolutionarily conserved lipid droplet functions.
Both Arf1 and Gbf1, an Arf guanine nucleotide exchange factor (GEF), are required for COPI recruitment from the cytosol to Golgi [69]. We also asked whether Arf1 and any of three pharmacologically related GEFs were required for lipid utilization. The Gbf1, Big1, and Big2 proteins are GEFs inhibited by Brefeldin A (BFA) [70]. BFA treatment and knockdowns of either Arf1 or Gbf1 (the latter confirmed at the protein level) resulted in lipid overstorage (Figure 9Q and 9R), whereas we observed no lipid overstorage following knockdown of Big1 or Big2 (unpublished data). Thus the COPI complex and critical regulators of COPI translocation are required for lipid utilization.
Lipid overstorage in the absence of COPI could be due to decreased release of NEFA from droplets, or increased synthesis of TG for storage, or both. In order to explore whether COPI is required for one or both of these general functions, we measured both NEFA release and esterification of NEFA into TG in AML12 cells (Figure 10A). As expected, we observed increased release of NEFA from cells treated with control siRNAs targeting Adrp and Tip47 transcripts, which is mediated by increased amounts of lipid droplet-associated ATGL [68]. In contrast, NEFA release decreased when Atgl lipase transcripts were targeted as controls. Additionally, we observed increased incorporation of NEFA into TG following Atgl knockdown, suggesting that the tremendous increase in TG seen in those cells is due to decreased NEFA release and continued synthesis of TG despite the reduced efflux. The modest increase in incorporation of NEFA into TG following COPI knockdown was insignificant. However, we observed approximately 40% of wild-type NEFA release in cells treated with siRNAs targeting either γCOPI or ζCOPI transcripts—in the same range as after Atgl knockdown (Figure 10A). In separate experiments, we also observed decreased NEFA release following Gbf1 knockdown, but not following Big1 or Big2 knockdown (Table S8). These data indicate that COPI is a novel regulator of lipolysis.
We also asked whether short-term pharmacological inhibition of COPI trafficking phenocopies the COPI knockdown phenotype in mouse cells, as we noted in Drosophila cells. We used COPI inhibitors Exo1 and BFA [39,66], both of which result in increased lipid storage. Both compounds reduced NEFA release to the same extent as the siRNAs targeting COPI subunit mRNAs (Figure 10B). To dissect the role of COPI in lipolysis, we used a combination of siRNAs targeting different genes in the lipolytic pathway, and Exo1 or BFA treatment, to mimic genetic epistasis experiments (a proven tool for dissecting functional relationships between members of the same or different pathways [71]). Combining siRNA-mediated knockdown of COPI members and BFA or Exo1 treatment did not enhance the decreased lipolysis phenotype (Figure10B), indicating that the observed effects following drug treatment are only COPI mediated. Additionally, these data suggest that there are no serious compound-based side effects vis-a-vis lipid droplets, even for the broad-spectrum inhibitor BFA (also note that other BFA-sensitive GEFs, Big1 and Big2, did not result in a lipid storage phenotype on knockdown). Decreased lipolysis could be due to decreased lipase activity at the lipid droplet. To determine whether that lipase was ATGL, we combined siRNAs targeting ATGL transcripts and BFA or Exo1 drug treatment (Figure 10B). If ATGL were responsible, then ATGL knockdown would have no effect on BFA- or Exo1-treated cells. Indeed, the lipolysis rate was not further decreased, suggesting that COPI-mediated lipolysis effects are mediated by ATGL. This conclusion is further supported by experiments in which we treated cells with siRNAs targeting ADRP and TIP47 transcripts in combination with either BFA or Exo1. In the absence of ADRP and TIP47, more ATGL is found at the lipid droplet surface [68]. We also found that Exo1 or BFA treatment rescues the effect of ADRP and TIP47 knockdown. This, along with the finding that COPI and ATGL are in the same pathway, suggests that COPI is an important positive regulator of ATGL.
Wild-type COPI could mediate release of NEFA from lipid droplets by altering the heterogeneous and dynamic collection of lipid droplet-associated proteins found in different cell types and conditions [72]. To further explore what happens to lipid droplets following COPI knockdown, we examined the distribution of TIP47 and ADRP on the lipid droplet surface. These are the only PAT proteins expressed in AML12 cells [68]. In control cells incubated with oleic acid, and control siRNAs, ADRP was associated with the lipid droplet surface whereas TIP47 was mostly found in smaller punctate cytoplasmic inclusions and more ill-defined cytoplasmic locations ([68] and Figure 11A). TIP47 and ADRP were not colocalized in untreated cells. Following siRNA treatments targeting α, β, β′, γ, and ζ COPI subunit or Gbf1 transcripts, both ADRP and TIP47 were observed on the same lipid droplets (Figure 11B–11H). Treating the cells with BFA had the same effect on TIP47 localization (Figure 12A and 12B). These data indicate that COPI is required for a wild-type pattern of PAT localization to the lipid droplet.
PAT proteins are tightly associated with the lipid droplet surface. In order to distinguish localization to the region of the lipid droplet from true localization to the lipid droplet surface, we treated cells with BFA after oleic acid feeding, and isolated lipid droplets by sucrose gradient ultracentrifugation. This treatment separates the lipid droplets from cytosol and other membrane fractions. To determine what proteins were on the lipid droplets, western blots were probed with antibodies detecting ADRP, TIP47, and ATGL, as well as the ATGL cofactor CGI-58. Whereas ADRP and CGI-58 remained quantitatively unchanged after BFA treatment, TIP47 protein levels in the lipid droplet fraction increased nearly 2-fold (Figure 12C). There was no change in TIP47 in the cytosolic fraction (unpublished). The cell-staining experiments showed a more dramatic increase in TIP47 at the ADRP-positive lipid droplets than we observed in the western blots, but importantly, both cell staining and western blotting show increased TIP47 on COPI inhibition. Strikingly, ATGL levels decreased to near or below the detection limit, suggesting that BFA treatment drives ATGL off the lipid droplet surface, or prevents ATGL association with the lipid droplet (Figure 12C). Thus, both cell staining and analysis of isolated droplets indicate that wild-type COPI limits abundance of TIP47 at the lipid droplet surface and is required for ATGL localization to the droplet surface. Taken together with the epistasis results demonstrating that COPI and ATGL function in the same pathway, these results indicate that COPI-mediated targeting of ATGL to the lipid droplet is required for lipolysis.
Positive regulation of lipolysis by the COPI retrograde-vesicle trafficking pathway was the most striking and unexpected result of our screen. We have found that interference with COPI function, either by RNAi or compounds, in Drosophila Kc167 or S3 cells, or in mouse 3T3-L1 or AML12 cells, results in increased lipid storage. Furthermore, recent and parallel studies in yeast [73] and Drosophila S2 cells [52] also suggested a role of COPI function in lipid droplet regulation. Interestingly, only the ε-subunit of the COPI complex failed to result in a lipid droplet deposition phenotype on knockdown. Although we cannot rule out limited RNAi efficacy or increased protein stability, εCOP was the only canonical COP subunit not resulting in a lipid storage phenotype in a parallel study using different cells and reagents [52], and we found that targeting of εCOP transcripts by RNAi in AML12 cells had a weak effect on lipid storage at best. Finally, εCOP is the only dispensable subunit in a recent study identifying COPI activity coupled with fatty acid biosynthesis as a host factor important for Drosophila C virus replication [74]. This is especially interesting, as certain enveloped viruses, including Hepatitis C virus, assemble on lipid droplets [75,76]. Taken together, these results indicate that six out of the seven wild-type COPI subunits mediate lipid storage by positively regulating lipolysis.
COPI could have a direct or indirect effect on lipid storage. The indirect mechanism is poorly defined, but if the Golgi is a “sink” for phospholipids derived from TG stores, then decreased Golgi function could simply decrease demand for TG substrate. If NEFA (from the media in fed cells, and from biosynthesis in unfed cells) conversion to TG continues, then increased lipid droplet volume would occur. It is also possible that canonical COPI function transporting lipids and proteins from the Golgi to the ER is ultimately responsible for lipid droplet utilization and protein composition at the lipid droplet surface. For example, COPI might be required for the particular phospholipid composition in hemimembranes formed on nascent droplets, which secondarily alter TIP47 and ATGL localization in mature lipid droplets.
However, evidence that Golgi function per se is not linked to lipid storage phenotypes, as well as direct association of COPI members and regulators with the lipid droplet or PAT proteins supports a more direct model. The COPI and COPII pathways have established roles as constitutive vesicle transport systems that cycle proteins as well as lipid from the Golgi to the ER (COPI), or vice versa (COPII) [37]. Interference with either of the COP trafficking systems results in disturbed ER and Golgi function [38,39]. The lipid overstorage phenotype was only seen in the case of interference with COPI trafficking. This indicates that the lipid overstorage phenotype is not a simple consequence of ER and Golgi function. Finally, in an indirect model in which COPI shuttles only between the Golgi and the ER, COPI should not be lipid droplet associated. However, COPI subunits are directly associated with the lipid droplet surface as shown by proteomics [6]. Additionally, Arf1 binds to ADRP, which is exclusively associated with the lipid droplet surface [77]. Arf79F, the Drosophila homolog of mammalian Arf1, also localizes to lipid droplets in Drosophila S2 cells [52].
We propose that COPI is likely to function directly at the lipid droplet surface and not indirectly through the Golgi (Figure 13). Perhaps COPI is a destination-specific transporter returning lipid droplet surface hemimembrane and Golgi membrane to the ER. The transport system that brings nascent lipid droplets from the ER to the lipid droplet has not been elucidated, but it is intriguing that the transport/PAT protein TIP47 is found preferentially on small lipid droplets. Small lipid droplets derived from the ER are thought to help build larger droplets by fusion. TIP47-coated droplets might form in the ER, and then COPI could return TIP47 to the ER after the lipid cargo is deposited. In this model, TIP47 becomes trapped at the lipid droplet surface in the absence of COPI.
Although we observed increased TIP47 on ADRP-positive droplets by both western blot and cell staining, the cell staining result was more dramatic. Our model might also explain why. The punctate staining of TIP47 in untreated cells could be due to TIP47 on nascent droplets that might also cofractionate with the larger ADRP-positive droplets in the western blots, leading to a less dramatic enrichment for TIP47 relative to ADRP in that experiment. However, we cannot rule out other explanations, such as nonlinear detection of antigen concentration or epitope masking in the cell staining experiments.
COPI perturbation increases stored TG by decreasing the lipolysis rate (this study, [52]) indicating that the wild-type COPI complex promotes lipolysis. We have shown that COPI directly or indirectly removes TIP47 from the lipid droplet surface and promotes ATGL localization to the droplet surface, where lipolysis occurs. ATGL has a key role in lipid droplet utilization, and ATGL association with the droplet is reduced by ADRP and Tip47 [68]. Our epistasis experiments combining siRNA-mediated ATGL knockdown and BFA or Exo1 compound treatment demonstrated that the decrease in lipolysis rate is due to loss of ATGL activity. COPI activity specifically alters lipid droplet surface composition by increasing the amount of TIP47 and reducing the amount of ATGL at ADRP-coated lipid droplets. We suggest that COPI negatively regulates localization of TIP47. TIP47 in turn prevents ATGL localization. The rescue of the double-knockdown phenotype of TIP47 and ADRP by BFA or Exo1 suggests that COPI has an independent feed-forward effect on ATGL levels at the lipid droplet surface.
Although we have focused our attention here on COPI, our systematic and genome-wide exploration of gene functions required for lipid storage in Drosophila significantly increases experimental access to the complex molecular processes regulating lipid storage and utilization. Further, the use of multiple screens using different cell types and different organisms greatly increases confidence in the genes in the intersection. Given widespread concerns about RNAi screening efficacy and off-target effects, as well as the time and effort required for downstream analysis, systematic use of multiple species and libraries to address a single biological question might be cost effective in addition to resulting in more durable datasets. Primary screens in Drosophila cells followed by secondary screens in mouse cells are much less expensive than a similar genome-wide screen in mammalian cells. Additionally, the availability of mutants in most Drosophila genes, along with demonstrated translation to mammalian systems, provides a valuable entry point for in-depth analyses in both fly and mouse; and eventually for the selection of therapeutic targets for emerging problems associated with obesity and other metabolic disorders.
We used the Harvard Drosophila RNAi Screening Center (DRSC, http://www.flyrnai.org) dsRNA collection, which covers more than 95% of the transcriptome (Release 3.2 BDGP) with a total of 17,076 dsRNAs [44] in duplicate. We seeded 1.5 × 104 Kc167 cells (DRSC) in 10 μl of serum-free Schneider's medium (GIBCO) in each well of microscopy-quality 384-well plates containing the pre-aliquoted dsRNAs (approximately 250 ng of dsRNA/well). Plates were spun at 1,200 rpm for 1 min and incubated for 45 min at 25 °C. We then added 40 μl of complete Schneider's medium supplemented with 10% FCS (JRH Biosciences), 50 units penicillin; and 50 μg of streptomycin/ml (GIBCO) and ±400 μM oleic acid (Calbiochem) complexed to 0.4% BSA (Sigma). Plates were sealed and incubated in a humidified incubator at 25 °C for 4 d. The cells were subsequently fixed for 10 min in 4% formaldehyde in PBS followed by a 10-min permeabilization step in PBS including 0.1% Triton X-100. For lipid droplet visualization and cell counting (nuclei), we incubated for 1 h with PBS including 5 μg/ml BODIPY493/503 (Molecular Probes) and 5 μg/ml DAPI or 5 μg/ml Hoechst33342 (Molecular Probes). After two washes with PBS including 0.01% Tween-20, cells were kept in 40 μl of PBS and visualized with a 20× objective on a Discovery1 automated microscope system (Molecular Devices).
A subset of 276 genes of the primary screen library were targeted by 362 additional dsRNAs (Table S10) generated from PCR products obtained from the Drosophila RNAi screening center of Harvard (DRSC). PCR fragments were reamplified using a modified T7 oligonucleotide (5′-GTA ATA CGA CTC ACT ATA GG-3′) and a touchdown PCR protocol. PCR products were subsequently used for in vitro transcription reactions using T7 RNA polymerase (Fermentas). Following DNAse-mediated digestion of the PCR template, dsRNAs were purified with Multiscreen PCR purification filter plates (Millipore). RNAi treatment was performed either as described for the primary screen in optical-quality 96-well plates (BD) with adjusted dsRNA and cell numbers, in duplicate (approximately 1 μg of dsRNA and 5 × 104 cells/well). Imaging was performed either with a BD Pathway 855 Bioimager automated microscope (BD) or with a Zeiss Axiovert200M (Carl Zeiss) and the OpenLab software (Improvision).
For the secondary mouse siRNA screen (Table S10), we used AML12 murine liver cells (Steven Farmer, Boston University) and 3T3-L1 fibroblast cells (ATCC) grown according to protocols of the American Type Culture Collection (ATCC). Assays were done in 96-multiwell plates (Fisher Scientific) at a density of 0.25 × 104 cells/well on growth medium supplemented with 200 μM oleic acid, which was added 18 h prior to fixation of the cells. Cells were transfected with Hiperfect transfection reagent (0.75 μl/well) (Qiagen) and experimental or ALLStars negative control siRNA oligonucleotides (10 nM), according to the manufacturer's instructions (Qiagen). Four days after transfection, cells were fixed and stained as described above for the Drosophila cells and imaged with a BD Pathway 855 Bioimager automated microscope system (BD).
Images of Drosophila cells (two sites/well in the primary screen; six sites/well for the secondary screen) were processed with a custom image segmentation algorithm (available from M. Beller upon request) written for the ImageJ software package [78]. After a sharpening and a brightness/contrast adjustment (for the BODIPY images; equal values for all images) or a gamma correction (for the DAPI images; same values for all images), a background subtraction followed by an Otsu thresholding step was run (Figure 1A–1D). Watershed processing to identify solitary particles followed. Finally, lipid droplets or nuclei were identified with the generic “analyze particles” function of the ImageJ software with the following settings: (1) settings for the nuclei: size from 10 to 10,000 pixels, 256 bins, outlines as well as measurement results displayed, measurements on the edges excluded, clear results, flood, and summary of the results; and (2) settings for the lipid droplets: identical parameters except size ranging from one to 200 pixels and a circularity from zero to one. For each detected particle, the size and area were measured. For each image, the total numbers of particles (“counts”) or cumulative measured area for all particles (“area”) are reported. A custom Perl script concatenated the summarized measurements, and the obtained information was used to calculate the ratio of lipid droplets per cells as a measure of lipid storage (“lipid droplet/nuclei (area)” or “lipid droplet/nuclei (counts)”).
Mammalian cell image analysis (four sites/well) was performed as described above with some adjusted settings reflecting the larger mammalian cell size as well as differences in imaging equipment (no brightness or contrast adjustments were applied). The generic “analyze particles” function of the ImageJ software was used with the following settings: (1) settings for the nuclei: size from 80 to 10,000 pixels, 256 bins, outlines as well as measurement results displayed, measurements on the edges excluded, clear results, flood, and summary of the results; and (2) settings for the lipid droplets: identical parameters, except the size ranging from one to 2,000 pixels and a circularity from zero to one.
The general thrust of the analysis is given below and is followed by a detailed description. Screen data are available (Table S4; http://lipofly.mpibpc.mpg.de/). Results were robust to data handling method (Table S1). Genes passing thresholding conditions (Tables S2 and S3) were used for the GO term analysis (Table S5). B-score p-values can be used to further restrict the gene lists shown in Tables S2 and S3.
Data analysis was performed with custom scripts written in the R language and packages provided by the Bioconductor project [79]. The lipid droplets and nuclei area measurements of the two images per well were used to calculate an averaged lipid area per nuclei area value per well. Additionally to the primary images, a number of wells required reimaging based on visual inspection (size of the complete dataset: N = 48,241 wells). To identify and extract images with bad quality, the values for lipid droplet (LD) area and count measurements as well as for the corresponding nuclei measurements of the two images per well were plotted against each other to look for variation within wells. In addition, the corresponding “LD area per nuclei area” and “LD count per nuclei count” ratios were plotted against each other per well. These plots showed 95 prominent outliers (segmentation artifacts/“bad” wells), which were removed (resulting N = 48,146 wells). The data values of reimaged wells were averaged.
The screen dataset was platewise normalized for within-plate and between-plate differences by four different algorithms. Because of the limited number of controls per plate, 98% of the wells per plate were used as a reference set in the normalization procedure as proposed in [47] in which the largest and smallest 1% values of the plate were removed to generate the reference set. Before data normalization, LD areas per nuclei area ratios were log-transformed. A classical robust Z-score normalization was performed first [zi = (xi − medianj)/madj, where zi is the Z-score of well i; xi is the raw value of well i; and medianj and madj are the median and median absolute deviation (MAD) of the plate j] in addition to the recently proposed strictly standardized mean difference normalization [SSMDi = (xi − meanj)/square root (2/nj − 2.5 × ((nj − 1) × SDj2))]. Those related algorithms were supplemented with both a fitted linear model normalization using the Prada package [45] and by B-score normalization [46]. Benjamini and Hochberg FDR-corrected p-values for all dsRNAs were calculated with the complete screen data (without the largest and smallest 1%) as a reference set. Scoring was done both on a platewise and screenwise manner. For the platewise hit identification, positives were identified by a quartile-based thresholding algorithm [48]. For this purpose, the first quartile (Q1), the median (Q2), and the third quartile (Q3) were calculated first. Afterwards, threshold T were calculated [Tupper = Q3 + c × (Q3 − Q2) and Tlower = Q1 − c × (Q2 − Q1), where c is a variable depending on the targeted error rate] [48]. The same hit selection strategy was also chosen for the screen-wide hit identification among the linear model normalized dataset. For the other normalization algorithms, fixed thresholds were selected. In all cases, threshold levels (as well as the c in the quartile-based thresholding) were chosen based on the identification rates of the internal controls brummer dsRNA, midway dsRNA, and wells with no oleic acid, which were present on every screening plate. The highest possible threshold was chosen capable of balancing both false-positive and -negative rates.
Identified Drosophila lipid regulating gene functions (Tables S2 and S3) were subjected to in silico analysis for enriched GO terms. For this purpose, we used the standard settings of the VLAD tool (Mouse Genome Informatics Web site [51]) using the complete Drosophila genome as a reference set. Results of the enrichment analyses were visualized by pruned GO term networks (pruning threshold = 4; collapsing threshold = 5), and results (pruning threshold = 3; collapsing threshold = 6) are additionally tabulated (Table S5).
Detailed lists of the scoring genes were annotated with the following information (Table S9): GO terms from FlyBase [80]; orthologs from FlyMine [81]; human disease gene orthologs from Homophila (http://superfly.ucsd.edu/homophila/, used with a significance threshold of E < 1 × 10–50, [31]; InParanoid [82] orthologs (http://inparanoid.sbc.su.se/cgi-bin/index.cgi); and Drosophila [4,33]; as well as mammalian [5,6,32,34] lipid droplet subproteome data.
A subset of genes identified in the genome-wide screen with a potential function in cellular lipid storage regulation was assayed by at least one additional dsRNA. In total, 276 genes were tested by targeting with 362 dsRNA sequences (Table S10). Because we were interested in validating the full range of phenotypes observed and not just the positives, we sampled across a broad range of B-scores. We performed directed retesting on the genes encoding COPI members. To test for COPI specificity, we used secondary dsRNA sequences targeting Arf family members not involved in COPI function as well as COPII vesicle transport encoding transcripts as controls. dsRNAs targeting those genes did not result in a phenotype in the primary screen. For a “positive” identification, we required that two independent nonoverlapping dsRNAs or siRNAs give the same phenotype. In addition, we tested mouse AML12 hepatocytes and mouse 3T3-L1 fibroblasts for an evolutionary conservation of the identified lipid storage modulators. Assuming that off-target effects are random, this also minimizes misleading off-target effects, and is certainly more stringent than the current standard of two positive RNAi reagents with retesting in the same species and cell type [60]. In total, 127 mouse genes covered by 312 siRNAs were tested (Table S10). Genes across the screen that were validated using the image-based analysis with additional RNAi reagents are listed in Table S7. Additional gene and COPI validation comes from small compound phenocopy, cell staining experiments, and measurements on lipid metabolism as outlined further below.
Lipid droplet area and nuclei area measurements obtained from the image segmentation procedure, which was carried out as described for the primary screen results, was used to express the ratio of lipid per cell. For each screen, plate data were median normalized. In order to identify genes modulating lipid storage, a basic thresholding of median ± 2 × MAD was used. Since the datasets were enriched for modulators of lipid storage, the median as well as MAD was calculated on the basis of control wells incorporated in the assay plates. For the Drosophila, AML12, and 3T3-L1 datasets, those wells contained no RNAi reagent, but were otherwise treated identical to the experimental wells. Screening plates also contained other control dsRNAs/siRNAs wells. The Drosophila secondary screen plates contained wells with dsRNAs targeting bmm or mdy as in the primary screen. In the case of the 3T3-L1 and AML12 cells, plates contained siRNAs targeting Atgl or a combination of two siRNAs targeting both Adrp and Tip47 transcripts [68]. Median ± thresholds were adjusted in order to fulfill the same prerequisites as in the primary screen, namely a maximum of identified controls with a minimum of false positives. False positives were scored based on the wells lacking RNAi reagent.
Small-molecule compound experiments were performed with embryonic Drosophila S3 cells (Bloomington Drosophila Stock Center [DGRC]), which showed excellent oleic acid feeding characteristics during RNAi assay development but inferior RNAi characteristics as compared to the Kc167 cells. S3 cells showed superior adherence during automated liquid handling in 1,536-well format. We dispensed 4 μl of cells at 1.25 × 106 cells/ml into LoBase Aurora COC 1,536-well plates (black walled, clear bottom) with a bottle-valve solenoid-based dispenser (Aurora) to obtain 5,000 cells/well. A total of 23 nl of compound solution of different concentrations were transferred to the assay plates using a Kalypsis pin tool equipped with a 1,536-pin array containing 10-nl slotted pins (FP1S10, 0.457-mm diameter, 50.8 mm long; V&P Scientific). One microliter of oleic acid (400 μM) was added, and the plate was lidded with stainless steel rubber gasket-lined lids containing pinholes. After 18–24-h incubation at 24 °C and 95% humidity, BODIPY 493/503 (Molecular Probes) was added to the wells to stain lipid droplets, and the Cell Tracker Red CMTPC dye (Molecular Probes) was added to enumerate cell number. Fluorescence was detected by excitation of the fluorophores with a 488-nm laser on an Acumen Explorer (TTP Lab Tech). The total intensity in channel 1 (500–530 nm) reflected lipid droplet accumulation. Cells were detected using channel 3 (575–640 nm) with 5-μm width and 100-μm depth filters. The ratio of the total intensity in PMT channel 1 over total intensity of channel 3 was also calculated. Percent activity was computed relative to an internal control (100% inhibited lipid droplet deposition due to the presence of 20 μM Triacsin C), which was added to 32 wells/plate.
Measurements of NEFA released from lipid droplets or incorporated into the TG fraction were performed as previously described [23,68,83]. Briefly, AML12 cells treated with or without specific siRNAs (10 nM) for 4 d were incubated overnight with growth medium supplemented with 400 μM oleic acid complexed to 0.4% bovine serum albumin to promote triacylglycerol deposition and [3H] oleic acid, at 1 × 106 dpm/well, was included as a tracer. In lipolysis experiments, re-esterification of fatty acids in AML12 cells was prevented by including 10 μM Triacsin C (Biomol), an inhibitor of acyl coenzyme A synthetase [67], in the medium. Quadruplicate wells were tested for each condition. Lipolysis was determined by measuring radioactivity released into the media in 1 h. For the lipid extraction and thin layer chromatography, the cell monolayer was washed with ice-cold PBS and scraped into 1 ml of PBS. Lipids were extracted by the Bligh-Dyer method [84], and 10% of the total lipid was analyzed by thin layer chromatography [83,85]. AML12 cells treated with or without specific siRNAs were additionally incubated with either vehicle (DMSO), 5 μM of Exo1 (12.5 mg/ml DMSO), or BFA (10 mg/ml DMSO) during the time of radioactivity release into the media (2 h). NEFA incorporation into the TG fraction and NEFA release are calculated as nanomoles/milligram protein (Table S8). Protein measurements were performed using a commercial BCA assay kit (Pierce Biotechnology) according to the manufacturer's instructions. Statistical significance was tested by impaired Student t test (GraphPad software).
Rabbit anti-TIP47 and goat anti-ADRP were used as previously published [9]. Antibodies targeting mouse ATGL were purchased from Cell Signaling Technology. The CGI-58 antibody was a gift from Dr. Osumi [29].
Cells were plated in four-well Lab-Tek chamber slides (Nunc) and incubated overnight with 400 μM oleic acid. In compound experiments, wells received vehicle (DMSO) or 5 μM BFA (10 mg/ml DMSO) treatment for 6 h. RNAi treatment prior to immunocytochemistry is outlined above. For ADRP and TIP47 staining, cells were fixed in 3% v/v paraformaldhyde/PBS for 15 min at room temperature. Staining was performed by published methods [9,86]. Cells were viewed with a confocal laser scanning microscope (LSM510; Carl Zeiss MicroImaging) using a 63× oil objective lens.
Eight 100-mm dishes for each condition were treated with 400 μM oleic acid overnight and further treated with DMSO or BFA (5 μM) for 6 h on the next day. Cells were washed three times with phosphate buffered saline (PBS; pH 7.4), scraped into PBS, and then pelleted by low-speed centrifugation. LD isolation was as reported [8]. The lipid fat cake was isolated and resuspended in 150 μl of PBS containing 5% SDS before 150 μl of 2× Laemmli sample buffer were added. For CGI-58 and ATGL western blots, those protein extracts were directly loaded. For ADRP and Tip47, the samples were diluted 200-fold (ADRP) or 20-fold (TIP47), respectively. A total of 35 μl were loaded then on each lane. X-ray films were used to detect the western blots. Quantification was done with ImageJ [78].
Drosophila RNAi screen hits: FBgn0000028, FBgn0000042, FBgn0000114, FBgn0000339, FBgn0000489, FBgn0000547, FBgn0000567, FBgn0001186, FBgn0001204, FBgn0001301, FBgn0002878, FBgn0003048, FBgn0003118, FBgn0003339, FBgn0003380, FBgn0003392, FBgn0003462, FBgn0003557, FBgn0003607, FBgn0003691, FBgn0004167, FBgn0004187, FBgn0004401, FBgn0004587, FBgn0004595, FBgn0004611, FBgn0004652, FBgn0004797, FBgn0004838, FBgn0004856, FBgn0004879, FBgn0005411, FBgn0005626, FBgn0005630, FBgn0010083, FBgn0010215, FBgn0010355, FBgn0010638, FBgn0010750, FBgn0011571, FBgn0011701, FBgn0013746, FBgn0014020, FBgn0015320, FBgn0015818, FBgn0015919, FBgn0016926, FBgn0016940, FBgn0019643, FBgn0020611, FBgn0020908, FBgn0021768, FBgn0022246, FBgn0023143, FBgn0024285, FBgn0024308, FBgn0024555, FBgn0024754, FBgn0025638, FBgn0026206, FBgn0026317, FBgn0026620, FBgn0026722, FBgn0026878, FBgn0027495, FBgn0027589, FBgn0027885, FBgn0027951, FBgn0028360, FBgn0028420, FBgn0028982, FBgn0029123, FBgn0029526, FBgn0029661, FBgn0029731, FBgn0029766, FBgn0029824, FBgn0029850, FBgn0029873, FBgn0029935, FBgn0030075, FBgn0030077, FBgn0030087, FBgn0030093, FBgn0030189, FBgn0030244, FBgn0030390, FBgn0030434, FBgn0030492, FBgn0030608, FBgn0030872, FBgn0030904, FBgn0031008, FBgn0031030, FBgn0031031, FBgn0031074, FBgn0031093, FBgn0031232, FBgn0031390, FBgn0031518, FBgn0031626, FBgn0031673, FBgn0031816, FBgn0031836, FBgn0031888, FBgn0031894, FBgn0032049, FBgn0032340, FBgn0032351, FBgn0032360, FBgn0032363, FBgn0032388, FBgn0032454, FBgn0032622, FBgn0032800, FBgn0032868, FBgn0032945, FBgn0033155, FBgn0033160, FBgn0033541, FBgn0034071, FBgn0034402, FBgn0034646, FBgn0034709, FBgn0034839, FBgn0034946, FBgn0034967, FBgn0035085, FBgn0035136, FBgn0035294, FBgn0035546, FBgn0035569, FBgn0035631, FBgn0036274, FBgn0036374, FBgn0036470, FBgn0036556, FBgn0036734, FBgn0036761, FBgn0036811, FBgn0037024, FBgn0037149, FBgn0037178, FBgn0037250, FBgn0037278, FBgn0037304, FBgn0037568, FBgn0037920, FBgn0037924, FBgn0038168, FBgn0038191, FBgn0038343, FBgn0038359, FBgn0038391, FBgn0038592, FBgn0038633, FBgn0038662, FBgn0039054, FBgn0039941, FBgn0039959, FBgn0039997, FBgn0040279, FBgn0040291, FBgn0040369, FBgn0040534, FBgn0040651, FBgn0040777, FBgn0042693, FBgn0050126, FBgn0050470, FBgn0051313, FBgn0051374, FBgn0051632, FBgn0051814, FBgn0052056, FBgn0052062, FBgn0052112, FBgn0052121, FBgn0052150, FBgn0052202, FBgn0052352, FBgn0052397, FBgn0052440, FBgn0052635, FBgn0052704, FBgn0052710, FBgn0052711, FBgn0052970, FBgn0053207, FBgn0053500, FBgn0053516, FBgn0058413, FBgn0061200, FBgn0083976, FBgn0083992, FBgn0085381, FBgn0086441, FBgn0086674, FBgn0086899, FBgn0243486, FBgn0259162, FBgn0259169, FBgn0259171, FBgn0259217, FBgn0259228, FBgn0259240, FBgn0259243, FBgn0000008, FBgn0000100, FBgn0000116, FBgn0000212, FBgn0000409, FBgn0000492, FBgn0000636, FBgn0000986, FBgn0001133, FBgn0001216, FBgn0001217, FBgn0001218, FBgn0001942, FBgn0002023, FBgn0002590, FBgn0002593, FBgn0002607, FBgn0002906, FBgn0002921, FBgn0003031, FBgn0003060, FBgn0003209, FBgn0003277, FBgn0003279, FBgn0003360, FBgn0003600, FBgn0003687, FBgn0003701, FBgn0003941, FBgn0003942, FBgn0004110, FBgn0004922, FBgn0004926, FBgn0005593, FBgn0005614, FBgn0005630, FBgn0005648, FBgn0008635, FBgn0010078, FBgn0010220, FBgn0010348, FBgn0010352, FBgn0010391, FBgn0010409, FBgn0010410, FBgn0010412, FBgn0010431, FBgn0010612, FBgn0010808, FBgn0011211, FBgn0011272, FBgn0011284, FBgn0011701, FBgn0011726, FBgn0011745, FBgn0011837, FBgn0012034, FBgn0013275, FBgn0013276, FBgn0013277, FBgn0013278, FBgn0013279, FBgn0013325, FBgn0013981, FBgn0014020, FBgn0014857, FBgn0015024, FBgn0015288, FBgn0015393, FBgn0015756, FBgn0015774, FBgn0015778, FBgn0015834, FBgn0016120, FBgn0016694, FBgn0016926, FBgn0017397, FBgn0017545, FBgn0017566, FBgn0017579, FBgn0019624, FBgn0019886, FBgn0019936, FBgn0020129, FBgn0020386, FBgn0020439, FBgn0020910, FBgn0022343, FBgn0022935, FBgn0023170, FBgn0023171, FBgn0023213, FBgn0023531, FBgn0024150, FBgn0024330, FBgn0024733, FBgn0024939, FBgn0025286, FBgn0025582, FBgn0025724, FBgn0025725, FBgn0026262, FBgn0026666, FBgn0026741, FBgn0027321, FBgn0027348, FBgn0027615, FBgn0028530, FBgn0028867, FBgn0028968, FBgn0028969, FBgn0029088, FBgn0029161, FBgn0029504, FBgn0029761, FBgn0029799, FBgn0029822, FBgn0029860, FBgn0029897, FBgn0030025, FBgn0030088, FBgn0030174, FBgn0030259, FBgn0030341, FBgn0030384, FBgn0030386, FBgn0030606, FBgn0030610, FBgn0030669, FBgn0030692, FBgn0030696, FBgn0030726, FBgn0030915, FBgn0030951, FBgn0030990, FBgn0031300, FBgn0031392, FBgn0031545, FBgn0031696, FBgn0031771, FBgn0031842, FBgn0031980, FBgn0032053, FBgn0032215, FBgn0032261, FBgn0032330, FBgn0032400, FBgn0032518, FBgn0032587, FBgn0032596, FBgn0032619, FBgn0032656, FBgn0032675, FBgn0032833, FBgn0032987, FBgn0033029, FBgn0033081, FBgn0033085, FBgn0033282, FBgn0033313, FBgn0033341, FBgn0033368, FBgn0033379, FBgn0033403, FBgn0033591, FBgn0033652, FBgn0033699, FBgn0033902, FBgn0033912, FBgn0034020, FBgn0034258, FBgn0034487, FBgn0034488, FBgn0034537, FBgn0034579, FBgn0034649, FBgn0034751, FBgn0034902, FBgn0034948, FBgn0034968, FBgn0034987, FBgn0035276, FBgn0035315, FBgn0035422, FBgn0035562, FBgn0035563, FBgn0035638, FBgn0035699, FBgn0035753, FBgn0035872, FBgn0035976, FBgn0036135, FBgn0036213, FBgn0036288, FBgn0036343, FBgn0036351, FBgn0036360, FBgn0036398, FBgn0036449, FBgn0036462, FBgn0036492, FBgn0036532, FBgn0036534, FBgn0036576, FBgn0036613, FBgn0036728, FBgn0036820, FBgn0036825, FBgn0036895, FBgn0036990, FBgn0037010, FBgn0037028, FBgn0037093, FBgn0037097, FBgn0037098, FBgn0037102, FBgn0037207, FBgn0037249, FBgn0037270, FBgn0037356, FBgn0037415, FBgn0037429, FBgn0037529, FBgn0037546, FBgn0037559, FBgn0037566, FBgn0037610, FBgn0037637, FBgn0037752, FBgn0037813, FBgn0037912, FBgn0037942, FBgn0037955, FBgn0038049, FBgn0038074, FBgn0038131, FBgn0038281, FBgn0038345, FBgn0038538, FBgn0038628, FBgn0038629, FBgn0038734, FBgn0038760, FBgn0038881, FBgn0038996, FBgn0039205, FBgn0039214, FBgn0039302, FBgn0039359, FBgn0039402, FBgn0039404, FBgn0039464, FBgn0039520, FBgn0039580, FBgn0039857, FBgn0040007, FBgn0040010, FBgn0040233, FBgn0040512, FBgn0040529, FBgn0040634, FBgn0040766, FBgn0040793, FBgn0043001, FBgn0043904, FBgn0050007, FBgn0050290, FBgn0050387, FBgn0051158, FBgn0051284, FBgn0051291, FBgn0051302, FBgn0051354, FBgn0051361, FBgn0051450, FBgn0051453, FBgn0051554, FBgn0051613, FBgn0051754, FBgn0051774, FBgn0051847, FBgn0052050, FBgn0052105, FBgn0052179, FBgn0052193, FBgn0052219, FBgn0052311, FBgn0052600, FBgn0052633, FBgn0052720, FBgn0052733, FBgn0052773, FBgn0052778, FBgn0052797, FBgn0053128, FBgn0053147, FBgn0053256, FBgn0053271, FBgn0053300, FBgn0053319, FBgn0058337, FBgn0062412, FBgn0062413, FBgn0083950, FBgn0085392, FBgn0085408, FBgn0085424, FBgn0085436, FBgn0086710, FBgn0086712, FBgn0086758, FBgn0086904, FBgn0250791, FBgn0250814, FBgn0250834, FBgn0250908, FBgn0259113, FBgn0259212, FBgn0259232, and FBgn0259246.
Mouse genes with a confirmed function in lipid storage regulation: MGI:107807, MGI:107851, MGI:1333825, MGI:1334462, MGI:1335073, MGI:1351329, MGI:1353495, MGI:1354962, MGI:1858696, MGI:1861607, MGI:1891824, MGI:1891829, MGI:1913585, MGI:1914062, MGI:1914103, MGI:1914144, MGI:1914234, MGI:1914454, MGI:1915822, MGI:1916296, MGI:1917599, MGI:1929063, MGI:2385261, MGI:2385656, MGI:2387591, MGI:2388481, MGI:2443241, MGI:3041174, MGI:3694697, MGI:88192, MGI:95301, MGI:98342, and MGI:99431.
|
10.1371/journal.pgen.1001190 | Hormad1 Mutation Disrupts Synaptonemal Complex Formation, Recombination, and Chromosome Segregation in Mammalian Meiosis | Meiosis is unique to germ cells and essential for reproduction. During the first meiotic division, homologous chromosomes pair, recombine, and form chiasmata. The homologues connect via axial elements and numerous transverse filaments to form the synaptonemal complex. The synaptonemal complex is a critical component for chromosome pairing, segregation, and recombination. We previously identified a novel germ cell–specific HORMA domain encoding gene, Hormad1, a member of the synaptonemal complex and a mammalian counterpart to the yeast meiotic HORMA domain protein Hop1. Hormad1 is essential for mammalian gametogenesis as knockout male and female mice are infertile. Hormad1 deficient (Hormad1−/−) testes exhibit meiotic arrest in the early pachytene stage, and synaptonemal complexes cannot be visualized by electron microscopy. Hormad1 deficiency does not affect localization of other synaptonemal complex proteins, SYCP2 and SYCP3, but disrupts homologous chromosome pairing. Double stranded break formation and early recombination events are disrupted in Hormad1−/− testes and ovaries as shown by the drastic decrease in the γH2AX, DMC1, RAD51, and RPA foci. HORMAD1 co-localizes with γH2AX to the sex body during pachytene. BRCA1, ATR, and γH2AX co-localize to the sex body and participate in meiotic sex chromosome inactivation and transcriptional silencing. Hormad1 deficiency abolishes γH2AX, ATR, and BRCA1 localization to the sex chromosomes and causes transcriptional de-repression on the X chromosome. Unlike testes, Hormad1−/− ovaries have seemingly normal ovarian folliculogenesis after puberty. However, embryos generated from Hormad1−/− oocytes are hyper- and hypodiploid at the 2 cell and 8 cell stage, and they arrest at the blastocyst stage. HORMAD1 is therefore a critical component of the synaptonemal complex that affects synapsis, recombination, and meiotic sex chromosome inactivation and transcriptional silencing.
| The biology of germ cells is intimately intertwined with meiosis. Meiosis I is a unique biological event, when chromosomes pair, recombine, and segregate. The synaptonemal complex is a protein lattice that enables chromosome pairing and recombination and is unique to meiosis I. Meiosis I requires a subset of factors that are unique to germ cells and meiosis. Germ cell–specific factors are known to play crucial roles during formation of the synaptonemal complex and include synaptonemal complex proteins SYCP1, SYCP2, and SYCP3, among others. We discovered a mouse HORMA domain containing protein, Hormad1 (Nohma), which is germ cell–specific and essential for male and female fertility. Mice deficient in Hormad1 have severe defects in early recombination, synapsis, and segregation—functions attributed to yeast HORMA domain containing protein, Hop1. Moreover, Hormad1 is likely a germ cell–specific component of the meiotic sex chromosome inactivation and transcriptional silencing complex.
| Mammalian meiosis is unique to germ cells and a critical step in sexual reproduction. Meiosis reduces the chromosome complement to haploidy in preparation for fertilization. The first meiotic division is unique in pairing of homologous chromosomes, homologous recombination, and formation of chiasmata. The reduction in chromosome numbers happens when homologous chromosomes segregate to opposite poles during the first meiotic division. Proper disjunction (separation) requires crossovers (manifested cytologically as chiasmata). The sister chromatids organize along structures called axial elements (AEs) and transverse elements connect AEs to form the synaptonemal complex (SC) [1]. SC is a proteinaceous structure that connects paired homologous chromosomes during prophase I of meiosis, and SC is critical for wild-type levels of crossovers to occur during meiosis. AEs are critical part of the SCs and mutations in proteins that form AEs disrupt sister chromatid cohesion, recombination, and chromosome segregation [2]–[4]. Proteins with HORMA domain are critical components of the axial elements [5]. HORMA domain proteins are predicted to form globular structure that may sense specialized chromatin states, such as those associated with double strand breaks (DSBs) or other forms of DNA damage [5]. Several mammalian proteins that contain HORMA domain, such as mitotic arrest deficient protein 2, MAD2, are essential for mitosis [6]–[7]. Mice lacking MAD2 unsurprisingly die during early embryogenesis [7]. In lower organisms, several meiotic specific HORMA proteins are known and all are critical for meiosis. These HORMA proteins are: Hop1 [8] and Red1 [9] in yeast; Him-3 [10] in nematodes; and Asy1 [11] in plants. Him-3 localizes to the axial cores of both synapsed and unsynapsed chromosomes. C. elegans Him-3 mutants are deficient in chromosome pairing, synapsis, and the regulation of double strand break repair [10], [12]–[13]. Synapsis in both male and female Asy1 mutants is disrupted [14]–[15]. In yeast, plants, and nematodes, HORMA domain proteins are critical components of the synaptonemal complex and essential for meiosis I. Others and we identified a previously uncharacterized gene that we named Nohma, later re-named to Hormad1 [16]–[18]. Hormad1 encodes a protein that contains a HORMA domain, and unlike Mad2, Hormad1 expression is germ cell–specific [16]. Mouse and human HORMAD1 are highly conserved and share 77% amino acid identity overall, and share 89% amino acid identity in the HORMA domain. Moreover, mouse and human HORMAD1 HORMA domains share 28% amino acid identity with Hop1 HORMA domain. Hop1 in yeast appears to bind near or at the sites of DSB formation and may modulate the initial DSB cleavage [19]. Hop1 mutants in yeast have reduced number of DSBs [20], and Hop1 may participate in recruiting DMC1, RAD51 and other proteins that are required for DNA repair during meiotic synapsis and recombination [19]–[20]. Phosphorylation of Hop1 by Mec1/Tel1 yeast kinases is important for interhomologue recombination and prevents DMC1-independent repair of meiotic DSBs [21]. Here we report that HORMAD1 is likely the mammalian counterpart of Hop1, and that HORMAD1 deficiency disrupts mammalian synaptonemal complex formation, meiotic recombination, and chromosome segregation.
We previously showed that Hormad1 RNA expression in testes began at postnatal day 10, with little expression detected at birth or postnatal day 5 [16]. Hormad1 RNA expression pattern coincided with the onset of meiosis, and appearance of primary spermatocytes in the developing testes. In situ hybridization with anti-sense Hormad1 riboprobe revealed that Hormad1 expression was confined to germ cells, and specifically spermatocytes, with no signal detected in spermatogonia or sertoli cells [16]. We generated antibodies against HORMAD1 and studied its protein localization pattern in testes. HORMAD1 localized exclusively in germ cells, specifically in zygotene, and early pachytene spermatocytes as previously described for the RNA expression [16].
Since HORMAD1 protein showed localization consistent with its potential role in meiosis I and contains the HORMA domain, we disrupted the Hormad1 gene to examine its requirement for germ cell development and meiosis in mouse. Hormad1 is located on chromosome 3 and composed of sixteen exons. We deleted exons 4 and 5 (Figure S1A), and this mutation is predicted to remove 33 amino acids from the highly conserved HORMA domain and to cause a frame shift mutation. Small amounts of truncated Hormad1 RNA transcripts were detectable on RT-PCR, and Western blots on testes extracts showed absence of HORMAD1 protein in knockout mice as expected (Figure S1B and S1C).
Female and male heterozygote matings produced expected Mendelian ratios, averaged 8.1±2 pups per litter (n = 20 breeding pairs) over a 6-month period, and remained fertile for at least 9 months. The litter size was statistically not significantly different from the wild-type average (8.4±2 pups per litter). Male and female mice heterozygous for the mutation (Hormad1+/−) were fertile with grossly normal male and female gonadal morphology and histology. However, both Hormad1−/− males and females were infertile with no pups produced over a period of 6 months from mating with wild-type female and male mice, respectively.
While ovaries showed no gross morphologic differences between the knockout and wild-type mice, Hormad1 knockout adult testes were significantly smaller than the wild-type testes (Figure S2A). Testes in the 7-day-old Hormad1−/− mice were grossly normal and weighed 8.0±2.0 mg/pair and did not significantly differ from the wild-type, 9.0±3.0 mg/pair of testes. By 4 weeks of postnatal life, the knockout testes (47±6 mg/pair) were 50% of the wild-type weight (94±1.7 mg/pair), and by 8 weeks the knockout testes (60±7.7 mg/pair) were 27% of the wild-type weight (225±2.7 mg/pair) (Figure S2B).
Histology at 6 weeks showed hypocellular seminiferous tubules with clumps of sertoli cells in the lumen. We observed spermatogonia and early spermatocytes, but no post-meiotic germ cells such as spermatids or spermatozoa (Figure 1A–1H). We therefore carefully examined spermatogenesis in Hormad1−/− mice. Spermatogenesis is a complex process that involves differentiating and proliferating self-renewing spermatogonia that differentiate into spermatozoa. Type A spermatogonia self-renew and can initiate differentiation into Type B spermatogonia which in turn differentiate into primary spermatocytes. Primary spermatocytes undergo meiosis I to form secondary spermatocytes. Secondary spermatocytes enter meiosis II and divide to produce haploid spermatids. We examined Hormad1 knockout testes histology during gonadal development to determine the stage at which spermatogenesis is disrupted. Identical testes weights at postnatal day 7, and similar histology between Hormad1−/− and wild-type testes argue that pre-spermatogonia in Hormad1−/− testes proliferate into Type A spermatogonia without major disruption. Immunohistochemistry with antibodies directed against PLZF and SOHLH1, markers that identify self-renewing (PLZF) and differentiating spermatogonia (SOHLH1), showed the presence of both proteins in the wild-type as well as the knockout animals, confirming that spermatogonia are unaffected (Figure 1J, 1K, 1N, 1O). At postnatal day 10, testes contain preleptotene/leptotene primary spermatocytes, and there was no gross difference between wild-type and Hormad1−/− testes. At 14 days, testes contain pachytene spermatocytes, and there was a significant difference between the wild-type and Hormad1−/− testes, with many apoptotic cells and few pachytene spermatocytes in Hormad1−/− testes (), and rising apoptotic index with age in the knockout as compared to the wild-type (Figure S3K). We counted leptotene, zygotene and pachytene spermatocytes in 6 week old wild-type and Hormad1−/− testes. Hormad1−/− testes showed declining number of spermatocytes beginning in stages II-III with 28±8 spermatocytes as compared to 52±12 in the wild-type (Figure 1A and 1E). No spermatocytes were noted in stages IV-IX in the Hormad1−/− testes (Figure 1F and 1G), and no significant difference was noted in the spermatocyte number in stages X-XII between the wild-type and knockout testes (Figure 1D and 1H). These results indicate that Hormad1 deficiency in the male gonad caused meiotic arrest at the pachytene stage.
Previous studies on HORMA domain proteins indicate their specific involvement in cell division. MAD2 is a ubiquitously expressed mammalian HORMA domain protein involved in both meiosis and mitosis [7], while yeast HOP1, RED1, nematode HIM3 and plant ASY1 genes are specifically involved in meiotic segregation, synapsis and recombination [2], [10]–[11], [13], [15], [22]. No mammalian counterparts to Hop1, Red1, Him3 and Asy1 have been functionally evaluated up to date. We previously hypothesized that HORMAD1 is a functional counterpart to Hop1, Him3 and Asy1 [16]. Critical components of the synaptonemal complex include meiosis specific SYCP1, SYCP2 and SYCP3 proteins. SYCP1 is a major component of the transverse filaments, while both SYCP2 and SYCP3 are components of the axial lateral elements [23]–[26]. To determine HORMAD1 localization during meiosis, and whether HORMAD1 localizes to the axial elements, or transverse filaments, we used antibodies against SYCP1, SYCP2 and SYCP3 to study their respective co-localization with HORMAD1. HORMAD1 co-localized with SYCP3 and SYCP2 but did not co-localize with SYCP1, which indicates that HORMAD1 is located along the axial elements (Figure 2A, 2C, and 2E). Recent studies also show that HORMAD1 localizes to the axial elements [17]–[18]. We also studied whether absence of SYCP2 affected HORMAD1 localization along the chromosomes. HORMAD1 localization is independent of major germ cell–specific components of the axial elements of the synaptonemal complex because neither Sycp2 (Figure 3A–3F) nor Sycp3 mutation affected HORMAD1 localization to the axial elements [18].
We examined localization of germ cell–specific synaptonemal complex proteins SYCP1, SYCP2 and SYCP3 in Hormad1−/− testes. SYCP1 is a major component of the transverse filaments and is known to form fibrillar structures in Sycp3 mutant spermatocytes [27]. However, the fibers are truncated, contain axial gaps, and do not associate with the centromeres of the meiotic chromosomes. We also observed truncated fibers with anti-SYCP1 antibodies in Hormad1−/− mice (Figure 2B). HORMAD1 is therefore not necessary for SYCP1 binding to the chromatin. We also examined localization of SYCP2 and SYCP3 proteins in Hormad1−/− mice. The localization of SYCP2 and SYCP3 to the chromatin was not significantly affected by the lack of HORMAD1 (Figure 2D and 2F). These results indicate that HORMAD1 is not necessary for SYCP2 and SYCP3 localization to the axial elements.
The deficiency in synaptonemal complex proteins such as SYCP3 is known to affect chromosome synapsis [27]. In order to determine the effect of HORMAD1 deficiency on chromosome synapsis during meiosis I, we utilized CREST sera. CREST sera labels centromeres and allows the determination of the pairing status during meiosis [28]. In the wild type spermatocytes, prior to the synapsis, 40 centromeres are usually observed in the leptotene stage. The number of visible centromeres become reduced as the synapsis of homologues progresses. At the completion of the synapsis in pachytene, 20 centromeric foci are usually observed corresponding to 19 autosomal homologues and partially paired X-Y chromosomes. We examined CREST foci formation in Hormad1−/− spermatocytes. Examination of over 100 Hormad1−/− spermatocytes and oocytes in meiosis I, revealed greater than 20 centromeric foci in both male and female germ cells, most containing 40 CREST foci (Figure 4A, 4D, and data not shown). These results indicate that Hormad1 deficient germ cells cannot complete homologous chromosome pairing, and Hormad1 is therefore critical for chromosome synapsis during meiosis.
Our experimental evidence strongly suggests that HORMAD1 localizes to the axial core and is yet another critical component of the synaptonemal complex. To determine the effect of Hormad1 deficiency on the structure of the synaptonemal complex, we visualized synaptonemal complexes during meiosis I in wild-type and Hormad1−/− spermatocytes using electron microscopy. In the wild-type, synaptonemal complexes were well visualized during the pachytene stage of meiosis (Figure 4B and 4C). Electron microscopy examination of one hundred and ten Hormad1−/− spermatocytes from three independent experiments revealed the lack of the typical tripartite synaptonemal complex structure (Figure 4E and 4F). Persistence of pre-synaptic number of centromeric foci (CREST staining) in Hormad1−/− spermatocytes, as well as non-visualization of the tripartite synaptonemal complex structure by electron microscopy, demonstrate that HORMAD1 is essential for chromosomal synapsis.
Previous studies have indicated that Sycp3 deficiency has subtle effects on meiotic recombination [27]. Early recombination events do not seem to be disrupted in Sycp3, as similar number of DMC1 foci are present in Sycp3 mutant and wild-type meiosis [29]. DMC1 is a meiotic specific recombinase that together with ubiquitously expressed RAD51 catalyzes homologous pairing and DNA strand exchange [30]–[31]. These early steps in recombination are critical for establishing the physical connections between homologous chromosomes during meiosis. Hop1 has been implicated in modulating the formation and processing of double stranded breaks [19]. We examined formation of DMC1, RAD51, and RPA foci in zygotene stage Hormad1−/− spermatocytes. There is a dramatic decrease in the number of DMC1 foci as compared to the wild-type (Figure 5A and 5B). We counted a total of 98.9±28.2 DMC1 foci in the wild-type spermatocytes (n = 50) and 9.28±3.9 DMC1 foci in the Hormad1−/− spermatocytes (n = 45). The number of RAD51 foci was also decreased from 189.3±31.8 in the wild-type spermatocytes (n = 50), to 69.3±34.5 in the Hormad1−/− spermatocytes (n = 40) (Figure 5C and 5D).
Following double-strand breaks formation by SPO11, RPA is recruited together with RAD51 to the single stranded DNA regions [32]. We counted a total of 194.5±64.5 RPA foci in the wild-type spermatocytes (n = 30) as compared to 70.1±38.5 foci in the Hormad1−/− spermatocytes (n = 20) (Figure 5E and 5F). These results indicate that early meiotic recombinational events were disrupted and not surprisingly, MLH1, a protein that forms foci in later stages of recombination and required for the formation of most of the crossovers (chiasmata) observed in mice [33], was dramatically reduced in Hormad1−/− spermatocytes (data not shown).
We also examined DMC1, RAD51 and RPA foci formation in female meiocytes at embryonic day 15.5 (E15.5). Embryonic ovaries contain zygotene to early pachytene oocytes at E15.5 [34]. We counted a total of 208.7±117.1 DMC1 foci in the wild-type E15.5 oocytes (n = 50), and 79.1±81.5 foci in the Hormad1−/− oocytes (n = 30) (Figure 6A and 6B), a total of 197.9±46.0 RAD51 foci in the wild-type oocytes (n = 50) versus 85.1±37.6 foci in the Hormad1−/− oocytes (n = 40) (Figure 6C and 6D) and a total of 317.16±135.3 RPA in the wild-type oocytes (n = 50) and 51.7±48.8 in the Hormad1−/− oocytes (n = 50) (Figure 6E and 6F). DMC1, RAD51 and RPA foci are therefore, similar to our observations in spermatocytes, significantly decreased in Hormad1 deficient female meiocytes.
These results indicate that homologous recombination is significantly affected in Hormad1−/− mammalian germ cells, as previously reported for HOP1 [35]. We also observed effects of Hormad1 deficiency on γH2AX staining (a phosphorylated form of histone H2AX), a well known surrogate marker for DSB formation [36]. In the leptotene stage, phosphorylation of H2AX is induced by SPO11 catalyzed DSBs in meiotic DNA, and γH2AX appears as large, cloud-like patterns thatdisappearat the pachytene stage [37]. At the leptotene stage, γH2AX staining in Hormad1−/− spermatocytes was significantly decreased (76% decrease in signal intensity) as compared to the wild-type (Figure 7A–7D). γH2AX staining was also significantly decreased in Hormad1−/− fetal oocytes (71% decrease in signal intensity) (Figure 6G and 6H). These results suggest that similar to Hop1 mutants, DSBs do not efficiently form in Hormad1 mutants.
HORMAD1 protein localization along autosomes was previously shown to be transient [17] (Figure 7B and 7F). HORMAD1 staining was highest along unsynapsed chromosome axis in the zygotene to pachytene stage and diminished significantly along autosomes in the pachytene [17]–[18] (Figure 7F and 7I). Interestingly, HORMAD1 localized strongly along desynapsed autosomes in diplotene meiocytes [17]. During the pachytene stage, HORMAD1 is faintly visible along the autosomes, but co-localizes strongly with γH2AX on the XY chromosomes, and specifically along the axial elements [17]–[18]. γH2AX is a phosphorylated form of histone H2AX, and a marker for DSBs [36], [38]. H2AX is phosphorylated throughout the chromatin in leptotene spermatocytes and by the pachytene stage [17]–[18], γH2AX staining is undetectable on autosomes and restricted to the sex body (Figure 7E and 7F) [39]. H2ax knockout shows the essential role for H2AX in sex body formation and meiotic sex chromosome inactivation [39]. Meiotic sex chromosome inactivation also involves ATR and BRCA1 dependent phosphorylation of H2AX [18], [40]. Interestingly, Hormad1 deficiency, similar to H2ax deficiency, abolishes the formation of the sex body (Figure 7G and 7H, and data not shown). The lack of sex body formation is most likely due to the disruption of Hormad1−/− spermatocytes prior to the pachytene. We examined the γH2AX localization in the wild-type and Hormad1−/− spermatocytes. Chromatin in Hormad1−/− spermatocytes stained with anti-γH2AXantibodies, but no preferential localization to the sex chromosomes was observed (Figure 7G and 7H). BRCA1 and HORMAD1 have recently been shown to co-localize in the sex body [18]. We determined whether BRCA1 localization is dependent on HORMAD1. BRCA1 protein could not be localized in Hormad1−/− spermatocytes (Figure 7I–7J, 7M). This finding is unlike H2ax knockout, where BRCA1 is still detected on the sex chromosomes despite the lack of the sex body [40]. We also examined ATR localization in wild-type and Hormad1−/− testes. In wild-type testes, HORMAD1 and ATR co-localized in the sex body, but we could not detect ATR in Hormad1−/− spermatocytes (Figure 7K–7L, 7N). Above data suggest that HORMAD1 may be involved in the recruitment of BRCA1, ATR and γH2AX to the sex chromosome.
Since ATR, BRCA1 and γH2AX are involved in the transcriptional silencing of sex chromosomes, we examined whether Hormad1 deficiency affects transcriptional repression. Previous studies have shown that H2ax and Brca1 deficiencies individually, lead to the over-expression of genes exclusively expressed from the X or Y chromosome [40]. We examined whether X-linked germ cell–specific genes were over-expressed in Hormad1−/− testes compared to wild-type testes. We performed quantitative real-time PCR on select autosomal genes (Hormad2, Rnh2 and Mov10l1) as well as X-chromosome derived genes (Usp26, Fthl17, Pramel3, Tex11, and Tex13) on wild-type and Hormad1−/− testes (Figure 8A–8H). Rnh2 and Mov10l1 are germ cell–specific transcripts, derived from autosomes, and were not differentially expressed between wild-type and Hormad1−/− testes (Figure 8B and 8C). In contrast, all of the germ cell–specific transcripts transcribed from the X chromosome were significantly elevated in Hormad1−/− testes over the wild-type. These include Usp26 (4.5 fold increase), Fthl17 (6.5 fold increase), Pramel3 (3.5 fold increase), Tex11 (2.2 fold increase), and Tex13 (2.8 fold increase) (Figure 8D–8H). Moreover, RNA expression microarray analyses comparing two week old Hormad1 deficient testes with corresponding wild-type, indicate that almost 20% of the up-regulated genes derive from the X chromosome (Figure 8I). Our results are remarkably similar to transcriptional de-repression observed in H2ax and Brca1 mutants [39]–[40], and indicate that HORMAD1 is a germ cell and meiosis specific factor critical in meiotic sex chromosome inactivation and transcriptional silencing.
HORMAD1 is a likely mammalian counterpart to the yeast HORMA domain meiotic protein, Hop1. Hop1 is phosphorylated by Mec1/Tel1, the budding yeast homologue to the mammalian ATR and ATM kinases, and this phosphorylation is thought to play an important role in inter-homologous recombination [21]. We therefore examined HORMAD1 expression in Atm deficient mice as well as ATM expression in Hormad1−/− animals. ATM is a serine/threonine-specific protein kinase that has been associated with cell cycle regulation, apoptosis, and response to DNA damage repair. ATM kinase activation is associated with increased auto-phosphorylation of ATM at multiple sites including serine 1981 [41]. We examined HORMAD1 protein expression in testes between postnatal days 5–21 (Figure 9A). Eight to ten day old testes contain spermatogonial cells as well as preleptotene/leptotene spermatocytes [34]. At postnatal day 14–18, testes contain pachytene spermatocytes and visible sex bodies [34]. HORMAD1 is known to be phosphorylated [17] (Figure 9B). Western blot analysis on testes extracts detected phospho-HORMAD1 beginning at postnatal day 14 (Figure 9A). Phospho-HORMAD1 protein was decreased in postnatal day 18 and 21 wild-type testes (Figure 9A). Phospho-HORMAD1 appearance correlates temporally with sex body formation. HORMAD1 phosphorylation was not affected by Atm deficiency (Figure 9A). These results indicate that ATM is not responsible for HORMAD1 phosphorylation. We also examined HORMAD1 localization in Atm−/− spermatocytes. We observed HORMAD1 localization to chromosomal axes in Atm deficient spermatocytes, as previously described by others [17] (Figure 9C and 9D). Since synaptonemal complexes do not form in Atm mutants, HORMAD1 association with unsynapsed chromosomes does not require ATM. The anti-ATM phospho-S1981 antibody did not detect phosphorylated ATM in Hormad1−/− testes (Figure 9E and 9F). These results suggest that HORMAD1 is upstream of ATM auto-phosphorylation, and therefore likely upstream of ATM kinase activation.
We have previously shown that Hormad1 RNA expression in the ovary was confined to the germ cell [16]. Meiosis I in the female gonad commences circa E13.5 and most oocytes arrest at the dictyate stage by the time of birth. Antibodies against HORMAD1 recognized HORMAD1 protein at E14.5 (leptotene) and E18.5 (arrest in diplotene) oocytes (Figure 10A and 10B), but little HORMAD1 protein was detected in the newborn ovary oocytes (Figure 10C), at the time when oocytes are arrested in diplotene. Deficiency in genes critical in meiosis can disrupt early ovarian development, as is the case for Dmc1, Msh5, Spo11 and Atm [42]–[44].
We therefore examined ovarian development in Hormad1−/− females. Antibodies against germ cell–specific transcriptional regulator SOHLH1 [45] stained wild-type and knockout oocytes throughout embryonic gonadal development with no significant differences noted (Figure 10A–10F). Moreover, the numbers of primordial, primary and secondary follicle counts did not significantly differ between the mutant and wild-type ovaries at post-natal day 8 (Figure S4). These results indicate that Hormad1 deficiency does not affect embryonic ovarian development, germ cell cyst breakdown, and primordial follicle formation. We also examined the histology of wild-type and Hormad1−/− ovaries between 2 and 30 weeks of life. Mice reach sexual maturity around 6 weeks of life, and mouse ovaries at this time consist of all of the follicular types including corpora lutea, an indication that the ovaries are ovulating. Postnatal Hormad1−/− ovaries were grossly indistinguishable from wild-type mice between 2 and 30 weeks of life, with abundant corpora lutea in the Hormad1−/− ovaries indicating that the normal process of oocyte maturation was not disrupted, and ovulation has occurred (Figure 11A–11H). We induced superovulation in knockout and wild-type mice with exogenous gonadotropins to determine whether subtler ovarian defects contributed to infertility in Hormad1−/− mice. Hormad1−/− females super-ovulated 28±11 eggs (n = 22), while wild-type animals superovulated 29±14 eggs (n = 13). We therefore did not observe significant difference between the number of eggs superovulated from wild-type versus knockout mice. These results indicate that ovarian development is grossly normal in Hormad1−/− mice, and that ovarian defects are unlikely to account for observed infertility.
We studied early embryonic development of fertilized Hormad1−/− oocytes because ovarian defects were unlikely to explain observed infertility. We recovered embryos from wild-type male matings with Hormad1−/− females at E0.5, E1.5, E2.5, and E3.5. Comparable numbers of morphologically indistinct 1-cell zygotes were recovered from oviducts of control and mutant female mice at E0.5 (Figure 12A–12C), and little difference was noted at the 2-cell stage, except for an increased number of 1-cell embryos in the knockouts, indicating a lag in the progression from the 1-cell to 2- cell stage (Figure 12D–12F). By E2.5, the number of normal appearing 8-cell stage embryos was significantly less in Hormad1−/− fertilized eggs as opposed to the wild-type (Figure 12G–12I), and no morphologically normal blastocysts were observed in the Hormad1−/− fertilized eggs at E3.5. It is interesting to note that at E3.5, a significant number of 4 and 8 cell stage embryos were observed in the knockout while only blastocysts were observed in the wild-type E3.5 embryos (Figure 12J–12L). We also tested, using Chicago Sky Blue 6B dye (Sigma, MO, USA) injection into the tail vein [46], whether blastocysts derived from Hormad1−/− fertilized eggs could implant. We did not detect implantation of Hormad1−/− fertilized eggs (Figure 12M). These results indicate that a defect in early embryogenesis led to premature loss of embryos.
We hypothesized that aneuploidy is the major cause of embryo wastage in Hormad1−/− females. Unlike meiosis in testes, errors during oocyte meiosis have milder effect on oocyte loss and apoptosis, as has been observed in Sycp3 and Smc1β knockouts [27], [29], [47]. The presence of growing oocytes in the Hormad1−/− ovaries, as opposed to early germ cell loss observed in testes during pachytene stage of meiosis, indicates greater tolerance of Hormad1 deficiency in the oocytes as compared to spermatocytes. The whole range of ovarian follicle types are present in the Hormad1−/− ovaries, including primordial follicles, primary, secondary and antral follicles. In contrast to other critical genes during meiosis, such as Dmc1, Msh5, Spo11 and Atm that cause early ovarian failure due to rapid loss of oocytes following disruption of meiosis I [42]–[44], Hormad1 deficiency does not activate apoptotic pathways and does not lead to gross premature loss of oocytes. We used chromosome specific mouse BAC (bacterial artificial chromosome) clones to perform fluorescent in situ hybridization (FISH) on germinal vesicle (GV) oocytes, 2 cell and 8 cell stage embryos to determine whether aneuploidy is significantly higher in Hormad1−/− embryos. Wild-type GV oocytes are arrested in meiosis I and contain bivalent chromosomes that consist of four chromatids. Examination of 58 GV oocytes from three independent knockout animals and 112 GV oocytes from three wild-type animals with BAC probes specific for chromosomes 19, 18 and X, revealed no significant differences, with presence of four chromatids for each of the chromosome examined as expected (Figure 13A–13C). Following fertilization and completion of meiosis II, each cell in the developing embryo should contain two chromosomes except for the sex chromosomes. We examined by FISH, mouse chromosomes 6 and X in the 2-cell wild-type and Hormad1−/− embryos. Sixty three cells examined for chromosome 6 in the Hormad1−/− 2-cell embryos revealed that 22% of the cells had 4 signals corresponding to chromosome 6, 35% had 3 signals corresponding to chromosome 6, 24% had 1 signal, and 5% had no signal. Wild-type 2-cell embryos were also examined by chromosome specific FISH, and 44 cells examined out of 46 showed only 2 signals, as expected. In 2-cell Hormad1−/− embryos, out of sixty three cells examined for chromosome X, 44% showed four, three or no signal. Wild-type 2-cell stage embryos, as expected, showed only 2 or 1 signal, consistent with either XX or XY sex of the cell. These results indicate widespread hypo and hyperdiploidy in 2-cell embryos (Figure 13D–13F).
We also examined cells from the 8-cell stage embryos. A total of 84 cells from the Hormad1−/− embryos were examined with FISH probes specific for chromosomes 11, X and 2. Hormad1−/− cells from the 8-cell stage, hybridized with BAC specific to chromosome 11, showed 4 signals in 10% of the cells, and 3 signals in 55% of the cells. Therefore, a total of 65% of the cells were either monosomic or trisomic for chromosome 11. In the wild-type 8 cell stage embryo, a total of 96 cells were scored for chromosome 11, and only 2 out of 96 (2%) cells showed a single signal, while 94 out of 96 cells (98%) showed two signals as expected. Similar results were obtained for chromosome 2 (Figure 13G–13I). Our results indicate widespread aneuploidy involving different chromosomes in both 2-cell and 8-cell embryos. Such aneuploidy causes early embryo demise and failure of implantation.
We also visualized second metaphase (M2) spindle using anti-β-tubulin antibodies in Hormad1−/− oocytes. During M2, sister chromatids migrate to the opposite pole to form functional, haploid gametes. Sister chromatids are bi-oriented on the spindle and congregate in the center of the spindle prior to separation (Figure 13J). Each sister kinetochore in a pair is attached to the opposite spindle, and such arrangement generates sufficient tension to cause proper segregation. M2 in Hormad1−/− oocytes is grossly disrupted with mis-orientation of chromatid attachment to the spindle observed in all M2 oocytes examined (Figure 13K) (n = 50). These results indicate importance of Hormad1 in proper spindle formation and segregation.
We discovered Hormad1 (Nohma), using in silico screen to identify germ cell–specific transcripts critical for gonadal development [16]. We showed that Hormad1 expression was confined to germ cells, and that Hormad1 transcripts were mostly concentrated in the pachytene spermatocytes and early oocytes. Here we show that Hormad1 is essential for both male and female fertility in the mouse. HORMA domain-containing proteins interact with chromatin, particularly chromatin associated with DNA adducts, and are critical mitotic spindle checkpoint proteins [5]. HORMA domain-containing proteins are critical regulatory proteins for mitosis as exemplified by MAD2 protein, and meiosis as exemplified by Hop1p [8], and Red1p [9] in yeast, Him-3 [10] in nematodes, and Asy1 [11] in plants. The 205-aa HORMA domain in the mouse HORMAD1, shares highest similarity to the Saccharomyces cerevisiae Hop1 protein (28% amino acid similarity). Yeast Hop1 mutants have defects in chromosome condensation, synapsis and recombination [4], [8], and Hop1p binds DSBs during meiotic prophase and appears to play an important role in interhomologue recombination. We previously hypothesized, based on homology to Hop1 in yeast, that Hormad1 plays an important role in mammalian meiosis. Our results indicate that HORMAD1 is the mammalian homologue to the Hop1 protein in yeast.
Our current studies, and those of others, have shown that HORMAD1 protein is confined to germ cells and co-localizes with synaptonemal complex proteins SYCP2 and SYCP3 on the chromatin as part of the axial core [27], [48]. Synaptonemal complex is a complex structure composed of multiple germ cell–specific and ubiquitously expressed proteins that connect paired homologous chromosomes. Similar to railroad tracks, the synaptonemal complex axial lateral elements (SYCP2, SYCP3) are connected to each other by proteins known as transverse filaments (SYCP1, TEX12) [1]. The axial/lateral elements play critical roles in chromosome condensation, pairing, and repress recombination pathways that involve sister chromatids [1]. Synaptonemal complex formation is associated with HORMAD1 depletion from the axes [17]–[18]. Our data show that HORMAD1 phosphorylation peaks at the time when HORMAD1 localization shifts to the sex body. ATM and ATR do not appear to be involved in HORMAD1 phosphorylation and binding to chromosomal axes. Previous study showed that Trip13, a homologue of yeast Pch2 kinase, may be involved in HORMAD1 depletion from synapsed chromosome axes [17]. Kinases, such as Mec1/Tel1 in yeast are important to effect inter-homologous recombination [21] by phosphorylating Hop1, but not much is known regarding HORMAD1. Future studies are necessary to understand HORMAD1 phosphorylation sites and responsible kinases.
Although HORMAD1 is not essential for the binding of well-characterized germ cell–specific synaptonemal complex proteins such as SYCP1, SYCP2 and SYCP3, HORMAD1 deficient spermatocytes are defective in synapsis and do not form recognizable synaptonemal complexes. Hormad1 deficiency causes a male meiotic arrest, that is similar to male meiotic arrests observed in other components of the synaptonemal complex and recombination including: Atm, Spo11, Sycp1, Sycp2, Sycp3 and Dmc1 [27], [30]–[31], [48]–[50]. These findings are not surprising as HORMAD1 co-immunoprecipitates with known axial proteins, SYCP2, SYCP3, REC8, and SMC1β [18]. Whether these interactions are biologically significant, and whether HORMAD1 interacts with other members of the synaptonemal complex, including newly discovered HORMAD 2 protein, is unclear. Interestingly, Hop1 interacts with another phosphoprotein, Red1 [35], [51], and whether HORMAD2 is the mammalian counterpart of Red1 remains to be seen.
Like its counterpart in yeast, Hop1, Hormad1 deficiency disrupts early and later stages of recombination as shown by the drastic diminution in the formation of DMC1, RAD51, RPA and MLH1 foci. These findings differ from the Sycp3 mutant, which affects later stages of recombination [29]. DMC1 is a meiotic specific recombinase that together with ubiquitously expressed RAD51, catalyzes homologous pairing and DNA strand exchange, and marks earlier stages of recombination than MLH1 foci [30]. HORMAD1 expression is not significantly affected in Dmc1 mutant spermatocytes [17]. These results indicate that unrepaired DSBs and synaptonemal complexes between non-homologous chromosomes seen in Dmc1 mutants, do not affect HORMAD1 localization. In Hop1 mutants, DMC1 foci are only faintly detected when compared to controls [19]. These findings have led to a suggestion that Hop1 binds at or near the sites of DNA double strand break and modulates the action and perhaps recruitment of recombinases such as DMC1 and RAD51 [19]. Moreover, Hop1 appears to be part of the meiosis specific surveillance system that monitors meiotic double stranded break repair [21]. Our observations support the possibility that HORMAD1 affects early recombination and may therefore perform similar functions in the mammalian meiosis.
During mammalian meiosis, X and Y chromosomes are inactivated and transcriptionally silenced (meiotic sex chromosome inactivation, MSCI) and chromatin condenses to form a sex (XY) body. The epigenetics of MSCI involves γH2AX, BRCA1 and ATR at a minimum. The most intriguing finding regarding HORMAD1 is its co-localization with BRCA1, ATR and γH2AX in the sex chromosomes during pachytene [18]. Hormad1 deficiency, similar to H2ax deficiency, abrogates formation of the sex body, as well as γH2AX localization to the sex chromosomes. Hormad1 deficiency also abrogates BRCA1 and ATR localization to the sex chromosomes. Interestingly, H2ax deficiency does not affect BRCA1 localization to the sex chromosome, despite anomalous sex body, while BRCA1 deficiency does affect both ATR and γH2AX localization to the sex body. These findings indicate that HORMAD1 is involved in the initial recruitment of BRCA1, ATR and γH2AX to the sex body. The mechanism whereby HORMAD1 affects ATR, γH2AX and BRCA1 localization is unclear, but our results are consistent with the conclusion that HORMAD1 is upstream of ATR, γH2AX and BRCA1, and establishes HORMAD1 as an essential component of the MSCI complex. Moreover, we have shown here that Hormad1 deficient testes preferentially over-express X-linked genes as observed in mice deficient for other components of the MSCI complex such as γH2AX and BRCA1 [17]. Almost 20% of genes preferentially up-regulated from the Hormad1−/− testes are derived from the X chromosome. Hormad1 is therefore the first germ cell–specific component known to play a role in MSCI.
Unlike males, meiosis I in females arrests during the embryonic development at the diplotene stage and is completed upon ovulation. The lack of dramatic germ cell apoptosis during ovarian follicle development has also been observed for Sycp2 and Sycp3 deficient mice [48], [52], and reiterates the laxity of oocytes to early meiotic errors, as opposed to spermatogenesis, where meiotic check point causes the dramatic phenotype observed in male gonads. Hormad1 deficiency does not affect folliculogenesis nor ovulation. These findings are unusual, given that Hormad1 affects recombination, and recombination associated proteins DMC1 and MLH1 are important in early oogenesis. Unrepaired recombination intermediates or defects in homologous chromosome pairing and synapsis are likely causes of early oocyte loss [53]. Spo11 deficient animals can ameliorate observed oocyte losses in recombination defective mutants, presumably due to the failure of Spo11−/− deficient mice to form DSBs that initiate meiotic recombination. Spo11 deficiency does not affect HORMAD1 association with the chromosome axes [17], a finding similarly observed for Hop1 localization in Spo11−/− yeast [9]. Moreover, Hop1 mutants are required for full levels of DSBs formation [37], [54]. It is therefore possible that early loss of oocytes in Hormad1 deficient animals does not occur in part because unrepaired DSBs do not form. Reduced staining of γH2AX in Hormad1 mutant leptotene stage oocytes argues that unrepaired DSB formation is indeed reduced. However, unlike Spo11−/− ovaries that have abnormal folliculogenesis, Hormad1−/− ovaries are not defective in folliculogenesis and other mechanisms must protect Hormad1 mutant oocytes from early loss.
Infertility in Hormad1−/− females is due to early embryo demise. This is in stark contrast to males, where spermatocytes are eliminated due to the pachytene defect. The greater ability of oocytes to tolerate meiosis I errors is well known [55], and exemplified by Sycp3 and Smc1β knockouts [52], [56], which have defects in synapsis and recombination. Mechanisms that underlie oocyte's insensitivity to such errors are not well understood [57]. Embryos derived from Hormad1−/− oocytes fertilized with wild-type sperm, cannot implant. Examination of post-fertilization events in fertilized Hormad1−/− oocytes, showed significant abnormalities in early embryo development, including growth lag, and severe disruption in development prior to the blastocyst stage. Very few blastocysts were produced in fertilized Hormad1−/− oocytes, and most were abnormal in morphology. We utilized FISH hybridization against specific chromosomes to show widespread hypo and hyperdiploidy during early embryonic development. These results are reminiscent of Sycp3 and Smc1β knockout embryos. SYCP3 and SMC1β are involved in the synaptonemal complex formation, are germ cell–specific, and implicated in human aneuploidy. Sycp3 deficient females are subfertile, but one third of the embryos die in utero due to aneuploidy [29]. Smc1β encodes a meiosis specific component of the cohesin complex, important in sister chromatid cohesion and recombination [58]. Smc1β deficient females are sterile, like Hormad1−/− females, and meiosis continues until metaphase II [59].
However, unlike Smc1β−/−, Hormad1−/− ovaries do not lose oocytes by 6 months of age, and postnatal oocyte counts do not differ significantly from the wild type. Sycp3−/− ovaries show significant decline in the primordial follicle pool at postnatal day 8, indicating accelerated apoptosis of oocytes as germ cell clusters break down to form primordial follicles [52]. DSBs, as indicated by the assembly of the histone variant γH2AX, form in Smc1β−/− and Sycp3−/− meiocytes and may lead to higher rate of oocyte loss in these mutants [47], [52]. Hormad1 deficient ovaries do not display loss of primordial follicles. Hormad1−/− ovaries may not be losing oocytes at an accelerated rate due to our observations that DSBs, as assessed by γH2AX staining, are substantially reduced in Hormad1 mutants. Similar to Hop1 mutants in yeast, lack of HORMAD1 may derepress Dmc1-independent inter-sister repair pathway, resulting in efficient DNA break repairs [21]. The reduced DSBs in Hormad1−/− meiocytes, and de-repression of Dmc1-independent inter-sister repair pathways, may not affect regular apoptotic pathways that eliminate substantial number of oocytes during germ cell clusters breakdown to form primordial follicles at the time of birth, resulting in no decrease in Hormad1−/− primordial oocyte numbers.
All murine experiments were carried out on the 129S7/SvEvBrd x C57BL/6 hybrid background. Litters were weaned at 3 weeks and breeding pairs set up at 6 weeks of age. One mating pair was placed per cage and inspected daily for presence of litters. All experimental and surgical procedures complied with the Guide for the Care and Use of Laboratory Animals and were approved by the Institutional Agricultural Animal Care and Use of Committee of Baylor College of Medicine and the Institutional Animal Care and Use of Committee of University of Pittsburgh.
Targeting construct, electroporation and ES cell selection was done as previously described [60]–[61]. Exons 4 and 5 of the mouse Hormad1 gene were replaced with a neomycin resistance gene flanked by 5.3 kb (5′) and 6.2 kb (3′) homologous sequences (Figure S1A). Targeted ES cells were verified by Southern blotting, and microinjected into the C57BL/6 blastocysts to produce chimeric mice that carried the mutation into the germ line. These mice were mated with C57BL/6 wild-type mice to generate Hormad1 heterozygous animals that were subsequently crossed to produce F2 offspring for functional analysis. PCR genotyping was performed using the following primers: WT-forward (5′- TCAAGACCAACCTGGGCTAC -3′) and WT-reverse (5′- CCATGTGGGTTGTAGGGAGT -3′) to amplify a 196-nucleotide wild-type band, and KO- forward (5′- TCAAGACCAACCTGGGCTAC -3′) and KO- reverse (5′- GGGGAACTTCCTGACTAGGG -3′) to amplify a 505-nucleotide mutant band (Figure S1A).
Testes were fixed in Bouin's solution (Sigma-Aldrich, MO, USA) and ovaries were fixed in 10% buffered formalin or 4% paraformaldehyde. Fixed tissues were embedded in paraffin, serially sectioned (5 µm) and stained with hematoxylin and eosin or with Periodic Acid Schiff (PAS). At least five pairs of testes and ovaries of each genotype were subjected to gross and microscopic analyses for each time point. Germ cell cysts, primordial, primary, and secondary follicles were defined as described [45]. We used antibodies against SOHLH1 [45], PLZF ( sc-22839, Santa Cruz, CA,USA), DDX4 (ab13840, Abcam, MA, USA), ATM phospho-S1981 (05-740, Millipore, CA, USA), and HORMAD1 proteins. Affinity purified, anti-HORMAD1 guinea pig and rabbit antibodies, were generated against part of HORMAD1 protein (amino acids 23-373) at Cocalico Biologicals (Lancaster, PA, USA).
Oocytes and spermatocytes for chromosome analyses were prepared essentially as described previously [62]–[63]. Ovaries and testes were incubated in trypsin-EDTA solution at 37°C for 15 min and washed briefly in PBS. Trypsinized ovaries and testes were pipetted repeatedly and centrifuged, followed by resuspension in PBS. Cell suspensions were placed on poly-L-lysine-coated slides containing 120 mM sucrose solution and 0.05% Triton X-100. The slides were fixed in 2% paraformaldehyde and 0.02% SDS for 1 hour at room temperature, washed in distilled water and air dried, and stored at −80°C before use. Immunostaining was preformed as described [62]–[63]. Slides were incubated with the primary antibody overnight at 4°C, washed with PBS and incubated with Alexa-488 and Alexa-595 (Invitrogen, CA, USA) secondary antibody (1∶300 dilutions) for 1 hour at room temperature. After washing with PBS, slides were mounted using VECTASHIELD medium with 4,6-diamidino-2-phenylindole (DAPI) (Vector Laboratories, CA, USA). SYCP2 rabbit and guinea pig immunoaffinity-purified antibody [48], DMC1 (sc-8973, Santa Cruz, CA, USA), RAD51 (ab1837, Abcam, CA, USA) and RPA (ab87272, Abcam, CA, USA) were used. Dr. Christer Höög (Karolinska Institutet, Sweden) kindly provided SYCP1 and SYCP3 rabbit immunoaffinity-purified antibodies. Dr. William R. Brinkley (Baylor College of Medicine, TX, USA) kindly provided anti-CREST human serum.
Wild-type and mutant oocyte and spermatocyte spreads were stained at the same time with the same mixture of antibodies. In each experiment, when comparing wild type and mutants, imaging of the cells was performed on the same day with the same microscope and camera settings. PerkinElmer Volocity software 5.3 was used to control for possible changes in illumination during the course of imaging and measurement of the immunofluorescence. We measured total immunofluorescence of γH2AX in identical-sized rectangles that were placed over the cell boundaries. Fifty individual leptotene stage spreads from wild type, mutant spermatocytes and oocytes were subjected to immunofluorescence analysis. A two-tailed non-parametric Wilcoxon–Mann–Whitney two-sample rank-sum test was used for sample comparison.
Testes were dissected into ∼0.25 cm pieces and fixed in 1x PBS with 2% formaldehyde and 3% glutaraldehyde (Ted Pella Inc., CA, USA), for 2 hr at room temperature. Samples were treated with 0.5% uranyl acetate and osmium tetraoxide, dehydrated with ethanol, and embedded in LX-112 medium (Ladd Research Industries, VT, USA). The samples were polymerized in a 70°C oven for 2 days. Ultrathin sections (70 –100 nm) were cut in a Leica Ultracut microtome (Leica, IL, USA), stained for 5 min in 1% aqueous uranyl acetate and 2 min in 1% aqueous lead citrate at room temperature in a Leica EM Stainer, and examined by a JEM 1010 transmission electron microscope (JEOL, MA, USA) at an accelerating voltage of 80 kV. Digital images were obtained using AMT Imaging System (Advanced Microscopy Techniques Corp, MA, USA).
Western blot analysis was performed essentially as described previously [64]. Postnatal day 5, 8, 14, 18 and 21 testes were collected and lysed in RIPA buffer (50 mM Tris-HCl, pH 7.4, 1% NP-40, 0.25% sodium deoxycholate, 150 mM NaCl, 1 mM EDTA with or without protease inhibitor cocktail (Roche)). Total protein (20 µg) was loaded onto the SDS–PAGE gel, transferred to nitrocellulose membrane, and immunoblotted with HORMAD1 rabbit antibody (1∶5000 dilutions). Varying amounts of protein phosphatase I (Sigma P7937, MO, USA) were incubated with 20 µg of 18 day old wild-type total testes lysates, for 30 min on ice. Phosphatase treated testes lysates were electrophoresed, transferred to nitrocellulose membrane and immunoblotted with anti-HORMAD1 rabbit antibody.
RT-PCR and quantitative real time PCR were performed as described previously [45], [65]. We used previously published oligonucleotide sequences corresponding to Rnh2, Mov10l1, Pramel3, Usp26, Fthl17, Tex11, Tex13 [66], and Hormad1 (forward 5′ CTGCTGACACCAAGAAAGCA 3′- and reverse 5′- CCTGGTGGTTGGTAATCTGG -3′) and Hormad2 (forward 5′ GCTCATCAGGGGCTAGACTG 3′- and reverse 5′- TGGTTCGCTGACCTTCTTCT -3′) primers. Q-PCR was performed using the SyberGreen PCR Master Mix (Bio-Red, CA, USA) and probe set specific for each gene under investigation. Each sample was analyzed in triplicate from at least three independent wild type and Hormad1−/− testes cDNA samples. The relative amount of transcripts was calculated by the ΔΔCT method as described by Applied Biosystems, and normalized to β-actin. The average and standard errors were calculated for the triplicate measurements, and the relative amount of target gene expression for each sample was plotted. Student's t test was used to compute the P values. Significance was defined as a P value<0.05.
RNA expression arrays on wild-type and Hormad1−/− 2 week old testes were performed on the Illumina BeadChip MouseWG-6 2.0 arrays and analyzed as previously described (GEO accession numbers, GSE21524) [67]–[68].
Mouse oocytes and pre-implantation embryos were collected by using standard protocols for timed mating [69]. Briefly, 4 to 6 weeks female mice were superovulated by injection of 5 IU of pregnant mare serum gonadotropin (PMSG) (Prospec, Rehovot, Israel) and 48 h later 5 IU of human chorionic gonadotropin (hCG) (Sigma, MO, USA). Females were placed individually with 10-week old males immediately after the injection of hCG. Collection of one cell, two cell, four cell, eight cell, and blastocyst stages was performed at 24 h, 42 h, 68 h, and 96 h after hCG injection, by flushing the oviducts and uterine horns with HEPES-buffered media under the microscope. Removal of the cumulus cells was achieved in HEPES-buffered media containing 1 mg/ml hyaluronidase (Sigma, MO, USA).
Oocytes and embryos were exposed to a hypotonic solution containing 0.1% sodium citrate, 0.1% bovine serum albumin for about 15 min and gently transferred onto the slides. Cells were fixed with fresh methanol: acetic acid (3∶1) solution dispensed carefully over the surface of the slide. The slides were air-dried and dehydrated at room temperature in a series of 70%, 85% and 100% ethanol solutions before FISH analysis.
Twenty BAC (bacterial artificial chromosomes) clones from a mouse RPCI-23 library representing sequences unique to specific chromosomes were selected from the UCSC Genome browser (http://genome.ucsc.edu) and NCBI (http://www.ncbi.nlm.nih.gov) databases (Table S1). Probe DNA extraction was performed according to the standard alkaline lysis method and labeled by standard nick translation with Spectrum Orange- or Spectrum Green-dUTP using a commercially available kit (Abbott/Vysis, IL, USA). Sequential fluorescence in situ hybridization experiments using a mixture of three probes were performed according to manufacturer's instructions (Abbott/Vysis, IL, USA). Each mixture contained a red, green and yellow (combined green/red) fluorescent probes. Probes were applied to slides, hybridized for 20 hr at 37°C, washed with 0.4×SSC/0.3% NP-40 for 2 minutes at 73°C and with 2×SSC/0.1% NP-40 for 1 minute at room temperature, and counterstained with DAPI. Digital FISH images were captured by a Power Macintosh G3 System using MacProbe software version 4.4 (Applied Imaging, CA, USA).
|
10.1371/journal.pntd.0002246 | Prevalence and Diversity of Bartonella Species in Commensal Rodents and Ectoparasites from Nigeria, West Africa | Bartonellae are fastidious bacteria causing persistent bacteremia in humans and a wide variety of animals. In recent years there is an increasing interest in mammalian bartonelloses in general and in rodent bartonelloses in particular. To date, no studies investigating the presence of Bartonella spp. in rodents and ectoparasites from Nigeria were carried out.
The aim of the current study was to investigate the presence of Bartonella spp. in commensal rodents and their ectoparasites in Nigeria. We report, for the first time, the molecular detection of Bartonella in 26% (46/177) of commensal rodents (Rattus rattus, R. norvegicus and Cricetomys gambianus) and 28% (9/32) of ectoparasite pools (Xenopsylla cheopis, Haemolaelaps spp., Ctenophthalmus spp., Hemimerus talpoides, and Rhipicephalus sanguineus) from Nigeria. Sequence analysis of the citrate synthase gene (gltA) revealed diversity of Bartonella spp. and genotypes in Nigerian rodents and their ectoparasites. Bartonella spp. identical or closely related to Bartonella elizabethae, Bartonella tribocorum and Bartonella grahamii were detected.
High prevalence of infection with Bartonella spp. was detected in commensal rodents and ectoparasites from Nigeria. The Bartonella spp. identified were previously associated with human diseases highlighting their importance to public health. Further studies need to be conducted to determine whether the identified Bartonella species could be responsible for human cases of febrile illness in Nigeria.
| Bartonella species are zoonotic vector-borne bacteria that typically parasitize the erythrocytes of mammalian hosts, resulting in long lasting infections. They are responsible for a wide range of clinical manifestations in both immunocompetent and immunocompromised hosts.
Rodents and a wide range of small mammals serve as reservoirs of bartonellae, usually with no apparent clinical manifestations. Close association between rodents and humans especially in rural communities as well as in the overcrowded cities facilitates transmission of these bacteria.
There have been no studies investigating the presence of Bartonella spp. in rodents and ectoparasites from Nigeria. The aim of the current study was to investigate the presence of Bartonella spp. in commensal rodents and their ectoparasites in Nigeria and its public health implications. We report, for the first time, the molecular detection of Bartonella in 26% (46/177) of commensal rodents and 28% (9/32) of ectoparasite pools from Nigeria. Sequence analysis of the citrate synthase gene (gltA) revealed diversity of Bartonella spp. and genotypes in Nigerian commensal rodents and their ectoparasites. The Bartonella spp. detected in this study were identical or closely related to Bartonella elizabethae, Bartonella tribocorum and Bartonella grahamii previously associated with human diseases highlighting their importance to public health.
| Bartonellae are Gram-negative facultative intracellular alpha-proteobacteria belonging to the family Bartonellaceae. Many Bartonella species have been affecting human life for centuries [1]. Since the first Bartonella species discovery, namely Bartonella bacilliformis, in 1905 by Alberto Leonardo Barton Thompson, more than 30 species of Bartonella were identified [2], [3]. Bartonella species have been found in a variety of mammals, and the numbers of Bartonella species and their respective reservoir hosts are constantly growing [4]. They are pathogens of emerging and reemerging significance, causing a wide array of clinical syndromes in human and animal hosts [5].
These bacterial species are transmitted between the reservoir and the final mammal host by hematophagous arthropods and insects such as fleas, sand flies, mites, lice and possibly ticks, usually by their bites [6]–[8]. The range of vectors involved in the transmission of the different species of this genus has not been fully characterized [9]. Bacteria belonging to the genus Bartonella are slow growers in vitro, and the most used diagnostic methods are isolation, serology and polymerase chain reaction (PCR). The use of sequencing on PCR amplicons has been recommended in order to detect new species, especially when dealing with uncommon clinical presentations and settings [4].
Bartonella DNA has been detected in various hosts and possible vectors in many countries including, Israel [10], [11], Indonesia [12], Nepal [13], Thailand [6], [14], China [15], Taiwan [16], Korea [17], USA [18]–[20], UK [21], [22], Spain [23] and The Netherlands [24]. In Africa there are reports from Kenya [25], the Democratic Republic of Congo and Tanzania [26], Algeria [27], [28], Egypt [29], [30], Gabon [31] and South Africa [32], [33]. However, there is no report of molecular screening of humans or animals and their ectoparasites for Bartonella spp. in Nigeria.
Although there are no case estimates of fever of unknown origin (FUO) in Nigeria, the condition remains a challenging medical problem and unraveling the diagnosis could be a daunting task when investigating for common infective and non-infective causes. Moreover, since Bartonella spp. are difficult to diagnose and are seldom included in the differential diagnosis list in cases of FUO, specific Bartonella sp. treatment is rarely instituted to patients with FUO.
The objectives of this study were to determine the possible infection of commensal rodents and their ectoparasites from Nigeria with Bartonella spp., to investigate the presence of zoonotic Bartonella spp. in these rodents and ectoparasites and to evaluate genetic heterogeneity of circulating Bartonella strains in this country.
The study protocol was read and approved by The National Veterinary Research Institute Vom Ethical Committee on Animal Use and Care. Permission to place the traps in the study area was granted by the residents. Animals were treated in a humane manner and in accordance with authorizations and guidelines for Ethical Conduct in the Care and Use of Nonhuman Animals in Research of the American Psychological Association (APA) for use by scientists working with nonhuman animals (American Psychological Association Committee on Animal Research and Ethics) in 2010.
Rodents were live trapped in domestic and peri- domestic areas in Vom (9°44′N/8°47′E) Nigeria between October–December 2011. A total of 177 rodents (48 Rattus rattus, 121 Rattus norvegicus, 6 Mus musculus and 2 Cricetomys gambianus) were captured. Trapping was done using wire cage traps baited with smoked fish and other food scraps set out in the evenings when rodents are known to leave their holes to scavenge in farmlands or nearby human habitations. Traps were checked for rodents early the next morning. Cages containing rodents were transported to the Parasitology Laboratory, National Veterinary Research Institute (NVRI) Vom Nigeria, where they were identified and classified by a zoologist. At the laboratory, the cages containing rodents were placed into a clear plastic bag, which was sealed at the opening. Halothane gas was applied into the bag and the activity of the rodents was monitored. Once the rodents were anaesthetized, they were removed from the cage and bled by cardiac puncture. Depending on the size, 0.5–3 ml of blood was drawn and aliquoted into an EDTA tube and labeled. Each rodent was checked for ectoparasites by brushing the fur with a tooth brush onto a white cardboard paper. Ectoparasites were placed in labeled vials containing absolute ethanol corresponding to the host from which they were removed and stored at −20°C. Both blood and ectoparasite samples were transported in a cool box to The Koret School of Veterinary Medicine, The Hebrew University of Jerusalem, Israel for analysis. The ectoparasites were morphologically identified by an entomologist (KYM) at the Department of Microbiology and Molecular Genetics in Jerusalem, Israel.
Two hundred microlitres of thawed whole blood sample was plated onto chocolate agar. The plate was incubated at 35°C and 5% CO2 and checked for growth of Bartonella species on alternate days for up to 30 days. Suspected colonies were randomly selected and separately sub-cultured onto different fresh agar plates to obtain pure colonies.
DNA was extracted from blood using BiOstic Bacteremia DNA Isolation Kit (MO Bio Laboratories, Inc USA) according to manufacturer's instructions.
The ectoparasites collected from each rodent species were pooled (2–3 arthropods per pool) according to genus and/or species. DNA was extracted from each pool using Illustra tissue and cell genomic Prep miniSpin kit (GE Healthcare UK Limited) according to manufacturer's instructions.
Pure cultured colonies of Bartonella sp. were aseptically scooped into microfuge tubes containing 50 µl of sterile Phosphate Buffered Saline (PBS). DNA was extracted from the bacterial colonies using Illustra tissue and cell genomic Prep miniSpin kit (GE Healthcare UK Limited) according to the manufacturer's instructions.
The oligonucleotide primers: forward BhCS871.p (5′ -GGGGACCAGCTCATGGTGG-3′) and reverse: BhCS1137.n (5′-AATGCAAAAAGAACAGTAAACA-3′) [34] were used for amplification of a 379 bp region of the Bartonella citrate synthase gene (gltA). Positive and negative controls were included in each PCR run. PCR was performed using reaction tubes, preloaded with a premier PCR master mix (Syntezza PCR-Ready High Specificity, Syntezza Bioscience, Israel). 50 µl total volume was used as follows: 3 µl of DNA template, 1 µl of 10 mM each primer, 1 µl MgCl2, 19 µl of ultra pure PCR water and 25 µl PCR master mix. Amplification was performed using a conventional thermocycler (Biometra, Goettingen, Germany) and the following program parameters: an initial denaturing at 95°C for five minutes, and 35 cycles of denaturation at 95°C for one minute, annealing at 56°C for one minute, and elongation at 72°C for one minute. Amplification was completed by holding the reaction mixture at 72°C for 10 minutes.
PCR products were tested for the presence of amplicons of the correct size by electrophoresis of 6 µl of the products on 1.5% agarose gels stained with ethidium bromide and checked under UV light for the size of amplified fragments by comparison to a 50 bp DNA molecular weight marker. Amplicons of the proper size were identified by comparison to the positive control lane on the gel.
Positive PCR products were purified using (EXO-SAP IT USB, Cleveland, Ohio, USA) and sequenced using the forward primer at the Center for Genomics Technologies, Hebrew University of Jerusalem, Israel. To avoid errors or misinterpretation of the sequencing results, we deleted primer sequences from the gltA sequences and removed all ambiguities in the sequences before sequence analysis was performed.
Analysis of DNA sequences and phylogenetic relationships were done using MEGA 5.
Sequences were aligned by MUSCLE and the evolutionary history was inferred using the Maximum Likelihood method based on the Tamura-Nei model [35]. The bootstrap consensus tree inferred from 200 replicates was taken to represent the evolutionary history of the taxa analyzed. Branches corresponding to partitions reproduced in less than 50% bootstrap replicates are collapsed. The percentage of replicate trees in which the associated taxa clustered together in the bootstrap test are shown next to the branches. All positions containing gaps and missing data were eliminated.
A total of 177 cardiac blood samples from four rodent species were examined in this study: 68.4% (121/177) from Rattus norvegicus rats; 26% (48/177) from Rattus rattus rats; 1.1% (2/177) from Cricetomys gambianus rats, and 3.4% (6/177) from Mus musculus mice.
One hundred and seventy ectoparasites comprising of 85 ticks, Rhipicephalus sanguineus (79) and Haemaphysalis leachi (6); 13 fleas, Xenopsylla cheopis (8), Ctenophthalmus spp. (5) and 62 Haemolaelaps spp. (gamasid mites) were recovered from the rodents. Ten additional Hemimerus talpoides (earwig sp.) were removed from the 2 C. gambianus captured (Table 1).
Due to contamination problems, bartonellae could be cultured from a small subset of 30 rodent blood samples only. Nine of the latter 30 blood samples produced typical bartonellae growth. The colonies were creamy white in color, small, moist with metallic sheen and tended to pit on the agar. Initial growth of Bartonella sp. cultures were seen after 5–7 days of incubation. Colonies were sub cultured onto new plates to obtain pure cultures, which were harvested and preserved in 10% glycerol at −80°C until molecularly analyzed.
Bartonella gltA gene fragments were detected in 46 of 177 (26%) rodent blood samples screened in this study. One of 2 C. gambianus (50%), 36 of 121 R. norvegicus (29.8%), and 9 of 48 R. rattus (18.8%) were positive for Bartonella sp. DNA. None of the 6 M. musculus examined was positive for Bartonella sp. gltA. Nine of 32 (28%) ectoparasite pools removed from 48/177 (27.1%) rodents were positive for Bartonella gltA DNA. All the ectoparasite species tested were positive for Bartonella sp. gltA except H. leachi (Table 1).
Forty six gltA sequences were obtained from blood, 3 from bacterial cultures and 9 from ectoparasite samples. Selected Bartonella sequences were deposited in GenBank under the following accession numbers: JX0265667–JX0265697 for blood, JX026972 for culture, and JX 026997–JX027006 for ectoparasites.
Sequences obtained were compared with Bartonella sp. sequences deposited in GenBank for sequence similarity. Thirty six sequences were obtained from R. norvegicus blood, 26 of which had 98–100% identity with GenBank deposited B. elizabethae sequence (n = 2,100% identity; n = 23, 99%; n = 1, 98%). Nine of the sequences obtained from R. norvegicus blood had 97–98% identity with GenBank deposited Bartonella tribocorum sequence, while 1 sequence had 98% similarity with GenBank deposited Bartonella grahamii. Nine sequences were obtained from R. rattus blood, 7 of which had sequence identity of 98–100% with GenBank deposited B. elizabethae sequence (n = 3, 100% identity; n = 1, 99%; n = 3, 98%) (Table 2). The sequence retrieved from the blood of C. gambianus had 99% identity with GenBank deposited B. elizabethae.
Bartonella gltA sequences obtained from one pool each of X. cheopis, R. sanguineus, and 3 pools of Haemolaelaps sp. had 97–100% similarity to B. elizabethae deposited in GenBank while a sequence from Ctenophthalmus sp. pool had 97% identity with B. tribocorum sequence deposited in GenBank. Interestingly, Bartonella sp. DNA with 99% sequence identity to B. elizabethae deposited in the GenBank was detected from one pool of H. talpoides earwigs that were removed from C. gambianus rats.
Bartonella spp. DNA was detected in 4 of 13 (30.8%) rodents from which the ectoparasites were removed. However, only one ectoparasite, H. talpoides removed from C. gambianus had the same percent sequence identity (100%) with that of the host. The DNA sequences from the ectoparasites had 97–99% identity with their first GenBank match (Table 3). The R. sanguineus pool that was positive for Bartonella spp. DNA was collected from R. norvegicus rat that was negative for Bartonella sp. DNA.
The phylogenetic relationship among the genotypes obtained in the present study and previously described Bartonella species is presented in Figure 1. Sequences of Bartonella sp. from this study formed 3 distinct clusters A-C along with B. elizabethae and B. grahamii (Fig. 1), but was distantly related to other sequences deposited in the GenBank. The first cluster (cluster A) consists of 4 sequences closely related to B. elizabethae. However, 5 other sequences that were 97–100% similar to B. elizabethae appear as single genotypes just below cluster A. Cluster B is made up of 2 sequences that were similar to B. grahamii deposited in the GenBank. The cluster C consists of 5 sequences that were 97–100% similar to B. tribocorum deposited in the genBank.
Sequences were coded based on rodent or ectoparasites species from which they were detected, accession numbers are in parentheses; RR = Rattus rattus; RN = Rattus norvegicus; CG = Cricetomys gambianus; CS = Ctenophthalmus sp; HT = Hemimerus talpoides; RS = Rhipicephalus sanguineus, MS = Haemolaelaps spp.; XC = Xenopsylla cheopis.
In this study, we report the molecular detection and genetic characterization of Bartonella species in rodents and ectoparasites from Nigeria, West Africa. Moreover, to the best of our knowledge, this is the first report of molecular investigation of Bartonella spp. in rodents and their ectoparasites in this country. The 26% prevalence of Bartonella DNA found in this study was higher than the 8.5% prevalence reported in small mammals from the Democratic Republic of Congo but lower than and 38% reported in Tanzania [26]. The differences between the findings in the latter studies and ours can be attributed to the fact that commensal rodents were screened in the current study while sylvatic rodents were screened in the DR Congo and Tanzania studies. Similarly, the 28% prevalence of Bartonella DNA by gltA PCR in ectoparasites in this study was slightly higher than the 21.5% reported in fleas from Algeria, targeting 3 genes and the inter-genic spacer (ITS) [28]. The high prevalence of detection of Bartonella spp. DNA in the ectoparasites attests to their role as vectors of these bacteria.
Several Bartonella spp. that were associated with human diseases were identified in this study, including B. elizabethae, B. grahamii and B. tribocorum. Bartonella elizabethae was found in patients with endocarditis [36]. Bartonella grahamii was associated with neuroretinitis or bilateral retinal artery branch occlusions [37]. A Bartonella genotype closely related (97%) to B. tribocorum was detected in the blood of human patient with fever from Thailand [25]. The finding of these zoonotic Bartonella spp. in commensal rodents from Nigeria demonstrates their importance as reservoirs for various zoonotic Bartonella species and warrants increased awareness of physicians and health care workers for these pathogens especially in unidentified febrile cases.
In this study, no DNA sequence similar to B. tribocorum was obtained from R. rattus rats. The detection of B. tribocorum only in R. norvegicus rats is in agreement with the earlier report of Márquez et al. [23] which supports the hypothesis that there is specificity of Bartonella spp. for their rodent hosts [15].
Although the role of R. sanguineus ticks in transmitting Bartonella spp. in nature is not proven [38] it is important to note that we detected Bartonella DNA in R. sanguineus ticks. Detection of Bartonella DNA in ticks was previously reported also by other authors [8], [17], [39]. The Bartonella spp. DNA detected from one R. sanguineus tick pool had 97 percent identity to B. elizabethae sequences deposited in GenBank. It is worthy to note that the host from which the R. sanguineus ticks were removed was negative for Bartonella spp. DNA. This suggests that the ticks might have acquired the bacteria during previous feeding on an infected host. The ability of the tick to transmit this organism to a susceptible host during the next feeding stage or to its progeny is worth further investigation.
Comparative analyses of the gltA sequences obtained from Bartonella spp. showed that commensal rodents in Nigeria harbor a diverse assemblage of Bartonella strains, some of which represent known Bartonella spp. and strains and others may represent distinct novel strains. Although only a portion of the citrate synthase gene (gltA) was used for phylogenetic analysis, this gene has been shown to be a reliable tool for distinguishing between closely related Bartonella genotypes [40]. By using this partial gene, it was possible to compare the variety of Bartonella genotypes isolated from rodents with homologous sequences of Bartonella strains found in other mammals, reported from other parts of the world. Finding considerable sequence diversity is typical for different species of Bartonella, although more characteristics are needed to describe novel Bartonella species [3].
In this study, the Bartonella genogroups identified in commensal rodents formed three separate clusters closely related to B. elizabethae but distantly related to other known Bartonella spp. Although BLAST searches shows some of the sequences had 97–100% similarity to B. tribocorum and B. grahamii sequences deposited in GenBank (Fig. 1). The findings of Bartonella sequences that were genetically distant from known GenBank deposited sequences requires further investigation in characterizing these genotypes and ascertaining whether they are pathogenic to animals and/or humans.
Pools of H. talpoides collected from C. gambianus in this study contained Bartonella DNA. Hemimerus talpoides (earwig sp.) are presumed to feed on the epidermis of their host or as a saprophytic on fungus from the skin of the host. The detection of Bartonella sp. DNA in this ectoparasite is interesting and requires further investigation [41].
In conclusion, this study has resulted in the identification and genetic characterization of Bartonella genotypes in commensal rodents and ectoparasites from Nigeria, West Africa. A high prevalence and diversity of Bartonella spp. and strains was detected in commensal rodents and their ectoparasites in this study. Several zoonotic Bartonella spp. including B. elizabethae, B. grahamii and B. tribocorum were identified for the first time in Nigeria highlighting their importance for public health in this country.
|
10.1371/journal.pbio.3000162 | A determining factor for insect feeding preference in the silkworm, Bombyx mori | Feeding preference is critical for insect adaptation and survival. However, little is known regarding the determination of insect feeding preference, and the genetic basis is poorly understood. As a model lepidopteran insect with economic importance, the domesticated silkworm, Bombyx mori, is a well-known monophagous insect that predominantly feeds on fresh mulberry leaves. This species-specific feeding preference provides an excellent model for investigation of host-plant selection of insects, although the molecular mechanism underlying this phenomenon remains unknown. Here, we describe the gene GR66, which encodes a putative bitter gustatory receptor (GR) that is responsible for the mulberry-specific feeding preference of B. mori. With the aid of a transposon-based, clustered regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated protein-9 nuclease (Cas9) system, the GR66 locus was genetically mutated, and homozygous mutant silkworm strains with truncated gustatory receptor 66 (GR66) proteins were established. GR66 mutant larvae acquired new feeding activity, exhibiting the ability to feed on a number of plant species in addition to mulberry leaves, including fresh fruits and grain seeds that are not normally consumed by wild-type (WT) silkworms. Furthermore, a feeding choice assay revealed that the mutant larvae lost their specificity for mulberry. Overall, our findings provide the first genetic and phenotypic evidences that a single bitter GR is a major factor affecting the insect feeding preference.
| The molecular mechanism underlying species-specific feeding preference in insects is poorly understood. The silkworm, Bombyx mori, is a typical monophagous plant-eating insect, but the genetic basis for its famous mulberry-specific feeding preference is unknown. Here, we identify gustatory receptor 66 (GR66) as a determinant of the silkworm’s mulberry-specific monophagy. GR66-mutant larvae generated by clustered regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated protein-9 nuclease (Cas9) acquired new feeding activity and showed the ability to feed on various plant species that are not normally consumed by the wild-type (WT) animals; a two-choice assay demonstrated that the mutant larvae had lost their feeding preference for mulberry. Our genetic and phenotypic evidence therefore demonstrates that GR66 is a major factor affecting the feeding preference of the silkworm.
| Chemosensory processes, including olfaction and gustation, are critical for host-plant selection in phytophagous insects [1,2]. Olfaction is responsible for host orientation, and gustation plays a central role in host selection [3,4]. Insect gustatory receptors (GRs), as well as olfactory receptors (ORs), therefore play critical roles in determining insect feeding preference. Most insect GRs are expressed exclusively in gustatory receptor neurons (GRNs) and transmit signals through GRNs to regulate insect feeding behaviors [5,6]. Insect GRs are known to recognize sugars, bitter compounds, and nonvolatile pheromones [7,8]. In Drosophila melanogaster, GR5a and GR66a are found in different populations of GRNs [5]. GR5a-positive GRNs respond to various sugars, and GR66a-positive GRNs respond to many bitter compounds [9,10]. In the butterfly, Papilio xuthus, a GR was reported to be involved in host-plant recognition for oviposition [11]. In addition, GRs are also required for the detection of CO2, nutrients, light, and temperature [12–14]. Large numbers of insect GRs have been identified in many insect species [15–24]. However, most GRs have not been functionally characterized, and the roles played by these GRs in insect feeding preferences remain unclear.
Based on the host-plant selection range, the feeding preferences of phytophagous insects are classified as monophagous, oligophagous, and polyphagous. Lepidoptera, the largest lineage of phytophagous insects, includes many important agricultural and forest pests that exhibit high diversity in terms of feeding preference. The domesticated silkworm, Bombyx mori, is a beneficial lepidopteran insect that has been a major contributor to silk production for thousands of years. One of the main characteristics of B. mori is its monophagous feeding preference, and silkworm larvae predominantly feed on fresh mulberry leaves (Morus alba L.). Several polyphagous silkworm mutant strains that feed on the leaves of various plants that are rejected by normal silkworms have been reported [25,26]. Genetic analysis of one representative strain, Sawa-J, revealed that a major recessive gene on the polyphagous (pph) locus was potentially responsible for this change in feeding preference [27]. However, the molecular mechanism underlying the monophagous feeding preference of B. mori is unknown, and whether GR genes are involved the feeding preference of silkworm remains to be determined. Recently, a complete set of 76 GR genes was identified in B. mori [28]. Among these genes, only three sugar GRs were functionally characterized [29–31], whereas most of the GRs remained functionally identified, including 66 putative bitter GRs [28].
The biological functions of most insect GRs are poorly understood, especially those of nondrosophilid insects, due to the lack of reverse genetic approaches for the study of these insect species. This is especially true for lepidopteran species, because RNA interference functions with variable efficiency in many species [32]. Recent advances in the development of targeted genomic manipulation tools provide great benefits for functional genomic research of lepidopteran insects. These genomic manipulation tools—including zinc-finger nucleases (ZFNs), transcription activator-like effector nucleases (TALENs), and the clustered regularly interspaced short palindromic repeats (CRISPR)/CRISPR-associated protein-9 nuclease (Cas9) system—have been extensively used to generate targeted mutations at single or multiple sites in many organisms in vitro and in vivo [33–35]. Among these tools, the CRISPR/Cas9 system is the most extensively used mutagenesis system due to its high mutagenic efficiency and simple procedure. Among lepidopteran insects, the CRISPR/Cas9 system has been successfully established in B. mori [36–38], Spodoptera litura [39], Plutella xylostella [40], and Helicoverpa armigera [41].
In the current study, we investigated the genetic basis for the feeding preference towards mulberry exhibited by silkworm. We mutated the GR66 gene, which encodes a putative bitter GR, in B. mori using the Cas9/small guide RNA (sgRNA) system. Homozygous GR66 mutant larvae exhibited expanded diets, indicating that the GR66 gene is responsible for mulberry-specific feeding behavior in the silkworm. Acquiring new feeding activity in the silkworm will contribute to modern sericulture as well as to the understanding of the molecular mechanisms of insect–host interactions.
It was reported that there are 76 putative GRs distributed on 16 of the 28 chromosomes of B. mori [28]. Among these genes, only one putative bitter GR gene, GR66, was identified as being located on the third chromosome. The genomic locus of this gene is within the putative pph locus of the polyphagous Sawa-J silkworm strain [27]. This finding indicates that GR66 might be the candidate gene for the pph locus and could be involved in the feeding preference of silkworm. We first investigated the relative mRNA levels of GR66 in different larval tissues using quantitative real-time PCR (qRT-PCR). It has been reported that most insect GRs are localized in the taste sensilla of the larval mouthparts [28,42] (Fig 1A). As expected, GR66 was predominantly expressed in larval maxillae (Fig 1B). The open reading frame (ORF) of the GR66 gene contains 1,140 base pairs and encodes a 380-amino-acid polypeptide. Bioinformatic analysis revealed that the GR66 protein consists of seven transmembrane domains with an intracellular N terminus, which is distinct from the structures of members of the G-protein-coupled receptor (GPCR) family (Fig 2B). We further investigated the cellular localization of this protein via transfection of an enhanced green fluorescent protein (EGFP)-fused GR66 expression plasmid into mammalian 293T cells. The results showed that the protein is localized on the cell membrane (Fig 1C).
To investigate the potential involvement of GR66 in the feeding preference of silkworm, we genetically ablated GR66 using a transposon-based, Cas9/sgRNA-mediated mutagenesis system [37]. Two independent transgenic lines were established by transposon-mediated germline transformation. One transgenic line expressed Cas9 under the control of the germ-cell–specific promoter Bmnos [37], and the other line expressed two sequence-specific sgRNAs targeting GR66 (Fig 2A) under the control of the BmU6 promoter [38]. Each line also expressed an IE1 promoter-derived fluorescent marker (EGFP in the Cas9-expressing line or DsRed2 in the sgRNA-expressing line) to facilitate the screening of positive individuals from the embryonic stage [37]. In the F1 hybrids between the Cas9 and sgRNA lines, somatic mutagenesis was identified by PCR-based analysis and subsequent sequencing. Mutants were generated at a single site or both sites (S1A Fig), indicating that successful mutagenesis was induced by the transgenic CRISPR/Cas9 system. Somatic mutants of GR66 showed no deleterious phenotype compared with the wild-type (WT) animals, indicating that knocking out GR66 did not interfere with silkworm development and fertility. To obtain heritable, nontransgenic, homologous mutants to assess feeding preference, a series of crossing strategies and PCR-based screening experiments were performed (S1B Fig) as described previously [43]. Finally, two independent homozygous lines with truncated GR66 proteins were established (S2 Fig). One mutant line (ΔGR66-1) had a 929-bp genomic DNA deletion at the GR66 locus, resulting in a 180-bp deletion in the ORF and to a truncated 319-aa protein, which was 60 aa shorter than WT GR66 protein (S2 Fig). The other mutant line (ΔGR66-2) had a 931-bp genomic deletion at the GR66 locus, resulting in a 182-bp deletion in the ORF and to a 312-aa protein, which was 67 aa shorter than the WT GR66 protein (S2 Fig). The truncated GR66 proteins of the ΔGR66-1 and ΔGR66-2 mutants contained only six or five transmembrane domains, respectively (Fig 2B). Because the truncated proteins did not have all seven transmembrane domains that are essential for the function of the membrane proteins [44,45], we presumed that both mutants lacked GR66 functions.
Consistent with the transgenic somatic mutants, homologous GR66 mutant silkworms were fully viable and fertile. We first used homozygous ΔGR66-2 newly moulted fifth-instar larvae to assess feeding behavior. After 24 h of starvation treatment to facilitate feeding sensitivity, both WT and homozygous GR66-2 mutant larvae were provided various food sources for 24 h (Fig 3A–3E), and then, the increase in weight and number of droppings were recorded (Fig 3G and 3H). The leaves of Mongolian oak (Quercus mongolica Fisch. ex Ledeb.), fruits of apple (Malus domestica) and pear (Pyrus spp.), and seeds of soybean (Glycine max) and corn (Zea mays) were subjected to analysis. Mulberry leaves were also used as a control. Both WT and mutant larvae ate the mulberry leaves and exhibited normal development (Fig 3A and S1 Movie). Leaves of Mongolian oak are known food sources of Chinese oak silkworm, Antherea pernyi, but are not consumed by B. mori. The ΔGR66-2 larvae ate the oak leaves (Fig 3B and S2 Movie), and droppings were observed (Fig 3H), but the body weights did not increase significantly (Fig 3G). The ΔGR66-2 larvae exhibited a 15.96% weight increase with approximately seven droppings per larva after feeding on apple, whereas the WT animals did not attempt to consume apple, and no droppings were observed (Fig 3C, 3G and 3H and S3 Movie). Furthermore, we found that the ΔGR66-2 larvae could also feed on pear (Fig 3D and S4 Movie), which belongs to the same family as apple, namely, Rosaceae. A 25.47% weight increase was observed for ΔGR66-2 larvae, whereas no significant increase was observed for WT animals (Fig 3G and 3H). The ΔGR66-2 larvae could feed on both fresh soybean and corn, with a 10.56% and 14.08% increase in weight, respectively, whereas no significant weight increase was observed for WT animals (Fig 3E–3H, S5 Movie and S6 Movie). After feeding, the larvae were dissected to confirm food digestion, and the results showed that the midguts were filled with the residues of the indicated foods (Fig 3A’–3F’). Additionally, the Mongolian oak leaf residue diffused into the anterior part of the midguts (Fig 3B’), indicating that Mongolian oak leaves could not be digested well. This finding also explained why the body weight did not increase significantly (Fig 3G). A similar result was obtained when the ΔGR66-1 mutant line was subjected to analysis (S3 Fig). Notably, none of the larvae could survive the entire fifth-instar stage when reared on food other than mulberry (S4 Fig), indicating that B. mori mostly adapted to mulberry leaves during long-term cultivation.
To further investigate the feeding preference of GR66 mutants, we performed a two-choice assay in prestarved fifth-instar larvae. Given a choice between mulberry leaves and Mongolian oak leaves, the WT larvae exhibited a strong preference for mulberry leaves and did not attempt to eat Mongolian oak leaves (Fig 4A and 4A’). In contrast, the ΔGR66 larvae exhibited similar feeding preferences for both mulberry leaves and Mongolian oak leaves (Fig 4B, 4B’, 4C and 4C’). In addition, a commercial artificial diet containing mulberry leaf powder and another artificial diet that lacked mulberry leaf (1:1 ratio of soybean powder to corn powder) were also used for a two-choice assay. Similar to the previous result, the WT larvae exhibited a strong preference for the artificial diet containing mulberry (Fig 4D and 4D’), whereas the ΔGR66 larvae exhibited similar feeding preferences for both artificial diets (Fig 4E, 4E’, 4F and 4F’). These results revealed that the GR66 mutant larvae had lost their specificity for mulberry, suggesting that GR66 is required for the mulberry-specific feeding preference of B. mori. In addition, we performed two-choice feeding assays with neonate larvae. Both the WT and GR66 mutant neonate larvae exhibited a strong preference for the artificial diet containing mulberry (S5 Fig). Although this phenotypic consequence remained to be elucidated, we speculated that food choice of neonate larvae are also strongly affected by ORs, because olfaction is responsible for host orientation [46].
Most insect GRs are located in the taste sensilla of the larval mouthparts, and it has been reported that the medial sensilla are responsible for sweet taste perception and lateral sensilla are responsible for bitter taste perception in Lepidoptera [28]. To investigate whether GR66 mutants exhibit altered responses to different tastes, electrophysiological recording analysis on contact chemosensilla was performed on taste sensilla, including the medial and lateral styloconic sensilla of fifth-instar larvae in the ΔGR66-2 line. We first investigated two sweet stimulants, namely, sucrose and myo-inositol, in the lateral sensilla. No difference was detected between WT and GR66 mutants at a concentration of 10 mM, indicating that GR66 depletion was irrelevant for the perception of these two sweet stimuli (S6A and S6B Fig). We subsequently investigated two bitter substances, namely, caffeine and salicin, in the medial sensilla at a concentration of 10 mM. The results showed that the electrophysiological response to these two substances was not affected by GR66 depletion (S6C and S6D Fig). We further tested the response to caffeine and salicin at different concentrations, and similar results were obtained (S6E and S6F Fig). These results indicated that the GR66 mutants did not exhibit altered responses to these typical sweet or bitter substances. Other compounds in mulberry leaves, especially the potential ligands of GR66, remain to be identified.
Molecular mechanisms of host-plant selection in phytophagous insects remain to be elucidated, and how GRs are involved in their feeding behaviors is poorly understood. To reveal the molecular mechanism underlying mulberry-specific herbivory in B. mori, we genetically ablated a putative bitter GR, GR66, via Cas9/sgRNA-mediated targeted mutagenesis. Homologous mutant larvae exhibited loss of mulberry specificity and the ability to feed on a wide range of food sources, indicating that GR66 is a determinant of the monophagous feeding preference of B. mori.
Increasing numbers of insect GRs have been identified, and their critical roles in detection of environmental stimulations have been reported [7–14]. In phytophagous insects, most reported GRs belong to putative bitter GR subfamily and they are necessary in the recognition of many plant secondary metabolites, which are normally bitter compounds [47]. In B. mori, the subfamily of the bitter GRs contains up to 66 genes and is the largest subfamily among the total 76 identified GRs in B. mori [28]. None of these putative bitter GRs had been functionally elucidated until the current study on GR66. Our data strongly suggest that GR66 is a major factor affecting the feeding preference of silkworm, because mutation of this gene could change the mulberry-specific herbivory of silkworm. We speculate that GR66 may serve as a feeding inhibitor in B. mori. This finding explains why GR66 mutagenesis could result in the acceptance of an expanded range of host-plant materials by the larvae. In WT animals, GR66 is active and inhibits the feeding behavior on nonhost materials, whereas certain compounds in mulberry leaves directly or indirectly repress GR66 activity, leading to initiation of such feeding behavior. Future validation of potential ligands of GR66 in mulberry leaves and identification of food components that dictate host specificity will be critical for elucidation of this species-specific feeding preference. In the current study, the ΔGR66 strains did not exhibit significant electrophysiological differences in the selection of sweet or bitter substances, including salicin. Our results were different from previously reported results for the polyphagous silkworm strain Sawa-J, which exhibited reduced sensitivity to the bitter compound salicin [26]. Because the pph locus in the Sawa-J strain has not been mapped to a single gene [26], the different electrophysiological phenotypes between the Sawa-J and ΔGR66-2 strains indicated that the putative involvement of different or additional genes, such as the many other GR genes in B. mori, should be taken into account to explain the monophagous feeding preference for mulberry. We presumed that the effects of these genes led to the Sawa-J strains and GR66 mutants exhibiting different responses to salicin. Additionally, it is possible that GR66 mutagenesis did not create completely null mutants (Fig 2), and truncated GR66 may still respond to salicin.
Mulberry leaves have been used as the only food source for mass rearing of silkworm for thousands of years. Due to limitations associated with labor and land consumption and seasonal cycles in the harvesting of fresh mulberry leaves, the development of silkworm strains that can feed on cost-effective diets instead of mulberry leaves has been pursued. Conversion of the monophagous silkworm to a polyphagous species by GR mutagenesis therefore provides a promising approach for the development of alternative food sources for mass rearing of silkworm. Furthermore, lepidopteran insects include a large number of agricultural and forest pests that exhibit high diversity in terms of feeding habits. Orthologous genes of GR66 or other GRs in lepidopteran insects may play key roles in the species-specific feeding preferences of these insects. Insect feeding preference is a very complicated biological process and is probably more complex than determined by a single gene. Large numbers of insect GRs remains to be functional elucidated, and they should also be considered to play important roles in feeding preference. Elucidation of the critical role of GRs in insect feeding preference will provide insights into the mechanisms underlying insect feeding behavior and insect–plant interactions, facilitating the development of novel strategies for pest management.
A multivoltine and monophagous silkworm strain, Nistari, was used in all the experiments. Larvae were fed fresh mulberry leaves at 25°C under standard conditions [48].
Heads were excised from newly hatched first-instar larvae of B. mori. The excised heads were washed in PBS and fixed with FAA solution (1:1:18 ratio of 37% to 40% formaldehyde to acetic acid anhydride to 50% ethanol). The fixed samples were dehydrated via exposure to gradually increasing concentrations of ethyl alcohol (50%, 60%, 70%, 80%, 90%, 95%, 100%) using a rotary machine. The heads were dried in a critical-point dryer and then coated with platinum prior to observation under a scanning electron microscope (JEOL).
Total RNA was isolated from the antennae, labra, mandibles, maxillae, labia, thoracic legs, and midguts of third-day fifth-instar (L5D3) larvae using TRIzol reagent (Invitrogen). The RNA was treated with DNase I (Invitrogen) to remove genomic DNA. One microgram of total RNA was used to synthesize cDNA using the ReverAid First Strand cDNA Synthesis Kit (Fermentas). Relative mRNA levels were determined by qRT-PCR using SYBR Green real-time PCR master mix (TOYOBO). The PCR conditions used were as follows: initial incubation at 95°C for 1 min, followed by 40 cycles of 95°C for 15 s and 60°C for 1 min. The primers used for qRT-PCR are listed in S1 Table. Another primer pair—namely, RP49-F and RP49-R (S1 Table)—was used as an internal control [48].
The ORF of BmGR66 was PCR-amplified using cDNA synthesized from the total RNA isolated from the maxillae at L5D3 as a template. The PCR products obtained were directly cloned into the pcDNA-3.0 vector to generate GR66-pcDNA3.0. To detect the expression of BmGR66 in human embryonic kidney 293T (HEK293T) cells, the ORF of GFP was cloned and incorporated in-frame upstream of BmGR66 with a flexible linker modifying the amino acids GGGGS. To construct the transgenic CRISPR/Cas9 system, we used the activator line pBac[IE1-DsRed2-Nos-Cas9] (Nos-Cas9), in which Cas9 was driven by a germ-cell–specific promoter, as described previously [37]. The plasmid pBac[IE1-EGFP-U6-BmGR66-sgRNA] (U6-sgRNA), used to express the sgRNA, was constructed as described previously [38]. The sgRNA targeting sites were designed as GN19NGG. The primers used for plasmid construction are listed in S1 Table.
HEK293T cells were cultured in Dulbecco's modified Eagle’s medium (DMEM, Thermo Fisher Scientific) supplemented with 10% fetal bovine serum (FBS) at 37°C and 5% CO2. For receptor localization analysis, HEK293T cells were seeded in 35-mm sterilized glass-bottom dishes and incubated for 24 h. EGFP-GR66-pcDNA3.0 was transfected into HEK293T cells using Lipofectamine 2000 (Invitrogen). After 24 h, the cells were fixed with 4% paraformaldehyde for 15 min and finally incubated with DAPI for 10 minutes. The cells were visualized by fluorescence microscopy on a Zeiss LSM 510 confocal laser scanning microscope attached to a Zeiss Axiovert 200 microscope using a Zeiss Plan-Apochromat 63×/1.40 NA oil immersion lens.
Germline transformation of silkworm was performed as described previously [48]. For the transgenic CRISPR/Cas9 system, the Nos-Cas9 line was crossed with the U6-sgRNA line, and genomic DNA was extracted from the Nos-Cas9:U6-sgRNA as previously described [38]. Subsequently, genomic PCR followed by sequencing was carried out to identify GR66 mutant alleles.
To establish a stable homozygous mutant line, the Nos-Cas9:U6-sgRNA (F1) were crossed with the WT. For the F2 progeny that lacked fluorescence, PCR-based genotyping was performed using genomic DNA extracted from adult legs as templates. Removal of legs did not interfere with moth survival and fertility. Details regarding the crossing procedure are shown in S1 Fig. Briefly, we backcrossed F1 somatic mutants with WT moths and used PCR to identify heterozygous F2 mutant animals. The selected F2 mutants were backcrossed with WT moths again. The progeny of this cross were approximately 50% heterozygotes and 50% WT animals. The F3 heterozygous animals were then sib-mated. The progeny of this cross were approximately 25% homozygous mutants, 50% heterozygous mutants, and 25% WT animals. The F4 homozygous mutants were then sib-mated to obtain 100% homozygous animals, which were used in subsequent experiments.
Newly moulted fifth-instar larvae were starved for 24 h prior to conducting the behavioural assay. After starvation, each larva was placed in a sterile culture dish separately. Different plant-derived food materials, such as mulberry (M. alba), Mongolian oak (Q. mongolica Fisch. ex Ledeb.), apple (M. domestica), pear (Pyrus spp.), soybean (G. max), and corn (Z. mays), were placed in the culture dishes. After 24 h, the weights of larvae were recorded, and the number of droppings was counted. Two-choice feeding preference tests were performed using plant leaves or artificial diets. Leaves of mulberry and Mongolian oak were placed on separate sides of the container 2 cm away from the middle. A two-choice feeding assay with an artificial diet containing mulberry leaf powder and an artificial diet that was 1:1 ratio of soybean powder to corn powder was performed as described above. Twenty newly moulted fifth-instar larvae after starvation for 24 h or a brood of neonate larvae were placed in the center. Photographs were taken at 0 and 60 min after release.
Tip recordings for insect contact chemosensilla were performed on the medial and lateral styloconic sensilla of fifth-instar B. mori larvae as described previously with some modification [49, 50]. Heads with the first thoracic segments were cut from newly hatched fifth-instar larvae that were starved for 24 h. An AgCl-coated silver loop was inserted into each head until pressure caused the mouthparts to open, and then the loop was connected to a copper miniconnector, which served as the recording electrode. A recording glass electrode filled with the stimulus solution was brought in contact with the tip of the styloconic sensillum under a dissecting microscope. Responses were recorded from both the medial and lateral styloconic sensilla on both sides of the head. Stimuli lasted 1 s and were separated by an interval of 3 min to allow for recovery and to minimize adaptation. The tip diameter size of the stimulating electrode was approximately 50 μm, which is suitable for stimulation of single styloconic sensilla. Action potentials (spikes) generated during the first second after stimulus onset were amplified by the amplifier (Syntech Taste Probe DTP-1; Hilversum, the Netherlands) and filtered (A/D-interface, Syntech IDAC-4; Hilversum, the Netherlands). The electrophysiological signals were recorded and analyzed with the aid of spike analysis programs for insect data (SAPID) Tools software, version 16.0 [51], as well as Autospike version 3.7 software (Syntech, Hilversum, the Netherlands). Solutions of sucrose, myo-inositol, caffeine, and salicin dissolved in 2 mM KCl were used as stimulants in the electrophysiological experiments. For each stimulant and corresponding sensillum responsive to the stimulant, 15 WT and mutant larvae that hatched from 3 to 5 different rearing batches were tested. A solution of 2 mM KCl served as a control. Data are presented as the means ± standard error of the means (SEMs).
All the experiments in this study were performed with at least three replicates. All the data are expressed as the mean ± SEM. The differences between groups were examined by either two-tailed Student t-test or two-way ANOVA. Statistically significant differences are indicated by asterisks.
|
10.1371/journal.pcbi.1002981 | Probabilistic Inference of Biochemical Reactions in Microbial Communities from Metagenomic Sequences | Shotgun metagenomics has been applied to the studies of the functionality of various microbial communities. As a critical analysis step in these studies, biological pathways are reconstructed based on the genes predicted from metagenomic shotgun sequences. Pathway reconstruction provides insights into the functionality of a microbial community and can be used for comparing multiple microbial communities. The utilization of pathway reconstruction, however, can be jeopardized because of imperfect functional annotation of genes, and ambiguity in the assignment of predicted enzymes to biochemical reactions (e.g., some enzymes are involved in multiple biochemical reactions). Considering that metabolic functions in a microbial community are carried out by many enzymes in a collaborative manner, we present a probabilistic sampling approach to profiling functional content in a metagenomic dataset, by sampling functions of catalytically promiscuous enzymes within the context of the entire metabolic network defined by the annotated metagenome. We test our approach on metagenomic datasets from environmental and human-associated microbial communities. The results show that our approach provides a more accurate representation of the metabolic activities encoded in a metagenome, and thus improves the comparative analysis of multiple microbial communities. In addition, our approach reports likelihood scores of putative reactions, which can be used to identify important reactions and metabolic pathways that reflect the environmental adaptation of the microbial communities. Source code for sampling metabolic networks is available online at http://omics.informatics.indiana.edu/mg/MetaNetSam/.
| We present a probabilistic sampling approach to profiling metabolic reactions in a microbial community from metagenomic shotgun reads, in an attempt to understand the metabolism within a microbial community and compare them across multiple communities. Different from the conventional pathway reconstruction approaches that aim at a definitive set of reactions, our method estimates how likely each annotated reaction can occur in the metabolism of the microbial community, given the shotgun sequencing data. This probabilistic measure improves our prediction of the actual metabolism in the microbial communities and can be used in the comparative functional analysis of metagenomic data.
| Metagenomics aims to analyze the microbial communities directly extracted from their living environment, bypassing the requirements of isolating and culturing the microbes. With the recent progress of the next generation sequencing (NGS) technologies, the shotgun sequencing of a whole microbial community has become a routine exercise. As a result, the list of metagenomics studies is growing rapidly [1], [2]. This provides ample opportunities for researchers to develop new computational methods to analyze the sequences from metagenomics projects.
To understand the functional and metabolic potential of a microbial community given the sequencing data, a key analysis is to predict - from raw NGS reads or assembled contigs - protein coding genes and their functions. Functional annotations are often achieved by similarity search (using BLASTX [3], or faster tools like BLAT [4] or RAPSearch [5]) against gene families collected in the databases of biological pathways, such as Kyoto Encyclopedia of Genes and Genomes (KEGG) [6], MetaCyc [7], or SEED [8] so that biological pathways can be reconstructed from the predicted functions. Although the principle is the same, different annotation systems may use different practices: for example, the HUMAnN pipeline directly predict gene families and pathways from short sequence reads based on similarity searches [9], while MG-RAST first predicts protein coding region from short reads de novo, and then predicts the functions of the predicted proteins based on similarity searches [10].
Differential functions or biological pathways can be identified by comparing annotations of metagenomes, providing insights into the differences of functionality of various microbial communities [11]–[13]. For example, in recent work, the community-level metabolic networks of the microbiome were constructed from metagenomic data, and both gene-level and network-level topological differences were identified as associated with the host-based environments [14]. For quantitative analysis, the abundances of genes (often measured as the reads counts) need to be normalized according to gene lengths (more reads will be sampled from longer genes), and the quantification of pathways needs to further consider the different sizes of the pathways (i.e., the number of gene families each pathway contains) and the overlaps among different pathways [15], [16].
In this paper, we present a computational method for inferring the functional activities in a metagenome on the basis of the metabolic reactions catalyzed by predicted genes from the dataset, instead of the genes themselves. By directly working on reactions in the context of a global network, our method is immune to the problem of pathway reconstruction caused by the overlaps between pathways - pathways are important for understanding the biological processes, however, their definition can be rather arbitrary, and the overlaps between pathways are artificially created. More importantly, our new method computes the likelihood of each reaction for all potential reactions catalyzed by predicted functions. Clearly, using all potential reactions can lead to an unfaithful estimation of the functionality of a microbial community: functional predictions are noisy and contain mistakes; on the other hand, there are genes that indeed have multiple functions [17], but not all these functions are carried out by the microbial community. Our previous approach MinPath [16], which has been incorporated in HUMAnN [9], improves pathway reconstruction for metagenomes by removing spurious pathways; MinPath, however, does not provide confidence for individual reactions inferred from metagenomic datasets.
We propose a probabilistic approach to estimate the likelihood of each reaction in a metagenome-scale metabolic network given predictions of enzymes. Our method computes the marginal probability of each reaction observed in a collection of randomly sampled subnetworks from the metagenome-scale metabolic network. In these subnetworks, for each annotated gene family, there exists at least one reaction that is carried out by the product of the gene (i.e. the enzyme). However, if the product of a gene is annotated to catalyze multiple reactions, some of these reactions may be excluded from the sampled subnetwork, as long as at least one of these reactions is included. We note that, according to this condition, each sampled subnetwork represents a putative reconstruction of the collective metabolic network of the metagenome, among which we assume the subnetworks containing fewer metabolites are more likely to represent the actual metabolism of the microbial community than the ones containing more metabolites. Based on this parsimony concept, we devised a Markov Chain Monte Carlo algorithm [18], by which we randomly sample a large set of subnetworks and estimate the likelihood of each reaction.
A microbial community adapts its collective metabolic profile to its living environment. Therefore, the similarity measure based on either protein content or metabolic activities in metagenomes can be used to cluster the metagenomes, consistent to similarity of the environments [19]. We applied our method to analyze 44 samples from several metagenomics studies: we used different measures to calculate the similarity of samples, and our results show that the distance measure based on the probability of reactions leads to the most discriminating clustering of the samples. Notably, the functional variations among metagenomes from different environmental niches cannot be fully explained by their differences in taxonomic composition, because the clustering of these metagenomes based on their metabolic taxonomic composition is not as discriminating as our method. We also show detailed comparison of the samples from two ecosystems, to demonstrate that how the probabilities of reactions can help identify important metabolic pathways that reflect the environmental adaptation of the microbial communities.
From the IMG/M metagenome repository [2], we downloaded 44 metagenomic datasets, which were acquired in 10 separate metagenomics studies of different host-associated or environmental ecosystems: human and animal gut, soil, ocean, freshwater and saline lake water (Table S1). All these studies were conducted by using Illumina sequencers with massive amount of reads acquired (short reads data file size ranging from 250 MB to over 200 GB). For each sample, IMG/M provides the assembled metagenome, and the protein-coding genes are characterized with additional functional annotations, such as KEGG ortholog groups of enzymes. Based on these identified KEGG ortholog groups and the KEGG reference metabolic pathways, we constructed a metagenome-scale annotated global metabolic network for each metagenomic sample. Note that these annotated global metabolic networks contain a similar number of multi-functional enzymes (Table S1). For each metabolic network, we applied the MCMC sampling method and computed the marginal probabilities of all annotated reactions. These probabilities can be used to compare the similarity of the microbial communities in the corresponding environments.
We clustered the 44 samples based on different similarity measures of their enzyme contents or metabolic reactions. Five types of measures were used to estimate the distance between the metagenomics samples and compare the clustering results (for details see Methods): 1) the Bray-Curtis dissimilarity that compares the quantities of the metabolic enzymes encoded in each pair of the metagenomics samples; 2) the binary distance between the binary vectors representing the presence/absence of each enzyme in the metagenomes; 3) the binary distance between the binary vectors representing the presence/absence of each reaction in the annotated global metabolic network based on the naive annotation of enzymes; 4) the taxonomic distance based on the phylogenetic composition of the prokaryotes involved in metabolism; and 5) the Euclidean distance that compares the marginal probabilities of the reactions estimated by using the MCMC algorithm.
The hierarchical clustering of the metagenomics samples (for details see Method) using the five distance measures are shown in Figure 1. It is clear that the clusters created by using (Figure 1 (b)) and (Figure 1 (e)) are more consistent with the actual environmental similarities than the ones using the other distance measures. Between these two, the clusters generated using are more accurate because it can separate all metagenomic samples based on their habitats while the other method failed to. Research has shown that lake water microbial communities are highly affected by inoculation of microbes from soils [20], therefore, soil samples and lake water samples are considered to be from similar environments in previous studies [21]. This correlates well with the -based clustering result, in which the soil samples and lake water samples are intermixed in one large cluster (Figure 1 (e)). Figure 1 (d) shows that the taxonomic composition derived from the genes involved in metabolic pathways cannot discriminate the microbial communities to their habitat groups very well, which implies that the functional similarities of the metabolisms cannot be completely attributed to the taxonomic composition of the metagenomes. The poor outcome when (enzyme quantities-based distance, Figure 1 (a)) is used as the distance measure indicates that prudence should be taken when using the enzyme abundances estimated in metagenome assemblies as a measure of metabolism in the microbial communities, even though it is shown to be useful in comparing relative abundance of metabolic functions in addition to binary functional reconstructions [9], [13], [14]. The poor performance of the clustering when the enzyme-coding genes are naively annotated to all the reactions (Figure 1 (c)) confirms our proposition that the naively annotated metabolic network cannot reflect the nature of the metabolic adaptation of the microbial community to its environment. The clustering results show that by computing the likelihood of the reactions occurring in the metabolism, the adaptation of the metabolism that was hidden in the global metabolic network can be revealed. This leads to a more accurate assessment of the functional similarities among the metagenomic samples.
To investigate how the distance measure improved the clustering results compared to , we focused our analysis on the metagenomic samples from two ecosystems: the permafrost soil samples from Alaska, and the saline lake water samples from an Antarctic deep lake. The first group of samples was collected from three different layers (two samples in each layer) in permafrost in the sediment of a creek in Alaska [22]. The second group of lake water samples was collected at six different depths ranging from 5 to 36 meters in Antarctic lakes [23]. Note that when using as the distance measure, 3 saline water samples were incorrectly grouped into the cluster of permafrost samples; but the distance measure can accurately separate the samples from the two environments (Figure 1, (c) and (e)).
We used statistical tests to assess whether a metabolic reaction is differentially likely to occur in the two environments. Because the presence/absence of the reactions in the metabolic network is represented by a binary vector, we identified 110 reactions using the Fisher's exact test [24] that are statistically different (P-value0.05) in the two groups (for details see Methods), indicating these reactions are likely to occur only in one of the two environments. We then used t-test to check if the marginal probability of each metabolic reaction is significantly different between the two groups. The t-test identified 447 reactions showing statistically different likelihoods to occur in the two environments (Table S2). The two tests agree on 109 reactions, and 338 reactions are considered to be different between these two sets of samples only by the t-test (Figure 2) on the likelihood of reactions, whereas only one reaction is detected as significantly different only by the Fisher's exact test.
Note that 166 of these 338 reactions are annotated to be catalyzed by one or more catalytically promiscuous enzymes in all of the metabolic networks (Table S2). In other words, there is no difference if we compare whether they exist (based on the annotation of genes) in the metabolic networks in both groups. However, the marginal probabilities of these reactions, which were assigned by the MCMC algorithm, are different among the two groups of samples, indicating these reactions show different likelihood to occur in the metabolism of the samples between the two groups. For example, one reaction in the Benzoate degradation pathway (KEGG reaction R06989) is observed in all 12 metabolic networks with different likelihoods in both environments; the reaction is on average 2.5-fold more likely to occur in the permafrost samples than in the lake water samples, if we compare their marginal probabilities (Table S2). This reaction is catalyzed by the enzyme muconate cycloisomerase (KEGG ortholog K01856), a promiscuous enzyme that also catalyzes four other reactions (Figure 3). All five reactions involve the isomerization of cis,cis-muconate and its derivatives (Figure S1). In particular, the reaction R06989, which is an important step in benzoate degradation, transforms cis,cis-muconate, which is enzymatically produced from catechol. The functions of benzoate and catechol metabolism are also found to be enriched in the permafrost microbial communities by other studies [25]. The results of the MCMC simulation show that the differences of the marginal probabilities of the other four reactions are much smaller compared to R06989. Also note that the probabilities of the five reactions are almost the same in the Antarctic lake samples, whereas, in the Alaska permafrost samples, the reaction that isomerize cis,cis-muconate (R06989) apparently has greater probabilities (Figure 3). This shows how the results of our method can be used to analyze the potential adaptation of the functions of promiscuous enzymes in different environments, which cannot be revealed when analyzing only the enzyme-encoding genes.
Another interesting observation is that the difference between the marginal probabilities of the 338 reactions can be used to correctly cluster the samples into two groups (Figure 2). In addition, if we focus on the samples extracted from Alaska permafrost, all 6 samples are correctly separated into three clusters, with each containing two samples from the same layer in the permafrost. This shows that those reactions only identified by comparing the estimated marginal probabilities contain the critical information in the metabolic adaptations of these samples to their environments.
Among those 447 reactions that are identified to be different by the t-test on the marginal probability, 327 reactions show higher marginal probabilities to appear in the Alaska permafrost samples than in Antarctic deep lake samples. The remaining 120 reactions show lower probabilities in the Alaska permafrost samples than the Antarctic deep lake samples (Table S2). We built two networks using these two sets of reactions. In these networks, vertices represent the reactions, and a pair of reactions is connected by an edge if there are one or more common metabolites in the two reactions(Figure 4). The connected components in these networks represent chains of metabolic reactions that can be considered to have significant higher probabilities to occur in the environment of one group compared to the other.
Several interesting chains of reactions were revealed in both networks. For example, the chain R07916-R04786-R04787 (R07916, R04786, R04787 are KEGG reaction IDs) has a higher probability to occur in the metabolism of microbial communities from Antarctic deep lakes (Figure 5), which is a part of the beta-carotene biosynthesis module belonging to the carotenoid biosynthesis pathway. Carotenoids are essential metabolites for photosynthetic bacterial because they provide photo-protection and accessory light harvesting [26]. The bacterial community in fresh water is known to carry out photosynthetic activities even in deep water. The Antarctic deep lake metagenomics study also revealed trace of photosynthetic microorganisms in their samples [23]. Therefore, it is not surprising to observe that photosynthesis related pathway modules have a higher likelihood to occur in the deep lake microbial communities than in the permafrost soil samples, which exist in an environment deprived of light. Several chains of reactions were found to have higher probabilities to occur in permafrost samples, among which were chains in methanogenesis, and keratan sulfate degradation. Slow rates of methanogenesis by cold-adapted methanogens occur in permafrost and active layer soils [22], and keratan is regarded as a carbon source for certain bacteria isolated from soil [27]. Note that these chains all contain reactions that are identified only by comparing the reaction marginal probabilities but not the enzymes or the existence of reactions. Some chains even contain only reactions that are identified by comparing the reaction marginal probabilities (Figure 4). Therefore, our method successfully expands the horizon of discovering important pathways that contain critical information of the adaptive metabolism of microbes.
In this paper, we focus on the analysis of metagenomics samples based on the metabolic reactions annotated to be catalyzed by the predicted genes in the metagenomes. We proposed a method that assigns marginal probabilities to reactions to estimate the likelihood of the reactions to occur in the metabolism of a microbial community. The Markov Chain Monte Carlo (MCMC) sampling algorithm establishes a framework that can be used to study the different aspects of metabolic networks. The subnetwork universe can be sampled by the MCMC algorithm with different constraints based on various assumptions. For example, in recent work, flux balance analysis (FBA) is used to constrain the viability of the sampled genome-scale metabolic networks in a MCMC based method [28].
The marginal probabilities are assigned by our method to reactions that are catalyzed by catalytically promiscuous enzymes. As shown in the results, the reactions occurring in a microbial community are a better representation of the metabolism in the community, because one reaction may be catalyzed by different enzymes encoded by different microbial organisms in the community. An extension of this method is to compare the likelihood of the reactions to be catalyzed by the same enzyme and allow us to investigate how promiscuous enzymes function in different environments.
By applying the parsimony assumption, our method successfully takes several intrinsic properties of the metabolic network into consideration. It should be noted that this method indirectly favors highly connected metabolic networks, where the number of non-terminal metabolites that can be produced and consumed by the microbial community is maximized. Similar assumptions have been used in other studies. For example, in the metabolic network reconstruction, gaps in metabolic paths are usually filled to decrease the number of isolated reactions or metabolites [29]. Notably, this assumption is particularly practical for the study of the metabolism in a microbial community rather than individual microbial organisms, because the microbes living in the same environment likely co-evolved into a condition under which microbes can make use of the metabolites from other microbes and only a small number of metabolites are required from the external environment by the whole community.
We note that our method lacks the compartmentalization of the biochemical reactions and the resolution of individual species. In previous studies, multi-species models have been applied to investigate the interactions within the community or between host and microbiomes [30]. In comparison, when studying the system-level behavior of the whole microbial communities, researchers often treat the microbial communities as individual adaptive organisms (also referred as supra-organism), ignoring the boundaries between species altogether [14], [31], [32]. In this study, we take a similar approach, which allows us to investigate the collective metabolic behavior of the microbial communities. This approach is also a necessity because genomic information is not available for all the species in the community and methods for decomposing complex metagenomic samples into compartmentalized organelles/prokaryotic cells are yet to be developed.
There are, however, several limitations of the method that are worth noting. In our metabolic network definition, the reactions are considered to be indirect, which indicate that all reactions are reversible. However, conditions in the cell are often such that it is thermodynamically infeasible for flux of reactions to flow in certain direction so the reaction becomes irreversible. Therefore, there might be dead-end metabolites in the network, the metabolites that are not the product of the other reactions or are not used by other reactions as substrates. In our model, they are misinterpreted as the metabolites that connect two reactions, which could decrease the accuracy of the reaction and pathway annotated by our method. Another issue is that this method does not consider the abundance of enzymes predicted from the metagenomic sequences. We observed large variance in the abundance of the same enzyme among samples as well as pairs of enzymes that share same metabolites as substrates or products. We are working on a revision of our current method to take the abundance of enzymes into account.
We start the problem formulation with a formal definition of a metabolic network.
Before we discuss the details of the MCMC algorithm, we introduce an inequality of likelihood of the metabolic subnetworks based on a parsimony assumption. We observe that within microbial communities, metabolic enzymes work collectively to catalyze a series of reactions to transform some compounds that are available in their living environment into other compounds that can be utilized to maintain their cellular functions. Although the enzymes act on different substrates and products, the products of some reactions are usually used as substrates in the subsequent reactions, resulting in a sequence of reactions devising the complex but efficient metabolic network. The adaption of a living microbial organism to its living environment often leads to a unique and nearly optimal metabolic network, in which only a small number of necessary compounds need to be taken from its environment, while the other compounds can be synthesized through metabolic reactions inside the microbial community. Based on these observations, we adopt a parsimony assumption: the valid subnetworks involving fewer compounds are more likely to represent the metabolism of the microbial community.
Equivalently, given two valid subnetworks and of the same annotated global network , where and , ifthen(2)
We construct a Markov Chain (MC) of annotated subnetworks to sample the valid annotated metabolic subnetwork in the universe and estimate the marginal probabilities of reactions. At each step of the random MC walk, a new metabolic subnetwork is generated by inserting a new reaction to the current subnetwork, or by deleting an existing reaction from the current subnetwork. We repeat the insertion/deletion until a valid new subnetwork is generated. The transition from the current subnetwork to a new valid subnetwork is accepted or rejected based on the parsimony inequality in equation (2). If the number of metabolites found in the new subnetwork is smaller than the number of metabolites in the current subnetwork, we accept the transition to the new subnetwork; otherwise, we accept the transition with a probability (called the candidate probability), where is the difference between the number of metabolites in the new and the current subnetwork.
It is straightforward to show that this type of transition is ergodic, i.e., any pair of subnetworks can be connected by a finite series of such transitions. The candidate probability ensures that the random walk samples the subnetworks based on the parsimony assumption, as defined in equation (2). According to the Metropolis-Hastings rule [33], the candidate probability also ensures the subnetwork samples are drawn from a probability distribution that is proportional to the likelihood of the subnetwork in the subnetwork universe .
The number of variables we are trying to estimate equals the number of reactions that are catalyzed by promiscuous enzymes. Metabolic networks constructed from metagenomic data normally contain hundreds of such reactions. Here we discuss several methodological issues of the Markov chain caused by the large number of variables in the sampling universe. First, every step in the random walk only changes the state of at most one reaction, therefore, the correlation between the two consecutive sampled subnetworks is obviously large. We use the subsampling (also called batch sampling) technique to reduce the correlation and to approximate independence between the successive samples of the Markov chain [18]. We subsample from the Markov chain with a deterministic batch size , meaning that we consider only one subnetwork from every sampled subnetworks. As shown in Figure 6 (a), when for a network containing 362 promiscuous enzymes (1183 enzymes in total), subsampling almost completely eliminates the correlation of successive sampled subnetworks. Another issue is the acceptance rate of the transitions in the Markov chain, which is the ratio of the number of accepted transitions from the current subnetwork to the new subnetwork, over the number of proposed transitions. This ratio was shown to affect the convergence pattern of the Markov chain [18], and it's heuristically recommended to be controlled to close to 1/4 for models with high dimensions [34]. The acceptance rate is affected by the candidate distribution, which in our model has the exponential distribution form , where is the difference between the number of metabolites in the new subnetwork and the current subnetwork. We choose the candidate distribution in this form because it can restrict the acceptance rate of our random walk approximately within the range of 0.20 to 0.24. Note that because we only sample the valid subnetworks, the rejected invalid subnetworks in our algorithm (Figure 7) are not counted when calculating the acceptance rate. Furthermore, it requires a large number of samples from the subnetwork universe for an accurate estimation of the marginal probability. This, in addition to the requirement of subsampling to avoid high correlation between the samples, requires a careful monitoring of the convergence of the Markov chain. We examine the ergodic average of the estimated marginal probability for all reaction in a metabolic network of 609 reactions that are catalyzed by multi-functional enzymes, and heuristically determine that with subsampling of batch size , the Markov chain converges after 10,000 subnetworks are sampled (Figure 6 (b)). To improve the convergence of our Markov chain, in practice, we choose to discard samples from the first 10 million steps of the Markov chain, i.e., the burn-in period, to make the random walk to start from a better point in the subnetwork probability space . Last but not least, because of the requirements of large batch size in subsampling, low acceptance rate, and large amount of samples, the running time of the Markov chain requires special examination. We find that when we control all these factors, the running time is linearly correlated to the number of reactions in the subnetwork, or in some sense the size of the subnetwork (Figure 6 (c)). For a large network with over 1,200 reactions, of which more than 600 are catalyzed by multi-functional enzymes, the algorithm can finish 250 million iterations in approximately 10 hours.
The formal Metropolis-Hastings algorithm is shown in Figure 7.
After sampling, we can compute the estimated marginal probability by(3)for such that , where is the total number of sampled valid subnetworks.
In addition to the marginal probability , we can also extract a subnetwork sample with the maximum likelihood in the samples, which is the subnetwork with the smallest number of compounds.
For any pair of metagenomes , the Bray-Curtis dissimilarity is , where is the sum of the lesser value for only those enzymes in common in both samples. and are the total number of enzymes counted in both samples [35]. In this paper, this distance is denoted as . The quantities of enzymes in each metagenomic dataset were obtained from IMG/M, computed based on the number of assembled contigs aligned to each family of enzymes. The Jaccard distance is used to estimate the binary distance of the samples, based on enzymes (denoted as ) and reactions (denoted as ). We also computed the Euclidean distance of reactions , which is based on the marginal probabilities of the reactions.
The taxonomic distances, denoted as were calculated in several steps: for each metagenome downloaded from IMG/M, to ensure the comparison is based on the organisms involved in the collective metabolic processes, we removed the genes that are not annotated with a metabolic function in the IMG/M KEGG ortholog group annotations. Then we used BLAST (version 2.1.18) to search the genes against the KEGG genes database (E-value ). From the BLAST results, we built a phylogenetic tree by gathering the genome of the top hits and mapping them to the Greengenes core set [36]. Using the phylogenetic trees, we calculated the pairwise taxonomic distances between samples with unweighted Fast UniFrac method [37], [38] (PyCogent [39] version 1.5.3), a metric that measures the phylogenetic relatedness of whole communities and has been widely applied in studies to compare taxonomic differences between complex microbial communities [40].
We used the Ward's minimum variance method as the linkage criteria in our hierarchical clustering, which tries to minimize the total within-cluster variance. Note that when applying other common linkage criteria in the hierarchical clustering, even though the performance varies, the order of performances using the four distance measure was still observed.
Fisher's exact test [24] is used to determine if there are nonrandom associations between two binary variables. For each reaction, we used Fisher's test to compare its presence/absence in the two groups of samples from the two environments. The p-value gives the exact probability of observing the particular ratio of the presence/absence of the reaction in the samples from the two environments, on the null hypothesis that the chances of the reaction to exist in both environments are the same. We consider reactions with p-value0.05 as ones that have significant different probability to be found in the two environments. We also used Fisher's test in comparing the binary representation of each enzyme in the two environments.
We used the independent two-sample t-test to determine whether the quantities of enzymes are statistically different in two environments under the assumptions that these quantities are independent and normally distributed, and the distributions of the quantities in the two groups of samples have the same variance. So for each enzyme, we consider its quantities in samples in the two environments as two groups of values, and the t-statistic determines whether the means of the two groups of values are different. Similarly, we also used t-test for analyzing whether each reaction has a different probability in the two environments.
The MCMC sampler of metabolic networks is implemented in Java, and based on the HYDRA MCMC library [41]. Running time of the program is dependent on the size of the network and the configuration of the MCMC sampler, including the subsampling size and burn-in period (Figure 6). For a large network with about 2,500 reactions, the sampling takes approximately 10 hours on a dual core Dell Latitude laptop, and requires a small amount of memory (50 MB). Source codes can be downloaded at http://omics.informatics.indiana.edu/mg/MetaNetSam/.
|
10.1371/journal.pbio.1001860 | RNA Mimicry by the Fap7 Adenylate Kinase in Ribosome Biogenesis | During biogenesis of the 40S and 60S ribosomal subunits, the pre-40S particles are exported to the cytoplasm prior to final cleavage of the 20S pre-rRNA to mature 18S rRNA. Amongst the factors involved in this maturation step, Fap7 is unusual, as it both interacts with ribosomal protein Rps14 and harbors adenylate kinase activity, a function not usually associated with ribonucleoprotein assembly. Human hFap7 also regulates Cajal body assembly and cell cycle progression via the p53–MDM2 pathway. This work presents the functional and structural characterization of the Fap7–Rps14 complex. We report that Fap7 association blocks the RNA binding surface of Rps14 and, conversely, Rps14 binding inhibits adenylate kinase activity of Fap7. In addition, the affinity of Fap7 for Rps14 is higher with bound ADP, whereas ATP hydrolysis dissociates the complex. These results suggest that Fap7 chaperones Rps14 assembly into pre-40S particles via RNA mimicry in an ATP-dependent manner. Incorporation of Rps14 by Fap7 leads to a structural rearrangement of the platform domain necessary for the pre-rRNA to acquire a cleavage competent conformation.
| Ribosomes are the cellular machines responsible for all protein synthesis. In eukaryotes, the assembly of ribosomes from their protein and RNA components is extremely complicated and involves more than 200 nonribosomal factors—three times the number of proteins in the mature complex. Among these factors, the Fap7 protein is particularly intriguing because it interacts with the small subunit ribosomal protein Rps14 and it exhibits adenylate kinase activity—a molecular function more commonly associated with regulating ATP/ADP levels than assembling protein–RNA complexes. Combining structural and biochemical analysis of the Rps14–Fap7 complex, we show that Fap7 uses protein side chains to mimic RNA contacts, rendering the interaction of Rps14 with ribosomal RNA or with Fap7 competitive and mutually exclusive. Once bound, Rps14 blocks the substrate-binding cavity of Fap7, and ATP hydrolysis will then break the Fap7–Rps14 complex apart. At the same time, the ribosome structure at the location where Rps14 binds is disrupted when the Fap7/Rps14 complex is formed, and this process is regulated by ATP binding and hydrolysis. Our model is thus that Fap7 temporarily removes Rps14 from the ribosome to enable a conformational change of the ribosomal RNA that is needed for the final maturation step of the small ribosomal subunit.
| Over 200 preribosomal factors are involved in the maturation of ribosomes. Most of these factors are essential for cell survival, but their precise molecular functions remain elusive (for reviews, see [1]–[3]). One of the last steps of maturation of the small subunit of the ribosome is the cytoplasmic cleavage of the 20S pre-rRNA at site D to generate 18S rRNA. This cleavage is carried out by the endonuclease Nob1 in 80S-like complexes composed of pre-40S particles and mature 60S [4],[5]. Cleavage also requires multiple pre-40S factors including the Nob1 binding protein Pno1/Dim2, the methyltransferase Dim1, the export factors Enp1 and Ltv1, and several NTPases including the Rio1 and Rio2 protein kinases, the Prp43 helicase and its cofactor Pfa1, the GTPase-related factor Tsr1, and the Fap7 NTPase.
The locations of maturation factors in the late pre-40S particles is emerging from in vivo RNA binding (CRAC), cryo-EM, and crystallographic studies on preribosomal particles [6]–[10], but detailed understanding of their functions remains limited. Amongst late pre-40S factors, the function of Fap7 is especially intriguing. Human hFap7 (also called hCINAP or AK6) bears structural homology to adenylate kinases (AKs) and harbors a broad AK activity [11],[12]. AKs catalyze the reversible transfer of the γ phosphate of adenosine nucleotide triphosphate (ATP) to an adenosine mono-phosphate (AMP), forming two molecules of adenosine di-phosphate (ADP). AKs play important roles in nucleotide metabolism [13], but the link between this enzymatic activity and ribonucleoprotein (RNP) assembly is enigmatic.
In yeast, yFap7 is necessary for the late cytoplasmic maturation steps of the 40S particle, and strictly required for the cleavage at site D [5],[14]. However, the association of yFap7 with the pre-40S particles is either weak or very transient [5],[14]. In the absence of yFap7, pre-40S subunits accumulate in 80S-like particles, containing 20S pre-rRNA. An active site mutant in yFap7 has the same phenotype, demonstrating a requirement for AK activity [5],[14]. The catalytic activity of hFap7 is also important in the assembly and/or stability of Cajal bodies [12],[15],[16], nuclear domains involved in the maturation of small nuclear RNP (snRNP) particles [17].
Fap7 forms a complex with Rps14 that is conserved between humans, yeast, and archea [14],[18],[19]. Rps14 is a component of the platform domain of the small subunit, and its C terminus forms an extended structure rich in basic residues, which binds the rRNA close to Helix 45 and site D. This C-terminal extension is essential for D site cleavage, and point mutations in yRps14 show effects on ribosome biogenesis similar to yFap7 depletion [14],[20]. Human hRps14 has at least two links to ribosomopathies and to cancer. Haploinsufficiency of hRps14 is a causal factor in myelodysplastic syndrome (MDS) and 5q syndrome, a genetic disorder related to Diamond Blackfan anemia that leads to severe anaemia, macrocytosis, and an increased risk of leukaemia [21],[22]. Additionally, hRps14 regulates the MDM2–p53 pathway by directly binding the acidic domain of MDM2 [23]. This interaction prevents MDM2 from targeting p53 for degradation, resulting in p53 activation and cell cycle arrest. This activity links deregulation of ribosome biogenesis in response to nucleolar stress to the inhibition of cell cycle progression. hFap7 is also an essential regulator of the hRps14–HDM2–p53 pathway by affecting the interaction between hRps14 and HDM2 [18].
To decipher the function of Fap7 in ribosome biogenesis, we functionally and structurally characterized the Fap7 protein alone and in complex with Rps14 and nucleotides. The combination of structural studies, enzymatic assays, RNA binding studies, and in vitro D site cleavage assays on purified preribosomes uncovered the function of Fap7 within pre-40S ribosomes.
When expressed alone, yRps14 showed poor solubility. We therefore designed polycistronic vectors to co-express Fap7 and Rps14 from the yeast Saccharomyces cerevisiae (yFap7/yRps14) and the archaeon Pyrococcus abyssi (aFap7/aRps14). A His6 affinity tag was fused to the N terminus of Fap7 for a first purification step by immobilized Nickel metal ion affinity chromatography (Ni-IMAC). Excess-free Fap7 was separated from the Fap7–Rps14 complex by gel filtration. Two crystal forms were obtained for the archaeal aFap7–aRps14 complex bound to ADP/Mg2+ and ATP/Mg2+, which diffracted to 2.1 Å and 2.4 Å, respectively. Phasing was performed by single anomalous diffraction on the ADP/Mg2+ crystal form with platinum-derived crystals diffracting to 4.0 Å (see phasing and refinement statistics in Table 1). Four copies of the Fap7–Rps14 complex are present in the asymmetric unit in both space groups, revealing a 1∶1 complex of aFap7 with aRps14 (Figure 1A).
The structure of bound aFap7 is similar to the previously reported human homologue hFap7, with 2.1 Å r.m.s.d. for 30% identity (Figure 1B, alignment in Figure S1A). All structural elements of the CORE domain (cyan in Figure 1A) adopt the canonical fold common to AKs [24]. aFap7 also contains the specific structures of the Fap7 AK family in the LID (residues 89–117) and NMP (residues 41–51) domains, which are involved in binding ATP and AMP, respectively (Figure 1A). The same numbering scheme for the secondary structure elements described for hFap7 has therefore been used (Ren et al. 2005) [11]. Complex formation buries 1,800 Å2 of aFap7, representing 18% of the total surface, with interactions over the LID and NMP domains, the C-terminal extension, and residues of the walker B motif (Figure 1A, protein contact map in Figure S2). aRps14 shares 60% sequence identity with yRps14 and can be superimposed on ribosome-bound yRps14 with 1.3 Å r.m.s.d. (Figure 1C, alignment in Figure S1B) [25]. The binding surface on aRps14 is largely comprised of the surface of the β-sheet in the globular core domain (gold in Figure 1A) and the basic residue rich C-terminal extension (Rps14-CE; red in Figure 1A, protein contact map in Figure S2). Comparison of the aFap7–aRps14 complex with ribosome-bound yRps14 did not reveal structural changes in the core domain, with the exception of the β4–α2 loop, whose RNA-bound conformation would clash with aFap7. In contrast, the aRps14-CE is completely remodeled by the interaction with aFap7 (Figure 1C). This large conformational change is possible because of the intrinsically disordered nature of the C-terminal extension. The protein contact map (Figure S2) shows that the interface can be roughly dived into three zones, representing the interaction of the core domains of Fap7 and Rps14, of the Rps14-CE with the aFap7-core, and of the Fap7-CE with the Rps14-core, which suggests that the C-terminal extensions of the two proteins play an important role in the complex.
Because of the high sequence identity between the archaeal and yeast proteins, the structure of the aFap7–aRps14 complex presents a good model for the yeast yFap7–yRps14 complex. However, a notable difference concerns the C-terminal extension of aFap7 (dark blue in Figure 1), which is not conserved in yeast and humans in length or sequence. In the aFap7–aRps14 complex, the C-terminal extension folds back onto the globular domain and contains an extra 3–10 helix (η8, dark blue in Figure 1), which packs both into the groove formed by the α5–α6 helices of the LID domain (light blue in Figure 1) on the β-sheet surface of aRps14 (see protein contact map in Figure S2). It contributes 500 Å2 of interaction surface with aRps14, including nine hydrogen bonds and one salt bridge. The η8 helix is present in all copies of the complex present in the asymmetric unit, in both ADP and ATP bound forms, and in exactly the same conformation. All these facts seem to indicate that the η8 helix is an integral part of the aFap7–aRps14 complex.
In order to further confirm that the interaction of the η8 helix with Rps14 is not an artifact of crystal packing, we have conducted structural studies in solution by small angle X-ray scattering (SAXS) on the aFap7–aRps14 complex. The resulting SAXS curves were fitted with a model built using our X-ray structure, with flexibility introduced in all missing N-terminal and C-terminal residues and poorly defined loops. The resulting fit shows a good agreement with the crystal structure (Table 2), showing that the η8 helix is located at the aFap7–aRps14 interface in solution (Figure 2A).
In order to both confirm the structure in solution of yeast Fap7–Rps14 complex and determine the contribution of the nonconserved C-terminal extension of yFap7, we conducted SAXS studies of yFap7 both free and bound to yRps14. The modeling of yFap7 was performed by introducing flexibility on the N-terminal residues of the tag, the LID domains, the β3–β4 loop, and the entire nonconserved C-terminal extension following the α7 helix, and on the β4–α2 loop in yRps14. The yFap7 terminal extension was modeled in the aFap7 conformation bound to aRps14, stacked on the NMP binding domain. The SAXS data recorded on the free yFap7 clearly indicates that the yFap7 structure contains unstructured regions, as the experimentally determined radius of gyration (Rg) of free yFap7 is comparable to that of the Fap7–Rps14 complex (from both yeast and archaea; Table 2). The modeling of yFap7 with these data clearly shows that a conformation of the C terminus packed on the globular domain is not compatible with the SAXS data (Figure 2C). The most representative structure fits well to the data and shows that the C-terminal extension is unstructured and does not form any contacts with the yFap7 core and NMP domains (Figure 2C). Because the initial conformation was chosen to “favor” the compact conformation during the modeling of the SAXS data, the unwinding of the C-terminal region in order to fit the SAXS data cannot be attributed to an artifact of the modelization protocol.
Modeling of the structure of the yFap7–yRps14 complex against the SAXS data shows that the complex adopts the same overall structure as its archaeal homologue, with both the same binding surfaces involved and the same orientation between the yFap7 and yRps14 proteins (Figure 2B). The archaeal complex can therefore be used with confidence to model the yeast complex. Surprisingly, the data suggest that the yFap7 C-terminal extension comprising helix η8 adopts the same conformation as the archaeal complex. Firstly, the Rg measured from the SAXS data recorded on the yFap7–yRps14 complex is comparable to that of the archaeal complex, indicating the absence of additional unstructured regions in the yeast complex. In addition, even though the conformation of the C-terminal extension was considered as flexible in the structure calculations, it remains in the same conformation as the aFap7 C-terminal extension in all calculated structures (Figure 2B). This indicates that despite the poor conservation of this region, the yFap7 C terminus participates in yRps14 binding in the same region as the archaeal complex. In agreement with this model, we were not able to purify measurable quantities of yFap7ΔC–yRps14 complex from a polycistronic vector expressing a yFap7 protein truncated of the last 20 residues (unpublished data), although this could be due to insolubility of yFap7ΔC or a poorer expression of yRps14 in this construct. In the archaeal system, truncation of the aFap7 C terminus after helix α7 (residue 157) had no effect on complex formation with aRps14 in our gel filtration assay (Figure 3A,B). However, performing the assay with aRps14ΔC, which shows a reduced but measurable affinity for aFap7 (Figure 3C), leads to a complete dissociation of the complex (Figure 3D). This indicates that the interaction between the Fap7 C terminus and Rps14 extends the binding surface and strengthens the binding affinity of the complex, but is not essential for the formation of the archaeal complex in vitro.
The structures obtained for aFap7–aRps14 in complex with ADP and ATP show subtle conformational differences. Both nucleotides bind in the aFap7 ATP binding pocket. The ADP-bound state of Rps14-bound aFap7 superposes perfectly with ADP-bound hFap7 for residues of the P-loop and LID domains in the ATP binding pocket (unpublished data). It therefore seems that, in the ADP-bound form, the binding of aRps14 does not lead to major structural changes in the aFap7 LID domain.
In the ATP-bound crystal form, two copies of aFap7 are bound by ATP and two others by ADP. The structures of the ATP- and ADP-bound complexes showed only small, local conformational differences in the actual ATP binding site. The base, ribose, α, and β phosphates of ATP superpose perfectly on those of ADP, and the residues of the P-loop are in the same conformation in both structures (Figure 4A). However, the presence of the γ phosphate of ATP is not compatible with the position of Arg100 (LID domain) in the ADP-bound state (Figure 4A). This arginine, involved in binding the α and β phosphate oxygen atoms in the ADP-bound structure, bridges the α and γ ATP phosphate oxygens. Compared to the ADP-bound structure, the γ phosphate pushes Arg100 away by approximately 1 Å. This effectively distorts the structure of the α5–α6 loop of the LID domain with a 0.9 Å r.m.s.d. compared to the ADP-bound conformation for residues 89–107 (Figure 4A). Interestingly, the conformational change in the LID domain modifies the positions of Tyr102 and Lys106 side chains, which interact with the Rps14-CE Asp124. In the Apo-hFap7 structure, Arg100 would clash with the ADP phosphates, and the LID domain is even more distorted (Figure 4A). This suggests that the LID domain undergoes major conformational rearrangements upon nucleotide binding and/or release and that Arg100 could sense the ATP binding pocket occupation state and relay this signal to the Rps14 binding interface through the LID domain residues Tyr102 and Lys106.
The AMP binding site on Fap7 homologues has not been experimentally determined, but comparative modeling with related AKs suggests that AMP is bound in the cavity formed by the LID and NMP domains [12]. For illustration, Figure 4B presents the structural model of aFap7 bound to diadenosine pentaphosphate (Ap5A), built by superposing aFap7 with human AK1A bound to Ap5A (Protein Structure Comparison Service Fold at European Bioinformatics Institute, http://www.ebi.ac.uk/msd-srv/ssm, authored by E. Krissinel and K. Henrick). Ap5A is an AK inhibitor that mimics a transition state during phosphate transfer from ATP to AMP. Within Ap5A, one adenosine and the α, β, and γ phosphates superpose perfectly with the bound ATP in the ATP binding site, while the remainder of the molecule matches the AMP substrate binding site between the Walker B, helix α2, LID, and NMP domains.
In the aFap7–Rps14 structure, the C-terminal residues of aRps14 (aRps14-CE) after the final β5 strand forms an unusual interaction with Fap7 that occludes the putative AMP binding site (Figure 4B). The Rps14-CE completely fills the cavity lined by residues from the LID, NMP binding, and Walker B domains, and would preclude nucleotide binding. Specifically, the ribose and base of AMP would sterically clash with aRps14 Gly125 and Thr126 residues (Figure 4B). In the aFap7–aRps14 structure, AMP is therefore excluded from the AMP binding site by aRps14.
The Rps14-CE forms a lasso-like structure, which creates an extensive interaction surface, contributing 800 Å2 (45% of the total binding surface) (Figure 4C). In order to confirm the importance of this interaction, we performed protein interaction assays analyzed by gel filtration, using aFap7, aRps14, and an aRps14ΔC construct truncated at residue 117. The aFap7–aRps14 complex could be reconstituted by mixing equimolar amounts of proteins, and this complex migrated as a single peak (Figure 3A). The aFap7–aRps14ΔC sample showed a much lower recovery of the complex, showing that the aFap7–aRps14 interaction was compromised when the aRps14-CE is removed (Figure 3C).
The interaction between the aRps14-CE and the aFap7-NMP domain involves hydrogen bonds between the main chain atoms of aRps14(Arg127/Lys129) and aFap7(Val48/Val50). The aRps14-CE contains six lysine and arginine residues and its interaction with aFap7 involves salt bridges to five conserved basic residues (Figure 4C). Salt bridges are found from acidic aRps14 residues to basic aFap7 residues from the LID domain (Glu109–Arg132), the NMP binding domain (Glu47 and Glu49–Arg127, Glu51–Lys129), and helix α6′ of the CORE domain of aFap7 (Asp122–Arg135, Asp125–Arg136). An intramolecular salt bridge is formed between Arg133 (aRps14-CE) and Asp117 (Rps14 core domain). The only acidic residue in aRps14-CE, Asp124, also forms a salt bridge to Lys106 in the aFap7 LID domain. yRps14–Asp124 does not make any interactions in ribosome-bound yRps14, so its high conservation suggests a major role in Fap7 binding during ribosome biogenesis. All residues involved in these salt bridges are conserved in the yeast and human proteins and found at equivalent positions in the hFap7 structure.
To experimentally verify that Rps14 competes for the AMP binding site of Fap7, we measured the AK activity of free yFap7 and the yFap7–yRps14 complex. As expected, yFap7 has an AK activity with Km and kcat values comparable to those reported for hFap7 (52 µM and 6.12×10−3 s−1, respectively; Figure 4D). In comparison, the yFap7–yRps14 complex showed almost complete inhibition of the Fap7 AK activity (Figure 4D), with a 94% reduction in enzymatic efficiency (kcat/Km). This indicates that yRps14 acts as a competitive inhibitor by blocking the AMP binding site and inhibits yFap7 AK activity. In agreement with these results, we found the same AK activity for the aFap7–aRps14ΔC complex and for free aFap7 (Figure S4A). However, because aRps14 ΔC has a lower binding affinity with aFap7, we cannot exclude that this observation of the measured activity arises from loss of interaction with aFap7.
Because Rps14 is a structural constituent of the ribosome, we compared the structure of yRps14 in complex with the ribosome or in complex with Fap7. In the context of the ribosome, yRps14 is positioned in the platform domain and buries 2,000 Å2 upon binding 18S RNA (Figure 5A). yRps14 binds the stem of helix 23 (H23) of 18S rRNA through the surface of the β-sheet on the β1 to β4 strands, the β1–β2, β2–β3, and β4–α2 loops, whereas the loop of rRNA helix 24 (H24) is bound by the β3–α1 loop (Figure 5B). In the ribosome, the Rps14-CE is totally buried and involved in extensive protein–RNA interactions, as frequently found for the charged C-terminal extensions of ribosomal proteins [26]. The six conserved, basic residues in the Rps14-CE (out of seven total) form salt bridges to phosphates of helix 24 and helix 45 (H24 and H45 in Figure 5B).
Surprisingly, the surface of Rps14 involved in rRNA binding corresponds closely to the interaction surface with Fap7 in the aFap7–aRps14 complex (Figure S3). The Rps14 interface involves 45 interfacing residues in the aFap7–aRps14 complex (green in Figure S3, left) and 50 in the complex with the 18S rRNA (purple in Figure S3, right). Comparison of these residues shows that 35 residues (over 70% of the total) are common between the two interfaces. All of the Rps14 structural elements involved in RNA binding are also involved in protein interactions, except for residues of the β3–α1 loop.
Closer inspection revealed that the specific protein–RNA contacts formed by Rps14 in the ribosome are replaced by contacts to protein side chains in the aFap7–aRps14 complex. For example, in ribosomes the globular domain of yRps14 binds the 18S rRNA via hydrogen bonding with Asn24 and salt bridges with Arg41, Arg84, and Lys90, whereas in the aFap7–aRps14 complex these residues contact aFap7–Asp117, Asp169, Glu112, and Asp125, respectively. This correspondence is even more dramatic for the Rps14-CE, where all six salt bridges to RNA are replaced by salt bridges involving protein side chains (Figures 4C and 5B). Fap7 therefore acts as an RNA mimic, which achieves specific protein–protein contacts by mimicking protein–RNA contacts.
To experimentally verify that Fap7 competes with RNA for Rps14 binding, we performed filter binding assays using a radioactively labeled 45 nt RNA oligomer corresponding to Helix 23 of the 18S rRNA from S. cerevisiae (residues 884–928, Figure 5C and Figure S4C,D,E,F,G) to mimic the binding site of Rps14. MBP-yRps14 was competent in binding this construct with an apparent affinity of 1.50±0.15 µM (Figure 5D, red), whereas yFap7 did not show detectable binding (Figure 5D, green). The yFap7–yRps14 complex was severely impaired in RNA binding (Figure 5D, blue), with a measured affinity of 14.8±4.5 µM. This confirms that in vitro Fap7 binding to Rps14 effectively occludes the RNA binding interface and can prevent Rps14 from binding a synthetic RNA construct. The consequence of this finding is that during ribosome maturation the structure of the platform domain must undergo a large change in protein and/or rRNA conformation to accommodate formation of the Fap7–Rps14 complex.
Depletion of yFap7 leads to accumulation of pre-80S–like complexes, composed of pre-40S particles associated with mature 60S subunits, but three components of the platform domain—Rps14, Rps26, and Rps1—are underrepresented in these particles [5]. To assess whether pre-80S complexes purified from yFap7-depleted cells represent ribosome assembly intermediates, we used an in vitro maturation assay to monitor 20S pre-rRNA cleavage by the yNob1 nuclease at site D [4]. In preribosomes purified from wild-type yeast using PTH–Nob1 as bait, cleavage was activated by addition of ATP (Figure 5E and quantifications in Figure S4I), as previously reported [4]. In contrast, preribosomes (in 80S-like Fap7-depleted conformation) purified by PTH–Nob1 in Fap7-depleted cells are not competent for site D cleavage (Figure 5E). These data confirm that preribosomes assembled in the absence of yFap7 are not substrates for the Nob1 nuclease probably due to incomplete assembly in the platform domain.
Fap7 depletion results in loss of Rps14 from pre-80S particles, suggesting that reintroduction of Rps14 might rescue site D cleavage in vitro. We reasoned that Fap7 could act as an assembly factor to reincorporate Rps14 in the ribosome. However, addition of yFap7 or yFap7–yRps14 in the absence or the presence of ATP/Mg did not lead to any activation of D-site cleavage, in wild-type preribosomes or preribosomes purified from Fap7-depleted cells. These results show that Fap7–Rps14 cannot remodel the complex into an active conformation. Mass spectroscopy analysis of the purified particles shows the preribosomes purified by PTH–Nob1 in wild-type and Fap7-depleted cells both contain the 40S and 60S ribosomal proteins in roughly the same proportions, but assembly factors are overrepresented in wild-type preribosomes (Figure S4H and Table S1). The lack of cleavage could therefore arise because factors required for Fap7 activity in vivo are absent in the in vitro assay, because the experimental conditions are not optimal for the activity of the Fap7–Rps14 complex or because the pre-80S complexes generated in the absence of Fap7 correspond to dead-end intermediates.
In the preribosomes purified in the absence of Fap7, we observed rRNA degradation bands, consistent with an altered structure of the 20S rRNA within the pre-40S particles (Figure 5E). Several rRNA structure-probing studies indicate conformational plasticity of the 3′ region of the 18S rRNA during ribosome biogenesis [7],[26],[27]. In the absence of yFap7 or in the Rps14–R134A mutant background, an 18S rRNA degradation product (referred to as 17S RNA) missing Helices 45 and 44 is observed [14],[19]. This indicates that compromising the function of Fap7–Rps14 causes the 20S pre-rRNA to adopt a nonnative conformation that is prone to degradation. This is in line with electron microscopy structures of the pre-40S particles, which show that the platform domain is assembled even though Rps14 is not at its final position [8]. Further experiments are needed in order to determine if Fap7–Rps14 has to be correctly incorporated prior to formation of pre-80S particles or if Fap7 is present in the pre-80S particles during the final translation-like steps that confer competence for site-D cleavage.
The binding interface of the aFap7–aRps14 complex involves residues important for both the catalytic activity of Fap7 and the RNA binding capacity of Rps14. The double D82A–H84A mutation in yFap7, located in the Walker B motif, was shown to inhibit 20S processing in the same way as FAP7 deletion [14], and the corresponding mutation in hFap7 (H79G) abolishes ATPase and AK activities [12]. This suggested that nucleotide binding and/or hydrolysis by Fap7 might regulate the binding of Rps14 to the rRNA. To test this hypothesis, we evaluated the roles of nucleotides in formation or dissociation of the Fap7–Rps14 complex.
To determine whether nucleotide binding strengthened or weakened the yFap7–yRps14 interaction, we used a pull-down assay with GST-tagged yRps14 bound to glutathione resin. The beads were incubated with yFap7 and/or RNA (H23 helix construct, Figure 5C) in the presence of various combinations of nucleotides and magnesium, the beads were washed with 30 volumes of buffer, and the bound proteins were revealed by SDS-PAGE. In the same conditions and where appropriate, bound RNA was precipitated and revealed on an acrylamide-urea gel. We find that in the absence of nucleotides or in the presence of ADP+Mg, yFap7 is able to bind yRps14 (Figure 6A, lanes 4 and 6). However, in the presence of ATP+Mg, almost no yFap7 was retained by yRps14, indicating that ATP+Mg severely reduces the affinity of yFap7 for yRps14 (Figure 6A, lane 5). In order to determine if this effect was due to ATP binding in the yFap7 ATP binding site or to ATP hydrolysis, we performed this experiment in the presence of the nonhydrolysable analogue AMPPNP (Figure 6A, lane 7). AMPPNP had no effect on yFap7 binding, indicating that hydrolysis and not binding is required for complex dissociation. This was further confirmed by performing the assay in the absence or presence of magnesium. The yFap7–yRps14 complex was dissociated (or failed to form) only in the presence of ATP+Mg (lanes 4 and 6), further confirming that ATP hydrolysis modulates the protein–protein interaction. To ensure that was not due to loss of nucleotide binding, we have verified by native gel electrophoresis using radiolabeled nucleotides that Fap7 binds AMPPNP and ATP−Mg with the same (AMPPNP) or a slightly lower (ATP−Mg) affinity compared to ATP+Mg (Figure S4B). No difference in the effect was observed at high (1 mM) or low (10 µM) nucleotide concentrations (Figure 6B), indicating that the effect was also not due to competition for the yRps14 binding site. Interestingly, no effect of ATP+Mg was observed when the assay was performed solely at 4°C. Incubation of the mix at room temperature is absolutely necessary, consistent with the model that the catalytic activity of the enzyme is responsible for the effect (unpublished data). This effect also explains why this effect was not seen in experiments performed on hyperthermophylic archaeal proteins at low temperatures [19].
We have monitored the stimulation of ATPase activity in our coupled enzyme assay (Figure 6E). The ATPase activity of yFap7 (5 µM) as a function of time was comparable to the background levels without yFap7. After 10 min incubation, Prp43 (a helicase harboring ATPase activity) or MBP-yRps14 were injected at the same concentration as yFap7. For Prp43, we observed an increased rate of production of ADP, reflecting the ATPase activity of the protein. Similarly, addition of yRps14 resulted in a rapid burst of ATP hydrolysis, followed by a linear increase in ADP formation, demonstrating that ATP hydrolysis by Fap7 is stimulated by yRps14 (Figure 6E). In contrast, addition of the preformed yFap7–yRps14 complex to the ATP buffer did not lead to a measurable increase in ATPase activity. This is in line with the observation that dissociation of the yFap7–yRps14 complex in the presence of ATP+Mg could only be observed by mixing the separately purified proteins, and not with the yFap7–yRps14 complex purified by co-expression (Figure S5). This explains why the co-purified complex, already in a stable conformation, could be crystallized in the presence of ATP+Mg and possibly shows that an external factor is necessary to trigger ATPase activity and dissociate the fully assembled complex in vivo.
In order to determine whether the nucleotide-dependent binding of Fap7 to Rps14 modulates the binding of Rps14 to RNA, this assay was repeated with GST-Rps14 that had been preincubated with an increasing ratio of RNA (Figure 6C and D). In agreement with the filter binding assays, yRps14 efficiently remained associated with the RNA construct in the absence of Fap7, and yFap7 was capable of displacing RNA from yRps14. yFap7+ADP was more effective in displacing RNA than yFap7 in the absence of nucleotides (compare lanes 5–7 and 11–13), while ATP+Mg failed to compete with RNA (lanes 8–10). This indicates that ADP and ATP modulate the affinity of the Fap7–Rps14 complex and could indirectly regulate the binding of Rps14 to RNA through this interaction.
Two mechanisms by which hydrolysis could dissociate the yFap7–yRps14 complex can be envisaged. The first explanation is that ATP hydrolysis induces a “static” modification of the structure (or of the oligomerisation state) of yFap7 and/or yRps14 that renders it incapable of forming the complex. One such modification compatible with the biochemical and structural results would be that Fap7 acts as a protein kinase to phosphorylate Rps14. The location of the Rps14-CE in the binding site for the AMP substrate of aFap7 would be consistent with this model. This mechanism would potentially be similar to the phosphorylation-dependent incorporation of Rps3 into pre-40S by the Hrr25 protein kinase [28],[29]. The γ phosphate of ATP is positioned at 5 Å from the Cα atoms of aRps14–Asp124, comparable to the distance to the AMP substrate. Moreover, a serine is found in the position equivalent to Gly125 in humans and yeast, and has been identified as phosphorylated in high-throughput assays [30]. The ability of yFap7 to phosphorylate yRps14 was assessed by incubating yFap7 and the yFap7–yRps14 complex with [γ-32P] ATP, but no evidence of phosphorylation of yRps14 or autophosphorylation of yFap7 was detected (unpublished data). However, phosphoaspartate regulation of Rio2 assembly has already been demonstrated [10], and because phosphoaspartates are very unstable, the phosphorylation of Asp124 warrants further investigation. This would also be consistent with the burst of ATPase activity seen in our assay upon addition of Rps14.
An alternative model would be a “dynamic” mechanism, in which complex dissociation is driven by conformational changes powered by ATP hydrolysis. This is potentially similar to the large-scale conformational changes in the ATP and AMP binding site observed upon substrate binding and nucleotide hydrolysis in AKs. This involves the transition between a closed (ATP- and AMP-bound) and an open (apo) conformation of the LID and NMP domains [31]–[33]. This transition is coupled to local unfolding (cracking) that results from the release, during catalysis, of intramolecular strain built up by residues in the LID domain [34],[35]. In Fap7, such a local unfolding event of the LID domain during catalysis is strongly predicted to disrupt the Rps14 binding interface. Conversely, nucleotide binding in the ATP binding pocket would stabilize the closed conformation and by extension the Fap7–Rps14 interaction surface. This model is consistent with the observation that ADP binding increases the affinity of yFap7 for yRps14 and that ATP hydrolysis destabilizes the complex.
A specific model for the conformational change can be proposed from the different structures of the aFap7–aRps14 complex. In the four copies of the complex present in the asymmetric unit in each space group, a second conformation of the aFap7–aRps14 complex is observed, which differs in the structure of three regions: the LID domain of aFap7, and the C-terminal extension and β4–α2 loop of aRps14. This structure represents a more open conformation of the NMP domain, a partially unstructured Rps14-CE (loss of four salt bridges) and a conformation of the aRps14 β4–α2 loop that would clash with a structured Rps14-CE (Figure 7A). We propose that the different conformations observed for the aFap7–aRps14 complex reflect changes relevant to Fap7 function and suggest a series of concerted motions that favor Rps14 disassembly: (1) Hydrolysis of ATP unfolds the LID domain helices, which affects interactions with both the Rps14-CE and the core Rps14 domain; (2) the Rps14-CE is destabilized and the salt bridges are broken; (3) the NMP domain β-hairpin shifts to an open conformation favoring the release of the aRps14-CE; and (4) the Rps14 β4–α2 loop “pushes” the Rps14-CE out of the substrate binding cavity (Figure 7A and Movie S1).
The Fap7–Rps14 complex is the first structure of a ribosome assembly factor bound to a ribosomal protein. The structure of the complex is very informative because it shows that the binding interface involves regions important both for the catalytic function of Fap7 and for RNA binding by Rps14. Fap7 appears to sequester Rps14 by blocking its RNA binding site in an ATP/ADP-dependent mechanism. Binding of Fap7 with Rps14 is not compatible with the position of Rps14 in the mature ribosome, and preribosomes purified in the absence of Fap7 do not contain stoichiometric amounts of Rps14. This suggests that Fap7 is involved in incorporating and/or repositioning Rps14 in the pre-40S particles. Fap7 would therefore function as a chaperone, enabling the final assembly of Rps14 in the ribosome at a particular stage in the maturation pathway. A model for this is presented in Figure 7B.
The remarkable correspondence between the RNA and Fap7 binding interfaces of Rps14 indicates that Fap7 chaperones the assembly of Rps14 by mimicking RNA. Interactions of Rps14 with protein side chains in the Fap7–Rps14 complex are replaced by RNA interactions in the mature ribosomal subunit. RNA mimicry is relatively rare, and only a few proteins are known to show this property in RNP assembly pathways [36],[37]. The Fap7–Rps14 complex is also notable because the binding of Rps14 inhibits the adenyl kinase catalytic function of Fap7. Inhibition of nucleases by proteins with RNA mimic function have already been observed for the immunity proteins of colicin nucleases [38], but Fap7 is unusual because the RNA mimic is the enzyme that is inhibited.
Although Fap7 harbors AK activity, the structure of the Fap7–Rps14 complex suggests that this enzymatic activity is not used directly in ribosome biogenesis. We speculate that in a common ancestor to Archaea and Eukaryotes, Fap7 had an AK-related function in the nucleus but was reprogrammed by gaining the capacity to bind Rps14, thus enabling Fap7 to regulate ribosome biogenesis. The biological relevance of nuclear AK activity has not been demonstrated directly, but AMP/ATP signaling networks are important in the nucleus and ribosome biogenesis is an energy-consuming process. Under stress conditions, such as nutrient deprivation, rapid inhibition of ribosome synthesis may be important to conserve energy. The AMP-activated protein kinase (AMPK) monitors nutrient availability and cellular energy status, through the AMP/ATP ratio, and regulates rRNA transcription, but is otherwise unrelated to Fap7 [39],[40]. It has also recently been proposed that AMPK responds to ADP/ATP ratios [40]–[42]. Moreover, nuclear transport can also be regulated by AK activity [43]. Because Rps14 and AMP could compete for the same binding site on Fap7, we propose that the in vivo function of Fap7, and therefore maturation of 40S ribosomal subunits, would be inhibited at elevated levels of AMP, which might signal reduced energy availability. A deregulation of the ADP/ATP ratios would also interfere with the Fap7–Rps14 association and/or dissociation and affect ribosome biogenesis. The reported role of Fap7 in oxidative stress responses [44] might also be mediated by the effects deregulated ADP/ATP ratios on Fap7–Rps14 association and dissociation. Notably, the inhibition of the late, cytoplasmic steps in 40S subunit synthesis in yeast leads to accumulation of pre-40S particles, which might therefore remain available for subsequent processing. This is in contrast to the inhibition of earlier, nuclear steps in ribosome synthesis, which leads to rapid degradation by RNA surveillance systems.
Rps14 is an essential structural component of the ribosome and, together with Rps5, is involved in helping to form or maintain the structure of the head/platform domain [45]. However, although mutations in the C-terminal extension of Rps14 interfere with site-D cleavage [20], depletion of Rps14 leads to earlier defects in pre-rRNA processing [46]–[48]. Rps14 therefore has a dual role in ribosome biogenesis, acting in both early and late preribosomal particles, as is the case for Rps5 [45]. This indicates that Rps14 is bound to the ribosome throughout the maturation process, so its function in 20S processing must be triggered by an external factor or conformational change. Fap7 is localized in both the nucleus and cytoplasm, suggesting that it too is bound early in preribosome maturation, although the timing of Fap7–Rps14 complex formation is unclear. A major implication of the structure of the aFap7–aRps14 complex is that Fap7 binding is not compatible with the structure of ribosome-bound Rps14. Thus, binding of Fap7 to Rps14 would prevent the final 40S subunit association of Rps14. Rps14, Rps1, and Rps26 are underrepresented in late preribosomes in the absence of Fap7 [5]. This argues that Fap7 does not simply block binding of these proteins, but is required for their correct assembly.
EM structures of preribosomes also show that the platform is not completely assembled: Rps14 is not in its final position, Pno1–Dim2 overlaps with the Rps16 binding site, and the rRNA, notably Helix 44 and the decoding site, is shifted from its position compared to the mature ribosome structure [8]. In addition, the presence of the 212 nt ITS1 region, on the 3′ end of the 18S rRNA, potentially creates steric hindrance in pre-40S particles and may distort the structure of the platform domain. A major conformational change may therefore be necessary to reposition the Nob1 at site D, and this was suggested to occur during the translation-like cycle in the pre-80S ribosomes [4]. The pre-40S EM structures suggest that Rps14 is already present in the platform domain before cleavage, and we speculate that prior to the conformational rearrangement, Fap7 bound to Rps14 functions to prevent premature assembly of the platform.
Our results suggest a model for the function of Fap7 in ribosome biogenesis (Figure 7B). We propose that in early pre-40S ribosomes prior to conformational reorganization of the platform domain, Rps14 is located in the platform domain but has not adopted its binding site in the mature ribosome. Prior to or during the structural rearrangement of the platform, Fap7 transiently binds Rps14, displacing bound RNA. Subsequent dissociation of the Fap7–Rps14 complex by ATP hydrolysis allows Rps14 to rebind the 18S rRNA, which may assist its folding into the mature conformation. After this step, the pre-40S subunits are ready for association with Fun12 and mature 60S subunits and can undergo the final site-D cleavage in pre-80S particles. Although our model is consistent with the experimental results for the Pyrococcus horikoshii archaeon [19], the model we propose is different because of the additional information provided by the structure of the complex and because biochemical assays, performed on the yeast proteins at physiological temperatures, uncovered a link between ATP hydrolysis and complex dissociation.
It remains to be determined how Fap7 and the Fap7–Rps14 complex associate with the ribosome. Because the Rps14 C terminus is specifically required for the late assembly steps, it is possible that Rps14 binds both rRNA and Fap7 through the Rps14-core domain and C-terminal extension, respectively. It is also possible that Fap7 is additionally anchored in the preribosomes by interaction with other proteins. Consistent with this, protein cross-linking identified Fap7 contacts with several proteins in the preribosomes [5]. An interaction between aFap7 and aNob1 has also been observed [19], but we have not been able to observe this interaction with the yeast homologues.
The model that Rps14 is initially bound in the vicinity of its final location but then repositioned by Fap7 is conceptually related to the assembly of Rps3 [29]. Rps3 is initially weakly associated with preribosomes but is dissociated by phosphorylation by the Hrr25 protein kinase. After the subsequent formation of the beak structure in the pre40S, dephosphorylated Rps3 reintegrates the ribosome in its final position. This assembly pathway is akin to the assembly of Rps14 by the Fap7 kinase, which is linked to a structural rearrangement in the platform domain. Given the complexity of preribosome assembly and the huge numbers of factors involved, each of these two-step assembly pathways may enhance overall speed and efficiency. Several other proteins have been reported to act as “place holders” that occupy ribosomal protein binding sites prior to recruitment of the ribosomal protein itself into the maturing structure [49],[50]. It has also been proposed that dispersed pre-rRNA regions can initially be brought together by ribosome synthesis factors, prior to binding by ribosomal proteins that will lock these regions into their final positions in the mature ribosome [51]. It therefore appears that ribosome assembly not only proceeds sequentially by addition of factors but also involves major rearrangements in ribosomal protein binding sites that have yet to be fully characterized.
Finally, the structure of the Fap7–Rps14 complex provides a molecular framework for the design of inhibitors of the interaction. These might serve to reactivate p53 expression in cancerous cells in which ribosome biogenesis has been deregulated. Our structure suggests that molecules targeted to the Rps14-CE binding cavity could effectively interfere with ribosome biogenesis in a similar manner to Rps14-CE mutations. The structure of the cavity offers the potential for designing much more selective inhibitors than those targeted to the ubiquitous ATP binding pocket. Disruption of the Fap7–Rps14 interaction should induce cell cycle arrest by both deregulating ribosome biogenesis and perturbing the Rps14–MDMd2–p53 pathway. This combination of p53-dependent and -independent effects could potential provide powerful chemotherapeutic approaches [52].
The open reading frames of both genes of Fap7 and Rps14 from S. cerevisiae and from P. abyssi were synthesized commercially by Genscript Corp (Piscatawy, NJ) and inserted as polycistronic constructs in pET21(a+) (Novagen). The pET21–Fap7–Rps14 constructs contain an N-terminal 6xHis-tagged Fap7 fusion protein. The constructs were expressed in Rosetta 2DE3 strain from Escherichia coli (Invitrogen) at 37°C in LB medium (Sigma) supplemented with ampicillin (100 µg ml−1) and chloramphenicol (25 µg ml−1) until OD600 nm between 0.6 and 0.8. Recombinant protein expression was induced by addition of 1 mM isopropyl β-D-1-thiogalactopyranoside. Cell cultures were incubated 4 h at 37°C and then harvested by centrifugation and resuspended in buffer A (20 mM Tris-HCl pH 8, 300 mM NaCl, 40 mM Imidazole) supplemented with 1 mM phenylmethylsulfonyl fluoride. Cells were lysed by using a French Press (4 runs at 11,000 p.s.i.), and lysate was centrifuged for 30 min at 20,000 rpm. The clear lysate was loaded onto a 5 ml HisTrap (GE Healthcare) connected to an ÄKTA purifier (GE Healthcare). Nonspecific proteins were removed by washing the column with buffer A, and the Fap7–Rps14 His-tagged complexes were then eluted with a linear gradient of imidazol (buffer B, 20 mM Tris-HCl pH 8, 300 mM NaCl, 500 mM Imidazole). Gel filtration was then performed on the fractions containing eluted proteins using buffer C [20 mM Tris-HCl pH 8, 200 mM NaCl, 1 mM Dithiothreitol (DTT)] on a Superdex 75 26/60 (GE Healthcare).
For GST-yRps14 and MBP-yRps14, the open reading frame of Rps14 from S. cerevisiae was cloned into pETM30 and pETM41, respectively (courtesy of A. Geerlof, EMBL Hamburg Outstation), and for His-yFap7, the open reading frame of Fap7 from S. cerevisiae was cloned into pRSF-DUET (from S. Granneman). For all the archaeal proteins, aRps14, aFap7, aRps14-ΔC, and aFap7ΔC, the open reading frames from P. abyssi were synthesized commercially by Genscript Corp (Piscatawy, NJ) and inserted in pETM11 (courtesy of A. Geerlof, EMBL Hamburg Outstation). All these constructs were expressed in Rosetta 2DE3 strain from E. coli (Invitrogen). Cells were grown in LB at 37°C to an OD of 0.5, and then protein expression was induced with 0.5 mM IPTG for 4 h at 37°C for archaeal proteins and 16 h at 18°C for yeast proteins. Cells were lysed by sonication, and the lysate was centrifuged for 30 min at 20,000 rpm.
For protein purification, the clear lysate was loaded onto a 5 ml HisTrap (GE Healthcare) connected to an ÄKTA purifier (GE Healthcare) for His-tagged proteins. Nonspecific proteins were removed by washing the column with buffer A′ (50 mM Hepes pH 7.5, 300 mM KCl, and 40 mM Imidazole), and the His-tagged complexes were then eluted with a linear gradient of imidazol (Buffer B′, 50 mM Hepes pH 7.5, 300 mM KCl, 500 mM Imidazole). Gel filtration was then performed on the fractions containing eluted proteins using buffer C′ (50 mM Hepes pH 7.5, 300 mM KCl, 1 mM DTT) on a Superdex 75 16/60 (GE Healthcare).
For IP assays, we prepared beads with GST-yRps14. Cells were harvested by centrifugation, and pellets were resuspended in 5–8 volumes of IP buffer (300 mM KCl, 50 mM Hepes pH 7.5, 0.5 mM EDTA, 1 mM DTT, 10% Glycerol) containing protease inhibitor complete EDTA free from Roche. Cells were broken by sonication, and the extract was clarified by centrifugation at 20,000 rpm for 30 min. Clarified extract was then incubated with pre-equilibrated glutathione sepharose beads from GE at a ratio of 0.5 ml of bead for each litter of extract, overnight at 4°C. Beads were then washed three times with 10 volumes of IP buffer and stored at 4°C.
The crystallization trials were performed using the hanging-drop vapor diffusion technique in 1 µl drops (with a 1∶1 protein∶precipitant ratio) equilibrated against 500 µl reservoir solution at 18°C. The two different crystal forms were obtained in 0.2 M MgCl2, 0.1 M Tris-HCl pH 8.5, 25% (w/v) polyethylene glycol 3350 with the aFap7–aRps14 complex at 3 mg ml−1 supplemented with 1 mM MgCl2 and either 1 mM ADP or 1 mM ATP.
Crystals were cryoprotected using three successive soaking steps with the reservoir solution containing 10%, 20%, and 30% ethylene glycol. X-ray data were collected at the Soleil Synchrotron (Saint-Aubin, France) on Beamline Proxima1 and at ESRF (Grenoble, France) on ID23-1. For phasing, crystals were soaked in a solution composed of the reservoir solution supplemented with 3 mM potassium tetrachloroplatinate(II). Data were collected at the absorption threshold of Platinum (1.0716 Å).
Native and SAD datasets were indexed using the program XDS [53], and experimental phasing and molecular replacement were carried out with the program Autosol and Phaser [54] from PHENIX [55] and described elsewhere (Loc'h et al. in preparation). Initial rebuilding was carried out with Buccaneer from the CCP4 program suite [56] and subsequent rebuilding and refinement with COOT [57] and the Refine module from PHENIX. The ATP crystal form was solved by molecular replacement with the ADP bound model of the complex using PHENIX. PDB coordinates were deposited with codes 4cwn and 4cw7 for the ADP- and ATP-bound structures.
SAXS data were collected on Beamline SWING at the Soleil Synchrotron at an electron energy of 13.32 keV. Protein samples were centrifuged at 13,200 g for 10 min before data acquisition, and concentration was measured with Nanodrop spectrophotometer (Thermo). Two modes of data collection were used: direct injection and size exclusion chromatography separation prior to data collection. For direct injection, samples at known concentration were loaded at a flow rate of 200 µL min−1 in the SAXS capillary flow cell. For size exclusion chromatography, samples were loaded on an Agilent SEC-3 size-exclusion chromatography column connected to an Agilent HPLC system at a flow rate of 200 µL min−1. In each case, 100 frames, with exposure times of 1 s per frame, were collected at 288 K in the flow cell connected to the HPLC system. The same buffer was used for all data collection (Tris-HCl 50 mM, NaCl 200 mM, pH 7.5). Buffer scattering was collected in the same conditions or using the gel filtration profile before the void volume.
Data reduction—that is, image conversion to 1D profile, absorption correction, scaling to absolute scale, buffer subtraction, averaging, and peak analysis—was carried out using the program FOXTROT, available on Beamline [58]. Further processing and data analysis was done using the programs of the ATSAS suite [59]. Guinier plot analysis was performed using PRIMUS [60] with scattering data at low angle regions with qmax*Rg<1.3. Pair distribution function was calculated with GNOM [61].
Ab initio analysis was performed with DAMMIF [62]. The 50 DAMMIF calculations were averaged with DAMAVER [63] to produce averaged and filtered shape. Homology models were made with MODELLER [64]. Modeling with SAXS data was done with DADIMODO [65]. Analysis of the fit between model and experimental data was done with CRYSOL [66]. The χ2 value (Chi) against the experimental SAXS profile and the residual are computed with the Crysol program and defined as: qk is the momentum transfer, Iexp the experimental intensity data, Im the intensity calculated from the model, σexp the experimental error, and c a scaling parameter.
The residual is defined as:
The data collection and processing statistics are reported in Table 2.
A kinetic enzyme-coupled method for assay of yFap7 AK activity was carried out as previously described for hFap7 [12]. AK and ATPase assays were performed on a safas UV/Vis spectrophotometer. When ATP was used as the substrate, the production of ADP is coupled to β-NADH oxidation via the action of pyruvate kinase (PK) and lactate dehydrogenase (LDH). The rate of β-NADH disappearance was monitored at 340 nm over a 15-min period and at 30°C. Reference samples, containing reaction mixture without yFap7, were used to subtract background absorbance mainly attributable to nonenzymatic ATP hydrolysis and the ATPase activity of puryvate kinase.
The final assay mixture (200 µL) consisted of 100 mM Tris-HCl pH 7.5, 60 mM KCl, 5 mM MgCl2, 0.2 mM β-NADH, 1 mM PEP (Phospho(enol)pyruvate—Sigma), 10 U/mL PK (Sigma), 14 U/ml LDH (Sigma), H2O, and 0.01–1 mM ATP. The AK activity of yFap7 (5–10 µM) with respect to ATP was measured in the presence of 0.3 mM AMP. Similar approaches have been carried out to study the ATPase activity of yFap7, the AK activity of yFap7 with respect to AMP, and the AK activity of yFap7 associated with yRps14.
Filter binding assays were performed and adapted as described [36]. The labeled RNA was heated at 65°C for 2 min and immediately placed on ice for 10 min and diluted in binding buffer containing 30 mM Tris-HCl pH 8.0, 150 mM KCl, 6 mM MgCl2, and 60 µg of E. coli tRNA per ml. Binding reactions consisted of 12 µl of RNA (10 fmol) and 8 µl of proteins (final concentration from 0.0036 to 3.6 µM). Binding reactions were incubated at 20°C for 15 min and then applied directly to filters containing the two membranes under gentle vacuum. Before and after application of the binding reactions, 200 µl of binding buffer without tRNA was used to equilibrate and rinse the system. Binding was quantified using a Molecular Dynamics phosphorimager and Image Gauge program (Fujifilm). The Intensity was corrected for background and fit for Kd using GNUPLOT (Thomas Williams, Colin Kelley et al., http://www.gnuplot.info) using the following equation:
The scale factor s was fitted along with Kd for the Rps14/RNA lane and applied to all other measurements.
To test interaction between yRps14, yFap7, and RNA, 30 µl of GST-yRps14–coated beads were used for each condition; beads were then resuspended in 1 ml of IP buffer and then treated as described.
Recombinant His-yFap7 was added and incubated for 1 h at 4°C; then beads were washed three times in IP buffer and one time in PBS. When nucleotides were used, they were incubated at the same time at 4°C for 1 h and then at room temperature (RT) for another 15 min before the washes. When both RNA and Fap7 were tested, RNA was first incubated with GST-yRps14+ beads for 1 h at 4°C, then His-yFap7 (with or without nucleotides) was added for another 1 h at 4°C, and then it was incubated at RT for 15 min before being washed.
For protein analysis, beads were resuspended in 50 µl of SDS loading buffer and boiled at 95°C for 5 min, and the supernatant was loaded on SDS-PAGE. For RNA analysis, RNAs were immediately extracted as described previously [4] and analyzed on 15% acrylamide–8 M Urea gel and stain with SYBR Safe.
Analytical gel filtration experiments for the detection of protein–protein interactions were carried out using an analytical Superdex S200 (GE) gel filtration column with a flow rate of 0.5 ml/min on an Äkta purifier system (GEHealthcare) in a 50 mM Hepes pH 7.5, 300 mM KCl buffer at 20°C. The protein concentrations were 20 µM and the total applied sample volume was 100 µl in all cases. Protein elution was followed by recording the UV adsorption at 280 nm.
WT, PTH-NOB1 or ΔFAP7, PTH-NOB1 cells were grown in 1 l of YNB glucose (2%) for 8 h to an OD600 of 0.5, collected, and washed with phosphate-buffered saline. Extracts were prepared in 500 µl of buffer A [50 mM Tris pH 7.5, 150 mM NaCl, 5 mM MgCl2, 0,1% NP-40, 1 mM DTT, and protease inhibitors (Roche)] using Zirconia beads as previously described [4]. Aliquots corresponding to 12 mg of proteins were added to 50 µl of immunoglobulin G (IgG)-sepharose beads (GE Healthcare) in a 1 ml final volume of buffer A. Immunoprecipitation was performed at 4°C for 1.5 h. Beads were then washed three times (5 min per wash) with 1 ml of buffer A at 4°C. Most of the supernatant was discarded and 50 µl of buffer X (50 mM Tris pH 7.5, 150 mM NaCl, 5 mM Mn2+, 0.1% NP-40, 1 mM DTT, 10% glycerol) added to the pellet (recovered by resuspension in 20 µl of remaining buffer A) to reach a final volume of 70 µl. We added 250 pmoles of recombinant yFap7 or yFap7–yRps14 complex when indicated and incubated it for 20 min at 20°C. Nucleotides were added when required at a final concentration of 1 mM. Reactions were incubated at 20°C for 30 min, and RNAs were then immediately extracted as described [4].
|
10.1371/journal.pntd.0001782 | Atypical Systemic Leishmaniasis to Be Considered in the Differential of Patients Presenting with Depressed Immunity | Systemic leishmaniasis has been known to present with prolonged fever, hepatosplenomegaly and wasting. Beside this classical form, a sub-clinical form has been identified. It is described with either one or two of the above symptoms missing; other findings have been reported instead, such as lymphadenopathy and anemia. In this report, we reveal a third unsuspected form which we are referring to as “atypical”.
Patients suspected to be immune-deficient were referred to our immunology specialized laboratory to study some aspects of their immune functions (not normally covered in the general laboratory). Multiple specialized tests were performed, including microscopic examinations using appropriate stains, and mainly cultures of biopsies on several types of specialized media. 19·4% of 160 patients were found to have close to normal laboratory profiles, but exhibited dysfunctional macrophages laden with Leishmania parasites.
Findings such as the ones we obtained allowed us to uncover the presence of patients with an atypical form of systemic leishmaniasis. It presents with symptoms masquarading a condition in which the immune system is non functional. This predisposes patients to recurrent secondary infections resulting in clinical pictures with a great variety of signs and symptoms. These findings alerted us to the fact that systemic leishmaniasis presents with a much wider spectrum of signs and symptoms than so far suspected and is far more common than diagnosed to date. Furthermore, among these 31 patients was a number of adults. This proved that in our area systemic leishmaniasis is surely not limited to the pediatric age group. Our recommendation is to entertain the diagnosis of atypical systemic leishmaniasis in any patient with an unexplained depressed immunity state and in whom no obvious immunologic defect can be identified.
| Systemic leishmaniasis is known to have in general two types of clinical presentations. The classical form is a full fledged disease characterized by fever, severe weight loss, enlargement of lymph organs and anemia. It renders patients very sick and is sometimes fatal. The second form, known as sub-clinical leishmaniasis, is a milder disorder. Although it presents with symptoms analogous to the ones classically associated with this infection, they are far less severe and certainly fewer in number. In this report, we are describing a third type of presentation, which is distinct from the above described. The hallmark of this third type of clinical form is depressed immunity. A number of patients suffering from repeated infections affecting various body organs and systems (skin, respiratory, and alimentary) were referred to us for advanced immune function testing. We were able to identify in a fraction of these patients Leishmania parasites. Upon appropriate treatment, these patients recovered. Systemic leishmaniasis was thus the culprit of their repeated episodes of falling ill. Hence our recommendation in this report is to consider the diagnosis of systemic leishmaniasis in patients suffering from signs and symptoms indicating depressed immunity (repeated episodes of fever or infections).
| Visceral Leishmaniasis (VL) or Kala-Azar was first described by Leishman in 1906. It is characterized by fever (periodic or continuous), cachexia, organomegaly (liver and spleen) causing the abdomen to protrude, severe wasting of the limbs and trunk, and pancytopenia [1]–[4]. Due to systemic parasite dissemination, the disease is in general fatal if untreated [5].
Reviewing the literature, there seems to exist two types of systemic leishmaniasis. A form presenting with the above described signs and symptoms i.e. a classical form, and another type of the disorder referred to as the sub-clinical. The latter was first alluded to by Leishman in 1906; then it was followed by several groups, starting from the late 1950s with Manson-Bahr [6], who recognized the presence of infected subjects carrying the parasite yet presenting with symptoms far milder. All of these groups based their diagnosis on positive immunological responses to the parasite [7]–[16].
The ratio of patients with sub-clinical disease to those with overt classical symptoms varied from one area to the other, and from one year to another. Its value (sub-clinical/overt) shifted from levels as low as 1∶2·4 to values as high as 18·5∶1 [14]. Part of the discrepancy in the reported figures is the result of several factors: First, the lack of agreement in defining this disorder; second, the limitations of the varied diagnostic tools, and third the geographic and demographic differences among populations (e.g. nutrition, genetics) [11], [14], [17].
In all cases, diagnosis of systemic leishmaniasis was dependent principally on the immune reaction to Leishmania antigens, especially that tissue examination does not always successfully demonstrate the parasite, neither do cultures of specimens from affected tissues.
As to our area, Bitar and Nachman reported 72 pediatric cases of leishmaniasis from five hospitals in Beirut (Lebanon). They were discovered between 1926 and 1964 (about 2 cases per year) [18]. Visceral Leishmaniasis presented then with symptoms and signs consistent with the typical textbook picture. This pathology invariably ended by the lymphatic system's insufficiency, hence total immune paralysis and death of the patient.
Over the last 13 years, 160 patients diagnosed with depressed immunity, and who underwent exhaustive laboratory investigations to no avail, were referred to our specialized laboratory for further evaluation of their immune system function.
Hence our objective in this report is to disclose to scientists in the field that systemic leishmaniasis has a clinical picture with far more protean signs and diverse symptoms.
This project received approval from the Institutional Review Board of the American University of Beirut Medical Centre (AUB–MC) as part of our epidemiologic surveillance study initiated in 1994, when the Leishmania study group was established. The project objectives were re-approved in 1999. We obtained written informed consent from all patients or from their parents/guardians (when they were less than 18 years of age). It was agreed not to reveal the identity of the patients should the results of their laboratory tests be used in any research study.
We reviewed all the cases referred to our laboratory situated at AUB–MC with the diagnosis of depressed immunity. Part of our routine in these cases was a questionnaire answered by each patient. We extracted the biodata with special attention to the major clinical features in the symptomatology. We also included the geographic area where the patients lived and live, as well as their personal and family histories. Travel in and outside the area was recorded as well.
As for the battery of laboratory tests carried out on every patient, these consisted of a complete blood count (CBC), immunoglobulin titers (total and subclasses), complement activity and titers (total, C3 and C4). In addition, we tested chemotactic activity using a Skin Window. That was performed at 2 time intervals, one slide at 4 hours for the acute response, and the other slide at 18 hours for the chronic response [19]. In suspicious cases, the window was modified to detect the presence of monocytes with leishmania organisms (report in preparation). Whenever the cellular response on this test was deficient and in the absence of any infectious agents affecting the cells, a Boyden chamber assay was performed to analyze the defect revealed by the Skin Window.
In addition, blood cultures on classical media (bacteria and fungi) and on Novy Nicolle McNeal medium (NNN medium) were also carried out.
The majority of patients had sections of biopsies from either a lymph node or a bone marrow, or had an aspirate of bone marrow and less often of the spleen. All of these samples were obtained in the course of the general investigation of their “defective” immunity and often their anemia. Available sections and/or aspirates were stained with leishman and acid fast stains for microscopic examination. Immunofluorescence for Leishmania parasites was used for confirmation in equivocal cases. For the latter, a positive control from axenic parasite cultures was always included.
31 of the 160 referred patients (19·4%) were found to suffer from an infection with Leishmania parasites. Presentation of these subjects and behavior of the parasite in this group constitute the focus of the current report.
The majority of the infected patients were from the North of Lebanon (48%), Akkar District, which was identified as a focus in 1999–2000 [20]. This locale was followed by Beirut area (23%) then by the South (16%) and last the Beqaa Valley (13%) (Figure 1).
The age at presentation of these cases ranged from a few months to 45 years. The majority by far (84·8%) were children in the age group between 2 months and 12 years, with a bimodal peak around 2 and 6 years. Worth stressing that there were a few adults. The gender was equally distributed, 16 males and 15 females. Duration of the presenting illness spanned from a few months to 11 years, in most cases lasting between 2 and 3 years. The symptoms varied in severity and nature of the targeted organ.
This biodata, including the geographic locale, age at presentation, gender, duration of illness in addition to the major clinical problems that brought the patient, is summarized in table 1.
The most frequent complaint was recurrent infections (74·2%) expressed by pathology either in the skin, the gastrointestinal tract, or the respiratory system. Skin infections (subcutaneous boils and/or deep abscesses), occurring anywhere on the body, were in the lead. To note, these were devoid of any detectable insect bite site.
Failure to thrive and Fever of Unknown Origin (FUO) came second in frequency. Splenomegaly was found in the minority of cases (Table 2).
Reviewing the results of the laboratory studies on these 31 patients revealed that the CBC parameters were all borderline normal except in one case. The immunologic investigation by and large was normal (notably, negative Leishmania serology). IgG titers were upper normal in a few cases; only in one case, IgE levels were remarkably elevated with an eosinophil count ranging between 6 and 13% (on repeated testing). Worth mentioning that none of the 31 patients had positive skin test for Leishmania and tuberculosis. As for the organ biopsy, it did demonstrate the parasite but its efficacy depended on the sampled tissue. Figure 2 depicts two photomicrographs illustrating intracellular parasites on a section of lymph node (A), and on a splenic aspirate (B).
The modified Skin Window was very efficient in demonstrating through its chronic phase the presence of the parasite within and without the monocytes in all 31 patients (Figure 3).
Every time we obtained a biopsy for microscopic examination, part of it went into culture on NNN medium. The efficiency of demonstrating the parasite using microscopy and/or culture is summarized in table 3.
The best test for detecting the parasite microscopically was the Skin Window. The next in efficiency (culture and/or microscopy) was Buffy Coat, which revealed the presence of parasites in 71% of the cases. Whole blood followed. Bone marrow aspirate's yield was positive only in one third of the cases which could be a reflection of its depressed activity. Although obtaining a splenic aspirate entails a serious risk (the reason why we limited it to 2 patients), we still think it is a very efficient tissue in yielding parasites.
Kala-Azar has been established to present with a well defined clinical picture, but for the last few decades suspicion of having systemic leishmaniasis with a symptomatology different from the classically described, has been gaining greater and greater support. Such cases with a variable degree of illness were detected on the basis of positive serology [7]–[16].
In our series, the patients presented with symptoms typical of immune deficiency in the presence of a normal immune profile. Still they were found to harbor Leishmania organisms. In contrast to the reported sub-clinical cases among whom some subjects were totally symptom free while others had a milder form of the classical disorder, our cases presented with depressed immunity exhibiting symptoms pertaining to one or two target systems or organs.
A novel clinical picture of Visceral Leishmaniasis started to unfold. It consists of depressed immunity predisposing the subjects to infections with a wide spectrum of organisms (bacterial, viral and fungal). They presented with symptoms varying in type and severity. Although these secondary infections tend to resolve (with or without treatment), yet they invariably recurr. Thus in most patients and before referral, systemic anti-bacterial treatment was extensively used, sometimes in association with surgical interventions especially for skin abscesses and other tumorous growths. Anti-fungal agents were mostly added in cases with gastrointestinal complaints. Improvement or partial relief of symptoms arising from the superimposed infections was reported in most of our cases, but no complete cure was ever attained. Furthermore, it is important to note that none of the patients evolved to display the full fledged picture of Kala-Azar.
This form of atypical presentation is distinct from what has been described so far as sub-clinical illness meaning that it is not a lighter classical disease. This entity is novel in the sense that all patients were suspected to be immune-compromised and hence were sent to us for further immune function studies. Incidentally while performing the Skin Window to test the acute and chronic immune cellular responses to injury, we detected leishman bodies within and without the monocytes and some of the lymphocytes constituting the chronic phase of this reaction.
Whereas other investigators relied on immunological tests (purely serological) for proving the presence of the parasite in the candidate patients, we favored demonstrating it microscopically within any of the appropriate tissues involved. Although the spleen when enlarged seemed to be the ideal organ to reveal the parasite, yet the difficulty and the high risk implicated in its aspiration discourages most investigators from sampling it. Buffy Coat examination has obviously several advantages and was efficient in a good number of patients; however its efficacy drops the more the patient is anemic.
Besides, this set of data support the fact that systemic leishmaniasis in Lebanon and probably the area exists to a much greater extent than what we reported in our previous surveillance study [20]. Furthermore, the cases were detected in referrals from only two medical centers. Admitting the fact that the number of patients we are reporting is small for solid statistical analysis, still we think it may not be futile to detect a trend that could characterize such patients. So far the skin seems to be the organ most frequently targeted by these parasites. This is not surprising since these preferentially reside in cells of the hair follicles. Other systems may represent the shock organs being frequently affected by infections as happens in defective body defenses. As expected, it is the younger subjects (infants and children) who are the population at high risk.
Our major recommendation is to entertain the diagnosis of atypical leishmaniasis irrespective of age or country of patients presenting with depressed immunity, especially so whenever thorough investigation covering a large spectrum of diseases fails and multiple shotgun style therapies are tried without any success.
|
10.1371/journal.ppat.1007442 | High-risk human papillomavirus oncogenes disrupt the Fanconi anemia DNA repair pathway by impairing localization and de-ubiquitination of FancD2 | Persistent expression of high-risk HPV oncogenes is necessary for the development of anogenital and oropharyngeal cancers. Here, we show that E6/E7 expressing cells are hypersensitive to DNA crosslinking agent cisplatin and have defects in repairing DNA interstrand crosslinks (ICL). Importantly, we elucidate how E6/E7 attenuate the Fanconi anemia (FA) DNA crosslink repair pathway. Though E6/E7 activated the pathway by increasing FancD2 monoubiquitination and foci formation, they inhibited the completion of the repair by multiple mechanisms. E6/E7 impaired FancD2 colocalization with double-strand breaks (DSB), which subsequently hindered the recruitment of the downstream protein Rad51 to DSB in E6 cells. Further, E6 expression caused delayed FancD2 de-ubiquitination, an important process for effective ICL repair. Delayed FancD2 de-ubiquitination was associated with the increased chromatin retention of FancD2 hindering USP1 de-ubiquitinating activity, and persistently activated ATR/CHK-1/pS565 FancI signaling. E6 mediated p53 degradation did not hamper the cell cycle specific process of FancD2 modifications but abrogated repair by disrupting FancD2 de-ubiquitination. Further, E6 reduced the expression and foci formation of Palb2, which is a repair protein downstream of FancD2. These findings uncover unique mechanisms by which HPV oncogenes contribute to genomic instability and the response to cisplatin therapies.
| High-risk human papillomavirus (HPV) causes nearly all cervical and many other anogenital cancers, and oropharyngeal cancers. As cisplatin is the most commonly used drug for cervical and HPV-associated oropharyngeal cancers, it is important to understand how HPV oncogenes disrupt the Fanconi anemia (FA) pathway involved primarily in the repair of cisplatin-induced DNA crosslinks. However, the mechanism by which HPV E6 and E7 attenuate the FA pathway is poorly understood. We demonstrate that E6/E7 expression disrupts crosslink repair and increase cisplatin sensitivity, and attenuate the FA pathway through multiple unique mechanisms. First, E6/E7 causes accumulation of FancD2, a central component of the FA pathway, at the sites away from DNA damage. This results in reduced recruitment of Rad51, another repair protein involved in the pathway. Second, E6 causes delayed FancD2 de-ubiquitination, an important process for effective repair. Third, E6 expressing cells decreases the expression and foci formation of Palb2 repair protein. Together, this work elucidates the mechanisms by which HPV attenuates the repair of DNA crosslinks increasing cisplatin cytotoxicity and efficacy in treating HPV-positive cancers.
| High-risk human papillomavirus (HR-HPV) E6/E7 oncoproteins are essential for the development of malignancies of the anogenital tract and oropharynx, with HPV16 being the predominant type [1]. Cervical and oropharyngeal cancers are the most common HPV-associated malignancies among females and males, respectively [2]. Persistent HPV infection destabilizes the cellular genome which can lead to cancer. Genomic instability is likely the result of the numerous interactions of HPV oncoproteins with host tumor suppressors and DNA damage repair (DDR) proteins. Recently, we demonstrated that high-risk HPV oncogenes attenuate double-strand break (DSB) repair by impairing the homologous recombination pathway [3]. To further elucidate the mechanisms by which HPV oncogenes impair DDR, the present study focuses on the impact of HPV16 oncogenes on the Fanconi anemia-BRCA (FA or FA-BRCA) pathway.
The FA-BRCA pathway is involved in the repair of intra or interstrand crosslinks (ICL), thereby maintaining genomic stability [4, 5] (S1 Fig). ICLs block DNA replication and transcription and, thus, are highly cytotoxic to cells if not repaired. The FA pathway is composed of 22 FA proteins (identified to date) [6]. When any one of the FA genes of the FA-BRCA pathway is mutated, individuals have a spectrum of disorders, called Fanconi Anemia, characterized by bone marrow failure, congenital malformations, cancer predisposition, and cellular sensitivity to ICL-inducing agents [7]. Upon exposure to DNA crosslinking agents, and during S-phase of the cell cycle, ATR/CHK1 signaling gets activated and helps in the formation of an ubiquitin ligase complex (FA core complex), which is composed of eight FA proteins (FancA, -B, -C, -E, -F, -G, -L, and -M) with other associated proteins (such as FAAP-24, MHF2). Among the FA core complex subunits, FancM recognizes the stalled replication fork at the ICL site and forms a landing platform for the core complex [8]. Activated ATR/CHK1 phosphorylates several FA proteins including FancM, FancD2, and FancI [9]. The FA core complex through its E3 ubiquitin ligase subunit FANCL and the corresponding E2 ubiquitin-conjugating enzyme (UBE2T/ FancT) catalyzes the monoubiquitination of the FancD2-FancI heterodimer. Monoubiquitinated FancD2 (FancD2-Ub) and FancI-Ub form discrete nuclear foci at double-strand breaks (DSB) created at ICL-stalled replication forks, and subsequently recruit downstream DNA repair proteins, including FancD1 (BRCA2), FancS (BRCA1), FancR (Rad51), FancN (Palb2), and FancJ (BRIP1). These proteins in nuclear foci co-operate with other DNA repair pathways such as nucleotide excision repair, homology recombination, and translesion synthesis to repair ICLs [8]. Once DNA is repaired, the FA pathway is turned off by de-ubiquitination of the FancD2/ FancI complex to prevent prolonged cell cycle arrest and cell death [10]. FancD2/ FancI de-ubiquitination is catalyzed by the ubiquitin-specific protease USP1, in conjunction with UAF1 (USP1-associated factor 1). While monoubiquitination of FancD2 is essential for ICL repair, its deubiquitination by the USP1-UAF1 complex is also critical for a functional FA pathway [10–14]. Knockout of either USP1 or UAF1 in mice causes an FA-like phenotype [11, 12, 14] and USP1 disruption or the absence of de-ubiquitination abrogates FancD2 foci formation and ICL repair and increases sensitivity to DNA cross-linkers [10–13].
Several studies have documented an interaction between HPV and the FA pathway. First, loss of FancA or FancD2 lead to hyperproliferation of HPV+ hyperplasia and increased proliferation of HPV genomes in organotypic cultures [15, 16]. Loss of FancD2 potentiates E7 driven cancers of the female lower reproductive tract, and head and neck in two separate studies using mouse models [17, 18]. Second, FA patients are susceptible to oral and anogenital squamous cell carcinomas [19], though the involvement of HPV in these cancers is controversial because of inconsistent detection of HPV DNA [20–23]. Third, HPV+ head and neck cancer cell lines show greater sensitivity to cisplatin compared to HPV negative cells [24].
The molecular mechanism(s) by which HPV interacts with the FA pathway is not well-understood. HR-HPV (mainly E7) was shown to upregulate several FA genes and activate the FA pathway by increasing FancD2 monoubiquitination and foci formation [25–27]. In contrast, HPV oncogenes were reported to perturb the functions of several FA proteins, including ATR, BRCA2/FancD1, BRCA1/FancS, and Rad51/FancR [3, 28–30]. Because HPV increases sensitivity to DNA crosslinkers [24], and a functional FA pathway restricts HPV replication [16], and FancD2 is preferentially recruited to the HPV episome leaving the cellular genome unrepaired [31], we hypothesized that HPV oncogenes impair the FA DNA repair pathway to facilitate the HPV life cycle. To date, studies have shown that HPV increases monoubiquitination of FancD2/ FancI [25–27, 31], but none have studied their deubiquitination pattern, which is as important as monoubiquitination for ICL repair [10–13]. Our study shows that HPV E6 caused delayed deubiquitination of FancD2 though it increased FancD2 monoubiquitination. Similarly, HPV E6 and E7 expressing cells increased FancD2 foci formation but impaired FancD2 colocalization to double-strand breaks. Collectively, our data demonstrate that HPV16 oncogenes abrogate the FA pathway, which further supports the hypothesis that HPV inhibits cellular DDR causing genomic instability.
We have previously shown that β-HPV 5/8 E6 and HPV16 E6 expression increases sensitivity to crosslinking agents such as cisplatin, mitomycin C and UVB [28]. To determine if the expression of E7 could also increase crosslinker sensitivity, we expressed HPV16 E6 and E7 individually or together in primary human foreskin keratinocytes (HFKs). Expression of HPV16 E6 and E7 was confirmed by qRT-PCR, as well as by immunoblot for their established targets p53 and pRB (S2 Fig). We found that both E6 and E7 expressing cells are hypersensitive to cisplatin compared to LXSN control (Fig 1A). When IC50 values were compared, E6 and E7 expressing cells were respectively 3.4 and 2 times more sensitive to cisplatin than LXSN.
Genetic epistasis analysis was conducted to determine whether the cisplatin hypersensitivity observed in E6/E7 expressing cells is due to a defect in the FA pathway. For this, we knocked down FancD2 and assessed cisplatin sensitivity (Fig 1B and 1C). FancD2 protein depletion by siRNA was confirmed by immunoblotting. FancD2 knockdown in LXSN control cells resulted in further increased sensitization to cisplatin, whereas there was little/no effect on cisplatin sensitivity in E6 or E7 expressing cells (Fig 1C). These results suggest that cisplatin sensitivity of HPV oncogene expressing cells is due to a defect in the FA pathway, particularly at or downstream of FancD2.
The ability of E6/E7 expressing cells to repair cisplatin-induced ICLs was investigated by utilizing a modified comet assay which has been widely used to evaluate ICL repair in vivo at the single cell level [32–34]. Cells were treated with cisplatin for 2 hr and then incubated in cisplatin-free medium. At 24, 48 and 72 hours post-treatment, cells were harvested and frozen to analyze all samples concurrently. Cells were thawed and irradiated to deliver a fixed level of random DSBs immediately prior to the comet assay (Fig 2A). Crosslinks hold the two strands of DNA together during alkaline denaturation and retard electrophoretic mobility of the irradiated DNA resulting in a reduced tail moment compared to untreated irradiated controls. The tail moment, which takes into account both tail length and amount of DNA in the tail, was at a basal level in untreated and unirradiated cells (Fig 2B). At 0 hr post-cisplatin treatment, the tail moment was decreased in all cell types compared to corresponding irradiated untreated controls. At 72 hr post-cisplatin treatment, LXSN cells regained the tail moment indicating that these cells were able to repair crosslinks. In contrast, E6E7 expressing cells did not improve the tail moment 72 hr after cisplatin treatment, suggesting that ICLs remained unrepaired in these cells (Fig 2C). Further, repair kinetics of cisplatin-induced ICLs were expressed as the percentage of crosslinks remaining at the time points assessed (Fig 2C). ICLs present in the cisplatin-treated sample were calculated by comparing the tail moment of treated and irradiated (Cp.IR) cells with irradiated (IR) samples and untreated (Ø) control samples (as described in Methods). Cisplatin-induced ICLs were removed efficiently in LXSN cells with ~35% of the ICLs remaining at 72 hr, whereas in E6 or E7 expressing cells, significantly elevated levels of ICLs persisted and remained unrepaired even at 72 hr. Increased formation of % ICLs in E6/E7 cells at 24, 48 and 72 hr may be due to inefficient removal of the DNA-platinum monoadducts or intrastrand adducts by the repair system and their possible conversion to higher order adducts[35]. Taken together, these data support the hypothesis that HPV oncogene expressing cells have a decreased ability to repair cisplatin-induced ICL.
To screen for defects in the FA pathway, we first examined the levels of FancD2 in HPV oncogene expressing cells. When untreated cells were analyzed, total FancD2 (Ub + non-Ub) levels were significantly increased in E6/E7 expressing cells compared to LXSN control (Fig 3A). When E6 or E7 were individually expressed, cells had about 2 times more total FancD2 compared to LXSN; but the level increased by ~4 fold when E6 and E7 were expressed together. Total FancI (Ub + non-Ub) levels were also proportionately increased in E6/E7 expressing cells compared to LXSN.
FancD2 monoubiquitination was evaluated in E6/E7 expressing cells as a readout to define an activated FA pathway. Ub-FancD2 or Ub-FancI can be distinguished from the non-Ub form as a retarded mobility on gels. E6 and E7 increased FancD2 and FancI monoubiquitination both at baseline and after cisplatin treatment (Fig 3A). Approximately a 3-fold increase in the Ub-FancD2: non-Ub FancD2 ratio was observed in E6/E7 expressing cells on cisplatin treatment. Similar results were observed following MMC treatment and UVB irradiation (S2C and S2D Fig). When cells were analyzed following different lengths of cisplatin treatment, LXSN cells showed Ub-FancD2 after 6 hr of cisplatin treatment; but Ub-FancD2 was found in E6 cells even without cisplatin treatment (S2E Fig). Further, we performed cellular fractionation analyses and prepared chromatin and soluble fractions from HFKs expressing HPV oncogenes or LXSN control. As expected, Ub-FancD2 or Ub-FancI was enriched dramatically in the chromatin-bound fraction of all transduced HFK cells. Importantly, an increased recruitment of Ub-FancD2/ FancI to chromatin was detected in E6 expressing cells compared to LXSN control and E7 expressing cells (Fig 3B).
We next wanted to determine whether high levels of FancD2 and FancI in E6 expressing cells were due to increased transcription or greater protein stability or both. No significant differences in FancD2 mRNA levels between LXSN and E6/E7 expressing cells were observed (S3A Fig). To determine protein stability, cells were exposed to the translation inhibitor cycloheximide (CHX) and protein levels were monitored over a 24 h period (S3B). LXSN showed a rapid FancD2 turnover with half-life (t ½) of ~3 hr and diminished to low levels in 24 hr after CHX addition. In contrast, E6 expressing cells had the t ½ of >24 hr and FancD2 levels remained elevated 24 hr following the addition of CHX. One study reported that the mono-ubiquitination of FancD2 promotes its stabilization in chromatin and cells that are deficient in monoubiquitination has significantly reduced FancD2 protein half-life [36]. Increased FancD2 stability in E6 cells may be due to high levels of mono-ubiquitinated form of FancD2 in these cells. FancI mRNA levels were significantly elevated in E6/E7 expressing cells compared to LXSN (S3A Fig). Consistently, the turnover rate of total FancI was considerably delayed in E6 cells compared to LXSN (S3B-C).
We next sought to know the effectors that may contribute to increasing FancD2 monoubiquitination in E6/E7 expressing cells. Phosphorylation of FancI S556 occurs upstream of, and enhances, FancD2 monoubiquitination [7]. Additionally, the proliferating cell nuclear antigen (PCNA) is monoubiquitinated by the RAD18 ubiquitin ligase in response to ICL lesions. Apart from its role in translesion synthesis repair, ubiquitinated PCNA (PCNA-Ub) is known to promote FancD2 monoubiquitination by either facilitating FancD2 recruitment onto chromatin via a direct physical interaction [36] or by promoting the recruitment of FancL and its E3 ubiquitin ligase activity on FancD2 [37]. Further, the protein UHRF1 (ubiquitin-like with PHD and RING finger domains 1) has recently been identified as a sensor of ICLs and is required for the recruitment of FancD2 to ICL sites [38, 39]. Phosphorylated FancI-S556, Ub-PCNA, and UHFR1 levels were increased in E6 expressing cells compared to LXSN following cisplatin treatment (Fig 3C). Elevated UHRF1 levels in E6 cells was the consequence of upregulation of UHRF1 transcription and decreased protein turnover rate (S3A–S3C Fig).
Depletion of ATR by siRNA knockdown decreased the phosphorylation of FancI at S556 and therefore reduced the Ub-FancD2 level (S4A and S4D Fig), suggesting a role for p-556 FancI in increasing the chromatin-bound fraction of Ub-FancD2 in E6 cells. On the other hand, depletion of UHRF1 did not reduce Ub-FancD2 level in E6 cells (S4B and S4E Fig), indicating elevated UHRF1 may not be involved in increasing Ub-FancD2. When PCNA was depleted, FancD2 was mono-ubiquitinated even in untreated E6 cells (S4C and S4F Fig). A report shows that RAD18, which ubiquitinates PCNA, plays a significant role in monoubiquitination of FancD2 even in the absence of PCNA [40]. Consistent with this report, our data indicates that Ub-PCNA is not essential in increasing Ub-FancD2 level in E6 expressing cells.
To further investigate the interaction of HPV with the FA pathway, the ability of E6/E7 expressing cells to form nuclear foci of FancD2 was quantified as the percentage of cells with >5 FancD2 foci in HFKs. FancD2 foci formation in cisplatin-treated E6 or E7 cells was elevated compared to LXSN controls (Fig 4A and 4B). Even without cisplatin treatment, there was increased FancD2 foci formation in E6 expressing cells. Phospho-H2AX foci were used as markers for DNA double-strand breaks (DSB).
We previously reported that HPV oncogene expressing cells impair the colocalization of Rad51 with pH2AX [3]. The same approach was used to investigate whether E6/E7 expressing cells affect the localization of FancD2 to DSBs. The expression of HPV oncogenes caused FancD2 to be localized away from DSBs (Fig 4A). In cells expressing E6 or E6+E7, only ~50% of FancD2 appeared to co-localized with pH2AX. E7 expressing cells showed a modest (~25%) but statistically significant reduction in colocalization of FancD2 to pH2AX compared to LXSN (Fig 4C). To complement these co-localization studies, we utilized U2OS-DR cells transduced with LXSN, E6, E7 and E6E7 and transiently transfected with an I-SceI expression vector, as previously described [3]. These U2OS cells have clonally integrated DR-GFP cassette consisting of two copies of nonfunctional GFP (S5A Fig) [41]. Exogenous expression of I-SceI produces a single DSB within the first GFP gene which contains an I-SceI recognition site. Thus, cells with single large pH2AX foci were selected and inspected for its colocalization with FancD2 (S5A Fig). An excellent colocalization of FancD2 with pH2AX was observed in LXSN cells, but there was ~50% and 20% reduction in colocalization of FancD2 in E6 and E7 expressing cells, respectively compared to LXSN cells (Fig 4D and 4E). These data suggest that FancD2 is forming repair complexes by localizing away from DSBs in E6/E7 expressing cells. Because E6/E7 induce replicative stress (data in progress), one possibility is that FancD2 complexes are localizing to single strand DNA breaks or stalled replication forks.
Mislocalization of FancD2 in E6 cells (Fig 4D and 4E) may hinder the recruitment of downstream proteins, such as Rad51 to sites of DNA damage. In fact, several lines of evidence indicate that FancD2 helps in the localization of Rad51 to DSBs. First, Ub-FancD2 colocalizes with Rad51 in nuclear foci during S phase [42]. Second, FancD2 promotes recruitment of Rad51 in nuclear foci by directly binding and stabilizing Rad51-DNA nucleoprotein filament for DSB repair [43, 44]. Third, FancD2 colocalized with Rad51 in cisplatin treated HFK LXSN control cells (S5B Fig). We previously reported that the colocalization of Rad51 with DSB is impaired in E6 expressing cells [3]. To determine whether the mislocalization of Rad51 observed in E6 expressing cells is epistatic to FancD2, the Isce-I colocalization assay was repeated, as described above (Fig 4C), in FancD2-depleted cells and immunostained with Rad51 and pH2AX. Western blot analysis confirmed FancD2 protein depletion in the siFancD2-transfected cells compared to siControl cells (Fig 4F). FancD2 depletion in LXSN cells caused a modest (~20%) but statistically significant reduction in colocalization of Rad51 to pH2AX in I-SceI induced DSB (Fig 4G and 4H), supporting the idea that FancD2 promotes Rad51 recruitment to DSB in normal cells. However, in E6 expressing cells, there was no further significant reduction in colocalization of Rad51 to pH2AX on FancD2 knockdown, suggesting that the Rad51 recruitment defect associated with E6 cells is due to FancD2.
To further screen for defects in the FA pathway, we investigated how E6/E7 affects the de-ubiquitination pattern of FancD2/ FancI upon DNA repair. Though FancD2/ FancI monoubiquitination is considered as a functional activator of the FA pathway, de-ubiquitination of FancD2/ FancI is also critical for effective ICL repair [10–13]. Since E6/E7 increases monoubiquitination of FancD2/ FancI, there may be a defect in de-ubiquitination. To address this, cells were treated with cisplatin or exposed to UV and allowed to recover for the indicated time (Fig 5A). Delayed de-ubiquitination of FancD2 was observed in E6 expressing cells during recovery after UVB exposure or cisplatin removal (Fig 5A–5C). Most of FancD2 was de-ubiquitinated following 24 hr of UVB and cisplatin release in LXSN cells but not in E6 or E6+E7 expressing cells. E7 expressing cells behaved more like LXSN.
As Ub-FancD2 persists abnormally in E6 cells, a series of experiments were designed to determine the basis of the delay in de-ubiquitination. First, we asked whether this phenotype was a consequence of decreased levels of the FancD2 deubiquitinating enzyme, USP1. In fact, the opposite result was obtained: E6/E7 expressing cells showed elevated USP1 with cisplatin treatment, while basal levels of USP1 did not differ among untreated cells (Fig 5D). These results suggest that delayed FancD2 de-ubiquitination in E6 cells is not because of lower USP1 expression. Second, a recent study indicated that de-ubiquitination of FancD2/ FancI by USP1 occurs only when the complex is no longer bound to chromatin [45]. Delayed FancD2 de-ubiquitination in E6 expressing cells may be due to retaining a chromatin-bound conformation of FancD2/FancI after DNA damage. To address this possibility, the association of FancD2/FancI with chromatin was examined using cell fractionation following cisplatin withdrawal or UVB exposure (Fig 5E). 24 or 48 hr following UVB exposure, E6 or E6+E6E7 expressing cells showed increased chromatin-bound Ub-FancD2/ FancI, whereas LXSN and E7 expressing cells showed predominantly soluble non-Ub FancD2/I. There was also increased chromatin-bound Ub- FancD2/ FancI at 18 and 24 hr following cisplatin treatment in E6 expressing cells compared to LXSN control. These data suggest that the delayed de-ubiquitination pattern observed in E6 expressing cells was due to FancD2/ FancI being retained in a chromatin-bound conformation where USP1 cannot work efficiently to de-ubiquitinate the complex, despite high levels of USP1.
Third, while FancI phosphorylation at Serine 556 promotes FANCD2 monoubiquitination, its phosphorylation at Serine 565 inhibits FANCD2 de-ubiquitination and impairs ICL repair [7]. Therefore, we investigated whether delayed de-ubiquitination of FancD2 was also due to elevated levels of phosphorylated-S565 FancI in E6 expressing cells. As expected, E6 or E6+E7 expressing cells showed increased p-S565 FancI on cisplatin treatment (Fig 5F). As phosphorylation of FancI occurs through the ATR-mediated pathway [7] and ATR/CHK1 activation increases FancD2 monoubiquitination [9], delayed FancD2 deubiquitination in E6 cells may be due to persistently activated ATR. Therefore, we examined the levels of p-ATR, p- S345 CHK1, p-S565 FancI as well as FancD2 monoubiquitination/ deubiquitination patterns in cells which had recovered for 24 hr in normal media after cisplatin withdrawal. In E6 or E6+E7 expressing cells, ATR/CHK1 was activated and persisted following cisplatin withdrawal, compared to LXSN (Fig 5G). These results are consistent with persistent levels of both ubiquitinated FancD2 and S565-phosphorylated FancI in E6 expressing cells following cisplatin release. Further, persistence of pATR foci and pCHK1 following UV exposure was seen in E6 cells compared to LXSN (S6A and S6B Fig). In E6 expressing cells, an ATR specific inhibitor (VE821) following cisplatin withdrawal caused an increase in FancD2 de-ubiquitination (Ub:non-Ub ratio of 0.6 and 0.51 at 18 and 24 hrs of VE821 treatment compared to a ratio of 1.74 and 1.40 at same time points in normal media), S6C and S6D Fig. These data demonstrate that persistently activated ATR/pCHK1/pFancI signaling could contribute to the delayed de-ubiquitination of FancD2 in E6 cells. Fig 5H shows the potential mechanisms for delayed de-ubiquitination of FancD2 in E6 cells.
To determine whether the effect of E6 on increased monoubiquitination and delayed de-ubiquitination of FancD2 was a consequence of aberrant cell cycle progression, LXSN and E6 cells were synchronized in early S-phase by double-thymidine block, released, and FancD2 mono- or de-ubiquitination patterns were examined as cells progressed through the cell cycle (Fig 6). A previous study reported that FancD2 undergoes monoubiquitination during the S-phase of the cell cycle [4]. Consistent with this study, in both LXSN and E6 cells, monoubiquitinated FancD2 was the major isoform present upon release from double-thymidine arrest into G1/S border and S-phase. Ub-FancD2 was noticeable till late S-phase, just prior to the decline of cyclin A. It is known that USP1 deubiquitinates FancD2 when cells exit S phase [12]. Ub-FancD2 was de-ubiquitinated when both LXSN and E6 cells exited S phase. Cell-cycle analysis by flow cytometry was performed at each time point to track cell-cycle progression (Fig 6D). These data suggest that E6 does not abrogate cell cycle-specific process of FancD2 monoubiquitination and de-ubiquitination.
To further address the regulation of FancD2 mono or de-ubiquitination, the potential role of p53 was examined. The p53 tumor suppressor is a major cellular target of HPV16 E6. One study reported that p53 downregulates FancD2 mRNA and protein levels [46]. As E6 degrades p53, we expected to see increased FancD2 mRNA and protein levels in E6 expressing cells. However, no significant differences in FancD2 mRNA levels between LXSN and E6 cells were observed (S3A Fig). But, since there was increased protein levels of total FancD2 in E6 expressing cells (Fig 3A), we hypothesized that this phenotype was dependent on the ability of E6 to degrade p53. To test this, HFK cells expressing a mutant of HPV16 E6 (8S/9A/10T) that is incapable of degrading p53 [47, 48] was used. Western blot analysis confirmed that E6 expression degrades p53, but the mutant failed to induce p53 degradation (Fig 7A, upper panel). Both cell lines were also evaluated for E6 expression by RT-PCR (Fig 7A, lower panel). Total FancD2 level was similar in mutant E6 and wild-type E6 cells (Fig 7B), indicating that E6 increased FancD2 protein level independently of its effects on p53 degradation. To confirm these results, cells were treated with Nutlin-3a, a Mdm2 inhibitor that increases p53 activity by preventing Mdm2-mediated proteasomal degradation. Nutlin treatment decreased total FancD2 level in LXSN cells but not in E6 expressing cells (Fig 7C). Similar downregulation of FancD2 was observed in lung fibroblast cells upon Nutlin-induced p53 activation, with no significant change in FancD2 level in p53-deficient counterparts [46].
Several publications suggested that FancD2 monoubiquitination is p53-independent. Several p53 defective cancer cell lines and chicken B lymphocyte cells lacking functional p53 were found to be fully competent for FancD2 monoubiquitination [49–51]. Interestingly, Rego et al. showed that FancD2 mono-Ub is p53-independent but dependent on p21, its downstream target [49]. In our study, although E6 showed increased Ub-FancD2 compared to LXSN, this increment was similar in mutant E6 cells, which fail to degrade p53 (Fig 7B). Hence, increased monoubiquitination of FancD2 by E6 is not a direct consequence of p53 degradation. To confirm these results, we used p53 knockdown LXSN cells and found that the Ub-FancD2 level was unchanged when compared to p53-sufficient LXSN (S7 Fig).
The exact role of p53 in FancD2 de-ubiquitination is not well-understood, although a study reported that p21, a p53 downstream target, represses USP1 transcription [49]. Since E6 increased USP1 protein expression on cisplatin treatment (Fig 5D), we argued that the mechanism by which E6 causes delayed FancD2 de-ubiquitination may be related to its effects on p53 degradation. To examine this, we performed similar experiments as in Fig 5A in cells expressing the mutated E6 that is incapable of degrading p53. At 48hr after UV exposure, FancD2 de-ubiquitination pattern in mutant E6 was similar to LXSN control cells (Fig 7D, lanes 3 and 9). Similarly, after 24hr release from cisplatin treatment, mutant E6 showed de-ubiquitinated FancD2 (Ub: Non-Ub ratio of 0.67), whereas wild-type E6 cells had predominately monoubiquitinated FancD2 (ratio of 1.7) (Fig 7E). This indicates that delayed FancD2 de-ubiquitination in E6 cells was related to the ability to degrade p53. To confirm that the observed effects were specific consequences of p53 degradation, we used shRNA to stably knockdown p53 in LXSN cells and analyzed FancD2 Ub/de-Ub pattern upon UV exposure or cisplatin withdrawal. Once again, FancD2 de-ubiquitination was markedly delayed in the absence of p53 (S8B Fig). These results strongly support a p53-dependent effect of E6 in causing delayed FancD2 de-ubiquitination.
We next examined whether p53 dependent delayed FancD2 deubiquitination in E6 cells is due to activated ATR/CHK1/pFancI signaling. pATR/pCHK1 levels were reduced in cells expressing mutant E6 that cannot degrade p53 compared to wild-type E6 cells (S8C Fig). Reduction in ATR activity resulted in decreased phosphorylation of FancI at S565 and increased FancD2 de-ubiquitination in mutant E6 cells.
To further understand how E6 or E7 impair the FA pathway and cause genomic instability, the expressions of proteins downstream of FancD2/I complex were evaluated. As a key regulator of the FA pathway, FancD2 recruits and coordinates functions of downstream FA repair proteins [52], including BRCA2/FancD1, BRCA1/FancS, Rad51/FancR, Palb2/FancN and BRIP1/FancJ. Previously, we demonstrated that neither BRCA2, BRCA1 nor Rad51 protein expression is impaired by the HPV oncogenes [3]. To determine whether the abundance of any other downstream protein is affected by HPV oncogenes, the levels of Palb2 were measured. Palb2 level was reduced when E6 was expressed separately or together with E7 (Fig 8A). Reduced Palb2 protein in E6 expressing cells was not caused by downregulation of Palb2 transcription but was due to increased protein turnover (S9C and S9D Fig). E6 exhibited shorter half-life (t ½) of ~12 hr compared to LXSN with t ½ of >24 hr. Depleted Palb2 level in E6 cells was not a consequence of p53 degradation (S9B Fig). Because E6 reduced Palb2 levels, the ability of Palb2 to form nuclear foci in response to cisplatin was examined. In LXSN control cells, the percentage of cells with >5 Palb2 foci peaked at about 90% following cisplatin treatment. In contrast, significantly fewer (~60%) E6 expressing HFKs formed Palb2 foci on cisplatin treatment (Fig 8B). Although fewer Palb2 foci were observed in E6 cells, these foci perfectly localize to pH2AX in I-SceI transfected U2OS-DR system (S10A Fig). Because E6 reduces Palb2 expression and foci formation, and Palb2 is also a component of FA nuclear foci for ICL repair, we asked whether the cisplatin hypersensitivity observed in E6 expressing cells is due to a defect in Palb2. Genetic epistasis analysis for cisplatin sensitivity was conducted by knocking down Palb2. Palb2 depletion in LXSN and E7 cells resulted in further sensitization to cisplatin (Fig 8C), whereas there was no effect on cisplatin sensitivity in E6 and E6+E7 expressing cells. These results suggest that cisplatin sensitivity seen in E6 expressing cells is, in part, due to a defect in Palb2 expression and foci formation.
Reduced Palb2 expression and foci formation in E6 cells (Fig 8A and 8B) may affect the interaction of Palb2 with other FA repair proteins, such as Rad51, required for resolution of ICLs. In fact, Palb2 physically and functionally interacts with Rad51. First, Palb2 colocalizes with BRCA2 foci, which also includes Rad51 [53–56]. Second, loss of Palb2 impairs Rad51 focus formation [57] and there is Palb2-dependent loading of BRCA2-Rad51 repair machinery at DSBs [56]. Therefore, we next examined whether the defect in localization of Rad51 observed in E6 expressing cells is due to decreased Palb2 expression or foci formation. In all HFK cells, including E6 expressing cells, Palb2 depletion caused a dramatic reduction in colocalization of Rad51 to pH2AX in I-SceI induced DSB (S10B Fig) indicating that the Rad51 recruitment defect observed in E6 cells is not due to impaired Palb2 expression and foci formation.
The FA DNA repair pathway is involved in guarding genome integrity, especially when challenged by endogenous and exogenous DNA crosslinking agents. Cisplatin is the most commonly used chemotherapeutic drug for cervical and oropharyngeal cancers. Here we provide mechanisms by which HPV oncoproteins attenuate the FA pathway and contribute to tumorigenicity and the response to cisplatin therapies in HPV-associated malignancies. Our present study (Fig 1) along with others [24] confirm the cross-linker hypersensitivity in HPV oncogene expressing cells. We also show that E6/E7 expressing cells have a defect in repairing cisplatin-induced ICLs (Fig 2). The epistatic analysis confirmed that ICL sensitivity in E6/E7 cells was due to a defect in FancD2 or downstream of FancD2 (Fig 1). However, FancD2 monoubiquitination and foci formation were increased in E6 or E7 cells (Figs 3 and 4), which is consistent with the previous work conducted in high-risk HPV+ and HPV oncogene expressing cells [25, 27, 31]. Further, high risk, but not low-risk, E7 was shown to increase FancD2 foci formation and activate the FA pathway [26, 27]. Strikingly, Ub-FancD2 was seen without cisplatin treatment in E6 expressing cells, which might be due to the presence of activated ATR which increases FancD2 monoubiquitination [9]. HPV oncogenes though, promoted the initiation of FA pathway by increasing FancD2 monoubiquitination and foci formation but hindered the completion of the repair by multiple mechanisms. E6 or E7 caused the accumulation of monoubiquitinated FancD2 at sites away from DNA damage. This subsequently hindered the recruitment of downstream protein Rad51 to DNA damage sites in E6 cells (Fig 4D–4H). Further, E6 mediated p53 degradation does not hamper cell cycle specific process of FancD2 modifications but abrogate the repair by delaying FancD2 de-ubiquitination (Figs 5–7). In addition, E6 reduced the expression and foci formation of Palb2 (Fig 8). Though there was a defect in FancD2 localization to DSBs in E6 cells, no mislocalization of Palb2 was observed, which is in support of a recent study reporting that FancD2 and Palb2 localize to DSBs independently of each other [58].
HPVs have been shown to recruit numerous cellular repair factors to their replication centers, mainly for HPV amplification [59–61]. Similarly, HPV E1 has been shown to interact with USP1-UAF1 complex to facilitate viral replication [62]. We speculate that increased Ub-FancD2 and foci formation may create an environment where FA repair proteins are readily available to support HPV replication. Consistent with our speculation, a study showed that higher levels of FancD2 and larger foci are predominantly recruited to HPV DNA rather than cellular genomes and localize to viral replication centers [31]. The preferential recruitment of cellular repair proteins to HPV replication centers would facilitate viral replication but could leave chromosomal DNA unrepaired. Our present work shows how HPV delays the repair of ICLs and attenuates the FA DNA repair pathway. Mislocalization of FancD2 observed in our study may represent a fruitless attempt to bring the FancD2 repair protein to non-existing viral replication centers. This would attenuate ICL repair regardless of the presence or absence of the viral genome because FancD2 helps in the recruitment of downstream FA repair proteins, including Rad51 in nuclear foci at the site of DNA damage. This is supported by our data demonstrating that mislocalized FancD2 in E6 cells causes a reduction in Rad51 recruitment to DSBs (Fig 4G–4H).
FancD2 monoubiquitination is necessary for ICL repair, but, by itself, is not sufficient for an efficient FA pathway. In other words, having more monoubiquitinated FancD2 within a cell does not necessarily improve ICL repair. This is exemplified by studies demonstrating that USP1-mediated FancD2 de-ubiquitination is required for both FancD2 foci formation and ICL sensitivity [10, 13] and for a functional FA pathway [7, 12, 45]. HPV E6 caused the delayed de-ubiquitination of FancD2, impairing ICL repair (Fig 5). HPV E6 cells showed increased phosphorylated S565 FancI and persistently activated ATR/CHK-1, which may, in part, cause delayed FancD2 de-ubiquitination in these cells. The delayed de-ubiquitination pattern observed in E6 expressing cells may be also due to a block at the step of release of FancD2 from chromatin which hinders USP1 deubiquitination activity.
ATR/CHK1 signaling is a key step in activating the FA pathway [5]. ATR is activated when the replication fork encounters DNA damage. ATR then phosphorylates several substrates, including FancD2 and FancI as well as members of the FA core complex. ATR-mediated FancI phosphorylation at Serine 556 (called ubiquitination-independent phosphorylation) occurs predominantly upstream of, and promotes, the monoubiquitination of FancD2/ FancI [7, 63]. On the other hand, FancI Serine 565 phosphorylation (called ubiquitination-linked phosphorylation) occurs downstream of Ub-FancD2/I and inhibits FancD2 de-ubiquitination [7]. Augmented ATR activity in E6 expressing cells resulted in increased phosphorylation of FancI at both sites S556 and S565. Increased p-S556 FancI promotes the monoubiquitination of FancD2, but at the same time elevated p-S565 FancI inhibits de-ubiquitination.
Together, we provide a mechanistic framework for HR-HPV oncogenes disruption of the FA DNA repair pathway (S11 Fig). This study advances our understanding of HPV tumorigenesis as well as tumor sensitivity to cisplatin during chemotherapy. Our study suggests a general model for the progression of HPV associated cancers. Over the multi-decade course of HPV-induced tumorigenesis, persistent expression of HPV oncogenes impairs the FA pathway. This leads to cells with destabilized genomes, allowing for rapid tumor progression. Previous studies are in support of the model that dysregulated FA pathway contributes to the development of cancers in patients without FA [64–66] and with FA [67, 68]. An extremely high incidence of cancer in FA patients [69] further suggest that the inactivation of FA pathway results in tumor progression.
Our work also has important therapeutic implications. Because continued expression of HPV oncogenes is required for cancer development, most HPV-associated tumors likely have defective FA pathway. These defects result in cisplatin hypersensitivity and demonstrate the mechanisms underlying the therapeutic efficacy of cisplatin in HPV-associated cancers. This also explains the reason underlying the better response rates of HPV+ oropharyngeal cancers than HPV negative head and neck cancers to cisplatin treatment. We believe that tumors that are cisplatin resistant, may have adapted other unexplored mechanisms to escape these defects. One possibility is that some cells with an intact FA pathway are positively selected during cancer progression, resulting in the growth of a cisplatin-resistant tumor.
The use of deidentified neonatal human foreskins for the study was approved by the Institutional Review Board at the Swedish Medical Center (Seattle, WA). Tissues from newborn circumcisions were collected from Swedish First Hill Birth Center, Seattle, WA.
Primary human foreskin keratinocytes (HFKs) were generated from deidentified neonatal human foreskins and grown in EpiLife Medium supplemented with 60 μM calcium chloride (ThermoFisher Scientific, MEPI500CA) and human keratinocyte growth supplement (ThermoFisher Scientific, S0015). HFKs derived from multiple donors were used to repeat the results. U2OS DR-GFP cells (a gift from Maria Jasin) were grown in DMEM supplemented with 10% FBS. U2OS DR-GFP cells contain a single integrated copy of the DR-GFP cassette [70].
Retrovirues were produced in 293T cells (Thermo Fisher Scientific) using plasmid constructs (LTR/VSV-G, CMV/tat, pJK3 and pLXSN vector- empty, E6, E7 or E6E7, or E6 8S/9A/10T mutant) and TransIT-293 transfection reagent (Mirus Bio) according to manufacturer’s protocol. Following retroviral transduction and G418 selection (50ug/ml) in HFKs or U2OS-DR cells, the expression of HPV16 oncogenes E6 and E7 was confirmed by qRT-PCR of 16 E6 and E7 and immunoblot to p53 and pRB (well-characterized cellular targets). The expression of p53 in HFKs was suppressed using transduction-ready p53 shRNA lentiviral particles (Santa Cruz Biotechnology, sc-29435-V). HFKs were transduced with lentiviral stock in media containing 10ug/ml polybrene. Following 48 hr of transduction, stably transduced cells were selected in 0.5 ug/ml puromycin and the efficiency of knockdown was monitored using immunoblot to p53.
Cells were treated with cisplatin (Selleck Chemicals, S1166), Mitomycin C (MMC) (Sigma, M4287), Nutlin-3a (Selleck Chemicals, S8059), VE-821 (Selleck Chemicals, S8007), cycloheximide (Millipore 239763), Ionizing radiation (IR) (GammaCell Cesium Irradiator -1000) or UVB (two FS20t12/UVB bulbs, Solarc Systems, Inc.). Cisplatin was dissolved at the stock of 1.5mM in 0.9% NaCl (sterile), Nutlin-3a and VE-821 at the stock of 10mM in DMSO, cycloheximide at the stock of 50mg/ml in DMSO, aliquoted and kept at -20° C protected from light.
Primary antibodies against p53 (Cell Signaling Technology, 9282), pRb (BD Pharmingen, 554136), FancD2 (Santa Cruz Biotechnology, FI17, sc-20022 or Abcam, ab2187), FancI (Santa Cruz Biotechnology, A-7, sc-271316), Phospho-Ser556 FancI and Phospho-Ser565-FancI (gifts from Ronald Cheung, Toshiyasu Taniguchi lab), pH2AX (Millipore, 05–636), Rad51 (Cosmo Bio Co. Ltd, 70–001), Palb2 (Bethyl Laboratories, A301-246A), Palb2 (a gift from Bing Xia, Rutgers Cancer Institute), Phospho-ATR (Ser428) (Cell Signaling Technology, 2853), Phospho-Chk1 (Ser345) (133D3) (Cell Signaling Technology, 2348), ATR (Cell Signaling Technology, 2790), CHK1 (2G1D5, Cell Signaling Technology, 2360), PCNA (PC10) (Cell Signaling Technology, 2586), Ubiquityl-PCNA (Lys164) (Cell Signaling Technology, 13439), UHRF1 (Santa Cruz Biotechnology, H-8, sc-373750), USP1 (C-term; a gift from Tony Huang, NYU School of Medicine), Cyclin A (in-house, Clurman lab), Beta actin (GeneTex, GTX110564 or Cell Signaling Technology, 5125), Vinculin (Sigma, V9131), and Histone H3 (Abcam, ab1791), GAPDH (14C10, Cell Signaling Technology, 2118)were used. Secondary antibodies were conjugated with horse radish peroxidase (mouse or rabbit, Cell Signaling Technology) or appropriate Alexa Fluor (Molecular Probes) were used.
Whole cell lysates were prepared by directly lysing cell pellets in SDS sample buffer (0.05 M Tris-HCl, (pH 6.8), 2% SDS, 6% β-mercaptoethanol) and boiling for 5 min. Cell fractionation was performed to obtain soluble and chromatin-bound fractions, as described [71]. Briefly, cell pellets were re-suspended in CSK + 0.1% Triton-X buffer (10mM PIPES, pH = 6.8, 100mM NaCl, 1mM EGTA, 1mM EDTA, 300mM Sucrose, 1.5mM MgCl2, 0.1% Triton-X-100 and protease inhibitors) and incubated on ice for 5 mins. After centrifugation (1500g for 5 min), the supernatant was collected and stored (soluble fraction). Pellet (Chromatin-bound fraction) was washed twice in CSK buffer and then resuspended in SDS sample buffer and boiled for 5 min. Samples were stored at -20 °C until quantification and immunoblotting. Proteins were quantified using Bradford protein assay (Bio-rad) and processed using NuPAGE LDS sample buffer (ThermoFisher Scientific) for loading equal amounts (15ug) onto the gels. SDS-PAGE electrophoresis was done using NuPAGE 3–8% Tris-acetate or 4–12% Tris-glycine gels (ThermoFisher Scientific) and proteins were transferred to Immobilon-P PVDF membranes (Millipore). Primary antibodies were diluted in blocking buffer (4% milk in TBS-Tween 20) or in 3% BSA in TBST for phosphorylated proteins and incubated overnight at 4 °C. Horseradish peroxidase (HRP)-conjugated anti-mouse or anti-rabbit secondary antibodies were used. Blots were developed using Clarity Western ECL substrate (Bio-Rad) and images were acquired using ChemiDoc MP Imaging System (Bio-Rad) and processed using Image Lab Software (Bio-Rad). Band quantification was performed using ImageJ, normalizing with a loading control.
qRT-PCR was conducted to evaluate the expression of HPV16 E6, E7, and GADPH genes, as described previously [3]. To evaluate the expression of FancD2 mRNA, two sets of primers were designed to amplify the longer FancD2 isoform from transcript sequence NM_033084.4. The first primer set adapted from Jaber et al. [46] was Forward 5’-AGACTGTCAAAATCTGAGGATAAAGAGA-3’ and Reverse 5’-TGGTTGCTTCCTGGTTTTGG-3’; and next set was designed as Forward 5’-CATGGCTGTTCGAGACTTC-3’ and Reverse 5’-CACAAAGAGACGCCCATAC-3’. Other primers used were FancI: Forward 5’-TGGCGGAGTTCTGTGATATGAG-3’ and Reverse 5’-CAGAGCAGGGGGAACCTTTG-3’, Palb2: Forward 5’- GCTCTTTTCGTTCTGTCGCC-3’ and Reverse 5’ TCTCCTTTAACTTTTCCTTCTCCTC-3’, and UHRF1: Forward 5’- TTCCCGCCGACACCAT-3’ and Reverse 5’ TCCTCCATCTGTTTGCCCC-3’.
To compare protein stability or turnover, cells were treated with 50 ug/ml cycloheximide (CHX), an inhibitor of protein biosynthesis and the protein levels were monitored over a 24 hr period. Cell lysates were collected at indicated time points followed by immunoblotting for FancD2, FancI, UHRF1 and Palb2. GAPDH was selected as a loading control whose level did not change following the addition of cycloheximide till a 24 hr period.
Cell survival over a range of cisplatin concentrations was measured using a crystal violet absorbance-based assay [72]. Briefly, cells were seeded on 24-well plate overnight and treated with cisplatin at indicated doses for 3 days. Cells were then washed once with PBS and fixed for 10 min at room temperature in 10% methanol and 10% acetic acid. Adherent cells were stained with 0.5% crystal violet in methanol. Plates were rinsed in water and allowed to dry completely. The adsorbed dye was re-solubilized with methanol containing 0.1% (wt/vol) SDS by gentle agitation for 2 hr at room temperature. Dye solution (200 μl) was transferred to 96-well plates and diluted (1:2) in methanol. Absorbance was measured at 595 nm. Cell survival was calculated by normalizing the absorbance relative to untreated controls.
siRNAs were transfected at a final concentration of 20nM in 6-well or 10 cm plates using Lipofectamine RNAiMAX (Invitrogen), following the manufacturer’s Reverse Transfection protocol. I-SceI colocalization and crystal violet assays were conducted, respectively 24 and 48 hr after siRNA transfection. The following siRNAs were used: si-FancD2: 5’- AACAGCCAUGGAUACACUUGA-3’; si-Palb2 (Santa Cruz Biotechnology, sc-93396), siATR (Santa Cruz Biotechnology, sc-29763), siPCNA (Santa Cruz Biotechnology, sc-29440), siUHRF1 (Santa Cruz Biotechnology, sc-7680) and siRNA Universal Negative Control#1 (Sigma, SIC001). Knockdown was confirmed by immunoblotting.
HFKs and U2OS DR-GFP cells (transduced with LXSN, 16 E6, 16 E7 and 16 E6E7) were grown in chamber slides overnight. Cells were either left untreated or treated with cisplatin or transfected with I-SceI expression vector (Addgene, 26477). Cells were then washed with 1X PBS and fixed and co-permeabilized using 2% paraformaldehyde and 0.5% Triton-X in PBS for 30 mins. After washing, cells were blocked with 3% BSA and 0.1% Tween20 in PBS for 1 hr. Cells were then subsequently incubated with appropriate primary antibodies (indicated in the text) in a moist chamber at 4° C overnight. The following day, cells were washed 3 times with 1X PBST (1X PBS + 0.1% Tween20) for 5 min each and then incubated for 1 hr at room temperature with Alexa Fluor secondary antibodies in dark. Primary and secondary antibodies were diluted in blocking buffer. Following secondary antibody incubation, cells were washed and mounted using coverslip in ProLong Diamond Antifade Mountant with DAPI (Molecular Probes). Cells were then visualized using a DeltaVision Elite confocal microscope (Applied Precision). The images were deconvolved using the Deltavision SoftWoRx program and were analyzed using ImageJ. Each experiment was repeated at least 3 times independently.
I-SceI colocalization assay in U2OS-DR cells (transduced with LXSN, E6, E7 or E6E7) were performed as described previously [3]. Briefly, cells were transfected with I-SceI expression vector using TransIT-LT1 reagent (Mirus) and were stained for pH2AX and the indicated protein. In the case of gene knockdown experiments, reverse siRNA transfection was done a day before transfecting I-SceI plasmid. A single large pH2AX foci created by I-SceI expression was inspected for its colocalization with indicated repair protein.
Repair of interstrand crosslinks was assessed by an alkaline comet assay [32–34] using Trevigen’s CometAssay (4250-050-K), following the manufacturer’s protocol with some modifications. Cells were treated with 1.5 uM cisplatin for 2 hr. At the end of treatment, cells were washed with PBS and incubated in fresh medium for the required post-incubation time or harvested immediately (time 0 h). To process all the samples concurrently and to eliminate assay variability, cells were harvested and cryopreserved in media containing 10% DMSO and 50% FBS, prior to performing comet assay. Immediately before analysis, cells were thawed in ice-cold DPBS and irradiated with 12 Gy gamma irradiation (using GammaCell Irradiator GC-1000) to introduce a fixed number of random DNA strand breaks and processed according to Trevigen’s alkaline Comet assay procedures. Briefly, irradiated or unirradiated cells were immediately plated in low melting agarose on slides. After hardening of the agarose gel, cells were lysed and subjected to alkaline electrophoresis for 30 min at 4° C. After air-drying, DNA was visualized using SYBR Gold and images were captured using TissueFAXS machine using FITC filter. Images were analyzed using OpenComet Plugin tool in NIH ImageJ [73]. Default background correction was used. The output images resulting from the automatic analysis were manually reviewed to eliminate outliers and select uniform comets. Updated tail moment values (Mean ± SE from >350 comets) were then used to calculate % ICL remaining.
The degree of DNA ICLs present in the cisplatin-treated sample was calculated by comparing the tail moment of cisplatin-treated and irradiated (Cp.IR) samples with irradiated untreated (IR) samples and unirradiated untreated (Ø) control samples. The level of ICL is proportional to the decrease in the tail moment in the irradiated cisplatin treated sample compared to the irradiated untreated control. To quantify ICL repair (or the percentage of ICL remaining), we employed the following formula:
%Decreaseintailmoment(orICLremaining)=[1−(Cp.IR–Ø)/(IR–Ø)]x100
where Cp.IR = mean tail moment of cisplatin-treated and irradiated cells; Ø = mean tail moment of untreated and unirradiated cells, and IR = mean tail moment of irradiated and untreated cells.
The data were expressed as the percentage of ICLs that remained at a specific time point where 0 hr was normalized to 100%.
Cells were synchronized at G1/S phase boundary by double-thymidine block and cell pellets were harvested for immunoblotting and flow cytometry analysis as described previously [4, 74], with some modifications. Briefly, HFK cells were treated with 2 mM thymidine (Sigma-Aldrich) in EpiLife media without growth supplements for 16 hours. Cells were then washed twice with PBS and released/grown in thymidine-free complete EpiLife media for 9 hours. Thereafter, cells were treated again with 2 mM thymidine in growth supplement-free EpiLife media for another 16 hours. Cells were washed twice with PBS and then released in complete media and harvested every 3 hours after release. Synchronized cells were analyzed at different time points by immunoblotting and analyzing DNA content by DAPI staining. Approximately 10,000 cells were analyzed using flow cytometry (BD FACS-Canto II), and flow histograms were generated using FlowJo software.
All statistical analyses were done using Student’s t-test (Unpaired two-tailed) in GraphPad Prism. P value < 0.05 was considered significant. Statistical significance at P < 0.05, P < 0.01, P<0.001 and P < 0.0001 are indicated as *, **, ***, **** respectively.
|
10.1371/journal.pgen.1006387 | Genetic Analysis of Collective Motility of Paenibacillus sp. NAIST15-1 | Bacteria have developed various motility mechanisms to adapt to a variety of solid surfaces. A rhizosphere isolate, Paenibacillus sp. NAIST15-1, exhibited unusual motility behavior. When spotted onto 1.5% agar media, Paenibacillus sp. formed many colonies, each of which moved around actively at a speed of 3.6 μm/sec. As their density increased, each moving colony began to spiral, finally forming a static round colony. Despite its unusual motility behavior, draft genome sequencing revealed that both the composition and organization of flagellar genes in Paenibacillus sp. were very similar to those in Bacillus subtilis. Disruption of flagellar genes and flagellar stator operons resulted in loss of motility. Paenibacillus sp. showed increased transcription of flagellar genes and hyperflagellation on hard agar media. Thus, increased flagella and their rotation drive Paenibacillus sp. motility. We also identified a large extracellular protein, CmoA, which is conserved only in several Paenibacillus and related species. A cmoA mutant could neither form moving colonies nor move on hard agar media; however, motility was restored by exogenous CmoA. CmoA was located around cells and enveloped cell clusters. Comparison of cellular behavior between the wild type and cmoA mutant indicated that extracellular CmoA is involved in drawing water out of agar media and/or smoothing the cell surface interface. This function of CmoA probably enables Paenibacillus sp. to move on hard agar media.
| Motility is crucial as it enables bacteria to seek out and colonize suitable environments. Here, a new isolate from the rhizosphere, Paenibacillus sp., exhibited unusual motility behavior. When spotted on low wet, hard agar media, the bacterium formed many colonies, each of which moved around actively like an individual organism. The cells in moving colonies had a large number of flagella, which drove colony movement. Newly identified large extracellular protein was essential to form moving colonies on hard agar media. This protein seems to facilitate motility by drawing water out of agar or smoothing the cell surface interface. On encountering a wet environment, the moving colonies disassembled quickly, and individual cells swam in the water layer, suggesting that moving colonies specially form under low wet conditions. The results describe a novel mechanism that explains how Paenibacillus sp. overcomes environmental challenges by moving on solid surfaces.
| Migration is a critical mechanism by which bacteria survive and thrive in a particular environment. Motility enables bacteria to search for nutrients, avoid toxic compounds, and seek out favorable environmental niches that they can then colonize. The organelles responsible for mobility, flagella, are common in bacteria [1]. The most common form of flagella-dependent motility, called swimming motility, only works in an aqueous environment. However, bacteria live not only in aqueous environments but also on a variety of biotic and abiotic solid surfaces. Therefore, many bacteria have developed mechanisms that facilitate movement on a solid surface. These include swarming, twitching, gliding, and sliding motility, which are mediated by flagella, Type IV pili, focal adhesion complexes, surface active molecules, or the expansive forces generated by growing cells [2–4].
Swarming motility is defined as flagella-driven “group” movement across a solid surface [3, 5, 6], and is observed in several bacterial families [5]. Swarming motility is clearly distinct from swimming motility, which is the flagella-driven movement of “individual cells” in an aqueous environment. Indeed, under laboratory conditions, swarming motility is usually observed in solid media containing agar at concentrations above 0.5%, whereas swimming motility is observed in liquid or solid media containing agar at 0.3% or lower [2,5]. Since the motion of flagella pushes the cell forward against the surrounding water, surface water is a critical element for swarming motility as well as for swimming motility. However, water in hard agar media is usually trapped within the agar matrix. To overcome this, swarmer cells attract water to the surface from the agar matrix [7–9]. A high cell density, cellular secretions, and flagella rotation help to attract water to the surface [7–11]. Swarming motility requires differentiation into specialized cells, which often exhibit hyperflagellation, cell elongation, and the secretion of wetting agents that attract water or reduce surface tension. Swarming cells actively move in a fluid layer within swarm colonies, and often form small moving groups, called rafts, in which the cells closely aligned along their long axis [3, 5, 6]. The formation of rafts facilitates movement on hard agar media partly by reducing viscosity/drag on individuals [12], but its mechanism and function are still unclear. Since rafts are unstable and frequently change their members and shape, no substance or matrix appear to maintain rafts [5, 6].
Paenibacillus is a genus of facultative anaerobic, Gram-positive, spore-forming bacteria that are closely related to the genus Bacillus [13]. Paenibacillus spp. are isolated from various environments, and are often associated with plants [14–18]. Many strains of Paenibacillus spp. stimulate plant growth; for example, by assisting nutrient acquisition through nitrogen fixation and by producing cytokinins, peptide antibiotics, and volatiles that elicit a defense response [15, 19–22]. Certain species of Paenibacillus, such as P. vortex, P. alvei, and P. dendritiformis exhibit motility that is sufficiently robust to enable movement across the surface of media containing >1.5% agar [23–25]. Although the ecological role of Paenibacillus bacteria is largely unknown, their strong motility ability probably provides considerable advantages with respect to colonization of their natural habitats. Moreover, as P. vortex also facilitates the spread of other non-motile microbes and fungi that bind to P. vortex colonies [26, 27], robust motile Paenibacillus bacteria probably contributes to the dispersal of other microbes in their habitats.
Colonies formed by robust motile Paenibacillus bacteria form intricate patterns on agar media. P. vortex forms a highly branched colony pattern on agar media [24]. When grown on hard agar media, hundreds to millions of P. vortex cells assemble and generate rotating colonies (also referred to as wandering colonies). These rotating colonies move forward, leaving behind trails of cells; the latter form the ‘branches’ of the colony pattern. P. alvei forms a nebula colony pattern on agar plates, which comprises randomly scattered clusters of bacteria [23]. Although its colony pattern is quite different from that of P. vortex, P. alvei also forms wandering colonies [3], indicating that P. vortex and P alvei share common mechanisms for motility. P. dendritiformis forms a chiral branching colony pattern, where cells are aligned parallel to their neighbors [25]. At the tips of the growing branches, each cell moves back and forth along their neighbor [28]. Since these cells do not form wandering colonies, the mechanism underlying the motility of P. dendritiformis appears to be different from those of P. vortex and P. alvei. The molecular mechanisms underlying motility and colony pattern formation are very interesting but are unclear because these Paenibacillus bacteria are not amenable to genetic manipulation. For instance, although flagella are thought to produce the driving force, this remains unproven.
Here, we describe the motility behavior of a genetically tractable isolate of Paenibacillus sp. NAIST15-1 and its genetic analysis. Paenibacillus sp. showed peritrichous hyperflagellation in response to growth on hard agar media. Hyperflagellated cells formed moving colonies that were able to migrate across the surface of solid media containing >1.5% agar. We found that the cmoA gene (colony movement A) was necessary for motility on hard agar media. CmoA is a large extracellular protein that envelops moving cells clusters. CmoA is conserved in Paenibacillus bacteria that form moving colonies. The cmoA mutant were unable to move on 1.5% agar media but able to spread on 0.6% agar media. However, its cellular behavior on 0.6% agar media was quite different from that of the wild-type strain. We discuss possible roles of CmoA in motility on hard agar media.
Paenibacillus sp. NAIST15-1 (hereafter referred to as Paenibacillus sp.) was isolated from spore-forming bacteria that showed antagonistic activity against a plant pathogen Fusarium oxyosporum from weed roots and associated soil in the course of the previous study [29]. The bacterium produced swollen sporangia (S1 Fig), which is typical of the genus Paenibacillus [13]. Consistent with this, the 16S rDNA sequence of this bacterium was quite similar to those of Paenibacillus bacteria (S2 Fig); it showed 99% identity with that of Paenibacillus alvei. These observations indicated that the bacterium belonged to the genus Paenibacillus. Paenibacillus sp. exhibited unusual colony formation. When inoculated onto the center of plates containing 2×YT/1.5% agar media, Paenibacillus sp. spread over the surface and formed many discrete colonies (Fig 1A). This colony scattering phenotype appears similar as that observed for Paenibacillus alvei, and is described as either “wandering colonies” or “nebula pattern formation” [3, 23]. We were interested in this phenotype and a molecular mechanism behind it. Our isolate Paenibacillus sp. was suitable for the analysis because it was amenable to genetic manipulation (Methods, see below). The colony spreading pattern was greatly affected by the agar concentration (Fig 1A). Paenibacillus sp. formed a featureless mat all over plates containing 0.3% or 0.5% agar, but formed a mat within the central portion of 1% agar plates, with multiple colonies around it. Colony spreading was partially inhibited on 2.0% agar, and completely inhibited when the agar concentration exceeded 2.5%. We then examined the behavior of motile cells at the leading edge zones of colonies grown on 0.3%, 0.5%, and 1.5% agar media by light microscopy (Fig 1B). Because cells grown on 0.3% agar media moved independently in three dimensions within a thick fluid layer, we were not able to obtain well-focused images. The behavior looks like swimming motility. Cells on 0.5% agar media were observable in a single plane, and often formed small groups that moved together. These groups were unstable and frequently changed their members and shape during movement. On 1.5% agar media, most cells formed groups in which the cells were closely aligned along their long axis; these cells often formed bundles (Fig 1B). The number of cells in these clusters varied from a few to over thousands, and the larger clusters contained stacks of cells (Fig 1B; lower-right panel). Neither these densely packed clusters nor large clusters were observed at the leading edge zones of colonies grown on 0.3% and 0.5% agar media. Among these clusters, larger clusters moved forward smoothly, while single cells and small clusters moved very slowly or not at all (S1 Movie and S2 Movie). Thus, cluster formation appears to facilitate cell movement on the surface of 1.5% agar media. While the clusters were moving forward, some cells often fell away from the clusters, suggesting that these detached cells comprise static single cells and small clusters at the leading edge zone. Swarm bacteria often exhibit elongated cellular morphology during swarming motility [5, 6, and references therein]. However, Paenibacillus sp. did not show elongated cell morphology on 1.5% agar media.
We next investigated how the formation of moving cell clusters led to the colony scattering phenotype on 1.5% agar. For this purpose, a suspension of Paenibacillus sp. was spotted onto 1.5% agar and the process of colony formation analyzed under a stereo microscope (S3 Movie). During the first few hours post-inoculation, multiple tiny colonies appeared at the inoculation site, which then, while growing, moved around actively over the surface of the media. The speed of the moving colonies was 3.6 μm sec-1 (an average of 12 colonies was monitored for 10 min). Each moving colony appeared to have polarity; a fixed forward region led colony movement. The shape of moving colonies was not constant and sometime became very long. Moving colonies sometimes coalesced and divided. Some colonies then began to rotate and form vortices. Vortex formation began at the inoculation site, which contained a high cell density, and all colonies formed vortices as the colony density increased. Both clockwise and counterclockwise vortices were observed. The rotation speed of the vortices gradually decreased until static colonies were formed. These observations show that moving microscopic clusters of cells grow up to form visible colonies that move actively on hard agar (Fig 2). We called these moving colonies “wandering colonies”, as first described by Henrichsen [3]. Each wandering colony then rotated and finally became a static round colony (Fig 2). These unusual behaviors are the reason why Paenibacillus sp. forms many discrete colonies on hard agar media. Moving colonies have also been described for Paenibacillus vortex, which forms a highly branched colony pattern on hard agar media [24]. At the tips of growing branches, tens to thousands of cells form a spiraling colony that moves forward on hard agar media [24]. However, unlike for Paenibacillus vortex, Paenibacillus sp. colonies moved forward without spiraling as wandering colonies; the formation of spiraling colonies by Paenibacillus sp. was observed in transition from the active moving phase to the sessile phase (S3 Movie). Thus, colony movement and spiraling are separate entities in Paenibacillus sp. at least under the growth conditions tested herein.
The entire genome of Paenibacillus sp. was sequenced using a massively parallel sequencing platform, MiSeq (Illumina). A total of 1.7 Gb of 300 base paired-end reads were assembled into 42 contigs (>1 kb) using CLC genomic workbench ver. 6.5 (CLC bio, Qiagen). The N50 value, which is a statistical measure of average length of contigs, was 264,241 bp and the longest contig was 824,207 bp. The draft genome of Paenibacillus sp. contains 6,768,284 bp, with a G+C content of 46.3%. The genome comprises 6,034 coding sequences, 5 ribosomal RNA partial sequences, and 78 transfer RNAs.
Flagellar genes were identified at five different loci within the Paenibacillus sp. genome. The largest cluster was the 30 kb fla/che operon, which contains genes for the flagellar hook-basal body complex, chemotaxis proteins, and an alternative sigma factor, σD (Fig 3). Another large flagellar gene cluster contains genes that encode anti-σD FlgM, the CsrA translational regulator, flagellin, and the filament cap. The other loci were the flhOP operon, which is probably required for hook assembly, and two stator operons (motAB and motCD). The composition and organization of these flagellar gene clusters were very similar to that of a Gram-positive model bacterium, Bacillus subtilis (Fig 3). These bacteria are expected to have a similar system for flagellar formation. However, B. subtilis has SwrA, which is required for transcriptional activation of the fla/che operon [30], but no swrA homolog gene was found in the Paenibacillus sp. genome. swrA is located between ftsEX-ctpB and uvrBA in the B. subtilis genome. Interestingly, these homologs are found upstream and downstream of motAB in the Paenibacillus sp. genome (S3 Fig).
Type IV pili are involved in motility in some bacteria [31]. Here, pili-related genes were also identified at two loci (S4A Fig). These gene products were similar to Flp pilus assembly proteins, Tad pilus assembly protein, and pre-pilin peptidase. However, our homology search did not identify all of the genes involved in Type IV pili biosynthesis, such as genes for pilin, and inner membrane core protein.
Genome sequencing suggests that Paenibacillus sp. possesses at least two potential organelles for motility: flagella and pili. To determine which might be responsible for motility, we carried out gene disruption. Flagellar genes fliF and hag encode the M ring of the flagellar basal body and flagellin, respectively. We constructed bacteria harboring in-frame deletions of fliF or hag using the pMAD plasmid [32]. The growth rates of the fliF and hag mutants in 2×YT liquid media were indistinguishable from those of the wild-type strain (S5 Fig). The motility of these disruption mutants was then tested on media containing agar concentrations varying from 0.3% to 1.5%. Disrupting fliF and hag completely abolished motility on all agar media (Fig 4). The deleted fliF and hag alleles were then restored using a pMAD plasmid carrying the wild-type sequence of fliF or hag. The resulting complemented strains exhibited normal motility (comparable with that of the wild-type strain) (S6 Fig). On the other hand, deletion of putative pili genes (PBN151_298, PBN151_299, PBN151_1478, and PBN151_1479) did not affect motility on 2×YT plates (S4B Fig). These results clearly show that flagella are required for motility on both soft and hard agar media.
Flagellar rotation depends on the stator protein complex [33]. Paenibacillus sp. possesses two operons for stator proteins, motAB and motCD. MotA and MotC are 35.6% identical, whereas MotB and MotD are 30.6% identical (S7A Fig). To test whether flagellar rotation is required for motility, we constructed disruption mutants of these operons. Whereas disrupting motAB did not affect motility on all agar media tested, disrupting motCD prevented motility on 1.0% and 1.5% agar media (Fig 4). Disrupting both motAB and motCD completely abolished motility on all agar media tested (Fig 4). We then cloned motAB or motCD in the multicopy plasmid pHY300PLK [34] and carried out complementation tests. Introduction of multicopy motAB (pHYmotAB) only partially restored motility to the motCD mutant, whereas introduction of pHYmotCD fully restored motility to the motCD mutant (Fig 5). We also carried out complementation tests using the motABmotCD quadruple mutant. Introduction of pHYmotAB restored motility to the motABmotCD mutant only on 0.3% and 0.5% agar media whereas introduction of pHYmotCD restored motility to the motABmotCD mutant on all agar media tested (Fig 5). These results indicate that either motAB or motCD is sufficient for motility on 0.3% and 0.5% agar media, but that only motCD can facilitate motility on 1.0% and 1.5% agar media. The requirement for both flagella and stator proteins indicates that flagellar rotation generates the impetus for motility on both soft and hard agar media.
The number of flagella is a key factor for robust motility [35–38]. Increasing the number of flagella allows bacteria to swarm on harder agar media or to swim through more viscous media. We hypothesized that increasing the number of flagella might be required for motility on hard agar media. To test this hypothesis, we first carried out fractionation experiments to identify the flagellin protein in Paenibacillus sp. protein extracts. Approximately 8 × 106 cells were spread over the surface of a 9 cm diameter 1.5% agar plate. After a 5 h incubation at 37°C, Paenibacillus sp. formed many moving colonies on 1.5% agar plates. These bacterial cells were suspended in buffer, collected, and the bacterial proteins separated into three fractions: secreted proteins, cell-surface associated proteins, and cellular proteins, as described in Methods. These fractions were then analyzed by SDS-PAGE (S8 Fig). We expected that flagellin would mainly be restricted to the cell-surface associated protein fraction, in which two strong protein bands (150 kDa and 28 kDa) were detected on SDS-PAGE gels (S8 Fig). LC-MS/MS analysis identified these two proteins as the S-layer protein (SpaA; Mw 111.5 kDa) and flagellin (Hag; Mw 29.6 kDa) (S9 Fig). The size of the S-layer protein on SDS-PAGE was much greater than its deduced molecular weight. The S-layer protein of Paenibacillus alvei is glycosylated [39, 40]. Also, Paenibacillus sp. possesses a putative S-layer glycosylation gene cluster (PBN151_2313 to PBN151_2300] in its genome. These observations suggest that the S-layer protein in Paenibacillus sp. is also glycosylated. The size of the flagellin protein on SDS-PAGE was very similar to its deduced molecular weight. These results confirm that flagellin is a major cell-surface associated protein. We then compared flagellin levels in the cell-surface associated protein fraction under three different growth conditions: liquid media and solid media containing 0.5% or 1.5% agar. Since 0.3% agar media are fragile, we used liquid media rather than 0.3% agar media for this analysis. Flagellin levels in 0.5% agar and 1.5% agar were the same (Fig 6A); however, those in liquid media were much lower at both the exponential- and early stationary growth phases (Fig 6A). To confirm this result, we tried to visualize flagellar filaments on cells grown in liquid or 1.5% agar media. To achieve this, a TCA (Ser) to TGC (Cys) substitution was introduced into the 161st codon of hag on the genome, which would allow the flagellin filaments to be labeled with a sulfhydryl-reactive fluorescent dye. The S161C substitution did not interfere with flagellin function because the hag S161C strain showed normal motility (S6 Fig). After fluorescent labeling, filaments were visible in the hag S161C strain but not in the wild-type strain (Fig 6B), indicating that the visualized filaments were indeed flagella. Consistent with the results of SDS-PAGE analysis, cells grown in liquid media had a few flagella, whereas those grown on 1.5% agar media exhibited peritrichous hyperflagellation (Fig 6B). We further compared the transcription of flagellar genes (hag, motAB, and motCD) in liquid, 0.5% agar, and 1.5% agar cultures. Northern blot analysis revealed that transcription of hag, motAB, and motCD was low in liquid, and strongly induced in 0.5% and 1.5% agar media (Fig 6C). Thus, hyperflagellation was supported by the transcriptional induction of flagellar genes on solid media. These results demonstrate that flagellar formation is strongly induced in response to growth on the surface of media containing agar at concentrations of 0.5% or above. Wandering colony formation was observable on media containing >1.0% agar, as described above. Flagellin levels and flagellar gene expression between 0.5% and 1.5% agar media were the same, indicating that flagellin levels per se are not a discriminator of wandering colony formation. Additional factors may be required for wandering colony formation on hard agar media.
When cellular proteins were fractionated, flagellin was observed not only in the cell-surface associated protein fraction but also in the secreted protein fraction in the wild-type strain (Fig 7, lanes 1 and 2). We found that one large, unknown protein (120 kDa) exhibited the same distribution as flagellin (Fig 7, lanes 1 and 2). Moreover, the 120 kDa protein was induced on solid media, as observed for flagellin (Fig 6A). Interestingly, the level of the 120 kDa protein was greatly reduced in the hag mutant (Fig 7, lanes 5 and 6). Therefore, we hypothesized that the 120 kDa protein might be involved in motility on hard agar media.
LC-MS/MS analysis revealed that the 120 kDa protein was the product of PBN151_2348 (S9 Fig), which is located downstream of the flagella gene cluster (Fig 3). PBN151_2348 is hereafter called cmoA (see below). To confirm that the 120 kDa protein was the product of cmoA, we constructed two new strains: an in-frame deletion mutant of cmoA and a cmoA-mCherry fusion strain. The fractionation analysis of cellular proteins revealed that deleting cmoA caused the 120 kDa protein to disappear from both the secreted and cell-surface associated protein fractions (Fig 7, lanes 7 and 8). The introduction of the mCherry tag into the 3-terminus of cmoA resulted in the disappearance of the 120 kDa protein and the appearance of a new 150 kDa protein (Fig 7, lanes 3 and 4). Unfortunately, the 150 kDa protein was hidden behind the abundant S-layer protein in the cell-surface associated protein fraction, and was only observable in the secreted protein fraction. The size of this protein (150 kDa) was consistent with the deduced molecular weight of the CmoA-mCherry fusion protein. These results confirm that the 120 kDa protein is the product of cmoA.
The motility assay revealed that the cmoA deletion mutant spread on 0.3% and 0.5% agar media but not on 1.0% and 1.5% agar media (Fig 4). The complemented strain exhibited normal motility, comparable with that of the wild-type strain (S6 Fig). Thus, CmoA is specifically required for motility on hard agar media. SDS-PAGE analysis showed that the cmoA mutant grown on 1.5% agar media expressed high levels of flagellin (Fig 7, lanes 7 and 8). Indeed, when water was poured on cmoA mutant cells grown on 1.5% agar, the cells immediately moved actively in water (S4 Movie). These results indicate that the cmoA mutation causes defects in motility on hard agar media without preventing flagellar formation. Based on these observations, we designated this gene cmoA (colony movement gene A).
CmoA is a large protein comprising 1,064 amino acids and contains an N-terminal signal sequence required for secretion, a single vWFA (von Willebrand factor type A) domain, and eight tandem IPT/TIG (immunoglobulin, plexins, transcription factors-like/transcription factor immunoglobulin) domains (Fig 8A). The vWFA domain is observed in cell adhesion and extracellular matrix proteins in eukaryotes [41], while the IPT/TIG domain is observed in cell surface receptors and transcription factors in eukaryotes [42]. However, their function in bacteria is unknown. By protein blast search on the National Center for Biotechnology Information (http://blast.ncbi.nlm.nih.gov), we identified 29 CmoA homologs; 24 were found in Paenibacillus spp., and the others were found in Brevibacillus thermoruber, Aeribacillus pallidus, Domibacillus indicus, Cohnella thermotolerans, and Clostridium sp. BL8. Except for Clostridium sp. BL8, these are Paenibacillus and its closest relatives. Among them, four species are reported to exhibit wandering colonies or the colony scattering phenotype in literatures: P. vortex [24], P. alvei CCM2051T [43], P. assamensis [44], and Paenibacillus sp. Y412MC10 [45], though while the colony phenotype of the other species is unknown. On the other hand, P. dendritiformis, P. polymyxa, P. larvae, all of which do not form wandering colonies, have no CmoA homolog. In Paenibacillus sp. cmoA is expected to constitute an operon with upstream genes yvyC, fliD, fliS, and PBN151_2349 in the flagellar gene cluster because there is no terminator sequence between these genes (Fig 8B). Among them, FliD and FliS are similar to the filament cap protein and the flagellar type III export chaperon specific for flagellin, respectively. YvyC and PBN151_2349 are unknown small proteins (111 and 99 amino acids, respectively). Except for PBN151_2349, the yvyC-fliD-fliS-(PBN151_2349)-cmoA operon is conserved in Paenibacillus vortex, Paenibacillus alvei CCM2051T, Paenibacillus sp. Y412MC10, and Paenibacillus assamensis that form wandering colonies. Each bacterium has a different small gene at the position corresponding to PBN151_2349. yvyC, fliD, and fliS are also conserved as members of the flagellar gene operon, yvyC-fliD-fliS-fliT, in Bacillus subtilis [46]. These observations indicate that cmoA may be a member of flagellar regulon.
CmoA has the N-terminal signal sequence required for secretion, and was detected in extracellular protein fractions, indicating that CmoA might have an extracellular function. To address this, we tested whether the cmoA mutant exhibited extracellular complementation. The secreted protein fraction was prepared from the wild-type strain grown on 1.5% agar, sterilized by passing through a membrane filter, and spread over the surface of a 1.5% agar plate. When the cmoA mutant was inoculated onto the center of the plate, it spread over almost all the plate within 18 h of incubation (Fig 8C). In contrast, the comA mutant did not spread on 1.5% agar plates supplemented with the secreted protein fraction prepared from the cmoA mutant (Fig 8C). As described above, the hag mutant exhibited decreased CmoA (Fig 7). The secreted protein fractions derived from the hag mutant restored motility to the cmoA mutant only partially (Fig 8C). Those secreted protein fractions did not affect the motility of the wild-type strain, and did not restore motility to the hag mutant (Fig 8C). These results indicate that extraneous CmoA can restore motility to the cmoA mutant. Thus, CmoA plays an extracellular function in terms of motility on hard agar media.
We next examined the localization of CmoA using the cmoA-mCherry strain. The CmoA-mCherry fusion is functional because the cmoA-mCherry strain exhibits normal motility (S6 Fig). To examine its localization during motility on hard agar media, the cmoA-mCherry strain was spotted onto 1.5% agar plates and grown until small colonies moved out from the inoculation site. Coverslips were placed directly onto the moving colonies and examined under a fluorescence microscope (Fig 8D). The bacterial cells showed strong fluorescence emitted by mCherry, which appeared between cells and enveloped clusters of cells. Weak fluorescence was also observed across the surface of the media, indicating that a small part of CmoA diffuses into the media during colony movement. No fluorescence signal was observed in wild-type strain under the same conditions.
We tried to examine the effect of the cmoA mutation on cellular behavior. However, we were unable to examine cellular behavior on 1.5% agar media. This is because motility mutants of cmoA and hag did not move when spread over the surface of 1.5% agar media; rather they formed cell aggregates by dividing within the inoculated spots. The shape of the aggregates was quite similar to that of moving cell clusters formed by the wild-type strain (S10 Fig). The cmoA mutant was able to spread on media containing up to 0.6% agar. The clusters formed by the wild-type strain on 0.6% agar media tended to be larger than those formed on 0.5% agar (S11 Fig). Therefore, we compared cellular behavior of both the wild-type strain and the cmoA mutant on 0.6% agar media using video light microscopy. The density of the wild-type cells at the leading edge zone of expanding colonies was very low (Fig 8E), and the cells frequently formed small clusters of several cells (S5 Movie). The clusters of wild-type cells moved more rapidly and smoothly than single cells, and moved forward into the unexplored region of agar media without stalling (S5 Movie). By contrast, the cmoA mutant formed a clear leading edge with a high cell density (Fig 8E). cmoA mutant cells in the area immediately inside the edge moved but did not form cell clusters (S6 Movie). When cmoA mutant cells reached the leading edge, some stalled and others turned back. The behavior of the cmoA mutant is very similar to that of other swarm bacteria, in that the cells are active within the swarming colonies but stall at the swarming edge on hard agar media. Swarming bacteria attract water from agar media and swarmer cells within the swarm colony move in the resulting fluid layer [7–10]. However, the fluid layer at the edge is very thin and cannot support cell movement until sufficient fluid is supplied from inside the swarm colony [10, 47, 48]. Thus, our observations indicate that expansion of the cmoA mutant probably depends on the fluid layer, as observed for other swarm bacteria. The cmoA mutant is probably unable to move into the unexplored region until sufficient fluid is supplied from inside the swarm colony. The leading edge region of the wild-type strain had very low cell density. Since a high cell density is thought to be important for extracting water from agar media and maintaining the resulting fluid [10], wild-type cells do not appear to acquire water in the same manner, as the cmoA mutant and other swarm bacteria. In fact, single cells of the wild-type strain were able to move slowly by themselves at the leading edge zone on 0.6% agar media (S5 Movie). These observations indicate that the wild-type strain has a water acquisition mechanism that does not depend on cell population, and that the cmoA mutant loses this mechanism. One possible explanation is that cell associated CmoA may facilitate extraction of water from the agar media and/or smooth the cell surface interface, which enables the wild-type strain to move on harder agar media.
We hypothesized that exogenous CmoA may stimulate motility of other bacteria. To test this possibility, we examined motility of B. subtilis, Paenibacillus alvei, and E. coli on plates supplemented with or without CmoA. The result showed that exogenous CmoA slightly enhanced B. subtilis motility: these cells formed slightly larger swarm colonies on 1% agar plates supplemented with the Paenibacillus sp. wild-type lysate than on plates supplemented with the cmoA mutant lysate (S12 Fig). A similar tendency was observed for the B. subtilis srfAC mutant, which does not produce the surface active compound surfactin involved in swarming motility, on 0.8% agar [49]. However, these effects were not very marked: therefore, we cannot conclude that CmoA stimulates B. subtilis motility. Exogenous CmoA did not affect motility of Paenibacillus alvei and E. coli (S12 Fig).
We compared cmoA expression in liquid and agar cultures. Northern blot analysis revealed that cmoA transcription was low in liquid culture and strongly induced in 0.5% and 1.5% agar cultures (Fig 8F), which was the same with flagellar genes, hag, motAB and motCD (Fig 6C). The size of cmoA transcript was much bigger than 23S rRNA (Fig 8E), suggesting that cmoA is probably cotranscribed with upstream genes yvyC, fliD, and fliS as expected from DNA sequence (Fig 8B). Wandering colonies were observable on media containing >1.0% agar. However, expression of cmoA and flagellar genes was the same between 0.5% and 1.5% agar media, indicating that the induction of cmoA and flagellar genes is insufficient to support wandering colony formation. As shown in S5 and S6 Movies, wild-type cells formed moving cell cluster in a CmoA-dependent manner on 0.6% agar. However, these moving clusters did not grow to visible wandering colonies on 0.6% agar media, suggesting that moving clusters are less stable on 0.6% agar than on 1.5% agar. We hypothesized that wetness might affect the stability of wandering colonies. To test this hypothesis, a small amount of water was gently poured on wandering colonies grown on 1.5% agar and cellular behavior was immediately observed. We found that wandering colonies disassembled easily and quickly in water (Fig 9), and individual cells moved randomly in water (S7 Movie). These observations indicate that wandering colonies are very sensitive to wetness and cannot maintain their shape in high wet conditions. These observations indicate that low wet conditions are required not only to induce cmoA and flagellar gene expression but also to maintain wandering colonies. Thus, Paenibacillus sp. forms wandering colonies specially to move on low wet surfaces.
Here, we describe cooperative motility behavior of Paenibacillus sp. NAIST 15–1 and its genetic analysis. Paenibacillus bacteria have commonly been isolated from diverse environmental samples, such as soil, hot springs, blood samples, honeybees, and plants [14–18, 22, 44, 45]. However, their genetic studies have been limited due to the difficulty in their genetic manipulation. We demonstrated that Paenibacillus sp. was amenable to genetic manipulation. Thus, this strain will be useful for genetic analysis of Paenibacillus bacteria. Paenibacillus sp. exhibited distinct types of flagella-driven motility, which was dependent on the agar concentration. Individual cells swam in liquid media or in a thick fluid layer on 0.3% agar media, but exhibited hyperflagellation and formed small moving groups, when grown on the surface of 0.5% agar media. Bacterial motility is often inhibited in solid media containing agar at concentrations above 1%. However, Paenibacillus sp. formed wandering colonies, which moved across the surface of media containing >1.5% agar.
We showed that flagella are essential for the motility of Paenibacillus sp. Despite its special motility behavior, the organization and composition of flagellar genes in Paenibacillus sp. are quite similar to those in B. subtilis. Flagella formation by Paenibacillus sp. was induced on hard agar media, which led to peritrichous hyperflagellation. Hyperflagellation was supported by the transcriptional induction of flagellar gene expression. Transcriptional induction of flagellar gene expression in response to growth on hard agar media is also observed in other swarm bacteria including B. subtilis, Vibrio spp., Proteus mirabilis, and Serratia marcescens [36, 50–53]. An artificial increase in the number of flagella allows bacteria to swarm on harder agars or to swim through more viscous media in B. subtilis, Salmonera enterica, Proteus mirabilis, Pseudomonas aeruginosa [35–38]. Thus, an increase in the number of flagellar is a key factor for motility on hard agar media. Flagellar stators generate the torque required for flagellar rotation using an electrochemical gradient of H+ or Na+ across the cytoplasmic membrane [33]. Paenibacillus sp. possesses two sets of flagellar stators, MotAB and MotCD. The gene disruption experiment showed that both MotAB and MotCD facilitate motility on 0.3% and 0.5% agar, but only MotCD facilitates motility on 1.0% and 1.5% agar. Multicopy MotAB was unable to restore motility to the motCD mutant on hard agar media. These observations indicate that MotCD plays a specific role in motility on hard agar media at least under the conditions tested herein. Some bacteria harbor multiple stators that play distinct roles in motility. Vibrio spp. has two types of flagella: polar and lateral. Polar flagella are driven by the Na+-dependent stator and provide the impetus for swimming motility [54]. Lateral flagella are driven by the H+-dependent stator and provide the impetus for swarming motility [54]. The formation of lateral flagella and their stators is transcriptionally induced in response to growth on hard agar media [53]. However, unlike Vibrio spp., transcription of both motAB and motCD operons of Paenibacillus sp. was induced on hard agar. A recent study shows that Paenibacillus sp. TCA20 also has two sets of stator operons, motA1motB1 and motA2motB2 [55]. Phylogenetic analysis of MotB1 and MotB2 along with other MotB stator subunits showed that MotB2 is highly similar to H+-dependent MotB stators, whereas MotB1 belongs to a different stator cluster from known H+-dependent or Na+-dependent MotB stators [55]. The MotB1-type stators are unique to Paenibacillus bacteria, and the MotA1MotB1 stator is suggested to use Ca2+ or Mg2+ as coupling ions [55]. Phylogenetic analysis showed that MotB of Paenibacillus sp. was classified into the group of MotB1-type stators, whereas MotD was classified into the group of H+-dependent MotB stators (S7B Fig). These observations indicate that MotAB and MotCD may use different coupling ions for flagellar rotation. Although at present the reason for the functional difference between MotAB and MotCD is unclear, the finding that flagella and stator proteins are required for motility strongly suggests that flagella rotation provides the impetus for Paenibacillus sp. motility on both soft and hard agar media.
Swarming motility is defined as flagella-driven group motility on a solid surface [3–6]. According to this definition, the motility by Paenibacilus sp. on hard agar media is classified as swarming motility, although it has unique characteristics. Since flagella rotation pushes a cell forward against the surrounding water, surface water is a critical element for swarming motility. Swarming bacteria on hard agar media can attract water to the surface from the agar matrix [7–10]. Cell growth (a high cell density), cellular secretions, and flagella rotation are thought to attract water to the surface, from where it diffuses throughout the swarming colony [7–11, 47, 48]. As a result, the swarming colony maintains a fluid layer within itself, in which each cell moves actively. However, the edge of the colony contains only a very thin layer of fluid [10]; thus swarmer cells stall there until fluid is supplied from the inside of the swarm colony [11, 47]. The flagella of stalled cells at the edge of the swarm colony frequently orient toward the unexplored region of the media and rotate; this likely pumps fluid outwards, thereby aiding expansion of the swarming colony [11, 47]. Thus, swarmer cells cannot move into unexplored regions of hard agar media without first expanding the fluid layer, and swarm bacteria usually formed single swarm colony when inoculated onto a single site on hard agar media. Paenibacillus sp. formed many clusters on hard agar media, which grew up to form visible wandering colonies. Cells in wandering colonies were associated with neighbor cells and did not move within wandering colonies. Cell clusters and wandering colonies were able to leave the fluid layer of the mother swarm colony and move around the unexplored region on hard agar media by themselves. As a result, Paenibacillus sp. formed multiple colonies even when inoculated into single site on hard agar media.
We identified CmoA as a protein specifically required for motility on hard agar media. CmoA is a large protein of 1,064 amino acids and comprises a single vWFA domain and eight tandem IPT/TIG domains. The vWFA and IPT/TIG domains are mainly conserved in eukaryotes and many of these domains are found in extracellular proteins such as cell adhesion proteins, extracellular matrix proteins, and cell surface receptors [41, 42]. Proteins containing these domains are also found in bacteria, but are not well characterized. The N-terminus of CmoA is a typical signal sequence for secretion. Indeed, CmoA was detected in extracellular fractions. Moreover, the motility of the cmoA mutant was restored in media supplemented with the secreted protein fraction derived from the wild-type strain, but not that from a cmoA mutant. Live imaging of cells expressing the CmoA-mCherry fusion protein plated on hard agar media showed that CmoA was mainly localized around the cells. CmoA-mCherry was present between the cells within clusters, but also covered entire clusters. These observations show that CmoA has an extracellular function, as observed for vWFA and IPT/TIG domain proteins in eukaryotes. The cmoA mutant was able to induce flagella in response to surface growth condition and moved on solid media containing up to 0.6% agar. However, its cellular behavior was quite different from that of wild-type cells. On 0.6% agar media, the wild-type strain formed small cell clusters that moved smoothly into the unexplored region of the media, whereas the cmoA mutant neither formed clusters nor moved smoothly into the unexplored region. Cells of the cmoA mutant moved actively inside the swarming colony, but stalled at the swarm edge and formed a clear leading edge. Interestingly, this behavior is similar to that observed for swarm cells of other bacteria such as B. subtilis and E. coli [11, 47–49], whose spread on hard agar media is dependent on the acquisition of fluid. Thus, the cmoA mutant seems to spread on hard agar media in a similar manner to other swarm bacteria. We also observed that the leading edge region of the wild-type strain had very low cell density. Under the condition, cell population-dependent water extraction from agar media is unexpected. In other word, Paenibacillus sp. has a water acquisition mechanism that does not depend on cell population. We propose that cell-surface associated CmoA may be involved in drawing water out of agar and/or smoothing cell surface interactions. Lacking this function, cmoA mutant cells may be unable to leave the mother swarming colony and move into the unexplored region on hard agar media. Proteus mirabilis secretes high molecular weight polysaccharide, which is responsible for extracting water from agar media [56]. E. coli secretes some osmolytes, probably lipopolysaccharides, to extract water [57]. Several swarm bacteria secrete surface active molecules, e.g. surfactin secreted by B. subtilis [49] and rhamnolipids by Pseudomonas spp. [58]. CmoA may be the first identified protein responsible for extracting water from agar media or for reducing surface tension to facilitate swarming motility.
We also observed that clusters of the wild-type cells moved much faster than single cells. Many swarm bacteria form cell clusters, called rafts, which also move faster than single cells in the fluid layer of swarm colonies. For instance, B. subtilis cells recruited to a raft moved with the group, whereas cells lost from a raft stop moving [49]. The formation of rafts is thought to facilitate movement on hard agar media partly by reducing viscosity/drag on individuals [12]. This though probably applies to cell clusters of Paenibacillus sp. As rafts of swarm bacteria are unstable and dynamically labile in terms of both members and shape, no specific substances or matrices appear to maintain rafts stability [5, 6, and references therein]. However, cell clusters of Paenibacillus sp. may have different features. Cell clusters of Paenibacillus sp. were able to leave the fluid layer of the mother colony and move alone into unexplored region. Moreover, on hard agar media containing >1% agar, those clusters grew up to form visible wandering colonies, in which cells are associated with neighbor cells. Therefore, the cell clusters of Paenibacilus sp. on hard agar media appear to be more stable than rafts of other swarm bacteria. We speculate that Paenibacillus sp. may have a mechanism for stabilizing cell clusters. Many swarm bacteria such as B. subtilis and E. coli, cannot move on media containing 1% agar. However, some bacteria, called robust swarmers, can move on media containing >1.5% agar [6, and references therein]. One such robust swarmer, Proteus mirabilis, forms highly organized multicellular rafts during swarming motility. Electron microscopy shows that adjacent cells in those rafts are connected via intercellular bundling of flagella [59]. Likewise, cells in wandering colonies of Paenibacillus vortex are also connected by filament networks, which are probably flagella. [24]. Since these electron microscope images were taken using fixation procedures, further confirmation using live cells will be required. On the other hand, these observations suggest that there may be a mechanism that stabilizes cell clusters in robust swarmers in which flagella may play an important role. The cmoA mutant did not form cell clusters and did not move on hard agar media containing >1% agar. CmoA was located around cells and enveloped cell clusters. Based on these observations, we speculate that CmoA may also play a role in stabilizing cell clusters. For example, cell-surface associated CmoA may interact with flagella, which form cell-cell connections. The social Amoeba, Dictyostelium discoideum forms multicellular fruiting bodies in response to nutrient exhaustion. During this process, cell-cell adhesion is mediated by binding between extracellular proteins TgrB1 and TgrC1 via mutual IPT/TIG domains [60]. CmoA itself may form cell-cell connection via its multiple IPT/TIG domains. However, cell-cell connections that support wandering colony formation are completely different from stable cell-cell connections observed in bacterial biofilms [61]. Wandering colonies were easily disassembled when water was added. The cell-cell connection in wandering colonies may be fragile or sensitive to wetting. Further work will be required to determine how Paenibacillus sp. forms wandering colonies and to examine role of CmoA in wandering colony formation.
Professor Ben-Jacob’s group isolated and extensively analyzed pattern forming Paenibacillus bacteria [23–25, and references therein]. However, these strains were isolated as contaminants from a B. subtilis stock and their natural habitat is unclear. Here, we isolated Paenibacillus sp. from the rhizosphere of natural weeds. We found that this type of Paenibacillus bacteria was easily isolated from natural weed roots and associated soils (S13 Fig). The natural habitat of pattern-forming Paenibacillus bacteria seems to be soil and its related environment including the rhizosphere. This seems to be the reason why Paenibacillus sp. exhibits complex motility behavior. Soil includes various substances, and its surface properties vary. The water content in soil is often variable and is sometimes very low. Thus, soil bacteria appear to employ multiple motility systems to overcome these environmental challenges. For instance, a soil bacterium, B. subtilis, exhibits flagellar-dependent swimming and swarming motility [49]. Recently, two groups reported that B. subtilis also exhibits sliding motility under conditions in which flagella do not work [62, 63]. Myxococus xanthus, which does not have flagella, also exhibits two types of motility, individual gliding motility and social twitching motility, which have different advantages on different surfaces [64]. Pseudomonas aeruginosa exhibits flagella-dependent motility, Type IV pili-dependent twitching motility, and sliding motility [65]. The results presented herein reveal that Paenibacillus sp. has evolved to use flagella under a wider range of conditions. To overcome conditions of low wetness, Paenibacillus sp. forms multicellular wandering colonies. When wandering colonies encounter wet conditions, they quickly disassembled and individual cells can swim or swarm in the water layer. These properties will be advantageous for Paenibacillus sp., enabling it to disperse its cells rapidly and widely in soil.
Plant weeds were uprooted from the ground within Nara Institute of Science & Technology (NAIST) and most of the soil around the roots removed. The plant roots were then cut, placed in 15 ml tubes, and soaked in 2 ml LB medium. After a brief vortex, the extract was incubated at 85°C for 10 min and then plated onto LB at appropriate dilutions. Twelve strains, all of which showed a colony scattering phenotype and antagonistic activity against Fusarium oxysporum, were isolated. Among these, Paenibacillus sp. NAIST15-1 was selected for analysis based on its strong growth and transformability. The 16S rDNA region was amplified from its genome DNA with primers 16S-F1 (5’-TTAGCGGCGGACGGGTGAGT) and 16S-R1 (5’-TGACGGGCGGTGTGTACAAG). Nucleotide data for partial 16S rDNA sequence has been submitted to the DDBJ/EMBL/GenBank databases under the accession number, LC185074.
The Paenibacillus sp. strains used in this study are listed in Table 1. All strains were derivatives of Paenibacillus sp. NAIST 15–1. Paenibacillus sp. was grown in LB (LB Lennox, Difco) or 2×YT (16g l-1 tryptone (Difco), 10g l-1 yeast extract (Difco), 5g l-1 NaCl) medium. For plates, media were solidified with the indicated concentrations of agar (for Microorganism Culture, Nakalai Tesque). Single colonies were isolated on 2.5% agar plates. Antibiotics were used at the following concentrations: erythromycin, 2.5 μg ml-1; tetracycline, 10 μg ml-1; and chloramphenicol, 2.5 μg ml-1. For the motility assay, strains were grown overnight in 2×YT liquid media at 28°C with shaking. One microliter of culture was then spotted onto the center of 2×YT plates. The overnight cultures generally contained an average of 8 × 107 cells ml-1. For the motility assay, freshly prepared plates were allowed to dry overnight on the laboratory bench and used the next day. For the protein and RNA analyses, overnight cultures were diluted 10-fold, and 100 μl were spread on each 9 cm diameter plate. Cells were then cultivated at 37°C for 5 h prior to analysis. Escherichia coli HB101 was used for plasmid construction and was cultivated in LB medium. Ampicillin was used at 50 μg ml-1 when necessary. Plasmids were prepared using the FastGene Xpress Plasmid PLUS kit (Nippon Genetics) prior to electroporation of Paenibacillus sp. Paenibacillus alvei NBRC3343 was obtained from National Institute of Technology and Evaluation. E. coli W3100, B. subtilis NCIB 3610, and its srfAC mutant were described previously [29].
Colony formation was analyzed by time-lapse light microscopy. Briefly, 1 μl of Paenibacillus sp. suspension was spotted onto 2×YT/1.5% agar medium in a 35 mm diameter plate. The plate was then incubated at 37°C in a microscope incubation system (Tokai hit). The process of cell movement and colony formation was observed under a SZX7 zoom stereo microscope (Olympus) connected to a DP70 digital camera (Olympus). Cell movement was observed under a DMRE-HC microscope (Leica) connected to the DP70 digital camera. Time-lapse images and movies were collected using DP70 controller software (Olympus). Movies were downsized using Windows movie maker (Microsoft). Fluorescent images were observed under the DMRE-HC microscope connected to a digital CCD camera (1300Y; Roper Science). Image acquisition and processing were performed using Metamorph (Universal Imaging Corporation).
Paenibacillus sp. was grown overnight in 5 ml of 2×YT at 28°C with shaking. Cells were then pelleted, suspended in 2 ml of saline-EDTA (150 mM NaCl, 50 mM EDTA, pH 8) supplemented with 0.5 mg ml-1 lysozyme, and incubated at 37°C for 20 min. Next, 1 μl of proteinase K (Takara Bio) was added and incubated at 37°C for 20 min. The lysate was then mixed with 200 μl of 10% SDS and incubated at 50°C until it became clear. Chromosomal DNA was extracted using phenol-chloroform-isoamyl alcohol and precipitated with ethanol. For draft genome sequencing, chromosomal DNA was purified using the Illustra bacteria genomic Prep Mini Spin Kit (GE Healthcare).
Genomic DNA from Paenibacillus sp. was fragmented (500 bp fragments) using a sonicator (Covaris S2, Covaris). A library for sequencing was prepared using a NEBNext DNA Library Prep Master Mix Set for Illumina (New England Biolabs) and paired-end sequenced (2 × 300 bp) on a MiSeq sequencer using MiSeq reagent kit v. 3 (Illumina K.K.). Raw reads were trimmed and de novo assembled using CLC Genomics Workbench 6.5 (CLC bio, Qiagen).
The raw read data were deposited in the DRA database at DDBJ with the following accession numbers: DRA ID: DRA003610; BioProject ID: PRJDB3476; and BioSample ID: SAMD00030729. The draft genome sequence and annotation for Paenibacillus sp. NAIST15-1 were registered in DDBJ/EMBL/GenBank with the following accession numbers: BBYF01000001–BBYF01000042 (42 entries).
Paenibacillus sp. was transformed by electroporation according to the method of Murray and Aronstein [66], with minor modifications. Briefly, Paenibacillus sp. was grown to OD600 = 0.2 in 2×YT at 37°C with shaking. The cells were then pelleted by centrifugation, washed three times with ice-cold 625 mM sucrose, and suspended in 1/200 culture volume of ice-cold 625 mM sucrose. Competent cells were aliquoted and stored at -80°C. Electroporation was performed using a Gene Pulser (Bio-Rad). For transformation, 45 μl of competent cells were mixed with 0.5–1 μg of plasmid DNA and transferred to an ice-cold 1 mm cuvette. The sample was then pulsed under the following conditions: voltage, 1.8 kV cm−1; capacitance, 25 μF; and resistance, 200 Ω. One milliliter of 2×YT was then added to the cell suspension and incubated overnight at 28°C with shaking. Transformants were selected on 2×YT/2.5% agar plates containing erythromycin (2.5 μg ml-1).
Gene disruption was performed using the plasmid pMAD, which contains a thermosensitive origin of replication for Gram-positive bacteria, erm, and bgaB, which encodes a thermostable β-galactosidase [32]. pMAD was a kind gift from Dr. Kazuya Morikawa (Tsukuba University) and Dr. Michel Débarbouillé (Institut Pasteur). DNA fragments involved in gene disruption via double crossover recombination were prepared by PCR and cloned into pMAD. The plasmid pEpGAP-mCherry [67] was used to construct the cmoA-mCherry strain. pEpGAP-mCherry was a kind gift from Dr. Neta Dean (Stony Brook University). The primers used for PCR are listed in S1 Table. A detailed explanation of the methods used for plasmid construction and gene disruption are provided in S1 File.
Overnight cultures grown at 28°C in 2×YT were diluted 10-fold and 100 μl spread over 2×YT/1.5% agar plates. After 5 h of incubation at 37°C, cells from two plate cultures were suspended in 600 μl of 10 mM Tris-HCl (pH 7.6) and collected in 1.5 ml tubes. The volume of the samples was usually 150–200 μl. Cells were separated by centrifugation at 15,000 rpm for 2 min. The supernatant was made up to 200 μl with buffer and used as the secreted protein fraction. The precipitated cells were suspended in 200 μl of 10 mM Tris-HCl (pH 7.6)/0.1% SDS and boiled for 1.5 min to extract cell-surface associated proteins. The samples were then centrifuged at 15,000 rpm for 5 min. The supernatant was used as the cell-surface associated protein fraction. The precipitate was resuspended in 10 mM Tris-HCl (pH 7.6)/0.1% SDS and the cells disrupted by sonication. The sample was centrifuged at 15,000 rpm for 5 min and the supernatant was used as the cellular protein fraction. The three fractions were then separated by SDS-PAGE (SuperSep Ace 10–20% gels; Wako Pure Chemical Industries), and proteins were stained with Coomassie Brilliant Blue (CBB).
After the cell-surface associated protein fraction was separated by SDS-PAGE, protein bands were stained with Coomassie Brilliant Blue and excised. Peptides were prepared for LC-MS/MS analysis by in-gel digestion with trypsin. LC-MS/MS analysis was performed using a LTQ-orbitrap XL system (Thermo Scientific), as described by Tanaka et al. [68]. The MS/MS spectra of the identified peptides were searched against our own database, which contains SpaA, flagellin, and CmoA sequences, and against non-redundant protein sequences in the National Center for Biotechnology Information (NCBI) database, using a MASCOT server (Matrix Science).
Cells were collected from 2×YT liquid or plate cultures and pelleted by centrifugation at 5,000 rpm for 2 min. Cells were then suspended in T-base buffer [69]. Samples were mixed with Alexa Fluor 594 C5 maleimide (Life Technologies; final concentration, 15 μg ml-1) and incubated at room temperature for 1 h in the dark. The samples were then washed twice with T-base buffer and observed under a fluorescent microscope.
Cells were collected from 2×YT liquid or plate cultures. Total RNA was extracted and Northern blot analysis performed as described previously [70]. The primers used for RNA probe preparation are listed in S1 Table.
Overnight cultures grown at 28°C in 2×YT were diluted 10-fold and 100 μl spread over 2×YT/1.5% agar plates. After 5 h of incubation at 37°C, cells from two plate cultures were suspended in 800 μl of 2×YT liquid media and collected in 1.5 ml tubes. After brief vortexing, the sample was separated by centrifugation at 15,000 rpm for 5 min. The supernatant was sterilized by passing through a 0.45 μm PVDF membrane filter and 100 μl of the resultant sample was spread over a 1.5% agar plate. The plate was left for 10 min on the bench and used for the swarm assay.
|
10.1371/journal.pntd.0004903 | A Primate APOL1 Variant That Kills Trypanosoma brucei gambiense | Humans are protected against infection from most African trypanosomes by lipoprotein complexes present in serum that contain the trypanolytic pore-forming protein, Apolipoprotein L1 (APOL1). The human-infective trypanosomes, Trypanosoma brucei rhodesiense in East Africa and T. b. gambiense in West Africa have separately evolved mechanisms that allow them to resist APOL1-mediated lysis and cause human African trypanosomiasis, or sleeping sickness, in man. Recently, APOL1 variants were identified from a subset of Old World monkeys, that are able to lyse East African T. b. rhodesiense, by virtue of C-terminal polymorphisms in the APOL1 protein that hinder that parasite’s resistance mechanism. Such variants have been proposed as candidates for developing therapeutic alternatives to the unsatisfactory anti-trypanosomal drugs currently in use. Here we demonstrate the in vitro lytic ability of serum and purified recombinant protein of an APOL1 ortholog from the West African Guinea baboon (Papio papio), which is able to lyse examples of all sub-species of T. brucei including T. b. gambiense group 1 parasites, the most common agent of human African trypanosomiasis. The identification of a variant of APOL1 with trypanolytic ability for both human-infective T. brucei sub-species could be a candidate for universal APOL1-based therapeutic strategies, targeted against all pathogenic African trypanosomes.
| African trypanosomes are protozoan parasites that affect both humans and animals in poor rural areas of sub-Saharan Africa, and are a major constraint on health and agricultural development. Disease control is principally dependent on the administration of drugs, which are old and largely unsatisfactory. Humans are naturally resistant to infection by most African trypanosomes species because of a lytic protein component in their blood, called APOL1. However, human-infective trypanosomes, T. b. rhodesiense in East Africa, and T. b. gambiense in West Africa, have evolved separate mechanisms to disarm this lytic protein and cause disease. Recently, variants of APOL1 were discovered in some primates that are able to kill the East African human disease-causing sub-species. These APOL1 variants form the basis of current attempts to create novel therapeutic interventions that can kill both animal and human-infective trypanosomes. In this study, we show that another variant of the same protein from a West African baboon species is able to kill, not only East African human-infective trypanosomes, but also the West African parasites, which causes the majority of human African trypanosomiasis cases. This new APOL1 variant could be a potential candidate for anti-trypanosomal therapies targeted at all pathogenic trypanosome species.
| African trypanosomes continue to exert a significant barrier to agricultural production and rural development across sub-Saharan Africa [1]. Due to a primate-specific innate trypanolytic mechanism, the majority of trypanosome species are unable to infect man. However, two sub-species of Trypanosoma brucei, T. b. rhodesiense and T. b. gambiense, have evolved distinct processes to resist this lysis and cause the debilitating and often fatal human form of African trypanosomiasis, known as sleeping sickness. The West African T. b. gambiense parasite typically causes a chronic disease profile, while the zoonotic T. b. rhodesiense sub-species, located in Eastern and Southern Africa, results in a more rapidly progressing, acute infection [2,3]. Seventy-million people over an area of 1.55 million km2 are at risk of contracting either of the two human-infective sub-species [4].
Current anti-trypanosomal drugs for medical and veterinary administration are largely unsatisfactory due to high toxicity, difficult treatment regimens, and emerging resistance [5–7]. Decades of drug development for African trypanosomiasis has produced safer refinements of existing therapies [7,8] and a number of promising novel drug candidates [9–11], but as yet no new anti-trypanosomal therapy has successfully passed phase III clinical trials. Furthermore, the adaptive immune response of vertebrates is rendered largely ineffective by the trypanosome’s ability to cyclically evade detection through variant surface glycoprotein (VSG)-mediated antigenic variation [12,13], placing a significant hurdle in the path of vaccine development. Broad-spectrum, safe, easily administered, and effective therapies to treat African trypanosomiasis are therefore still needed. The recent discovery of primate serum proteins that are able to kill both animal and human-infective trypanosomes is now offering opportunities for novel therapeutic approaches [14,15].
It has been known for over a century that the serum of humans and a small number of other Catarrhine primates are highly toxic to most African trypanosome species [16,17]. The molecular basis of this innate immunity in man has been elucidated and centres on two trypanolytic serum complexes, Trypanosome Lytic Factor 1 (TLF-1) [18,19] and TLF-2 [20,21], which share the same core protein components: haptoglobin-related protein (HPR) and apolipoprotein L1 (APOL1). HPR bound to haemoglobin mediates TLF-1 endocytosis via the haem-scavenging, haptoglobin-haemoglobin receptor (HpHbR) on the trypanosome’s surface [22–25]. Difficulty in purifying TLF-2 ex-vivo, has hindered discovery of exactly how this complex is bound and internalised by the parasite but it is known that it does not require HpHbR [21,26]. Despite differences in uptake, both TLF-1 and TLF-2 utilize the same lytic component in the form of the ionic channel-forming protein, APOL1 [22,27,28]. Following internalization, APOL1 undergoes a pH-dependant conformational change in the endolysosome pathway which releases it from the TLF complex [29,30], and promotes insertion into parasite membranes [31,32]. The exact mechanism of APOL1-mediated lysis that follows remains to be elucidated. In one recent model APOL1 insertion was found to disrupt both lysosomal and mitochondrial membranes, inducing an apoptosis-like cell death [33]. In contrast, an alternative model proposes that endosome recycling of APOL1 to the neutral environment of the parasite’s plasma membrane accelerates cation-selective channel activity and promotes lysis by osmotic swelling [34].
The Trypanosoma parasites responsible for animal trypanosomiasis are rapidly killed by this innate defence system, whereas the human sleeping sickness parasites, T. b. rhodesiense and T. b. gambiense, are able to resist lysis. In T. b. rhodesiense, resistance is effected by the VSG-derived, serum resistance associated (SRA) protein [35,36] which binds to the C-terminal domain of APOL1 in the endolysosome pathway preventing channel-mediated lysis [27,37–39], plausibly by impeding correct membrane insertion of APOL1 [34,40].
The mechanism of human serum resistance in T. b. gambiense has taken longer to unravel. T. b. gambiense typically grows to very low parasitemia and is difficult to adapt to laboratory models. An additional complicating factor is that T. b. gambiense shows two distinct "groups" that differ in genotype and phenotype [41–44]. The classic, clonal T. b. gambiense type [45], labelled “group 1” and found in West and Central Africa, is the predominant human-infective sub-species, responsible for 97% of all reported human cases [46]. T. b. gambiense group 1 strains are invariably resistant even after prolonged passage in laboratory rodents [42,47] and the mechanism underlying this resistance appears multifactorial, with at least three independent contributing components so far identified. Firstly the reduction of TLF-1 uptake through reduced expression and polymorphism of the HpHbR receptor that reduces binding affinity [48–50]; secondly, expression of a VSG-related T. b. gambiense-specific glycoprotein (TgsGP) which is essential, but not sufficient, for resistance [51] and which may increase resistance to APOL1 pore-mediated lysis by stiffening trypanosomal membranes [52]; and thirdly, faster APOL1 degradation has been proposed, through the action of cysteine peptidase [52,53]. A second, more virulent type of T. b. gambiense was identified in Cote d’Ivoire and Burkina Faso in the 1980’s [42,44] and defined as “group 2”, but has since virtually disappeared and may now be extinct. Studies of the limited number of group 2 strains that have been isolated indicate that these parasites are closely related to West Africa T. b. brucei [41,43,44,54] and exhibit a variable human serum resistance phenotype, in a manner superficially similar to T. b. rhodesiense [42,47,48]. Although the underlying resistance mechanism remains elusive it does not appear to involve a reduction in TLF-1 uptake [48] or the SRA [55] or TgsGP gene [56,57].
Unlike humans and gorillas [58,59], from which they diverged around 25 million years ago [60], several members of the Cercopithecidae (Old World monkey) family appear intrinsically resistant to T. b. rhodesiense [58,59,61]. Both serum and APOL1 from the East African baboon species, Papio hamadryas, has been demonstrated to effectively lyse human-infective T. b. rhodesiense [14,58]. This difference in innate immunity between Homo sapiens and P. hamadryas, has been pinpointed to the position of a single amino acid in the baboon APOL1 C-terminus which prevents the parasite’s SRA protein from binding and neutralising APOL1 lytic activity [62]. Furthermore, a nearly identical mutation has now also been detected in the C-terminus of APOL1 variants of some humans with African ancestry whose serum exhibits lytic activity against T. b. rhodesiense but not T. b. gambiense [63].
This led to the hypothesis that as T. b. gambiense is found only in West Africa, another variant of APOL1 may exist in some West African primates that is able to kill T. b. gambiense. In this study we examined the serum and APOL1 protein of a West African baboon species, Papio papio, suggested to be refractory to T. b. gambiense infection, with the ability to eliminate parasites in a laboratory infection [64]. Here we demonstrate that serum and recombinant protein from the P. papio APOL1 ortholog lyses representative strains of all sub-species of T. brucei in an in vitro assay system. The identification of an APOL1 variant with broad trypanolytic ability against T. brucei sub-species, including the most prevalent T. b. gambiense type, may provide a potential reagent for the development of universal APOL1-based therapeutic agents.
Representative bloodstream form cell lines were selected for each subspecies from a collection at the University of Glasgow and have been previously described. STIB247 is a T. b. brucei strain originally isolated from a hartebeest in Serengeti, Tanzania in 1971 [65]. The T. b. rhodesiense strain EATRO98 was isolated by the East African Trypanosomiasis Research Organization (EATRO) from a human in Nyanza, Kenya in 1961 [66]. T. b. gambiense group 2 strain STIB386 (MHOM/CI/78/TH114) was originally isolated in 1978 from an infected patient in Côte d'Ivoire [67]. ELIANE (MHOM/CI/52/ITMAP 2188) is a T. b. gambiense group 1 strain isolated from a human in Côte d’Ivoire in 1952 [68]. Additional T. b. gambiense group 1 strains tested were human isolates, PA (MHOM/CG/80/ITMAP1843/PA) from Republic of the Congo in 1975 [43], BIM (MHOM/CM/75/ITMAP1789/BIM) from Cameroon in 1975 [43], and TOBO (MHOM/CI/83/DAL596/TOBO) and S1/1/6 RI from Côte d'Ivoire in 1983 [69] and 2002 [70], respectively. All bloodstream form culture lines were maintained in vitro in modified HMI9 medium [71] supplemented by 1.5 mM glucose, 1 mM methyl cellulose, 250 μM adenosine, 150 μM guanosine and 20% foetal bovine serum (FBS). Expression of the SRA human serum resistance gene in T. b. rhodesiense EATRO98 was maintained under selection with 1% normal human serum. Ectopic expression of functional T. b. brucei HpHbR in ELIANE was previously generated using the tubulin-targeting TbbHpHbR pTub-phelo construct (strain ELIANE TbbHpHbR-/+) [51], and maintained under phleomycin selection. Bloodstream form isolates BIM and S1/1/6 RI were grown from stabilate in donor BALB/C mice (Harlan, United Kingdom) and trypanosomes purified from blood by differential centrifugation as previously described [72]. Cells were maintained as for bloodstream culture cells lines at 37°C in 5% CO2 for up to 24 hours until use. All animal procedures were carried out in accordance with the Animals (Scientific Procedures) Act of 1986. Subspecies classification for T. b. gambiense group 1 strains was confirmed by a positive PCR result for the T. b. gambiense specific glycoprotein (TgsGP) gene and T. b. rhodesiense by a positive PCR result for the subspecies-specific serum resistance-associated (SRA) gene, as previously described [48]. T. b. brucei and T. b. gambiense group 2 strains were confirmed by a combination of negative SRA/TgsGP PCR results, the human serum sensitivity phenotype and their microsatellite genetic profile [73].
Sera Laboratories International, UK, provided pooled adult P. papio baboon serum derived from two individuals. Additional P. papio baboon serum, derived from a single adult male individual, was provided by Matrix Biologicals, UK. Normal human serum was obtained from a consented human donor and subject to appropriate ethical approval. The APOL1 protein levels in all serum samples are unquantified.
Trypanosomes were diluted to 5 x 105 parasites per ml in modified HMI9, with the addition of human serum or P. papio serum serially diluted in foetal bovine serum (FBS), or FBS only, to a total concentration of 20%. Assays were performed in a final volume of 200 μl in a standard 96 well plate at 37°C in a CO2-equilibrated incubator. The number of viable motile trypanosomes was quantified at 24 hours by haemocytometer counts under the microscope in triplicate, for at least three independent experiments. The percentage viability of parasites in human or P. papio serum was normalised relative to the FBS control for each cell line to account for inherent differences in strain growth rate. Dose–response curves and IC50 values were determined using GraphPad Prism software (version 7.0).
The H. sapiens (accession no. CCDS13926.1) or P. papio (accession no. KC197810) APOL1 open reading frame (ORF) was synthesised and supplied by GeneArt life technologies in an Invitrogen Gateway-compatible entry vector. The entry vector containing the APOL1 cDNA sequence, minus the N-terminal signal peptide (H. sapiens, residues 28–398; P. papio, residues 28–288) was cloned into pDest17 destination vector, which added an N-terminal 6xHis-tag, and transformed into BL21- AI competent E. coli. Protein expression was induced using 0.2% L-Arabinose for 16 hours at 37°C. Cells were lysed with urea lysis buffer (8 M urea, 20 mM Tris-HCl, 0.5 M NaCl, 5 mM imidazole, pH 8) and the cellular detritus removed by centrifugation at 5000g for 15 minutes. A small aliquot was removed for analysis by SDS-PAGE and Western blot with 1:5000 HRP-conjugated mouse anti-His antibody (Qiagen) and the remainder was used for protein purification under denaturing conditions. Denatured 6x His-tagged APOL1 protein was purified by passing the cell lysate through a gravity-flow Ni-Sepharose column (Gravitrap, GE Healthcare), and washing several times with urea lysis buffer pH 8 supplemented with increasing concentrations of Imidazole (5 mM-50 mM). Finally, bound protein was eluted with urea lysis buffer pH 8 containing 500mM imidazole. The eluate was dialyzed overnight against 20mM acetic acid and 0.05% tween and concentrated using 10,000 MW Vivaspin columns (Sartorius). Purity and concentration of the final purified protein was checked using a Qubit fluorometer (Thermo fisher) and SDS-PAGE (S1 Fig), then the concentration adjusted to 1 mg/ml and stored in aliquots at 4°C.
To assess survival in recombinant APOL1, trypanosomes were diluted to 5 x 105 parasites per ml in modified HMI9 containing 20% FBS and incubated with a dilution series of recombinant human or P. papio APOL1. The recombinant APOL1 was formulated in protein-free buffer (20mM acetic acid, 0.05% tween) and added in a volume of 10 μl to a final assay volume of 200 μl in a standard 96 well plate. A control containing an equivalent volume of protein-free buffer was also prepared. Assays were performed at 37°C in a CO2-equilibrated incubator, and the number of viable motile trypanosomes in each well was quantified at 24 hours by haemocytometer counts under the microscope in triplicate for at least three independent experiments. Cell counts in recombinant APOL1 were compared to control wells containing protein-free buffer only to determine percentage survival. In each assay, cells were incubated in 20% normal human serum as a positive control. Dose–response curves, IC50 values and one-way ANOVA were determined using GraphPad Prism software (version 7.0). Where indicated, trypanosomes were pre-incubated with 10 mM ammonium chloride (NH4Cl), a weak base, for 30 minutes at 37°C to reverse acidification of the endolysosome system prior to the addition of recombinant APOL1.
Samples for IFA were prepared as follows. All incubation steps unless stated otherwise were performed in a humidor at room temperature. Bloodstream form trypanosomes were diluted in HMI9 medium containing 20% FBS at a concentration of 106 parasites/ml and incubated with 50 μg/ml purified recombinant H. sapiens or P. papio APOL1 for two hours at 37°C. After this period, cells were washed once in serum-free HMI9 medium, and settled onto glass slides before fixing in 1% paraformaldehyde for 10 minutes. Samples were permeabilised using 0.1% Triton X-100 in PBS for 20 minutes then incubated in blocking solution (2% BSA in PBS) for 20 minutes. After washing three times in PBS, slides were incubated for 40 minutes with 1:500 mouse anti-p67 antibody (gift from Jay Bangs, Department of Microbiology and Immunology, University at Buffalo, NY, USA) in blocking solution. Washes were repeated and then primary antibody was detected using 1:1000 goat anti-mouse AlexaFluor594 secondary antibody (Life technologies) incubated for 40 minutes in blocking solution. To detect His-tagged APOL1 slides were washed three times in PBS and then incubated for 40 minutes with 1:500 AlexaFluor488 mouse anti-penta-His antibody (Molecular Probes, Invitrogen) in blocking solution. Following a final three washes the cells were treated with 50% glycerol, 0.1% DAPI, 2.5% 1, 4-diazabicyclo [2.2.2] octane (DABCO) in PBS, protected with a coverslip and sealed with acetone. Slides were imaged using the Deltavision Core system and SoftWorx package (Applied Precision) with standard filter sets (DAPI/FITC/Texas-Red and Light transmission). Approximately 30 serial sections through each trypanosome were taken for each filter. The images were composited and the brightness, contrast and color levels normalised between samples and exposures using the ImageJ software package (US National Institute of Health).
The University of Glasgow ethical review board approved the use of human serum in this study. The human serum volunteer gave written informed consent.
Trypanolytic activity against the human-infective East African T. b. rhodesiense sub-species has been demonstrated for sera from several members of the Cercopithecidae family, including baboons, mandrills and sooty mangabeys [14,37,58,59]. To date however, no primate has been identified with lytic activity against West African T. b. gambiense parasites. To determine the trypanolytic ability of serum from the West African Guinea baboon, P. papio, representative examples of the different T. brucei sub-species, were incubated for 24 hours in vitro, with a dilution series of P. papio or human serum. The strains selected included five different isolates of classic T. b. gambiense group 1, the cause of 97% of reported HAT cases [46], from a number of different disease foci in West Africa. As illustrated in Fig 1A, normal human serum efficiently lysed T. b. brucei bloodstream parasites (IC50; 0.0005%) in a 24 hour assay, but not strains of the human-infective T. b. rhodesiense or T. b. gambiense subspecies. In contrast, P. papio (pooled sera) was completely lytic to all tested strains, including both T. b. gambiense group 1 and 2 isolates, at concentrations ≥ 10% (Fig 1B). The sensitivity of T. b. brucei to P. papio pooled serum (IC50; 0.00035%) was comparable to that of T. b. rhodesiense (IC50; 0.00038%). T. b. gambiense group 1 and 2 strains however, were killed significantly less potently, with an IC50 approximately 70-fold (IC50; 0.024% serum, T. b. gambiense group 2) or 2000-fold (IC50; 0.46–1.68% serum, T. b. gambiense group 1) higher than that of the other sub-species, although still at a sub-physiological concentration. The trypanolytic activity of P. papio was also confirmed against a smaller collection of T. brucei strains using an alternative source of P. papio sera derived from a single male individual, which killed T. b. gambiense at a lower concentration > 2% (S2 Fig), presumably reflecting variation between individual animal samples.
APOL1 has been demonstrated to be the lytic factor in normal human serum [22,27,28], and T. b. rhodesiense-lytic orthologs of APOL1 have now been identified in the serum of a number of Old World monkey species, including species of the Papio baboon genus [14,37,58]. Furthermore this lytic activity of Papio APOL1 against T. b. rhodesiense has been demonstrated to be the result of a single polymorphism [62]. We therefore hypothesize that the broad lytic ability of P. papio may be attributable to a functional variant of this protein. Sequenced APOL1 cDNA was used as a template for the production of recombinant variants of P. papio and human APOL1 protein (S3 Fig-Amino acid alignment). Representative strains of the different T. brucei sub-species were incubated in the presence of purified P. papio and human recombinant protein to determine if APOL1 alone had demonstrable trypanolytic ability. Titrated human recombinant APOL1 protein completely lysed T. b. brucei parasites after 24 hours (IC50; 1.013 μg/ml), at concentrations comparable to the physiological levels of APOL1 reported for normal human serum [74–76], but had no lytic effect on strains of the human serum resistant parasites, T. b. rhodesiense, T. b. gambiense group 1 or T. b. gambiense group 2 (Fig 2A). In contrast, recombinant P. papio APOL1 protein exhibited trypanolytic activity against representative strains of all T. brucei sub-species (Fig 2B with additional T. b. gambiense group 1 strains assays provided in S4 Fig). Furthermore, strains of all sub-species tested appeared equally susceptible to the effect of recombinant P. papio APOL1, with no significant difference in IC50 observed (one-way ANOVA, F (3, 24) = 1.741, p = 0.19). Notably, as has been observed for human APOL1, this lytic activity is inhibited by the addition of the acidotropic agent ammonium chloride to the assay (Fig 2A and 2B). Ammonium chloride is a weak base that raises endolysosomal pH, thereby preventing pH-dependant conformational changes to APOL1 that are predicted to be essential to efficient ion-channel mediated lysis [32,34,77]. This corresponding inhibition of APOL1-mediated lysis for both orthologs is further indicative of a conserved mechanism of action. In summary these assays demonstrate that the P. papio APOL1 ortholog in isolation exhibits trypanolytic ability against all tested examples of the human-infective T. brucei sub-species. Although there may be other, as yet uncharacterized factors that contribute to the lytic ability of P. papio serum, the APOL1 ortholog is a significant trypanolytic component.
A reduced sensitivity to lysis was observed for both the predominant T. b. gambiense group 1 and minor group 2 strains, relative to T. b. brucei and T. b. rhodesiense, when incubated with P. papio serum, but not recombinant APOL1 protein. We postulated that for T. b. gambiense group 1, this difference might be the result of disparity in the rate of uptake of APOL1 versus APOL1–containing trypanolytic factors by these parasites. In normal human serum, HPR bound to haemoglobin, acts as ligand to facilitate TLF-1 uptake via the T. brucei HpHbR receptor [23,78]. However, a defining feature of T. b. gambiense group 1 strains is a decrease in TLF-1 internalisation as a result of reduced HpHbR expression and a conserved L210S substitution that reduces the binding affinity of HpHbR for its ligand [50,79]. Reduced TLF uptake via HpHbR contributes to the invariant human serum resistant phenotype of these parasites, although alone is insufficient to impart resistance to human serum [78] due to the existence of other speculated receptors for TLF-1 [80,81], and the additional TLF-2 particle in human serum for which the uptake mechanisms remain unknown [21,82,83]. In contrast, recombinant APOL1 is internalised by non-specific fluid phase endocytosis and trafficked through the endolysosome pathway, thus completely circumventing the HpHbR receptor [27,48].
The number of molecules in the TLF complex and its exact structural composition in baboon serum is currently unresolved, but a representative baboon species, P. hamadryas, has been demonstrated to have similar constitutive components (HPR and APOL1) to human TLF [14]. As T. b. gambiense group 1 parasites have a reduced uptake of human TLF-1 but the other subspecies do not we postulated that a similar mechanism could reduce the uptake of P. papio TLF particles by T. b. gambiense group 1 strains, which is corrected by direct incubation in recombinant APOL1 protein. To investigate this we repeated the serum resistance assays using a T. b. gambiense ELIANE strain expressing a functional T. b. brucei HpHbR receptor (ELIANE TbbHpHbR -/+), that was previously generated by our laboratory and demonstrated to take up comparable amounts of TLF-1 to T. b. brucei [51]. As previously observed, expression of the functional T. b. brucei HpHbR receptor alone was insufficient to convert the phenotype of T. b. gambiense to human serum sensitivity and this clone (TbbHpHbR -/+ T. b. gambiense) retains full resistance to normal human serum (Fig 3A). However, it exhibits a 1000-fold increased sensitivity to P. papio serum (relative to the wild-type T. b. gambiense group 1 ELIANE strain), producing an IC50 value (0.0005%) comparable to that observed for the T. b. brucei and T. b. rhodesiense sub-species (Fig 3B and S2 Fig). Taken together, the serum and APOL1 assays indicate that diminished TLF uptake via the HpHbR receptor, rather than higher innate resistance to P. papio APOL1-mediated lysis underlies the increased resistance to P. papio serum observed for T. b. gambiense group 1 strains.
In T. b. gambiense group 2, in contrast, an as yet uncharacterised HpHbR–independent mechanism/s determines human infectivity. T. b. gambiense group 2 strains, including the STIB386 isolate used in this study, have been shown to express the HpHbR gene at level comparable with T. b. brucei, with no demonstrable reduction in TLF-1 uptake [48]. Consequently, the reduced sensitivity to P. papio serum lysis, but not APOL1 protein, also observed for these HpHbR-functional parasites, further indicates that important differences exist in the cell biology of between T. b. gambiense group 2 and T. b. gambiense group 1 strains that determine sensitivity to these primate lytic factors.
Human recombinant APOL1 is taken up by fluid phase endocytosis and trafficked through the endocytic pathway to the endolysosome, the initial activation site of APOL1, in all T. brucei sub-species [27,48]. This results in lysis of T. b. brucei but not of T. b. rhodesiense or T. b. gambiense [48], which each possess mechanisms to resist the lytic effects of APOL1 [35,48,51,52]. To determine if P. papio APOL1 is localised through the parasite endolysosome pathway in a similar manner to that demonstrated for human APOL1, uptake of both recombinant proteins was compared in T. b. brucei and T. b. gambiense group 1 parasites using a fluorescent antibody to detect the His-tagged recombinant APOL1 protein. The cells were then examined by microscopy, in conjunction with the lysosomal marker p67. In order to achieve images of APOL1 uptake we used high concentrations of APOL1 (material and methods) to counteract possible degradation of APOL1 in the lysosome. Consistent with previous experiments of serum and APOL1 uptake in our laboratory [48,49,51], no lysosomal swelling was observed. As shown in Fig 4, both human and P. papio APOL1 are internalised by T. b. brucei and T. b. gambiense after a two hour incubation and are observed to co-localise with an antibody directed against the lysosomal membrane protein p67, indicative of the parasite endolysosome pathway [84,85]. These observations, in parallel with the ablation of lysis observed after co-incubation with acidotropic agent, ammonium chloride in APOL1 lysis assays, suggest that as previously demonstrated for human APOL1, exposure of the protein to the low pH of the endolysosomal pathway is also a requirement for trypanolytic activity of the baboon APOL1 ortholog.
The ancient co-evolutionary engagement of African trypanosomes with their mammalian hosts has shaped an innate lytic molecule in man that protects from infection with most African trypanosomes. In response, the extensive antigenic repertoire of T. brucei [86] has provided a rich resource from which to evolve counter-measures to APOL1-mediated lysis on at least two occasions; SRA in T. b. rhodesiense in East Africa [35,36,87], and TgsGP in T. b. gambiense group 1 in West Africa [51,52,88]. In this study we present a novel APOL1 variant from a species of West African baboon that killed examples of all T. brucei sub-species, including T. b. rhodesiense, T. b. gambiense group 2, and T. b. gambiense group 1, the agent of most current cases of human African trypanosomiasis. The identification of such genetic variants, capable of killing both animal and human-infective parasites presents new opportunities for unconventional approaches to disease treatment and control, using APOL1-based biological therapies.
Previous studies have identified APOL1 orthologs in a subset of Old World monkeys [14,62], and an APOL1 variant with a key similarity in some humans with African ancestry [63], that encode proteins lytic to T. b. rhodesiense. In both variants, evidence suggests protection is mediated by the position of a single lysine residue in the C-terminal protein domain that obstructs coiled-coil interactions with SRA, thus allowing APOL1-directed lysis to proceed unimpeded [62]. Unfortunately in humans, the two amino acid deletion that alters the SRA-binding region in this APOL1-G2 variant come with an associated fitness cost: a 7–29-fold increased risk of developing a wide spectrum of kidney disorders in individuals carrying two copies of a variant allele [63,89–92]. The exact biological mechanism underlying this APOL1-associated nephropathy is not yet known but appears to be specific to the human variants. Engineered versions of the human APOL1 variant transiently expressed in a mouse model caused significant toxicity to the organ of expression (liver), which was not observed with baboon APOL1 or human APOL1 modified to introduce only the protective baboon lysine to the C-terminus [62]. This is an encouraging result, and such baboon-like APOL1 variants are now the focus of efforts to create suitable mechanisms of delivery, such as the conjugation of APOL1 protein to an antibody fragment targeted to parasite surface antigens [93] and an ambitious project to create targeted transgenic cattle expressing variant APOL1 [15].
These variants could be used to protect the reservoir host species from zoonotic T. b. rhodesiense sleeping sickness in addition to animal trypanosomiasis, which places severe restrictions on agricultural production and rural development in Sub-Saharan Africa [1]. Unfortunately, they will have a limited effect on the overall burden of human sleeping sickness. None of the APOL1 variants used in these experiments are able to kill the major human pathogen T. b. gambiense group 1 which places a population of 57 million people in West and Central Africa at risk of disease [4], less than 5% of whom are currently under surveillance [94]. Furthermore, there is a risk that the proposed interventions could result in the creation of a vacant ecological niche that increases the incidence of T. b. gambiense group 1 in domestic livestock through selective removal of susceptible competitor species such as T. b. brucei, T. congolense, T. vivax and T. b. rhodesiense.
We have addressed these concerns directly in this study by examining the serum of a West African baboon species P. papio that overlaps in distribution with that of T. b. gambiense, and which had been suggested to self-cure T. b. gambiense group 1 infection [64]. In that study primates infected with T. b. gambiense group 1 parasites exhibited a serological response that decreased throughout the course of the experiment and had no detectable parasitemia, consistent with an initial infection, followed by rapid parasite clearance and self-cure. In our study P. papio serum is able to lyse T. b. gambiense in 24 hours in vitro. The difference in timing of parasite killing between the in vivo and in vitro experiments, which could be due to a number of different factors such as parasite sequestration, is a well-recognised phenomenon. It is possible that parasites avoid lysis by residing in sites of low APOL1 concentrations (for example at the bite site in the skin) in the animal before eventually being cleared. This factor must be taken into account when attempting to develop APOL1-based therapies as in vitro assays do not always reflect the complexity of in vivo cell biology. The introduction of improved bioluminescent imaging to quantify parasite burden could be used to test in vivo for complete parasite clearance.
We have shown that the lytic effect of P. papio serum can be reproduced with an ortholog of the trypanolytic primate defence protein, APOL1, which demonstrates the uptake and localisation characteristics of other previously identified APOL1 proteins [27,48]. The trypanolytic action of this P. papio APOL1 variant against T. b. rhodesiense can be attributed to the C-terminal lysine mutation that is conserved among several members of the Cercopithecine subfamily that includes baboon, mandrills and mangabeys [62]. However the mechanism by which it counters T. b. gambiense, which has evolved multiple contributing mechanisms of human serum resistance, remains more elusive. All T. b. gambiense group 1 parasites share a mutated HpHbR with reduced affinity for one of the human APOL1-containing particles (TLF-1) via the HPR ligand [48–50], although a second particle, TLF2, appears to have alternative, as yet unresolved, mechanism(s) of internalisation [24,80,81,83]. The exact composition of TLF in baboon serum has not been clarified. However analysis of an HPR-affinity purified HDL sub-fraction from P. hamadryas baboon serum detected a TLF-equivalent particle that contains the same structural components as human TLF [14]. Furthermore, when transiently expressed in mice, all three components were required for maximum lytic activity against T. b. rhodesiense [14], suggesting HPR-HpHbR may play a role in uptake of baboon TLF. Here we show that T. b. gambiense group 1, although still fully susceptible to sub-physiological concentrations of P. papio serum, was 1000-fold less sensitive than T. b. brucei sub-species. This difference was ablated when functional T. b. brucei HpHbR was restored to the T. b. gambiense parasite, supporting a role for P. papio TLF uptake via both HpHbR-mediated endocytosis as well as unidentified alternative mechanisms, possible shared with those already proposed for human TLF [24,80,81,83].
Secondly, the TgsGP gene has been demonstrated to be essential for human serum resistance in T. b. gambiense group 1, as gene deletion renders the parasites sensitive to human serum lysis [51,52]. In contrast to the T. b. rhodesiense SRA protein, TgsGP and APOL1 do not appear to interact directly. Instead, TgsGP is proposed to bolster T. b. gambiense resistance to human APOL1 pore-forming activity through a process of plasma membrane stiffening [52]. A third mechanism by which T. b. gambiense might resist the actions of NHS, through enhanced APOL1 degradation within the endolysosomal system, has also been proposed [52]. Modulation of expression levels of the cysteine protease Cathepsin L and its inhibitor (ICP) has demonstrated an important role for cathepsin-mediated degradation of APOL1 in human serum resistance [53]. Difference in expression levels of these genes has not been detected in T. b. gambiense, however a lower pH is observed within the early endosomes that is predicted to accelerate their proteolytic activity relative to T. b. brucei [52]. Intriguingly, we observed equal sensitivity of all strains tested to P. papio APOL1-directed lysis, suggesting that the activity of TgsGP, and APOL1 degradation by cysteine peptidases, that effectively hinders human APOL1 in T. b. gambiense, poses no such barrier to the P. papio variant. This raises interesting questions about how exactly P. papio APOL1 is able to overcome these factors? Many details of the action of the TgsGP protein in particular remain cryptic. Despite its essential role in human serum resistance in T. b. gambiense, ectopic expression of T. b. gambiense TgsGP alone in T. b. brucei is insufficient to confer resistance to human serum [51,88]. There is evidently a role for other, as yet unidentified processes, in T. b. gambiense human serum resistance, which are absent or incomplete in T. b. brucei.
Sequence analysis has revealed that baboon and human APOL1 orthologs share only 58% amino acid sequence identity [14]. Despite this, in the recently elucidated example of baboon serum lysis of T. b. rhodesiense it was demonstrated that a single amino acid substitution conserved between baboon species is responsible for APOL1 evasion of SRA binding [62]. Uncovering the mechanism by which P. papio has developed its broad trypanolytic ability may offer further insights into the workings of T. b. gambiense human serum resistance, as well as aid in the design of an improved APOL1 therapy that could target all pathogenic trypanosomes across Sub-Saharan Africa. Such universal therapies that can treat both animal and human pathogens are particularly appropriate to the “one health” approach, currently advocated by WHO, FAO, and OIE, that integrates medical and veterinary health policy and research for addressing zoonotic diseases.
The Guinea baboon P. papio is found only in a limited area of western equatorial Africa, where its range overlaps with that of T. b. gambiense group 1. Five other baboons are represented in the Papio genus of which serum for only one, the east African P. cynocephalus (yellow baboon) has been previously tested against T. b. gambiense parasites, and was reported to be non–lytic [37]. Unfortunately APOL1 sequence is currently unavailable for comparative analysis with this species or the southern African P. ursinus (Chacma baboon) and P. kindae (Kinda baboon) species from Central Africa. Of the remaining Papio species, APOL1 sequences from cDNA have been successfully obtained for P. hamadryas (Hamadryas baboon) from North East Africa, and Central African P. anubis (Olive baboon) [62], the closest related species to P. papio in a recent phylogenetic study of mitochondrial DNA [95]. Amino acid alignments of P. papio APOL1 with these available sequences indicate ~98.5% identity to P. hamadryas and 93.5% to P. anubis (S3 Fig). A study in which C-terminal polymorphisms of P. anubis were incorporated into human recombinant APOL1 were observed to be lytic to T. b. rhodesiense but not T. b. gambiense [37], however full length APOL1 transcripts, unavailable at the time of the study, have not been tested. For P. hamadryas, serum and APOL1 have not yet been tested against T. b. gambiense, however a laboratory infection of two individual baboons with a strain of T. b. gambiense group 1 suggested hamadryas baboons to display a level of trypanotolerance to infection [64]. Future studies in which the sensitivity of T. brucei subspecies to serum and APOL1 from the other baboon species, followed by the construction of chimera mutants are now needed to help resolve the crucial polymorphisms responsible for T. b. gambiense lysis, as has been successful for T. b. rhodesiense.
|
10.1371/journal.pcbi.1004942 | Competitive Dynamics in MSTd: A Mechanism for Robust Heading Perception Based on Optic Flow | Human heading perception based on optic flow is not only accurate, it is also remarkably robust and stable. These qualities are especially apparent when observers move through environments containing other moving objects, which introduce optic flow that is inconsistent with observer self-motion and therefore uninformative about heading direction. Moving objects may also occupy large portions of the visual field and occlude regions of the background optic flow that are most informative about heading perception. The fact that heading perception is biased by no more than a few degrees under such conditions attests to the robustness of the visual system and warrants further investigation. The aim of the present study was to investigate whether recurrent, competitive dynamics among MSTd neurons that serve to reduce uncertainty about heading over time offer a plausible mechanism for capturing the robustness of human heading perception. Simulations of existing heading models that do not contain competitive dynamics yield heading estimates that are far more erratic and unstable than human judgments. We present a dynamical model of primate visual areas V1, MT, and MSTd based on that of Layton, Mingolla, and Browning that is similar to the other models, except that the model includes recurrent interactions among model MSTd neurons. Competitive dynamics stabilize the model’s heading estimate over time, even when a moving object crosses the future path. Soft winner-take-all dynamics enhance units that code a heading direction consistent with the time history and suppress responses to transient changes to the optic flow field. Our findings support recurrent competitive temporal dynamics as a crucial mechanism underlying the robustness and stability of perception of heading.
| Humans have little difficulty moving around in dynamic environments containing other moving objects. Previous research has demonstrated that moving objects may induce biases in perceived heading in some circumstances. Nevertheless, heading perception is surprisingly robust and stable. Even when large moving objects occupy much of the visual field and block our view of the future path, errors in heading judgments are surprisingly small—usually less than several degrees of visual angle. Furthermore, perceived heading does not abruptly shift or fluctuate as moving objects sweep across the observer’s future path. The aim of the present study is to investigate the qualities of our visual system that lead to such robust heading perception. We simulated two existing models that specify different heading mechanisms within the visual system and found that they could not capture the robustness and stability of human heading perception in dynamic environments. We then introduced the competitive dynamics model that succeeds due to its reliance on recurrent, competitive interactions among neurons that unfold over time that stabilize heading estimates. Our results suggest that competitive interactions within the visual system underlie the robustness and stability of human heading perception.
| Humans move through an often cluttered world with ease. We effortlessly walk and drive through busy streets in everyday life without colliding with other moving pedestrians. These competencies require an accurate, reliable, and stable perception of the direction of self-motion (i.e., heading). Although heading perception is inherently multisensory, with contributions from the vestibular [1–3] and motor [4] systems, vision represents the dominant sensory modality for many animals [5–7]. Forward self-motion along a linear trajectory produces a field of radially expanding optic flow that emanates from a singularity known as the focus of expansion (FoE), which coincides with the direction of travel in the absence of eye movements.
It is well established that the primate visual system is sensitive to and uses information in optic flow to perceive heading. Single neurons in the dorsal medial superior temporal area (MSTd) [8–10], ventral intraparietal area (VIP) [11,12], and other brain areas, are tuned to the direction of self-motion through three-dimensional (3D) space. Such neurons are sensitive to radial fields of motion with different FoE positions that encompass much or all of the visual field that is experienced during self-motion [13,14]. At the population level, the largest proportion of neurons is tuned to FoE positions that correspond to straight-ahead headings [10], which is to be expected for a system that depends on optic flow to perceive the direction of self-motion during locomotion.
Much of what is known about heading perception comes from psychophysical experiments wherein human subjects view computer displays of simulated self-motion and judge the perceived direction of travel once the trial concludes. Remarkably, humans can judge their heading to within 1–2°, even when the environment contains as few as ten dots [15,16]. Heading judgments remain accurate despite the presence of substantial amounts of noise or the removal of large portions of the flow field [17,18]. The accuracy of human heading judgments is consistent with the tuning of MSTd neurons that shows maximal sensitivity around the straight ahead [2,19].
Heading perception is not only accurate, it is also remarkably robust and stable. These qualities warrant further investigation and are the focus of the present study. The robustness and stability of heading perception are especially evident in dynamic environments containing independently moving objects. Regions of the optic array corresponding to moving objects generally contain optic flow that is inconsistent with the background optic flow and uninformative about heading. Nonetheless, heading perception is biased by moving objects by no more than a few degrees. Objects that approach the observer in depth (approaching objects; Fig 1A) induce a bias in the direction opposite the object motion of ~3° [20]. When objects maintain a fixed depth with respect to the observer as they move laterally (fixed-depth objects; Fig 1B), heading perception is biased by ~1° in the direction of object motion [21]. Objects that recede in depth from the observer as they move across the observer’s future path (retreating objects; Fig 1C) yield a heading bias in the direction of object motion of less than 3° (Layton & Fajen, in preparation).
These biases are surprisingly small when one considers the conditions in which they are induced. The moving objects in the aforementioned experiments were generally large and moved near the observer, such that a sizeable proportion of the visual field contained discrepant optical motion. They often crossed the observer’s future path, thereby occluding the region of the optic array near the background FoE (as in Fig 1), which is known to be the most informative region for heading perception [16,22]. Moving objects that approach the observer in depth generate a radial pattern of optic flow with a FoE of their own that may be offset from the background FoE by much more than a few degrees. In some circumstances, all of these potential complications may occur at the same time. In Fig 1D, for example, the moving object occupies approximately half of the visual field, occludes the background FoE (indicated by the blue dot), and generates radial motion with a FoE (red dot) that is offset from the background FoE by 20°. The fact that heading perception under these conditions is biased by no more than a few degrees attests to the robustness of the visual system.
Furthermore, such biases are induced only when objects cross or move near the observer’s future path [20,23]. Objects that move far away from the future path do not influence heading perception. Nonetheless, our experience when an object approaches and eventually crosses our future path is not that heading abruptly shifts; that is, humans do not perceive themselves as moving in one direction at one instant and then in a different direction at the next instant when the object begins to cross the path. Heading perception is more stable and less susceptible to fluctuations.
Previous research on heading perception in the presence of moving objects [20,21,24] has focused on the sources of bias. The fact that heading perception is as reliable and stable as it is under such conditions has been largely overlooked. Nonetheless, these qualities of heading perception are worthy of investigation. Understanding the mechanisms that underlie the robustness and stability of heading perception was the primary aim of the present study.
We test the hypothesis that the robustness and stability of heading perception is rooted in a particular form of temporal dynamics within the visual system—specifically, recurrent competitive interactions that unfold over time among units in area MSTd. As a simple demonstration that heading perception has an important temporal component, consider a scenario in which a moving object approaches and crosses the observer’s future path from the left or right. Recall that heading perception is biased under these conditions but that the bias is surprisingly small. One possible reason why the bias is not larger than it is, is that heading perception is based on the temporal evolution of the optic flow field, including not only the period of time when the object crossed the path but also prior to this point, before the more informative regions of the flow field were occluded by the object. Although this may seem obvious, existing models of heading perception [20,25] have no features for capturing heading perception as a process that evolves over time. As we explain below, these models estimate heading based on the instantaneous flow field and generate a new estimate that is independent of the previous one at each successive instant.
We tested the role of temporal dynamics in an experiment in which human subjects made heading judgments in the presence of an object that approached from the side and crossed the observer’s path at the end of the trial [23]. Stimulus duration was varied between 150 and 1500 ms. Importantly, within each object trajectory condition, the last 150 ms was the same across stimulus durations, but conditions with longer durations also included the earlier part of the event leading up to the last 150 ms prior to occlusion of the background FoE. If heading perception is based on the temporal evolution of the optic flow field, the heading bias should be weaker in the longer duration conditions because the visual system should be able to use the information from the earlier part of the trial—before the object occluded the path—to improve the accuracy of the estimate.
Indeed, when stimulus duration was short (i.e., when subjects only saw the last part of the trial), they exhibited a very large heading bias (~6°). However, the bias was dramatically reduced when stimulus duration was longer–that is, when the earlier part of the trial before the object crossed the path was also included in the stimulus. The findings indicate that heading perception is based on the evolution of the optic flow field and the ability to integrate information over time underlies the surprising accuracy and stability of heading perception. In other words, temporal dynamics offers a candidate solution to the problem posed above about why heading perception is not more biased by moving objects than it is, and why we do not experience abrupt transitions in perceived heading.
These findings provide compelling evidence that heading perception is based on the evolution of the optic flow field, but are not especially informative about the nature of the underlying neural mechanisms. In the present study, we used modeling and simulation to test the sufficiency of a particular type of mechanism involving on-center/off-surround recurrent interactions among MSTd neurons. Neurons in a recurrent network send inhibitory feedback to other neurons to balance the excitatory feedback they send themselves to potentiate their activity. Recurrent dynamics among neurons in MSTd are compatible with the neuronal decay time constant in MSTd (81 msec), which is as much as five times slower than that of areas from which MSTd receives input and indicates a persistence in the heading signal long after the visual optic flow signal ceases [26]. Competition among heading-sensitive neurons in MSTd over time through recurrent interactions may exert nonlinear effects on the heading signal—the contrast of the heading signal could be enhanced and the uncertainty about heading reduced. In the language of Bayesian inference, MSTd neurons may update the network’s belief about heading over time [27]. As we demonstrate below, these interactions that unfold over time may serve as a mechanism to stabilize heading perception, even when the visual signal is temporarily disrupted.
Our approach is to compare the performance of a model with recurrent temporal dynamics in MSTd against two models without this property, focusing on whether these models capture the spatio-temporal robustness of human heading perception in dynamic environments. The model with recurrent temporal dynamics is an updated version of the model introduced by Layton, Browning, & Mingolla [28]. The other two models are the motion pooling model developed by Warren & Saunders [20] and the differential motion model developed by Royden [25]. These models were chosen because they are representative of existing biological modeling approaches and because they were designed to estimate heading in the presence of independently moving objects. In all three models, the similarity is computed between the optic flow field (or a transformation thereof) and a number of vector field templates containing radial expansion with different FoE positions. Each template resembles the canonical receptive field organization of a MSTd cell selective to a particular FoE location in the visual field and is center-weighted—motion vectors nearby the preferred FoE position are weighted greater than those located further away. Such templates have characteristics that are consistent with those that develop in models using supervised and unsupervised learning [29–32]. Motion pooling models have demonstrated that matching these biologically inspired global motion templates with the patterns of optic flow that arise during self-motion provides a plausible means for cells in MSTd to extract heading [33–36] (but see [37]). The template match feeds MSTd units with their inputs and the preferred FoE position of the most active unit reflects the heading estimate of the model.
The differential motion model [25] is distinct from the others in that the template match is performed on a field of difference vectors rather than the optic flow field. Differential motion models were originally proposed to account for the ability to perceive heading while making eye movements, which introduce rotation into the flow field. Any instantaneous optic flow field can be decomposed into translational and rotational components [38]. A vector’s translational component depends on the corresponding point’s depth in the environment, whereas the rotational component does not. Therefore, subtracting nearby motion vectors that correspond to points at different depths within the environment eliminates the rotational component and results in a scaled version of the translational component. Because the translational component is informative about the observer’s heading and the rotational component is not, certain difference vectors may be used to recover a heading estimate. Rieger & Lawton [39] developed the first differential motion algorithm to compute heading based on a decomposition of translational and rotational flow when differential motion parallax is present. Hildreth [40] later extended the approach with a voting procedure to account for the presence of moving objects. A number of neural models that decompose flow into translational and rotational components have successfully simulated properties of MT and MST [41,42] (but see [43]).
The differential motion model developed by Royden and simulated in the present study [25] is a refinement of an earlier version [44] to situate the differential motion algorithm in a biological framework. The field of difference vectors in the Royden model is obtained by processing the optic flow field with motion sensitive units with antagonistic surrounds whose properties resemble those of cells in primate MT—. These operators respond optimally when a sharp change in speed occurs within the receptive field along the preferred motion direction, which may coincide with a sudden change in depth and result in motion parallax: near background motion results in a faster optical speeds than far background motion. The heading estimate in the differential motion model is the direction that corresponds to the preferred FoE of the most active center-weighted template.
Both the motion pooling model [20] and the Layton et al. model (as well as its successor introduced here) compute the template match directly on the optic flow field. We refer to the latter as the competitive dynamics model to highlight its unique feature—that it is a dynamical system that continuously integrates optic flow within a competitive network of MSTd neurons [28]. This differs from the differential motion and motion pooling models, which only process vector fields at independent points of time.
Two other models, neither of which can be classified as motion pooling or differential motion models, warrant mentioning. First, the analytic model of Raudies & Neumann [45] relies on neither motion differences nor templates to account for the pattern of human heading biases in the presence of moving objects, but rather a weighted combination of segmentation cues derived from the flow field. Heading bias arises in the model even without segmentation cues because the moving object induces a discrepancy compared to analytic parameters that describe the observer’s self-motion in a static environment. The pattern of bias produced by the model does not resemble that of humans, but segmenting the optic flow field by accretion/deletion, expansion/contraction, and acceleration/deceleration improves the correspondence. Second, Saunders & Niehorster [46] cast the problem of estimating heading in the presence of moving objects into a Bayesian context whereby the objective is to estimate the translational and rotational components of an ideal observer from optic flow along with the depth of points in the scene. The model estimates the posterior probability that an observer moves along a particular heading by multiplying the likelihoods that each motion vector in the optic flow pattern was independently generated by a particular combination of observer translation and rotation parameters. The model accounts for human heading bias in the presence of approaching and fixed-distance objects. We will not give further consideration to either model in our simulations below because both process vector fields at independent points of time and because our focus in the present study is on neural models.
We simulated the differential motion, spatial pooling, and competitive dynamics models under a variety of conditions to test for robustness and stability in heading estimates (see Methods section for details about the models and simulations). To anticipate the results, we found that the differential motion and spatial pooling models yield erratic, sometimes wildly fluctuating heading estimates over time. Furthermore, simply adding temporal smoothing of optic flow signals to these models does not capture the spatio-temporal characteristics of human heading perception. In contrast, the estimates from the competitive dynamics model are less biased by moving objects, less variable, more stable, and more similar to human heading estimates in the presence of moving objects. Taken together, the findings imply that competitive interactions within MSTd are a plausible mechanism to account for the robustness and stability of human heading perception.
Previous efforts to evaluate the differential motion and motion pooling models have focused on how accurately they reproduce patterns of human heading judgments [20,25,28]. Both models succeed in capturing the heading bias in humans for approaching objects, but only the differential motion model has been shown to match human judgments for fixed-depth objects. It has been argued that the motion pooling model fails to capture the human heading bias for fixed-depth objects [25,28], but this has not actually been formally tested. Neither model has been evaluated in the retreating object scenario. More importantly, the robustness and stability of these models has never been systematically explored.
In this section, we examine model estimates of heading during self-motion in the presence of moving objects that cross the observer’s future path while approaching (Fig 2A and 2B), maintaining a fixed-depth (Fig 2C), or retreating (Fig 2D). The blue and gold curves in each plot show the mean heading error over time for the differential motion and motion pooling models, respectively, with lighter shaded regions indicating ±1 SE. The color of the horizontal bar at the top of each subplot in Fig 2 indicates when the moving object is crossing the observer’s future path (red), as well as the portions of the trial before (orange) and after (green) crossing.
First, we consider simulations with an approaching object that crosses the observer’s future path at a 15° angle (7.5° object FoE offset) (Fig 2A). The typical bias in human judgments in the presence of objects that approach at comparable angles is about 2.5° in the direction opposite object motion [23]. Note that subjects in the human experiments made judgments after viewing the entire stimulus, so the existing data are not informative about how perceived heading evolves over time as the object changes position in the visual field. As such, we represent typical human performance in Fig 2 using a single dot positioned at the far right of the figure with a dashed line of the same color for reference.
The motion pooling model initially yields unbiased heading estimates but quickly exhibits a ~6° bias in the direction opposite of object motion as the object approaches and crosses the future path. This is consistent with the human data in direction but greater in magnitude. The heading bias remains relatively stable while the object occludes the background FoE. The differential motion model also yields a bias (~7°) in the direction opposite object motion during object crossing, but the bias arises later. That is, the differential motion model yields accurate heading estimates for a longer period of time while the object occludes the heading direction compared to the motion pooling model.
Although the heading biases from both models are only slightly greater than those exhibited by humans, model performance deviates from human judgments much more dramatically at larger object trajectory angles. When the object approaches along a 70° angle (35° object FoE offset), the differential motion model exhibits a bias that exceeds that of humans by a factor of 10 (Fig 2B). The motion pooling model yields a large bias in the direction of object motion, followed by a dramatic reversal in the opposite direction. The large initial bias in the direction of object motion (positive in Fig 2B) was unexpected because the motion pooling model is known to exhibit heading bias in the opposite direction for approaching objects (i.e., toward the object FoE or negative in Fig 2A) [20,24]. The positive bias arises from the strong rightward radial motion of the background flow ahead of the leading edge of the moving object, which activates templates weighted to the far right of actual heading. The level of activation is only moderate because the rightward flow is not a perfect match for templates in that direction. Nonetheless, the activation level is higher than for other templates, including those more closely aligned with the object FoE. This is because as the object draws closer in depth, the spatial distribution of dots nearby the object FoE becomes sparser. Radial templates weighted nearby the object FoE are only weakly activated because of the limited amount of motion. Eventually, as the object continues to cross the observer’s path, it occludes enough of the background flow to diminish activation of templates weighted to the far right. At this point, the most active templates are those closely aligned with the object FoE, causing the heading estimate to abruptly reverse and exhibit a bias in the direction opposite object motion.
Although the direction of the bias generated by the motion pooling model was initially unexpected, the magnitude of bias in both models is not surprising given that both models estimate heading based on the instantaneous flow field, which is dominated by discrepant object motion toward the end of the trial. Nonetheless, human heading judgments are far more robust even when objects approach at larger angles [24].
Fig 2C shows the simulation results with a fixed-depth object. Both the differential motion model and motion pooling model yield unbiased or weakly biased heading estimate before and after the objects crosses the observer’s path. During the crossing period, the differential motion model produces a 1–2° heading bias in the direction of object motion, consistent with human heading judgments. On the other hand, the motion pooling model produces a weak bias in the direction opposite object motion, which is not consistent with human heading judgments. Variability is slightly larger in the differential motion model, except for the spike that occurs in the motion pooling model estimates when the object begins to cross the future path.
Fig 2D depicts the simulation results for the retreating object scenario. The differential motion model yields accurate heading estimates until shortly before the object crosses the path, at which point there is a sharp rise in the bias. The 1–2° bias in the direction of object motion is consistent with the human data (Layton & Fajen, in preparation). Model variability is comparable to that obtained for the fixed-depth object. The heading error generated by the motion pooling model gradually ramps up while the object is approaching the observer’s path, and then sharply reverses to a bias in the direction opposite object motion. The reversal and subsequent gradual bias reduction occur because templates in the model respond to the radial-like motion pattern created by the trailing edge of the object and the background. The most active template in the model tracks the position of trailing edge, which progressively moves toward the heading direction, resulting in a weakening of the bias.
Although the mean estimates from the differential and motion pooling models match those of human observers in some conditions, the model and human estimates differ dramatically in direction and/or magnitude under other conditions. Furthermore, these models do not exhibit the stability that is characteristic of human heading perception. Heading estimates from both models often changed very abruptly, increasing or decreasing by many degrees of visual angle in less than 100 ms. If human heading perception was subject to such wild fluctuations, moving objects would induce easily noticeably shifts in perceived heading as they approach and cross the future path. Yet both psychophysical studies and introspection while driving or walking in busy environments suggest that heading perception is far more stable and that any changes in perceived heading are small and too gradual to be noticed.
Because the differential motion and motion pooling models were not designed to integrate optic flow over time, we explored whether smoothing the activation of model MSTd units over time with a moving average would address some of the issues with the stability of heading estimates. This was implemented by applying a 3 (‘Low’), 6 (‘Med’), or 9 (‘Hi’) frame moving average to the activation produced by each unit in the 2D MSTd array. As illustrated in Fig 3, temporal smoothing was not effective. In both models, the large heading biases, abrupt changes, and reversals remained even with a high degree of temporal smoothing. Of course, the degree of smoothing could be further increased, but that would not qualitatively change the estimates and would introduce significant lag into the signal, making the model sluggish in response to actual changes in heading.
We now introduce the competitive dynamics model, which is based on the model of Layton et al. [28], and explore whether it better captures the robustness and stability of human heading perception. The model contains areas that correspond to the primate retina, lateral geniculate nucleus (LGN), primary visual cortex (V1), medial temporal area (MT+), and the dorsal medial superior temporal area (MSTd) (see Methods for details). These areas are organized into three functionally distinct stages: sensitivity to change in the retina; motion detection in LGN, V1, and MT+; and self-motion estimation in MSTd. The self-motion estimation mechanisms are the same as those in the model of Layton et al. [28], but the stages for sensitivity to change and motion detection are new. As a large dynamical system, populations of neural units in each area obey systems of Hodgkin-Huxley-like ordinary differential equations. In other words, model cells in each area temporally integrate the network response to the optic flow time history and the bottom-up signal derived from the presently available optic flow.
The main feature that differentiates the competitive dynamics model from the differential motion and motion pooling models is the use of recurrent competitive dynamics among model MSTd cells that unfold over time. Units in model MSTd obey on-center/off-surround recurrent competitive dynamics, which refine the heading estimate over time. Each unit competes for its heading representation: a unit enhances the signal for its preferred heading through self-excitation and suppresses other heading signals generated through the pattern of activity produced by other units in the network. As a dynamical system, the network takes time to develop a reliable heading estimate, which persists for some time—even if the optic flow signal is interrupted. The competitive dynamics that refine the heading estimate and the persistence of activity over time in the competitive dynamics model may hold the key to robust heading perception.
To test the performance of the model, we ran simulations under the same conditions used to test the previous models. In Fig 4, we plot the mean heading error produced by the competitive dynamics model for the approaching, fixed-depth, and retreating object conditions. The mean heading bias reaches ~2.5° in the direction opposite object motion for the object approaching along a 15° angle, ~4° in the same direction for an object approaching along a 70° angle, ~2.5° in the direction of object motion for the fixed-depth object, and ~4° in the direction of object motion for the retreating object. Like the differential motion model, the competitive dynamics model yields heading biases that consistently match those of human observers in direction. Unlike the differential motion and motion pooling models, however, heading error builds up in each condition over several hundred milliseconds and does not exhibit large, sudden excursions or reversals in the direction of bias. Furthermore, the competitive dynamics model yields far more accurate, stable, and human-like heading estimates when objects approach at extreme angles (e.g., 70°) compared to the differential motion and motion pooling models (compare Figs 4B and 2B, noting the difference in the scale of the y-axis). In summary, the competitive dynamics model better replicates the complex pattern of human heading biases across variations in object trajectories, and yields heading estimates that are relatively stable and change gradually.
Before we can conclude that the improvement in performance is due to recurrent competition, it is necessary to rule out other differences between the competitive dynamics model and the differential motion and motion pooling models. As such, we simulated a version of the competitive dynamics model without connectivity between model MSTd units. The thick fuchsia curves in Fig 5 depict the model performance on the approaching, fixed-depth, and retreating objects simulations when we lesioned connections within model MSTd. Without recurrent competition, heading estimates change more abruptly, are strongly influenced by the object FoE in the presence of the objects that approach along 15° and 70° trajectories, and are in the incorrect direction in the presence of the fixed-depth and retreating objects. Together, these findings support the hypothesis that competitive interactions in MSTd play a crucial role in the robustness of human heading perception.
Our second additional test of robustness and stability uses video rather than analytic optic flow as the input signal. Flow detected from the video is inherently noisier and sparser than that that is analytically specified. Since decreases in the optic flow density do not weaken the influence of moving objects on heading judgments, this serves as an important test of model robustness.
For this set of simulations, the object approached from a 35° angle and the motion gave rise to a pseudo FoE when the object occluded the future path at the end of the trial. This differs from the stimuli used in the previous pseudo FoE simulations wherein the object crossed the future path earlier in the trial. The object was cylindrical and traveled along a ground plane, which resembled the condition from Experiment 3 of Layton & Fajen [24] when the object was laterally shifted by 0.75 m. The estimates from the differential motion model fluctuate about zero before the moving object crosses the observer’s path, and then turn sharply negative (Fig 8). The motion pooling model yields smaller biases, but also shows fluctuations throughout the trial. In contrast, mean heading error for the competitive dynamics model is close to zero until the object crosses the path, at which point a small negative error (as in human performance) emerges. These results highlight the robustness of recurrent competition that unfolds over time. Even though the density and position of the dot motion may vary considerably frame-to-frame in the optic flow detected from video (yielding changing global motion patterns), on-center/off-surround competition in the competitive dynamics model stabilizes the heading estimates.
Human heading perception is remarkably stable even when vision is temporarily interrupted, such as during eye blinks, or in the more extreme scenarios when the entire optic flow field suddenly changes, such as when large moving objects occupy most of the visual field (e.g. a train crossing a driver’s future path). To test how extreme perturbations to the optic flow field affect model heading estimates, we replaced optic flow of simulated self-motion through a static environment with full-field laminar flow for 1, 2, 5, or 10 contiguous frames (Fig 9). We did not plot the results for the motion pooling model because heading estimates were accurate, except for during the period of laminar flow when the maximal activation shifted instantaneously from the central to most laminar template. We could not simulate the differential motion model, because the laminar flow did not contain depth variation.
Fig 9 shows how full-field laminar flow perturbations affect the absolute heading error produced by the competitive dynamics model. For this set of simulations, we show individual trials (one per condition) rather than averages across multiple trials. The duration of the laminar flow perturbation had a graded effect on the maximal absolute heading error–in general, longer laminar flow perturbations yielded larger heading errors. Heading estimates restabilized in each case, but the model required longer periods of time to recover from the longer perturbations. This occurred because the recurrent mechanisms in the competitive dynamics model not only integrate the presently available optic flow, but also the response to the optic flow time history. This leads to the prediction that misperceptions in heading that may result from prolonged extreme disruptions to the optic flow field endure for a longer period of time than those caused by shorter disruptions. In general, however, the mechanisms in competitive dynamics model tolerated even the most extreme optic flow disruptions, greatly mitigating heading errors as compared to the motion pooling model.
The simulations presented thus far have focused on self-motion in the presence of moving objects. However, the stability of heading also improves over time during self-motion through static environments. This was demonstrated by Layton & Fajen [23], who showed that the variability in human heading judgments was affected by the duration of the stimulus, with the greatest variability occurring when trial duration was short (150 ms) and variability decreasing and eventually reaching a plateau at 500 ms.
Fig 10 depicts the variance of the model MSTd population across the template-space, which indicates the quality of the heading estimate, over 1 sec of simulated self-motion through a static environment. Similar to the variability in human heading judgments, the population variance is high early in the trial, indicating uncertainty in the heading estimate due to the broadness of the MSTd activity distribution. Also like human judgments, model variance decreases over time and plateaus at approximately 500 msec. These simulations suggest that even in static environments, recurrent mechanisms within MSTd, such as those in the competitive dynamics model, may play an important role in refining and stabilizing heading estimates over time.
While much research has focused on the accuracy of human heading perception, the factors that underlie its hallmark robustness have been largely ignored. Heading perception is not only accurate, but remarkably robust—a property that is illuminated by research on self-motion in the presence of independently moving objects. In the present article, we focused on the question of why heading perception isn’t biased by more than a few degrees and doesn’t abruptly shift when an object crosses the observer’s future path.
Simulations of self-motion in the presence of laterally moving objects that approached, maintained a fixed-depth from, or retreated from the observer revealed that differential motion and motion pooling models yield unstable heading estimates over time, often changing abruptly when the object approached the future path. This was not surprising, considering that the model computations process optic flow in independent instants over time. Introducing temporal smoothing in the models was not sufficient to eliminate the abrupt fluctuations. On the other hand, the competitive dynamics model presented here, which contains competitive interactions among neural units in MSTd as well as temporal dynamics, exhibited gradual changes in heading estimates over the course of several hundred milliseconds after an object occluded or disoccluded the observer’s future path. The competitive dynamics model not only captures the pattern of human heading bias observed when approaching, fixed-depth, and retreating objects near the observer’s future path, but also the temporal stability and resilience that makes human heading perception robust. Recurrent competitive interactions that unfold over time among neurons in MSTd may hold the key to the stability and robustness of human heading perception.
The primary aim of this study was not merely to introduce a new model that generates more accurate heading estimates, but rather to examine the general principles that underlie robust and stable heading perception. The most obvious difference between the competitive dynamics model and the motion pooling and differential motion models is the introduction of temporal dynamics. However, this feature of the competitive dynamics model does not merely smooth the heading estimate over time. As demonstrated above, simply adding temporal smoothing to the motion pooling and differential motion models barely reduces the fluctuations in heading estimates. Rather, the competitive dynamics model integrates a collection of neural mechanisms (i.e., divisive normalization, on-center/off-surround interactions, recurrent connectivity, and thresholding) that are fundamental, ubiquitous operations performed by populations of neurons all throughout cortex [47–50]. In this section, we explain how these mechanisms offer a more principled account of the robustness and stability of human heading perception.
The MSTd network in the competitive dynamics model implements a recurrent competitive field, wherein neural interactions lead to the divisive normalization of the population activity [51]. Divisive normalization represents an important property whereby the total energy in the network is conserved and the dynamic range automatically adjusts to process the input pattern, irrespective of gain fluctuations [52]. Automatic gain control plays a particularly important role in stabilizing network dynamics in a number of scenarios, such as in the pseudo FoE video simulation, wherein the reliability of the optic flow signal may vary considerably over time.
In model MSTd, units compete via on-center/off-surround interactions for their preferred heading direction, determined by the FoE position of their receptive field template. Each unit enhances its own heading signal through recurrent self-excitation and inhibits competing heading signals through recurrent inhibition (Eq 21). Crucially, the estimated heading arises through divisive or shunting interactions that unfold over time to balance the MSTd population activity and the continuously evolving bottom-up optic flow signals. These recurrent interactions result in soft winner-take-all dynamics within model MSTd and the simultaneous representation of multiple possible heading estimates at any point in time. In both the 70° approaching object (Fig 4B) and pseudo FoE simulations (Fig 7A), soft winner-take-all dynamics led to candidate heading estimates in multiple locations. This allowed the competitive dynamics model to balance, rather than abruptly switch between, estimates based on their salience over time. A similar mechanism allowed Layton & Browning [53] to capture the effects of spatial attention on the tuning of MSTd neurons.
While others have proposed models that depend on recurrent mechanisms in MSTd [54–56], there are important differences between these models and the competitive dynamics model. First, activity normalization and feedforward/feedback signal integration occur within the competitive dynamics model as a single process, tightly linked to the network dynamics. By contrast, the other models decompose these operations into three separate stages within each model area (e.g. MSTd). Second, the continuous integration of optic flow over time represents a distinctive property only captured by the competitive dynamics model. Computations performed by other models of MSTd with recurrent mechanisms occur at equilibrium, when units reach a steady state at each point in time. The competitive dynamics model is a dynamical system that is agnostic as to whether a steady state is ever reached: units respond to both the changes in the optic flow input and the evolving state of interacting V1, MT+ and MSTd networks. This exclusive property of the competitive dynamics model is consistent with how MT relies on a temporal solution to overcome the aperture problem [57] and how human heading perception depends on the temporal evolution of the optic flow field [23]. The fact that optic flow acceleration/deceleration played an important role in accounting for human judgments in the analytical model of Raudies & Neumann [45] strengthens the evidence that the temporal evolution of the optic flow field plays an important role in heading perception.
Returning to the question of why heading perception does not abruptly shift when an object crosses the observer’s future path, the explanation provided by the competitive dynamics model is that the competitive dynamics among MSTd neurons take time to unfold. Optic flow signals arriving at the present time interact with the present state of the network that reflects the information about the heading direction detected over the course of the recent time history. For example, when many MSTd units are simultaneously active, there is greater uncertainty about the heading direction across the population and a reliable optic flow signal may quickly influence the heading estimate (Fig 10). On the other hand, if few MSTd units are active, meaning there is a high degree of confidence in the heading estimate across the population, it may take some time for even a strong distinct optic flow signal to influence the heading estimate. This property contrasts with the behavior of other models whereby the activation of model MSTd always reflects an instantaneous transformation of the optic flow field. The time course of competitive dynamics likely interacts with the relatively slow response latencies (median: ~190 msec) of MSTd neurons to radial expansion [58,59].
The fact that the competitive dynamics model captures human heading perception so well highlights the importance of these competitive mechanisms. The motion pooling and differential motion models implement algorithms that are sufficiently generic that they could in principle be carried out on vector field representations of the optic flow field without regard to neural systems. For example, the motion differencing operations performed by the differential motion model could just as well be carried out on raw optic flow vectors without the interpretation that MT units perform the operation. On the other hand, the competitive dynamics model explicitly models individual neurons and their interactions in dynamically evolving networks—the computations and neural interpretation are inextricably linked. We emphasize that the upshot is not that the competitive dynamics model is superior to the other models, but that the competitive mechanisms that implement more realistic neural dynamics play a central role in the robustness and stability of heading perception.
To account for human heading judgments in the presence of moving objects, the differential motion model segments the optic flow field on the basis of local speed differences [25] and the model of Raudies & Neumann [45] segments on the basis of accretion/deletion, expansion/contraction, and acceleration/deceleration. This raises the question of whether the visual system requires segmentation to perceive heading in the presence of moving objects. The fact that the competitive dynamics model captures patterns of known human heading judgments without any explicit segmentation of the optic flow field, shows that, at least in principle, segmentation may not be necessary and recurrent mechanisms in MSTd are sufficient. While segmentation likely plays a fundamental role in object motion perception [60], several lines of evidence do not support a role in heading perception. First, neurons in MSTd do not appear to extract the translational component when the optic flow field contains both translation and rotation [43], which is a core prediction of differential motion models [25,39,40,44]. Second, the MT cells that possess antagonistic surrounds that are proposed by Royden to perform the differential motion computations and that could be used to extract the accretion/deletion segmentation cue used by Raudies & Neumann do not appear to project to heading sensitive cells in MSTd [61–64]. Third, sensitivity to several of the “segmentation cues” of Raudies & Neumann may be achieved without explicit segmentation stages. Radial and spiral templates that realize the properties of MSTd receptive fields extract information from optic flow about expansion/contraction and spatial curvature, respectively. In addition, the competitive dynamics model extracts information about acceleration/deceleration and temporal curvature by integrating the optic flow field over time.
Although our focus has been on visual mechanisms that underlie robust heading perception, self-motion perception is inherently multi-sensory. The majority of heading sensitive cells in primate MSTd are primarily driven by visual input but modulated by vestibular signals [65]. That is, the heading tuning curve of MSTd neurons tends to be sharper and more selective when self-motion occurs in the presence of optic flow compared to in darkness [66]. The multi-sensory tuning of MSTd neurons may contribute to the robustness of heading perception during ordinary locomotion in a manner that goes beyond the visual mechanisms explored in existing models. For example, as many as half of MSTd neurons demonstrate an enhanced response when the visual and vestibular signals are consistent with one another (“congruent cells”) and others demonstrate a diminished response when the multi-sensory signals are in conflict (“opposite cells”) [67]. The multimodal response differences between these populations of MSTd neurons may increase the robustness of heading perception by discounting optic flow, such as that produced by a moving object, that does not agree with non-visual self-motion directions [68]. Despite the coarseness of vestibular tuning in MSTd, it may be sufficient to resolve whether a region of optic flow within the visual field arises due to self- or object motion [10,69] and allow heading perception to persist when optic flow is intermittently unavailable or unreliable. The availability of proprioception during active self-motion likely provides another redundant signal to facilitate heading perception [4]. More neurophysiological and modeling work needs to be performed to clarify how multisensory signals interact to contribute to the robustness of heading perception.
In the present article, we simulated biological models of heading perception to investigate why perceived heading in humans is only biased by several degrees in the presence of moving objects and why the perceived heading does not abruptly shift when the object crosses the observer’s future path. We found that passive temporal smoothing alone was not sufficient in accounting for the characteristic robustness of human heading perception. However, recurrent competitive interactions that unfold over time among model units in area MSTd resulted in stable heading estimates.
We generated 1.5 sec (45 frame) sequences of optic flow that simulated self-motion toward two frontoparallel planes initially positioned 800 cm and 1000 cm away in depth from the observer. This environment was selected to accommodate the differential motion model, which performs best when the scene contains depth discontinuities. Each plane consisted of 3000 dots and occupied the entire visual field at the outset. The observer moved along a straight-ahead heading at 200 cm/sec.
In sequences that contained a moving object, the object was rectangular (150 cm x 150 cm) and consisted of 320 dots. Its trajectory was parameterized in terms of a starting lateral offset from the observer’s path, starting relative depth to the observer, speed, and heading-relative trajectory angle. Table 1 specifies the parameters of object trajectories.
In the video simulation, we detected optic flow in the video (320 px x 240 px) using the Horn-Schunk algorithm built into MATLAB’s computer vision toolbox, which served as the input to the differential motion and motion pooling models [70]. The competitive dynamics model detected motion from the video directly. Note that the Horn-Schunk optimization includes a regularization term in the objective function that smoothes the estimated motion field. Compared to correlational [71] or motion energy [72] approaches to motion detection, which do not require smoothing of the motion vector field, the Horn-Schunk algorithm should generate motion and heading estimates in motion pooling and differential motion models that are no less stable and accurate.
To ensure that the variability is not due to an insufficient sample size, we ran each of the models on a number (usually 25) of the optic flow sequences that were identical except for the initial random placement of the dots until the variance in the heading estimates plateaued.
We also simulated all three models in a static environment (no moving object) to confirm that they were operating as expected and that there were no systematic biases. These simulations revealed a mean heading error close to zero with low variability for all three models.
The differential motion [25] and motion pooling [20] models were implemented in MATLAB according to their published specifications. We changed several parameters to ensure that the models performed as expected on our visual displays. For example, we set the model MSTd Gaussian pooling variance parameter (σ) to 19 px in the model of Royden (2002) to ensure that heading estimates were in the direction of object motion for the fixed-depth object and in the direction opposite object motion for the approaching object. We changed the same parameter in the model of Warren & Saunders [20] to 25 px to achieve the best performance across the visual displays that we tested. Model MT in both models consisted of units that tiled the input dimensions of the visual displays. To generate the opponent operators with different differencing axes in the model MT of Royden [25], we filtered the input with 7x7 rectified Gabor kernels. MSTd units had overlapping receptive fields and were centered every two pixels along each of the input dimensions. Parameter values remained the same in all models and simulations. Analytical optic flow computed by a pinhole camera projection served as input to the differential motion and motion pooling models [38]. We derived model heading estimates by considering the location of preferred FoE of the maximally active MSTd unit along the horizontal cross-section that contained the observer’s heading. Heading error was computed by subtracting the location of the preferred FoE of the maximally active MSTd unit from that which coincides with the observer’s heading direction.
The model presented here encompasses the V1-MT+-MSTd processing stages of the Layton et al. model [28]. Updates have been performed to the front-end so that the model detects optic flow from video input using stages that correspond to those along the primate magnocellular pathway. Moreover, algorithmic simplifications used in the Layton et al. model have been replaced so that each stage consists of networks of coupled Hodgkin Huxley type ordinary dynamical equations. Our model builds on the STARS and ViSTARS models [73–75].
Fig 11 schematically depicts an overview of the model. The model consists of three main stages: detecting changes in luminance (model Retina and LGN), detecting motion (model V1 and MT+), and estimating self-motion (model MSTd). The details of these stages are described in the following sections. Fig 12 shows the response of each model area to simulated self-motion in a static environment toward two frontoparallel planes.
The ordinary differential equations described in the following sections model the dynamics of cells across multiple brain areas along the magnocellular pathway of primate cortex. Equations often assume the form of a recurrent competitive field [51]:
dxidt=−xi+(1−xi)(f(xi)+Ii+)−xi(∑k≠if(xk)+Ii−)
(1)
The firing rate x of unit i in the network layer described by Eq 1 obeys shunting dynamics, which implement a number of important dynamical properties, such as divisive normalization and boundedness [51,52,76]. Eq 1 contains a passive decay term −xi, excitatory input term (f(xi)+Ii+) that is shunted by (1 − xi) to ensure the firing rate remains bounded above by 1, and inhibitory input term (∑k≠if(xi)+Ii−) that is shunted by xi to bound the firing rate from below by 0. The variables Ii+ and Ii− denote excitatory and inhibitory inputs to unit i, respectively. In Eq 1, the function f(xi) regulates the recurrent self-excitation or inhibition that the unit receives from others in the same network. In model networks, f is a sigmoid function that gives rise of soft winner-take-all dynamics [51,77]. We use the sigmoid
f(w;f0)=w2w2+f02
(2)
that achieves its half-maximum value of 0.5 at f0.
The output of network layers may be thresholded by Γ according to the following function g
g(w;Γ)=[w−Γ]+
(3)
where the notation [∙]+ indicates the half-wave rectification max(∙,0).
Network equations apply to all units in the layer and as such we use matrix bold notation. For example, x denotes the array of cells at each spatial location (i,j). Connectivity between units in different network layers are either connected 1-to-1 or through a Gaussian kernel that defines the convergence of feedforward or feedback signals. When units are connected 1-to-1, the unit with a receptive field centered at position (i,j) projects to a unit in another layer at position (i,j). The Gaussian kernel Gσ,s defines how connections converge from one layer onto the next when the spatial extent of the receptive field is larger than that of its input.
In Eq 4, the operator ∙ defines the dot product, σ indicates the standard deviation, and s defines the radius of the kernel. Convolutions in the following sections are always centered at the position of each unit (i,j). In some cases, the Gaussian kernel that has radius s and is elongated in the direction d corresponding to the angle θ, which we define as Wσ¯,s,d.
In Eq 5, ∇ is the rotation matrix ∇=[cosθsinθ−sinθcosθ], θ=d4, Σ=[σx00σy], and σ¯=σx/σy. The kernel Wσ¯,s,d is normalized to sum to unity.
Units in areas LGN and V1 uniformly tiled 1-to-1 the spatial dimensions of the visual display. The overlapping receptive fields of MT+ and MSTd units were spaced according to their speed sensitivity. Units in MT+ tuned to speeds of 1, 2, and 3 px/frame had receptive fields uniformly distributed throughout the visual input array at every single, second, and third pixel, respectively. MSTd units had receptive fields centered every 6 px, doubling the maximum offset of MT+.
The pattern of luminance at time t in the visual signal I(t) is transformed into signals of increments J+(t) and decrements J−(t), which represent the change in the input across successive frames. These signals correspond to the coding of luminance increases and decrease by ON and OFF retinal ganglion cells.
The notation ⌈∙⌉ refers to taking the ceiling of the operand.
Model LGN units L±(t) respond to transient changes in the visual signal, but are not direction selective [78]. ON and OFF LGN units remain sensitive to the luminance increments and decrements in their retinal inputs, respectively [79]. The following equation describes the activity of LGN units:
L±(t)=[R±(t)Z±(t)]+
(8)
where R± indicates the population of units that perform a leaky integration of its retinal inputs, which is gated by the habituative transmitter Z±. Habituative gates, sometimes called dynamic or depressing synapses, prevent responses to persistent inputs by modeling the slow-term deletion of a neuron’s neurotransmitter stores [80,81]. Tonic inputs depress the habituative gates, which when multiplicatively combined with R±, lead to rapid suppression of the LGN activity L±. Habituative gates slowly recover to their full capacity in the absence of input. In effect, L± responds well to motion and weakly to stationary inputs.
In Eqs 9 and 10, ϵLGN,R corresponds to the inverse time constant of each cell R±, ϵLGN,Z corresponds to the inverse time constant of each gate Z±, and λ indicates the transmitter depletion/repletion rate. For our simulations, we fixed ϵLGN,R = 2 sec−1, ϵLGN,Z = 0.01 sec−1, and λ = 10.
The detection of motion direction occurs through a three stage process that corresponds to simple and complex cells in V1, and cells in area MT+ with excitatory surrounds (Fig 13). First, motion is detected by simple cells using a Reichardt or correlation-based mechanism based on the arrival of signals from LGN with different conduction delays and receptive field locations [71] (but see [73,74,82–84] for an alternative biological mechanism that relies on nulling inhibition). The motion signal is refined through short-range feedforward on-center/off-surround pooling of simple cell activity by complex cells (Fig 13, bottom two panels). Finally, a feedback loop between V1 complex cells and MT+ cells disambiguates local motion signals (i.e. solves the aperture problem) through the spatial pooling of complex cells by units in MT+ tuned to the same motion direction and the suppression of complex cells tuned to dissimilar motion directions (Fig 13, top two panels).
An asterisk appears adjacent to the inhibitory feedback connections from MT to V1 in Fig 13 because the model proposes that the net effect on V1 complex cells is inhibitory, not that the individual feedback projections are inhibitory. In fact, feedback projections to V1 are likely mostly excitatory and target excitatory neurons [85]. MT feedback projections target several V1 layers, including layer 6 [86], and neurons therein project to inhibitory interneurons [87] and those involved in feedforward processing [88] in layer 4C. Given that layer 4 neurons project to layer 2/3 [89], which contains complex cells, the circuit may serve a modularity role on complex cells. Even though the MT-V1 feedback projections may be excitatory, they may exert an inhibitory effect on complex cells, consistent with predictions from the competitive dynamics model. Modeling the laminar microcircuitry of V1 extends beyond the aims of the present paper, so we use inhibitory MT feedback as a simplification (see [90] for a laminar model of V1).
The feedback loop between MT and V1 proposed by Bayerl & Neumann [91] bears some similarity to the one described here, but, contrary to the predictions of the competitive dynamics model, the feedback exerts an excitatory rather than an inhibitory influence on V1 units. Feedback signals from MT that suppress complex cells sensitive to dissimilar motion directions rather than enhance complex cells sensitive to similar motion directions prevent the biologically implausible scenario of a runaway positive inter-areal feedback loop between V1 and MT and are consistent with the reduced suppression in V1 neurons following MT inactivation [92,93].
In the following sections, we describe the details of model areas V1, MT and MSTd.
Signals from model LGN with spatially displaced receptive field centers (Δ→) that possess a spectrum of conduction delays (δ→) converge onto V1 [94]. In other words, each V1 unit acquires its speed and direction tuning based on the spatiotemporal correlation present in its convergent afferent LGN signals. For simplicity, we consider the two conduction delays δ0 and δ1 that implicate motion detection across successive frames: δ0 = ⌊t⌋ and δ1 = ⌊t − 1⌋, where ⌊∙⌋ indicates taking the floor of the operand.
V1 units were tuned to speeds (s) of 1, 2, or 3 px/frame in the 8 cardinal directions (d): up, down, left, right, and the four diagonals. The speed-direction tuning was determined by summing the LGN signal centered at position (i,j) derived from the present time with the delayed signal centered at position (i − Δx,j − Δy) derived from the optic flow on the previous frame. We use the set Φ(s,d) to refer to the nonzero spatial offsets from position (i,j) that implicate motion with speed s in the direction d. For example, a V1 simple cell tuned to unit speed motion in the rightward direction would have an offset Δ→∈Φ(1,1), where Δ→=(−1,0). In other words, comparing the delayed signal displaced one unit to the left to the present signal yields sensitivity to rightward motion. To account for the increased distance along the diagonals, we scaled these signals by a factor of s. To ensure that the V1 unit receives input from the most correlated input (Eq 10), the signal is thresholded with ΓLGN = 0.45 and squared [48,72]. The set of spatial offsets |Δ→∈Φ(s,d)| grows with speed, but not all offsets factor into the output signal, so we normalized by the number of active contributions (1|As,d±>0|) (Eq 11). This adaptive normalization approximates the homeostatic plasticity process known as synaptic scaling [95].
The bottom-up input B± to V1 is computed based on the afferent LGN signals L± according to the following equations:
As,d,i,j±(t,Δ→)=[Li,j±(δ0)+Li−Δx,j−Δy±(δ1)−ΓLGN]+2
(11)
Bs,d,i,j±(t)=1|As,d±>0|∑Δ→∈Φ(s,d)As,d,i,j±(t,Δ→)
(12)
Simple cells in model V1 perform a leaky integration of their adaptive spatiotemporal inputs
dSs,d±dt=ϵS(−Ss,d±+(1−Ss,d±)Bs,d±)
(13)
where the units have an inverse time constant ϵS = 5 sec−1.
Fig 12B shows the pattern of activity of direction selective simple cells tuned to unit speed at the end of a self-motion simulation through a static environment.
Model complex cells are tuned to a direction and speed, but unlike simple cells, are insensitive to contrast polarity. Complex cell units refine and enhance their directional selectivity by locally pooling over simple cells with the same direction and speed tuning parallel to the preferred motion direction (Fig 13). The units also receive shunting inhibition from nearby simple cells whose receptive fields are positioned in the orthogonal direction. This feedforward connectivity between model simple and complex cells enhances responses to uniformly moving dots or surfaces and implements the collinear facilitation and orthogonal suppression properties of V1 complex cells [96,97].
V1 complex cell units compete across preferred motion direction in the following contrast enhancing network:
Qs,d=g(Ss,d++Ss,d−,ΓC)2
(14)
dCs,ddt=−Cs,d+(1−Cs,d)(Cs,d2+∑(I,J)(Wσ¯MT,s,θ∥d,i−I,j−JQs,d,I,J))
−Cs,d(∑k≠d(Cs,k2+Ys,k)+∑(I,J)(Wσ¯MT,s,θ⊥d,i−I,j−JQs,d,I,J)).
(15)
In Eq 15, Wσ¯MT,s,d is the anisotropic Gaussian kernel elongated in the direction d (σ¯MT=15), the notation θ ∥ d means the angle θ of the preferred motion direction, the notation θ ⊥ d means the angle θ+π2 orthogonal to the preferred motion direction, and Y defines the feedback a complex cell receives from area MT+ (see Eq 18):
Ys,d=∑(I,J)(GσV1,3s,i−I,j−JMs,d,I,J)
(16)
where σV1 = 0.5. Complex cells receive inhibitory feedback from model area MT+ that suppresses the activity of units with preferred directions k that differ from the complex cell preferred direction d. Note that each matrix Ms,d has been scaled back into the coordinate system of the visual display before performing the convolution.
Fig 12C depicts the pattern of activity of complex cells tuned to unit speed at the end of the self-motion simulation.
The complex cell output from V1 Os,d is thresholded by Γv1mt = 0.01, squared, and transformed by the sigmoid function f(∙;γv1mt), with an inflection point γv1mt = 0.01.
Units in model MT+ perform a long-range spatial pooling over complex cell signals. These cells inherit their directional tuning from V1 complex cells, but integrate motion within their larger receptive fields. Analogous to V1 complex cells, the receptive field of units in MT+ is elongated in the direction parallel to preferred motion direction [97].
Units in MT+ respond well when complex cells with a consistent directional tuning are active within the receptive field (Fig 13, top panel). Over time, the suppressive feedback loop between V1 and MT+ resolves ambiguity in the detected motion direction (Fig 13, top two panels). Fig 12D depicts the MT+ cells tuned to unit smallest scale at the end of the self-motion simulation.
The output signal Nd from MT+ collapses across speed, after scaling each matrix Ms,d into the coordinates of MSTd. Units are collinearly pooled, parallel to their preferred motion direction, thresholded with ΓMT = 10−3, and subsequently squared:
Nd=g(∑(I,J)(Wσ¯MT,2,θ∥d,i−I,j−J∑sMs,d,I,J;ΓMT))2
(19)
Fig 12E shows the output signal from MT+.
Although a full characterization of MSTd neuron receptive fields remains an ongoing challenge [48], the set of spiral templates spanned by linear combinations of radial and circular motion patterns has proven successful in capturing the pattern sensitivity across the population [98]. Analyses performed with a prior version of the competitive dynamics model revealed that linear self-motion did not activate spiral templates; such templates may play an important role in perceiving self-motion when traveling along a curved path [77]. Even in the presence of objects moving along a linear trajectory, the optic flow does not contain the spatial or temporal curvature required to activate spiral templates [45]. Because the range of conditions considered in the present study focus on observer and object motion along linear trajectories, we restricted the model MSTd template space to radial patterns.
Our model MSTd does not include cells with sensitivity to laminar or “planar” flow [14], but the role of such cells should be investigated in future research. Whereas objects that approach the observer along nearly parallel trajectories generate radial motion patterns within the object contours (Fig 1A), fixed-depth objects (e.g. Fig 1B) and objects that approach with large path angles generate more laminar flow patterns. Through potential interactions with radial cells, activation of cells in MSTd tuned to laminar flow by these moving objects may stabilize heading signals and reduce bias. In this regard, cells tuned to laminar flow may participate in a redundant mechanism that complements or factors into the recurrent competition used in the competitive dynamics model. Future physiological and modeling work should clarify the role of cells tuned to laminar flow for heading perception, particularly in the presence of moving objects.
Model MSTd matches an array of templates T± tuned to radial expansion (+) and contraction (-) with the bottom-up signal from model MT+ Nd. The templates Ti,j± have FoE/FoC positions centered on every position (i,j) within the spatial grid tiling of MSTd. The specification of expansive and contractive templates has been described previously [77]. In the following equation, the match is normalized by the energy of each template:
Vi,j±=1∑(I,J)(Ti,j,I,J±)∑d∑(I,J)(Ti,j,I,J±Nd,I,J)
(20)
MSTd cells compete in a soft winner-take-all network across the polarity of radial motion (expansion versus contraction) and over 2D space: Each cell has a firing threshold ΓMST = 0.3 and sigmoid inflection point fMST = 0.001. The kernel GσMST,7 specifies how units with neighboring singularity selectivities compete with one another, with σMST = 10. The dynamics of the MSTd cell tuned to radial expansion or contraction with FoE/FoC selectivity at position (i,j) is defined as follows:
dPi,j±dt=−Pi,j±+(1−Pi,j±)(f(g(Pi,j±;ΓMST);fMST)+Vi,j±)
−Pi,j±(∑w≠±∑(n,m)≠(i,j)GσMST,7,n,mf(g(Pn,mw;ΓMST);fMST))
(21)
Fig 12F shows the activity of expansion-sensitive MSTd cells at two different points in time. The heading estimate h* from model MSTd is determined by considering the FoE selectivity of the most active cell tuned to expansion along the horizontal cross-section x that contained the heading direction. While other expansion-selection units were often active, the most active unit had a centrally positioned FoE selectivity due to the vertical radial symmetry of the optic flow displays simulated.
|
10.1371/journal.pcbi.1005157 | Fused Regression for Multi-source Gene Regulatory Network Inference | Understanding gene regulatory networks is critical to understanding cellular differentiation and response to external stimuli. Methods for global network inference have been developed and applied to a variety of species. Most approaches consider the problem of network inference independently in each species, despite evidence that gene regulation can be conserved even in distantly related species. Further, network inference is often confined to single data-types (single platforms) and single cell types. We introduce a method for multi-source network inference that allows simultaneous estimation of gene regulatory networks in multiple species or biological processes through the introduction of priors based on known gene relationships such as orthology incorporated using fused regression. This approach improves network inference performance even when orthology mapping and conservation are incomplete. We refine this method by presenting an algorithm that extracts the true conserved subnetwork from a larger set of potentially conserved interactions and demonstrate the utility of our method in cross species network inference. Last, we demonstrate our method’s utility in learning from data collected on different experimental platforms.
| Gene regulatory networks describing related biological processes are thought to share conserved interaction structure. This assumption motivates a great deal of work in model systems–where discovery of gene regulation may be more experimentally tractable–but is difficult to directly evaluate using existing methods. The presence of shared structure in a well studied model system or process should make the problem of network inference in a related process easier, but this information is not often applied to the discovery of global gene regulatory networks. Further, to be able to successfully translate findings between different organisms, it is important to be able to identify where regulatory structure is different. We provide a method based on penalized fused regression for inferring gene regulatory networks given prior knowledge about the similarity of interactions in each network. This method is demonstrated on synthetic data, and applied to the problem of inferring networks in distantly related bacterial organisms. We then introduce an extension of the method to deal with the condition of uncertainty over the degree of regulatory conservation by simultaneously inferring gene conservation and interaction weights.
| As the volume and variety of genome scale data continues to increase in quantity and quality, the goal of accurately modeling gene regulatory networks has become attainable [1–3]. Large-scale data collection efforts have contributed to the development of high quality networks which accurately represent biological processes, but most processes and organisms remain uncharacterized at the network level. Furthermore, as new technologies are developed and some old ones are replaced, such as RNAseq and microarray, it becomes important to be able to combine data from multiple platforms, lest we lose valuable information from existing studies. The problem of inferring related—but not necessarily identical—structure from related—but not identical—data is ubiquitous in biology. Multi-source network inference has applications for learning multiple networks in related species, for learning networks associated with distinct processes within the same species, and for learning networks based on heterogeneous data sources. Moreover, as it becomes possible to learn genome-wide regulatory networks, we can begin to compare and to test whether there is conservation of networks across species and biological processes. Our use of model organisms to study biological processes and diseases relevant to humans relies on the assumption of conservation; yet this has not been effectively tested at the genome scale.
We present two methods for network inference based on linear estimates of gene expression dynamics, extending existing dynamical-systems methods for network inference [1, 4, 5]. The core of both methods is the observation that biological information about the relatedness of genes can be used to select which network coefficients should be similar to one another in a multi-source network inference problem (ie orthologous TFs should regulate orthologous genes), and that these constraints can be efficiently represented as penalties in a least-squares regression problem.
Numerous studies have shown that functional conservation exists in gene regulatory networks even across large evolutionary distance [6–9]. Our first method—fused L2—takes advantage of this similarity by imposing an L2 penalty on the differences between a priori similar interactions (termed fusion penalty). These constraints favor network configurations in which orthologous genes have similar regulators. Because network inference problems are typically under-constrained, these additional constraints allow data in one species to improve network inference performance in another.
Existing multi species approaches often use orthology as a proxy for functional conservation [10–14], or attempt to learn functional similarity via expression data [15]. Orthology can be approximated using readily identifiable sequence similarity, which is often a useful predictor of functional similarity [16, 17]. However, many genes will have evolved different functions and therefore may have new regulatory interactions. For example, gene duplications may lead to neofunctionalization [18] of the duplicated genes. Or, when comparing regulation across cell lines, changes in chromatin configuration may affect our hypotheses about the similarity of interactions between pleiotropic TFs and target genes across cell types (a within-species analog to neo-functionalization) [19].
Identifying interactions that are present in one species but not another is of direct biological interest, but existing approaches to network inference are unable to effectively test the hypothesis of conserved subnetworks. Observing a large difference in the weights of regulatory interactions obtained though independent inference of multiple networks is perhaps the best (least biased) evidence against conservation of orthologous regulatory interactions (cases where target and regulator have orthologs across species). However, this is sometimes weak evidence, as network inference is typically under-constrained [20], meaning there could be a different set of networks for which conservation does hold, and which fit the data almost as well. Solving the networks jointly with fusion addresses this problem, but may be biased when gene function is not conserved.
Our second method—adaptive fusion—attempts to solve the problem of identifying evolutionary divergence using a non-concave saturating fusion penalty to simultaneously infer the constrained networks and to learn which constraints should be relaxed (ie which parts of the network are genuinely different). This penalty is based on statistical techniques intended to provide unbiased regularization penalties for regularized regression [14, 21]. We extend these techniques fused regression, and provide an algorithm that approximates the solution to the resulting non-convex loss function.
We develop two algorithms for solving efficiently multi-output least-squares regression problems with pairwise L2 fusion penalties on entries of the coefficient matrix, and discuss conditions under which each is suitable. We also introduce—in the form of adaptive fusion—the idea of a saturating penalty function on fusion constraints, and estimate the solution to the resulting optimization problem through iterative application of the fused L2 algorithm. We start with a discussion of the fused L2 and adaptive fusion algorithms, and describe their performance on synthetic datasets intended to represent related gene regulatory networks. We then demonstrate the applications of the fused L2 network inference algorithm on biological data, first demonstrating gains in cross-species network inference using the bacteria species Bacillus subtilis and the distantly related Bacillus anthracis, then moving on to multi-platform network inference using B. subtilis, and finally intra-species fused network inference using priors based on operons in B. subtilis. We then discuss adaptive fusion and show the technique’s ability to identify incorrect orthology information which has been introduced to a biological dataset, suggesting the technique’s applicability to discovering neo- and sub-functionalizations.
We used gene-expression data from Bacillus subtilis and B. anthracis in order to assess performance gains of fused regression on real data. Our B subtilis data set consists of 360 time-series and steady-state observations of 4891 genes, 4100 of which are protein coding [25], during the life cycle. Our B. anthracis dataset consists of 72 time-series and steady-state observations of 5536 genes comprising data from distinct points in the life cycle and iron-starvation conditions. There were 247 known transcription factors (TFs) in the B. subtilis dataset, and 248 TFs in the B. anthracis dataset. We obtained 1,870 one-to-one orthologs from Inparanoid [26], 95 of which are transcription factors, which produced 177,650 fusion-constraints between gene interactions within the two species. This number represents only 14.7% of the regulatory interaction matrix in B. subtilis and 12.9% in B. anthracis.
To assess network inference performance, and for use as priors, we used a gold standard of 3,040 known B. subtilis interactions with corresponding activation and repression sign. Of these 3,040 priors, 968 had corresponding interactions in B. anthracis. Based on our simulation results, we can expect the greatest gains in network-inference performance from fusion when the species of interest has a small number of available conditions, but data is abundant in a related species. However, in order to evaluate performance objectively a gold-standard of known interactions is necessary. As a result, we can only evaluate network recovery for B. subtilis, and B. subtilis also has the majority of our conditions. In order to simulate the data-poor regime, we subsampled our B. subtilis data. We divided our B. subtilis data into k folds, and then for each fold fit a network to the B. subtilis data from that fold alone fused to the entire 72 B. anthracis conditions (Fig 4a). Though overall performance is hindered by our subsampling of B. subtilis data (a necessary procedure to allow evaluation of networks) we demonstrate marked improvement in learning the B. subtilis network when using fused regression (Fig 4a). Notably, these performance gains occur mostly at low values of recall. That is: the highest confidence part of the network is inferred more accurately, with minimal gains for interactions which are more uncertain. Because these interactions are likely to be the focus of followup experiments and validation, gains here are more valuable for prioritizing the order in which interactions are investigated.
We tested a combination of our fused regression approach with a method for estimating transcription factor activities (TFA). Rather than modeling gene expression using transcription factor mRNA abundance, we fit gene expression as a function of transcription factor activity, as applied to B. subtilis by Arrieta-Ortiz et al [4]. TFA activity estimates transcription factor activities that are modulated through mechanisms such as dimerization and interaction with required factors. TFA activity estimates have been shown prior to be better predictors of TF function than expression level alone in several contexts including similar network inference tasks [34] [4]. We estimate TFA based on known regulatory interactions using network component analysis [35]. To test the integration of this approach with our fused regression, we assessed the combination of B. subtilis datasets, as in Fig 5a, with the incorporation of TFA estimation. We randomly divided the prior known interactions in half, and used half to learn TFA and to generate priors on network structure. The remaining interactions were reserved as a gold standard for validation. As in previous studies, we observed a marked improvement in network inference when using transcription factor activity (Fig 6). We also obtained AUPR improvement when using fused regression alongside TFA; gains from sharing information across datasets using fused regression were preserved and even enhanced by using TFA.
Although our approach is generalizable to a wide variety of multi-source network inference problems, we begin with the concrete example of network inference in two related species. Our approach to multi-species network inference is based on the hypothesis that gene regulation in related species is governed by similar but not necessarily identical gene regulatory networks, due to conservation of function through evolution. We represent conservation of network function by introducing constraints into the objective function for network inference that penalize differences between the weights of regulatory interactions believed to be conserved. These constraints favor the generation of similar networks for related species, and in the generally under-constrained regime of network inference can improve the accuracy of network recovery. We then go on to introduce a method to test the assumption of conserved network structure, and to relax the associated constraints on pairs of interactions for which the data does not support conservation. Finally, we demonstrate the flexibility of the method by using fusion constraints based on operon membership to improve network inference performance.
Gene expression data, such as microarray or RNA seq, provide information about the relationship between genes by allowing an experimenter to measure correlations in expression value over time or across conditions. Many sources of information—such as the knowledge that two genes are related through orthology or belong to the same operon—provide additional information about the relationships between these gene-gene relationships. For example, two genes that belong to the same operon are likely to have a similar set of regulators [32], but knowing that two genes are members of a polycistronic transcript does little to inform the identity (strength, sign) of those regulators. Meta-information about the structure of gene regulatory networks, specifically which pairs of interactions are a priori likely to be similar to one another, can provide a powerful set of constraints to improve network inference performance [10, 55]. We present a general framework for gene regulatory network inference that incorporates this meta-information—termed fusion constraints—and apply the technique to the problem of simultaneous inference of regulatory networks in multiple species (B. subtilis and B. anthracis), as well as to the problems of combining data from multiple experimental platforms and information about operon structure.
A number of existing approaches have applied fused regression to related problems in network inference. TreeGL applied fused LASSO to the problem of estimating partial correlations of gene expression in a breast cancer dataset consisting of multiple cell types [52]. They imposed fusion between the coefficients of models learned for different cell types whenever there was an edge between those cell types in a genealogy graph describing the cancer cells’ development. A similar approach was recently applied to the problem of inferring TF to gene regulatory weights from data sources describing multiple environmental conditions in E. Coli, Mycobacterium tuberculosis, and Mus musculus [53]. In this formulation, networks associated with each data source were fit simultaneously, with an L1 constraint on the similarity of (arbitrarily ordered) adjacent networks. In both of these approaches, fusion was restricted to be between corresponding entries of the coefficient matrices. In terms of our formulation in the cross-species case, this is equivalent to the constraints that would be generated by a one-to-one orthology mapping between species, limiting to the (typical) case where orthology is partial or not one-to-one. By optimizing a more general objective function—in which fusion constraints can be placed on arbitrary pairs of coefficients associated with otherwise unrelated regression tasks—we can extend this work to the cross-species case.
B. subtilis and B. anthracis are distantly related bacterial species with limited gene orthology. Nevertheless we show that network recovery in B. subtilis can be improved through the inclusion of expression data from B. anthracis. Many previous methods for cross-species network inference operate on the conserved subset of orthologous genes [56]. This assumption may be appropriate with very closely related species, but could not be applied in this domain, where a large fraction (62% and 67%) of the B. subtilis and B. anthracis genomes do not have clear orthologs. Our method, in contrast, can obtain improvements in network inference performance even when the conserved subset of genes is small.
This approach is particularly interesting in light of the diversity of important model organisms used in modern biology. Different model systems provide different advantages and disadvantages for experimental design [57], but in many cases work in less used systems is hampered by lack of available data. Our intent is to provide a principled method for combining data from diverse sources, so that results in specialized systems can be integrated with data from well studied organism.
Although it is important to take advantage of the similarities of related organisms for generating improved models of gene regulation, it is also critically important to understand how systems differ from one another. Existing approaches to the genome-wide testing of the assumption that orthologous genes have similar regulators learn regulatory networks separately, then compare to identify conservation [58]. Because network inference is typically under-constrained, fitting a network that describes a particular set of experimental observations amounts to sampling a single network from a large set of networks that fit the data equally (or almost equally) as well. As a result, the existence of a difference between corresponding regulatory interactions in a pair of experimentally derived networks is weak evidence that a difference truly does exist. Uncoupled global network inference algorithms are a very weak tool for uncovering evolutionary divergence. Our method explicitly favors recovering networks for which evolutionarily corresponding interactions are similar. As a result, the failure to obtain networks that confirm evolutionary conservation is much more direct evidence that the networks have truly diverged.
Although fused regression allows more accurate identification of pairs of non-conserved interactions, the weights obtained for these interactions will be biased towards one another. We have described a method—adaptive fusion—that attempts to address this bias by simultaneously learning the networks and constraint weights. This method is based on minimizing a saturating penalty function on fusion constraints, similar to a class of penalties that have been developed to minimize bias in regularized regression [14, 21]. The result of adaptive fusion is both a network and a new set of fusion constraints, the weights of which can be interpreted as describing conservation structure across the networks. For the multiple species case, relaxation of fusion constraints represents orthologs which do not share similar interactions presumably due to evolution of regulatory circuitry [59]. When jointly learning networks describing processes in different cell lines, this may identify interesting context-specific behavior. Genes may be fused together on the basis of similar binding sites or chromatin features, and the relaxing of the fusion penalty indicates divergence of gene function.
Because our model shares its basic assumptions about the role of transcription factors in gene expression dynamics with models developed for single-species network inference, we are able to leverage techniques developed for the single-species estimation of transcription factor activity [34]. The performance gains of this additional step in the cross-species case are significant. Our approaches—fused L2 and adaptive fusion—represent a very general framework for simultaneous network inference and the incorporation of structured biological priors. These priors—incorporated into our method as fusion constraints—allow the use of rich sources of biological knowledge, such as orthology and operon structure, which have informed experimental design, but are typically not incorporated into genome wide network inference algorithms. By accommodating the simultaneous inference of multiple related networks, we can improve network inference performance by allowing the efficient reuse of data from similar, but not necessarily identical, sources. A method for pooling data from multiple sources holds the promise of vastly expanding the quantity of data available for analysis, particularly in less commonly used model systems. At the same time these methods allow us to test our assumptions on how similar biological systems relate to one another, by allowing us to rule out conservation in a principled way, and at the genome-wide scale.
|
10.1371/journal.ppat.1006524 | Herpesviruses shape tumour microenvironment through exosomal transfer of viral microRNAs | Metabolic changes within the cell and its niche affect cell fate and are involved in many diseases and disorders including cancer and viral infections. Kaposi’s sarcoma-associated herpesvirus (KSHV) is the etiological agent of Kaposi’s sarcoma (KS). KSHV latently infected cells express only a subset of viral genes, mainly located within the latency-associated region, among them 12 microRNAs. Notably, these miRNAs are responsible for inducing the Warburg effect in infected cells. Here we identify a novel mechanism enabling KSHV to manipulate the metabolic nature of the tumour microenvironment. We demonstrate that KSHV infected cells specifically transfer the virus-encoded microRNAs to surrounding cells via exosomes. This flow of genetic information results in a metabolic shift toward aerobic glycolysis in the surrounding non-infected cells. Importantly, this exosome-mediated metabolic reprogramming of neighbouring cells supports the growth of infected cells, thereby contributing to viral fitness. Finally, our data show that this miRNA transfer-based regulation of cell metabolism is a general mechanism used by other herpesviruses, such as EBV, as well as for the transfer of non-viral onco-miRs. This exosome-based crosstalk provides viruses with a mechanism for non-infectious transfer of genetic material without production of new viral particles, which might expose them to the immune system. We suggest that viruses and cancer cells use this mechanism to shape a specific metabolic niche that will contribute to their fitness.
| The metabolic state within a cell and its local environment is altered in many diseases and disorders including those caused by viral infections. The gamma-herpesviruses Kaposi’s Sarcoma Associated Herpesvirus (KSHV) is a viral agent associated with a large number of human malignancies. KSHV was shown to manipulate the metabolism of host cells and to induce similar metabolic changes to those found in non-viral cancers. Our work demonstrates that KSHV not only regulates host cell metabolism, but also alters the metabolism of neighbouring non-infected cells. We report that exosomes secreted from KSHV-infected cells selectively transfer the viral miRNAs into neighbouring cells. While these cells remain uninfected, the viral miRNAs are active in them to induce the Warburg effect. Moreover, our data show that this exosomal transfer of miRNAs transforms the exosome-recipient cells into ‘feeder cells’ producing a microenvironment that is more supportive for host cell growth. Finally, we found that this flow of genetic information increases the angiogenic potential of non-infected cells and therefore, could further enrich the infected cell microenvironment to support their fitness. This previously undescribed mechanism provides important insight into cancer and viral pathology and suggests new avenues for therapeutic intervention.
| Altered metabolism is regarded as a hallmark of cancer. It is thought that cancer cells rewire metabolic pathways in such a way that biosynthetic processes are balanced against ATP production to support high rates of proliferation [1]. One of the most characteristic metabolic hallmarks of tumour metabolism is aerobic glycolysis. Despite the inefficiency of glycolysis in energy production, the glycolytic phenotype provides cancer cells with several advantages such as increased biosynthesis of intermediate macromolecules and anti-apoptosis and signalling through metabolites [2]. Recently, it has been suggested that cancer cells, in addition to their intrinsic metabolic alteration can also induce aerobic glycolysis in adjacent stromal cells, a phenomenon termed the ‘reverse Warburg Effect’ [3, 4]. The reverse Warburg effect emphasises the importance of tumour stromal cells in supplying energy metabolites and chemical building blocks to the rapidly proliferating cancer cells.
Oncogenic viruses cause more than 15% of human cancers and it is predicted that effective treatment against them will lead to 25% fewer cancers in developing countries and 7% in developed countries [5]. Pathogenicity of these viruses involves the hijacking of host cellular pathways, including those controlling cell metabolism, suggesting they can use as a model system to study cancer development.
Kaposi’s sarcoma herpesvirus (KSHV) is the etiological agent of Kaposi’s sarcoma (KS) and certain lymphoid neoplasms. KS is the most common neoplasm in HIV-1-infected individuals and also induces significant morbidity in other immunosuppressed individuals (e.g., post organ transplantation) and in populations where KSHV infection is endemic (para-Mediterranean regions) [6]. To date KSHV infection does not have an effective treatment and new therapeutic approaches are needed. KS cells, as other cancers, have a distinct metabolism and KSHV was shown to alter several metabolic pathways in its host cell [7–10]. We have recently shown that the KSHV-encoded microRNAs (miRNAs) induce aerobic glycolysis in infected cells through regulation of key cellular genes involved in mitochondrial activity and regulation of glucose metabolism [11]. Interestingly, interaction between KSHV infected cells and their microenvironment was shown to be important for primary effusion lymphoma growth in vivo[12]. It was recently shown that the KSHV miRNAs are present in exosomes isolated from KS patient and KS mouse model [13] but the role of this in KS biology is still unknown. We suggest that exosomes present a new possible platform allowing KSHV in infected cells to interact with their microenvironment and improve its host cell fitness.
Exosomes are spherical structures sealed with membranes released from the endosomal compartment of most cell types. They vary in size and molecular composition, depending on the cell of origin [14]. The functional impact of exosomes is imparted by the molecular components (protein and RNA cargo) they carry [15]. Exosome uptake is thought potentially to modulate many physiological and pathological processes including cell growth, immune regulation, angiogenesis and metastasis [16, 17]. Exosomes uptake was also shown to affect metabolism; exosomes secreted from cancer associated fibroblast were suggested to modulate cancer cells metabolism by transferring miRNAs and metabolites [18], and breast cancer secreted exosomes were shown to reduce glucose metabolism in cells in the pre-metastatic niche by transferring miR-122 [19]. Interestingly, it has been suggested that viruses, including KSHV, modulate the secretion of exosomes from infected cells [13, 20–22].
Here we have used KSHV infection as a model to study whether cells utilise the exosomal pathway to regulate the metabolism of their microenvironment. We show that KSHV-infected primary lymphatic endothelial cells secrete exosomes containing viral-encoded miRNAs. These exosomes transfer the miRNAs to surrounding uninfected cells to induce a reverse Warburg effect. We show that these miRNAs function in the recipient cells to regulate known target genes and this results in reduced mitochondria biogenesis and induction of aerobic glycolysis. Most importantly, we found that this metabolic cross-talk between infected and neighbouring non-infected cells has physiological implications in KSHV life cycle supporting the growth of latent infected cells. Finally, our results show that regulation of microenvironment metabolism by miRNA transfer might be a global mechanism used by other viruses and cancer cells.
Taken together our results reveal a novel mechanism whereby virally infected cells and cancer cells regulate the metabolism of surrounding cells via miRNA transfer. This allows these cells to control their microenvironment without producing new viral particles, which might trigger host immune recognition.
KSHV miRNAs regulate mitochondria and glucose metabolism in infected cells [11]. In addition, it has been demonstrated that patient- and mouse model-derived exosomes carry in them the KSHV miRNAs [13]. We hypothesised, therefore, that KSHV modulates exosomes secretion to regulate the metabolism of neighbouring cells.
We test our hypothesis in lymphatic endothelial cells (LEC), the precursor cells for KS. We infected primary LEC with BAC16-derived WT KSHV (KLEC) or miRNA cluster-deleted KSHV (ΔmiR-KLEC) [23]. Cells were selected to produce a 100% infected population and cultured for 6 days to allow establishment of latency (Fig 1A and S1A Fig). Post selection, both KLEC and ΔmiR-KLEC were found to have similar KSHV genome copy number (S1B Fig) and expression levels of other latent genes (S1C Fig). Growth media was then collected every 48 hours for a period of 14 days after each infection and extracellular vesicles were purified by standard differential centrifugation from 150ml of media (S1D Fig). To determine if these nano-vesicles are exosomes we performed imaging, particle sizing, and measured biochemical properties of purified vesicles released from infected and non-infected LEC. Size and population characterisation using NanoSight Nanoparticle Tracking Analysis technology measured particles with diameter distribution between 50-150nm, corresponding to the expected size of exosomes (S1E Fig). This analysis also showed that 150ml of growth media contains between 2-3x1011 particles/ml (S1F Fig). Purity of the exosome preparations was determined by both electron microscopy (Fig 1B) and immunoblot analysis for the known exosome markers CD63, CD9 and ALIX (Fig 1C and S1G Fig).
Because our hypothesis is a non-cell autonomous regulation on cell metabolism, we confirmed that our purified exosomes are not contaminated with KSHV particles. Immunoblot analysis showed no detectable KSHV envelope-associated protein ORF8 in the purified exosome fraction (Fig 1D, bottom panel, 3 left lanes) and no viral particles were identified by electron microscopy. As expected, CD63 can be detected also in the KSHV sample (Fig 1D, top panel, last lane) since it is collected from the media of activated cells using high speed centrifugation. Finally, after incubation with purified exosomes, we could not detect the KSHV DNA in uninfected cells, and these cells did not become GFP positive nor kanamycin resistant, as would be expected had they become infected with the recombinant virus. Taken together, we concluded that the purified exosome fraction is KSHV free and therefore any effect this fraction has, is not due to de novo KSHV infection.
To test whether the KSHV miRNAs are present in secreted exosomes, RNA was extracted from LEC-, KLEC- and ΔmiR-KLEC-derived exosomes. To ascertain that the RNA is protected within the vesicles, exosomes were treated with 0.4μg/μl RNase for 10 min at 37°C prior to RNA extraction [24]. qRT-PCR analysis using the KSHV-miR LNA PCR primer sets (Exiqon), showed all 12 KSHV encoded miRNAs are present in KLEC derived exosomes (S1H Fig). Importantly, using these primer sets we could not detect any signal for RNA extracted from exosomes derived from LEC. Similarly, we could only detect miR-K12-10 and miR-K12-12, which are encoded out of the miR-cluster, in exosomes derived from ΔmiR-KLEC (S1I Fig). To better quantify the viral miRNAs in these exosomes as well as the effect of KSHV infection on transfer of cellular miRNAs in exosomes, we sequenced the small RNAs from LEC and KLEC as well as from exosomes secreted from these cells. We detected around 1800 different miRNAs in LEC and KLEC and around 1200 miRNAs in exosomes secreted from these cells. Importantly, we found that while in KLEC, the viral miRNAs present around 5% of the total miRNAs reads, in exosomes secreted from them, these miRNAs are responsible for around 10% of the total miRNAs reads (Fig 1E). In addition, we found differences in the expression profile of the viral miRNAs between the cells and exosomes. For example, KSHV miR-K12-10a-3p, K12-4-3p and K12-8-3p are over represented in the exosomes while KSHV miR-K12-11-3p and K12-4-5p are underrepresented (Fig 1F and S1 Table). Moreover, specific cellular miRNAs that are over represented in KLEC compared to LEC such as miR-145-5p and 143-3p (S2 Table), were not found to be enriched in KLEC-derived exosomes (S3 Table). On the other hand, we identified cellular miRNAs which are enriched in KLEC-derived exosomes compared to LEC-derived exosomes although these are not enriched in the respective cells (S2 and S3 Tables). For example, hsa-miR-216a is highly enriched in KLEC derived exosomes. miR-216a was suggested to function as an oncomiR and to induce epithelial-mesenchymal transition (EMT) by targeting PTEN and SMAD7 [25]. This suggests that transfer of human miRNAs in exosomes secreted from KSHV infected cells may play an additional role in KSHV pathogenicity. Taken together these results suggest that KSHV manipulates its host cells secretion system to selectively enrich the packaging of the viral miRNAs together with specific cellular miRNAs in exosomes secreted from infected cells.
Having established the presence of KSHV miRNAs in exosomes we sought to explore their trafficking and more importantly, biological function. We initially tested whether exosomes secreted from KLEC are taken up by non-infected LEC, by labelling internal protein in purified exosomes using the Exo-Green (System Bioscience). Untreated LEC were first labelled using CellMask Deep Red Plasma Membrane Stain (molecular probes), then incubated with labelled exosomes and analysed using confocal microscopy. As shown in Fig 2B, upon incubation with labelled exosomes, GFP signal can be detected within the cells, indicating uptake of these exosomes. Flow cytometry analysis of these cells showed that positive staining is still detectable 24 hours’ post uptake by target cells (S2A Fig). Importantly, uptake of KLEC-derived exosomes results in transfer of the viral miRNAs to these cells (Fig 2C).
miRNAs function by binding to their target genes to inhibit their translation and induce their mRNA degradation. Therefore, we next tested if the KSHV miRNAs are active in the cytosol of recipient cells to inhibit the expression of known target genes. We have previously shown that the KSHV miRNAs function as cluster to regulate EGLN2 and HSPA9 and induce aerobic glycolysis [11]. We inserted the 3’ UTRs of these genes downstream of a luciferase coding sequence to generate a reporter of KSHV miRNA activity. Cells expressing this reporter were incubated with exosomes, secreted from LEC or KLEC for 48 hours (1x109 particles). Incubation with KLEC-derived exosomes resulted in a ~25–30% reduction in luciferase activity relative to cells incubated with control LEC-secreted exosomes (Fig 2D). To further confirm the activity of the KSHV miRNAs in receptive cells we specifically tested the activity of miR-K12-10, which is expressed separately from the other miRNAs, using blue fluorescent protein (BFP) fused to 8 repeats of the miRNA target site. When these cells were incubated with KLEC-secreted exosomes, we found a ~25% reduction in BFP intensity compared to incubation with exosomes secreted from LEC (Fig 2E). To further test if these KLEC-derived exosomes regulate the expression of these genes under more physiological exosomes-transfer conditions we used Transwell plates [26] to co-culture non-infected LEC (bottom compartment) with either ΔmiR-KLEC or KLEC (upper compartment). We then analysed the RNA from the non-infected LEC and found 30% and 40% down regulation of EGLN2 and HSPA9 mRNAs levels respectively (Fig 2F). Taken together these results show that the KSHV encoded miRNAs are transferred to non-infected cells and maintain their ability to down-regulate specific target genes.
We have previously shown that the KSHV miRNAs induce aerobic glycolysis in infected cells. We therefore predicted that they would have a similar effect in exosome recipient cells. To test this hypothesis, we educated non-infected LEC by either growing them in the presence of isolated exosomes for 48 hours (Fig 3A) or by co-culturing them with infected cells using Transwell plates (Fig 3B). We first measured the oxygen consumption rate of educated LEC using the Seahorse XF24 Analyzer. The Seahorse Extracellular Flux Analyzer determines oxygen consumption rate (OCR) and extracellular acidification rate (ECAR), in order to assess cellular functions such as oxidative phosphorylation and glycolysis. We educated non-infected cells with increasing numbers of exosomes and found a dose dependent reduction in oxygen consumption of educated cells with maximal effect using 2.5x109 exosomes (S3A Fig). We therefore decided to use this number of exosomes for cell education for all future experiments.
Education using KLEC-derived exosomes reduced the baseline oxygen consumption in recipient cells by ~30% compared to education using LEC-derived exosomes (Fig 3C and S3A Fig). Importantly, exosomes derived from ΔmiR-KLEC did not affect oxygen consumption (Fig 3C), suggesting the viral miRNAs are the driving force behind this phenotype. Similarly, after co-culturing KLEC and LEC for 5 days, LEC were found to have reduced oxygen consumption compared to uneducated LEC (S3B Fig). This suggests that transfer of the KSHV miRNAs into non-infected cells reduces mitochondrial respiration. Similarly, we also observed a 30% increase in glucose uptake, consistent with increased aerobic glycolysis (Fig 3D). Mitochondria are key players in normal glucose metabolism during aerobic conditions and as part of the Warburg effect, many cancer types show altered mitochondrial activity [27]. Since we have previously shown that expression of the KSHV-encoded miRNAs reduces mitochondria biogenesis [11], we tested whether KLEC-derived exosomes have a similar effect on mitochondrial volume by loading cells with MitoTracker together with Calcein AM. MitoTracker is a fluorescent dye that labels mitochondria within live cells utilising the mitochondrial membrane potential. It therefore allowed us to calculate mitochondrial volume (MitoTracker staining) relative to total cell volume (Calcein staining) (Fig 3E). Upon incubation with KLEC-derived exosomes we found a ~40% decrease in mitochondria volume in educated cells compared to cells incubated with LEC derived exosomes (Fig 3F). Importantly, we did not detect any significant difference between cells educated using LEC- and ΔmiR-KLEC-derived exosomes, strengthening our notion that the KSHV encoded miRNAs are responsible for this phenotype. The hypoxia-induced factor alpha (HIF1α) is a known regulator of glucose metabolism [28, 29] and can mediate the Warburg effect in cancer cells [30]. KSHV has been shown to activate HIF1α and HIF2α during latency [31] and we have observed that expression of KSHV miRNAs induces HIF1α stabilisation [11]. As shown in Fig 3G, HIF1α expression was increased in cells educated using KLEC-derived exosomes compared to those educated using LEC- or ΔmiR-KLEC-derived exosomes. To further characterise the metabolic effect induced by exosomes secreted from KSHV infected cells, we performed targeted quantitative analysis using capillary electrophoresis mass spectrometry (CE-MS). This analysis showed significant increase in lactate and pyruvate as well as decrease levels of TCA cycle metabolites and ATP, in cells educated by KLEC derived exosomes (Fig 3H and 3I and S3C Fig), supporting our notion that these exosomes reduce mitochondrial activity in educated cells.
Importantly, when educated cells were grown for additional 5 days without exosomes, the metabolic phenotype was reversed and their oxygen consumption was comparable to that of untreated LEC (S3D Fig). This supports our notion that these cells are not infected by KSHV and that this phenotype depends on constant transfer of miRNAs from infected cells.
Finally, we tested whether exosomes secreted from KLEC have the same metabolic effect on other cell types relevant to KS. Education of Human Umbilical Vein Endothelial Cells (HUVEC), with exosomes extracted from KLEC induced aerobic glycolysis, as shown by reduced oxygen consumption and mitochondria volume (S3E and S3F Fig).
Taken together these results suggest that transfer of the KSHV miRNAs via exosomes induces aerobic glycolysis, reduces mitochondria biogenesis and leads to HIF1α stabilisation in surrounding non-infected cells.
Many viruses other than KSHV express miRNAs, and we speculated that miRNA transfer via exosomes might be a general mechanism used by viruses to regulate their microenvironment. Epstein Barr Virus (EBV) is a human gamma herpes virus which, like KSHV, is the etiological agent for several lymphoid malignancies [32]. EBV encodes at least 40 miRNAs, which were shown to be present in exosomes secreted from EBV transform cells [21, 33]. The EBV encoded miRNAs are thought to have many target genes in common with KSHV [34–36] and EBV-miR-BART1 has been suggested to regulate metabolism-associated genes [37]. To test whether exosomes secreted from EBV infected cells have similar effect to those secreted from KSHV infected cells, we collected exosomes from the growth media of EBV positive and negative AKATA cell lines (S4A and S4B Fig). We have found that EBV-encoded miRNAs were present in only exosomes secreted by the EBV positive AKATA cells (Fig 4A). Educating human fibroblasts using these exosomes for 48 hours resulted in a 25% decrease in oxygen consumption (Fig 4B), stabilisation of HIF1α (Fig 4C and 4D) and expression of its target gene VEGFA (Fig 4E). This suggests EBV can also use exosomes to alter its microenvironment metabolism in a similar way to KSHV.
Regulating energy metabolism using miRNAs is not exclusive to viruses, and many cellular miRNAs are also known to control energy metabolism [38]. miR-210, for example, is known to regulate cell metabolism, is associated with mitochondrial defects and glycolytic phenotype [39–41], and was suggested to be secreted in exosomes under hypoxic conditions [42]. To test if miR-210 can be transferred in exosomes to alter the metabolism of cells within the microenvironment we forced the expression of miR-210 in HEK293T and HCT116 cell lines. Exosomes secreted from these cells had much higher levels of miR-210 compared to exosomes derived from cells infected with a control vector (S5A Fig). miR-210 directly targets the iron-sulphur assembly proteins ISCU1/2 [43]. To test if miR-210 can be transferred in exosomes and be active in recipient cells, we tested ISCU1 mRNA levels in human fibroblasts, educated using exosomes secreted from control or miR-210 over expressing cells. Education using exosomes contain high levels of miR-210 leads to a ~50% decrease of ISCU1 mRNA levels (S5B Fig). Importantly, these educated cells also reduced their oxygen consumption by 30–40% (S5C Fig).
Taken together these results suggest that transfer of miRNAs via exosomes is a general mechanism that can be used by cells in a variety of pathological contexts to regulate the metabolism of cells in their microenvironment.
How might metabolic transformation of the microenvironment enhance the fitness of KSHV? One possibility is that it sensitises surrounding cells to viral infection. To test this, we infected different educated cells with KSHV (Fig 5A). Contrary to our expectations we observed a 50% decrease in infection of cells educated using KLEC-derived exosomes compared to those educated using LEC exosomes (Fig 5B). Similarly, DNA analysis of cell educated using KLEC and ΔmiR-KLEC exosomes showed a similar decrease in the viral copy number in cells educated using KLEC derived exosomes (Fig 5C). To test whether this unexpected effect is due to the metabolic changes induced by these exosomes, we mimicked the exosome metabolic effect by expressing the HIF1α P402A/P564A stable mutant [44] in LEC (Fig 5D). Over-expression of HIF1α had a similar effect to exosome treatment, leading to a 40% reduction in the viral copy number 48 hours’ post KSHV infection (Fig 5E). Thus, it appears that aerobic glycolysis inhibits KSHV infection, and that exosomes secreted from infected cells in fact prevent KSHV spreading into new cells.
The fact that uptake of exosomes secreted from LEC reduces viral spreading suggests other benefit for the virus. It has been suggested that the reverse Warburg effect in stromal cells supports growth of cancer cells [3, 4]. We therefore hypothesised that inducing aerobic glycolysis in nearby non-infected cells supports the growth of KSHV-infected cells. To test this hypothesis, we co-cultured KLEC in Transwell plates with HUVEC educated with different exosomes (Fig 5F). We found that growing in the presence of cell per-educated with KLEC derived exosomes, promoted KLEC growth by ~40% compared to growing in the presence of cells educated with exosomes from ΔmiR-KLEC (Fig 5G). HUVEC over-expressing stable mutant HIF1α had a similar effect on KLEC growth (Fig 5H), supporting the notion that this increased growth is indeed due to the reverse Warburg effect.
In cancer models, the reverse Warburg effect is thought to support cancer cell growth by promoting the secretion of energy-rich metabolites such as lactate and pyruvate from non-cancerous neighbours. These metabolites are taken up by the cancer cells and used in the mitochondrial TCA cycle, thereby promoting efficient energy production and higher proliferative capacity [3]. Since we have found that uptake of exosomes secreted from KLEC leads to increased levels of lactate and pyruvate in educated cells (Fig 3H and 3I), we tested whether lactate supports KLEC growth. We found that supplementing the growth medium with Lactate promotes KLEC growth, though to a lesser extent than with HUVEC co-culture (Fig 5I), suggesting that lactate responsible for part of this phenotype. Consistent with this, we found that KSHV infection leads to over expression of the monocarboxylate transporters MCT1 and 2 (S6 Fig), supporting our notion that these cells uptake high energy molecules such as lactate and pyruvate to support their growth.
Taken together, these results suggest a metabolic feedback where KSHV infected cells induce aerobic glycolysis in cells in their microenvironment, and those as a result secrete high energy metabolites that support the KSHV infected cells.
Kaposi's sarcoma (KS) is a highly-vascularised tumour supporting large amounts of neo-angiogenesis. It has been proposed that KSHV directly induces angiogenesis in a paracrine fashion [45]. Consistent with this KSHV infection of endothelial cells in culture induces a number of host pathways involved in activation of angiogenesis and a number of KSHV genes themselves can induce pathways involved in angiogenesis.
We have previously shown that expression of the KSHV miRNAs leads to stabilisation of HIF1α in infected cells [11]. Here we found that exosomal transfer of KSHV miRNAs leads to similar affect also in non-infected cells (Fig 3G). Since HIF1α is as a master regulator of angiogenesis[46], we hypothesis that KSHV uses exosomes to induce angiogenesis also in non-infected cells. To test this, we determined the angiogenic ability of non-infected LEC using an endothelial cell tube-formation assay. As shown in Fig 6A and 6B, LEC educated by KLEC-derived exosomes have greater angiogenic potential compared to LEC educated by LEC or ΔmiR-KLEC derived exosomes. This suggests that viruses can use exosomes secretion as a mechanism to enrich their growth environment by increasing the angiogenic potential of non-infected endothelial cells.
It was previously shown that exosomes collected from patient blood or KS models can induce migration [13]. In order to directly test if this phenotype is due to transfer of the viral miRNAs we educated HUVEC using exosomes from LEC, KLEC and ΔmiR-KLEC, and tested their migration capability using wound assay. We found that while HUVEC educated using LEC or ΔmiR-KLEC exosomes migrate similarly, HUVEC educated using KLEC derived exosomes migrate faster (Fig 6C and 6D).
Taken together these results suggest that KSHV uses exosomes to induce angiogenesis and migration of non-infected cells around its host cell, and present a potential mechanism allowing the virus to enrich its microenvironment.
Viruses have long served as tools in molecular and cellular biology to study a variety of complex processes. In this study, we reveal a novel mechanism by which oncogenic herpesviruses can regulate the nature of their microenvironment, which has implications for cancer cell biology.
We have found that KSHV not only regulates its host cell metabolism, but also alters the metabolism of neighbouring non-infected cells. We report that exosomes secreted from latently infected primary LEC selectively transfer the viral miRNAs into neighbouring cells. While these cells remain uninfected, the viral miRNAs are active in them and down regulate expression of their target genes. This results in a metabolic shift toward aerobic glycolysis and reduced mitochondria biogenesis. Moreover, our data show that this exosomal transfer of miRNAs transforms the exosome-recipient cells into ‘feeder cells’ producing a microenvironment that is more supportive for host cell growth. We suggest this is due to secretion of high energy molecules, such as lactate and pyruvate, that can be used by the infected cells (Fig 7). Moreover, we found that this flow of genetic information increases the angiogenic potential of non-infected cells and therefore, could further enrich the infected cells microenvironment to support their fitness.
Our data suggest that KSHV uses a miRNAs-based mechanism to manipulate the metabolism of cells in its microenvironment. This is based on: (i) our previous observation that expression of the KSHV miRNAs is sufficient to induce aerobic glycolysis, (ii) our observation that the KSHV miRNAs are active in recipient cells to regulate the same metabolic target genes as in KSHV infected cells and (iii) the fact that exosomes secreted from ΔmiR-KLEC, which do not contain the viral miRNAs, do not have the same effect. Nevertheless, while our data show that the KSHV miRNAs are the driving force behind this metabolic phenotype, the possibility that other components in these exosomes might support it, still needs to be further investigated.
To some extent it is not unexpected that KSHV miRNAs will be present in exosomes secreted from latent cells, since these miRNAs are highly expressed in them. Indeed, it was shown that these miRNAs are present in exosomes collected from plasma of KS patient or KS mouse models [13]. While identification of viral miRNAs in the blood might be useful for diagnostic purposes, it is hard to appreciate the biological advantage for the virus by transferring these miRNAs in the blood stream. Here we show for the first time that the KSHV miRNAs are selectively enriched in exosomes secreted from infected cells. Critically we show the physiological functionality of this exosomal transfer in shaping the metabolic feature of the infected cells microenvironment and the advantage of this local transfer for virally infected cells. Although exosomes secreted from LEC and KLEC contain both viral and cellular miRNAs, our results support the notion that the KSHV miRNAs are the driving force behind this metabolic transformation since it is not induced by exosomes secreted from cell infected with a mutant virus, lacking the miRNAs cluster (ΔmiR-KLEC). Our results also show that although the KSHV miRNAs are highly expressed in infected cells, much lower levels are sufficient to regulate their target genes and to induce aerobic glycolysis in surrounding cells. These results are consistent with other studies showing that exosomes can transfer miRNAs in sufficient levels to regulate their target genes in the recipient cells [19, 47–49]. We therefore suggest that infected cells express the viral miRNAs in much higher levels than those required to regulate their target genes, to ensure their inclusion in the secreted exosomes.
Our results suggest that regulation of cell metabolism by miRNAs transfer is not unique for KSHV but may present a more general mechanism used by other viruses and cancer cells. EBV, a close relative of KSHV, encodes at least 40 miRNAs, many of which regulate the same genes and pathways as the KSHV miRNAs [35]. We show here that EBV can transfer these miRNAs in exosomes and these can also affect mitochondrial respiration in exosome recipient cells. It was previously shown that exosomes secreted from EBV infected cells also transfer proteins that might be involved in altering the metabolism of recipient cells [22]. While our results do not rule out this possibility, the fact that EBV and KSHV are suggested to share many of their miRNAs target genes supports our model that viral miRNAs transfer is the driving force behind this metabolic regulation.
Herpesviruses account for most of the viral encoded miRNAs, though other DNA and even RNA viruses also encode miRNAs [50]. miRNAs are likely to be invisible to the adaptive immune response. Therefore, transferring miRNAs via exosomes would be advantageous during persistent infection since it allows viruses to recruit cells in their vicinity without producing and releasing new viral particles, a process that requires energy and that exposes them to the immune system. Moreover, many cellular miRNAs are also known to be involved in regulation of energy metabolism [38]. miR-210 is highly expressed under hypoxic conditions to alter cellular processes including cell cycle regulation, mitochondria function, apoptosis and angiogenesis [51]. Hypoxia can arise because of oxygen diffusion limitation in avascular primary tumours or due to abnormal tumour microvascularization. For these reasons, these cells might also have poor nutrient supply. Our results suggest the miR-210, which is over-expressed under these conditions, can be transferred by exosomes into cells in the tumour microenvironment. Inducing the Warburg effect in their microenvironment by transferring miR-210 into normoxic non-cancer cells can support the growth of the hypoxic cancer cells by supplying high-energy molecules such as lactate and pyruvate. Our results with miR-210, raise the possibility that other onco-miRs can be transferred into the microenvironment to support the tumour growth.
Our results suggest that altering their host metabolism by miRNA transfer is a novel mechanism used by oncogenic viruses to influence their hosts. One outcome of this, is growth support for the virus host cell. However, cell metabolism has also been shown to be involved in regulation of other cellular processes that are relevant for KS development and prolong infection. Cell metabolism and specifically oxidative metabolism and glycolysis were also shown to regulate both innate and adaptive immune systems [52]. Moreover, it was recently shown that glucose consumption by tumours metabolically restricts T cells, thereby allowing tumour progression [53]. Our results suggest that viruses might use exosomes to create a hypoglycaemic microenvironment that similarly suppress the immune response against infected cells. Thus, our results raise the possibility that viruses use exosomes to shape the metabolism of their microenvironment during persistent infection, as a mechanism to evade the immune system.
It was previously shown that exosomes secreted from breast cancer cells can inhibit glycolysis in the pre-metastatic niche [19] and that exosomes secreted from cancer associated fibroblast can induce aerobic glycolysis in nearby cancer cells [18]. Here we show the opposite effect, whereby uptake of exosomes from herpes viruses-infected cells induce aerobic glycolysis in surrounding normal cells. This is the first-time exosomes are shown to induce the reverse Warburg effect and presents a new mechanism by which cancer cells can recruit cells in their microenvironment to support their growth.
KSHV miRNAs were shown to regulate many other cellular processes such as cytokine responses, immune recognition, cell survival, transcriptional reprogramming and angiogenesis [54–56]. Therefore, transfer of the viral miRNAs via exosomes may alter a wide spectrum of processes, adding to the complexity of the relationship between viruses and their hosts. The cross talk between KS cells and their microenvironment warrants further study, with a view to identifying novel therapeutic targets.
We present a novel mechanism allowing viruses to regulate the metabolism and migration of cells in their vicinity in a way that supports their fitness by transferring their genetic material via exosomes. We suggest that viruses and cancer cells use this mechanism to shape a specific metabolic niche that will favour their proliferation. It also implies that, similar to cancer cells, latently infected cells depend on their environments for sustained growth. Targeting this miRNA transfer-based metabolic cross-talk between diseased cells and their microenvironment could therefore open a new therapeutic window.
HUVEC and LEC were purchased from Promocell and grown in endothelial growth medium 2 and MV2 (Promocell) respectively. Both cell types were used for experiments before passage 8. To exclude exosomes derived from the FBS, it was subjected to centrifugation of 120,000g for 3 hours. iSLK producing cells were kindly provided by Rolf Renne (University of Florida). KSHV producing iSLK cells were cultured in DMEM (Invitrogen), supplemented with 10% FBS, 1μg/ml Puromycin, 250μg/ml Geneticin and 1200μg/ml Hygromycin. HEK293T (ATCC) and HCT116 (ATCC) cells were cultured in DMEM (Invitrogen), supplemented with 10% FBS. EBV producer line Akata (kindly provided by Paul Farrell, Imperial College) and EBV negative Akata (kindly provided by Andrew Bell, University of Birmingham) were cultured in RPMI 1640 (Invitrogen) supplemented with 10% FCS.
Wild type and ΔmiR-cluster KSHV were prepared from iSLK cells as previously described [23]. Early passage LEC were infected and selected using 50ug/ml Hygromycin B (Invitrogen). Cells were tested for 100% infection (GFP positive) before carrying on any experiment. Repeats for each experiment were performed using different KSHV infections.
Uninfected LEC and LEC infected with WT or ΔmiR-cluster were grown up to approximately 80% confluency. Infected cells were cultured for at least 6 days before collecting media to verify latency establishment, and were grown up to passage 8. Medium was replaced every 48, and kept at 4°C for up to 7 days. The medium was subjected to centrifugation of 300g for 5 minutes, 2000g for 10 minutes and concentrated with a Centricon Plus-70 filter (Millipore) according to the manufacturer’s instructions. The media was then subjected to centrifugation of 10,000g for 1 hour, and was filtered using 0.22μM filters. To purify exosomes, the sample was subjected to ultracentrifugation at 120,000xg for 1 hour and washed once with PBS. Exosomes pellet was resuspended in PBS, quantified for protein concentration and particle number, and stored at -80°C.
The exosome suspensions in PBS were incubated on formvar- and 2 nm carbon-coated copper grids overnight at 4°C in a humidified chamber. They were then washed twice in ddH20 by dipping onto the surface of a water droplet and then stained with 2% aqueous uranyl acetate for 2.5 minutes. The stain was drawn off with cartridge paper to leave a thin negative stain. The sections were examined in a Jeol 1010 microscope. Images were taken with a Gatan Orius SC100B charge-coupled device camera and analysed with Gatan Digital Micrograph.
Characterisation of vesicle size distribution and concentration was performed using Nanoparticle Tracking Analysis (NTA) (Malvern Instruments, Nanosight NS300). Sample size distributions were calibrated in a liquid suspension by the analysis of Brownian motion via light scattering. Nanosight provides single particle size and concentration measurements.
Total RNA was extracted from exosomes or from their respective parental cells (3 biological replicates for each condition) using the miRNeasy mini-kit (Qiagen) and was quantified using Qubit RNA HS assay kit. 200ng or 500ng of RNA from exosomes or cells respectively, were used for small RNA libraries using the NEBNext Multiplex Small RNA Library Prep Set for Illumina (NEB) according to the manufacturer’s instructions. Constructed libraries were assessed using the BioAnalyzer 2100 (Agilent) and KAPA Library Quantification Kit (KAPA Biosystems). Quantified libraries were then sequenced using the single-end sequencing protocol at 36bp.
Expression of human and KSHV miRNA reads were quantified using PaTMaN [57] rapid aligner to the miRbase release 21 database of miRNA sequences [http://mirbase.org] allowing for 1 mismatch (or counting only perfect matches if more than one miRNA was matched). Normalisation and differential expression were performed using the DESeq2 method [58], whilst multiple testing correction employed independent hypothesis weighting (IHW) for greater statistical power [59].
For education experiments 105 non-infected cells were incubated for total of 48 hours when 2.5x109 exosomes were added every 24 hours. For biological replicates of all experiments, we used exosomes from separate KSHV infections. For Transwell co-culture experiments, indicated cells were grown on both compartments for total of 5 days and media was replaced every other day. For exosome-tracking experiments, purified exosomes were fluorescently labelled using Carboxyfluorescein succinimidyl diacetate ester (System Biosciences). Labelled exosomes were washed in 20 ml of PBS, collected by ultracentrifugation, and resuspended in PBS.
Cells were seeded in XF 24-well cell culture microplates (Seahorse Bioscience) at 4 × 104 cells/well (0.32 cm2) in 200μl growth medium and then incubated at 37°C/5% CO2 for 20–24 hours. Assays were initiated by removing the growth medium from each well and replacing it with 600μl of assay medium pre-warmed to 37°C. The cells were incubated at 37°C for 30 minutes to allow media temperature and pH to reach equilibrium before the first-rate measurement. Prior to each rate measurement, the XF24 Analyzer gently mixed the assay media in each well for 3 min to allow the oxygen partial pressure to reach equilibrium. Following mixing, OCR and ECAR were measured simultaneously for 4 min to establish a baseline rate. The assay medium was then gently mixed again for 3 min between each rate measurement to restore normal oxygen tension and pH in the microenvironment surrounding the cells. Uncoupled, maximal and non-mitochondrial respiration was determined after the addition of 5 μM oligomycin, 1 μM carbonyl cyanide 4-(trifluoromethoxy)phenylhydrazone (FCCP) and 2 μM antimycin-A. All chemicals were from Sigma-Aldrich.
Glucose uptake was measured by incubating cells with 30μM glucose analogue 6-NBDG (Invitrogen) for 15 minutes. Cells where then washed and trypsinized and their fluorescence (λex: 465 nm, λem: 540nm) was measured by flow cytometry.
Cells were loaded with 5μM Calcein-AM and 50nM MitoTracker Red (Invitrogen; 37 °C, 30 minutes) in growth media for 30 minutes and Z-series of images were acquired using a Zeiss LSM 510 system (Carl Zeiss, Inc., Cambridge, UK), as previously described [11]. Maximal projection of images was used to quantify the area of green (Calcein) and red (mitoTracker Red) signal. Mitochondrial area was defined relative to cytoplasmic area as ‘area red/area green’. Images were analyzed using the MetaMorph Microscopy Automation & Image Analysis Software (Molecular Devices). The two channels (Calcein-AM and MitoTracker Red) were separated and threshold in order to acquire two separate binary images. To ensure reproducibility, for each biological repeat, we analysed images of at least 10 fields, when each field contain between 5–10 cells.
Metabolites were extracted according to the Human Metabolome Technologies, Inc. protocol. Shortly, 2x106 cells were washed twice using 5% mannitol solution and metabolites were extracted by adding methanol for 30 second. The extracted solution was centrifuged 2300g at 4°C for 5 minutes, filtered and evaporated using centrifugal evaporator. Analysing the ionic metabolites including in these cells by CE-TOFMS and CE-QqQMS was performed by Human Metabolome Technologies, Inc. The results in the manuscript present 3 biological repeats, where cells were educated using exosomes from different KSHV infections.
Cells were lysed in RIPA buffer (300mM Sodium Chloride, 1% NP-40, 0.5% Sodium deoxycholate, 0.1% Sodium dodecyl sulphate and 50mM Tris pH 8.0). Exosomes were lysed directly into 1x Laemmli sample buffer (Bio-Rad Laboratories). Equal amounts of protein were resolved on Mini-PROTEAN TGX Precast gels (Bio-Rad Laboratories). Antibodies against CD63 (Invitrogen), α-TUBULIN (Sigma-Aldrich), HIF1α (BD Transduction Laboratories), LAMIN A/C (Santa Cruz Biotechnology), EBV-GP125 (GeneTex) and KSHV-ORF8 (ThermoFisher Scientific) were detected with IRDye secondary antibodies (LI-COR) or HRP-conjugated secondary antibodies. Images were acquired using the Li-COR Odyssey Imaging System or the GE Healthcare Imagequant LAS 4000.
Genomic DNA for qPCR was extracted using the QIAamp DNA mini-kit (Qiagen). Cell total RNA was extracted using either the RNeasy mini-kit or the miRNeasy mini-kit (Qiagen). Exosomal RNA was extracted using the miRNeasy mini-kit (Qiagen). KSHV genome copy numbers were quantified by qPCR as previously described [11].
cDNA synthesis for qRT–PCR quantification of mature miRNAs was performed using the Exiqon Universal cDNA Synthesis Kit II according to the manufacturer's instructions. Detection of the mature KSHV miRNAs was performed using the KSHV-miR LNA PCR primer sets (Exiqon). Where indicated cellular small nucleolar RNA RNU66 or S5 rRNA were used as a reference RNA. Importantly, the KSHV LNA PCR primer sets do not give any background detection for negative control (such as non-infected cells or ΔmiR-KLEC).
The reporter plasmids for the 3’UTRs of the indicated genes were previously described [11]. Cells expressing each of the reporter plasmids were incubated for 48 hours with the indicated exosomes for 48 hours. Cells were harvested according to the Dual-Luciferase Reporter assay system (Promega). Luciferase activity was measured using a Fluoroskan Ascent FL luminometer (ThermoScientific). Firefly activity was normalised to internal Renilla luciferase levels.
Growth factor–reduced Matrigel (Becton Dickinson) was placed in 96-well tissue culture plates (75μl/well) and allowed to gel at 37°C for 30 minutes. Then 1x104 LEC, pre-incubated with exosomes derived from LEC, KLEC or ΔmiR-KLEC, were added to each well and incubated at 37°C for 24 hours. Morphological changes were visualised using phase contrast microscope.
Educated HUVEC were seeded in 12 well plates and grown overnight to confluence. A 200μl tip was used to make a straight scratch and plates were immediately placed in Nikon Biostation CT or Zeiss Cell Observer. Images were acquired for 16 hours and analysed for scratch area using ImageJ.
miR-210 was cloned using the Gateway Cloning protocol (Invitrogen). Shortly, the mature microRNA was amplified using the primers:
The PCR product subcloned into the gateway entry vector pENTR/pTER+ [60]. The miRNA was further cloned into the 3rd gen lentiviral promoter-less Gateway destination vector pLenti X1 Puro DEST using the Gateway LR Clonase II enzyme mix (Invitrogen).
|
10.1371/journal.pcbi.1005356 | Single-molecule protein identification by sub-nanopore sensors | Recent advances in top-down mass spectrometry enabled identification of intact proteins, but this technology still faces challenges. For example, top-down mass spectrometry suffers from a lack of sensitivity since the ion counts for a single fragmentation event are often low. In contrast, nanopore technology is exquisitely sensitive to single intact molecules, but it has only been successfully applied to DNA sequencing, so far. Here, we explore the potential of sub-nanopores for single-molecule protein identification (SMPI) and describe an algorithm for identification of the electrical current blockade signal (nanospectrum) resulting from the translocation of a denaturated, linearly charged protein through a sub-nanopore. The analysis of identification p-values suggests that the current technology is already sufficient for matching nanospectra against small protein databases, e.g., protein identification in bacterial proteomes.
| Protein identification is the key step in many proteomics studies. Currently, the most popular technique for intact protein analysis is top-down mass spectrometry which recently enabled high-throughput identification of many proteins and their proteoforms. However, this approach requires large amounts of materials and is currently limited to short proteins, typically less than 30 kDa. On the other hand, nanopore sensors promise single molecule sensitivity in protein analysis, but an approach for the identification of a single protein from its blockade current (nanospectrum) has remained elusive, since the signal from the sensors relates to the amino acid sequence of the protein in a poorly understood way. In this work we describe the first algorithm for protein identification based on nanospectra associated with translocation of proteins through pores with sub-nanometer diameters. While identification accuracy currently does not allow reliable processing of complex protein samples yet, we believe, that the rapidly improving experimental protocols along with the new computational algorithms will transform into a viable protein identification approach in the near future.
| When Church et al. [1] proposed to use nanopores for sequencing biopolymers, they had envisioned both DNA and proteins sequencing. However, the progress in protein sequencing turned out to be much slower since it is more difficult to force proteins through a pore systematically and measure the resulting signal [2]. These difficulties underlay the experimental and computational challenges of Single Molecule Protein Identification (SMPI).
Nanopores promise single molecule sensitivity in the analysis of proteins, but an approach for the identification of a single protein from its nanospectrum has remained elusive. The most common approach to nanopore sequencing relies on the detection of the ionic –current blockade signal (nanospectrum) that develops when a molecule is driven through the pore by an electric field. Preliminary work [3, 4] was limited to analyzing protein conformations in pure solutions rather than identifying proteins in a mixture. Subsequent steps demonstrated that nanopores can detect protein phosphorylations [5] as well as conformations and protein-ligand interactions [6]. Recent studies on combining nanopores with aptamers have shown limited success for protein analysis [7]. Proposals for electrolytic cell with tandem nanopores and for single molecule protein sequencing have been made, but not yet implemented [8–11].
Recently, the sequence of amino acids in a denatured protein were read with limited resolution using a sub-nanometer-diameter pore, sputtered through a thin silicon nitride membrane [12]. Protein translocations through the pore modulated the measured ionic current, which was correlated with the volumes of amino adids in the proteins. However, the correlation was imperfect, making it difficult to solve the problem of reconstructing a protein from its nanospectrum with high fidelity.
Developing computational and experimental methods for analyzing nanospectra derived from a electrical signals that produced when a protein translocates through a sub-nanopore could enable a real-time sensitive approach to SMPI that may have advantages over top-down mass spectrometry for protein identification. Despite difficulty and expense (requiring especially powerful magnets) to implement it, top-down mass spectrometry has been used in a few labs around the world to identify intact proteins and their proteoforms. However, it is about 100-fold less sensitive than bottom-up mass spectrometry, which can be used to detect attomoles of material [13]. In stark contrast, a sub-nanopore has been used to discriminate residue substitutions in a single molecule with low fidelity [12].
Similar to mass-spectrometry, where de novo protein sequencing (based on top-down spectra) remains error-prone [14, 15], the challenge of de novo deconvoluting nanospectra into amino acids sequences of proteins is currently unsolved. However, protein identification based on top-down spectra (i.e., matching a spectrum against all proteins in a protein database) is a well-studied topic. For example, top-down protein identification tools ProsightPC [16] and MS-Align+ [17] reliably identify proteins, report p-values of resulting Protein-Spectrum Matches (PrSMs), and even contribute to improving gene annotations by discovering previously unknown proteins [18].
In this paper, we describe the first algorithm for protein identification based on nanospectra derived from current blockades associated with denaturated, charge linearized translocation of protein through pores with sub-nanometer diameters. Our Nano-Align algorithm matches nanospectra against a protein database, identifies Protein-Nanospectrum Matches (PrNMs), and reports their p-values. Our analysis revealed that the typical p-values of identified PrNMs vary from 10−4 to 10−6, which is already sufficient for a limited analysis of nanospectra against small bacterial proteomes.
The software is publicly available at http://github.com/fenderglass/Nano-Align.
The details regarding the experiments and methods used to acquire electrical current blockade signals from the translocation of single protein molecules through sub-nanopores have been described elsewhere [12]. To summarize, first, a pore with sub-nanometer cross-section was sputtered through thin silicon nitride membrane supported on a silicon chip using a tightly focused, high-energy electron beam in a scanning transmission electron microscope (Fig 1). The thickness of the membranes ranged from 8 to 12nm. Then the silicon chip supporting the membrane was embedded in a multiport microfluidic device that allowed for independent electrical access to the cis and trans-sides of the sub-nanopore by two Ag/AgCl electrodes. To perform electrical measurements, the sub-nanopore was immersed in 0.2 − 0.3 M NaCl and a transmembrane voltage in the range between 300 − 700 mV was applied. The resulting pore current was measured using an Axopatch 200B amplifier controlled with Clampex 10.2 software. Finally, recombinant denatured protein, along with 2 ⋅ 10 − 3% sodium dodecyl sulfate that imparted a nearly uniform negative charge to the protein, were added to the microfluidic reservoir (c.a. 20 fmoles of protein) and subsequently blockades in the open pore current associated with single molecules translocating through the pore were observed. It was determined that a lower transmembrane bias voltage improved the signal-to-noise ratio (SNR) and lengthened the median duration of the blockades, but it also increased the propensity for the pore to clog. Multi-level events associated with residual native protein structure or multiple molecules competing for the pore were occasionally observed, but were manually culled from the data pre- analysis [12].
Five proteins were analyzed by measuring the blockade currents through sub-nanopores: a recombinant chemokine CCL5 of length 68 AAs; two variants of the H3 histone designated as H3.2 and H3.3, which consist of the chain of 136 AAs, differing only by residue substitutions at positions 32, 88, 90 and 91; a tail peptide of the H3 histone (residues 1-20) and a fourth histone, H4 of length 103 AAs. More details about the datasets are given at the ‘Datasets’ section below.
When a single molecule of protein translocates through the sub-nanopore, its amino acids block the flow of ions, causing a change in the open pore current Iopen. The fraction of occupied pore volume Vmol/Vpore (where Vpore and Vmol are volumes of the pore and molecule inside this pore, respectively) was assumed to be proportional to the fractional blockade current, which is calculated as |I − Iopen|/Iopen, where I is the raw current during the translocation. The raw signal measurements from the pore were pre-processed as follows: first, the discretized pore signal, sampled at 250 kHz, was split into the separate blockades, each one representing a translocation of a single protein (Fig 2)
Only events with sufficient duration to detect single-AA duration features were selected. Typical blockade duration analyzed here ranged from 1 to 20 milliseconds, as shorter times did not permit accurate discrimination of intra-event features due to the measurement bandwidth. The mean fractional blockade current varied from 0.05 to 0.5 for different nanospectra. Recorded signals exhibited fluctuations that were associated with different structural features of a protein translocating through the pore.
Since the electrolytic current through the pore is associated with the occupied pore volume, one of the major factors that influences the signal is the volume of amino acids that occupy the sub-nanopore near the waist [19]. The estimates of amino acid volumes were obtained from crystallography data [20]. Since the pore can simultaneously accommodate multiple amino acids, it was assumed that the fluctuations in a blockade were proportional to a linear combination of amino acids volumes in the pore waist. In particular, we found that the mean volume of amino acids yielded a good approximation of the empirical signal values. Thus, given a protein P of length |P|, we split it into overlapping windows of size k (or k-mers) and generate a theoretical nanospectrum MV(P) as a vector of dimension |P| + k − 1 by taking the average volume of |P| − k + 1 k-mers and extra 2 * (k − 1) shorter prefix and suffix substrings from the beginning and end of a protein. These extra prefix and suffix substrings correspond to the start and the end of a translocation, when the pore is occupied by less than k amino acids. For example, for k = 3, the “protein” KLMNP results in a vector of length seven corresponding to the following substrings: K, KL, KLM, LMN, MNP, NP, and P.
Experimental analysis of peptides with post-translational modifications and mutations [12] revealed changes in the specific regions of the recorded signal traces, that corresponded to approximately four amino acids in length. In addition, simulations of the electric field in a 0.5x0.5 nm2 diameter, 8 nm thick pore in an SiN membrane indicated that the vast majority of the field was confined within 1.5 nm of the pore near the waist at the center of the membrane, which gives roughly the same estimate of the number of amino acids. Thus, the Mean Volume (MV) model assumes that each fluctuation in the blockade current corresponds to a read of a quadromer (short prefixes and suffixes of a protein correspond to shorter mers), which results in the best fit (among all reasonable values of k) with experimental nanospectra.
Generally, the MV model results in theoretical nanospectra correlated with the empirical data. The mean Pearson product-moment correlation coefficient between a consensus of experimental nanospectra (an average of multiple protein translocations, as described below) and the corresponding MV model was ranging from 0.25 to 0.45 for various datasets. However some regions show large deviations between theoretical and experimental nanospectra, which may be associated with additional attributes such as hydrophilicity or charge. In particular, our analysis revealed that such discordant regions were enriched with small amino acids, which have volumes below the median value (see Fig 3) for illustration and ‘Characterizing errors of the models’ section below for the detailed discussion). Since we acquired multiple nanospectra originating from multiple known proteins, an alternative approach for generating theoretical nanospectra was to use a supervised learning paradigm. We used a Support Vector Regression (an SVM-based regressor) to establish the correspondence between a k-mer inside the pore and a signal it generates [21]. Given an empirical nanospectrum E recorded from a protein P, we tiled P into overlapping quadromers qi and discretized E into |P| + 3 points. Thus, each qi had an associated experimental signal value ei.
Next, the feature space of the model has to be defined. Following the ideas of the MV model, it is natural to assume that blockade current is affected by the composition of amino acids in a quadromer, rather than their order (however, the dependence might be non-linear). As many of the 20 proteinogenic amino acids have similar volumes, we partitioned them into four volume groups (Fig 4) and defined a feature vector fi of a quadromer qi as the composition of amino acids from each group (as a tuple of length four). For example, an amino acid quardromer GQLD has zero amino acids from Large group (> 0.2nm3), two from Intermediate group (between 0.15 and 0.2 nm3), one from Small group (between 0.11 and 0.15nm3) and one from Minuscule group (< 0.11nm3), and is converted to a feature vector (0, 2, 1, 1). This choice of the feature space reduced the overfitting effect and increased coverage of the training dataset (there are only 35 distinct quadromer compositions in the defined feature space versus 204 = 160 000 amino acid quadromers).
Using a set of pairs (fi, ei) we trained an SVR regressor with the Radial Basis Function kernel (implemented in an open-source library libsvm [22]). The Support Vector Regression (SVR) model takes a peptide P as input and outputs an SVR-based theoretical nanospectrum SVR(P) (Fig 3). The mean Pearson correlation coefficient between the theoretical and empirical nanospectra (consensus) for the SVRmodel was varying from 0.38 to 0.68 for different datasets, confirming the improvement over the MV model. The parameters of the SVR model were chosen through cross validation experiments and are equal to C = 1000, γ = 0.001, and ϵ = 0.01.
The analysis of error patterns of the SVR model revealed a bias in the signal estimation that was correlated with the hydrophilicity of the amino acids (see ‘Characterizing errors of the models’ section). Also, Bhattacharya et al. [23] recently reported that water molecules affect the signal of DNA translocating through the nanopore since hydrophilic amino acids are more likely to acquire a water molecule and change the effective volume [24]. Thus, it is desirable to include amino acid hydrophilicity into the model.
Motivated by these finding, we explored an alternative approach for supervised learning by using the Random Forest (RF) regression [25, 26] for theoretical nanospectra generation. In comparison to the SVR model, the resulting Random Forest (RF) model is more robust to outliers and exhibit less overfitting [27], which allowed us to use the volumes of all 20 amino acids as features. According to this RF model, each quadromer qi from the training set is converted to a feature vector fi, where each element of the vector is a pair of volume and hydrophilicity of the corresponding amino acid.
We used an open source implementation of the Random Forest regressor from Scikit-learn package [28] to build the described model. The model performed well on the training sets, but the accuracy was poor on the test proteins with different amino acid sequence and composition. This was mainly caused by the fact that only a few among all possible amino acid quadromers were observed in the training sets. However, under assumption that nanopore current does not depend on the order of amino acids, it is possible to significantly expand the training sets by randomly permuting amino acids within quadromers. Specifically, prior to model we randomly permuted each fi vector, leaving the same corresponding qi value. This dataset expansion significantly improved the performance of the RF model on to training testing datasets. See Fig 3 for examples of theoretical nanospectra in the MV, SVR, and RF models.
Given an experimental nanospectrum S and a protein P, we transformed S into a vector S → by splitting S into |P| + 3 regions and taking the average value inside each of them. The vector S → was then normalized by subtracting the mean and dividing by the standard deviation. Under the hypothesis that P has generated S →, we estimated the proportion of explained variance by computing R2 coefficient of determination between S → and the model output. Given a database of proteins DB, a protein P(S, DB) is defined as a protein with the maximum R2 against S among all proteins from DB. A pair formed by the protein P(S, DB) and the nanospectrum S defines a putative Protein-Nanospectrum Match (PrNM).
Single protein correlation analysis indicated that proteins were correlated more with themselves on average (Fig 5). In contrast, we did not observe such correlation in the open pore current, indicating that there is an inherent signal in blockades. However, electrolytic current through the pore is affected by many factors, such as uncorrelated time-dependent fluctuations in the ionic current and electrical instrument noise, which results in noisy nanospectra. Averaging multiple nanospectra from the same protein resulted in significant noise reduction and increased accuracy of PrNM identification. This effect is similar to improvements in peptide identifications that are achieved by clustering of mass spectra in traditional proteomics [29, 30].
Typically, clustering of 5 − 10 nanospectra results in a consensus nanospectrum that significantly improves the signal-to-noise ratio over a single nanospectrum (the mean Pearson correlation coefficients between theoretical and empirical nanospectra increased 1.5—2-fold for various datasets). Since each of the existing datasets of nanospectra originated from a single pure protein, we randomly partitioned the dataset of nanospectra into clusters and performed identification of consensus nanospectra instead of a single nanospectrum.
In traditional proteomics, the precursor mass assists top-down protein identification since it greatly reduces the computational space that has to be searched in the protein database. Likewise, information about the protein length would be very useful for SMPI, but estimating the protein length based on a nanospectrum originating from a sub-nanopore is a non-trivial problem since the existing experimental protocol does not control the translocation speed that may vary widely as evident from the blockade duration.
Our analysis revealed that protein translocations modulate the blockade current, which was captured by the measurements. Each blockade, associated with the translocation of a protein showed a characteristic number of fluctuations during the duration of the blockade. It turned out that the fluctuation frequency (described below) was correlated with the protein length and the other features, such as amino acid composition.
We explored a possibility of the separation of a sample of nanospectra into clusters corresponding to different proteins. From a sample of different proteins, we estimated the fluctuation frequency of each nanospectrum as the number of peaks (local maximums) divided by the duration of the blockade. The distribution of fluctuation frequencies (Fig 5b) revealed that each protein in our datasets has a characteristic peak in the distribution. To separate the nanospectra into clusters based on the fluctuation frequency one can apply the Gaussian Mixture model to estimate the protein lengths from nanospectra and to improve the efficiency of SMPI.
Analyzing a mixture of multiple proteins is conceptually harder than analyzing the existing experimental datasets of nanospectra that all originated from pure protein solutions. Since it is unknown what protein gives rise to what nanospectrum in a mixture, it is difficult to cluster nanospectra for a reliable identification. Further, orientation of each molecule must be deduced prior to clustering since each protein can translocate through the pore in two different directions.
However, it is possible to cluster nanospectra based on their estimated fluctuation frequency to differentiate proteins with different lengths. As multiple proteins may have a similar length, it is important to further split some length-based clusters into finer protein-based clusters. We believe, that this could be done by applying clustering algorithms which automatically estimate the number of clusters (e.g. Affinity Propagation [31]). Evaluating the results of clustering in the case of complex mixtures was problematic since all available experimental datasets of nanospectra were generated from the pure protein solutions.
We benchmarked Nano-Align using nanospectra from five short human proteins: H3.2, H3.3, H4, CCL5 and H3 tail peptide (Table 1). The nanospectra from H3.2, H3.3 and H4 were acquired using the two similar pores whereas the nanospectra for CCL5 and H3 tail were acquired using two different pores with different sizes. The proteins were split into three pairs: (CCL5, H3 tail), (H4, H3.2) and (H3.3, H3.2). For each pair of proteins, the SVR and RF models were trained using the protein with higher number of nanospectra and the accuracy of identifications was estimated using the other protein from the pair. The first two pairs represented proteins that were very different in both length and amino acid composition, thus minimizing the overfitting effect. The third pair represented highly similar proteins, that only differ in four amino acids.
To evaluate the accuracy of SMPI, we constructed decoy protein database for each dataset from the correct protein and randomly generated proteins of the same length and amino acid composition as the correct protein. The size of decoy database varied from 105 to 5 ⋅ 106 for different datasets, depending on the identification accuracy and the number of nanospectra in the dataset. The p-value of a PrNM was approximated as the percentage of proteins from the database scoring higher than the correct protein against the given nanospectrum.
Below we show results for the SVR and RF models only since they turned out to be significantly more accurate than the MV model for all datasets. Fig 6 shows median p-values for SVR and RF models as a function of the number of nanospectra in a cluster. As expected, both models showed the improvement in the accuracy with the increase in the cluster size. The p-values for the pair (CCL5, H3 tail) were high for both models (0.03–0.05 for a consensus of size 10). However, the dataset (H4, H3.2) showed a significant improvement for the RF model (p-values of the order of 10−4 for a consensus of 10 nanospectra), while the accuracy of the SVR model was comparable to the previous dataset. Finally, the RF model showed high accuracy on (H3.3, H3.2) dataset, with p-values below 10−5 for the nanospectra clusters of size five.
The RF model consistently outperformed the SVR model on the datasets that were generated using pores of similar sizes, which suggests that the decision trees are better suited for SMPI due to their robustness against outliers. Also, amino acid hydrophilicity proved to be a valuable predictor of the pore signal. The RF model performed slightly worse than the SVR model on the dataset generated using two different pores, suggesting that it is more sensitive to the experimental conditions. The fact that the RF model performed better on the proteins that were more similar to the training proteins is not surprising, but rather highlights the importance of choice of the training set, which should have substantial coverage of the data.
Additionally, we benchmarked the RF model performance using a database containing real human proteins. We extracted all proteins of length between 100 and 160 from the human proteome (about 20% of the human proteome) and performed the identification of H3.3 spectra against this reduced database. On average, the true protein was ranked five against all other proteins (for a cluster of size five). An example of database hits is given in the Table 2. Interestingly, all high-scoring proteins belong to H3 histone family and differ by only few amino acids. While the search space was artificially reduced, this experiment already provides a justification for analysis of unknown nanospectra against small bacterial proteomes, after further improvements in the protein length estimation discussed above.
For each of the three models (MV, SVR and RF) we measured the bias with respect to different features of amino acids. Using H3.2 dataset (that provides the best amino acid coverage) we calculated the signed error defined as the mean difference between the empirical and theoretical nanospectra. For each amino acid, the signed error was measured among the associated quadromers. Fig 7 shows the volume-related bias of the MV model. This bias could be explained by the fact that larger amino acids have more influence on the pore signal than smaller amino acids. The SVR model and RF model show no bias with respect to amino acid volumes. A similar analysis revealed a bias with respect to amino acid hydrophilicity in the SVR model. The MV model did not show a clear dependence, possibly due to the dominant effect of the volume bias. The RF model showed no statistically significant bias related to hydrophilicity.
We presented the first algorithm for Single Molecule Protein Identification using a signal generated by a protein translocation through a sub-nanopore. We also proposed three models for generating theoretical nanospectra and concluded that the Random Forest model results in the most accurate identifications. The typical estimated p-values of identification accuracy were ranging from 10−4 to 10−6, which is already sufficient for a limited analysis of nanospectra against small bacterial proteomes containing a few thousands proteins. The comparison of algorithm performance on different datasets suggests that the model sensitivity will further improve when more nanospectra originated from different proteins become available.
Cysteine (Cys) was the highest source of error in all three models for H3.2. Likewise, Cys was an above average source of error in CCL5 [12] but, it was a below average source of error in the similar sequence of CXCL1. Thus, it seemed unlikely that only the size affects the error. On the other hand, both Cys and Met, which exhibit higher number of prediction errors are at the high end of the hydropathy index and have only few waters (4 and 10, respectively) binding them [32], which may indicate that water affects the blockade current. In addition, it has been speculated that charge could also affect the duration and magnitude of a blockade [12, 33]. Whereas it seems likely that both charge and water play a role in the blockade current, measurements and the MV model testing these ideas have been inconclusive so far.
While SMPI is currently not in a position to compete with top-down proteomics, this technology is still in its infancy. Furthermore, due to the inherent single molecule sensitivity, there are several avenues of research that can be addressed uniquely by SMPI that offer protein-discrimination from very small samples (attomoles). Thus, SMPI has a potential to emerge as a new technology for accurate protein identification.
|
10.1371/journal.pgen.1002008 | Diverse Roles and Interactions of the SWI/SNF Chromatin Remodeling Complex Revealed Using Global Approaches | A systems understanding of nuclear organization and events is critical for determining how cells divide, differentiate, and respond to stimuli and for identifying the causes of diseases. Chromatin remodeling complexes such as SWI/SNF have been implicated in a wide variety of cellular processes including gene expression, nuclear organization, centromere function, and chromosomal stability, and mutations in SWI/SNF components have been linked to several types of cancer. To better understand the biological processes in which chromatin remodeling proteins participate, we globally mapped binding regions for several components of the SWI/SNF complex throughout the human genome using ChIP-Seq. SWI/SNF components were found to lie near regulatory elements integral to transcription (e.g. 5′ ends, RNA Polymerases II and III, and enhancers) as well as regions critical for chromosome organization (e.g. CTCF, lamins, and DNA replication origins). Interestingly we also find that certain configurations of SWI/SNF subunits are associated with transcripts that have higher levels of expression, whereas other configurations of SWI/SNF factors are associated with transcripts that have lower levels of expression. To further elucidate the association of SWI/SNF subunits with each other as well as with other nuclear proteins, we also analyzed SWI/SNF immunoprecipitated complexes by mass spectrometry. Individual SWI/SNF factors are associated with their own family members, as well as with cellular constituents such as nuclear matrix proteins, key transcription factors, and centromere components, implying a ubiquitous role in gene regulation and nuclear function. We find an overrepresentation of both SWI/SNF-associated regions and proteins in cell cycle and chromosome organization. Taken together the results from our ChIP and immunoprecipitation experiments suggest that SWI/SNF facilitates gene regulation and genome function more broadly and through a greater diversity of interactions than previously appreciated.
| Genetic information and programming are not entirely contained in DNA sequence but are also governed by chromatin structure. Gaining a greater understanding of chromatin remodeling complexes can bridge gaps between processes in the genome and the epigenome and can offer insights into diseases such as cancer. We identified targets of the chromatin remodeling complex, SWI/SNF, on a genome-wide scale using ChIP-Seq. We also identify proteins that co-purify with its various components via immunoprecipitation combined with mass spectrometry. By integrating these newly-identified regions with a combination of novel and published data sources, we identify pathways and cellular compartments in which SWI/SNF plays a major role as well as discern general characteristics of SWI/SNF target sites. Our parallel evaluations of multiple SWI/SNF factors indicate that these subunits are found in highly dynamic and combinatorial assemblies. Our study presents the first genome-wide and unified view of multiple SWI/SNF components and also provides a valuable resource to the scientific community as an important data source to be integrated with future genomic and epigenomic studies.
| Chromosomes undergo a wide variety of dynamic processes including transcription, replication, repair and packaging. Each of these activities requires the recruitment and congregation of a particular set of factors and chromosomal elements. For example visualization of nascent mRNA in HeLa cells has led to a model of transcription units being clustered into “factories” thereby facilitating optimal engagement of RNA Polymerase II (Pol II) and coordination with other crucial holoenzyme complexes [1]–[3]. In addition to RNA Pol II and transcription factors, transcriptional assemblages include proteins critical to regulating chromatin. The accessibility of nuclear proteins to DNA is often controlled by ATP-dependent chromatin remodeling complexes, which are thought to play a role in a number of different cellular transactions by reshaping the epigenetic landscape.
The SWI/SNF (switch/sucrose nonfermentable) chromatin remodeling proteins were first discovered in Saccharomyces cerevisiae as components of a 2 MDa complex that repositions nucleosomes for vital tasks such as transcriptional control, DNA repair, recombination and chromosome segregation [4], [5]. Mammalian SWI/SNF is comprised of approximately ten subunits and the combinations of these subunits, some of which have multiple isoforms, enable multiple varieties of SWI/SNF complexes to exist both within a given cell and across cell types [6]. Among these subunits either of the two ATPases, Brg1 or Brm, is sufficient to remodel nucleosome arrays in vitro, however maximal nucleosome remodeling activity is achieved when the SWI/SNF subunits BAF155, BAF170 and Ini1 are present in a 2∶1 stoichiometry relative to Brg1 [7]. Whereas the ATPases have an obvious catalytic function, the roles of the other SWI/SNF subunits are largely obscure. Several reports indicate that BAF155 and BAF170 provide scaffolding functions for other SWI/SNF subunits as well as regulating their protein levels [8], [9]. SWI/SNF also contains β-actin and the actin-related protein BAF53, suggesting a possible bridge to nuclear organization or signal transduction, e.g. through phosphatidylinositol signaling [10], [11]. Phosphatidylinositol 4,5-bisphosphate has been shown to bind to Brg1 and promote binding to actin filaments [12]. Mutations resulting in loss of Ini1 function are associated with rare but aggressive pediatric cancers [13], [14]. The SWI/SNF subunits Brg1 [15] and ARID1A [16]–[18] are likewise thought to have tumor suppressor roles based on mutations recovered from other tumor types. Curiously, Ini1 alone has a unique and largely undefined role in HIV-1 infection that includes binding of Ini1 to HIV-1 integrase and the cytoplasmic export of Ini1 and its incorporation into HIV-1 particles [19]–[21].
The role of SWI/SNF components in cancer and tumor suppression is poorly understood despite extensive study. Detailed investigations of individual loci have implicated SWI/SNF in various transcriptional pathways including the cell cycle and p53 signaling [22], insulin signaling [23], and TGFβ signaling [24], as well as signaling through several different nuclear hormone receptors [25]. Although in vitro experiments and single-gene studies have been informative and have laid the foundation for understanding chromatin remodeling, a global analysis of targets of SWI/SNF is expected to yield a more extensive view into the biological roles of SWI/SNF components and their involvement in human disease.
In this study we present two complementary global analyses of SWI/SNF subunits to provide a more systematic view of SWI/SNF functions. First we performed ChIP-Seq with the ubiquitous SWI/SNF components Ini1, BAF155, BAF170 as well as the Brg1 ATPase. Second, in a parallel set of studies we performed mass spectrometry identification of proteins that co-immunoprecipitate with SWI/SNF components. Using our ChIP-Seq results the resulting chromosomal locations were integrated with published annotations to yield a more complete understanding of SWI/SNF on a genome-wide scale. We find SWI/SNF components frequently occupy transcription start sites (TSSs), enhancers, CTCF regions and many regions occupied by Pol II. Further analyses of the SWI/SNF regions we identified by ChIP-Seq reveals that SWI/SNF factors target genes and signaling pathways involved in cell proliferation and cancer. Our investigation of SWI/SNF protein interactions detected not only the expected co-occurrences of individual SWI/SNF factors with each other but also with cellular components such as nuclear matrix proteins, key transcription factors and centromere proteins implying a ubiquitous role in gene regulation and nuclear function. We find an overrepresentation of both SWI/SNF-associated chromosomal regions and proteins in cell cycle and chromosome organization. Collectively our results suggest that SWI/SNF is at the nexus of multiple signal transduction pathways, essential chromosomal functions and nuclear organization.
We identified the targets of four SWI/SNF components, Ini1 (SMARCB1), Brg1 (SMARCA4), BAF155 (SMARCC1) and BAF170 (SMARCC2), using ChIP-Seq. Chromatin complexes were isolated from HeLa S3 nuclei following independent immunoprecipitations with antibodies for each factor. Each of these antibodies was characterized by both immunoblot and mass spectrometry analyses (see Materials and Methods). Reads that mapped uniquely to the genome were retained (29–33 million reads per data set; Table 1) and significant binding regions were identified using the PeakSeq program with q-value<0.05 [26]. The peaks were compared to a similarly-sized data set of uniquely mapped ChIP DNA reads obtained from control immunoprecipitation experiments using normal IgG (i.e. a control serum that is not directed to any known antigens). Using this approach we identified many Ini1-, Brg1-, BAF155- and BAF170-associated regions (Table 1).
The majority of SWI/SNF binding occurs near (±2.5 kb) protein-coding genes, a distribution that is significant relative to a random target list (p<1×10−16; Genome Structure Correction (GSC) test [27]; see Materials and Methods). Several examples of SWI/SNF positioning relative to genic regions are shown in Figure 1 and Figure S1. In order to further examine SWI/SNF locations with respect to gene-rich and gene-poor regions we obtained a set of histone H3K27me3 domains that were identified in HeLa cells (Table S1; [28]) because this chromatin mark often occurs in gene-poor and repressed (i.e. heterochromatin) regions. Although most SWI/SNF-binding occurs outside H3K27me3 domains, we observed that SWI/SNF is occasionally found in heterochromatin regions, as shown in Figure 2. In this example a 7.5 Mb heterochromatin region on Chr16 contains a single gene, the neuronal cadherin CDH8, that is repressed and lacks RNA Pol II, however several SWI/SNF binding regions are found nearby.
We have performed considerable analyses of the targets for the individual SWI/SNF factors, particularly with respect to elements representing several major classes of genomic features including promoters (Ensembl protein-coding genes), RNA Pol II sites [26], CTCF sites [28], and predicted enhancers [29]. All of these features were identified in HeLa cells (Table 1, Tables S1 and S2; see Materials and Methods). In comparisons between the individual target lists for Ini1, Brg1, BAF155 and BAF170 with promoters, RNA Pol II sites, CTCF sites and enhancers we found that each SWI/SNF factor is significantly overrepresented for each of these major classes of genomic elements (p<1×10−16, GSC test, see Materials and Methods). To arrive at a single unified and more conservative list of SWI/SNF locations, we first took the union of all regions for Ini1, BAF155, BAF170 and Brg1, resulting in 69,658 SWI/SNF regions. We then trimmed this list to a high-confidence set of 49,555 sites by eliminating those regions where either only a single SWI/SNF subunit was present or that those regions that did not co-occur with either promoters, RNA Pol II sites, CTCF sites or predicted enhancers. We used this list of 49,555 SWI/SNF regions for all subsequent analyses unless otherwise noted (Table S3). The four major classes of genomic features mentioned above were overrepresented in both the 69,658 SWI/SNF regions as well as the more conservative list 49,555 SWI/SNF regions (p<1×10−16, GSC test).
We next examined the configurations of our 49,555 SWI/SNF regions (Figure 3A and Table 2). Ini1, BAF155 and BAF170 have been described as forming a ‘core’ based on their ability to stimulate remodeling activity of the Brg1 ATPase in reconstitution experiments [7]. Among our data 30,310 regions (61%) have two or more SWI/SNF components and 9,760 regions (20%) contain the core of Ini1, BAF155 and BAF170; for the purposes of this study we call this the ‘core set’. Among putative complexes comprised of two or more SWI/SNF subunits, we observed BAF155 was the subunit most common to each binding region. Only 770 SWI/SNF subunit co-occurrences were recovered that lacked BAF155 as compared to 6,467 for BAF170 and 14,824 for Ini1. This finding is consistent with several previous studies showing that BAF155 is important for SWI/SNF complex stability [8], [9]. BAF155 may increase the stability of the complex during assembly, or BAF155 may be easier to detect by ChIP.
One of the primary functions of chromatin remodeling complexes is to assist in gene regulation. Among the SWI/SNF regions in our high-confidence union set of 49,555 sites, 29% correspond to the 5′ ends of protein-coding Ensembl genes, 40% correspond to Pol II sites, 17% correspond to CTCF sites and 43% correspond to predicted enhancer regions (Figure 3B; Table 3). The various combinations of these four elements account for a total of 90% of the SWI/SNF union regions; 4,800 (10%) of the SWI/SNF regions are unclassified using the above elements. Similar trends were observed for the 9,760 SWI/SNF “core” regions where Ini1, BAF155 and BAF170 all co-occur (Table 3). None of these four particular SWI/SNF subunits or any combinations thereof exhibited a differential preference for one type of element (Table S4).
There are some differences between the SWI/SNF core and union regions. The SWI/SNF core regions are overrepresented for RNA Pol II (p<9.9×10−16; hypergeometric test) and 5′ ends (p<6.5×10−211; hypergeometric test) relative to all of the SWI/SNF high-confidence union regions; however the SWI/SNF high-confidence union regions are overrepresented for enhancer regions relative to the Ini1-BAF155-BAF170 core (p<2.4×10−67; hypergeometric test). Neither the SWI/SNF core nor the high-confidence union regions were over- or underrepresented for CTCF sites relative to each other (p>0.05; hypergeometric test).
Enhancers are often characterized by long-range interactions. We examined the locations of SWI/SNF binding regions in the 150 kb CIITA region where numerous chromosomal looping interactions have been mapped at high resolution in HeLa cells using 3C (Chromosome Conformation Capture). Brg1 has been previously mapped at several sites in this locus in these cells [30]. Superimposition of these 3C data on our SWI/SNF ChIP-Seq data (Figure 4) reveals that all six of the 3C interacting regions in the CIITA locus (−50 kb, −16 kb, −8 kb, pIV, +40 kb and +59 kb) are bound by SWI/SNF components. Moreover certain individual SWI/SNF component binding regions that appeared initially as orphans may now be seen as part of a complete complex when joined with a distal element. For example Ini1 at pIV when joined with BAF155 and BAF170 regions at the −16 kb element forms a SWI/SNF core. Thus in the CTIIA locus SWI/SNF regions are often associated with 3C regions and many of the regions bound by individual factors may in fact be part of entire SWI/SNF complexes inside the nucleus.
Overall our ChIP-Seq results are summarized in Table 1, Table 2, Table 3, and Figure 3 and indicate that SWI/SNF likely contributes to gene regulation through many different avenues in light of its binding to promoters, enhancers and CTCF sites. Furthermore SWI/SNF may facilitate looping interactions among these various elements as it has been shown in vitro that SWI/SNF can interact simultaneously with multiple DNA sites and generate loops between them [31]. Interestingly we found a slightly higher presence of the SWI/SNF core at TSSs and with Pol II than the SWI/SNF union regions with these elements (Table 3). Thus a complete core of Ini1, BAF155 and BAF170 may be required for effective promoter function whereas only a subset of these factors may be required for enhancer function. Alternatively a full SWI/SNF core may be more difficult to recover from a single enhancer element as compared to a more compact promoter region due to the enhancer's presumed interaction with many different distal elements.
As detailed above SWI/SNF regions are enriched for Pol II. To explore the prevalence of SWI/SNF with transcriptional machinery we asked whether the converse would also be true, namely if regions bound by RNA polymerases are enriched for SWI/SNF. Indeed Pol II regions are enriched for SWI/SNF binding regions (p<1×10−16, GSC test). Although Pol II overlaps extensively with SWI/SNF it differs from SWI/SNF in its concordance with CTCF and enhancer regions (Table S5). Pol II regions lacking SWI/SNF show a five-fold decrease in CTCF sites and a two-fold decrease in enhancer regions as compared to those Pol II regions containing SWI/SNF.
We further compared our SWI/SNF regions with binding intervals identified for RNA polymerase III (Pol III), which in addition to transcribing tRNA and other non-protein coding RNAs has an emerging role in the formation of boundary elements [32], [33]. Pol III localization data were obtained from published ChIP-Seq studies using HeLa cells ([34], [35]; Tables S1 and S6) and constitute 478 known and novel Pol III-associated regions. Pol II is often associated with Pol III (Table 4; reviewed in [32]. Therefore we examined whether SWI/SNF was associated with Pol III binding regions independently of Pol II. Of the 478 Pol III regions, 253 Pol III intervals lack Pol II and among these 39% (98/253) contain one or more SWI/SNF components. These results suggest that SWI/SNF association with Pol III can occur independently of Pol II.
Overall 65% (309/478) of Pol III regions and 84% (19,541/23,320) of Pol II regions have at least one SWI/SNF factor associated with them. The Ini1-BAF155-BAF170 core is found at 41% (195/478) of Pol III regions and 52% (12,079/23,320) of Pol II regions. From the colocalizations of SWI/SNF, Pol II and Pol III we see that there is substantial overlap among these factors yet each of these factors also has distinct characteristics.
SWI/SNF is known to act as both an activator and repressor of transcription [36]. We examined the locations of four SWI/SNF components relative to transcribed regions in HeLa S3 cells using the RNA-Seq data of Morin et al. [37], Ini1, Brg1, BAF155 and/or BAF170 are present at or near the 5′ ends (±2.5 kb) of 71 to 92% of active protein-coding genes. As noted above, SWI/SNF occupancy in promoters is similar to that of Pol II and each of the factors is individually enriched in promoter regions (p<1×10−16, GSC test). Although the majority of Ini1, Brg1, BAF155 and BAF170 target genes are expressed, an appreciable fraction of gene targets have little or no detectable mRNA in HeLa cells. A closer examination of the union regions where a SWI/SNF component is located in the promoter of an inactive gene reveals that 58% (2,063/3,565) of these promoters are co-associated with Pol II suggesting transcriptional stalling (reviewed in [38], [39].
Considering that SWI/SNF components bind near many expressed regions and that SWI/SNF factors occur in a multitude of configurations (Figure 3 and Table S3), we examined transcript expression levels for all possible combinations of Ini1, Brg1, BAF155 and BAF170 occurrences. Using the RNA-Seq data of Morin et al. [37], we examined transcript expression levels corresponding to each of these configurations (Figure 5). We see that the highest levels of transcription are associated with the following four configurations: 1) the complete core of Ini1, BAF155 and BAF170; 2) the complete core plus Brg1; 3) Ini1 and BAF155 only and 4) Brg1, BAF155 and BAF170. Although BAF155 is the subunit that is common to all of the configurations associated with the highest levels of transcription, it does not appear to be the sole driver of transcriptional activity. Compared against each other, all three components of the core complex taken individually have nearly indistinguishable profiles. Despite the involvement of Brg1 in two of the four configurations with the highest expression levels, most other configurations involving Brg1 are restricted to profiles associated with the lowest expression levels. One inference from these data is that certain combinations of SWI/SNF subunits are likely synergistic in promoting transcription whereas other combinations may be inhibitory or unstable.
We also examined SWI/SNF occurrences relative to 48,403 non-canonical small RNAs from HeLa cells (≤156 bp; Table S1) where most (83%; p<1×10−16, GSC test) of these small RNAs are near protein-coding genes [40]. Approximately one third (30%) of this entire small RNA set is within 1 kb of a target from our high-confidence union list of 49,555 SWI/SNF regions. The incidence of small RNA-SWI/SNF co-associated regions was nearly equivalent in protein-coding genes and intergenic regions. From this we surmise that SWI/SNF may contribute to gene regulation of a variety of transcripts, many of which are newly annotated and of unknown function.
Prior research has shown that a variety of signaling cascades are linked to SWI/SNF [25]. To gain further insights into potential actions of SWI/SNF components we examined the underlying Gene Ontology (GO) and Kyoto Encyclopedia of Genes and Genomes (KEGG) designations of their gene targets to determine significantly overrepresented annotations and pathways (Table 5 and Table S8). SWI/SNF gene targets were associated with ‘Pathways in cancer’ and several specific cancers types, e.g. chronic myeloid leukemia and pancreatic cancer. A number of signaling pathways and cellular processes that are “hallmarks of cancer” [41] were also overrepresented among the gene targets of Ini1, Brg1, BAF155 and BAF170. These include the Wnt, ErbB, p53, MAPK, and insulin signaling pathways, and processes endemic to oncogenesis and cancer progression such as DNA repair, the cell cycle and apoptosis. From these analyses we surmise the recruitment of SWI/SNF components is likely to influence the molecular basis of cancer through several potential mechanisms.
The SWI/SNF-enriched pathways are highly interconnected. Using the 49,555 SWI/SNF targets we identified a total of 24 KEGG signaling or biochemical pathways (Figure 6, yellow nodes). Interestingly, these pathways partition into three groups (Figure 6A–6C). Two of the groups (Figure 6A and 6B) comprise sets of pathways exhibiting at most one degree of separation, e.g. ‘inositol phosphate metabolism’ and ‘amino sugar and nucleotide sugar metabolism’. The third group (Figure 6C) consists of three pathways that are unrelated to any other pathways in the KEGG database. As displayed in Figure S2 directly related pathways such as ‘p53 signaling’ and the ‘cell cycle’ have shared components and many of the genes encoding these components are occupied by SWI/SNF factors. Thus, our results demonstrate that SWI/SNF is involved in many closely related signaling pathways and cellular processes and may help serve to coordinate expression of genes involved in these processes.
The genomic binding data demonstrates that SWI/SNF localization is coupled with a broad range of functional elements, suggesting that SWI/SNF may also be found with a broad range of associated proteins. To further examine the scope of SWI/SNF's roles in the nucleus we analyzed proteins associated with SWI/SNF subunits using co-immunoprecipitation followed by mass spectrometry. The SWI/SNF components Ini1, BAF155, BAF170, Brg1, Brm and ARID1A were immunoprecipitated from HeLa S3 nuclei, the resulting proteins were gel-separated and peptides were generated for analysis by mass spectrometry (See Materials and Methods; Table S9). In addition to the factor-specific antibodies, parallel immunoprecipitations were performed using non-specific IgG antibodies. Proteins identified in these “control IgG” immunoprecipitations were excluded as potential SWI/SNF co-purifying factors.
We identified a total of 101 proteins that were specifically associated with at least one of the SWI/SNF components assayed (Figure 7, turquoise edges; Table S10). Of the non-SWI/SNF subunits detected, 5 of these interactions were found previously in HeLa cells (e.g. estrogen receptor alpha [42], and 96 were new to this study. Interestingly one of the novel interactions we observed in HeLa cells, BAF155 with NUF2, has been previously observed in yeast between the yeast homolog of BAF155 (SWI3) and NUF2 [15]. Using the 101 nodes that we identified as proteins co-purifying with SWI/SNF in our undirected approach we ascertained overrepresented GO categories (Table 6). Several of these designations such as ‘cell cycle’ and ‘chromosome organization’ coincide with the categories obtained from GO and pathway analyses of SWI/SNF ChIP-Seq targets, suggesting the possibility of highly-interactive network structures.
Many of the proteins that were novel to this study reinforce and expand upon other published reports of SWI/SNF characterizations. For example SWI/SNF components have been localized by immunofluorescence to mitotic kinetochores and spindle poles [43], and Brg1-deficient mice show dissolution of pericentromeric heterochromatin domains [44]. From our immunoprecipitations BAF155 and BAF170 were associated with a number of kinetochore and centrosomal proteins (e.g. BUB1B, CENPE and NUF2, Figure 8, green circles). The role of SWI/SNF in the maintenance of kinetochore and spindle function is unknown. We detected a variety of transcription factor activators and repressors (e.g. NFκB1, NFκB2, RelA, PML and NFX1) as well as DNA repair (ERCC5 and RAD50) and cell cycle (e.g. CCNB3 and CDCA2) proteins (Figure S3). Some of the SWI/SNF interacting proteins themselves interact with one another. For example we detected several different proteins integral to estrogen and insulin signaling (Figure 7; Table S10). We also identified proteins associated with only one SWI/SNF factor; these may either be interactions with a specific SWI/SNF component or an inability to detect the protein in the immunoprecipitations.
We developed an expanded network of SWI/SNF associations by including proteins that were found by others to co-purify with SWI/SNF subunits (Figure 7, black edges). Only those factors that showed a one-degree separation with a SWI/SNF component in HeLa cells are displayed and all interactions are annotated in Table S10. SWI/SNF interacting proteins are associated with numerous UniProt keywords (Figure 8; [45]). Overall these results suggest a role for SWI/SNF components in a wide array of nuclear processes and diseases. Some of these processes may take place in nuclear substructures. Higher order chromatin structure is facilitated by the nuclear lamina and tethering of genes to the nuclear periphery is one epigenetic mechanism of gene regulation [46], [47]. Intriguingly we and others have detected SWI/SNF components with various nuclear envelope-associated proteins (Figure 7 and Table S10) including lamin A, EMD (emerin) and BAF/BANF1 (Barrier to Autointegration Factor, which although similar in name is not a SWI/SNF subunit). Two of the nuclear membrane proteins, SYNE1 and C14orf49, that we isolated in association with BAF155 are part of LINC complexes that link the nucleoskeleton and cytoskeleton [48], [49].
Numerous studies point to a high degree of functional organization in cell nuclei [46]. Emerging nuclear organization models would benefit greatly from a catalogue of processes and chromatin characteristics mapped to particular genomic elements. For example, the nuclear lamins are thought to influence chromatin organization, DNA replication and transcription [47], [50]. Our immunoprecipitation results demonstrating that SWI/SNF components are associated with lamin A/C (Figure 7 and Table S10) along with immunoprecipitation, immunolocalization and cell fractionation experiments from others demonstrating an association between SWI/SNF and nuclear lamina (e.g. emerin Figure 7; [51]) prompted us to investigate whether SWI/SNF and the lamins can be located to the same genomic sequences.
We isolated lamin A/C and lamin B ChIP DNA from HeLa S3 nuclei and performed ChIP-chip on tiling arrays covering the ENCODE pilot regions (see Materials and Methods and Table S1). Most of the 1,770 lamin A/C regions mapped to H3K27me3 domains (76%; 1,337/1,770) whereas the 1,270 lamin B regions were less commonly associated with H3K27me3 (29%; 372/1,270). Comparing regions where signal was detectable for the SWI/SNF, lamin A/C and lamin B experiments revealed that SWI/SNF has a much higher overlap with lamin B than lamin A/C. We found that 38% (297/784) of SWI/SNF sites are within 100 bp of a lamin B region whereas only 5% (41/784) of SWI/SNF sites are within 100 bp of a lamin A/C region (Table 7). For both lamin types the colocalization with SWI/SNF regions is significant relative to random target lists (lamin B, p<1×10−16; lamin A/C, p<1×10−15; GSC tests). SWI/SNF-lamin B intersecting regions contained approximately the same proportion of CTCF sites in the ENCODE regions as did all SWI/SNF sites in the ENCODE regions (p>0.05 hypergeometric test; Table 7). Enhancers are underrepresented in the SWI/SNF-lamin B regions relative to all SWI/SNF locations in the ENCODE regions (p<1.9×10−36; hypergeometric test). The SWI/SNF-lamin B regions are overrepresented for Pol II (p<2.9×10−39; hypergeometric test) and 5′ ends (p<7.3×10−37; hypergeometric test) relative to all SWI/SNF locations in the ENCODE regions.
In crosslinked chromatin SWI/SNF is detected primarily with lamin B, but as noted from the above mass spectrometry experiments, in solubilized, non-cross-linked cells SWI/SNF is detected with lamin A/C and not lamin B (Figure 1 and Figure 7). We interpret these results to indicate that SWI/SNF, lamin A/C and lamin B co-associate in different nuclear contexts but are all part of a broader interacting network with specific sub-associations.
SWI/SNF and the lamins have each been implicated in DNA replication (see above; [52], [53]). One of the proteins we detected as associated with SWI/SNF is the replication protein RepA and another regulator of DNA replication, geminin, has been found to co-purify with SWI/SNF in HeLa cells (Figure 7, red circles; [54]). We investigated whether there might be a relationship among SWI/SNF, lamins and DNA replication origins. We obtained a set of 282 DNA replication origins identified in HeLa cells for the ENCODE regions ([55]; Table S1). Of these 282 replication origins, 90 (32%) occur within 100 bp of a SWI/SNF region (p<1×10−16, GSC test), 86 (31%) occur at the 5′ ends of protein-coding genes and 151 (54%) occur within 100 bp of a lamin B region. In contrast to lamin B, only 17% (48/282) of the replication origins were near a lamin A/C region. These results are consistent with nuclear staining patterns observed in mouse 3T3 cells showing colocalization between lamin B and sites of DNA replication whereas the same colocalization patterns were not observed for replication foci and lamin A [52].
Of the 86 replication origins in promoter regions, 88% (76/86) intersected a lamin B region and most (78% or 67/86) were within a 100 bp of a SWI/SNF region. These data indicate that SWI/SNF components are located near many DNA replication origins, particularly those located in promoter regions. The coincidence of chromatin remodeling factors, promoters, lamins and replication origins at the same subset of genomic regions suggests that these loci may be particularly favorable for the formation of both DNA and RNA polymerase assembly and chromatin tethering. As shown in Figure 1 the interplay among these elements as well as with Pol II, CTCF and heterochromatin regions is complex and interwoven, such that each may share many different supporting and counteracting roles.
SWI/SNF performs a crucial function in gene regulation and chromosome organization by directly altering contacts between nucleosomes and DNA. In the work presented here we undertook a two-pronged approach (ChIP-Seq and IP-mass spectrometry) to move towards a more thorough understanding of these functions. Our ChIP-Seq analyses demonstrate that SWI/SNF components overlap extensively with important regions that require tight control of the dynamics of nucleosome occupancy such as promoters, enhancers and CTCF sites. Not only does the SWI/SNF complex change the accessibility of DNA but it also acts in concert with an extensive host of cooperating factors, thereby facilitating combinatorial control among various genomic elements. In addition to our ChIP-Seq results, the diversity and number of proteins that co-purify with SWI/SNF as identified in our mass spectrometry experiments further supports SWI/SNF's involvement with a variety of functionally distinct complexes.
RNA polymerases II and III are extensively colocalized with SWI/SNF components. Studies of transcription in HeLa cells have estimated that the number of active RNA II polymerases exceeds the number of transcriptionally active sites by at least one order of magnitude, leading to the proposal of “transcription factories” [1]–[3]. The number of RNA Pol II transcription factories in HeLa cells has been estimated between 5,000 and 8,000 where each factory can be typified by several looped loci, their resulting transcripts and distal elements such as enhancers. We infer that SWI/SNF regions are prevalent in transcriptional assemblages and their associated regulatory loops, given that >90% of our high-confidence union targets are associated with genic or regulatory regions and that 65% of Pol III and 84% of Pol II regions colocalize with at least one SWI/SNF factor (Table 4, Tables S5 and S6).
Interestingly we observed that SWI/SNF components often occur independently of each other and in various configurations across the genome, and similarly our mass spectrometry data point to heterogeneity of SWI/SNF complexes. We speculate that several mechanisms may underlie these various configurations and their associated genomic features, including 1) synergism or antagonism of the individual SWI/SNF factors in influencing expression (e.g. Figure 5); 2) failure to detect individual subunits due to epitope masking as a consequence of variation with local environments; 3) the capture of incomplete complexes that may in fact be completed upon superposition of genome-wide 3C data once such data become available (e.g. Figure 4); 4) the existence of SWI/SNF sub-complexes that deviate from the conventional composition of SWI/SNF assemblies (e.g. [56]) or 5) the capture of intermediates in a multistep assembly or remodeling process. This last view is consistent with a model of stochastic assembly that may occur through intermediate interactions and that has been described for several other large, multifactor complexes such as RNA polymerases and associated transcription factors [57], spliceosomes [58], and DNA repair complexes [59].
As shown in Figure 6 SWI/SNF occurs throughout many interconnected pathways. The assembly of functional SWI/SNF complexes at many locations in the genome may require the activation of one or more of these related pathways. Consequently some of the SWI/SNF associated regions we observed may reflect constitutive binding of partially assembled complexes that may be poised to receive additional signal inputs for subsequent regulatory activity. Indeed it has been shown that SWI/SNF components are present at regulatory regions even in the absence of stimulatory conditions or tissue-specific cofactors. For example Brg1 is present constitutively at the interferon-inducible genes IFITM3 [60] and CIITA [30] in unstimulated HeLa cells, which is consistent with our own finding of Brg1 and Ini1 at IFITM3 and various combinations of BAF155, BAF170, Ini1 and Brg1 at different elements in CIITA. In solution SWI/SNF factors are associated constitutively with RelB (HEK293 cells, [61]), RelA, NFkB1 and NFkB2 (HeLa cells, this study), the glucocorticoid receptor (T4D7 cells, [62]) and estrogen receptor alpha (HeLa cells, this study and [42]; SW13 cell extracts, [63]). The prevalence of SWI/SNF and the high degree of connectivity of its overrepresented pathways implies that SWI/SNF may assist in many related processes and may even facilitate crosstalk across many constituents of the transcriptional machinery. Notably SWI/SNF binds in the genes of its own subunits (Table S19) suggesting that SWI/SNF may contribute to auto- and cross-regulation of its subunit levels. Loss-of-function of a particular subunit, as may occur in certain cancers, could initiate oscillations and alter the relative abundance of the levels of the other SWI/SNF subunits through a variety of feedback and feed-forward loops. Aberrant SWI/SNF expression has been proposed to result in new combinatorial assemblies of SWI/SNF, some of which may deleterious [64].
The gene attributes revealed by our ChIP-Seq data substantiate that SWI/SNF is proximal to targets that comprise sets of fundamental biological processes. Many of the functional categories we found to be significantly overrepresented have disease implications, especially as related to cancer (Figure S2). For example failures in DNA repair and unchecked cell cycle activity are common characteristics of pre-cancerous cells, and our SWI/SNF analyses identified the p53 and MAPK signaling pathways, which are well known for maintaining checkpoint functions. Growth dysregulation particularly in the context of hormone signaling is another common cancer phenotype. Extracellular growth signals are transduced from the cell membrane to the nucleus by the ErbB, insulin and phosphtidylinositol signaling pathways, all of which we recovered as overrepresented (Table 5). The existence of phosphoinositide signaling in the nucleus and the ability of Brg1 to act as an effector for phosphatidylinositol 4,5-bisphosphate (PIP2) raises the prospect of several levels of control of this signaling pathway with respect to SWI/SNF [65], a hypothesis that can be examined in future studies.
Several of the overrepresented pathways we identified through our ChIP-Seq analyses share proteins detected in SWI/SNF co-purification experiments, thereby providing a resource to explore potential, highly-interactive network structures. For example we found that genes with products critical for ‘nucleotide excision repair’ were enriched using our SWI/SNF union list (Figure 6). Within this pathway the excision repair protein ERCC5 co-purified with both BAF155 and BAF170 in our IP (immunoprecipitation)-mass spectrometry experiments. The excision repair protein, XPC, associates with SWI/SNF in response to UV irradiation in HeLa cells, and BRCA1 and ATR also cooperate with SWI/SNF in DNA repair (Figure 7; Table S10; [66]). Thus we speculate SWI/SNF may participate in DNA repair through both transcriptional regulation as well as recruitment to regions undergoing repair.
Our study uses two strategies to attempt to comprehensively collect a SWI/SNF interaction network. We limited our network to a single model system, HeLa cells, because many attributes of SWI/SNF have been documented in these cells and it has been noted that SWI/SNF associations vary by cell type [67]. We extensively collated SWI/SNF protein interactions described in the literature. This undertaking was necessary because many of the proteins described in the literature as co-associated with SWI/SNF factors are not represented in interaction databases such as BioGRID, Molecular Interactions Database (MINT), IntAct, Human Protein Reference Database (HPRD), Nuclear Protein Database (NPD) and Interologous Interaction Database (I2D). Therefore we attempted to comprehensively collect such information to overcome these limitations. In total 158 SWI/SNF interacting proteins have been described in HeLa cells (Figure 8 and Table S10), which is similar to the number of SWI/SNF interacting proteins that have been described in other cell types [67]. Published molecular associations that were not discerned here might be due to interactions that are: 1) transient or of low affinity, 2) dependent on a specific set of biochemical conditions or 3) undetectable due to masking by the presence of more abundant protein(s) of similar size. In working with protein interaction data, similar degrees of overlap have been noted when comparisons are made across data sets [68], [69] and even in a well-studied model such as yeast, mass spectrometry analyses have found a plasticity of complexes and many previously undetected interactions [70]–[72]. From the ChIP-Seq and ChIP-chip results we expected that CTCF and lamin B may be among the proteins that co-associate with SWI/SNF, however neither of these factors was recovered in any of the non-directed experiments (Table S10), including a CTCF immunoprecipitation-mass spectrometry experiment performed in HeLa cells. In addition to the above considerations one possibility is that CTCF or lamin B may associate more strongly with one of the SWI/SNF factors not studied, e.g. BAF53A or one of the BAF60 subunits.
SWI/SNF is most often described in a chromatin remodeling context however data derived from a variety of sources suggests that SWI/SNF has other facets. It is possible that not all of SWI/SNF's functions involve DNA localization and therefore other types of global experiments, such as the IP-mass spectrometry, are valuable as first steps towards recognizing previously unknown roles. Unlike cytoplasmic compartments, nuclear compartments are not separated by a physical barrier but rather are functional assemblies that are typically organized around sets of molecules engaged in common functions. Data from both ChIP-Seq and IP-mass spectrometry illuminate the sectors in which SWI/SNF operates and the integration of these two methods is better than each alone for furnishing a broad comprehension of SWI/SNF action. For example ChIP-Seq enables the global identification of SWI/SNF chromosomal elements except for those regions with highly repetitive sequence such as human centromeres (Figure 2A). In this respect IP-mass spectrometry is complementary to ChIP-Seq because it strongly suggests that SWI/SNF occurs at kinetochores as evidenced by its co-purification with CENPE, NUF2, BUB1B and CLASP2 (Figure 7 and Figure 9). In addition to kinetochore proteins the SWI/SNF co-purification experiments also uncovered proteins from other substructures including centrosomes, microtubules, the nuclear periphery and PML nuclear bodies, the latter of which is characterized by cryptic foci of PML (promyelocytic leukemia protein) and has been implicated in a variety of diseases [73]. The ChIP-Seq and IP-mass spectrometry data are synergistic as well. Notably both methods found an overrepresentation of regions or proteins enriched for ‘cell cycle’ and ‘chromosome organization’. One possible inference from these studies is that SWI/SNF is well positioned to integrate signals across multiple signaling pathways both by its presence in a variety of cellular structures and its role in gene regulation through chromatin remodeling.
A fraction of SWI/SNF complexes co-associate with elements of the nuclear periphery where they are well situated to contribute to the nuclear organization and position-dependent gene expression (Figure 7; [74]). We found that in crosslinked cells SWI/SNF localizes more widely with lamin B than lamin A whereas in non-crosslinked cells SWI/SNF co-purifies with lamin A. As mentioned above lamin B may have escaped detection in SWI/SNF protein interaction studies. A related possibility is that SWI/SNF may exist in different nuclear pools that have varying solubilities and associations, such that recovery of particular SWI/SNF complexes depends upon the proteins with which SWI/SNF is associated. For example lamins A and B are known to have different nucleoplasmic mobilities and localization patterns [50], [52]. Immunolocalization experiments in HeLa nuclei have revealed that the A/C- and B-type lamins form distinct meshworks with occasional points of intersection [50], which is consistent with the interspersed patterns of lamin A/C and B that we detected (Figure 1). Hence it is reasonable to expect that SWI/SNF associated with lamin A would behave differently than when associated with lamin B. We surmise that in a chromatin context the dominant association of SWI/SNF with the nuclear lamins occurs in regions where lamin B is present. The purification of SWI/SNF with lamin A may indicate other biological roles, such as cell cycle progression or nuclear assembly [75], [76].
Gaining a more detailed understanding of SWI/SNF's activities in or near various heterochromatin environments will be central to comprehending nuclear events over the cell cycle as well as during development. Among the numerous molecular and epigenetic factors that have been found to affect heterochromatin formation or maintenance, the heterochromatin protein 1 alpha (HP1α, also known as CBX5; Figure 7) and Polycomb complexes (PcG) are of particular relevance to SWI/SNF [77]–[79]. Polycomb complexes promote gene silencing by catalyzing the trimethylation of H3K27 in its target regions, and SWI/SNF antagonizes this epigenetic silencing [80]. It is tempting to speculate that SWI/SNF found near the edges of H3K27me3 domains (Figure 1A and 1C) may be contributing to the establishment or maintenance of boundary elements. SWI/SNF may also engage in heterochromatin dynamics through its interaction with HP1α, which is often located in the centromeric regions (reviewed in [81]). Curiously HP1α interacts with the lamin B receptor [82] thus providing a potential bridge between heterochromatin and the inner nuclear membrane. Both H3K27me3 and lamin B are associated with spatially regulated genes whose conversion between active and inactive states depends on access to their regulatory regions, as may be conferred by SWI/SNF.
The work presented here provides new insights into the scope of SWI/SNF's influence in gene regulation and nuclear organization. The integration of numerous studies is beginning to reveal the complexities contributing to the regulation of any given locus. Contemporary models of transcriptional control propose that a series of factors transiently associate with a regulatory region before a decisive event tilts these intermediate reactions towards a productive outcome [57], [83]. SWI/SNF may contribute to such intermediate reactions or trigger switches between inactive and active states. The capacity for SWI/SNF to preserve many aspects of homeostasis also makes it vulnerable to being ensnared for aggressive cell proliferation. Our work demonstrates that SWI/SNF in particular and perhaps chromatin remodeling proteins in general will contribute unique insights to our understanding of gene regulation and disease mechanisms through the integration of target regions, spatial positioning and functional annotations. For example the co-occurrence of SWI/SNF with centrosomes, microtubules, kinetochores and the nuclear periphery may suggest that a pool of SWI/SNF is sequestered by these structures during mitosis to assist in the post-mitotic reformation of chromosomal territories. Our collective findings help inform a comprehensive view of SWI/SNF function as well as form a valuable compendium for future studies of nuclear functions as related to chromatin remodeling.
Suspension HeLa S3 cells were cultured by the National Cell Culture Center (Biovest International Inc., Minneapolis, MN) in modified minimal essential medium (MEM), supplemented with 10% FBS at 37°C in 5% CO2, to a density of 6×105 cells/mL. Cells were fixed with 1% formaldehyde at room temperature for 10 min. Fixation was terminated with 125 mM glycine (2 M stock made in 1x PBS). Formaldehyde-fixed cells were washed in cold Dulbecco's PBS (Invitrogen) and swelled on ice in a 10-mL hypotonic lysis buffer [20 mM Hepes (pH 7.9), 10 mM KCl, 1 mM EDTA (pH 8.0), 10% glycerol, 1 mM DTT, 0.5 mM PMSF, and Roche Complete protease inhibitors, Cat#1697498]. To isolate nuclei, whole cell lysates were homogenized with 30 strokes in a 7 mL Dounce homogenizer (Kontes, pestle B). Nuclear pellets were collected by centrifugation and lysed in 10 mL of RIPA buffer per 3×108 cells [RIPA buffer: 10 mM Tris-Cl (pH 8.0), 140 mM NaCl, 1% Triton X-100, 0.1% SDS, 1% deoxycholic acid, 0.5 mM PMSF, 1 mM DTT, and protease inhibitors]. Chromatin was sheared with an analog Branson 250 Sonifier (power setting 2, 100% duty cycle for 7×30-s intervals) to an average size of less than 500 bp, as verified on a 2% agarose gel. Lysates were clarified by centrifugation at 20,000× g for 15 min at 4°C.
Clarified nuclear lysates from 1×108 cells were agitated overnight at 4°C with 20 µg of one of the following antibodies: 1) anti-Ini1 (C-20), Santa Cruz Biotechnology, sc-16189; 2) anti-BAF155 (H-76), Santa Cruz Biotechnology, sc-10756; 3) anti-BAF170 (H-116), Santa Cruz Biotechnology, sc-10757; 4) anti-Brg1 (G-7), Santa Cruz Biotechnology, sc-17796; 5) anti-lamin A/C (H-110), Santa Cruz Biotechnology, sc-20681; 6) anti-lamin B antibody, EMD Biosciences, NA12; or 7) normal IgG, Santa Cruz Biotechnology, sc-2025. Antibody incubations were followed by addition of either protein A (Millipore #16-156) or protein G agarose beads (Millipore #16-266). Beads were permitted to bind to protein complexes for 60 min at 4°C. Immunoprecipitates were washed three times in 1x RIPA, once in 1x PBS, and then eluted in 1xTE/1%SDS. Crosslinks were reversed overnight at 65°C. ChIP DNA was purified by incubation with 200 µg/ml RNase A (Qiagen #19101) for 1 h at 37°C, with 200 µg/ml proteinase K (Ambion AM2548) for 2 h at 45°C, phenol:chloroform:isoamyl alcohol extraction, and precipitation with 0.1 volumes of 3 M sodium acetate, 2 volumes of 100% ethanol and 1.5 µL of pellet paint (Novagen #69049-3). ChIP DNA prepared from 1×108 cells was resuspended in 50 µL of Qiagen Elution Buffer (EB). Three biological replicates were prepared per antibody.
ChIP-Seq libraries were prepared and sequenced as previously described [26], [84]. Biological replicates for each factor were converted into separate and distinct libraries. To summarize, ChIP DNA samples were loaded onto Qiagen MinElute PCR columns, eluted with 15 µL of Qiagen buffer EB, size-selected in the 100–350 bp range on 2% agarose E-gels (Invitrogen) and gel-purified using a Qiagen gel extraction kit. DNA was end-repaired and phosphorylated with the End-It kit from Epicentre (Cat# ER0720). The blunt, phosphorylated ends were treated with Klenow fragment (3′ to 5′ exo minus; NEB, Cat# M0212s) and dATP to yield a protruding 3′-‘A’ base for ligation of Illumina adapters (100 RXN Genomic DNA Sample Prep Oligo Only Kit, Part# FC-102-1003), which have a single ‘T’ base overhang at the 3′ end. After adapter ligation (LigaFast, Promega Cat# M8221) DNA was PCR-amplified with Illumina genomic DNA primers 1.1 and 2.1 for 15 cycles by using a program of (i) 30 s at 98°C, (ii) 15 cycles of 10 s at 98°C, 30 s at 65°C, 30 s at 72°C, and (iii) a 5 min extension at 72°C. The final libraries were band-isolated from an agarose gel to remove residual primers and adapters. Library concentrations and A260/A280 ratios were determined by UV-Vis spectrometry on a NanoDrop ND-1000 spectrophotometer (Thermo Fisher Scientific). Purified and denatured library DNA was capture on an Illumina flowcell for cluster generation and sequenced on an Illumina Genome Analyzer II following the manufacturer's protocols [85].
Immunoprecipations were performed using the same conditions as for ChIP experiments except the HeLa S3 cells were not crosslinked. In addition to the ChIP antibodies described above we also used anti-Brm, Abcam Cat# ab15597 and anti-BAF250a (PSG3), Santa Cruz Biotechnology, sc-32761. Complexes were resolved on BioRad 4–20% precast Tris-HCl gels (Cat#161-1159) such that a single gel was used for each specific antibody and normal IgG immunoprecipitation pair. Gels were silver stained using Pierce SilverSNAP stain for mass spectrometry (Cat#24600) and each lane was excised into 10–12 molecular weight regions. Gel slices were destained, dried in a Savant speed-vac and digested overnight at 42°C with Sigma's Trypsin Profile IGD kit for in-gel digests (Cat# PP0100). Following the overnight incubation the liquid was removed from each gel piece and volume reduced by drying to approximately 10 µL. The individual gel slices were analyzed separately.
The samples were subjected to nanoflow chromatography using nanoAcquity UPLC system (Waters Inc.) prior to introduction into the mass spectrometer for further analysis. Mass spectrometry was performed on a hybrid ion trap LTQ Orbitrap mass spectrometer (Thermo Fisher Scientific) in positive electrospray ionization (ESI) mode. The spectra was acquired in a data dependent fashion consisting of full mass spectrum scan (300–2000 m/z) followed by MS/MS scan of the 3 most abundant parent ions. For the full scan in the orbitrap the automatic gain control (AGC) was set to 1×106 and the resolving power for 400 m/z of 30,000. The MS/MS scans were done using the ion trap part of the mass spectrometer at a normalized collision energy of 24 V. Dynamic exclusion time was set to 100 s to avoid loss of MS/MS spectral information due to repeated sampling of the most abundant peaks.
Sequence data from MS/MS spectra was processed using the SEQUEST database search algorithm (Thermo Fisher Scientific). The resulting protein identifications were brought into the Scaffold visualization software (Proteome Software) where the information was further refined resulting in improved protein id conformation. Scaffold search criteria were set at 98% probability and required at least 2 unique peptides per id.
All ChIP-Seq data sets (Ini1, Brg1, BAF155, BAF170, and Pol II) were scored against a normal IgG control using PeakSeq [26] with default parameters (q-value<0.05) to determine an initial set of enriched regions. These lists were then filtered by removing those regions that did not meet all of the following requirements: 1) the q-value from PeakSeq was further restricted to a q-value of<0.01; 2) a minimum of 20 sequencing reads per peak from the specific antibody ChIP; 3) an enrichment of 1.5-fold of the specific antibody relative to the normal IgG control; and 4) an excess of at least 10 of the specific antibody reads relative to the normal IgG control reads. Enriched regions satisfying these criteria comprised our initial list of enrichment sites for each factor (Table 1 and Tables S11, S12, S13, S14, S15, and S16). Among these data sources, Pol II and the normal IgG control have been published as part of prior studies and are available in GEO (accession numbers GSE14022 and GSE12781, respectively) [26], [84]. Data for Ini1, Brg1, BAF155 and BAF170 can be accessed through GEO series GSE24397.
After obtaining our initial list of enriched regions for each factor subjected to chromatin immunoprecipitation, we generated a union list of SWI/SNF component targets. Using the method described in Euskirchen et al. [86], we formed the union of Ini1, BAF155, BAF170, and Brg1 enriched regions as identified by ChIP-Seq and merged any unioned regions that were separated by ≤100 bp. Each union region was then classified by whether it intersected with one or more of BAF155, BAF170, Ini1, and Brg1. The resulting list consists of 69,658 SWI/SNF union regions (Table S2).
We compared our ChIP-Seq target lists for the 69,658 SWI/SNF union regions against genomic features at which chromatin remodeling is expected to play a prominent role: RNA polymerase II sites [26], 5′ ends of Ensembl protein-coding genes, CTCF sites [28], and regions predicted to be enhancers in HeLa cells [29]. We also compared individual SWI/SNF component lists against each other. Only those SWI/SNF regions which intersect another SWI/SNF component or which intersect at least one of the above genomic features were retained for the ‘high-confidence’ union list. For gene promoter regions, we define overlap as a target region with at least 1 shared bp within ±2.5 kb of the annotated transcription start site (TSS). SWI/SNF region intersections were calculated both for all genes in the Ensembl 52 database build using annotations from NCBI36 (human genome build hg18) as well as for a subset of genes that Ensembl identifies as protein-coding. The resulting target list consists of 49,555 ‘high-confidence’ SWI/SNF union regions (Table S3). Union regions containing all three of the BAF155, BAF170, and Ini1 subunits are designated as the 9,760 ‘core’ SWI/SNF regions (Table 3).
To determine the co-occurrences of features of interest we used a similar intersection strategy as was used for determining the high-confidence SWI/SNF regions. For all pairwise comparisons, one of the two data sets was extended by 100 bp on each side of the region and then intersected against the other, non-extended dataset. We required an overlap of at least 1 bp to deem two regions as associated. Using a Perl script, the intersection results for all comparisons were combined to form the co-occurrence table. The same procedure was followed to generate SWI/SNF-centric (Tables S2 and S3), Pol II-centric (Table S5) and Pol III-centric (Table S6) co-occurrence tables.
Using the HeLa RNA-Seq data of Morin et al. [37], we subdivided each list by the expression status of the corresponding gene targets. Expressed genes were defined as any Ensembl gene with an associated Ensembl transcript having an adjusted depth of ≥1, representing an average coverage of 1x across all bases in the transcript. A total of 9,711 expressed protein-coding genes satisfied these criteria.
We created a series of lists based upon the combinations of SWI/SNF components that could co-occur using the 49,555 high-confidence SWI/SNF regions derived from Table S3. Using the RNA-Seq data of Morin et al. [37], we intersected each list against the 5′ ends of transcripts queried by that study and recorded the corresponding adjusted depth for any transcript with a 5′ end within ±2.5 kb of a SWI/SNF region. Morin et al. treats adjusted depth as a measurement of transcription level for the corresponding transcript. For each list, these measurements were used to build a series of violin plots showing the probability distribution of transcription levels associated with different compositions of SWI/SNF subunits. Note that each SWI/SNF region from Table S2 can only be assigned to one list (e.g. a region containing BAF155, BAF170, and Ini1 is not also assigned to the list of regions containing BAF155 and BAF170).
Overrepresented GO categories [87] and KEGG pathways [88] were determined using DAVID tools [16]. Figures S2 and S3 were drawn using KGML-ED [89].
The ENCODE tiling arrays (NimbleGen Systems Inc., Madison, WI) interrogate the regions from the pilot phase of the ENCODE project [90] and tile the non-repetitive forward strand DNA sequence with 50-mer oligonucleotides spaced every 38 bp (overlapping by 12 bp) for a total of approximately 390,000 features. For array hybridizations ChIP DNA samples from 1×108 cells were labeled according to the manufacturer's protocol by Klenow random priming with Cy5 nonamers (lamin A/C or lamin B ChIP DNA) or Cy3 nonamers (normal IgG ChIP DNA). Biological replicates, defined as ChIP DNA isolations prepared from distinct cell cultures, were each hybridized to separate microarrays. Each lamin data set consists of three biological replicates. ChIP DNA labeling and array hybridizations were conducted by the NimbleGen service facility (Reykjavik, Iceland). Briefly, arrays were hybridized in Maui hybridization stations for 16–18 h at 42°C, and then washed in 42°C 0.2% SDS/0.2x SSC, room temperature 0.2x SSC, and 0.05x SSC. Arrays were scanned on an Axon 4000B scanner.
For each pair of arrays the files (in GFF file format) corresponding to the two channels for ChIP DNA (635 nm) and reference DNA (532 nm), were uploaded to the TileScope pipeline for normalization and scoring [91]. Data were scored with the following TileScope program parameters: quantile normalization of replicates, iterative peak identification, window size = 500, oligo length = 50, pseudomedian threshold = 1.0, p-value threshold = 4.0, peak interval = 1000, and feature length = 1000. Regions called by Tilescope were then filtered and corrected for multiple hypothesis testing by false discovery rate (FDR). To generate our set of background regions for FDR analysis, we randomly shuffle the probe values within each replicate, ensuring that the same probes are swapped for each replicate. This shuffled data set is then used as input to Tilescope and the scores compared against the lamin A/C and the lamin B data sets. The final lists of enriched regions for lamin A/C and lamin B have a final FDR of 0.1. Target coordinates were converted to hg18 using the UCSC ‘liftOver’ utility (http://genome.ucsc.edu/cgi-bin/hgLiftOver). Lamin A/C and lamin B data are available through GEO series GSE24382 and Tables S17 and S18.
To facilitate comparisons between sequencing and array data we retained only those regions that could be queried by both platforms. To this end, we first identified sequences represented on the ENCODE tiling array that possess less than 25% mappability in ChIP-Seq experiments using 30 bp reads. Any enriched regions in the lamin A/C and the lamin B data sets that were entirely contained within these regions of low mappability were removed from our lists, as corresponding signal levels are unlikely to be detected accurately via ChIP-Seq. Mappability was determined using a 30 bp read length and reported in 100 bp windows according to [26]. The end result is a list of lamin A/C and lamin B enriched regions identified by ChIP-chip in areas of the genome that can be queried by ChIP-Seq. Accordingly, regions that are not represented on the ENCODE tiling arrays were also removed from our SWI/SNF ChIP-Seq experiments for this comparison. Because our ChIP-Seq data covers the entire genome, we began by restricting our enriched SWI/SNF regions only to those that occur in the ENCODE pilot regions. We further refined our ChIP-Seq data set by discarding any SWI/SNF regions that occur in a region of the tiling array for which a signal level of 0 was observed via ChIP-chip. Once our SWI/SNF, lamin A/C, and lamin B lists were limited to those regions that could be queried by both platforms, we intersected the remaining lamin regions and the SWI/SNF regions using the same method that generated the all features table for enhancers, Pol II, and other elements, as described above. Similar procedures were followed for intersections with DNA replication origins identified in the ENCODE regions using tiling arrays [55].
To determine whether SWI/SNF components, core regions, and union regions are enriched for factors such as enhancers, small RNAs, lamin A/C and B, CTCF sites, Pol II regions, Pol III sites, 5′ ends and DNA replication origins, we used the genome structure correction test (GSC). This test determines the significance of observations where there “exists a complex dependency structure between observations” and was specifically designed for large-scale genomic studies [27]. Given two lists of genomic regions to compare and a list of coordinates defining the overall sample space (i.e. the length of each chromosome), a p-value for the significance of the overlap of the two lists is calculated and we report this value where noted.
All data produced for this study can be accessed through GEO and accession numbers for individual series are provided in the relevant sections. Alternatively, data from the lamin ChIP-chip experiments and the Ini1, Brg1, BAF155, and BAF170 ChIP-Seq experiments can be accessed through GEO using the SuperSeries accession number GSE24398.
|
10.1371/journal.pgen.1002704 | Extent, Causes, and Consequences of Small RNA Expression Variation in Human Adipose Tissue | Small RNAs are functional molecules that modulate mRNA transcripts and have been implicated in the aetiology of several common diseases. However, little is known about the extent of their variability within the human population. Here, we characterise the extent, causes, and effects of naturally occurring variation in expression and sequence of small RNAs from adipose tissue in relation to genotype, gene expression, and metabolic traits in the MuTHER reference cohort. We profiled the expression of 15 to 30 base pair RNA molecules in subcutaneous adipose tissue from 131 individuals using high-throughput sequencing, and quantified levels of 591 microRNAs and small nucleolar RNAs. We identified three genetic variants and three RNA editing events. Highly expressed small RNAs are more conserved within mammals than average, as are those with highly variable expression. We identified 14 genetic loci significantly associated with nearby small RNA expression levels, seven of which also regulate an mRNA transcript level in the same region. In addition, these loci are enriched for variants significant in genome-wide association studies for body mass index. Contrary to expectation, we found no evidence for negative correlation between expression level of a microRNA and its target mRNAs. Trunk fat mass, body mass index, and fasting insulin were associated with more than twenty small RNA expression levels each, while fasting glucose had no significant associations. This study highlights the similar genetic complexity and shared genetic control of small RNA and mRNA transcripts, and gives a quantitative picture of small RNA expression variation in the human population.
| Genetic information is transmitted to the cell only through RNA molecules. A special class of RNAs is comprised of the small (up to 30 nucleotide) ones, known to be potent regulators of various cellular processes. At the same time, they have not been as widely studied as messenger RNAs—we do not know how much variation in their sequence and expression level occurs naturally in human populations or how this variability influences other traits. We measured small RNA levels and genetic variability in fat tissue from 131 individuals by high-throughput sequencing. We could associate the expression levels with genetic background of the individuals, as well as changes in metabolic traits. Surprisingly, we found no large scale influence of small RNA variation on mRNA levels, their main regulatory target. Overall, our study is the first to give a quantitative picture of the naturally occurring variation in these important regulatory molecules in human fat tissue.
| A world of noncoding RNA molecules has been uncovered in the last decades, expanding our understanding of functional elements in the genome [1]. After it was found that the small (∼15–30 nt) noncoding RNAs can directly modulate protein levels [2], [3], and via that, almost any cellular process [4], they have been subject to vigorous study, leading to the recognition that several different types of small RNAs can act as posttranscriptional regulators [5].
MicroRNA genes (miRNAs) were the first animal small RNA genes to be discovered [6], and over 1,500 examples have been found in humans to date [7]. The primary miRNA transcript has a stem loop structure that is recognised and cleaved via RNA processing enzymes to produce a double stranded duplex [8]. The mature miRNA strand is loaded into a complex containing Argonaute family proteins and guided to targeting, while the other strand is assumed to be degraded. miRNAs target mRNA transcripts via base pair complementarity, typically in the 3′ untranslated region [8], [9], but also coding sequence [10]. This targeting can induce transcript cleavage, degradation, destabilisation, or repression of translation, thus modulating protein levels. Small nucleolar RNAs (snoRNAs) are typically longer genes (60–300 nt) that facilitate RNA editing within ribosomal or spliceosomal RNAs [11]. However, their full sequences can also be processed into snoRNA derived RNAs that exert a similar mode of action as miRNAs [12], [13], [14].
The recent ability to quantify levels of small RNA expression invites questions about the extent and causes of their variability in the human population. Importantly, the quantity and quality of transcripts are the only way genetic variation can influence phenotype. Thus, the genetic contribution to small RNA expression trait variability has to be assessed for accurate understanding of transmission of heritable information. Such questions have already been successfully addressed for mRNA expression levels, where variability between tissues [15], populations [16], and diseased and healthy individuals [17], as well as the contribution of genotype [16], [18], [19], [20] have been thoroughly characterised. Previous studies have found genetic contribution to miRNA levels in both human fibroblasts [21] as well as adipose tissue [22] using miRNA microarrays. However, other types of small RNAs have not been assayed, and a full account of small RNA sequence and transcriptome variability in a reference cohort is missing.
Small RNA expression can be viewed as a primary genetic trait to be mapped in isolation, but also as a quantitative trait with downstream influences on gene expression and other phenotypes. Recent studies have been successful in combining information about genotype and intermediate phenotypes (such as mRNA levels [17], [23], [24] or inferred cellular activations [25]) to understand how the genetic signal is mediated. In this light, it is especially interesting to analyse small RNA transcript levels as intermediate traits potentially causative for downstream effects, as both miRNAs and snoRNAs have already been implicated in many human disease phenotypes ranging from obesity and autism to cancer [26], [27], [28], [29], [30], [31], [32].
The MuTHER (Multi-Tissue Heritability Resource) cohort was established with the aim of analysing the genetics of gene expression in multiple human tissues in over 800 individuals [19], [20], [33]. This cohort is a subset of the UK Twins [34], and has extensive information on genotype and gene expression, as well as a plethora of clinical phenotypes. We set out to characterise small RNA variability in 131 abdominal fat samples from MuTHER resource using high throughput sequencing technology. We quantified the content of the small RNA transcriptome, the extent of sequence and transcript level variation, the relative levels of miRNA expression from both arms of the molecule, as well as coexpression of miRNAs from the same cluster. Since high density genotype data, mRNA levels from the same RNA sample as well as obesity-related phenotypes were available for these individuals, we associated these measurements with the small RNA levels to find out about the extent of genetic control, mRNA and miRNA expression correlates, and relation of small RNAs and global metabolic traits.
We sequenced subcutaneous adipose tissue small RNAs of 131 females from the UK TWINS cohort [34] included in the MuTHER study [19] on the Illumina GAII platform (Materials and Methods, data available at the EGA, submission ID EGAS00001000212). After filtering, quality control, and mapping, we obtained 331 million total reads, with a median of 2.3 million reads per sample aligning to the genome (Materials and Methods, Figure S1). The majority of the reads (93%) mapped to annotated mature miRNA sequences (mirBase v17 [7]), with the rest divided between tRNAs (2%), snoRNAs (0.6%), lincRNAs (0.3%), and other noncoding RNA features annotated in Ensembl v63 [35] (Table S1). This distribution is expected, as we size-selected for 15–30 base pair fragments, which excludes other functional RNA species except for degradation products. In addition, we found reads mapped to loci previously unannotated for noncoding RNA transcription. We identified 12 novel miRNA gene candidates using MapMi ([36], Materials and Methods, Dataset S1), and 701 short (<100 bp) regions with at least 1000 total mapped reads across all samples (2% of all mapped reads, Table S2). These regions were significantly enriched in DNAse hypersensitivity sites (237/701, one-tailed binomial p<10−10, Materials and Methods), which often harbour enhancer elements that are known to give rise to short transcripts [37]. The rest overlapped exons (149/701, one-tailed binomial p<10−10) and introns (190/701, not significant), with 233 regions arising from intergenic sequence.
We quantified expression levels of 418 known miRNA gene products, 239 tRNAs, 173 snoRNAs, 111 lincRNAs and 107 other RNAs that had at least 1000 total sequencing reads (Figure 1A, Table S1). For further analyses, we focused on miRNAs and snoRNA derived sequences as the only known functional molecules in our selected size range. The adipose tissue small RNA transcriptome is of medium complexity, with a median of 17 species of molecules required to account for 75% of the mapped reads (Figure 1B). The most highly expressed small RNAs (Figure 2A) have previously been associated with adipose development (mir-143-3p [38], mir-21-5p [12]), angiogenesis (mir-126-3p [39], mir-378a-3p [40]), and erythropoiesis (mir-24-3p [41], mir-451a [42]). We compared the average expression levels in adipose tissue to public human small RNA sequencing data from B-cells [43], liver [44], pigment cells [45], pooled thymocytes, bone marrow, CD34+ progenitor cells [46], lung, kidney, skeletal muscle, heart, pancreas, frontal orbital gyrus, spleen, and liver tissue [47] after processing them with our pipelines (Materials and Methods, Table S3). While seven of the ten most highly expressed small RNA genes and gene families were highly expressed in all tissues (let7 family, mir-24-3p, mir-378a-3p, mir-21-5p) other highly expressed small RNAs (mir-143-3p, mir-126-3p) were specific to adipose tissue (average q-value of pairwise comparisons <0.1, Materials and Methods). In total, there were 12 miRNAs with significantly higher expression (q<0.1) compared to mean of every other tissue, and no such snoRNAs, with mir-126-3p, mir-340-3p, mir-190a, and mir-335-3p showing the strongest specificity signal (Figure 2B).
Next, we called variants from the RNA sequence data (Materials and Methods), and found one mature miRNA and two snoRNA polymorphisms, all with independent evidence from whole genome sequencing of the UK10K cohort (personal communication, UK10K Consortium) (Table S4). All three found variants had relatively low (<11%) minor allele frequency (MAF). Assuming Hardy-Weinberg equilibrium, and equal expression from both gene copies, the miRNA sequence variant represents a fraction of 7×10−5 of the 14,005 mature miRNA and star sequence sites that could pass our filters, consistent with previous reports of strong purifying selection in the functional small RNA regions [48]. The same regions in the UK10K project harboured 13 called polymorphic sites, 9 of which had MAF<1%. We detected one of these sites using small RNA sequencing (MAF = 11%), and did not find the rest. Based on the MAF of each UK10K DNA variant, and expression levels of the small RNAs, we expected to recover one additional site (Materials and Methods). While it is possible that other polymorphisms are present in sequences coding for miRNA and snoRNA products, the derived alleles were not observed on at least 10 reads in our data, and could thus not be reliably detected. In addition to genetic variants, we found three A to I RNA editing events in the mature miRNA regions (Materials and Methods, Table S4). These sites were the 7th, 8th, and 9th bases of the mature product, and edited in 25, 18, and 11 percent of the reads, indicating that additional variability is tolerated in the functionally important seed region. We also observed bases at the ends of mapped reads not matching the genome in line with previous reports ([49], Table S5), but as similar discrepancies were not observed at comparable frequency in the data from other tissues, we considered them more likely to be sequencing or library preparation artefacts than true RNA modifications.
As mature miRNA sequences and analogous snoRNA products function via base pair complementarity, there is selective pressure against accumulating variants in their regions. Previous reports from DNA sequence data have confirmed increased conservation of miRNA sequence compared to intronic and intergenic background, but also a more pronounced effect for more highly expressed genes. We also observed a lack of miRNAs with at least 1000 reads on average and UCSC primate conservation score of less than 0 (Figure 3A, p<3×10−5, chi-squared test, Materials and Methods). Moreover, we assessed if the variability in the expression levels is under similar influence. Indeed, we observed a lack of small RNAs with expression variance of at least 5, and a conservation score below 0 (Figure 3B, nominal p<3×10−4, chi-squared test), suggesting that selection acts on not just average expression, but also expression variation.
After analysing the variability of RNA expression levels within and between tissues, we next addressed inter-individual variation. First, we tested whether experimental confounders influenced small RNA expression variability between samples. To this end, we performed principal components analysis of log-transformed, normalised read counts (Materials and Methods, Tables S6 and S7), and associated first twenty components (PCs) to known covariates of sample multiplexing tag, library batch, sequencing flow cell, RNA integrity score (RIN, [51]), and RNA concentration (Materials and Methods). We found significant associations (Bonferroni-corrected p<0.05) for RIN (PC1), library batches (PC1, 2, 3, 6, 10, 11, 12, 13, 18), and two multiplexing tags (PC5, 14). As these components capture major directions of variation in the data, we included the associated covariates measured for all samples (age, library batch and multiplexing tag) in eventual analyses. Since it has been demonstrated that unmeasured confounders similarly have an influence on expression levels [52], [53], we tested whether applying the Bayesian factor analysis package PEER [54], to account for these confounders, increases the number of discoveries. As we already corrected for 30 known covariates, including the additional inferred factors did not increase findings in downstream analyses, and were not used.
To identify miRNA and snoRNA genes whose expression is driven by cis-acting genetic variation, we performed association tests between their transcript levels and SNPs within 100 kb of the transcript (Materials and Methods). We found significant cis-eQTLs (nominal p<2.4×10−4, FDR<5%) for eight of 418 miRNAs and six of 173 snoRNAs (Table 1). In comparison, 462 eQTLs were found for 27,499 mRNA probes in the same tissue and cohort with comparable sample size and FDR in a previous study [19], suggesting a similar level of genetic control for mRNA and small RNA transcript levels.
We validated our eQTL findings in an independent cohort of 70 human samples with array-based miRNA expression data from abdominal adipose tissue [22]. Five of the eight miRNAs with an eQTL in our study were assayed in this study, with three of them replicating (nominal p<0.05, Table S8) and p-values across the full set of eQTLs tested in each study concordant (Spearman rank correlation p<8×10−4, Figure S2). However, we found no overlap between our significant cis-eQTL results and 12 significant (p<0.05 from 10,000 permutations) miRNA cis-eQTLs reported in human fibroblasts [21], likely due to a different set of expressed genes and lack of replication power. As mRNA studies in larger cohorts have found only up to a third of genetic associations to be tissue-specific [19], [55], we also expect many of the small RNA eQTLs to have an effect in other tissues in better powered studies.
The full MuTHER cohort of 776 individuals was profiled for mRNA levels from adipose tissue using the same RNA sample for the individuals in our study, as well as skin and lymphoblastoid cell lines from a separate RNA sample. Thus, we could directly assess any overlap in genetic control of transcripts of different type and across multiple tissues. We found seven of our small RNA eQTL SNPs to also be significantly associated with a nearby mRNA probe (Table 2). The mRNA transcripts were the nearest annotated transcript to the two miRNAs and two snoRNAs, but at least one and up to four annotated mRNA transcripts away from the rest of the snoRNAs. Further, in three of the eight cases, the mRNA and small RNA did not share the direction of the SNP effect. This suggests nontrivial shared genetic control, either via enhancer or promoter, or a single transcript that is spliced to form multiple genes.
As our cohort has been phenotyped for DEXA-derived measurements of percentage trunk fat mass (PTFM), BMI, fasting insulin, and fasting glucose (summaries in Table S9), we examined the association between small RNA expression and these obesity-related phenotypes (Materials and Methods). We found 47, 41 and 23 out of the 591 tested small RNAs to be associated with PTFM, BMI and fasting insulin respectively (per-trait FDR<5%). As these traits are highly correlated (Pearson's r>0.45 for all pairwise comparisons), there is also considerable overlap in the associated small RNAs between the traits (Table 3).
Fourteen small RNAs were highly significantly associated with at least one of the phenotypes (FDR<0.1%, Table 3, miRNA targets and functional enrichment analysis [56] in Table S10, Figure S3, S4). As a complement to the association analysis, we also contrasted small RNA gene expression levels between lean (BMI<25; n = 45) and obese (BMI>30; n = 36) subjects, and found that 43 small RNAs showed significant differences between the two groups (FDR<0.05, p<5.7×10−3, Table S11), including all significant hits from Table 3.
Four of the phenotype-associated small RNAs have previously been associated with metabolic phenotypes and/or adipogenesis. In a recent study, mir-1179 was found to be significantly associated (FDR<5%) with metabolic syndrome case control status, with lower expression levels in cases [22]. Here, we report similar associations between mir-1179 and obesity phenotypes with lower expression levels associated with increasing BMI, PTFM and FI, all of which are major components of metabolic syndrome. Mir-21, here significantly associated with obesity phenotypes (Table 3), has been reported to be involved in regulation of adipogenesis and lipid metabolism through its gene targets TGFBR2 and PPARalpha respectively [57]. Furthermore, mir-21 as well as mir-146b have been reported to be expressed at higher levels in skin tissue from diabetic mice [58], and in response to glucose stimulation in mouse adipocytes [59], [60]. Overexpression of mir-29 isoforms in mouse adipocytes resulted in an insulin resistant phenotype [59]. In a recent study, carried out in mouse islets, isoforms of mir-29 were found to contribute to the beta-cell-specific silencing of MCT1 (SLC16A1) expression required for appropriate insulin secretion [61]. In our study, mir-29b-2-5p was significantly associated BMI and fasting insulin, but not PTFM. Immune processes have previously been found to be enriched among mRNAs associated with metabolic phenotypes [17], [62], and mir-146a involved in inflammatory processes [63] and innate immunity [64] was here found to be associated with PTFM, BMI and insulin.
We overlapped the SNPs for our 14 significant cis-eQTLs (cis-SNPs), with SNPs that are directly associated with obesity-related phenotypes in published genome wide association study (GWAS) data [65]. Four of the cis-SNPs were associated (nominal p<0.05) with body mass index (BMI) [65], one each with waist-hip-ratio adjusted for BMI (WHRadjBMI) [65], low density lipoprotein (LDL) high density lipoprotein (HDL), and total cholesterol (TC), and none with triglycerides (TG) [66] (Table 4).
While on the whole, none of the cis-SNPs were genome-wide significant in the GWAS data, they were significantly enriched for nominally significant (p<0.05) SNPs in the BMI GWAS results ([65], binomial p = 0.007), indicating either their pleiotropic effect, or metabolic trait regulation through small RNA expression levels. rs2440129 was nominally significant in the BMI GWAS lookup [65], while mir-195-3p was significantly associated with both rs2440129 in cis (FDR<5%, p<2.4×10−5), as well as BMI (FDR<5%, p<3.9×10−3) and PTFM (FDR<5%, p<4.1×10−3), suggesting a mechanism for the rs2440129 association. Rs6658641 has a significant (FDR<5%, p<1.6×10−4) cis association with mir-197-3p in our data (Table 1), GNAI3 mRNA in three tissues (Table 2), as well as nominally significant associations to metabolic traits in GWAS. As mir-197 has been reported to regulate the expression of tumour suppressor gene FUS1 [67] and to be upregulated in type two diabetes patients [32], it is plausible that the effect of rs6658641 genotype on downstream expression and metabolic traits is mediated via the miRNA expression level.
miRNA genes are either processed from intronic mRNA sequence, or transcribed from endogenous promoters [68]. A single miRNA promoter can give rise to a transcript that includes a cluster of miRNAs that are then individually cleaved [69]. We tested whether pairwise correlations between expression levels of miRNAs in the same cluster (defined by Saini et al. [68] to be within a 10 kb block) are larger than those between random miRNAs, and found significant enrichment of positive correlation (Materials and Methods, Figure S5). The median of median pairwise correlations between cluster member expression levels was 0.37, compared to 0.03 of random miRNA sets of same size (p<10−8, Mann-Whitney U test). On the other hand, we found little evidence for relation between miRNA expression level and expression of its nearest mRNA probe. The distribution of correlation coefficients was centered on zero, without a heavy tail of positive correlation (Figure S6), a statistically significant difference to distribution of random small RNA-mRNA pairs (p>0.37, Mann-Whitney U test), or a trend for higher correlation for less distant probes. This shows that mRNA transcript levels are not good predictors of intronic miRNA levels in our dataset, and suggests that more miRNAs are expressed from an endogenous promoter than commonly appreciated, in line with recent findings [69], [70], [71].
One of the two modes of miRNA action is directly regulating the transcript level via influencing the stability of the transcript, or direct cleavage [72]. To test whether variability in the miRNA expression levels is related to variability in its target mRNA expression, we calculated correlations between miRNA expression levels and their validated mRNA targets from miRecords [73] or predicted mRNA targets from tarBase v5 [74] both with and without accounting for experimental confounders in mRNA and miRNA data sets (Materials and Methods). To our surprise, we found that the average correlation between miRNA expression levels and their 522 validated targets was −0.012, and their 194,205 predicted targets −0.004. While these averages are statistically significantly less than 0 (one-sample t-test p<0.05 and 10−5 respectively), they indicate no strong enrichment of extreme negative correlations compared to random miRNA-mRNA pairs (Figure S7). We also tested whether the miRNA seed sequences are overrepresented in the 3′ UTR regions of the mRNA expression levels most negatively correlated to the miRNA using Sylamer ([75], Methods). Again, we found no evidence for significant enrichment (all q-values>0.5). This suggests that at a genome-wide level inter-individual variation of small RNA expression levels in our reference cohort does not have a detectably large effect on mRNA expression.
The mature miRNA is processed from a double-stranded RNA hairpin by the Dicer RNAse [72], with the other arm assumed to be degraded [76]. The basis for choosing one of the hairpin arms as a mature product, and the extent to which the alternate arm (the less commonly observed product, previously also referred to as the star sequence) is functional, are not well understood [77], [78], [79]. To assess the extent of expression of both arms, and the variability of the relative expression ratio, we quantified the expression level of the alternate arm for 63 miRNAs. Other miRNA genes had only one arm detectably expressed, and only eight out of the 63 alternate arms were expressed at average level of at least 250 reads per sample. For seven miRNA genes, the alternate arm was on average more highly expressed compared to the mature product according to miRBase (Figure S8). Looking at variation between individuals, we found 12 mature sequence expression levels to be significantly correlated with their alternate arm sequence expression level (|Spearman's rho|>0.4, nominal p<2×10−5). For mir-186 and mir-29a, high abundance of the alternate arm sequence was indicative of low mature sequence levels, suggesting mutually exclusive selection of the arms. As the arm choice is suggested to be influenced by the nearby RNA context [78], [80], we tested for whether DNA variants in the region are correlated with the relative abundance of sequence from the two arms. We found SNP rs13174179 to be associated with the expression difference of miR-378 arms (nominal p = 5.3×10−4, FDR<10%).
We have presented the largest small RNA sequencing dataset in a human reference cohort to date, and demonstrated the extent, causes, and consequences of the variability in small RNA expression.
In spite of the medium complexity of the small RNA transcriptome, we quantified close to 1,000 different small RNA species. The highly expressed small RNAs fell into two categories in terms of inter-tissue variability - adipose-specific, and ubiquitously expressed microRNAs, corroborating previous observations [81], [82]. We confirmed that small RNA sequences have low genetic variability. This finding was especially pronounced for small RNAs highly expressed in the tissue we assayed, as only three derived alleles and three editing events were found. Additional genetic variants have been seen using DNA sequencing methods, but their potential functional impact remains to be assessed in other tissues where the genes are expressed above background level. Purifying selection acting on highly expressed as well as highly variable small RNAs was evident from their high conservation throughout the mammalian lineage, reiterating the importance of these functional molecules.
Unexpectedly, some of the largest sources of variability in our data were due to the experimental protocol. The barcoding method used in this study, whereby the indexing tag and the unknown RNA are sequenced in the same read, caused a bias in terms of the profile of small RNAs that were captured. This could be addressed by using a generic 5′ adaptor and one that incorporates the indexing tag via PCR, such that the RNA sequence and the indexing tag are determined in separate reads, or performing the reverse transcription step directly on the flow cell [83]. Similar issues with tag bias have been observed and addressed in recent work published after the experiments reported here were carried out [84], [85]. Additional limitations for the library preparation were the quality and quantity of the starting material. Although not always feasible in a clinical situation, every attempt should be made to ensure that the quality of the total RNA is of a very high standard (minimum RIN of 8), and it is subject to minimal handling and freeze/thaw cycles prior to library construction. These considerations forced us to employ statistical methods to account for batch effects due to multiplexing tags, and to drop 37 samples from our initial design due to poor RNA quality.
Differences in sample preparation and sequencing platform introduce technical variation that biases and reduces the power of direct comparisons between small RNA sequencing studies [86]. We limited such confounding effects on our assessment of small RNA expression tissue specificity by using only Illumina short read data from other studies, and treating their raw reads in an identical manner to our samples. While we do not expect this to fully mitigate the problem, we do not expect that the residual bias produces the reported large differences between tissues. These considerations do not affect the rest of our analysis, for which the small RNA and mRNA data were collected from the same RNA samples, and genotyping and phenotyping were performed on the same individuals.
Another important issue for comparing RNA levels between samples and finding genetic associations was mapping bias due to sequence variants. Previously uncharacterised polymorphisms resulted in fewer reads mapped to samples with derived alleles, which also created a significant eQTL at a known linked SNP. We recommend projects using small RNA sequencing to employ our technique of including known genetic variation in the reference sequence, and to use an ambiguity aware aligner, such as NovoAlign, to avoid such pitfalls.
Correcting for these technical issues, we were able to explore the biological causes of small RNA expression variation. We found genetic associations at a rate comparable with mRNA transcripts, and replicated them in an independent cohort. Unexpectedly, we found eight cases of a locus genotype influencing expression levels of a nearby mRNA and a nearby small RNA, where in four of these cases the two were unlikely to share a transcript as they were separated by at least one additional transcribed region. This highlights that cis, or proximal signal does not have to be contained to the near vicinity of the transcript, and that distal regulatory sites are shared between multiple genes.
We also looked for coordinated transcription by direct correlation of nearby transcripts. Small RNAs are known to be expressed in clusters from a shared promoter, as well as cleaved from intronic RNA sequence [68]. While we found support for increased correlation between miRNAs from the same cluster, we did not see a global signal for correlation between intronic miRNAs and their nearest mRNA probe expression. Previous results have shown a strong relationship between average tissue mRNA expression level and the intronic miRNA expression [81], but our results suggest the additional variability around the average level is not as tightly linked, possibly due to an independent promoter of the miRNA, or additional postprocessing regulation of the spliced mRNA transcript.
Finally, phenotypic and environmental differences can and do elicit changes in the transcriptome. To this end, we found 51 small RNA genes whose expression level is significantly associated with metabolic phenotypes available for our cohort. Given the strength of the observed signal, it is not possible without additional information to distinguish between causal, reactive, and common cause models for the relationship between the expression and phenotype traits. Studies in mouse models and human cohorts have shown that environmental factors, such as diet, can influence the expression of both mRNA [87] as well as small RNA [88] in adipose tissue. We used fat biopsies taken from individuals who had been instructed to fast the day of the biopsy to control for potential confounding effect of the daily food consumption, but long term dietary behaviour was not available for these samples and thus could not be analysed. Modelling potential hidden causes of variation in the expression data did not increase the number of discoveries, suggesting that even if the environmental factors were observed, they could not be accounted for in a simple linear manner. Despite this, we can not infer in general that the phenotypic variability is due to changes in small RNA expression. In some cases however, previous findings suggest a plausible regulatory effect of small RNAs on phenotypes as highlighted in the results.
The MuTHER cohort was set up with the aim to assess heritability of gene expression in different tissues using twins. However, as using highly related subjects reduces the power to map eQTLs using association, we focused our resources on unrelated individuals in the clinically relevant adipose tissue for which related phenotype data and an eQTL replication cohort were available. Analysing multiple tissues, or employing a co-twin design to provide heritability estimates and immediate replication of the results could be followed up from this pilot study.
A major goal of this study was to assess the effect of naturally occurring variation in miRNA expression levels on the mRNA levels. However, we found no evidence for miRNA expression variation to be correlated with target mRNA variation. This negative result cannot be due to the amount of noise in our data alone, as we could successfully detect genetic effects and phenotype correlations. Thus, the strength of association between natural variation of miRNA expression and variation in their target mRNA expression is limited to a smaller scale than that of genetic control or downstream effects of global metabolic phenotypes. This lack of tight target regulation supports the growing body of evidence [22] that quantitative variation of small RNA expression within a tissue does not have even a moderately sized effect on its target mRNA levels, and is consistent with a primary role of miRNAs being to buffer mRNA levels, for example to a random fluctuations of transcriptional regulators.
The small effect size of drastic miRNA level perturbation via knockdown, transfection, or overexpression of a single miRNA on its target mRNA expression levels has already been shown in several recent studies in human cell lines. For example, the median log2 expression level change of the top 150 TargetScan conserved targets was 0.096 (6.9%) for mir-29 knockdown in fetal lung fibroblasts [89], 0.131 (9.5%) for mir-145 transfection of MB-231 breast cancer cells [90], 0.173 (12.7%) for mir-30 overexpression in melanoma cell lines [91], and 0.465 (38.0%) for mir-7 overexpression in A549 cancer cells [92]. Thus, even for these extreme perturbations of miRNA levels, the observed effects on the target mRNAs are not pronounced. It is therefore not surprising that the naturally occurring inter-individual variation also does not have a large effect.
For the first time, we were able to assess the expression variation of both microRNA arms. We found that while the alternative arms (star sequences) are not highly expressed in general, there are several of them that are not degraded, and are expressed at appreciable levels. We also observed examples of high mature miRNA expression being correlated with low expression of the alternate arm, and a relatively strong genetic signal for arm choice of one miRNA.
The unrelated individuals included in this study are part of the MuTHER study of Caucasian females (median age 58) recruited from the UK Adult Twin Registry (TwinsUK, [34]). Punch biopsies (8 mm) were taken from a relatively photo-protected area adjacent and inferior to the umbilicus, subcutaneous adipose tissue was dissected followed by DNA and RNA extraction as described in [20]. For inclusion in this study the requirements were that the individuals were not under hormone replacement therapy, and did not have confirmed Diabetes Mellitus Type 2. Subjects were instructed to fast on the day of the biopsy to avoid potential biases due to food consumption. We used genotypes obtained, filtered and imputed to HapMap2 as described in [20]. The previously published gene expression values [20] were obtained using the Illumina Human HT-12 V3 BeadChips, followed by filtering and normalisation, and are available at the ArrayExpress [93] (www.ebi.ac.uk/arrayexpress) under accession number E-TABM-1140. Metabolic phenotypes were measured at the same time point as the biopsies and were collected as previously described, including BMI [94], DEXA measurements of percentage trunk fat mass, fasting glucose [95] and fasting insulin [96].
Only samples with good quality total RNA (no visible degradation in BioAnalyzer profile and RIN scores in excess of 6.7) were selected for small RNA isolation. Low molecular weight RNA (<40 nucleotides) was size-selected from between 0.5 to 1.0 µg total RNA using a flashPAGE Fractionator (Ambion, Austin, TX, USA). The recovered small RNAs were first ligated to the Illumina v1.5 small RNA 3′ adaptor (Illumina, Inc., San Diego, CA, USA) using T4 RNA ligase 2- truncated (New England Biolabs, Ipswich, MA, USA). This was followed by a second ligation, using T4 RNA ligase 1, to one of twelve modified Illumina SRA 5′ adaptors, each with a six-base index tag at the 3′ end (Sigma-Aldrich, Haverhill, UK). Both ligation steps were performed according to the Illumina v1.5 protocol. The 5′ and 3′ adaptor-ligated small RNAs were immediately reverse transcribed, amplified and size-selected as described in the Illumina v1.5 protocol. The completed cDNA libraries were pooled (12 libraries per pool) in equimolar amounts and were sequenced using 37 base reads on the Illumina GAII platform.
Raw sequencing data was obtained in FASTQ format, and processed with R [97](Bioconductor [98], [99], Biostrings and ShortRead [98], [99] packages) and python scripts. We first assigned the raw reads to their corresponding multiplexing tags. For this, we calculated the edit distance of the first six bases to all 12 index tag sequences used in the study, considering 0.25 as the distance between N and any other base. Reads with edit distances of at least 2.75 to all tags were discarded as well as those with the same minimum edit distance to more than one tag. The remaining reads were assigned to the library corresponding to the shortest edit distance, and their first six bases were removed before proceeding. The next step consisted of locating and trimming sequences matching the small RNA 3′ adaptor using the trimLRPatterns function and allowing for mismatches of up to 20% of the alignment length. The first 12 bases of the 3′ adaptor sequence were allowed to align to any location within the short reads, and if no alignment was found, a shortened adaptor sequence was realigned iteratively by removing one base from the 3′ end and anchoring the alignment to the 3′ end of the short reads. To further clean up the short reads to help avoid ambiguous mappings, any window of five bases with at least three Ns was located and the read was trimmed starting at the position of the first N. Any occurrence of an N within two bases of the 3′ end of the read was also trimmed. The reads were then low-complexity filtered to remove those with > = 90% of a single base. After all filtering steps, reads with less than 16 bases were discarded. All remaining read sequences should correspond to short RNA molecules present in the samples, and length histograms were produced to confirm the enrichment of a miRNA peak around 22 bases.
Accurate quantification of small RNA molecule counts from read data is challenging due to genetic variation in the sequence, ambiguities in read mapping, and frequent contamination by large numbers of adaptor dimers. To solve these problems, we used a multi-stage mapping approach to exclude contaminating molecules that could be due to the library preparation kit, prioritise alignments to known small RNAs, and take genetic variation into account.
First, we aligned the known small RNA molecule sequences against the human reference genome (NCBI build 37), retrieved all known variants in the mapped regions from the UK10K sequencing data (July 2011 release, personal communication, UK10K Consortium), and created an individual sequence of each small RNA, with variable bases denoted with the corresponding IUPAC ambiguity codes. We included all mapped regions for RNAs that mapped to more than one genomic location. We then created five synthetic reference genomes (all with the ambiguous bases at variable sites) corresponding to:
We mapped reads to these references using BWA [100] (bwa aln -n 2 -o 1) and with novoalign v. 2.07.11 (http://www.novocraft.com, parameters -h 60 60 -t30 -s -m -l 16 -R 0 -r A 30). The latter is aware of sequence ambiguities, but the former was more sensitive at detecting reads aligning to the contaminating sequences. We also tested Bowtie [101], but did not use it due to the inability of the tested version to handle indels.
For the mapping calls, we excluded reads mapping to contaminating sequence with either method. For both aligners, we then took all the alignments in the highest stratum of references (miRNA>ncRNA>pseudogenes>genome), and picked the ones with the smallest edit distance. Conservatively, we only retained alignments of a read if both aligners agreed on all the aligned locations. For reads mapping genome-wide, but not any known ncRNAs, we created the set of uncharacterised RNA loci covered by at least one read in at least one sample without gaps, and assigned reads to their corresponding uncharacterised loci.
Finally, we quantified the expression level of each ncRNA and unannotated locus by counting the number of reads aligning to it. If a read mapped between k alternative sequences or loci in one reference, we added 1/k to the count of each. We trimmed the data matrix to contain only RNAs that were observed at least 1000 times across all individuals, or at least 100 times in a single individual. We discarded individuals with less than 500,000 mapped reads. This retained 131 individuals, including 129 individuals with at least 800,000 reads, and 119 individuals with at least 1,500,000 reads.
This multi-stage approach excludes mapping contaminating sequences to the reference, avoids allelic imbalance due to ability to map by incorporating information on genetic variation, and resolves potential mapping ambiguities to reflect our belief of how small RNA molecules are generated.
To use the read counts quantitatively, we normalised the data to have a comparable total number of reads for each individual. We estimated a size factor s for library j as the median inflation factor across all genes: sj = mediang (ngj/GMj(ngj)), where GM stands for the geometric mean, and ngj is the read count of gene g for individual j as recommended by Anders and Huber [102]. For further analyses, we used the log2-transformed corrected values log2(ngj/sj) to account for heteroskedasticity in the data.
We downloaded UCSC genome browser [50] tracks for DNASE hypersensitivity sites, and ENSEMBL gene structures for human genome version 37, and calculated their total length, as well as overlap with the loci giving rise to unannotated small RNA molecules. We calculated the significance of the enrichment of unannotated regions in the track from the probability of observing at least as many overlaps of the 701 unannotated regions given the frequency of bases covered by each track using a standard binomial test.
We tested for significance of the correlation coefficient r between covariates and principal components of the raw read count data as well as log-transformed and normalised data by calculating a statistic t = r ((1−r2)/129)−0.5, and calculating the (two-tailed) probability of observing at least as extreme a value, and Bonferroni correcting for 131 tests (one for each PC). We called the correlation significant, if the corrected p-value was less than 0.05, corresponding to |r|>0.31.
For each sample, we created sorted BAM files from the alignment output, and called segregating sites using Samtools v.0.1.12 (samtools pileup –vcf ) [103]. We then combined the list of all called variable sites across all samples with sites from the UK10K project (June 2011 release, personal communication, UK10K Consortium), created pileup files at them for each sample (samtools pileup –l sites.tab –f ref.fa), and combined all the information into a single table giving the number of times each nucleotide was observed in every sample for each site. The sites were filtered to have information from at least 20 samples, have at least one sample with at least 10 observed alleles, and have at least one sample with at least 20% non-reference allele frequency. We further discarded three sites as likely false positives – one had 23 observed non-reference alleles, with 16 in one sample and no DNA evidence (see below), and the other two were variants in the last base of the mature miRNA, consistent with a modified degradation product. For validating the genotypes using genome sequencing data, we constructed DNA read pileup files at same sites for 40 of our samples sequenced in the UK10K cohort. We called a site to be a DNA polymorphism, if it had at least five DNA sequencing reads supporting the non-reference allele. An A to I edit was called if there were no more than two DNA sequencing reads with a G allele, and both A and G alleles were observed at least 90% of the samples, implying extreme deviation from Hardy-Weinberg equilibrium.
We applied the MapMi pipeline [36] to find potential novel miRNA loci. We retrieved the sequences of unannotated genomic regions, calculated their corresponding RNA secondary structure using RNAfold [104], and applied the MapMi classifier to obtain a structure score s. We calculated the self-containment score c of hairpins with s>35 as described in [105], and retained hairpins satisfying s*c>35. We then mapped all reads to the filtered candidate hairpins using bowtie, allowing for zero mismatches, and manually assessed the structural characteristics, genomic context, and alignment pileup shape for each candidate hairpin.
Mammalian conservation scores were downloaded from the UCSC genome browser [50]. A chi-square test with one degree of freedom was used to test the deviation of the fraction of highly expressed (average log-scale expression>10) unconserved (conservation score <0) genes from expectation. Similar test was used for highly variable (log scale variability>5) unconserved genes.
We subtracted off the linear fit of sample covariates (library batch and multiplexing tag), from the log-transformed, normalised data, and calculated the Pearson correlation coefficient between the residual expression levels and other small RNAs, miRNA star expression levels, and mRNA probe expression levels from [20]. For mRNA levels, we used both raw measurements, as well as residuals after correcting for global variance components using PEER. We also tested for correlation with uncorrected expression levels, and using linear models as described below, but found no additional enrichment of statistical signal.
miRNA binding specificity is controlled through binding of its seed region (bases 1–8 of the mature miRNA) with seed complementary regions (SCRs) in the 3′ UTR of mRNAs. Binding is enhanced if a SCR matches the first seed nucleotide with an adenosine, irrespective of the seed nucleotide [72]. As the strongest statistical associations have been reported for regions of length 6, 7 and 8 (the full region), we combined analyses for these seed lengths.
For each miRNA, we ordered probes and their associated 3′ UTRs by correlation of probe expression values to the miRNA, with correlations calculated in four different ways as described above. For each possible 8-nucleotide sequence s8 ending in an adenosine, we considered its middle 6-mer s6, the two constituent 7-mers s7,1 and s7,2 and the 8-mer s8 itself as the seeds. For each of these four seeds s and given n, we used Sylamer 08-185 [75] to calculate a hypergeometric p-value pn(s) to assess the extent to which the number of their SCR incidences in the top n of the ordered 3′ UTRs deviated from the expected. Potential nucleotide composition biases were accounted for using third order Markov correction (flag -m 4). The seed enrichment score for s8 was calculated as maxn(−log10 pn(s6)−log10 pn(s7,1)−log10 pn(s7,2)−log10 pn(s8)) using a grid of values for n.
For each miRNA that produced a ranked list of probes, the null distribution of observed scores was estimated by fitting an extreme value distribution for all calculated adenosine-ending 8-mer scores using the R function fgev from the evd package. For the miRNA used to generate the list, the significance of its influence on the mRNA expression was evaluated by testing its seed enrichment score against the estimated null. q-values were calculated for the ordered list of miRNA p-values.
We selected series GSE18651, GSE19737, GSE27718, GSE14507 from the Gene Expression Omnibus [106] in which a particular miRNA was directly perturbed, either by knockdown or overexpression. We downloaded the normalized expression data using Bioconductor package GEOquery [107], and performed differential expression analysis using limma [108] to sort the genes according to fold-change in response to the perturbation. We first validated that the mRNA expression changes actually represent the direct effect of a miRNA on its targets. To do so we used Sylamer [75] to search for enrichment of seed-matches in the 3′UTR sequences in the appropriate portion of the genelist, i.e. in knockdown experiments targets should be up-regulated, upon overexpression targets should be down-regulated. We then obtained the targets of each miRNA according to TargetScan v5 [109], and calculated their median fold-change in the corresponding experiment. We tested different sets of targets, prioritizing by evolutionary conservation (PCT) or by context-score, and selecting the 150 targets with the best scores. In all cases the median fold-change of these target sets was quite low, representing changes of 5–38%. Selecting more targets led to a reduction in the median fold-change. We also calculated the median fold-change of all possible targets, taking the full set of transcripts with at least a 7mer seed-match in their 3′UTR. These larger sets had the lowest median fold-changes, representing a 2–8% change in expression. All this confirms the notion that miRNAs do not act as on-off switches on the majority of their targets. Even in experiments that dramatically alter miRNA abundance, the average effect upon targets is modest.
Associations between snoRNA and miRNA expression and mean genotypes (expected minor allele count under IMPUTE posterior probabilities, MAF>5%, IMPUTE info value>0.8) or phenotypes were tested using a linear model implemented in R [97]. Cis-eQTL analysis was limited to SNPs located within 100 kB either side of the transcript. The linear model was adjusted for age, multiplex tag and library batch. The significance of the genotype or phenotype effect was calculated from the Chi-square distribution with 1 degree of freedom using −2log(likelihood ratio) as the test statistic. False discovery rate (FDR) was calculated using the qvalue package implemented in R 2.11 [97]. Corrections for multiple testing were done using q-values to control the false discovery rate (FDR) at 5%. To calculate the FDR, the associations between the 591 small RNAs and all the cis-located SNPs for each small RNA, were considered.
To test for difference in small RNA expression between obese (BMI>30) and lean (BMI<25) individuals we treated BMI, for subjects falling into one of the two BMI groups, as a binary categorical variable. Linear models where fitted with small RNA expression level as response variable and the lean/obese categorical variable as the predictor while adjusting for relevant covariates (age, library batch, multiplex tag). Significance of the effect size estimates of the lean/obese predictor was determined by a likelihood ratio test, and FDR was calculated using the qvalue package.
|
10.1371/journal.ppat.1001118 | A Major Determinant of Cyclophilin Dependence and Cyclosporine Susceptibility of Hepatitis C Virus Identified by a Genetic Approach | Since the advent of genome-wide small interfering RNA screening, large numbers of cellular cofactors important for viral infection have been discovered at a rapid pace, but the viral targets and the mechanism of action for many of these cofactors remain undefined. One such cofactor is cyclophilin A (CyPA), upon which hepatitis C virus (HCV) replication critically depends. Here we report a new genetic selection scheme that identified a major viral determinant of HCV's dependence on CyPA and susceptibility to cyclosporine A. We selected mutant viruses that were able to infect CyPA-knockdown cells which were refractory to infection by wild-type HCV produced in cell culture. Five independent selections revealed related mutations in a single dipeptide motif (D316 and Y317) located in a proline-rich region of NS5A domain II, which has been implicated in CyPA binding. Engineering the mutations into wild-type HCV fully recapitulated the CyPA-independent and CsA-resistant phenotype and four putative proline substrates of CyPA were mapped to the vicinity of the DY motif. Circular dichroism analysis of wild-type and mutant NS5A peptides indicated that the D316E/Y317N mutations (DEYN) induced a conformational change at a major CyPA-binding site. Furthermore, nuclear magnetic resonance experiments suggested that NS5A with DEYN mutations adopts a more extended, functional conformation in the putative CyPA substrate site in domain II. Finally, the importance of this major CsA-sensitivity determinant was confirmed in additional genotypes (GT) other than GT 2a. This study describes a new genetic approach to identifying viral targets of cellular cofactors and identifies a major regulator of HCV's susceptibility to CsA and its derivatives that are currently in clinical trials.
| Identification of cellular cofactors and their mechanisms of action is a fundamental aspect of virus-host interaction research. Screening of genome-wide small interfering RNA libraries has become an efficient way of systematically discovering cellular cofactors essential for various aspects of viral life cycle. We and others have recently demonstrated that cyclophilin A (CyPA) is an essential cofactor for hepatitis C virus (HCV) infection and serves as the direct target of a new class of clinical anti-HCV compounds, cyclosporine A (CsA) and its derivatives, that are devoid of immunosuppressive function. Here we report the identification of a key regulator of HCV's dependence on CyPA and susceptibility to CsA using a novel genetic screening approach that can potentially be applied to additional cellular cofactors and other viruses. The effectiveness of this approach, termed cofactor-independent mutant (CoFIM) screening, was further supported by results obtained with a parallel CsA-based selection using additional genotypes of HCV. This paper reports a new technology with which we discover and characterize the major determinant of HCV's sensitivity to CyPA inhibitors, which are currently being tested in clinical trials.
| Successful completion of the life cycle of a virus depends not only on the function of proteins encoded by the virus but also on cellular cofactors. The availability of genome-scale small interfering RNA (siRNA) libraries and high-throughput screening technology has permitted systematic efforts to discover cellular proteins important for viral infections in cell-culture systems. Typically, proteins with apparent functional relevance to the particular virus are characterized in detail after the discovery, but the mechanisms of action for many other cofactors remain undefined. An important step toward the illustration of the mechanism is to identify the viral agent through which a cofactor functions. Although in rare cases, when small chemical compounds are available whose target is the cellular cofactor, screening for compound-resistant mutant virus can provide valuable information about the viral target, but this approach is not useful for the majority of cofactors. Here we report a new genetics approach, which we designate cofactor-independent mutant (CoFIM) screening, to identify the viral targets of cellular cofactors through the selection of mutant viruses that can replicate in cells where a particular cellular cofactor is knocked down. Because our approach does not rely on prior knowledge of the function of the cofactor or the availability of chemical inhibitors, it may be broadly applicable to cellular cofactors with unknown mechanisms of action.
Various cellular proteins have been implicated in the life cycle of hepatitis C virus (HCV), identified mostly by protein-protein interaction and/or siRNA library screening [1]. A critical role for cyclophilins (CyPs) in HCV replication was first suggested by the direct antiviral effect of cyclosporine A (CsA) [2], [3]. We and others then identified CyPA as the main CyP isoform that serves as an essential cofactor for HCV infection [4], [5], [6] and the principal mediator of CsA resistance [4], [5]. In addition, the peptidyl-prolyl isomerase (PPIase) motif of CyPA has been found to be important for HCV replication [5], [6], [7]. Although CyP inhibitors are currently being studied in clinical trials as a novel class of anti-HCV drugs, the viral target of CsA and the substrate of CyPA's PPIase activity are not well characterized. Nonstructural proteins 5B (NS5B), NS5A, and NS2 have all been proposed to be potential targets of CyPA [8], [9], [10], [11], [12]. In addition, resistance-mapping studies using subgenomic replicons have so far generated a fragmented picture of determinants of CsA susceptibility [5], [9], [12]. In this study, we isolated a mutant JFH-1 full-length virus that escaped the inhibition by shRNAs targeting CyPA. Characterization of the mutant virus revealed a critical dipeptide motif and several surrounding prolines in domain II of NS5A to be the principal modulators of CyP dependence and CsA sensitivity, in a cell culture model of HCV infection (HCVcc) [13], [14], [15], [16].
Although CsA-resistant replicons have been isolated [5], [9], [12], attempts to obtain CsA-resistant full-length HCVcc had been unsuccessful. Because we previously observed a potent block of HCVcc infection of Huh-7.5 cells expressing a small-hairpin RNA (sh-A161) that downregulated CyPA expression by more than 90% [4], we devised a CoFIM selection for CyPA using this shRNA to obtain a HCV mutant that was less dependent on CyPA. In the first selection, we infected Huh-7.5 cells with JFH-1 at a low multiplicity of infection (MOI<0.01) and cultured the cells until 100% of the cells became positive for HCV staining. We reasoned that the relatively long period (10–14 days) of infection spreading time might maximize the diversity of the HCV quasispecies and increase the number of preexisting mutations. We then transduced into the infected Huh-7.5 cells a lentiviral vector expressing sh-A161 (Figure 1A). After a 2-wk antibiotic selection of shRNA-expressing cells, more than 95% of the cells cleared viral infection as a result of suppression of CyPA, but a small percentage (<5%) of cells remained positive for HCV core staining. Continued culturing of these cells for one more week resulted in a population of cells that were 100% infected despite efficient CyPA knockdown, indicating the emergence of a mutant virus that could replicate efficiently with significantly lower levels of CyPA. Viral particles produced from these selected cells were collected and then used to infect Huh-7.5 cells expressing shRNAs directed at firefly luciferase (sh-Luc), CyPA (sh-A161), or CyPA and CyPB (sh-Broad). Although wild-type (WT) JFH-1 could only infect the sh-Luc cell line, the selected virus, designated J-LA, infected all three cell lines with high efficiency (Figure 1B). These results indicated that J-LA was less dependent on CyPA and did not develop the ability to use CyPB as a substitute cyclophilin in the presence of CyPA knockdown. We then measured the CsA sensitivity of JFH-1 and J-LA and observed a greater than 16-fold shift of CsA sensitivity (Figure 1C), consistent with our previous finding that the CsA resistance was correlated with reduced dependence on CyPA in the setting of a subgenomic replicon [4]. The reduced dependence on CyPA was specific, as J-LA virus was not able to infect Huh-7.5 cells containing a shRNA targeting either one of two other essential HCV cofactors (Phosphatidylinositol 4-kinase alpha and Occludin-1) (Figure 1D) [17], [18], [19], [20], [21], [22], [23], [24], [25].
Three additional independent CoFIM selections (#2, 3, 4 in Table 1) with different parameters produced additional preparations of mutant viruses that were capable of infecting CyPA knockdown cells. The order of HCV and shRNA introduction into the cells was reversed in two of the experiments to reveal whether the same mutant viruses can emerge with preexisting selection pressure. Interestingly, the process took much longer (6 rather than 3 wk) to produce sh-A161-resistant virus when JFH-1 RNA, transcribed in vitro with T7 phage polymerase from a DNA plasmid as template, was electroporated directly into sh-A161 cells, probably because of the lack of diversity and preexisting mutations in the input RNA. Sequence analysis of these mutant viruses revealed related mutations in a dipeptide motif (D316 and Y317) in the domain II of NS5A in all four mutant viral samples. In addition, an I31T mutation in E2 was identified in three of the four selections, whereas the remaining selection revealed a T33A mutation, also in the E2 protein (Table 1). We engineered mutant viruses containing either the E2 mutation I31T or the NS5A mutations and tested their ability to infect sh-A161 cells. The mutant virus containing E2:I31T was not able to infect sh-A161 cells efficiently, and further passage of this virus in sh-A161 cells resulted in the emergence of the NS5A:Y317N or NS5A:Y317H mutation (#5 in Table 1). These data suggested that the E2:I31T mutation was not able to confer reduced dependence on CyPA, consistent with a recent report that this mutation increased infectivity of HCVcc by reducing lipoprotein association with viral particles, a process not known to involve CyPA [26]. The NS5A mutations, however, did confer reduced CyPA dependence on WT virus when engineered into either the JFH-1 (Figure 2A) or the J6-JFH background (Figure S1). Neither WT virus was able to replicate in sh-A161 cells, but mutant viruses harboring D316E, Y317N, or both mutations replicated efficiently in these cells. While the single mutants replicated 5- to 10-fold less in sh-A161 cells than in sh-Luc cells, a combination of the two generated a genome (DEYN) that replicated several fold more efficiently in sh-A161 cells than in the control cells (Figure 2A). The same pattern was observed when the viral genomes were introduced by means of infection rather than electroporation of RNA: the DEYN virus, produced from either sh-Luc or sh-A161 cells, infected sh-A161 cells several fold more efficiently than it did sh-Luc cells, and the single mutants again had an intermediate phenotype (Figure 2B). Of note, approximately same amount of viruses were produced by WT and the mutant genomes in sh-Luc cells (Figure 2C), indicating that these mutations do not significantly impact viral assembly. Furthermore, sh-A161 cells replicating DEYN mutant RNA showed no defect in producing viral particles (Figure 2C, stripe bar) that were able to infect naïve sh-A161 cells (Figure 2B, stripe bar), suggesting that the DEYN virus is capable of completing the entire viral life cycle efficiently in this CyPA knockdown environment. In addition, other selected mutations (Y317H, Y317R) at the Y317 position also conferred reduced CyPA dependence on the J6-JFH virus (Figure S1). Finally, the double mutant (DEYN) exhibited an approximately 20-fold resistance to CsA treatment but remained sensitive to IFN treatment (Figure 2D, Figure S2). Together, these data identify NS5A as a major viral target of CyPA and demonstrate that the DY motif in NS5A is a major determinant for CyPA dependence and CsA susceptibility.
A near-perfect repeat of the peptide containing the DY motif and the four prolines occurs immediately downstream in the JFH-1 NS5A protein (Figure S3A), but we did not recover any mutations in this second DY motif in any of our screens. Moreover, mutation of this second DY motif (D329EY330N) resulted in a replication-deficient virus (Figure S3B). These data suggest a functional divergence of the two repeated DY motifs.
Several independent studies recently demonstrated that the PPIase motif of CyPA is essential for HCV replication [5], [6], [7], suggesting the presence of critical proline residues that serve as substrates for CyPA. We reasoned that the DY motif, which locates to a major CyPA-binding peptide in this domain [11] but not proline themselves, probably confer CsA resistance through the proline substrates of CyPA, and that consequently the mutation of the relevant prolines would affect DEYN's ability to replicate in CyPA knockdown cells. To identify such proline residues, we changed the individual prolines (14 total, Figure 3A) in the domain II of NS5A into alanines, both in the WT and the DEYN background of the J6-JFH genome (Table 2). The mutant genomes were then electroporated into both sh-Luc and sh-A161 cells for measurement of replication capacity. The replication phenotypes of these mutants fell into three categories. Mutations of the first group of prolines had no effect on either the WT's or the DEYN's ability to replicate in both cell lines; i.e., a proline mutant in the WT background would replicate in sh-Luc cells but not in sh-A161 cells, whereas the same proline mutant in the DEYN background would replicate in both cells with high efficiency (Figure 3B). Mutations of the second group of prolines had minimal effect on the WT virus but reversed the replication capacity of the DEYN virus in sh-Luc and sh-A161 cells (Figure 3C); i.e., the DEYN + proline mutant replicated less efficiently in sh-A161 cells than in the control cells. Mutation in the third group, which included only one proline (P310), had the most profound effect: P310A reduced the replication capacity of the WT virus to the level of the GND pol- virus in both cells (Figure 3D left panel, 3E), it also completely abolished DEYN's ability to replicate in sh-A161 cells (Figure 3D, right panel); moreover, it reduced the replication capacity of DEYN virus in the control cells more than any other single proline mutant (Figure 3D, right panel). These data suggested a functional interaction between the DY motif and the four prolines in groups II (prolines 315, 319, 320) and III (proline 310) in determining CyP dependence and CsA susceptibility, and a possible use by CyPA of multiple prolines as substrates on the NS5A protein. The importance of prolines 310 and 315 in HCV replication was further supported by results from double and quadruple mutants in the DEYN background. DEYN was not able to rescue the replication of either the quadruple mutant (with all 4 prolines mutated to alanine) or a double mutant containing P315A and P310A (Figure 3F).
To determine whether the DEYN mutations affect the CsA sensitivity of NS5A's binding to CyPA, we first established a biochemical binding assay for the CyPA-NS5A interaction. Recombinant forms of WT and three mutant CyPA proteins that each contained a single mutation in the active site of the PPIase (Figure 4A) [27] were purified as hexahistidine-tagged proteins and mixed with lysates of JFH-1 infected cells for an affinity pull-down experiment. WT CyPA efficiently bound to full-length NS5A in this assay. The F113A mutant retained approximately 10–20% of NS5A-binding, while mutations R55A and F60A completely abolished NS5A-binding by CyPA (Figure 4B). These results demonstrated the importance of CyPA's PPIase motif in NS5A-binding. We next added increasing amounts of CsA to the binding reaction and observed that the CyPA-NS5A interaction is sensitive to CsA. However, DEYN mutations did not significantly alter the CsA-sensitivity of this interaction (Figure 4C). These data argue against the possibility that the DEYN mutations conferred CsA resistance to the virus by altering CyPA-NS5A binding. We also analyzed the possible connection between CyPA-independence and the phosphorylation status of NS5A. In the biochemical binding assay, the CyPA interacted with both forms (hyper- and basal-phosphorylated) of NS5A to approximately the same extent (Figure S4A). In addition, neither DEYN mutations nor CsA treatment changed the relative ratio of these two phosphorylated forms (Figure S4B). We conclude from these experiments that CyPA's action is probably unrelated to the phosphorylation status of NS5A. We next determined whether the CsA resistance conferred by these mutations was related to altered cleavage kinetics at the NS5A-NS5B junction, as suggested by a previous study [5]. The DEYN mutations in JFH-1 were transferred into a polyprotein expression plasmid, which was then used to express HCV proteins for the measurement of cleavage kinetics. These mutations produced no significant effect on the rate of NS5A-NS5B cleavage (Figure 4D), suggesting that the DEYN mutant identified here uses a mechanism distinct from those of the previously reported cleavage mutants (NS5A:V464A; NS5A:V464L) that could also replicate, albeit to a lower level, in a CyPA-knockdown cell line [5].
The domain II of NS5A has been shown to be largely disordered [28]. We first determined whether this domain, which contains the DEYN mutations and the putative proline substrates, had the potential to adopt different structural conformation. Circular dichroism analysis of recombinant NS5A D2 resulted in spectra indicative of a largely unordered structure (Figure 5). When CD melting experiments were performed, however, an isodichroic point between 209 nm –211 nm was observed in both the WT and the mutant D2 (Figure 5A, B), indicating the existence of at least two different structural conformers in this domain for both proteins. The CD spectra of WT and DEYN mutant proteins, however, were very similar in these melting experiments, indicating that there is no gross structural difference between WT and mutant D2 proteins at the resolution of CD analysis. We then analyzed a 20-mer peptide that contained the DY motif and the four relevant prolines (G304-E323) (Figure 5C) to detect any difference in local conformation introduced by the mutations at this putative CyPA substrate site. Indeed, the spectral amplitude corresponding to the DEYN mutant was more negative and shifted to longer wavelengths compared to the WT peptide (Figure 5C), indicating that the DEYN peptide may adopt a more extended structure with less turn characteristics relative to the WT peptide. To pinpoint the amino acid residues contributing to this structural difference, we characterized each peptide using two dimensional NMR spectroscopy. Specifically, we used 2D NOESY spectra, in which cross peaks arise between protons that are close in space (generally <5 Å). It is well known that helical or turn and β-strand or extended conformations in peptides yield distinct NOE patterns based on 1H-1H distances in these secondary structures. While the WT peptide adopts a mostly random coil structure, NOE data clearly indicated a more extended structure between residues 316 and 318 for the DEYN peptide. Specifically, we observed sequential αN and NN NOEs for each non-proline residue in the WT peptide, indicating that this peptide adopts a conformational ensemble containing both extended and turn-like conformations (Figure 6A). Interestingly, the amide resonances of residues A311 and W312 exhibited large up-field shifts for a small, random-coil peptide, consistent with the unusual resonance shifts of the A311 and W312 residues observed in a full-length D2 construct [11], suggesting that the peptide may adopt similar conformations as the NS5A protein. We observed sequential αN NOEs for each non-proline residue in DEYN and sequential NN NOEs between residues A311-R314 but did not see sequential NN NOEs between E316-N318 (Figure 6B). This suggests that the DE and YN substitutions bias the peptides towards more extended, β-strand-like conformations compared to the WT sequence.
We also determined if CyPA acted on these peptide substrates differently because of the structural difference that we observed. Adding either the WT or the DEYN peptide to 15N-labeled CyPA significantly perturbed the chemical shifts for the residues in the same regions of CyPA (Figure S5) indicating that these two peptides bound to the same region of CyPA, consistent with equivalent binding in the biochemical binding assay. Adding a catalytic amount of CyPA (∼1∶10) to the peptides caused chemical shifts in the following residues: W312, A313, D/E316,Y/N317, E323 (Figure 6, top). With the exception of E323, the peptide residues that shifted upon addition of CyPA were all located between P310 and P320, further supporting the notion that the segment containing these four prolines is a major CyPA substrate relevant for HCV replication. Moreover, we observed a sequential NN NOE between D316-Y317 in WT peptide with or without CyPA, indicating the presence of a turn formation between these residues that is not dependent upon CyPA (Figure 6A, bottom). In contrast, NOE analysis of DEYN peptide showed the absence of an E316-N317 NN NOE, suggesting a reduced tendency for turn formation in the absence of CyPA. Adding a catalytic amount of CyPA to the DEYN peptide induced the formation of a similar turn between residues E316-N317 as evidenced by the presence of an E316-N317 NN NOE (Figure 6B, bottom). These data further validate the DY motif as a major modulator of CyPA's action.
The putative peptide substrate of CyPA, centered around the DY motif, is the most conserved region in NS5A domain II of GT 1a, 1b, and 2a (Figure 7A). To determine whether this major determinant of CsA sensitivity that we identified in GT 2a is functionally conserved across genotypes, we conducted a CsA-resistance selection in a GT 1a replicon. Using a GT 1a replicon labeled with red fluorescent protein (RFP), we isolated resistant cells with high replication levels in the presence of CsA, a NS5B inhibitor (200 nM of HCV-796), or an equivalent volume of solvent (DMSO) (Figure 7B). The sensitivity of the selected cells to three antiviral molecules, the protease inhibitor BILN-2061, HCV-796, and CsA, was measured (Figure 7C). All cell lines were equally sensitive to BILN-2061 and those selected in HCV-796 were 74-fold less sensitive to HCV-796 but remained as sensitive to CsA. In contrast, the replicon cells selected in the presence of CsA exhibited specific resistance to CsA, and the levels of resistance directly tracked the concentration of the compound applied during selection.
The CsA-resistant replicon was also cross resistant to CsD, a CsA derivative that is a potent inhibitor of CyPA with dramatically reduced calcineurin binding [29], indicating that the resistant replicon overcame CyPA inhibition rather than calcineurin recruitment (Figure 7D). To determine the source of CsA resistance, we tested the CsA sensitivity of a naïve GT 1b replicon in the CsA-resistant cells. A WT 1b replicon with a Renilla luciferase reporter was transfected into the GT 1a replicon cells selected at 3 µM CsA. The sensitivity of the 1b replicon to CsA is determined by monitoring of the luciferase signal. The 1b-luciferase replicon was equally sensitive to CsA in WT or CsA-resistant 1a-RFP replicon cells (Figure 7E). This result demonstrated that the CsA resistance was driven by mutations in the replicon and not in the host cell.
A D320E mutation, which is the equivalent of the D316E mutation in JFH-1, was observed in CsA-resistant cell lines selected with both 2 µM (50% of the bacterial clones sequenced had the mutation) and 3 µM CsA (100% of the bacterial clones sequenced had the mutation) but not in the control cell lines. When the D320E mutation was engineered into a WT GT 1a replicon, it conferred a statistically significant 2.7-fold shift in EC50 of CsA but no shift in the EC50 of the control antiviral compound 2′CmeA (Figure 7F). In addition, the D320E 1a replicon was able to replicate in sh-A161 cells, albeit to a lower level than in the control cells, while the WT 1a replicon could not (Figure 7G). Replication in control cells was not affected by the DE mutation.
We next determined whether the DEYN mutations could also increase CsA resistance in the GT 1b background. We have previously shown that a point mutation in the NS5B gene of a GT 1b replicon (NS5B:I432V) could confer a low level (1.5 fold) of CsA resistance [9]. Combining DEYN with the NS5B mutation generated a mutant replicon (DEYNI432V/Con1) that was significantly more resistant (>4 fold) (Figure 8A). In addition, either DEYN alone or DEYN combined with NS5B:I432V conferred the ability of the Con1 replicon to efficiently replicate in sh-A161 cells in a colony-formation assay (Figure 8B). These data, together with recent reports of the D320E mutation in CsA-resistant GT 1b replicons [30], [31], strongly suggest that the DY motif represents key residues in a functionally conserved determinant of CyPA-dependence and CsA susceptibility across genotypes.
Using a newly developed genetic-selection approach, we obtained for the first time a full-length, CsA-resistant HCV genome that can replicate more efficiently in CyPA-knockdown cells than in control cells. We also identified a specific dipeptide motif that is a major controller of CyPA dependence and CsA susceptibility. Mutation of this DY motif not only renders HCVcc less dependent on CyPA and less sensitive to CsA treatment but also changes the conformation of the NS5A peptide that binds to CyPA. In addition, four prolines that represent putative substrates of CyPA were mapped to a contiguous peptide surrounding the DY motif with the most critical proline located six amino acids upstream of the aspartic acid residue.
Increased RNA binding by HCV replicase and delayed NS5A-NS5B cleavage have been proposed as potential mechanisms of CsA resistance [5], [30], correlated with reduced dependence on CyPA [4]. The mechanism of the DEYN mutations, however, appears to be distinct from either of these, as no difference was observed in either CyPA binding or NS5A-NS5B cleavage. The functional mapping and biophysical characterization of the putative CyPA substrate on NS5A D2 suggests the following molecular mechanism for DEYN-conferred CsA-resistance and reduced CyPA dependence: The WT NS5A D2 structure is largely disordered, containing a mixture of turn-like and more extended conformations at the CyPA substrate site. Cis Xaa-Pro peptide bonds further reduce the population of extended conformers in this region. CyPA catalyzes cis-trans peptide bond interconversion in NS5A, affecting the population of turn and extended conformers. We propose that the latter are required for the normal biological function of NS5A as part of a replicase component. DEYN mutations cause the structural shift toward this more extended form, reducing the dependence on CyPA to induce such a structure. Of note, aspartate, present in the WT sequence, is a strong N-capping residue for helix formation and mutation to glutamate reduces the helical propensity for these residues, increasing the population of structure conformers with less turn features in the DEYN mutant compared to the WT. It is interesting that CyPA introduces turn formations in the DEYN peptide, presumably through mass action due to the larger population of the extended conformers in this mutant.
The DEYN mutations confer the most marked CyPA-independent phenotype reported thus far and do so in a full-length virus with all viral proteins present. NS2 has recently been shown to increase CsA sensitivity in either the full-length or a replicon background [5], [10]. We did not identify any mutations in the NS2 region in our screens, and the DEYN mutation conferred resistance both in the presence (full-length) or the absence (NS3-NS5B subgenomic replicon, Figure S6) of NS2. In addition, the resistance-conferring mutations that we previously identified in NS5B of a GT 1b replicon [9] were not required for DEYN to confer reduced dependence on CyPA, suggesting that the NS5A DY motif is the primary regulator of CyPA dependence and CsA susceptibility. Note that residual amounts of CyPA exist in Huh-7.5 sh-A161 cells [4], and the DEYN mutant virus is inhibited by CsA at high concentrations (>8 µg/ml), suggesting that either the remaining CyPA or other minor contributors to CsA sensitivity such as other CyP isoforms expressed at low levels [4], [32] facilitate the replication of the DEYN virus in sh-A161 cells. Of note, sh-A161 only inhibits the expression of CyPA, but not other CyP isoforms (Figure S7). To address definitively the issue of CyPA independence, a CyPA-knockout cell line similar to the one generated by Braaten et al. [33] that is also susceptible to HCV infection may be necessary. In this regard, we propose to apply the CoFIM designation to viral mutants with either complete independence or reduced dependence on cellular cofactors.
Proline residues in NS5A that are important for HCV replication have been identified previously, both in GT1b [34] and GT2a [35]. In a cell-culture adapted GT1b replicon, two prolines (P314 and P324) in domain II that are equivalent to P310 and P320 identified in this study were found to be critical for colony formation of the replicon. In addition, several residues in close proximity to these prolines were also essential for replication [34]. It would be interesting to determine whether mutation of these residues can influence CyPA-binding or the related conformation change of NS5A. Finally, a polyproline motif (PP2.1) that locates to the low-complexity sequence [36] between domains II and III of NS5A can regulate both RNA replication and viral assembly [35], raising the possibility of additional CyPA substrates outside NS5A domain II. We speculate that the isomerization of one or more of these prolines of NS5A by CyPA is required for the proper function of NS5A, which is currently not understood, in the life cycle of HCV.
CsA sensitivity appears to vary in different genotypes [37], [38]. An independent selection with a GT 1a replicon identified D320E, which is equivalent to D316E in JFH-1, as an important contributor to CsA resistance. Similar observations have been obtained with GT 1b replicons whereby the proline residues surrounding the D320 residue are also important for replication of the 1b genotype [30], [31], [34]. Analysis of the HCV sequences deposited at the European HCV Database (euHCVdb) revealed that the frequency of changes at both the DY motif and the four prolines in natural isolates are low (Table 3). The most variable position is NS5A 316 and its equivalent position, which had substitutions in only 22 out of 2764 isolates. And although there are some isolates that harbor either the D to E or the Y to N change at the NS5A DY motif, we did not find a single natural isolate that contains both changes. To further address the in vivo relevance of the DY motif, it would be important to determine whether mutations at this site can emerge from patients in clinical trials treated with CsA derivatives such as DEBIO-025, NIM811, or SCY-635 [39], [40], [41].
So far as we know, this is the first report of an shRNA-based selection of viral mutants with reduced dependence on a cellular cofactor. The general steps of Co-FIM selection starts with the identification of a cellular cofactor, suppression of which would lead to viral inhibition. The next step would be the selection of viral variants that can replicate in the constant pressure of shRNA-mediated knockdown of the cofactor. This can be done either by introducing viral particles or genomes into stable knockdown cells (selection # 3, 4, and 5) or transducing infected cells with lentiviral vectors expressing shRNAs (selection #1 and 2). Sequencing and identification of mutations in the selected viruses are then followed by standard reverse genetics approaches to verify the effect of the mutations. Obviously, a cellular cofactor that is also essential for cell survival would not be amenable for this selection. Also, it might be very difficult for a virus to bypass entire cellular pathways that are important for its replication. For HCV, these may include membrane-reorganizing factors and proteins involved in lipid metabolism, to name a few.
In addition to target identification, the CoFIM screen may also reveal novel cellular pathways involved in the viral life cycle and identify cofactor hierarchy. For example, cross-resistance to knockdown of multiple cofactors may place distinct factors in the same pathway for viral infection and reveal previously unidentified interactions. The strong agreement of our shRNA-based results with those from independent compound-based selection indicates that CoFIM screening may be applicable to a wide range of cellular cofactors in the absence of small-molecule inhibitors or knowledge of mechanism.
Huh-7.5 cells and J6-JFH (p7-RLuc) were provided by Charles M. Rice. JFH-1 constructs were provided by Takaji Wakita. Huh-7.5 sh-Luc, sh-A161, and Huh-7-Lunet/T7 cells have been described previously [4], [5]. The genotype 1b–PI-luc replicon, used for the supertransfection experiment, and genotype 1a replicons have been described previously [42]. Monoclonal antibodies against JFH-1 NS3 and NS5A were made in house. Compounds BILN-2061 and 2′CmeA were purchased from Acme Biosciences (Palo Alto, CA), HCV-796 from Curragh Chemistries (Cleveland, OH), IFN-α from Sigma-Aldrich (St. Louis, MO), and CsD from Enzo Life Sciences International, (Plymouth Meeting, PA).
Huh-7.5 cells were infected with JFH-1 virus at MOI of <0.01 for 4 h. The input virus was removed, and the cells were cultured for 10–14 days, during which a small sample was removed every 3 days for immunofluorescence staining for HCV core and NS3 until all cells were positive for HCV. The infected cells were then transduced with lentiviral vector sh-A161 and were selected in the presence of 1.2 µg/ml puromycin for 3 wk. We first examined the stable cells for CyPA/CyPB with western blotting to confirm knockdown and then examined them for presence of HCV-positive cells by NS3-staining. The cells were then cultured for another week, during which a small sample was removed by every 2 days for determination of the percentage of infected cells. Supernatant containing mutant viruses was collected once all the cells again became 100% infected.
Huh-7.5 cells stably expressing sh-A161 were challenged with JFH-1 viral particles or genome RNA transcribed in vitro. The cells were then cultured for 3–6 wk, during which a small sample was removed every 3 days for NS3 staining. Supernatant containing mutant viruses was collected once >90% of the cells became positive for HCV NS3.
After 40 days of serial passaging in media supplemented with increasing concentrations of CsA, cells fluorescing in the top 15% of the population were isolated by flow cytometry. After outgrowth in the high concentration of CsA, RNA isolation, RT-PCR, and sequencing were performed by Tacgen (Hayward, CA). To determine drug sensitivity, we seeded stable replicon cells in 96-well plates at a density of 5×103 cells per well or transiently transfected replicon cells at 1.3×104 cells per well and treated them with compounds for 3 days, after which the cells were assayed for luciferase or NS3 activity [43]. Activity levels were converted into percentages relative to the untreated controls (defined as 100%) and data were fit to the logistic dose response equation y (a/(1+(x/b)c) with XLFit4 software (IDBS, Emeryville, CA).
Binding of recombinant His-CyPA to NS5A was performed using total lysate collected from Huh 7.5 L-Luc cells containing either the JFH-1 virus or full-length JFH-1 genomic replicons (pFGR-JFH-1 WT or pFGR-JFH-1 DE/YN). Lysates, generated by lysing 8×105 cells in 600 µl of IP buffer (50 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM EDTA, 0.5% NP-40, 1 mM PMSF, 1 mM DTT, 1x protease inhibitor cocktail), were incubated with His-select Nickel affinity gel (Sigma) prior to binding. Three-hundred micrograms of His-CyPA protein was preincubated with increasing amounts of cyclosporine A for 30 min at 4°C. The recombinant protein was then mixed with 50 µl of cleared lysate and incubated at 4°C or 1 hour. Freshly equilibrated resin was then added to each sample and allowed to bind for 30 min at 4°C. Beads were pelleted at 5000×g for 1 min and the supernatant removed. Three washes were performed using wash buffer (50 mM sodium phosphate, 10 mM imidazole, 250 mM sodium chloride, pH 8.0) followed by elution with wash buffer containing 250 mM imidazole. The flowthrough and eluted (CyPA-bound) fractions were then analyzed by SDS-PAGE and western blotting.
RT-PCR and sequencing. Sequences of primer sets used in RT-PCR reactions to amplify JFH-1 genome are available upon request. PCR products were sequenced directly or as cloned inserts in a pCR2.1-TOPO vector (Invitrogen, CA).
These methods have been described previously [4], [42].
The metabolic labeling and immunoprecipitation were performed essentially as previously described [5]. Huh-7 Lunet/T7 cells were transfected with subgenomic NS3 to NS5B JFH-1 expression constructs derived from the WT or DEYN mutant or with the empty vector (pTM1-2). After 22 h, cells were starved for 1 h in methionine/cysteine-free medium and then pulse-labeled for 1 h with [35S] methionine/cysteine (150 µCi/ml Express Protein labeling mix). Cells were lysed immediately (0) or incubated with nonradioactive medium for 1 or 2 additional hours (chase). Immunoprecipitation was performed with a monoclonal NS5A-specific antibody. Samples were processed by SDS-PAGE (8%), followed by fluorography and autoradiography.
Site-directed mutagenesis was carried out with a QuikChange kit (Stratagene, CA). Sequences of mutagenesis primers for all the prolines and the DY motif are available upon request. Fragments containing the mutated NS5A sequence were also cloned into a pTM NS3-3′ plasmid to produce subgenomic NS3 to NS5B JFH-1 expression constructs containing the corresponding mutations (pTM NS3-3′DEYN JFH-1).
One microgram of in vitro transcribed WT and mutant RNAs were electroporated into 4×106 Huh-7.5 cells stably expressing shRNAs. For the 1a replicons, replication was measured transiently at 4, 48, 72 and 96 hrs post-electroporation. The 1b replicons were allowed to form colonies after three weeks of selection and the resulting colonies were stained with crystal violet. CsA treatment of the stable 1b replicon cell lines were performed as described previously [9].
Luciferase assays was performed according to the manufacturer's instructions for the Dual-Glo Luciferase Assay System from Promega (Madison, WI) with two exceptions: the Luciferase Assay System was used for the Genotype 1b supertransfection experiments, and the Renilla Luciferase Assay System was used for the Genotype 1a D320E mutant. Of note, the J6-JFH-1 reporter virus carries a Renilla luciferase gene that is not a target of sh-Luc, which is directed against the firefly luciferase gene.
HCV core ELISA was performed according to the manufacturer's instructions for the HCV Antigen ELISA kit (Ortho-Clinical Diagnostics, Japan).
Circular dichroism data were collected with a Jasco J-810 spectropolarimeter (Jasco Inc., MD) equipped with Peltier temperature control. The spectra were recorded as averages of 5 scans in 20 mM sodium phosphate, 50 mM NaCl, pH 6.5 buffer at 0.1-nm resolution and 0.05-cm path length. The wt and DEYN NS5A-D2 protein (6.3 µM) and peptide (50 µM) sample concentrations were determined by absorbance at 280 nm and the following extinction coefficients: 14,105 M−1/cm−1 and 15,595 M−1/cm−1 for DEYN and wt protein and 5500 M−1/cm−1 and 6990 M−1/cm−1 for DEYN and wt peptides, respectively. The NS5A-D2 thermal melt data were collected at a scan rate of 15°C/h with a 10-min equilibration time prior to collection of wavelength spectra at 5, 15, 25, 40 and 60°C. All data were acquired in at least triplicate, with baseline correction and no curve smoothing. The circular dichroism instrument was calibrated with ammonium (+)-camphor-10-sulfonate by a two-point calibration method described previously [44].
A synthetic WT peptide corresponding to residues 304-323 of NS5A (304GFPRALPAWARPDYNPPLVE323) and a peptide with the double mutations D316E and Y317N were obtained from NEO BioScience (Cambridge, MA). Purity was determined to be greater than 98% by high-performance liquid chromatography and mass spectrometry. Peptide NMR spectra were collected on 1 mM samples in aqueous buffer at 4°C on a Bruker Avance spectrometer operating at 700 MHz. Total correlation (TOCSY) spectra were collected using a clean-DIPSI mixing sequence (100 ms mixing time) [45] with excitation sculpting solvent suppression [46] as 2048×256 complex matrices with spectral width of 7692.3 Hz in both dimensions. NOESY spectra (400 ms mixing time) were collected under identical spectrometer conditions using excitation sculpting solvent suppression [46]. The indirect dimensions were extended during processing using linear prediction or covariance techniques [47], [48]. The expression and purification of 15N-labeled human CyPA was carried out as previously described [11] with the additional purification step of high-performance liquid chromatography (Superdex 75, GE Healthcare). 15N-labeled CyPA (∼0.5 mM) was exchanged into NMR buffer (50 mM NaH2PO4/Na2PO4, pH 6.3, 40 mM NaCl, 2 mM EDTA, and 1 mM DTT), and peptide was added sequentially to produce the desired final molar equivalents. 2D 1H,15N-HSQC NMR spectra [49] were acquired at 22°C on a Bruker Avance spectrometer operating at 500 MHz 1H frequency. 2D spectra were collected as 1024×200 complex matrices with spectral widths of 833.33 Hz and 2500 Hz in the 1H and 15N dimensions, respectively, from 16 scans per complex t1 point. Spectra were processed with nmrPipe [50] and analyzed with NMRView [51].
|
10.1371/journal.ppat.1005178 | Trans-generational Immune Priming Protects the Eggs Only against Gram-Positive Bacteria in the Mealworm Beetle | In many vertebrates and invertebrates, offspring whose mothers have been exposed to pathogens can exhibit increased levels of immune activity and/or increased survival to infection. Such phenomena, called “Trans-generational immune priming” (TGIP) are expected to provide immune protection to the offspring. As the offspring and their mother may share the same environment, and consequently similar microbial threats, we expect the immune molecules present in the progeny to be specific to the microbes that immune challenged the mother. We provide evidence in the mealworm beetle Tenebrio molitor that the antimicrobial activity found in the eggs is only active against Gram-positive bacteria, even when females were exposed to Gram-negative bacteria or fungi. Fungi were weak inducers of TGIP while we obtained similar levels of anti-Gram-positive activity using different bacteria for the maternal challenge. Furthermore, we have identified an antibacterial peptide from the defensin family, the tenecin 1, which spectrum of activity is exclusively directed toward Gram-positive bacteria as potential contributor to this antimicrobial activity. We conclude that maternal transfer of antimicrobial activity in the eggs of T. molitor might have evolved from persistent Gram-positive bacterial pathogens between insect generations.
| In some insects, the immunological experience of mothers is transferred to their otherwise naïve offspring, protecting them against infection. Such a maternal effect has likely evolved from selective pressure imposed by the persistence of some microbial pathogens in the environment between insect generations. If microbes are not transmitted vertically from mother to the offspring, only those able to survive in the external environment have the highest probability to infect the offspring. Therefore, early levels of immune protection transferred by mothers to their offspring might be specific of these microbes. In this study, we found that enhanced levels of antimicrobial activity in the eggs of immune challenged females of the mealworm beetle, Tenebrio molitor, were only active against Gram-positive bacteria, whatever the microorganism used for the maternal challenge. Furthermore, immune challenged females with fungi rarely transferred antimicrobial activity to their eggs. The analysis of the proteins conferring antibacterial activity in the eggs of bacterially immune-challenged mothers revealed the presence of tenecin 1, an antibacterial peptide active against Gram-positive bacteria only. These results suggest that maternal transfer of antimicrobial activity in the eggs in T. molitor might have evolved from the persistence of Gram-positive bacterial pathogens between insect generations.
| Maternal effects are of paramount importance for offspring fitness when mothers adjust the phenotype of their offspring to match the environment that they are likely to experience [1]. For instance, in many different groups of animals, the immunogenic experience of the mother is transferred to otherwise “naïve” offspring and can protect it against infection [2, 3]. Maternal transfer of immunity has been particularly well studied in vertebrates, in which infected females can transfer specific antibodies to the offspring via the placenta and milk in mammals during lactation, or via the egg yolk in birds, reptiles and fish [3, 4]. As newly born vertebrates have limited abilities to produce antibodies, this maternal transfer of immunity protects them while their own immune system becomes mature. In addition, it may further serve as signal to up-regulate the immune system of the offspring later on [3, 4].
Invertebrates lack the antibodies that vertebrate females transfer to their offspring. However, maternal transfer of immunity, also referred as trans-generational immune priming (TGIP), occurs in invertebrates too [5], suggesting that it has to be achieved by other, yet unknown, mechanisms. The effects of TGIP have been revealed through enhanced levels of immune activity and/or an increased survival to infection in primed offspring. TGIP has been found for a variety of arthropods against multiple classes of microbes and parasites [6–19]. Its trans-generational effects can be found across all life-stages of the protected progeny: from laid eggs [10, 16, 17], during the larval development [9, 11, 14, 18] and persisting even until the adult stage [12, 13,15], although its manifestation may depend on the ontogenetic stage of the offspring [18]. Furthermore, the immune protection provided to the offspring can exhibit variable levels of specificity.
Specificity measures the degree to which TGIP discriminates parasites on different levels of relatedness between the parental infection and that of the offspring. Previous studies suggest that a wide spectrum, from cross-reactive (non-specific) [6] to highly specific TGIP [7, 13] occurs in arthropods. So far specific TGIP has been demonstrated only in adult offspring and its occurrence at earlier ontogenetic stages is currently unknown. As immune defences vary with developmental stage in several insect systems [20–23], specific TGIP is also likely to vary with the ontogeny of the primed offspring. This might be particularly critical for eggs born to mothers soon after the maternal infection, as they are likely first exposed to the maternal pathogens. However the occurrence of specific TGIP in eggs is currently unknown.
In the bumblebee Bombus terrestris and in the mealworm beetle Tenebrio molitor mothers immune challenged with bacteria produce eggs containing enhanced levels of antimicrobial activity [10, 16, 17]. In insects, antimicrobial activity across all the life stages (including in eggs) generally relies on the presence of antimicrobial peptides (AMPs) and lysozymes [24–26]. Some of these molecules have a rather large spectrum of antimicrobial activity whereas others are more selective towards Gram-positive, Gram-negative or fungi [24]. Whether antimicrobial activity found into the eggs protects specifically the eggs from the pathogen that previously infected the mother is not known. However, for both bumblebees and the mealworm beetles, females and their eggs are sharing the same environment and are therefore likely exposed to the same pathogens. In this context, a specific protection in the eggs against the prevailing pathogen in the environment experienced by the mother should be particularly advantageous. Hence, within the range of specificity that AMPs and lysozymes convey to fight microbial infection, we may expect that antimicrobial activity in the eggs to be specific to the microbe that previously immune challenged the mother.
While the mechanisms through which eggs are immune protected by immune challenged mothers are currently unknown, we may propose three non-exclusive hypotheses, which support specificity of egg protection. First, immune challenged mothers may passively transfer antimicrobial peptides from their hemolymph to their eggs. Second, they may transfer mRNAs coding for immune related molecules. Under these two hypotheses, eggs should therefore exhibit a spectrum of antimicrobial activity similar to that of their mother. This might be two reasonable hypotheses since, in T. molitor, enhanced levels of antimicrobial activity are observed in eggs laid at the time where antimicrobial activity is detected in the hemolymph [17, 27]. Third, antibacterial activity found in the eggs might result from the expression of immune related molecules in the eggs themselves. Insect eggs, could be induced to produce immune effectors genes including antimicrobial peptides [25, 26, 28, 29] and it has been recently shown, in Galleria mellonella, that microbial immunogens that challenged mothers could be transported to ovaries and eggs where they could stimulate the expression of immune genes [29]. The production of antimicrobial factors in the eggs would differ according to the maternal challenge.
In this study, we first tested whether an immune challenge to female T. molitor affects levels of egg antimicrobial activity in a pathogen-specific manner. To this end we immune challenged females with a large range of inactivated microbial pathogens and tested the resulting antimicrobial activity of their eggs against several microorganisms, including those used to challenge mothers, using inhibition zone assays [30]. The microorganisms used were either related to one another as defined by kingdom (fungi versus bacteria), Gram type (Gram-positive versus Gram-negative) and species within the same genus. We then identified the molecular substances conferring antimicrobial activity in the eggs of bacterially immune-challenged mothers using a proteomic approach. Unexpectedly from our above hypothesis, we found that enhanced levels of antimicrobial activity in the eggs of immune challenged mothers were only active against Gram-positive bacteria but not the Gram-negative bacteria we tested, whatever the microorganism used for the maternal challenge. However, whereas maternal immune challenges with bacteria always resulted in increased levels of anti-Gram-positive activity in the eggs, no such results could be found for maternal challenges using fungi. Furthermore, the analysis of the proteins responsible for antibacterial activity in the eggs revealed the presence of tenecin 1, an antibacterial peptide known to show antibacterial activity against Gram-positive bacteria only [31]. This work suggests that pathogenic Gram-positive bacteria able to persist between generations are likely responsible of the evolution of TGIP in eggs of T. molitor.
We tested specificity of the maternal transfer of antimicrobial activity to the eggs by first focusing on the effects of maternal bacterial challenges on egg anti-bacterial activity. Eighteen to 23 virgin females (10 ± 1 day post emergence) per immune treatment received a single injection of a suspension of inactivated Gram-positive (Arthrobacter globiformis or Bacillus thuringiensis) or Gram-negative bacteria (Escherichia coli or Serratia entomophila) (“female treatment”) in saline solution. We chose these bacteria because they vary in their cell-wall components (DAP-type and Lys-type peptidoglycan; S1 Table, [32]) that are determinant for the expression of antimicrobial peptides (AMP) in insects [33], and because they are known to be either entomopathogenic (Bacillus sp., Serratia sp.) or ubiquitous (A. globiformis, E. coli). A group of control females was treated in the same way, but with the omission of microorganisms as a procedural control for effect of the injection (sham control mothers) and an additional group of non-injected females (naïve control mothers) was used as control. Females were then paired with a virgin and immunologically naïve male of the same age and allowed to produce eggs. Only eggs laid from day 3 to 7 post injection were collected to test their antibacterial activity because females are protecting most of their eggs at this period of time [17]. The antibacterial activity of egg extracts was tested against a range of bacteria that included those used for the maternal immune challenge (with the exception of S. entomophila replaced by Serratia marcescens, more often used for antimicrobial tests in laboratory) and Bacillus subtilis using standard zone of inhibition assays (“egg assay”) [30]. Egg antibacterial activity was first analysed by testing the probability of detecting a zone of inhibition in egg extracts for each female treatment and for each egg assay. In addition, among the egg extracts for which we detected an inhibition zone, we checked whether the diameter of this zone was different according to female treatment and egg assay.
We found no significant interaction between the maternal treatment and the egg assay on the probability of detecting a zone of inhibition (Generalized Linear Model—GLM: female treatment * egg assay: χ220, 503 = 1.37, p = 0.12), suggesting that the eggs were not better protected against the specific microorganism that females had previously encountered. However, it was influenced by both the female treatment (χ29, 523 = 26.1, p < 0.001) and egg assay (χ29, 523 = 83.01, p < 0.001, Fig 1). Females challenged with bacteria produced a higher proportion of protected eggs than both naïve and sham females (see Fig 1 for corresponding Tukey's HSD test results). Concerning the egg assays, A. globiformis and B. subtilis were the bacteria that were the most susceptible to the antibacterial activity transmitted to the eggs by the challenged females, whereas the probability of detection of a zone of inhibition when the eggs were exposed to B. thuringiensis was significantly lower (see Fig 1 for corresponding Tukey's HSD test result). No zone of inhibition could be detected on E. coli and S. marcescens, while we can detect anti-Gram-negative antibacterial activity in the hemolymph of similarly immune challenged adult beetles using the same method (S1 Fig).
Among the egg extracts for which we detected an inhibition zone, the size of the latter was influenced by the female treatment in interaction with the egg assay (Linear Model—LM: F16, 159 = 4.8, p < 0.001). However, there seems to be no support for specificity, since the egg extracts tested on A. globiformis did not produce larger zones of inhibition when they came from A. globiformis-challenged mothers than when they came from other mothers challenged with other bacteria species (Fig 2, see S2 Fig for corresponding test results). Interestingly, the egg extracts of B. thuringiensis challenged females were the ones responsible for the biggest zone of inhibition on A. globiformis (Fig 2). In the case of the egg extracts tested on B. subtilis and B. thuringiensis, the size of the zone of inhibition they produced was similar between female bacterial treatments, and were not significantly different compared to naïve and PBS-challenged females (Fig 2).
Thus all the bacteria species we used to immune challenge mothers induced enhanced levels of antibacterial activity in the eggs. However, whatever the type of bacteria (Gram-negative or Gram-positive) that challenged the mothers, antimicrobial activity found in the eggs was only active against Gram-positive bacteria.
Test of specificity of the maternal transfer of antimicrobial activity to the eggs was further extended to fungal maternal challenges using the same method as above. Females (22 to 23 per immune treatment group) were immune challenged with killed yeasts of Candida albicans, or spores of the entomopahogenic fungus, Metarhizium anisopliae. Immune challenged females with killed A. globiformis and sham injected ones were also included in this experiment as positive and negative controls, respectively. The resulting egg extracts were tested against yeasts (C. albicans), fungi (M. anisopliae), Gram-positive bacteria (A. globiformis and B. subtilis) and Gram-negative bacteria (E. coli, S. marcescens), using standard zone inhibition assays.
As in the previous experiment, we found no significant interaction between the maternal treatment and the egg assay on the probability of detecting a zone of inhibition (GLM: female treatment * egg assay: χ228, 348 = 0.07, p = 1), but the female treatment (χ210, 358 = 27.9, p < 0.001) and the egg assay (χ210, 358 = 29.7, p < 0.001) had a significant effect. A. globiformis-challenged females produced a higher proportion of protected eggs compared to C. albicans and M. anisopliae-challenged females (Fig 3). Similarly to our previous experiment, there was no evidence of specificity since an antibacterial activity could only be detected against A. globiformis and B. subtilis, and never on E. coli, S. marcescens, C. albicans and M. anisopliae (Fig 3). In addition, contrary to A. globiformis-challenged females, eggs of C. albicans-challenged, females did not produce any zone of inhibition on B. subtilis, and, among the eggs of the M. anisopliae-challenged females only those of a single female produced a zone of inhibition (see Fig 3 for corresponding test results).
There was also no effect of the female treatment in interaction with the egg assay on the size of zone of inhibition produced by the eggs among the clutches that exhibited antimicrobial activity (LM: F4, 40 = 3.37, p = 0.07). However, both female treatment (F3, 41 = 22.59, p < 0.001) and egg assay (F3, 41 = 67.85, p < 0.001) had a significant effect (Fig 4). Females challenged with A. globiformis produced the eggs with the highest antibacterial activity compared to females challenged with C. albicans and M. anisopliae (Fig 4). Consistently with the previous experiment, A. globiformis-challenged females produced eggs with a larger zone of inhibition on A. globiformis compared to B. subtilis (Fig 4, see S3 Fig for corresponding test results).
These results show that maternal challenges with fungi rarely led to the presence of antimicrobial activity in the eggs than maternal challenge with bacteria. However, when they do, antimicrobial activity found in the eggs is restricted toward Gram-positive bacteria, as for the bacterial immune challenges.
Based on above results, we aimed to identify the substance(s) responsible for egg antibacterial activity from eggs produced by bacterially immune challenged mothers. To this purpose, we used egg extracts from immune challenged females with killed A. globiformis (N = 8), B. thuringiensis (N = 16), E. coli (N = 13) or S. entomophila (N = 11) and those from sham-injected (N = 21) and naïve (N = 7) control females.
Egg extracts exposed to proteinase K lost their antibacterial activity on A. globiformis test plates (S4 Fig), confirming the proteinaceous nature of the egg antibacterial compounds. Then, in order to determine whether the maternal treatment influenced the eggs proteome, we compared the protein profile of the above egg extracts using Acid Urea—Polyacryamide Gel Electrophoresis (AU-PAGE) [34] and Tricine-SDS PAGE [35]. The protein profiles produced by AU-PAGE revealed the presence of an additional band of protein(s) (N1) in all egg extracts of immune challenged mothers with B. thuringiensis, E. coli and S. entomophila, in 88% of eggs from mothers immune challenged with A. globiformis, and never in eggs of sham injected and naïve control mothers (Fig 5). Similar results were obtained using Tricine-SDS PAGE analysis (S5 Fig).
Protein band(s) containing the protein(s) responsible for antibacterial activity in the eggs were then localized using a Gel Overlay Assay [34]. This assay consisted in transferring an AU-PAGE gel after electrophoresis onto a zone of inhibition test plate seeded with A. globiformis to test the antibacterial activity of each protein band. The assay revealed the presence of a single zone of bacterial growth inhibition corresponding to the N1 band in the colloidal blue-stained gel run in parallel. This assay revealed that the N1 band contained proteins with anti-Gram positive activity (S6 Fig). To get better resolution on the size of the protein band, a slice of AU-PAGE gel from the antibacterial region was electrophoresed into a Tricine SDS-PAGE gel, and resolved into a single band of about 5 kDa, which is similar to the molecular mass of D1 band proteins (S5 Fig).
Two protein bands corresponding to N1 on AU-PAGE gels after migration of egg extracts from B. thuringiensis and E. coli–injected females, and one corresponding to D1 on Tricine SDS-PAGE gels obtained after migration of egg extracts from S. entomophila–injected female (S5 Fig) were analysed by mass-spectrometry (LC-MS/MS) for protein identification. This analysis revealed the presence of tenecin-1 in both the N1 and D1 bands, whatever the bacterial maternal treatment (Table 1). Tenecin-1 is a 4.5 kDa antimicrobial peptide from the defensing family inhibiting the growth of Gram-positive bacteria [31].
We report three novel findings about maternal transfer of immunity in the eggs of the mealworm beetle T. molitor following a microbial immune challenge of the mothers. First, we found that levels of antimicrobial activity in eggs resulting from maternal challenge differed according to the kingdom of the microbes used for the maternal immune challenge. Whereas the bacteria species we used to immune challenge mothers induced enhanced levels of antimicrobial activity in the eggs (Fig 1), the use of fungi rarely did (Fig 3). Second, whatever the type of the microbes that challenged the mothers (fungi, Gram-negative or Gram-positive bacteria), antimicrobial substances found in the eggs were only active against Gram-positive bacteria. Third, we provide reasonable evidence that antibacterial activity directed toward Gram-positive bacteria in eggs of bacterially immune-challenged mothers is caused by the presence of tenecin 1, an antimicrobial peptide for which no other function than an activity against Gram+ bacteria has been reported [31]. Hence, antimicrobial activity in T. molitor eggs is mainly induced when mothers are challenged by bacteria and targets Gram-positive bacteria.
At first glance, our results contrast with those of a previous study, which found that the maternal immune protection of the offspring of the red flour beetle, Tribolium castaneum, was pathogen specific [13]. In this species, survival was increased in the adult offspring when the bacteria strain used on the progeny was the same than the one used in the maternal challenge. This would mean either that the degree of specificity of TGIP can differ between even closely related species, or that the immune protection of the offspring becomes pathogen specific at a later life stage. The effects of TGIP indeed differ across life stages [18]. In T. molitor for example, TGIP results in elevated levels of antimicrobial activity in eggs and larvae, but not in adults where increased haemocyte counts were observed [9, 15, 16]. Such a qualitative variation in the expression of enhanced immunity across life stages of the offspring may have several putative causes. For instance, it is possible that physiological constrains during the ontogeny of the developing offspring restrain the involvement of some immune effectors at certain life stages. Alternatively, ecological differences between eggs, newborn larvae and adults may differentially expose the offspring to microbes present in the environment. As a consequence, the immune effectors involved in the protection of the offspring at each life-stage might be optimal. Further investigations are needed to test these hypotheses.
It is currently unknown whether antibacterial activity in T. molitor eggs originates from the mothers or from the laid eggs themselves, although both origins are not exclusive. In various vertebrate and invertebrate species, maternally expressed immune factors are transferred into or onto the offspring, even in absence of any experimental maternal challenge [36–38]. As enhanced levels of antimicrobial activity are observed in eggs laid at the time where antimicrobial activity is detected in the hemolymph of T. molitor females [17, 27], one could reasonably hypothesize that the antimicrobial activity in the eggs originates from the antimicrobial peptide passively transferred by females to the eggs. In that case, its spectrum of activity would be a reflection of the local disease environment and should mirror that of the female’s hemolymph. In T. molitor, exposure to any kind of bacterial peptidoglycan (Lys-type or DAP-type) induces a strong anti-Gram-positive activity in the hemolymph of nymphs and adults ([27, 39] and S1 Fig). Similarly, exposure to DAP-type peptidoglycans (carried by Gram-negative and Bacillus bacteria) induces anti-Gram-negative activity ([39] and S1 Fig). If immune challenged females passively transfer antimicrobial substances in their eggs and assuming that all of them are equally transferred in eggs, then both anti-Gram-positive and anti-Gram-negative activity should be found in eggs of challenged females. In this study, anti-Gram-positive activity was found in eggs of bacterially challenged females, whatever the bacterial challenge. However, we failed to detect anti-Gram-negative activity in the eggs of females challenged with Gram-negative or Bacillus bacteria while we can detect it in the hemolymph of females using the same method (S1 Fig). While this would argue against a passive transfer of antibacterial peptides to the eggs by mothers, it should be noted that anti-Gram-negative activity in the hemolymph of mothers is much weaker than the anti-Gram-positive one (S1 Fig). This could result in very low or undetectable anti-Gram-negative activity in the eggs if mothers passively transfer antimicrobial activity to their eggs. Therefore, we cannot rule out the possibility that mothers can passively transfer antibacterial peptides in the eggs. Alternatively, antibacterial activity found in the eggs might be produced in the eggs themselves, in which the expression of genes encoding antimicrobial factors would differ according to the occurrence of a maternal challenge. Insect eggs, including coleopteran species, are able to express immune effectors genes such as antimicrobial peptides [25, 26, 28, 29]. In G. mellonela, immune gene expression in eggs was recently proposed to be induced by the depostion of the immunogens that challenged mothers in the eggs, which potentially support the expression of specific transgenerational immune responses [29]. If T. molitor eggs express genes coding for antimicrobial factors, they may thus exclusively express tenecin 1, suggesting that egg response to immunogens differs from that of their mother. Additional analyses found that transcripts of tenecin-1 were found to be abundant in eggs of immune-challenged females compare to those of control females (S1 Text). While it suggests that the peptides can be produced in the eggs, it does not rule out the possibility that transcripts can be provided by the mothers. Further studies will determine whether T. molitor eggs can express genes encoding antimicrobial factors, such as lysozymes and antimicrobial peptides, and whether such expression differ according to the maternal treatment. Our results might also be explained in a context of “microbiota effect”, when resident microbes replicate before being vertically transmitted upon infection of the host. Heritable bacteria are widespread in insects [40] and could persist at an undetectable concentration to the host immune system. Here, one might imagine that T. molitor females could harbour Gram-positive bacteria that are triggered into vertical transmission upon infection by any bacterial species, explaining the presence of antibacterial activity directed toward Gram-positive bacteria in eggs of immune challenge females. However, to our knowledge, T. molitor has yet never been reported to house covert bacterial infection or symbiotic bacteria [41]. Furthermore, we found no convincing indications of the presence of covert bacterial infection in ovaries and eggs of both control and immune challenged T. molitor females (S2 Text).
From an ecological and evolutionary point of view, the pattern of antimicrobial protection of the eggs may result from selective pressures imposed by the dominant microbial threat of the mealworm beetle. Insect eggs are known to suffer from microbial infection [42, 43] and our results suggest that Gram-positive bacteria might be more threatening than any other microbes in the natural environment of T. molitor. Microbial pathogens of T. molitor are currently barely known. To our knowledge, only three bacterial species have been described so far as pathogens of T. molitor and all are Gram-positive bacteria: Bacillus cereus, Bacillus thuringiensis and Brevibacillus laterosporus [44]. Other might be discovered in the future. Further studies will determine whether these or other Gram-positive bacteria are indeed the most abundant or threatening parasites for T. molitor, whether they can infest T. molitor eggs and whether antimicrobial activity provisioned in the eggs is beneficial against these microbes.
Even if Gram-positive bacteria are not the most abundant, they are probably the group of bacteria that are the most able to persist in the external environment by forming endospores. Endospore formation is limited to several genera of Gram-positive bacteria that include entomopathogenic bacteria such as Bacillus and Brevibacillus [45]. In the context of persisting diseases across generations, we may expect that when microbes are not transmitted vertically from mother to the offspring, only those able to survive in the external environment, such as endospore forming bacteria, have the highest probability to infect the offspring. Therefore, early levels of immune protection transferred by mothers to their offspring might be specific of these microbes. However, because fungi also form persistent spores, the above argument should hold for them too, especially knowing that insect eggs can suffer from fungi infections [42]. In contrast, our results show that fungi were weak inducers of TGIP in the eggs of T. molitor. One might hypothesise that immune effectors mobilized against fungi could not be transferred to the eggs or that females may use other maternal effects to avoid the infection of their eggs by entomopathogenic fungus [46].
To summarize, we found that maternal transfer of antimicrobial activity in eggs of T. molitor is mainly induced by bacterial immune challenges. However, whatever the microbial pathogen that challenged the mothers, antimicrobial substances transferred to the eggs were only active against Gram-positive bacteria because of the presence of tenecin 1. Our results suggest that this antimicrobial peptide found in the eggs is unlikely to originate from the female’s hemolymph. To our knowledge, our study is the first to provide a substantial characterization of the molecular mechanism involved in TGIP in an insect. Besides, we propose that maternal transfer of antimicrobial activity in the eggs of T. molitor might have evolved from the persistence of Gram-positive bacterial pathogens between insect generations. Detailed characterization of the community of pathogens associated with T. molitor is required to test this hypothesis.
Mealworm beetles originated from an outbred stock culture maintained in our laboratory in bran flour added with ad libitum access to water and regularly added with proteins (piglet flour), apple and bread. Pupae were then collected from these stock cultures and adults were maintained individually after emergence in a Petri dish supplied with bran flour a piece of apple and water for ten days. The experiments used virgin adult beetles of controlled age (10 ± 1 day post emergence). Due to practical limitations, the study was conducted in two successive experiments. The first experiment focused mainly on the effects of maternal bacterial challenges on the eggs anti-bacterial activity whereas the second experiment mainly focussed on the effects of maternal fungal immune challenges. Females were immune challenged (“female treatment”) by injection of a 5-μL suspension of inactivated microorganisms in phosphate buffer saline (PBS 10 mM, pH 7.4) after being chilled on ice for 10 min. Control females were treated in the same way, but with the omission of microorganisms as a procedural control for effect of the injection (sham control mothers). In the first experiment, an additional group of non-injected females (naïve control mothers) was used as control. Only sham control mothers were used in the second experiment because the first experiment reveals that in antimicrobial activity in eggs of sham control mothers was similar to that of naïve control mothers. Immediately after their immune treatment, the females were paired with a virgin and immunologically naïve male of the same age and allowed to produce eggs in a Petri dish supplied with wheat flour, apple and water in standard laboratory conditions (25°C, 70% RH; dark). Egg were collected every second day from day 3 to day 7 post injection because females are protecting most of their eggs at this period of time [17]. Two samples of 10 to 12 eggs per female were collected at random during this period of time to prepare egg extracts which antimicrobial activity was tested against the above microorganisms using inhibition zone assays [30].
Bacteria and yeasts were grown overnight at 28°C in LB and Sabouraud liquid respectively. M. anisopliae was cultured on Sabouraud-agar plates for four days at 28°C. Microorganisms were then inactivated in 0.5% formaldehyde prepared in PBS for 30 minutes, rinsed three times in PBS, and their concentration adjusted to 108 microorganisms per ml using a Neubauer improved cell counting chamber. Aliquots were kept at -20°C until use.
Egg extracts were prepared by smashing eggs into an acetic acid solution (0.05%, 5 μl per egg) followed by a 2 min centrifugation at 3600 rpm. Supernatants were divided into six to seven aliquots and kept at -20°C until antibacterial tests could be carried out.
Antimicrobial activity in egg extracts was measured using standard zone of inhibition assays [30]. Microorganisms were cultured as described above and seeded to 1% agar plates at a final concentration of 105 microorganisms / ml (LB-agar plates for bacteria, Sabouraud-agar plates for C. albicans and M. anisopliae). Six millilitres of seeded medium was poured into a Petri dish and sample wells were then made using a Pasteur pipette fitted with a ball pump. Two microliters of egg extract was added to each well. Plates were then incubated at 28°C for 24 or 48 h according to the tested microorganism after which the diameter of inhibition zones was measured. Antimicrobial activity of egg extracts obtained from the two random samples of eggs (see above) produced by each female was tested in triplicate for each microorganism (egg assay) and the mean of these 6 values was used as data point.
The proteinaceous nature of the antimicrobial compounds in egg extracts of bacterially immune challenged mothers was assessed by testing the persistence of their antibacterial activity against A. globiformis after being incubated with proteinase K (200 μg/mL) for 2 hours at 37°C. The influence of maternal immune challenge on the proteome of their eggs was examined by comparing the protein profile from egg extracts (obtained each from 5 eggs) produced by immune challenged mothers with A. globiformis, B. thuringiensis, E. coli, S. entomophila with that of sham control and naïve control mothers using Acid Urea—Polyacryamide Gel Electrophoresis (AU-PAGE) [34] and Tricine-SDS PAGE [35]. For the AU-PAGE analysis, extracts were mixed in loading buffer (9M urea in 5% acetic acid with methyl green as tracking dye) and loaded on a 15% AU-PAGE, which was pre-run at reversed polarity for 1h at 150 V using 5% acetic acid as running buffer. After 100 min of migration at 150V at reversed polarity, the gels were stained using a colloidal blue solution [47]. Tricine-SDS PAGE gels consisted in a 16.5% separating and an 8% stacking gels. After a 140 minutes run at 90V, the gels were stained using colloidal blue. The antimicrobial activity of all the extracts we used was assessed in parallel using zone of inhibition assays (see above).
Protein band(s) responsible for antibacterial activity in the eggs of bacterially immune challenged mothers were localized using a Gel Overlay Assay [34]. Egg extracts (10 eggs diluted in 25 μl of 0.05% acetic acid) were subjected to non-denaturing AU-PAGE in duplicate, as described above (except that loading buffer did not contain methyl green). The gel was cut into two identical halves. One half was rinsed twice for 10 min with 10 mM Sodium Phosphate Buffer pH 7.4 to remove excess of acetic acid and urea. The rinsed gel was then placed on LB-agar Petri dish prepared with low-electroendosmosis-type agarose containing A. globiformis (105 bacteria per ml). After incubation for 3h allowing transfer of the electrophoresed polypeptides to the underlying agar plate, the electrophoresis gel was removed and the Petri dish was incubated for 36 hours at 28°C to allow bacterial growth. Zones into which antimicrobial proteins or peptides had diffused were identified as bacterial colony-free regions. The second half of the gel was stained using colloidal blue solution.
Protein bands of interest from egg extracts produced by immune challenged females with B. thuringiensis, E. coli and S. entomophila were analysed by mass-spectrometry (LC-MS/MS). Native proteins (N1 band) and denatured ones (D1) were excised from colloidal blue stained AU-Page and Tris-SDS gels, respectively. Gel pieces were distained (NH4CO3 50mM/CH3CN 50/50) and digested with trypsin (0.3 μg/digestion) at 30°C for 12 hours. Peptide sequences were determined by mass spectrometry performed using a LTQ Velos instrument (Dual Pressure Linear Ion Trap) equipped with a nanospray source (ThermoFisher Scientific) and coupled to a U3000 nanoLC system (ThermoFisher Scientific). Proteins identification was performed with the MASCOT Algorithm from the Proteome Discoverer software v1.1 (ThermoFisher Scientific) against the UniProtKB/Swiss-Prot database (Dec 2012).
Specificity of the transgenerational immune response was analysed by first testing the probability of detecting a zone of inhibition in egg extracts for each female treatment and for each egg assay. Among the egg extracts for which we detected an inhibition zone, we checked whether the diameter of this zone was different according to female treatment and egg assay. The probability of detecting a zone of inhibition according to the female treatment and the egg assay was first analysed using a Generalized Linear Model (GLM) fitted with a binomial distribution (presence/absence of a zone of inhibition produced by a clutch for a given microorganism). Then the size of the zone of inhibition of the protected egg extracts according to the female treatment and egg assay was analysed with a Linear Model (LM). Egg extracts from the same female were assessed on different microorganisms, causing pseudoreplication and ideally the above analyses should have included the female ID as a random factor in a Generalized Linear Mixed Model (GLMM) and a Linear Mixed Model (LMM). However, the integration of a random factor in these models leads to their overparameterization, and the impossibility to test for the interaction between female and egg treatments. Since the results of the regular or mixed-effect models including the simple effects without interactions were consistent, we decided to continue our analyses with a simple GLM and a LM on these response variables. For all analyses, we checked if the residuals of the models were normally distributed and their variance homogeneous, in order to confirm the choice of a given distribution and explanatory variables. We performed backwards-stepwise regression as a mean for model simplification. We started by including both female treatment and egg assay and their interaction, and proceeded to remove non-significant terms. Comparisons of fits of the different models are provided in S3 Text. Differences between each female treatment and egg assay were highlighted by estimating the degree of overlap of the 95% CI, following the recommendations of [48]. The difference was considered significant when the 95% CI did not overlap on more than half of their length (p<0.05) [49]. The confidence intervals are represented on the bar plots of Figs 1 and 3. For Figs 2 and 4, they are represented with the estimates of the models (package ggplot2) in S2 and S3 Figs, respectively. All the data were analysed using R software [49]. The GLMM and LMM were performed with le lme4 package [50]. All data files are available from the Dryad database at doi:10.5061/dryad.4g40g. [51].
The accession number for the T. molitor antimicrobial peptide tenecin 1 is Q27023.
|
10.1371/journal.pcbi.1005888 | Specific excitatory connectivity for feature integration in mouse primary visual cortex | Local excitatory connections in mouse primary visual cortex (V1) are stronger and more prevalent between neurons that share similar functional response features. However, the details of how functional rules for local connectivity shape neuronal responses in V1 remain unknown. We hypothesised that complex responses to visual stimuli may arise as a consequence of rules for selective excitatory connectivity within the local network in the superficial layers of mouse V1. In mouse V1 many neurons respond to overlapping grating stimuli (plaid stimuli) with highly selective and facilitatory responses, which are not simply predicted by responses to single gratings presented alone. This complexity is surprising, since excitatory neurons in V1 are considered to be mainly tuned to single preferred orientations. Here we examined the consequences for visual processing of two alternative connectivity schemes: in the first case, local connections are aligned with visual properties inherited from feedforward input (a ‘like-to-like’ scheme specifically connecting neurons that share similar preferred orientations); in the second case, local connections group neurons into excitatory subnetworks that combine and amplify multiple feedforward visual properties (a ‘feature binding’ scheme). By comparing predictions from large scale computational models with in vivo recordings of visual representations in mouse V1, we found that responses to plaid stimuli were best explained by assuming feature binding connectivity. Unlike under the like-to-like scheme, selective amplification within feature-binding excitatory subnetworks replicated experimentally observed facilitatory responses to plaid stimuli; explained selective plaid responses not predicted by grating selectivity; and was consistent with broad anatomical selectivity observed in mouse V1. Our results show that visual feature binding can occur through local recurrent mechanisms without requiring feedforward convergence, and that such a mechanism is consistent with visual responses and cortical anatomy in mouse V1.
| The brain is a highly complex structure, with abundant connectivity between nearby neurons in the neocortex, the outermost and evolutionarily most recent part of the brain. Although the network architecture of the neocortex can appear disordered, connections between neurons seem to follow certain rules. These rules most likely determine how information flows through the neural circuits of the brain, but the relationship between particular connectivity rules and the function of the cortical network is not known. We built models of visual cortex in the mouse, assuming distinct rules for connectivity, and examined how the various rules changed the way the models responded to visual stimuli. We also recorded responses to visual stimuli of populations of neurons in anesthetized mice, and compared these responses with our model predictions. We found that connections in neocortex probably follow a connectivity rule that groups together neurons that differ in simple visual properties, to build more complex representations of visual stimuli. This finding is surprising because primary visual cortex is assumed to support mainly simple visual representations. We show that including specific rules for non-random connectivity in cortical models, and precisely measuring those rules in cortical tissue, is essential to understanding how information is processed by the brain.
| Much of our current understanding of local cortical connectivity in neuronal circuits of the neocortex is based on the presumption of randomness. Anatomical methods for estimating connection probabilities [1,2] and techniques for using anatomical reconstructions to build models of cortical circuits [3–7] are largely based on the assumption that connections between nearby neurons are made stochastically in proportion to the overlap between axonal and dendritic arborisations [8].
On the other hand, a wealth of evidence spanning many cortical areas and several species indicates that cortical connectivity is not entirely random. In species that display smooth functional maps in primary visual cortex (V1), such as cat and macaque monkey, long-range intrinsic excitatory connections tend to preferentially connect regions of similar function [9–13]. Although rodents exhibit a mapless, “salt and pepper” representation of basic visual features across V1 [14], non-random connectivity is nonetheless prevalent both within and between cortical layers [15–20], reflecting similarities in functional properties [21–25] or projection targets [26–28].
Despite multiple descriptions of specific connectivity in cortex, the rules underlying the configuration of these connections are not entirely clear. Whereas strong connections are more prevalent between neurons with similar receptive fields, the majority of synaptic connections are made between neurons with poorly-correlated receptive fields and poorly correlated responses [24]. This sea of weak synaptic inputs might be responsible for non-feature-specific depolarisation [24] or might permit plasticity of network function [20].
However, another possibility is that weak local recurrent connections reflect higher-order connectivity rules that have not yet been described. Recent reports have highlighted the facilitatory and selective nature of plaid responses in mouse V1 [29–31]. Many neurons in mouse V1 respond to plaid stimuli in accordance with a simple superimposition of their responses to the two underlying grating components (i.e. “component cell” responses [32]). However, a significant proportion of neurons that are visually responsive, reliable and selective exhibit complex responses to plaid stimuli that are difficult to explain with respect to simple combinations of grating components [30]. We hypothesised that responses to complex stimuli in mouse V1 could be a result of local combinations of visual features, through structured local recurrent excitatory connectivity. These rules could be difficult to detect through anatomical measurements, if they comprised only small deviations from predominantly like-to-like connectivity.
Here we examined whether small tweaks to recurrent connectivity rules could alter visual representations in cortex, by analysing the computational properties of cortical networks with defined rules for local connectivity. We simulated visual responses to grating and plaid stimuli in large networks with properties designed to resemble the superficial layers of mouse V1, assuming distinct connectivity schemes. We then compared the response patterns and visual representations predicted by the network simulations with those recorded in vivo in mouse V1, to test the predictions arising from our models.
Specifically, we evaluated two broad classes of connectivity patterns, where specific local excitatory connectivity is defined according to the visual response properties of neurons (Fig 1):
Despite the small difference in network configuration, these distinct rules give rise to radically different visual representations of plaid stimuli, both in terms of complexity of visual response selectivity of individual neurons and regarding facilitation versus suppression in response to these compound stimuli. We found that the complexity of plaid responses in mouse V1 was reproduced in our simulations when assuming the feature-binding connectivity scheme, with local connections grouping multiple feedforward response properties, but not when assuming purely like-to-like connections.
Under the assumption that the configuration of local recurrent connections in cortex might lead to differential processing of simple and compound visual stimuli, it is important to quantify the relationship between responses to grating and plaid stimuli in visual cortex. Plaid stimuli are often constructed from a single choice of relative component angle (90° orthogonal gratings), leaving open the possibility that a richer set of plaid stimuli would help to classify neurons with complex responses.
We therefore probed mouse V1 with grating component stimuli composed of grating stimuli with 16 drift directions, and three full sets of plaid stimuli composed of 45°, 90° and 135° relative grating component orientations. We recorded responses from layer 2/3 neurons in V1 using two-photon imaging of animals expressing GCaMP6m (Fig 2A–2F; 8 animals, 8 sessions, 441 / 879 responsive / imaged neurons; see Methods). We defined a modulation index (MI) to quantify the degree of facilitation or suppression elicited by plaid stimuli over grating stimuli, for single cortical neurons; large positive values for MI indicate strong facilitation in response to plaid stimuli, whereas large negative values indicate strong suppression (see Methods). Visual responses to the full set of plaid stimuli were dominated by facilitation, and were significantly more facilitatory than when considering only the set of 90° plaids (Fig 2I; median modulation index MI 0.098 ± [0.081 0.12] vs 0.011 ± [-0.0060 0.027]; p < 1⨉10–10, Wilcoxon rank-sum; all following values are reported as median ± 95% bootstrap confidence intervals unless stated otherwise).
The presence of stronger facilitation when comparing responses to the full set of plaid stimuli with responses to 90° plaids alone, is consistent with our earlier finding that some neurons in mouse V1 are highly selective for particular combinations of grating components [30]. Accordingly, we used a plaid selectivity index (PSI) to quantify how selective were the responses of single neurons over the set of plaid stimuli (see Methods). The PSI was defined in analogy to orientation or direction selectivity indices (OSI or DSI), such that values of PSI close to 1 indicate that a neuron responds to only a single plaid stimulus out of the set of presented plaid stimuli. Values of PSI close to 0 indicate that a neuron responds equally to all plaid stimuli. Responses to the full set of plaid stimuli were highly selective; significantly more selective than predicted by a component model generated using all plaid and grating stimuli (Fig 2J; median PSI 0.38 ± [0.36 0.41] vs 0.30 ± [0.28 0.31]; p < 1⨉10–10, Wilcoxon rank-sum) and indeed significantly more selective than responses to the 90° plaids alone (Fig 2J; median 90° PSI 0.25 ± [0.23 0.28]; p < 1⨉10–10 vs all plaids, Wilcoxon rank-sum).
Therefore, probing visual cortex with a dense set of plaid stimuli reveals richer visual responses than when probed with a set of only 90° plaids. Indeed, recent results suggest that using an expanded set of plaid stimuli evokes more pattern-cell responses in mouse V1 [31]. Consistent with this finding, our results show that using a dense set of plaids does not make responses to compound stimuli trivial to predict based on component responses. In addition, we found that visual responses were more facilitatory and more selective than when measured using 90° plaids alone.
How are selective, facilitatory responses to plaid stimuli generated in V1? As we suggested previously, one possibility is that specific grating component representations are combined through local excitatory connectivity [30]. In mouse V1, synaptic connection probability is enhanced by similarity of orientation preference [21,23,25], suggesting that local excitatory connections may group together neurons with common preferred orientations. Connection probability is even more strongly modulated by neuronal response correlations to natural visual stimuli; i.e., the likelihood for a synaptic connection is higher for neuronal pairs responding similarly to natural scenes [21,22,24].
If connections in mouse V1 were strictly governed by preferred orientation, then neurons with similar orientation preference should also predominately have similar responses to natural movies, and vice versa. We recorded visual responses in populations of neurons labelled with the synthetic calcium indicator OGB in anesthetized mouse V1 (5 animals, 129 / 391 responsive neurons with overlapping receptive fields / total imaged neurons; S1A–S1C Fig; see Methods). We used signal correlations to measure the similarity between the responses of pairs of neurons with identified receptive fields (S1A Fig) to drifting grating (S1B Fig) and natural movie (S1C Fig) visual stimuli (see Methods).
We found that neuronal pairs with high signal correlations to natural scenes, which are most likely to be connected in cortex [21,22,24], showed only a weak tendency to share similar orientation preferences (S1D and S1E Fig; pairs with OSI > 0.3; p = 0.8, Kruskall-Wallis). This is consistent with earlier findings in cat area 17 (V1), which showed a poor relationship between responses to gratings and natural movies [34].
Similarly, under a like-to-like connectivity rule, synaptically connected neurons in mouse V1 should share both similar orientation preference and responsiveness to natural movies. We therefore compared response correlations and preferred orientations for pairs of mouse V1 neurons, which were known to be connected from in vivo / in vitro characterisation of functional properties and connectivity (data from [24] used with permission; 17 animals, 203 patched and imaged cells, 75 connected pairs). Consistent with our results comparing responses to gratings and natural movies, connected pairs of cells with similar orientation preference were not more likely to share a high signal correlation to flashed natural scenes (S1F Fig; p = 0.54, Kruskall-Wallis). Also consistent with earlier findings [21,23], we observed a positive relationship between synaptic connectivity and similarity of orientation preference (S1G Fig; p = 0.045, Ansari-Bradley test). However, strongly connected pairs (strongest 50% of excitatory post-synaptic potentials—EPSPs—over connected pairs) were not more similar in their preferred orientation than the remaining pairs (p = 0.17, Ansari-Bradley test vs weakest 50% of connected pairs). Connected pairs spanned a wide bandwidth of preferred orientations, with more than 20% of connections formed between neurons with orthogonal preferred orientations. Spatial correlation of receptive fields is a comparatively better predictor for synaptic connectivity than shared orientation preference, but a majority of synaptic inputs are nevertheless formed between neurons with poorly- or un-correlated responses [24]. We conclude that similarity in orientation preference only partially determines connection probability and strength between pairs of neurons in mouse V1.
This weak functional specificity for similar visual properties can be explained by two possible alternative connectivity rules. In the first scenario, local excitatory connections in cortex are aligned with feedforward visual properties, but with broad tuning (Fig 1A; a like-to-like rule). As a consequence, all connections show an identical weak bias to be formed between neurons within similar tuning, and the average functional specificity reported in S1G Fig and elsewhere [21,24] reflects the true connection rules between any pair of neurons in cortex.
Alternatively, local excitatory connections may be highly selective, but follow rules that are not well described by pairwise similarity in feedforward visual properties. For example, subpopulations of connected excitatory neurons might share a small set of feedforward visual properties, as opposed to only a single feedforward property (Fig 1B; a feature-binding rule). In this case, connections within a subpopulation could still be highly specific, but this specificity would be difficult to detect through purely pairwise measurements. If pairwise measurements were averaged across a large population, any specific tuning shared within groups of neurons would be averaged away.
We designed a non-spiking model of the superficial layers of mouse V1, to explore the effect of different connectivity rules on information processing and visual feature representation within the cortex. Non-spiking linear-threshold neuron models provide a good approximation to the input current to firing rate (I–F) curves of adapted cortical neurons [35]; model neurons with linear-threshold dynamics can be directly translated into integrate-and-fire models with more complex dynamics [36,37], and in addition form good approximations to conductance-based neuron models [38]. A full list of parameters for all models presented in this paper is given in Table 1.
The dynamics of neuronal networks defined with particular specific synaptic connectivity rules remain generally unknown, although some results suggest that specific connectivity leads to reduced dimensionality of network activity patterns [40]. Here we explored the relationship between specific connectivity and network dynamical properties in a non-linear, rate-based network model incorporating realistic estimates for recurrent excitatory and inhibitory connection strength in layer 2 / 3 of mouse V1.
To explore the basic stability and computational consequences of functionally specific excitatory connectivity, we built a small five-node model (four excitatory and one inhibitory neurons; “analytical model”; Fig 3). Connections within this model were defined to approximate the average expected connectivity between populations of neurons in layer 2 / 3 of mouse V1. Excitatory neurons were grouped into two subnetworks, and a proportion s of synapses from each excitatory neuron was reserved to be made within the same subnetwork.
When s = 0, E↔E synapses were considered to be made without specificity, such that each connection in the small model approximated the average total connection strength expected in mouse V1 in the absence of functional specificity. When s = 1, all E↔E synapses were considered to be selectively made within the same subnetwork, such that no synapses were made between excitatory neurons in different subnetworks. Connections to and from the inhibitory node were considered to be made without functional specificity in every case, mimicking dense inhibitory connectivity in mouse visual cortex [41–44]. The general form of the weight matrix is therefore given by
W=[aabb−wieaabb−wiebbaa−wiebbaa−wieweiweiweiwei−wI∙fI]
(2)
where wS = wE ⋅(1− fI) ⋅s is the specific weight component, wN = wE ⋅(1− fI) ⋅(1−s) is the nonspecific weight component, wE is the total synaptic weight from a single excitatory neuron, wI is the total synaptic weight from a single inhibitory neuron; fI = 1/5 is the proportion of inhibitory neurons; a = wS / 2+wN / 4 is the excitatory weight between neurons in the same subnetwork; b = wN / 4 is the excitatory weight between neurons in different subnetworks; wie = wI ⋅(1− fI) / 4 is the nonspecific inhibitory to excitatory feedback weight; and wei = wE ⋅ fI is the nonspecific excitatory to inhibitory weight.
Amplification in the network with specific connectivity is selective (Fig 3B and 3C): neurons within a subnetwork recurrently support each other’s activity, while neurons in different subnetworks compete. Therefore, which sets of neurons will be amplified or will compete during visual processing will depend strongly on the precise rules used to group neurons into subnetworks. We therefore examined the impact of like-to-like and feature-binding rules on responses in our analytical model. The excitatory network was partitioned into two subnetworks; connections within a subnetwork corresponded to selective local excitatory connectivity within rodent V1. Under the like-to-like rule, neurons with similar orientation preferences were grouped into subnetworks (Fig 3D).
We tested the response of this network architecture to simulated grating and plaid stimuli, by injecting currents into neurons according to the similarity between the orientation preference of each neuron and the orientation content of a stimulus (see grating labels in Fig 3D and 3E). When a stimulus matched the preferred orientation of a neuron, a constant input current was injected (Ii (t = ι); when a stimulus did not match the preferred orientation, no input current was provided to that neuron (Ii (t) = 0). When simulating the analytical model, the input current ι = 1.
Under the like-to-like rule, responses of pairs of neurons to simple grating stimuli and more complex plaid stimuli were highly similar (Fig 3D). Amplification occurred within subnetworks of neurons with the same preferred orientation, and competition between subnetworks with differing preferred orientation [53,55] (visible by complete suppression of response of neurons in lower traces of Fig 3D).
Alternatively, we configured the network such that the rules for local excitatory connectivity did not align with feedforward visual properties (a feature-binding rule). We formed subnetworks by grouping neurons showing preference for either of two specific orientations (Fig 3E). When this feature-binding connectivity rule was applied, neuronal responses to grating and plaid stimuli differed markedly (cf. top vs bottom panels of Fig 3E). Selective amplification was now arrayed within populations of neurons spanning differing orientation preferences, and competition occurred between subnetworks with different compound feature preferences. Importantly, a feature-binding rule implies that neurons with the same preferred orientation could exist in competing subnetworks. While their responses to a simple grating of the preferred orientation would be similar and correlated (Fig 3E; indicated by a high response correlation measured over grating responses ρg), the same two neurons would show decorrelated responses to a plaid stimulus (Fig 3E; indicated by a low response correlation measured over plaid responses ρp). We conclude that changes in pairwise response similarity, provoked by varying the inputs to a network, can provide information about the connectivity rules present in the network.
The results of our simulations of small networks suggest that rules for specific local connectivity can modify the correlation of activity between two neurons in a network, depending on the input to the network. The question arises of how connectivity rules shape distributed representations of visual stimuli, when examined across a large network and over a broad set of stimuli. We therefore simulated the presentation of grating and plaid visual stimuli in a large-scale non-linear, rate-based model of the superficial layers of mouse V1. Individual neurons were modelled as described above for the small scale network (Eq (1)).
To construct the large-scale simulation model of mouse V1, 80,000 linear-threshold neurons were each assigned a random location in physical space ui∈T2 where T defines the surface of a virtual torus of size 2.2×2.2 mm. Excitatory and inhibitory neurons were placed with relative densities appropriate for layers 2 and 3 of mouse cortex [56]. Approximately 18% of neurons were inhibitory; [57,58]; see Tables 1 and 2 for all parameters used in these models. Excitatory neurons were assigned an orientation preference θ drawn from a uniform random distribution, mimicking the “salt and pepper” functional architecture present in rodent visual cortex [14].
We simulated the presentation of grating and plaid stimuli in our large-scale network model of mouse V1. We quantified response similarity between pairs of neurons as suggested by the results of the small network simulations: by measuring pairwise response correlations over a set of grating stimuli (ρg), and separately over a set of plaid stimuli (ρp; see Methods).
In the network that implemented a like-to-like connection rule for recurrent excitatory connectivity (Fig 4A and 4B), pairs of neurons showed similar responses to both grating and plaid stimuli (Fig 4B; R2 = 0.83 between ρg and ρp), in agreement with the analytical like-to-like model of Fig 3D.
However, in the network that implemented a feature-binding connection rule, where in addition to spatial proximity and similarity in preferred orientation subnetworks were defined to group neurons of two distinct preferred orientations (Fig 4C and 4D), neurons showed reduced correlation in response to plaid stimuli (Fig 4D, R2 = 0.13 between ρg and ρp), in agreement with the analytical feature- binding model of Fig 3E. Different configurations of local recurrent excitatory connectivity produced by like-to-like or feature-binding rules can therefore be detected in large networks, by comparing responses to simple and compound stimuli.
Consistent with our analytical models, networks without functionally specific connectivity did not give rise to decorrelation (S3B Fig; R2 = 0.72 between ρg and ρp). This shows that decorrelation between plaid and grating responses in our models does not arise simply due to random connectivity, but requires the active mechanism of selective amplification through feature-binding subnetwork connectivity. Inhibitory responses were untuned in our simulations (blue traces in Fig 4A and 4C), in agreement with experimental observations of poorly-tuned inhibition in mouse V1 [42,58,65,66].
Our analytical network results show that in principle the configuration of local excitatory connectivity, whether aligned with or spanning across feedforward visual properties, has a strong effect on visual representations (Fig 3). Our large-scale simulations show that these effects can be detected in large networks as differences in the pairwise correlations of responses to simple and compound visual stimuli (Fig 4). We therefore aimed to test which connectivity scheme is more likely to be present in visual cortex, by examining responses of neurons in mouse V1.
Using two-photon calcium imaging, we recorded responses of populations of OGB-labelled neurons in mouse V1 to a set of contrast-oscillating oriented grating stimuli over a range of orientations, as well as the responses to the set of plaid stimuli composed of every possible pair-wise combination of the oriented grating stimuli (Fig 5; 5 animals, 5 sessions, 313 / 543 responsive / total imaged neurons; see Methods). Responses to plaid stimuli in mouse V1 suggest that stimulating with a denser sampling of compound stimulus space leads to a better characterisation of response selectivity [31] (Fig 2). Accordingly, we probed responses in mouse V1 under stimuli analogous to those used in the model simulations, with a dense coverage of plaid combinations over a set of finely-varying grating orientations.
We found that consistent with our earlier findings examining 90° drifting plaid stimuli [30], responses to grating stimuli did not well predict responses to plaid stimuli. Pairs of neurons with similar preferred orientation but with highly differing responses to plaid stimuli were common (Fig 5B and 5C; R2 = 0.05 between ρg and ρp; OSI > 0.3). The degree of decorrelation we observed in mouse V1 was considerably higher than predicted by the like-to-like model, and was more consistent with the feature-binding model (Fig 5E).
Decorrelation induced by plaid responses and the lack of a relationship between grating and plaid responses in mouse V1 were not a result of unreliable or noisy responses in vivo. We included in our analysis only neurons that were highly reliable, and responded significantly more strongly than the surrounding neuropil (see Methods). As a further control, we used experimentally recorded responses to grating stimuli to generate synthetic plaid responses for mouse V1 that would result from a cortex with like-to-like subnetwork connectivity (Fig 5D, inset; see Methods). Our control data were generated from single-trial responses of single V1 neurons, and therefore included the same trial-to-trial variability exhibited by cortex. This control analysis indicates that a like-to-like rule among V1 neurons would result in a higher correlation of grating and plaid responses than experimentally observed (Fig 5D; median R2 = 0.77 ± [0.767 0.775] between ρg and ρp; n = 2000 bootstrap samples; compared with R2 = 0.05 for experimental results; p < 0.005, Monte-Carlo test).
Importantly, this control analysis is not restricted to our like-to-like rule, but makes similar predictions of highly correlated grating and plaid responses for any arbitrary model that combines grating components to produce a plaid response, as long as that rule is identical for every neuron in the network [30]. This is because if a single consistently-applied rule exists, then any pair of neurons with similar grating responses (high ρg) will also exhibit similar plaid responses (high ρp). In contrast, neurons that are connected within the feature-binding model combine different sets of grating components, depending on which subnetwork the neurons are members of.
Neurons in mouse V1 exhibited a wide range of facilitatory and suppressive responses to plaid stimuli, roughly equally split between facilitation and suppression (Fig 5F and 5G; 45% vs 42%; MI > 0.05 and MI < –0.05). The proportion of facilitating and suppressing neurons in mouse V1 was similar to that exhibited by responsive neurons in our feature-binding model (Fig 5G; V1 versus F.B., p = 0.17; two-tailed Fisher’s exact test, nV1 = 313, nF.B. = 809). In contrast, neither the like-to-like model nor a model without functionally specific connectivity exhibited significant facilitation in responsive neurons, and both were significantly different from the distribution of facilitation and suppression in mouse V1 (Fig 5G; p < 0.001 in both cases; two-tailed Fisher’s exact test, nL-to-L = 729, nRnd = 729). The wide range of facilitatory and suppressive responses observed in mouse V1 is more consistent with a feature-binding rule for local connectivity, compared with a like-to-like rule or a network without functionally specific connectivity.
Whereas feedforward mechanisms for building response properties in visual networks have been extensively studied, it is not well understood how visual responses are shaped by local recurrent connections. We hypothesised that the configuration of local recurrent cortical connectivity shapes responses to visual stimuli in mouse V1, and examined two alternative scenarios for local connection rules: essentially, whether local excitatory connections are made in accordance with feedforward visual properties (“like-to-like”; Fig 1A), or whether local excitatory connections span across feedforward visual properties to group them (“feature-binding”; Fig 1B). We found that highly selective and facilitatory responses to plaid stimuli observed in mouse V1 (Fig 2, Fig 5; [30]) are consistent with tuning of recurrent connections within small cohorts of neurons to particular combinations of preferred orientations. Moreover, responses in mouse V1 are inconsistent with a simple configuration of cortical connections strictly aligned with feedforward visual responses.
We found that the precise rules that determine local connections among neurons in cortex can strongly affect the representation of visual stimuli. The feature-binding rule we examined embodies the simplest second-order relationship between connectivity and preferred orientation, and was chosen for this reason. We cannot rule out more complicated connectivity rules as being present in mouse V1, but we have shown that a simple like-to-like rule cannot explain responses to plaid visual stimuli. Random, non-functionally specific connections were also unable to explain complex plaid responses in mouse V1 (S3 Fig).
How can the detailed statistics of “feature-binding” rules be measured in cortex? Existing experimental techniques have been used to measure only first-order statistical relationships between function and cortical connectivity [18,21–24,42]. Unfortunately, current technical limitations make it difficult to measure more complex statistical structures such as present under a feature-binding connectivity rule. Simultaneous whole-cell recordings are typically possible from only small numbers of neurons, thus sparsely testing connectivity within a small cohort. Even if simultaneous recordings of up to 12 neurons are used [17], identifying and quantifying higher-order statistics in the local connectivity pattern is limited by the low probability of finding connected excitatory neurons in cortex. Nevertheless, our feature-binding connectivity model is consistent with the results of functional connectivity studies (S1 Fig).
In addition, our results highlight that small changes in the statistics of local connectivity can have drastic effects on computation and visual coding. Introducing a small degree of specificity, such that a minority of synapses are made within an excitatory subnetwork, is sufficient to induce strong specific amplification and strong competition to the network, even though a majority of the synapses are made randomly without functional specificity (Fig 3A–3C). Under our feature-binding model 68% of synapses are made randomly; approximately 27% are made under a like-to-like rule and the remaining 5% are used to bind visual features. Clearly, detecting the small proportion of synapses required to implement feature binding in V1 will be difficult, using anatomical sampling techniques that examine only small cohorts of connected neurons.
A recent study functionally characterised the presynaptic inputs to single superficial-layer neurons in mouse V1, using a novel pre-synaptic labelling technique [67]. Consistent with our results for preferred orientation (S1F and S1G Fig), they found that presynaptic inputs were similarly tuned as target neurons but over a wide bandwidth. The majority of synaptically connected networks were tuned for multiple orientation preferences across cortical layers, similar to the feature-binding networks in our study.
We implemented an alternative approach, by inferring the presence of higher-order connectivity statistics from population responses in cortex. This technique could be expanded experimentally, by presenting a parameterised battery of simple and complex stimuli. Stimuli close to the configuration of local connectivity rules would lead to maximal facilitation and competition within the cortical network. Importantly, our results strongly suggest that simple stimuli alone are insufficient to accurately characterise neuronal response properties in visual cortex.
Our theoretical analysis and simulation results demonstrate that functionally specific excitatory connectivity affects the computational properties of a cortical network by introducing amplification of responses within subnetworks of excitatory neurons, and competition in responses between subnetworks (Fig 3A–3C). Several recent studies have demonstrated that visual input is amplified within the superficial layers of cortex [68–70], and recent results from motor cortex suggest competition between ensembles of neurons [71]. Our modelling results indicate that some form of selective local excitatory connectivity is required for such amplification to occur through recurrent network interactions, under reasonable assumptions for anatomical and physiological parameters for rodent cortex (Fig 3A–3C; S2 Fig). This still leaves in question whether the particular configuration of selective excitatory connectivity plays a role.
Our simulation results showed that the effects of amplification and competition on cortical responses are tuned to the statistics of local connectivity. This implies that complex visual stimuli for which the composition of stimulus components matches the statistics of a subnetwork will undergo stronger amplification than other non-matching visual stimuli (Fig 6). In our feature-binding model, the statistics of subnetwork connectivity were defined to reflect combinations of two preferred orientations chosen from a uniform random distribution. This combination of two orientations is similar to the visual statistics of plaid stimuli with arbitrarily chosen grating components. As a result, plaid stimuli gave rise to stronger amplification than single grating components alone, when the composition of the plaid matched the composition of connectivity within a particular subnetwork. This led to a facilitatory effect, where some neurons responded more strongly to plaid stimuli than to the grating components underlying the plaid stimuli. Conversely, competition between subnetworks led to weaker responses to some plaid stimuli, for neurons that “lost” the competition. Competition could therefore be one cortical mechanism underlying cross-orientation suppression in response to plaid stimulation.
In contrast, suppression in the like-to-like and random non-specific models occur because the energy in the stimulus is spread across two grating components, and is not combined by the network to form strong plaid selectivity. In the like-to-like model, competition occurs between representations of the two oriented grating components of the plaid, causing additional suppression. The presence of amplified, strongly facilitating plaid responses in mouse V1 is therefore consistent with the existence of subnetworks representing the conjunction of differently-oriented edges.
Could the complexity of plaid texture responses in mouse V1 be explained by convergence of differently tuned feedforward inputs from layer 4 onto single layer 2/3 neurons, similar to the proposed generation of pattern-selective responses in primate MT [32,72]? Building plaid responses in this way would imply that layer 2/3 neurons would respond to multiple grating orientations, since they would receive approximately equal inputs from at least two oriented components. However, layer 4 and layer 2/3 neurons are similarly tuned to orientation in rodent V1 [63,64], in conflict with this feedforward hypothesis.
In addition, if responses to complex stimuli were built by feedforward combination of simple grating components, then the response of a neuron to the set of grating stimuli would directly predict the plaid response of that neuron. This would then imply that two neurons with similar responses to plaid stimuli must have similar responses to grating stimuli. However we found this not to be the case experimentally; two neurons with similar responses to grating components often respond differently to plaid textures or to natural scenes (S1D Fig; Fig 5A and 5B; [30]).
We cannot rule out the influence of feedback projections on shaping responses to plaid stimuli. The time resolution of calcium imaging is too slow to differentiate between feedforward, recurrent local, and feedback responses based only on timing. However, top-down feedback inputs are considered to be suppressed during anesthesia [73]; in contrast, we observed complex responses to plaid stimuli in anesthetized animals. Since our proposed mechanism for feature binding relies on recurrent amplification, relatively few excitatory synapses are required to reproduce complex plaid responses. In contrast, non recurrent influences such as feedforward or feedback projections would require comparatively more synapses to achieve a similar pattern of plaid responses. There are more local recurrent excitatory synapses in V1 layer 2 / 3 than there are available excitatory synapses in feedback projections to V1 (22% recurrent excitatory synapses in layer 2 / 3 vs a maximum of 17.2% feedback synapses; [1]). In addition, putative feedback inputs would need to be wired with high functional specificity; this degree of anatomical specificity has not been demonstrated experimentally.
Non-specific connectivity between excitatory and inhibitory neurons, as assumed in our simulation models, is consistent with the concept that inhibitory neurons simply integrate neuronal responses in the surrounding population [74], and is also consistent with experimental observations of weakly tuned or untuned inhibition in rodent visual cortex [42,52,58,65,66]. Although specific E↔I connectivity has been observed in rodent cortex [16,28], the majority of E↔I synapses are likely to be made functionally non-specifically in line with the high convergence of E→I and I→E connections observed in cortex [41,42,65].
In our models, shared inhibition is crucial to mediate competition between excitatory subnetworks (Fig 3); inhibition is untuned because excitatory inputs to the inhibitory population are pooled across subnetworks. Poorly tuned inhibition, as expressed by the dominant class of cortical inhibitory neurons (parvalbumin expressing neurons), therefore plays an important computational role and is not merely a stabilising force in cortex.
Other inhibitory neuron classes in cortex (e.g. somatostatin or vaso-intestinal peptide expressing neurons) have been shown to exhibit feature-selective responses [58,75,76]. Recent computational work examined the influence of multiple inhibitory neuron classes with different physiological and anatomical tuning properties in a model for rodent cortex [77]. They examined the role of inhibitory connectivity on divisive and subtractive normalisation of network activity in a network with specific, orientation-tuned inhibitory connectivity. They found that specific inhibitory feedback could lead to divisive normalisation of network activity, while non-specific inhibitory feedback could lead to subtractive normalisation.
However, the computational role of specific inhibition is likely to rest on the precise rules for connectivity expressed between excitatory and inhibitory neurons. If the rules for E↔E and E↔I connections align, then a specific inhibitory population could act as a break on excitation within a subnetwork, and could allow more specific anatomical connectivity to persist while maintaining the balance between excitation and inhibition in cortex. The functional profile of this balancing pool would be highly tuned, and be similar to that of the excitatory neurons in the subnetwork, suggesting a physiological signature of specific inhibitory feedback that could be sought experimentally. Alternatively, if E↔I connection rules result in counter-tuned specificity, these connections would act to strengthen competition between subnetworks.
As discussed above, our like-to-like model of orientation-tuned selective excitatory connectivity coupled with non-specific inhibitory feedback is similar in network topology to classical ring models of orientation tuning in visual cortex (e.g [53,54,78]). The principal difference in our model is the embedding of functionally selective connectivity within spatially-constrained anatomical connectivity. We showed that under model parameters chosen to be realistic in mouse V1, only a small fraction of excitatory synapses must be specific in order to introduce selective amplification and competition within the network.
Several previous models of recurrent cortical connectivity designed for columnar visual cortex have incorporated selective excitatory connectivity, either with connectivity relying on purely anatomical constraints (e.g [79]) or mimicking the spatially periodic, long-range lateral excitatory projections found in monkey, cat and other species (e.g. [80–83]). Similarly to our models, these works emphasise that feature integration can occur within V1 through recurrent processing of visual stimuli. These earlier models examined how specific synaptic connectivity between spatially separated neurons across visual space can perform operations that link representations of similar visual features such as contour integration, and can underlie competition between dissimilar visual features [80,84]. The principal difference to our models is that we examined how local excitatory connections between neurons representing overlapping regions of visual space can underlie facilitatory binding of dissimilar visual features. Our models therefore examine the consequences of higher-order patterns in local recurrent connectivity on visual coding.
In visual cortex of primates, carnivores and rodents, orientation tuning develops before postnatal eye opening and in the absence of visual experience [85,86].
Local recurrent connections develop after the onset of visual experience and maintain their plasticity into adulthood [85,87–91]. Statistical correlations in natural scenes might therefore lead to wiring of subnetworks under an activity-dependent mechanism such as spike-time dependent plasticity (STDP) [92–96]. Along these lines, examinations of the development of specific excitatory connections after eye opening found that similarities in feedforward input were progressively encoded in specific excitatory connections [22].
We expect that, as the specificity of lateral connections forms during development, the emergence of compound feature selectivity will gradually occur after the onset of sensory experience. This hypothesis is consistent with experience-dependent development of modulatory effects due to natural visual stimulation outside of the classical receptive field, as observed in mouse V1 [97]. A complete factorial combination of all possible features occurring in natural vision is clearly not possible. However, the most prominent statistical features of cortical activity patterns could plausibly be prioritised for embedding through recurrent excitatory connectivity. At the same time, competition induced by non-specific shared inhibition will encourage the separation of neurons into subnetworks. In our interpretation, single subnetworks would embed learned relationships between external stimulus features into functional ensembles in cortex, such that they could be recovered by the competitive mechanisms we have detailed.
In prefrontal cortex, compound or mixed selectivity of single neurons to combinations of task-related responses has been found in several studies [98,99]. This is proposed to facilitate the efficient decoding of arbitrary decision-related variables. Binding feedforward cortical inputs into compound representations, as occurs in our feature-binding model, is therefore a useful computational process with general applicability. Our work suggests that specific local excitatory connectivity could be a general circuit mechanism for shaping information processing in cortical networks.
Experimental procedures followed institutional guidelines and were approved by the Cantonal Veterinary Office in Zürich or the UK Home Office.
Procedures for urethane anesthesia, craniotomies, bulk loading of the calcium indicator, as well as for in vivo two-photon calcium imaging and in vitro recording of synaptic connection strength were the same as described previously [24,30,100,101].
Visual stimuli for receptive field characterisation, drifting gratings and plaids and masked natural movies were displayed on an LCD monitor (52.5 × 29.5 cm; BenQ) placed 10–11 cm from the eye of the animal and covering approximately 135 × 107 visual degrees (v.d.). The monitor was calibrated to have a linear intensity response curve. Contrast-oscillating grating and plaid stimuli were presented on an LCD monitor (15.2 × 9.1 cm; Xenarc) placed 9 cm from the eye of the animal and covering 80 × 54 v.d. The same screen was used for stimulus presentation during intrinsic imaging to locate visual cortex and during two-photon imaging. The open-source StimServer toolbox was used to generate and present visual stimuli via the Psychtoolbox package [33,105].
Stimuli for receptive field characterisation comprised a 5 × 5 array of masked high contrast drifting gratings (15 v.d. wide; overlapping by 40%; 9 v.d. per cycle; 1 Hz drift rate; 0.5 Hz rotation rate) presented for 2 s each in random order, separated by a blank screen of 2 s duration, with 50% luminance (example calcium response shown in S1A Fig). Frames were averaged during the 2 s stimulus window to estimate the response of a neuron.
Full-field high-contrast drifting gratings (33.33 v.d. per cycle; 1 Hz drift rate) were presented drifting in one of 8 directions for 2 s each in random order, separated by a 6 s period of blank screen with 50% luminance (example calcium response shown in S1B Fig). Frames were averaged during the 2 s stimulus window to estimate the response of a neuron.
Full-field 50% contrast drifting sine-wave gratings (25 v.d. per cycle; 1 Hz drift rate) were presented drifting in one of 16 directions for 1 s each in random order (calcium responses shown in Fig 2). Full-field drifting plaid stimuli were constructed additively from 50% contrast sine-wave grating components (25 v.d. per cycle; 1 Hz drift rate; 1 s duration; Fig 2). Three frames were averaged following the peak response (384 ms window) to estimate the response of a neuron.
Full-field natural movies consisted of a 43 s continuous sequence with three segments (example calcium response shown in S1C Fig).
Full-field contrast-oscillating square-wave gratings and plaid stimuli were composed of bars of 8 v.d. width which oscillated at 2 Hz between black and white on a 50% grey background, and with a spatial frequency of 20 v.d./cycle (example calcium response shown in Fig 5A). On each subsequent oscillation cycle the bars locations shifted phase by 180°. Static gratings were used to avoid introducing a movement component into the stimulus. A base orientation for the gratings of either horizontal or vertical was chosen, and five orientations spanning ±40 deg. around the base orientation were used. Contrast-oscillating plaids were composed of every possible combination of the five oscillating grating stimuli, giving 5 grating and 10 plaid stimuli for each experiment. A single trial consisted of a blank period (50% luminance screen) presented for 20 s, as well as presentations of each of the gratings and plaids for 5 s each, preceded by 5 s of a blank 50% luminance screen, all presented in random order. Frames from 0.25 s to 4.75 s during the stimulus period were averaged to estimate the response of a neuron.
Analysis of two-photon calcium imaging data was conducted in Matlab using the open-source FocusStack toolbox [33]. During acquisition, individual two-photon imaging trials were visually inspected for Z-axis shifts of the focal plane. Affected trials were discarded, and the focal plane was manually shifted to align with previous trials before acquisition continued. Frames recorded from a single region were composed into stacks, and spatially registered with the first frame in the stack to correct lateral shifts caused by movement of the animal. Only pixels for which data was available for every frame in the stack were included for analysis. A background fluorescence region was selected in the imaged area, such as the interior of a blood vessel, and the spatial average of this region was subtracted from each frame in the stack. The baseline fluorescence distribution for each pixel was estimated by finding the mean and standard deviation of pixel values during the 10 s blank periods, separately for each trial. Regions of interest (ROIs) were selected either manually, or by performing low-pass filtering of the OGB (green) and sulforhodamine (red) channels, subtracting red from green and finding the local peaks of the resulting image.
A general threshold for responsivity was computed to ensure that ROIs considered responsive were not simply due to neuropil activity. The responses of all pixels outside any ROI were collected (defined as “neuropil”), and the Z-scores of the mean ΔF/F0 responses during single visual stimulus presentations were computed per pixel, against the baseline period. A threshold for single-trial responses to be deemed significant (ztrial) was set by finding the Z-score which would include only 1% of neuropil responses (α = 1%). A similar threshold was set for comparison against the strongest response of an ROI, averaged over all trials (zmax). These thresholds always exceeded 3, implying that single-trial responses included for further analysis were at least 3 standard deviations higher than the neuropil response. Note that this approach does not attempt to subtract neuropil activity, but ensures that any ROI used for analysis responds to visual stimuli with calcium transients that can not be explained by neuropil contamination alone.
The response of an ROI to a stimulus was found on a trial-by-trial basis by first computing the spatial average of the pixels in an ROI for each frame. The mean of the frames during the blank period preceding each trial was subtracted and used to normalise responses (ΔF/F0), and the mean ΔF/F0 of the frames during the analysed trial period was computed. The standard deviation for the baseline of a neuron was estimated over all ΔF/F0 frames from the long baseline period and the pre-trial blank periods. ROIs were included for further analysis if the ROI was visually responsive according to trial Z-scores (maximum response > zmax) and reliable (trial response > ztrial for more than half of the trials). The response of a neuron to a stimulus was taken as the average of all single-trial ΔF/F0 responses.
Receptive fields of neurons recorded under natural movie and drifting grating stimulation were characterised by presenting small, masked high-contrast drifting gratings from a 5 × 5 array, in random order (see above; S1A Fig). A receptive field for each neuron was estimated by a Gaussian mixture model, composed of circularly symmetric Gaussian fields (ρ = 7.5 v.d.) placed at each stimulus location and weighted by the response of the neuron to the grating stimulus at that location. The centre of the receptive field was taken as the peak of the final Gaussian mixture. Neurons were included for further analysis if the centre of their receptive field lay within a 7.5 v.d. circle placed at the centre of the natural movie visual stimulus. Example single-trial and trial-averaged calcium responses to natural movie stimuli are shown in S1C Fig.
The similarity in response between two neurons was measured independently for grating and plaid stimuli. The set of grating responses for each neuron were composed into vectors R1g and R2g, where each element of a vector was the trial-averaged response of a neuron to a single grating orientation. The similarity in grating responses between two neurons was then given by the Pearson’s correlation coefficient between R1g and R2g: ρg = corr(R1g, R2g) (see S1B Fig, inset). The similarity in response to plaid stimuli was computed analogously over the sets of trial-averaged plaid responses R1p and R2p: ρp = corr(R1p, R2p) (see Fig 5A, inset). Similarity was only measured between neurons recorded in the same imaging site.
The similarity between neurons in their responses to movie stimuli (ρm) was measured by computing the signal correlation as follows. The calcium response traces for a pair of neurons were averaged over trials. The initial 1 s segment of the traces following the onset of a movie segment were excluded from analysis, to reduce the effect of transient signals in response to visual stimulus onset on analysed responses. The Pearson’s correlation coefficient was then calculated between the resulting pair of traces (ρm; see S1C Fig, inset). Note that correlations introduced through neuropil contamination were not corrected for, with the result that the mean signal correlation is positive rather than zero. For this reason we used thresholds for “high” correlations based on percentiles of the correlation distribution, rather than an absolute correlation value. The similarity between neurons in their responses to flashed natural stimuli (ρCa; S1F Fig) was measured as the linear correlation between the vector of responses of a single neuron to a set of 1800 natural stimuli [24].
The Orientation Selectivity Index (OSI) of a neuron was estimated using the formula OSI = [max(Rg)−min(Rg)]/sum(Rg), where Rg is the set of responses of a single neuron to the set of grating stimuli. The OSI of a neuron ranges from 0 to 1, where a value of 1 indicates that a neuron responds only to a single grating stimulus; a value of 0 indicates equal, nonselective responses to all grating stimuli.
The Plaid Selectivity Index (PSI) of a neuron, describing how selective a neuron is over a set of plaid stimuli, was calculated using the formula PSI = 1−[−1 + ∑jRp,j/max(Rp)]/[#(Rp)−1] where #(Rp) is the number of stimuli in Rp [30]. The PSI of a neuron ranges from 0 to 1, where a value of 1 indicates a highly selective response, where a neuron responds to only a single plaid stimulus; a value of 0 indicates equal, nonselective responses to all plaid stimuli.
A plaid Modulation Index (MI), describing the degree of facilitation or suppression of a neuron in response to plaid stimuli, was calculated using the formula MI = [max(Rp)−max(Rg)]/[max(Rp)+max(Rg)], where Rp is the set of responses of a single neuron to the set of plaid stimuli [30]. The MI of a neuron ranges from -1 to 1. Values of MI < 0 indicate stronger responses to grating stimuli compared with plaid stimuli; values of MI > 0 indicate stronger responses to plaid stimuli. A value of MI = -1 indicates that a neuron responds only to grating stimuli; a value of MI = 1 indicates that a neuron responds only to plaid stimuli.
The proportion of facilitating and suppressing neurons was compared between mouse V1 and model responses using two-tailed Fisher’s exact tests. The population of responsive neurons was divided into three groups: facilitating (MI > 0.05); suppressing (MI < -0.05); and non-modulated (-0.05 < = MI < = 0.05). These categories were arranged into three 2 × 3 contingency tables, with each table tested to compare facilitation and suppression between mouse V1 and one model.
We used single-cell, single-trial responses to oscillating contrast grating stimuli to explore whether we could distinguish between correlated and decorrelated responses to plaid stimuli, given experimental variability and noise. For each cell in the experimentally-recorded data set, we used the set of grating responses Rg to generate plaid responses Rp for the same cell, under the assumption that the response to a plaid was linearly related to the sum of the responses to the two grating components. For each plaid, we randomly selected a single-trial response for each of the grating components of the plaid. The predicted single-trial plaid response was the sum of the two grating responses. We generated 100 bootstrap samples for each experimental population, with each sample consisting of the same number of trials and neurons as the experimental population. We then quantified the relationship between grating and plaid responses as described for the experimental data.
We used a sample size commensurate with those used in the field, and sufficient for statistical analysis of our observations. No explicit sample size computation was performed. Two-sided, non-parametric statistical tests were used unless stated otherwise in the text.
|
10.1371/journal.pgen.1000260 | Dissection of a QTL Hotspot on Mouse Distal Chromosome 1 that Modulates Neurobehavioral Phenotypes and Gene Expression | A remarkably diverse set of traits maps to a region on mouse distal chromosome 1 (Chr 1) that corresponds to human Chr 1q21–q23. This region is highly enriched in quantitative trait loci (QTLs) that control neural and behavioral phenotypes, including motor behavior, escape latency, emotionality, seizure susceptibility (Szs1), and responses to ethanol, caffeine, pentobarbital, and haloperidol. This region also controls the expression of a remarkably large number of genes, including genes that are associated with some of the classical traits that map to distal Chr 1 (e.g., seizure susceptibility). Here, we ask whether this QTL-rich region on Chr 1 (Qrr1) consists of a single master locus or a mixture of linked, but functionally unrelated, QTLs. To answer this question and to evaluate candidate genes, we generated and analyzed several gene expression, haplotype, and sequence datasets. We exploited six complementary mouse crosses, and combed through 18 expression datasets to determine class membership of genes modulated by Qrr1. Qrr1 can be broadly divided into a proximal part (Qrr1p) and a distal part (Qrr1d), each associated with the expression of distinct subsets of genes. Qrr1d controls RNA metabolism and protein synthesis, including the expression of ∼20 aminoacyl-tRNA synthetases. Qrr1d contains a tRNA cluster, and this is a functionally pertinent candidate for the tRNA synthetases. Rgs7 and Fmn2 are other strong candidates in Qrr1d. FMN2 protein has pronounced expression in neurons, including in the dendrites, and deletion of Fmn2 had a strong effect on the expression of few genes modulated by Qrr1d. Our analysis revealed a highly complex gene expression regulatory interval in Qrr1, composed of multiple loci modulating the expression of functionally cognate sets of genes.
| A major goal of genetics is to understand how variation in DNA sequence gives rise to differences among individuals that influence traits such as disease risk. This is challenging. Most traits are the result of a complex interplay of genetic and environmental factors. One of the first steps in the path from DNA to these complex traits is the production of mRNA molecules. Understanding how sequence differences modulate expression of different RNAs is fundamental to understanding the molecular origins of complex traits. Here, we combine classic gene mapping methods with microarray technology to characterize and quantify RNA levels in different crosses of mice. We focused on a hotspot on chromosome 1 that controls the expression of a large number of different types of RNAs in the brain. This hotspot also controls many disease traits, including anxiety levels, and vulnerability to seizure in mice and humans. We show that this hotspot is made up of several distinct functional regions, one of which has an unusually strong and selective effect on aminoacyl-tRNA synthetases and other genes involved in protein translation.
| The distal part of mouse Chr 1 harbors a large number of QTLs that generate differences in behavior. Open field activity [1], fear conditioning [2], rearing behavior [3], and several other measures of emotionality [4],[5] have been repeatedly mapped to distal Chr 1. This region is also notable because it appears to influence responses to a wide range of drugs including ethanol [6], caffeine [7], pentobarbital [8], and haloperidol [9]. In addition to the behavioral traits, a number of metabolic, physiological and immunological phenotypes have been mapped to this region (table 1) [10]–[36]. This QTL rich region on mouse distal Chr 1 exhibits reasonably compelling functional and genetic concordance with the orthologous region on human Chr 1q21–q23. Prime examples of genes in this region that have been associated with similar traits in mouse and human are Rgs2 (anxiety in both species), Apoa2 (atherosclerosis), and Kcnj10 (seizure susceptibility) [37]–[42].
Studies of gene expression in the central nervous system (CNS) of mice have revealed major strain differences in the expression level of numerous genes located on distal Chr 1, e.g., Copa, Atp1a2, and Kcnj9 [26], [43]–[45]. These differentially expressed genes are strong candidates for the behavioral and neuropharmacological traits that map to this region. We have recently shown that sequence variants near each of these candidate genes are often responsible for the prominent differences in expression [26],[46],[47]. In other words, sequence differences near genes such as Kcnj9 cause expression to differ, and variation in transcript level maps back to the location of the source gene itself. Transcripts of this type are associated with cis-QTLs.
These expression genetic studies have also uncovered another unusual characteristic of mouse distal Chr 1. In addition to the extensive cis-effects, a large number of transcripts of genes located on other chromosomes map into this same short interval on distal Chr 1 [47],[48]. These types of QTLs are often referred to as trans-QTLs. The clustering of trans-QTLs to distal Chr 1 has been replicated in multiple crosses and CNS microarray datasets [47]. We refer to this region of Chr 1, extending from Fcgr3 (172.5 Mb) to Rgs7 (177.5 Mb) as the QTL-rich region on Chr 1, or Qrr1. It is possible that these modulatory effects on expression are the first steps in a cascade of events that are ultimately responsible for many of the prominent differences in behavior and neuropharmacology. For example, Qrr1 modulates the expression of several genes that have been implicated in seizure (e.g., Scn1b, Pnpo, Cacna1g), and this may be a basis for the strong influence Qrr1 has on seizure susceptibility [41].
In this study, we exploited 18 diverse array datasets derived from different mouse crosses to systematically dissect the expression QTLs in Qrr1. The strong trans effects are consistently detected in CNS tissues of C57BL/6J (B6)×DBA/2J (D2) and B6×C3H/HeJ (C3H) crosses, but are largely absent in ILS/Ibg (ILS)×ISS/Ibg (ISS) and C57BL/6By (B6y)×BALB/cBy (BALB), and in all non-neural tissues we have examined. We applied high-resolution mapping and haplotype analysis of Qrr1 using a large panel of BXD recombinant inbred (RI) strains that included highly recombinant advanced intercross RI lines. Our analyses revealed multiple distinct loci in Qrr1 that regulate gene expression specifically in the CNS. The distal part of Qrr1 (Qrr1d) has a strong effect on the expression of numerous genes involved in RNA metabolism and protein synthesis, including more than half of all aminoacyl-tRNA synthetases. Fmn2 and Rgs7, and a cluster of tRNAs are the strongest candidates in Qrr1d.
The Chr 1 interval, from 172–178 Mb, harbors 32 relatively precisely mapped QTLs for classical traits such as alcohol dependency, escape latency, and emotionality (Mouse Genome Informatics at www.informatics.jax.org, Table 1). To compare the enrichment of QTLs in Qrr1 with that in other regions, we counted classical QTLs in 100 non-overlapping intervals covering almost the entire autosomal genome (table S1). These intervals were selected to contain the same number of genes as Qrr1. Numbers of QTLs ranged from 0 to 23, and averaged at 9.16±5.37 (SD). Compared to these regions, Qrr1 had the highest QTL number, over 4 SD above the mean, and over three times higher than average.
In this section, we summarize the number of expression phenotypes that map to Qrr1 in different tissues and mouse crosses. The results are based on the analysis of 18 array datasets that provide estimates of global mRNA abundance in neural and non-neural tissues from six different crosses. These crosses are—(i) BXD RI and advanced intercross RI strains derived from B6 and D2, (ii) CXB RI strains derived from B6y×BALB, (iii) LXS RI strains derived from ILS and ISS, (iv) B6×C3H F2 intercrosses, and (v & vi) two separate B6×D2 F2 intercrosses. These datasets were generated by collaborative efforts over the last few years [46], [47], [49]–[52] and some were generated more recently (e.g., the Illumina datasets for BXD striatum and LXS hippocampus, and BXD Hippocampus UMUTAffy Exon Array dataset). All datasets can be accessed from GeneNetwork (www.genenetwork.org).
We mapped loci that modulate transcript levels and selected only those transcripts that have peak QTLs in Qrr1 with a minimum LOD score of 3. This corresponds to a generally lenient threshold with genome-wide p-value of 0.1 to 0.05, but corresponds to a highly significant pointwise p-value. Because we are mainly interested in testing a short segment on Chr 1, a pointwise (region-wise) threshold is more appropriate to select those transcripts that are likely to be modulated by Qrr1. Qrr1 covers approximately 0.2% of the genome and extends from Fcgr3 (more precisely, SNP rs8242852 at 172.887364 Mb using Mouse Genome Assembly NCBI m36, UCSC Genome Browser mm8) through to Rgs7 (SNP rs4136041 at 177.273526 Mb). We defined this region on the basis of the large number of transcripts that have maximal LOD scores associated with markers between these SNPs.
Hundreds of transcripts map to Qrr1 with LOD scores ≥3 in neural tissue datasets of BXD RI strains, B6D2F2 intercrosses, and B6C3HF2 intercrosses (table 2). The QTL counts in Qrr1 are far higher than the average of 15 to 35 expression QTLs in a typical 6 Mb interval. The fraction of QTLs in Qrr1 is as high as 14% of all trans-QTLs, and 5% of all cis-QTLs in the whole genome (table 2). The enrichment in trans-QTLs in Qrr1 is even more pronounced when the QTL selection stringency is increased to a LOD threshold of 4 (genome-wide p-value of approximately 0.01). For example, 27% of all highly significant trans-QTLs in the BXD cerebellum dataset are in Qrr1 (table 2). The BXD hippocampus dataset that was assayed on the Affymetrix Exon ST array is an exception—there are over a million probe sets in this array, and the percent enrichment of QTLs in Qrr1 appears to be relatively low. Nevertheless, about 1000 transcripts map to Qrr1 in this exon dataset.
In contrast to the CNS datasets, relatively few transcripts map to Qrr1 in non-neural tissues of the BXD strains and B6C3HF2 intercrosses. While the number of cis-QTLs is still relatively high (1–3%), Qrr1 has limited or no trans-effect in these datasets (table 2).
Qrr1 does not have a strong trans-effect in the LXS and CXB hippocampus datasets (table 2). This indicates that the sequence variants underlying the trans-QTLs do not segregate to nearly the same extent in the LXS and CXB RI panels as they do in B6×D2 and B6×C3H crosses. This contrast among crosses can be exploited to parse Qrr1 into sub-regions and identify stronger candidate genes.
The trans-QTLs in Qrr1 are highly replicable. A large fraction of the transcripts, in some cases represented by multiple probes or probe sets, map to Qrr1 in multiple CNS datasets. For example, there are 747 unique trans-QTLs with LOD scores greater than 4 (genome-wide p-value≤0.01) in the BXD hippocampus dataset (assayed on Affymetrix M430v2 arrays). Out of these highly significant trans-QTLs, 155 are in Qrr1 and the remaining 592 are distributed across the rest of the genome (figure 1). We compared the trans-QTLs in the hippocampus dataset with a similar collection of trans-QTLs (LOD≥4) in the cerebellum dataset (assayed on Affymetrix M430 arrays). Only 101 trans-QTLs in the hippocampus are replicated in the cerebellum (for trans-QTLs that were declared as common, the average distance between peak QTL markers in the two datasets is 1.6 Mb). But it is remarkable that of the subset of common trans-QTLs, 64 are in Qrr1 (figure 1). The replication rate of trans-QTLs in Qrr1 is therefore about 6-fold higher relative to the rest of the genome. When we compared the BXD hippocampus dataset with the B6C3HF2 brain dataset (assayed on Agilent arrays), we found 54 trans-QTLs common to both datasets (for the common trans-QTLs, the average distance between peak markers in the two datasets is 2.7 Mb). Strikingly, out of the 54 trans-QTLs common to both crosses, 52 are in Qrr1 (figure 1).
Among the transcripts with the most consistent trans-QTLs are glycyl-tRNA synthetase (Gars), cysteinyl-tRNA synthetase (Cars), asparaginyl-tRNA synthetase (Nars), isoleucyl tRNA synthetase (Iars), asparagine synthetase (Asns), and activating transcription factor 4 (Atf4). These transcripts map to Qrr1 in almost all datasets in which the strong trans-effect is detected. Gars, Cars, and Nars are aminoacyl-tRNA synthetases (ARS) that charge tRNAs with amino acids during translation. Asns and Atf4 are also involved in amino acid metabolism—Asns is required for asparagine synthesis and is under the regulation of Atf4, which in turn is sensitive to cellular amino acid levels [53]. Other transcripts that consistently map as trans-QTLs to Qrr1 include brain expressed X-linked 2 (Bex2), splicing factor Sfrs3, ribonucleoproteins Snrpc and Snrpd1, ring finger protein 6 (Rnf6), and RAS oncogene family member Rab2.
Qrr1 contains 164 known genes. The proximal part of Qrr1 is gene-rich and has several genes with high expression in the CNS (e.g. Pea15, Kcnj9, Kcnj10, Atp1a2). The middle to distal part of Qrr1 is relatively gene sparse and consists mostly of clusters of olfactory receptors and members of the interferon activated Ifi200 gene family. Though comparatively gene sparse, the middle to distal part of Qrr1 contains a small number of genes that have high expression in the CNS—Igsf4b, Dfy, Fmn2, and Rgs7.
A subset of 35 genes were initially selected as high priority candidates based on the number of known and inferred sequence differences between the B6 allele (B) and D2 allele (D) and based on expression levels in multiple CNS datasets (table 3). Eleven of these candidates contain missense SNPs segregating in B6×D2 crosses. We also scanned Qrr1 for variation in copy number [54],[55]. Graubert et al. [55] reported segmental duplication in Qrr1 with a copy number gain in D2 compared to B6 near the intelectin 1 (Itlna) gene at 173.352 Mb. We failed to detect any expression signatures of a copy number variation around Itlna in any of the GeneNetwork datasets. However, we did identify an apparent 150 kb deletion across the Ifi200 gene cluster (175.584–175.733 Mb). Affymetrix probe sets 1426906_at, 1452231_x_at, and 1452349_x_at detect Ifi204 and Mnda transcripts in B6 but not in D2. The expression difference is robust enough to generate cis-QTLs with very high LOD scores (>40). This gene cluster has low expression in the CNS (Affymetrix declares this probe sets to be “not present”), but high expression in tissues such as hematopoietic stem cells and kidney, in which the trans-effect of Qrr1 is not detected. The Ifi200 gene cluster was therefore excluded as a high priority candidate.
Transcripts of 26 of the 35 selected candidate genes map as cis-QTLs (LOD≥3) in the BXD CNS datasets (table 3). These putatively cis-regulated genes are among the strongest candidates in the QTL interval. The D allele in Qrr1 has the positive effect on the expression of Sdhc, Ndufs2, Adamts4, Dedd, Pfdn2, Ltap, Pea15, Atp1a2, Kcnj9, Kcnj10, Igsf4b, and Grem2. Increase in expression caused by the D allele ranges from about 10% for Adamts4 to over 2-fold for Atp1a2. In contrast, the B allele has the positive effect on the expression of Pcp4l1, Fcer1g, B4galt3, Ppox, Ufc1, Nit1, Usf1, Copa, Pex19, Wdr42a, Igsf8, Dfy, Fmn2, and Rgs7. Increase in expression caused by the B allele ranges from about 7% for Usf1 to 40% for Pex19.
Individual probes were screened to assess if the strong cis-effects are due to hybridization artifacts caused by SNPs in probe targets. Thirteen candidate genes with cis-QTLs were then selected for further analysis and validation of cis-regulation by measuring allele specific expression (ASE) difference [56]. This method exploits transcribed SNPs, and uses single base extension to assess expression difference in F1 hybrids. By means of ASE, we validated the cis-regulation of 10 candidate genes—Ndufs2, Nit1, Pfdn2, Usf1, Copa, Atp1a2, Kcnj9, Kcnj10, Dfy, and Fmn2 (table 4). Adamts4 and Igsf4b failed to show significant allelic expression difference. In the case of Ufc1, the polarity of the allele effect failed to agree with the ASE result (D positive at p-value = 0.02).
The BXD CNS datasets were generated from a combined panel of conventional RI strains and advanced RI strains that were derived by inbreeding advanced intercross progeny. The advanced RIs have approximately twice as many recombinations compared to standard RIs and the merged panel offers over a 3-fold increase in mapping resolution [57]. This expanded RI set combined with the relatively high intrinsic recombination rate within Qrr1 [58] provides comparatively high mapping resolution. Mapping precision can be empirically determined by analyzing cis-QTLs in multiple large datasets, particularly the BXD Hippocampus Consortium, UMUTAffy Hippocampus, and Hamilton Eye datasets. These three datasets were selected because they have expression measurements from six BXD strains with recombinations in Qrr1. These strains—BXD8, BXD29, BXD62, BXD64, BXD68, and BXD84—collectively provide six sets of informative markers and divide Qrr1 into six non-recombinant segments, labeled as segments 1–6 (haplotype structures shown in figure 2).
As cis-acting regulatory elements are usually located within a few kilobases of a gene's coding sequence [59], we used the cis-QTLs as an internal metric of mapping precision by measuring the offset distance between a cis-QTL (position of peak QTL marker) and the parent gene (figure 3). For cis-QTLs with LOD scores between 3–4 (genome-wide p-value of 0.1–0.01) the mean gene-to-QTL peak distance is 900 kb. The offset decreases to a mean of 640 kb for cis-QTLs with LOD scores greater than 4 (p-value<0.001). Very strong cis-QTLs with LOD scores greater than 11 (p-value<10−6) have a mean gene-to-QTL peak distance of only 450 kb. In all, 60% of cis-QTLs we examined have peak linkage on markers located precisely in the same non-recombinant segment as the parent gene, and 30% have peak linkage on markers in a segment adjacent to the parent gene (dataset S1). These cis-QTLs provide an empirical metric of mapping precision within Qrr1.
Mapping precision of cis-QTLs is comparatively higher in the BXD hippocampus dataset (average offset of only 410 kb), and we used this set to examine the trans-QTLs (LOD≥3) at higher resolution. The trans-QTLs in Qrr1 were parsed into subgroups based on the location of peak LOD score markers (figure 4). This method of resolving trans-QTLs effectively grouped subsets of transcripts into functionally related cohorts. For instance, all the QTLs for the aminoacyl-tRNA synthetases (ARS) have peak LOD scores only within the distal three segments of Qrr1 (figure 5). This consistency in QTL peaks for transcripts of the same gene family is itself a good indicator of mapping precision. In addition to the ARS, numerous other genes involved in amino acid metabolism and translation map to the distal part of Qrr1 (e.g., Atf4, Asns, Eif4g2, and Pum2).
We divided the trans-QTLs into two broad subgroups—those with peak QTLs on markers in the proximal part of Qrr1 (Qrr1p; 172–174.5 Mb or segments 1, 2, 3 in figure 2), and those with peak QTLs on markers in the distal part of Qrr1 (Qrr1d; 174.5–177.5 Mb or segments 4, 5, and 6 in figure 2). While Qrr1p is relatively gene-rich, only 35% of the trans-QTLs (129 out of 365 probe sets) have peak LOD scores in this region. The majority of trans-QTLs—about 65% (236 out of 365 probe sets)—have peak QTLs in the relatively gene-sparse Qrr1d.
The two subsets of transcripts—those with trans-QTLs in Qrr1p and those with trans-QTLs in Qrr1d—were analyzed for overrepresented gene functions using the DAVID functional annotation tool (http://david.abcc.ncifcrf.gov/). This revealed distinct gene ontology (GO) categories enriched in the two subsets (dataset S2). Enriched GOs among the transcripts modulated by Qrr1p include GTPase-mediate signal transduction (modified Fisher's exact test p = 0.001), and structural constituents of ribosomes (p = 0.003). Transcripts modulated by Qrr1d are highly enriched in genes involved in RNA metabolism (p = 4×10−7), tRNA aminoacylation (p = 1×10−5) and translation (p = 2×10−5), RNA transport (p = 0.003), cell cycle (p = 0.004), and ubiquitin mediated protein catabolism (p = 0.006). Other GO categories show enrichment in both Qrr1p and Qrr1d. For example, genes involved in RNA metabolism and ubiquitin-mediated protein catabolism are also overrepresented among the transcripts modulated by Qrr1p (p = 0.002 for RNA metabolism and p = 0.005 for ubiquitin-protein ligases). This may either be due to limitations in QTL resolution, or due to multiple loci in Qrr1p and Qrr1d controlling these subsets of transcripts.
A remarkable number of transcripts of the ARS gene family map to Qrr1. A total of 16 ARS transcripts have trans-QTLs at a minimum LOD score of 3 in one or multiple BXD, B6D2F2, and B6C3H CNS datasets (table 5). In almost all cases, QTLs peak on markers on the distal part of Qrr1. Except for Hars, the B allele in Qrr1 consistently increases expression by 10% to 30%. In the case of Hars, the D allele has the positive additive effect and increases expression by about 10%.
We examined all probes or probe sets that target ARS and ARS-like genes in the B6×D2 CNS datasets. The Affymetrix platform measures the expression of 34 ARS and ARS-like genes; 24 of these map to Qrr1 at LOD scores ranging from a low of 2 to a high of 12. Even in the case of the suggestive trans-QTLs (i.e., LOD values between 2 and 3), the B allele in Qrr1 has the positive effect on expression. The ARS family is also highly represented among trans-QTLs in the B6C3HF2 brain dataset. Thirty-seven probes in this dataset target the tRNA synthetases, eleven of these have trans-QTLs in Qrr1d (LOD scores ranging from 2 to 20), and almost all have a B positive additive effect (exceptions are Hars and Qars). The co-localization of trans-QTLs to Qrr1d, the general consensus in parental allele effect, and their common biological function indicate that there is a single QTL in the distal part of Qrr1 modulating the expression of the ARS. It is crucial to note that this genetic modulation is only detected in CNS tissues.
In the LXS hippocampus dataset, Qrr1 has only a limited trans-effect on gene expression. Despite the weak effect, expression of Dars2 (probe ID ILM580427) maps to the distal part of Qrr1 at a LOD of 3. Although this is only a weak detection of the ARS QTL in the LXS dataset, it nonetheless demonstrates the strong regulatory effect of Qrr1 on the expression of this gene family. In the case of the CXB hippocampus dataset, not a single trans-QTL for the ARS is detected in Qrr1.
In addition to the high overrepresentation of transcripts involved in translation and RNA metabolism, several transcripts known to be transported to neuronal processes or involved in RNA transport also map to Qrr1d, including Camk2a, Bdnf, Cdc42, Eif4e, Eif4g2, Hnrpab, Ppp1cc, Pabpc1, Eif5, Kpnb1, Rhoip3, Stau2, and Pum2 [60]–[63]. An interesting example is provided by the brain derived neurotrophic factor (Bdnf). Two alternative forms of Bdnf mRNA are known—one isoform has a long 3′ UTR and is specifically transported into the dendrites; the other isoform has a short 3′ UTR and remains primarily in the somatic cytosol [64]. The Affymetrix M430 arrays contain two different probe sets that target these Bdnf isoforms. Probe set 1422169_a_at targets the distal 3′ UTR and is essentially specific for the dendritic isoform, and probe set 1422168_a_at targets a coding sequence common to both isoforms. Although both probe sets detect high expression signal in the hippocampus, only the dendritic isoform maps as a trans-QTL to Qrr1d. This enrichment in transcripts that are transported to neuronal processes raises the possibility that this CNS specific trans-effect may be related to local protein synthesis.
Prompted by the many ARS transcripts that consistently map to Qrr1d, we searched the genomic tRNA database [65] for tRNAs in this region. Interestingly, distal Chr 1 is one of many tRNA hotspots in the mouse genome and several predicted tRNAs are clustered in the non-coding regions of Qrr1 (figure 2). The majority of these tRNA sequences are in the proximal end of Qrr1, over 2 Mb away from Qrr1d. We scanned the intergenic non-coding regions in Qrr1d for tRNAs using the tRNAscan-SE software [65] and uncovered tRNAs for arginine and serine, and three pseudo-tRNA sequences between genes Igsf4b and Aim2 (175.204–175.257 Mb) in Qrr1d (dataset S3). Transfer RNAs are involved in regulating transcription of the ARS in response to cellular amino acid levels [66] and are functionally highly relevant candidates in Qrr1d. Polymorphism in the tRNA clusters (e.g., possible copy number variants, differences in tRNA species) may have significant impact on the expression of the ARS.
Trans-regulation of large number of transcripts by Qrr1 is a strong feature of crosses between B6 and D2—both the BXD RI set and B6D2F2 intercrosses—and in the B6 and C3H intercrosses. The feature is much weaker in the large LXS RI set and in the small CXB panel. The effect specificity demonstrates that a major source of the Qrr1 signal is generated by variations between B and D, and B and C3H alleles (H) but not by variations between the ILS and ISS alleles (L and S, respectively), and B and BALB alleles (C). This contrast can be exploited to identify sub-regions that underlie the trans-QTLs [67].
SNPs were counted for all four pairs of parental haplotypes—B vs D, B vs H, B vs C, and L vs S—and SNP profiles for the four crosses were compared (figure 6). Qrr1 is a highly polymorphic interval in the B6×D2 crosses. The flanking regions, however, have few SNPs (170–172.25 Mb proximally, and 177.5–179.5 Mb distally) and are almost identical-by-descent between B6 and D2. The B6×BALB crosses, despite being negative for the trans-effect, have moderate to high SNP counts in Qrr1 and share a SNP profile somewhat similar to B6×D2 crosses. The B6×C3H crosses also have moderate to high SNP counts in Qrr1, with a relatively higher SNP count in Qrr1d compared to Qrr1p. In contrast, in the LXS, Qrr1p is more SNP-rich than Qrr1d. Most notably, the segments that harbor the tRNAs and candidates Fmn2, Grem2, and Rgs7 are almost identical by descent between ILS and ISS. This SNP comparison indicates that the strongest trans-effect is from Qrr1d. A possible reason why the trans-effect is not detected in the CXB RI strains, despite being SNP rich in Qrr1, is that the crucial SNPs underlying the trans-QTLs may not be segregating in this cross or that undetected copy number variants make important contributions to the Qrr1 effects. A final explanation may be that the small CXB dataset (13 strains) is simply underpowered.
We used the specificity of cis-QTLs in the multiple crosses to identify higher priority candidates in Qrr1. The assumption is that candidate genes whose transcripts have cis-QTLs (LOD score above 3) in the B6×D2 and B6×C3H crosses but not in the LXS and CXB RI strains are stronger candidates for trans-QTLs that are detected in the former two crosses but not in the latter two crosses. In contrast, cis-QTLs with the inverse cross specificity are less likely to underlie these trans-QTLs. Based on this criterion, there are four high-ranking candidates in Qrr1p—Purkinje cell protein 4-like 1 (Pcp4l1), prefoldin (Pfdn2), WD repeat domain 42 a (Wdr42a), and Kcnj10 (table 3). There are only two high-ranking candidates in Qrr1d—formin 2 (Fmn2), an actin binding protein involved in cytoskeletal organization, and regulator of G-protein signaling 7 (Rgs7) (table 3).
Both Fmn2 and Rgs7 are almost exclusively expressed in the CNS and are high priority candidates for the CNS specific trans-QTLs. A point of distinction between the two candidates is that while expression of Rgs7 maps as a cis-QTL only in the B6×D2 and B6×C3H crosses, expression of Fmn2 maps as a cis-QTL in B6×D2 and B6×C3H crosses, and in the CXB RI strains in which the trans-effect is not detected (table 3). Based on the pattern of specificity of cis-QTLs in multiple crosses, Rgs7 is a more appealing candidate. However, Fmn2 has known missense SNPs that segregate in the B6×D2 (Glu610Asp, Pro1077Leu, Asp1431Glu) and B6×C3H crosses (Val372Ala). There are no known missense mutations in Fmn2 in the CXB and LXS RI strains, and no known missense mutation in Rgs7 in any of the four crosses.
Linkage disequilibrium (LD) is a major confounding factor that limits fine-scale discrimination among physically linked candidates in a QTL. To further evaluate the two high-priority candidates in Qrr1d—Fmn2 and Rgs7—we implemented a partial correlation analysis [68] in which the effect of genotype at Qrr1d was controlled. For this analysis, we computed the partial correlation coefficient between cis-regulated transcripts and each trans-regulated transcript after regression against the Qrr1d genotype. This partial correlation reveals residual variance that links cis candidates with trans targets, independent of genetic variance at Qrr1d. We computed the partial correlation between Rgs7 and Fmn2, and 14 transcripts representative of the different GOs that map to Qrr1d (dataset S4). The highest partial correlations are between Fmn2 and Rnf6 (r = 0.68, p-value<10−13), Atf4 (r = 0.6, p-value<10−9), Asns (r = 0.55, p-value<10−7), Ube2d3 (r = 0.5, p-value<10−6), Hnrpk (r = 0.5, p-value = 10−5), Rab2 (r = −0.5, p-value = 10−5), and Gars (r = 0.5, p-value = 10−5). The strongest correlate of Fmn2 is Rnf6, a gene involved in regulating actin dynamics in axonal growth cones [69]. Although not unequivocal, this analysis provides stronger support for Fmn2 than for Rgs7.
Fmn2 is almost exclusively expressed in the nervous system [70] and is a strong candidate for a trans-effect specific to neural tissues. However, its precise function in the brain has not been established. Fmn2-null mice do not have notable CNS abnormalities [71], but to evaluate a possible role of Fmn2 on expression of genes that map to Qrr1d, we generated array data from brains of Fmn2-null (Fmn2−/−) and coisogenic (Fmn2+/+) 129/SvEv controls. At a stringent statistical threshold (Bonferroni corrected p<0.05), only eight genes have significant expression differences between Fmn2−/− and Fmn2+/+ genotypes (table 6). Five out of the eight genes, including Pou6f1, Usp53, and Slc11a, have trans-QTLs in Qrr1d. Deletion of Fmn2 had the most drastic effect on the expression of the transcription factor gene Pou6f1, a gene implicated in CNS development and regulation of brain-specific gene expression [72],[73]. Expression of Pou6f1 maps as a trans-QTL (at LOD score of 3) to Qrr1d in the hippocampus dataset, and its expression was down-regulated more than 44-fold in the Fmn2−/− line. While the expression analysis of Fmn2-null mice does not definitively link all the trans-QTLs to Fmn2, variation in this gene is likely to underlie some of the trans-QTLs in Qrr1d. The possible compensatory mechanism in the Fmn2-null CNS, and the different genetic background of the mice (129/SvEv) are factors that may have contributed to the weak detection of trans-effects in the knockout line.
We examined the intracellular distribution of FMN2 protein in neurons using immunocytochemical techniques. All hippocampal pyramidal neurons on a culture dish exhibited distinct and fine granular immunoreactivity for FMN2. The cell body itself had the strongest signal (figure 7A). This fine punctate labeling extended into proximal dendrites and could be followed into distal dendrites. In some instances very thin processes, possibly the axons, were also labeled.
The strong trans-effect that Qrr1 has on gene expression is a likely basis for the classical QTLs that map to this region. For example, the major seizure susceptibility QTL (Szs1) has been precisely narrowed to Qrr1p [74]. We found that 10 genes already known to be associated with seizure or epilepsy have trans-QTLs with peak LOD scores near Szs1 and in Qrr1p. These include Scn1b, Cacna1g, Pnpo, and Dapk1 (Table S2) [75]–[84]. In every case, the D allele has the positive additive effect on the expression of these seizure related transcripts, increasing expression 5% to 20%. The two potassium channel genes, Kcnj9 and Kcnj10, are the primary candidates [74]. Both are strongly cis-regulated. The tight linkage between these genes (within 100 kb) limits further genetic dissection, but in situ expression data from the Allen Brain Atlas (ABA, www.brain-map.org) provides us with a powerful complementary approach to evaluate these candidates [85]. Kcnj9 (figure 8A) is expressed most heavily in neurons within the dentate gyrus, whereas Kcnj10 (figure 8B) is expressed diffusely in glial cells in all parts of the CNS. The seizure-related transcripts with trans-QTLs near Szs1 are most highly expressed in neurons, and all have comparatively high expression in the hippocampus. Furthermore, expression patterns of six of the seizure transcripts that map to Qrr1p show spatial correlations with Kcnj9. Dapk1 and Cacna1g (figure 8C) have expression pattern that match Kcnj9 with strong labeling in the dentate gyrus and CA1, and weaker labeling in CA2 and CA3. In contrast, Socs2 (figure 8D), Adora1, Pnpo, and Kcnma1 complement the expression of Kcnj9 with comparatively strong expression in CA2 and CA3, and weak expression in CA1 and dentate gyrus.
Qrr1 is a complex regulatory region that modulates expression of many genes and classical phenotypes. By exploiting a variety of microarray datasets and by applying a combination of high-resolution mapping, sequence analysis, and multiple cross analysis, we have dissected Qrr1 into segments that are primarily responsible for variation in the expression of functionally coherent sets of transcripts. The distal portion of Qrr1 (Qrr1d) has a strong trans-effect on RNA metabolism, translation, tRNA aminoacylation, and transcripts that are transported into neuronal dendrites. Fmn2, Rgs7, and a cluster of tRNAs are strong candidates in Qrr1d. We analyzed gene expression changes in the CNS of Fmn2-null mice and detected a profound effect on the expression of a small number of transcripts that map to Qrr1d, particularly on the expression of the transcription factor Pou6f1. We have shown that the FMN2 protein is highly expressed in the cell body and processes of neurons, and is a high priority candidate in Qrr1d.
The two inwardly rectifying potassium channel genes—Kcnj9 and Kcnj10—are strong candidates for the seizure susceptibility QTL in Qrr1p that has been unambiguously narrowed to the short interval from Atp1a2 to Kcnj10 [74]. In BXD CNS datasets, Qrr1 also modulates the expression of a set of genes implicated in the etiology of seizure and epilepsy, including Pnpo, Scn1b, Kcnma1, Socs2, and Cacna1g. Polymorphisms in the Kcnj9/Kcnj10 interval that influence expression of these genes are excellent candidates for the Szs1 locus.
The in situ expression data in the ABA shows a striking spatial correlation between expression of Kcnj9 and other seizure-related transcripts that have trans-QTLs in Qrr1p. The complementary expression of Kcnj9 and the seizure-related transcripts (figure 8) make Kcnj9 a stronger candidate than Kcnj10. Kcnj9 has over a 2-fold higher expression in D2 [our data],[and cf. 26,86], a seizure prone strain, compared to B6, a relatively seizure resistant strain, suggesting that the proximal cause of Szs1 may be high expression of this gene, perhaps due to the promoter polymorphism discovered by Hitzemann and colleagues [26].
Fine mapping of complex traits have often yielded multiple constituent loci within a QTL interval [87],[88]. Our mapping analyses of expression traits also show that multiple gene variants, rather than one master regulatory gene, cause the aggregation of expression QTLs in Qrr1. Subgroups of genes with tight coexpression can be dissected from the dense cluster of QTLs. Most notable is the strong trans-regulatory effect of Qrr1d on genes involved in amino acid metabolism and translation, including a host of ARS transcripts. However, there are limits to our ability to dissect Qrr1, and genes associated with protein degradation and RNA metabolism map throughout the region. In part this may be due to inadequate mapping resolution, but it may also reflect clusters of functionally related loci and genes [89]. At this stage we are also unable to discern whether there is a single or multiple QTLs within Qrr1d. While it is likely that a single QTL modulates the expression of the ARS, there may be additional gene variants in Qrr1d that modulate other transcripts involved in translation and RNA metabolism. With increased resolving power it may be possible to further subdivide transcripts that map to Qrr1p and Qrr1d into smaller functional modules.
There may be multiple loci in Qrr1 that modulate different stages of protein metabolism in the CNS. Maintenance of cellular protein homeostasis requires finely tuned cross talk between transcription and RNA processing, the translation machinery, and protein degradation [90]–[92], gene functions highly overrepresented among the transcripts that map to Qrr1. While these are generic cellular processes, there are unique demands on protein metabolism in the nervous system. Neurons are highly polarized cells and specialized mechanisms are in place to manage local protein synthesis and degradation in dendrites and axons [93]. The nervous system is also particularly sensitive to imbalances in protein homeostasis [94],[95], a possible reason why the trans-effects of Qrr1 are detected only in neural tissues.
Transfer RNAs are direct biological partners of the ARS, and the cluster of tRNAs in the highly polymorphic intergenic region of Qrr1d (figure 6) is an enticing candidate. In addition to their role in shuttling amino acids, tRNAs also act as sensors of cellular amino acid levels and regulate transcription of genes involved in amino acid metabolism and the ARS [66]. There is tissue specificity in the expression of different tRNA isoforms [96], and we speculate that the tRNA cluster in Qrr1d is specifically functional in neural tissues.
Rgs7, a member of the RGS (regulator of G-protein signaling) family, is another high-ranking candidate in Qrr1d. RGS proteins are important regulators of G-protein mediated signal transduction. Rgs7 is predominantly expressed in the brain and has been implicated in regulation of neuronal excitability and synaptic transmission [97],[98]. Although RGS proteins are usually localized in the plasma membrane, RGS7 has been found to shuttle between the membrane and the nucleus [99]. This implies a role for RGS7 in gene expression regulation in response to external stimuli.
Our final high-ranking candidate in Qrr1d is Fmn2. It codes for an actin binding protein exclusively expressed in the CNS and oocytes, and is involved in the establishment of cell polarity [70],[71]. In Drosophila, the formin homolog, cappuccino, has a role in RNA transport and in localizing the staufen protein to oocyte poles [100]–[102]. It is possible that FMN2 has parallel functions in mammalian neurons. Interestingly, Staufen 2 (Stau2), a gene involved in RNA transport to dendrites [62], maps to Qrr1d in BXD CNS datasets. Furthermore, deletion of formin homologs in yeast results in inhibition of protein translation [103], compelling evidence for an interaction between the protein translation system and formins. Evidence for a role for Fmn2 in dendrites also comes from our immunocytochemical analysis that clearly demonstrates the expression of FMN2 protein in dendrites. Taken together, Fmn2 is a functionally relevant candidate gene in Qrr1d and may be related to RNA transport and protein synthesis in the CNS.
The microarray datasets used in this study (table 2) were generated by collaborative efforts [46], [47], [49]–[52]. All datasets can be accessed from www.genenetwork.org. They provide estimates of global mRNA abundance in neural and non-neural tissues in the BXD, LXS, and CXB RI strains, B6D2F2 intercrosses, and B6C3HF2 intercrosses. Detailed description of each set, tissue acquisition, RNA extraction and array hybridization methods, and data processing and normalization methods are provided in the “Info” page linked to each dataset. In brief, the datasets are:
The conventional BXD RI strains were derived from the B6 and D2 inbred mice [104],[105]. The newer sets of advanced RI strains were derived by inbreeding intercrosses of the RI strains [57]. The parental B6 and D2 strains differ significantly in sequence and have approximately 2 million informative SNP. A subset of 14,000 SNPs and microsatellite markers have been used to genotype the BXD strains [106],[107]. We used 3,795 informative markers for QTL mapping. Thirty such informative markers are in Qrr1 and we queried these markers to identify strains with recombinations in Qrr1; genes with strong cis-QTLs (Sdhc, Atp1a2, Dfy, and Fmn2) were used as additional markers. Smaller sub-sets of markers were used to genotype the two F2 panels (total of 306 markers for the whole brain, and 75 markers for the striatum F2 datasets).
The LXS RI strains were derived from the ILS and ISS inbred strains. They have been genotyped using 13,377 SNPs, and some microsatellite markers [108]. 2,659 informative SNPs and microsatellite markers were used for QTL mapping.
The CXB panel consists of 13 RI strains derived from C57BL/6By and BALB/cBy inbred strains. A total of 1384 informative markers were used for QTL mapping.
The B6×C3H/HeJ F2 intercrosses have been genotyped using 13,377 SNPs and microsatellite markers, and 8,311 informative markers were used for QTL mapping.
Majority of the BXD and LXS tissues (cerebellum, eye, forebrain, hippocampus, kidney, liver, and striatum for the HQF Illumina dataset) were dissected at the University of Tennessee Health Science Center (UTHSC). Mice were housed at the UTHSC in pathogen-free colonies, at an average of three mice per cage. All animal procedures were approved by the Animal Care and Use Committee. Mice were killed by cervical dislocation, and tissues were rapidly dissected and placed in RNAlater (Ambion, www.ambion.com) and kept overnight at 4° C, and subsequently stored at −80 degree C. Tissue were then processed at UTHSC or shipped to other locations for processing.
For the tissues that were processed at UTHSC (all BXD and LXS CNS tissues except HBP Affymetrix striatum), RNA was isolated using RNA STAT-60 (Tel-Test Inc., www.tel-test.com) as per manufacturer's instructions. Samples were then purified using standard sodium acetate methods prior to microarray hybridization. The eye samples required additional purification steps to remove eye pigment; this was done using the RNeasy MinElute Cleanup Kit (Qiagen, www.qiagen.com). RNA purity and concentration was evaluated with a spectrophotometer using 260/280 nm absorbance ratio, and RNA quality was checked using Agilent Bioanalyzer 2100 prior to hybridization. Array hybridizations were then done according to standard protocols.
We have re-annotated a majority of Affymetrix probe sets to ensure more accurate description of probe targets. Each probe set represents a concatenations of eleven 25-mer probes, and these have been aligned to the NCBI built 36 version of the mouse genome (mm8 in UCSC Genome Browser) by BLAT analysis. We have also re-annotated the Illumina probes and incorporated these annotations into GeneNetwork. Each probe in the Illumina Mouse–6 and Mouse–6.1 arrays is 50 nucleotides in length, and these have been aligned to NCBI built 36.
We used the strain average expression signal detected by a probe or probe set. QTL mapping was done for all transcripts using QTL Reaper [47]. The mapping algorithm combines simple regression mapping, linear interpolation, and standard Haley-Knott interval mapping [109]. QTL Reaper performs up to a million permutations of an expression trait to calculate the genome-wide empirical p-value and the LOD score associated with a marker. We selected only those transcripts that have highest LOD scores, i.e., genome-wide adjusted best p-values, on markers located on Chr 1 from 172 to 178 Mb. This selected transcripts that are primarily modulated by Qrr1 but excluded transcripts that have QTLs in Qrr1 but have higher LOD scores on markers located on other chromosomal regions. Cis- and trans-QTLs were distinguished based on criteria described by Peirce et al. [47]. To identify trans-QTLs common to multiple datasets, we selected probes/probe sets that target the same genes and have peak LOD scores within 10 Mb in the different datasets.
We screened all Affymetrix probe sets with cis-QTLs in Qrr1 for SNPs in target sequences. This step was taken to identity false cis-QTLs caused by differences in hybridization. As probe design is based on the B6 sequence, such spurious cis-QTLs show high expression for the B allele, and low expression for the D allele. Our screening identified only two probe sets in which SNPs result in spurious local QTLs—1429382_at (Tomm40l), and 1452308_a_at (Atp1a2). The majority of cis-QTLs in Qrr1 are likely to be due to actual differences in mRNA abundance. We did not detect a bias in favor of the B allele on cis-regulated expression and the ratio of transcripts with B- and D- positive additive effects is close to 1∶1.
To measure expression difference between the B and D alleles, we exploited transcribed SNPs to capture allelic expression difference in F1 hybrids [56] using a combination of RT-PCR and a single base extension technology (SNaPshot, Applied Biosystems, www.appliedbiosystems.com). For each transcript we analyzed, Primer 3 [110] was used to design a pair of PCR primers that target sequences on the same exon and flanking an informative SNP.
We prepared four pools of RNA from the hippocampus, and four pools of genomic DNA from the spleen of F1 hybrids (male and female B6×D2 and D2×B6 F1 hybrids). To avoid contamination by genomic DNA, the four RNA pools were treated with Turbo DNase (Ambion, www.ambion.com), and then first strand cDNA was synthesized (GE Healthcare, www.gehealthcare.com). The genomic DNA samples were used as controls, and both cDNA and genomic DNA samples were tested concurrently using the same assay to compare expression levels of B and D transcripts.
We amplified the cDNA and genomic DNA samples using GoTaq Flexi DNA polymerase (Promega Corporation, www.promega.com). PCR products were purified using ExoSap-IT (USB Corporation, www.usbweb.com) followed by SNaPshot to extend primer by a single fluorescently labeled ddNTPs. Fluorescently labeled products were purified using calf intestinal phosphatase (CIP, New England BioLabs, www.neb.com) and separated by capillary electrophoresis on ABI3130 (Applied Biosystems). Quantification was done using GeneMapper v4.0 software (Applied Biosystems), and transcript abundance was measured by peak intensities associated with each allele. Ratio of B and D allele in both cDNA and gDNA pools was computed, and t-test (one tail, unequal variance) was done to validate expression difference and polarity of parental alleles.
GeneNetwork has compiled SNP data from different sources—Celera (http://www.celera.com), Perlegen/NIEHS (http://mouse.perlegen.com/mouse/download.html), BROAD institute (http://www.broad.mit.edu/snp/mouse), Wellcome–CTC [107], dbSNP, and Mouse Phenome Database (http://www.jax.org/phenome/SNP). SNP counts were done on the GeneNetwork SNP browser.
A partial correlation is the correlation between X and Y conditioned on one or more control variables. In this study, first order partial correlation was used to detect the interaction between trans-regulated transcripts and cis-regulated candidate genes conditioned on the genotype (marker rs8242481 at 175.058 Mb). If x, y and z are trans-regulated transcripts, cis-regulated transcript, and genotype in the QTL, respectively, then the first order partial correlation coefficient is calculated as—where rxy can be either Pearson correlation or Spearman's rank correlation between x and y. We employed the Spearman's rank correlation because the expression levels of many transcripts do not follow a normal distribution.
The significance of a partial correlation with n data points was assessed with a two-tailed t test on where r is the first order correlation coefficient, and k is the number of variables on which we are conditioning.
Cultured hippocampal neurons from male B6 mice, prepared as described in Schikorski et al. [111] and cultured for 23 days, were fixed with 4% paraformaldehyde and 0.1% glutaraldehyde in HEPES buffered saline (pH 7.2) for 15 min. Cell membranes were permeabilized with 0.1% triton X-100 and unspecific binding sites were quenched with 10% BSA for 20 min at room temperature (RT). Neurons were incubated with a polyclonal anti-FMN2 antibody (Protein Tech Group, www.ptglab.com) diluted to 0.3 µg/ml at RT overnight. An anti-rabbit antibody raised in donkey (1∶500, Invitrogen; http://www.invitrogen.com) conjugated with the fluorescent dye Alexa488 was used for the detection of the first antibody. All regions of interest were photographed with identical illumination and camera settings to allow for a direct comparison of the staining in labeled and control neurons.
The Fmn2−/− mice were generated using 129/SvEv (now strain 129S6/SvEvTac) derived TC-1 embryonic stem cells. Chimeric mice were backcrossed to 129/SvEv [70]. The Fmn2-null and littermate controls are therefore coisogenic. To validate the isogenicity of regions surrounding the targeted locus [112], we genotyped the Fmn2+/+, Fmn2+/−, and Fmn2−/− mice using ten microsatellite markers located on, and flanking Fmn2 (markers distributed from 172 Mb to 182 Mb). These markers are D1Mit455, D1Mit113, D1Mit456, D1Mit356, D1Mit206, D1Mit355, D1Mit150, D1Mit403, D1Mit315, and D1Mit426. With the exception of a marker at Fmn2 (D1Mit150), all alleles in null, heterozygote, and wildtype animals were identical.
RNA was isolated from whole brain samples of Fmn2+/+ and Fmn2−/− mice, and assayed on Illumina Mouse-6 array slides (six samples per slide). We compared five samples from Fmn2−/− nulls, and five samples from Fmn2+/+ wildtype. Equal numbers of each genotypes were placed on each slide to avoid batch confounds. Microarray data were processed using both raw and rank invariant protocols provided by Illumina as part of the BeadStation software suite (www.illumina.com). We subsequently log-transformed expression values and stabilized the variance of each array. To identify genes with significant expression difference between the Fmn2−/− and Fmn2+/+ cases, we carried out two-tailed t-tests and applied a Bonferroni correction for multiple testing, and selected probes with a minimum adjusted p-value<0.05.
Classical QTLs counts are based on the April 2008 version of Mouse Genome Informatics (MGI: www.informatics.jax.org) [113]. Search for tRNAs was done using tRNAscan-SE 1.21 (http://lowelab.ucsc.edu/tRNAscan-SE/) [65]. GO analysis was done using the analytical tool DAVID 2007 (http://david.abcc.ncifcrf.gov/) [114]. Overrepresented GO terms were identified and statistical significance of enrichment was calculated using a modified Fisher's Exact Test or EASE score [115]. We used the Allen Brain Atlas to analyze expression pattern in the brain of young C57BL/6J male mice (www.brain-map.org) [85],[116].
In RI strains, non-syntenic associations can lead to LD between distant loci [89],[106]. In the BXDs, we detected such non-syntenic associations between markers in Qrr1 and markers on distal Chr 2 and proximal Chr 15. As a result of these associations, some transcripts that have strong cis- or trans-QTLs in Qrr1 tend to have weak LOD peaks, usually below the suggestive threshold, on distal Chr 2 and proximal Ch15. However, there is no bias for genes located in these intervals in LD with Qrr1 to have trans-QTLs in Qrr1.
The Qrr1 segment has been reported to have paralogues on mouse Chrs 1 (proximal region), 2, 3, 6, 7, 9, and 17 [117],[118]. We examined if the trans-QTLs in Qrr1 are of genes located in these paralogous regions. However, genes located in the paralogous regions are not overrepresented among the trans-QTL.
|
10.1371/journal.pgen.1002637 | Heritability and Genetic Correlations Explained by Common SNPs for Metabolic Syndrome Traits | We used a bivariate (multivariate) linear mixed-effects model to estimate the narrow-sense heritability (h2) and heritability explained by the common SNPs (hg2) for several metabolic syndrome (MetS) traits and the genetic correlation between pairs of traits for the Atherosclerosis Risk in Communities (ARIC) genome-wide association study (GWAS) population. MetS traits included body-mass index (BMI), waist-to-hip ratio (WHR), systolic blood pressure (SBP), fasting glucose (GLU), fasting insulin (INS), fasting trigylcerides (TG), and fasting high-density lipoprotein (HDL). We found the percentage of h2 accounted for by common SNPs to be 58% of h2 for height, 41% for BMI, 46% for WHR, 30% for GLU, 39% for INS, 34% for TG, 25% for HDL, and 80% for SBP. We confirmed prior reports for height and BMI using the ARIC population and independently in the Framingham Heart Study (FHS) population. We demonstrated that the multivariate model supported large genetic correlations between BMI and WHR and between TG and HDL. We also showed that the genetic correlations between the MetS traits are directly proportional to the phenotypic correlations.
| The narrow-sense heritability of a trait such as body-mass index is a measure of the variability of the trait between people that is accounted for by their additive genetic differences. Knowledge of these genetic differences provides insight into biological mechanisms and hence treatments for diseases. Genome-wide association studies (GWAS) survey a large set of genetic markers common to the population. They have identified several single markers that are associated with traits and diseases. However, these markers do not seem to account for all of the known narrow-sense heritability. Here we used a recently developed model to quantify the genetic information contained in GWAS for single traits and shared between traits. We specifically investigated metabolic syndrome traits that are associated with type 2 diabetes and heart disease, and we found that for the majority of these traits much of the previously unaccounted for heritability is contained within common markers surveyed in GWAS. We also computed the genetic correlation between traits, which is a measure of the genetic components shared by traits. We found that the genetic correlation between these traits could be predicted from their phenotypic correlation.
| Obesity associated traits such as central adiposity, dyslipidemia, hypertension, and insulin resistance are major risk factors for type 2 diabetes and cardiovascular complications [1]. The constellation of these traits has been termed metabolic syndrome (MetS). Understanding the genetic factors underlying these traits and how they are correlated is clinically important. Large-scale genotyping investigations such as genome-wide association studies (GWAS) are useful tools for identifying genetic factors. However, significant genetic variants discovered in GWAS explain only a small proportion of the expected narrow-sense heritability, h2, defined as the ratio of additive genetic variance to phenotypic variance [2]. This discrepancy underlies the debate concerning “missing” genetic factors among the common variants [3], [4].
The main approach of GWAS has been to identify significant single-nucleotide polymorphisms (SNPs) by examining each SNP individually for significance. The h2 attributed to that marker is then given by 2f(1−f)a2, where f is the frequency of the marker and a is the additive effect. To reduce the chance of false positives, a stringent p-value criterion has been adopted (typically p = 5*10−8, based on an adjusted p-value of 0.05 for one-million tests). It has been suggested that this selection criterion is too conservative [5] and that some of the missing heritability may be linked to genetic markers of small effect that fail this stringent cutoff.
Alternatively, the narrow sense heritability explained by the common SNPs, hg2, may be estimated by adapting a linear mixed-effects model [6], [7] that is used to estimate h2. This model decomposes the phenotypic variance into genetic and residual variance components. Usually, the model is applied to related individuals where the genetic relationships are estimated by using family pedigree or genetic markers [8], [9]. Yang et al. [6], [7] pointed out that hg2 could be estimated using genetic relationships obtained from the common SNPs for unrelated individuals. The main assumed difference between hg2 and h2 is due to the difference in linkage disequilibrium (LD) between the common SNP markers and the rest of the genome, with the assumption that closely related individuals would be in greater LD than unrelated individuals. Thus, heritability estimated with the genetic relationships of unrelated individuals is attributed to the common variants while that estimated with genetic relationships of related individuals is attributed to the entire genome. While the method does not identify single variants, it provides the maximum expected variance expected by the set of markers or the relative complement of the set (e.g., common versus rare variants). Recently, it has been shown that a large proportion of h2 is explained by the common single-nucleotide polymorphisms (SNPs) for several traits using this model [6], [7]. Here, we showed that large proportions of the phenotypic variance for several metabolic syndrome (MetS) traits were also captured by the common SNPs. Among these, we validated the height and body-mass index estimates by Yang et al. [6], [7] in independent GWAS populations. We also quantified the genetic correlation between traits explained by the common SNPs.
We estimated h2 and hg2 for height and body-mass index (BMI) in the Framingham Heart Study population (FHS), and height and seven metabolic syndrome traits (MetS) traits: BMI, waist-to-hip ratio (WHR), systolic blood pressure (SBP), fasting glucose (GLU), fasting insulin (INS), fasting triglycerides (TG), and fasting high-density lipoprotein (HDL) in the Atherosclerosis Risk in Communities population (ARIC) (ARIC MetS estimates shown in Table 1). Our base FHS population consisted of 4,240 subjects and our base ARIC population consisted of 8,451 subjects (see Methods and Tables S1 and S2 for a description of the populations). The genetic relationship between pairs of subjects was estimated using 436,126 genome-wide common SNP markers for ARIC and 320,118 SNPs for FHS (see Methods for details).
We first estimated h2 for related individuals with relationships between 0.35 and 0.65, derived empirically from the SNP markers, for height and BMI in the ARIC and FHS populations (see Methods for derivation of the relationship matrix). This resulted in 3,663 subjects (6,706,953 pairs of subjects) for FHS and 530 subjects (140,185 pairs of subjects) for ARIC. We found h2 to be 0.77 (s.e. 0.03) for height and 0.39 (s.e. 0.04) for BMI in FHS, and 0.88 (s.e. 0.09) for height and 0.34 (s.e. 0.12) for BMI in ARIC. The estimated h2 were consistent with values obtained using phenotypic regression (data not shown) and previous results [6], [7], [10], [11].
We then compared these values to estimates for hg2 for unrelated individuals with relationships less than 0.025 (see Methods for derivation of the relationship matrix). This resulted in 1,489 subjects (1,107,816 pairs of subjects) for FHS and 5,647 subjects (31,882,962 pairs of subjects) for ARIC. As mentioned above, hg2 provides an estimate of the heritability explained by common variants because of presumed lesser linkage disequilibrium between the common SNPs and the rest of the genome as compared to related individuals. We found hg2 to be 0.50 (s.e. 0.18) for height and 0.10 (s.e. 0.18) for BMI in FHS, and 0.46 (s.e. 0.05) for height and 0.14 (s.e. 0.05) for BMI in ARIC. These values are consistent with previously estimated values [6], [7]. Using the average across FHS and ARIC estimates, this implied that the common SNPs accounted for approximately 58% of h2 for height and 33% for BMI. To assess whether including more common SNPs would explain more of the h2, we examined how hg2 depended on the number of SNPs. As shown in Figure S1, the mean and standard error of the hg2 estimate for height in the ARIC population appeared to stabilize after approximately 300,000 SNPs.
We then estimated h2 and hg2 for the MetS traits in the ARIC population using the same subjects as above (see Table 1). We validated our h2 estimates by using phenotypic regression between related individuals for some of the traits (data not shown). The median h2 was 0.33, the minimum was 0.23 (INS), and the maximum was 0.48 (HDL). The median hg2 was 0.13, the minimum was 0.09 (INS), and maximum was 0.24 (SBP). Comparing the medians suggested that hg2 explains ∼39% of the h2 for these MetS traits. We found that the common SNPs explained large proportions of the h2: 41% of h2 for BMI, 46% for WHR, 30% for GLU, 39% for INS, 34% for TG, 25% for HDL, and 80% for SBP.
We next estimated the genetic correlations between MetS traits using a bivariate (multivariate) model (see Tables S3 and S4 for covariances). Table 2 shows the genetic and residual correlations for related individuals using bivariate models. The genetic correlation is the additive genetic covariance between traits normalized by the geometric mean of the individual trait genetic variances. The residual correlation is similarly estimated using the residual covariance and variances. For related individuals, we found significant genetic correlations for BMI-WHR, WHR-INS, GLU-INS, INS-TG, and TG-HDL and significant residual correlations between BMI-WHR, BMI-INS, BMI-HDL, WHR-INS, INS-HDL, and TG-HDL. Table 3 shows the genetic and residual correlations for the unrelated individuals. We found significant genetic correlations for BMI-WHR and TG-HDL and significant residual correlations for all of the estimates except SBP-HDL. The genetic correlations for unrelated individuals were proportional to the genetic correlations for related individuals (see Figure S2) with a proportionality constant of 0.44 (s.e. = 0.15 ; two-tail t-distribution p-value with 20 d.f. = 8.2*10−3). The phenotypic correlations between traits were similar for related and unrelated individuals and are shown in Table 4. These values were also consistent with the reported estimates in the National Heart Lung and Blood Institute-Family Heart Study (NHLBI-FHS), which included Framingham Heart Study and ARIC families [11].
We validated our genetic correlation estimates using bivariate models for each pair of traits by analyzing all 7 MetS traits simultaneously for the unrelated individuals in a single multivariate model. This 7 trait multivariate model was much more expensive computationally so we used a less stringent convergence rule. The results were similar to the bivariate model (see Table S5 and S6) although the genetic correlation increased and their error decreased for a number of the estimates. In addition to the significant genetic correlations in the bivariate models, we also found the genetic correlation for BMI-INS to be significant in the 7 trait model.
We then examined the relationship between the genetic and phenotypic correlations (see Figure S3). For related individuals, we found that the phenotypic correlations rp were proportional to the genetic correlations rg with a proportionality constant of 1.2 (s.e. = 0.16; two-tail t-distribution p-value with 20 d.f. = 3.1*10−7). For unrelated individuals, we found that the phenotypic correlations were proportional to the genetic correlations with a proportionality constant of 0.85 (s.e. = 0.19 ; two-tail t-distribution p-value with 20 d.f. = 2.3*10−4). The direct proportionality between rp and rg implies that the ratio rg/rp is approximately constant for the MetS traits.
We used a recently developed approach to analyzing GWAS data and provided new estimates for the total amount of additive genetic information contained in the common SNPs for MetS traits. The approach uses a linear mixed-effects model to estimate the additive genetic variances and correlations between traits. The model relies on knowing the genetic relationships between the individuals analyzed. Previously, this had been obtained from family pedigrees. Visscher et al. [9] and Yang et al. [6] observed that the genetic relationships could be computed from the GWAS SNPs. They also presumed that the heritability estimated for unrelated individuals with low SNP correlation are explained mainly by these common SNPs because the linkage disequilibrium between the common SNPs and the rest of the genome is weak. This would be in contrast to related individuals with high SNP correlation where linkage disequilibrium is strong. Thus, heritability estimated with the genetic relationships of unrelated individuals is attributed to the common SNPs while that estimated with the related individuals is attributed to the entire genome. This then creates a major distinction between h2 and hg2. We computed both in the same population. However, differences between estimates of h2 and hg2 may also arise due to differences in environmental influences and non-additive genetic effects that may bias the estimates. Provided that these biases are small then the ratio of hg2 to h2 provides an estimate of the proportion of narrow sense heritability captured by the common SNPs.
We confirmed previous findings that a large proportion of h2 is explained by the common SNPs. Our hg2 estimates for height and BMI in two independent analyses (i.e. ARIC and FHS) were consistent with previously reported values [6], [7]. Our h2 estimates for BMI, GLU, INS, TG, HDL, and SBP were similar to the findings of the large family National Heart, Lung, and Blood Institute (NHLBI) Family Heart Study [11], which included Framingham Heart Study and ARIC families. We found that hg2 explained a large proportion of h2 across the MetS traits, and hg2 explained approximately 39% of the h2 for these traits. We estimated that the common SNPs explain 58% of h2 for height, 41% for BMI, 46% for WHR, 30% for GLU, 39% for INS, 34% for TG, 25% for HDL, and 80% for SBP. Our hg2 findings are striking compared to traditional GWAS approaches where significant common SNPs have been shown to explain only 4% of h2 for BMI with 32 SNPs, 11% for GLU with 14 SNPs, 20% for TG with 48 SNPs, 25% for HDL with 60 SNPs, 3% for SBP with 10 SNPs, and 12% for height with 180 SNPs [12]–[16]. Height had the largest absolute hg2, which was consistent with having a large h2. Surprisingly, SBP had the largest proportion of h2 explained by the common SNPs while only a few percent of this has been uncovered by traditional GWAS. However, the standard error of hg2 for SBP was large and reducing this error will be important for further investigation. Conversely, our analysis suggested that the SNP markers already identified for TG and HDL may contain the maximum heritability expected from the common SNPs.
Our analysis of hg2 against the number of SNPs suggested that the mean and standard error of hg2 for height is well estimated by approximately 300,000 markers and that including more markers would have little effect for this trait and perhaps others. The standard error of hg2 also increased with SNP number. This may seem paradoxical but can be explained by recalling that the estimate for hg2 is proportional to the regression coefficient of the square of the phenotype differences versus the genetic relationship (i.e. Haseman-Elston regression) [8]. The standard error of hg2 is thus inversely proportional to the variance of the genetic relationship. Since the latter is estimated from the common SNPs, this variance is expected to decrease as the number of SNPs increases thereby increasing the standard error [6].
Using the bivariate (multivariate) model [17], [18] we estimated the genetic and residual correlations between the MetS traits. Among these, we found that the genetic correlations in related and unrelated individuals for BMI and WHR were significantly different from zero. This is consistent with both traits as indirect measures of body fat and common health risks [19]. Previously, Rice et al., 1994 [20] found significant genetic correlations between BMI and SBP among normotensive nonobese families. This suggested a common genetic etiology to their physiological relationship through hyperinsulinemia resulting in increased renal reabsorption of sodium and sympathetic activation [20]. We found a large genetic correlation among related subjects, although it was not significant because of the large error. This was consistent with the large family study by the NHLBI that did not find a significant genetic correlation [8]. Perusse et al, 1997 [21] argued that cross-trait resemblance between BMI and lipids is mostly environmental. In concordance, we did not find significant genetic correlations between either BMI or WHR and TG and HDL for either related or unrelated individuals (see Table 3 and Table 4) while residual (which includes environmental) correlations were significant for BMI–HDL. We found that the residual covariance accounted for a minimum of 71% (derived from the estimates in Table 4 and Table S3) of the phenotype covariance between BMI or WHR and the lipid measurements for related individuals. Genetic correlations between TG and HDL were also large, which is consistent with their direct physiological relationship [22]. This is also consistent with the findings from a recent GWAS meta-analysis whose results showed that 50% of the significant markers for TG were also significant for HDL (derived from Supplementary Tables 6 and 11 in [16]), and with a genome-wide LOD correlation analysis [23]. While we found some significant genetic correlations among both related and unrelated subjects, the variance was large for these estimates and greater statistical power is needed for better accuracy.
We found that the genetic correlation was directly proportional to the phenotypic correlation, which was an unexpected, empirical finding. Previously, a linear relationship between the correlations was hypothesized by Cheverud for sets of traits with common functions, and shown empirically for a number of traits [8], [24]–[26]. While this finding is interesting from an evolutionary genetics perspective, it may also serve a useful purpose in the maximum likelihood computation of the linear mixed-effects model by providing initial genetic correlation (i.e. covariance) estimates based on the phenotypic correlations.
In summary, we provided evidence that the common SNPs explain large proportions of the variance for several MetS traits in agreement with previous findings for some of these traits [6], [7]. This is consistent with the original premise of GWAS that a large proportion of phenotypic variation for common traits may be due to common variants [27]. However, an amendment to this premise is that it is likely to be many common variants with small effect. This is supported by recent meta-analyses with larger sample sizes that have identified more associated common SNPs. This approach can serve as a first approximation of the total heritability expected from common SNPs given a genome-wide set of markers and requires fewer subjects to achieve significant results. We also found genetic associations that will be useful for single gene and systems biology studies. Future studies with greater power will provide estimates for weaker multivariate genetic associations and provide greater precision for the estimates presented here.
Our main study population was the Atherosclerosis Risk In Communities (ARIC) population. The ARIC population consists of a large sample of unrelated individuals and some families across North America. The population was recruited from four centers across the United States: Forsyth County, North Carolina; Jackson, Mississippi; Minneapolis, Minnesota; and Washington County, Maryland. For this study, we restricted our analysis to the European-American group. The population was recruited in 1987 from the general population consisting of subjects aged 45 to 64 years. The ARIC population consisted of 8,451 subjects.
Quality control and genotype calls for common SNPs were evaluated previously for ARIC using the Affymetrix Human SNP Array 6.0. We selected bilallelic autosomal markers based on the following criteria: missingness <0.05, Hardy-Weinberg equilibrium (p<10−6) and minor allele frequency >0.05. Subjects with missingness >0.05 were removed. This resulted in 436,126 retained markers.
Quality control measurements from dbGAP (GENEVA ARIC Project Quality Control Report Sept 22, 2009) indicate significant population stratification between self-identified white (European-ancestory kind group) and black populations when projected onto HapMap components. Furthermore, principal-components analysis of the European-ancestory group by dbGAP showed that no component explained more than 0.1% of the population variance. For this study we only analyzed the European-ancestory group and treated it as a single population.
ARIC phenotypes were adjusted for age, sex, and study center. Only single measurements from visit 1 were used for these subjects. We only used subjects with negative diabetes status and with genotype and phenotype information for all traits. This resulted in 8,451 subjects. We standardized all the traits. We first log-transformed BMI, glucose, insulin, triglycerides, HDL, and systolic blood pressure. All laboratory measurements are under fasting conditions. Population trait statistics are in Table S1.
We estimated h2 and hg2 for height and BMI in the Framingham Heart Study population (FHS). The FHS population is a large multi-generational dataset that started in 1948 in Framingham, Massachusetts in the United States. It consists of a number of ethnicities predominantly from the United Kingdom, Ireland, Italy, and Western Europe [28]. Markers were screened similarly to ARIC and we also removed any SNPs that did not overlap with the ARIC set, which results in 320,118 SNPs. We used principal components analysis of the linkage disequilibrium (LD) pruned genetic relationship matrix to identify components with variance >0.1%. LD pruning was as in the ARIC 2009 report. This resulted in 73,432 retained SNPs. We found three significant components that were then used as covariates in the REML model. For consistency with ARIC, we restricted the age range at time of exam to 45 to 65 years and randomly selected a single measurement in the case of multiple measurements. Phenotypes were adjusted for age, sex, and generation prior to the REML estimation and standardized. We first log-transformed BMI. Population trait statistics are in Table S2. Our base FHS population consisted of 4,240 subjects.
We determined h2 using the linear mixed-effects model (see derivation below) and related individuals defined as genomic relatedness between 0.35 and 0.65. We assume that the common SNPs are in greater linkage disequilbrium among related individuals and, as such, can be used to estimate the total additive-genetic variance across the allele spectrum as suggested by Visscher et al., 2006 [9]. We constrained the relationship matrix to have at least one related pair per subject. This was done by pruning the entire population relationship matrix by randomly selecting a row and removing the row and its corresponding column if no genomic covariance in the row was between the cutoff values. For all pairs, including unrelated individuals, we used their empirically defined relationship. This resulted in 530 individuals being selected for analysis in ARIC and 3,663 individuals in FHS.
h2 was estimated with h2 = varg/(varg+vare), where varg and vare are the genetic and residual variance components estimated by the REML model using related individuals. The error was estimated from the inverse Fisher Information (see linear mixed-effects model below) and propagated using a first-order Taylor expansion.
We used the linear mixed-effects model and only unrelated individuals to estimate the additive-genetic variance attributable to the common SNPs (hg2). Unrelated individuals were defined as subjects with maximum genomic correlation of <0.025. The genomic relationship matrix was then produced as above based on this cutoff. The cutoff was taken from Yang et al. 2010 [6] and is less than the expected coefficient of relatedness between 2nd cousins. For these estimates we used the same group of 5,647 unrelated individuals for all estimates in ARIC and 1,489 individuals in FHS. hg2 was estimated as hg2 = varg/(varg+vare), where varg and vare are the genetic and residual variance components estimated by the REML model using unrelated individuals. The standard error was estimated as above. The height hg2 versus SNP number analyses were performed over allele frequency range of 0.05 to 0.5 in order of increasing and decreasing frequency.
The genetic correlation (rg) is defined as , where (varg(ti)) is the additive genetic variance of trait i and covariance (covg(ti,tj)) is the additive genetic covariance between the traits. The variances and covariances are estimated directly in the multivariate linear mixed-effects model. The error was computed from the estimated errors of the variances and covariance using a first-order Taylor expansion. The residual and phenotypic correlations were analogously defined. Phenotype correlations and error were estimated by linear regression of the standardized phenotypes.
The mean and errors for proportionality constants between the genetic and phenotypic correlations were determined by randomly sampling over the distributions of the parameter estimates (i.e. Monte Carlo method) assuming that the error around the mean parameter estimate was normally distributed and that the parameters were independent. We then fit a linear function with the y-intercept fixed at 0 (after first confirming that it was not significantly different from zero).
We assessed significance for correlation coefficients (r) using the standardized Fisher transformed estimate of r: arctan(r)/arctan(s.e.(r)). We estimated the two-tailed p-value from a normal distribution and significance was determined by p<0.05 and Bonferroni corrected for 21 hypotheses.
Significance for regression coefficient () was estimated using the standardized coefficient . We estimated the two-tailed p-value from a t-distribution and 20 degrees of freedom and significance was determined by p<0.05.
Preprocessing of SNPs and phenotypes was done using PLINK [29] (v1.07,http://pngu.mgh.harvard.edu/purcell/plink/) and MATLAB (2010b, MathWorks, Natick, MA). REML optimization was executed using software written in MATLAB.
We considered the following multivariate linear mixed-effects model for m individuals, n loci and t traits [6]–[8], [17], [18], [30]:where yi is a m×1 vector of trait i for m individuals, Xi is an m×s fixed effects matrix for trait i, vi is a s×1 vector of fixed effects parameters for trait i, Z is an m×n matrix of standardized genotypes, ui is an n×1 vector of random effects for trait i satisfying ui∼N(0,G) and ei is an m×1 vector of residual effects satisfying ei∼N(0,R), with matrix blocks Gij = covgijIn and Rij = coveijIm and Il is the l×l identity matrix. This model can be used for single or multiple traits. For two traits, it is called a bivariate model. The model is identical to that used by [6], [7], [17].
We considered only bi-allelic SNPs in Hardy-Weinberg equilibrium. Denote the minor allele by q and the major allele by Q. Let the minor allele frequency at locus i have frequency pi. We assign a value of 2 for genotype qq, 1 for genotype qQ and 0 for genotype QQ. The Hardy-Weinberg mean frequency for the genotype at locus i is 2pi and the variance is 2pi(1−pi). The standardized genotype entries have values of (2−2pi)/(2pi(1−2pi))1/2 for qq, (1−2pi)/(2pi(1−2pi))1/2 for qQ, and −2pi/(2pi(1−2pi))1/2 for the QQ genotype.
The log of the likelihood function is given bywhere the covariance matrix can be expressed as a tensor product with m×m blocks V−1ij and A is the genetic relationship matrix. Following Yang et al. [6], we used a modified covariance matrix for A, , where the diagonals of A are computed using the formulaWe use the restricted maximum likelihood (REML) approach [8] where the gradients of the log likelihood are given bywhere Iij is a tm×tm dimensional matrix with zero entries except for a m×m identity matrix at block location i, j, and , where .
We solved the REML equations using an EM algorithm [8], which was given byfor iteration k+1 in terms of iteration k. We iterated until the rate of change of the log likelihood function was less than about 10−4. We also checked that the rate of change of the square of the covariance predictions was less than 10−8. We checked our results against the software developed by Yang et al. (GCTA) [31] for the univariate model.
For the multivariate model, we transformed to a coordinate system where the covariance matrices were diagonal [8] to speed up the computation. Let zj be the set of phenotypes for individual j. We used the canonical transformation such that and . Q can be computed from the formula where , (S is the matrix of left eigenvectors of GR−1). The transformed genetic covariances are given by and the residual covariances are It. Each step consisted of taking a single step with the univariate EM algorithm for the transformed additive genetic and residual variance followed by a transformation back to the original coordinates. We iterated until the maximum of the magnitudes of the components of the gradient of the log likelihood function was less than approximately .
In our computations, we used both the direct EM algorithm and the canonically transformed algorithm because even though the transformed algorithm was in principle faster, it sometimes had poor convergence properties if the initial guess was not sufficiently close to the maximum likelihood value. We ensured that both give the same results. For computational efficiency, the results shown are computed from the bivariate model for the different trait pairs. We confirmed our results with a multivariate model that included all traits.
Our error estimates were given by the inverse of the Fisher information matrix F, which we computed by evaluating the Hessian of the log likelihood at the maximum likelihood predictions. F is a t(t+1)×t(t+1) dimensional matrix with rows corresponding to the genetic and residual variances and covariances (where covij was set equal to covji) and with block elements (that are not all contiguous) given byfor and .
|
10.1371/journal.ppat.1002667 | The Lipopolysaccharide from Capnocytophaga canimorsus Reveals an Unexpected Role of the Core-Oligosaccharide in MD-2 Binding | Capnocytophaga canimorsus is a usual member of dog's mouths flora that causes rare but dramatic human infections after dog bites. We determined the structure of C. canimorsus lipid A. The main features are that it is penta-acylated and composed of a “hybrid backbone” lacking the 4′ phosphate and having a 1 phosphoethanolamine (P-Etn) at 2-amino-2-deoxy-d-glucose (GlcN). C. canimorsus LPS was 100 fold less endotoxic than Escherichia coli LPS. Surprisingly, C. canimorsus lipid A was 20,000 fold less endotoxic than the C. canimorsus lipid A-core. This represents the first example in which the core-oligosaccharide dramatically increases endotoxicity of a low endotoxic lipid A. The binding to human myeloid differentiation factor 2 (MD-2) was dramatically increased upon presence of the LPS core on the lipid A, explaining the difference in endotoxicity. Interaction of MD-2, cluster of differentiation antigen 14 (CD14) or LPS-binding protein (LBP) with the negative charge in the 3-deoxy-d-manno-oct-2-ulosonic acid (Kdo) of the core might be needed to form the MD-2 – lipid A complex in case the 4′ phosphate is not present.
| Capnocytophaga canimorsus, a commensal bacterium in dog's mouths, causes rare but dramatic infections in humans that have been bitten by dogs. The disease often begins with mild symptoms but progresses to severe septicemia. The lipopolysaccharide (LPS), composed of lipid A, core and O-antigen, is one of the most pro-inflammatory bacterial compounds. The activity of the LPS has so far been attributed to the lipid A moiety. We present here the structure of C. canimorsus lipid A, which shows several features typical for low-inflammatory lipid A. Surprisingly, this lipid A, when attached to the core-oligosaccharide was far more pro-inflammatory than lipid A alone, indicating that in this case the core-oligosaccharide is able to contribute significantly to endotoxicity. Our further work suggests that a negative charge in the LPS-core can compensate the lack of such a charge in the lipid A and that this charge is needed not for stabilization of the final complex with its receptor but in the process of forming it. Overall the properties of the lipid A-core may explain how this bacterium first escapes the innate immune system, but nevertheless can cause a shock at the septic stage.
| Capnocytophaga canimorsus, a usual member of dog's mouths flora [1] was discovered in 1976 [2] in patients who underwent dramatic infections after having been bitten, scratched or simply licked by a dog. The most common syndrome is sepsis, sometimes accompanied by peripheral intravascular coagulation and septic shock [3]. C. canimorsus is a Gram-negative rod belonging to the family of Flavobacteriaceae in the phylum Bacteroidetes [4], [5]. Human infections occur, worldwide, with an approximate frequency of one per million inhabitants per year [6].
C. canimorsus are able to escape complement killing and phagocytosis by human polymorphonuclear leukocytes and macrophages [7], [8]. Whole bacteria are also poor agonists of Toll-like receptor (TLR) 4, which results in a lack of release of pro-inflammatory cytokines by macrophages [9]. In addition to these “passive” features, C. canimorsus have been shown to harvest glycan moieties from glycoproteins at the surface of animal cells, including phagocytes [10], [11], [12], in addition they also deglycosylate human IgG [12].
One of the most pro-inflammatory bacterial compounds is the lipopolysaccharide (LPS, endotoxin) [13], consisting of three domains: lipid A, the core-oligosaccharide and the O-polysaccharide (O-antigen). As a potent activator of the innate immune system, LPS can induce endotoxic shock in patients suffering from septicemia. Recognition of LPS by the host occurs via the TLR4/MD-2/CD14 receptor complex [14], [15], [16], at which two proteins, CD14 and LBP, have been shown to enhance the response to LPS by transporting single LPS molecules [17], [18], [19], [20]. It has been shown that the lipid A moiety of the LPS is sufficient for TLR4 binding and stimulation [21], [22]. The interaction of lipid A and its receptor was unraveled by x-ray crystallography pioneering studies of complexes between MD-2 and the lipid A analog Eritoran [23] or lipid IVA [24]. The identification of the binding sites of lipid A to MD-2 and also to the Leucine-rich repeat (LRR)-domains of TLR4 [21] is a landmark achievement that enables a deeper understanding of the structure-function relationship between LPS/lipid A and its receptors. According to these data, the 1 and 4′ phosphates of the lipid A backbone, which form charge interactions with TLR4 and MD-2, are the key elements for receptor activation [21], [25], even though for some of the interactions conflicting data have been reported [26]. It was further shown that the β-hydroxymyristate chain at position 2 forms hydrogen bonds and hydrophobic interactions with TLR4. At present, there is no evidence that the LPS-core plays any major role in binding to TLR4; only a 10- to 100-fold difference in endotoxicity of lipid A and LPS has been reported for E. coli, Porphyromonas gingivalis or Proteus mirabilis [27], [28], but these small differences have been attributed to changes in solubility, even if solid experimental proof is lacking. The core-oligosaccharide has so far never been shown to alter TLR4/MD-2 binding of a specific lipid A, only slight changes in MD-2 binding have been reported [29].
In this work, we investigated the lipid A structure of C. canimorsus in order to clarify its contribution to the septicemia and shock provoked by these bacteria. Very few lipid A structures have actually been solved in the Cytophaga/Flavobacterium group, with the exception of the lipid A from Elizabethkingia meningoseptica (former Flavobacterium meningosepticum) [30]. Already some time ago, the acyl chains present in the LPS of Cytophaga bacteria have been identified as [13-Me-14:0 (i15:0), 13-Me-14:0(3-OH)(i15:0(3-OH), 16:0(3-OH) and 15-Me-16:0(3-OH) (i17:0(3-OH)] [31], whereat i15:0 is iso-pentadecanoic acid (13-methyltetradecanoic acid, 13Me-14:0), i15:0(3-OH) represents iso-(R)-3-hydroxypentadecanoic acid [(R)-3-hydroxy-13-methyltetradecanoic acid, 13Me-14:0(3-OH)]; 16:0(3-OH) is (R)-3-hydroxyhexadecanoic acid and i17:0(3-OH) represents iso-(R)-3-hydroxyheptanoic acid [(R)-3-hydroxy-15-methylhexanoic acid, 15-Me-16:0(3-OH)]. Here we show that lipid A of C. canimorsus consists of the penta-acylated hybrid backbone 2,3-diamino-2,3-dideoxy-d-glucose (β-d-GlcN3N′) linked (1′→6) to α-d-GlcN where the 4′ phosphate group is missing and the 1 phosphate is linked to an ethanolamine group, forming a P-Etn. Not unexpectedly, this lipid A was of very low endotoxicity but, surprisingly, when bound to the core [lipid A-core (LA-core)] it became 20,000 fold more endotoxic. In agreement with this observation, we show that the LPS core promotes the binding of C. canimorsus lipid A to MD-2. This is the first example of a core-oligosaccharide dramatically changing the endotoxicity of lipid A, in which the carboxy group of Kdo probably takes over the function of ionic binding of the missing 4′ phosphate in the lipid A.
GlcN and GlcN3N were found in a ratio of approx. 2∶1 (Table 1). Based on the notion that by gas-liquid chromatography (GLC) analysis synthetic GlcN3N expressed a response factor of about 50% when compared with GlcN (or Galactosamine (GalN) as internal standard), it was inferred that GlcN and GlcN3N are present in equimolar amounts in the lipid A backbone, suggesting the presence of a “hybrid backbone” in C. canimorsus lipid A (Table 1). Total fatty acid analysis revealed i15:0, i15:0(3-OH), 16:0(3-OH), and i17:0(3-OH) in a molar ratio of approximately 1∶1∶1∶2 in lipid A preparations. Analysis of ester-bound acyl chains indicated the presence of i15:0 and i15:0(3-OH) in approximately equimolar amounts, indicating that one 16:0(3-OH) and two i17:0(3-OH) residues are primary acyl chains N-linked to the lipid A backbone (Table 1). This result suggests a penta-acylated lipid A species.
The reversed phase HPLC profile of the lipid A sample is shown in Fig. S1. Peak 2 expressed a molecular ion at m/z 1716.30, which is in excellent agreement with a lipid A containing i15:0, i15:0(3-OH), 16:0(3-OH), and two moles of i17:0(3-OH) attached to the lipid A backbone (GlcN3N-GlcN), which also carries one P-Etn residue. The second major fraction (peak 5) at m/z 1594.29 was compatible with lipid A lacking the P-Etn. Based on peak intensities (peaks 2 and 5) about 40% of the P-Etn was liberated, most likely from the lipid A under the hydrolysis conditions used (Fig. S1).
All lipid A fractions investigated expressed a certain heterogeneity with respect the chain length of acyl chains (-CH2- groups), as all MS showed peak “clusters” differing by 14 u, thus suggesting acyl chain heterogeneity (Table 2, Fig. S2). Combined GLC/mass spectrometry (GLC-MS) analysis of the acyl chains revealed that the mass difference of Δm/z = 14 u was not due to the exchange of one single, prominent shorter acyl chain [e.g. 16:0(3-OH)→i15:0(3-OH)]. Instead, the lipid A showed a certain structural “fuzziness” with respect to the size and position of the individual acyl chains, which, according to this finding, appeared to be statistically distributed over all positions with no specific structural variation.
The ESI-MS data of the wt strain shown in Table 2 indicated identical mass at m/z 1716.30 for peaks 2 and 3. As these lipid A fractions differed in their retention time, we conclude that they represent different structural isomers as they could be baseline-separated by HPLC. This HPLC analysis in combination with ESI-MS data thus shows that structural heterogeneity might not be solely related to the chain length of one acyl chain, but also to its position within the lipid A backbone.
In order to allocate the type of the hybrid lipid A backbone, the acyl chain distribution over the lipid A backbone, and the attachment side of the P-Etn, electrospray-ionization Fourier transform ion-cyclotron resonance (ESI FT-ICR) MS/MS in the positive mode was run [32]. The triethylammonium salt of HPLC purified lipid A at m/z 1820.40 was selected as precursor ion (Fig. S3). Infrared multiphoton dissociation (IRMPD)-MS/MS generated one abundant characteristic B-fragment oxonium-ion of the non-reducing end at m/z 907.77, which is in excellent agreement with the mass value calculated for GlcN3N with i15:0, 16:0(3-OH), and i17:0(3-OH) attached (m/z 907.77). This fragmentation pattern also showed that P-Etn is attached at the reducing end - most likely at position C-1. Thus the lipid A in C. canimorsus is penta-acylated with an acylation pattern of three being attached to the “non-reducing” GlcN3N′ and two to the reducing GlcN sugar (3+2) in the lipid A hybrid backbone.
The lipid A was studied further by high-field NMR spectroscopy using correlation spectroscopy (COSY), total correlation spectroscopy (TOCSY), rotating-frame nuclear Overhauser effect spectroscopy (ROESY), 1H,13C-heteronuclear single-quantum coherence (HSQC), 1H,31P-heteronuclear multiple-quantum coherence (HMQC), and 1H,31P-HMQC-TOCSY experiments. The results are depicted in the supplement (Table S1). The 1H,13C-HSQC spectrum (Fig. 1) showed two H-1,C-1 cross-peaks at δ 4.28/103.4 and 5.29/92.8 for GlcN3N′ and GlcN, which were distinguished by correlations between protons at nitrogen-bearing carbons and the corresponding carbons (C-2′ and C-3′ of GlcN3N′ and C-2 of GlcN, at δ 52.9, 54.6, and 51.4, respectively). 3J1,2 coupling constants of 8.0 and 2.9 Hz for the H-1 signals at δ 4.28 and 5.29, were determined from the 1H NMR spectrum and showed that GlcN3N is β- and GlcN α-linked. The H-1 signal of α-GlcN was additionally split due to coupling to phosphorus (2J1,P 7.9 Hz), thus indicating that α-GlcN is phosphorylated with P-Etn and β-GlcN3N′ represents the “non-reducing” end of the lipid A backbone. The β1′→6-linkage between the two amino sugars was evident from strong cross-peaks of H-1′ of GlcN3N′ with protons H-6a′,6b′ of GlcN at δ 3.64 and 3.87 in the ROESY spectrum. The location of the P-Etn residue at position 1 of α-GlcN was further confirmed by 1H,31P-HMQC and 1H,31P-HMQC-TOCSY (Fig. S4) as well as ROESY experiments, which showed correlations between H-1 of GlcN at δ 5.29 and H-1a,1b of ethanolamine (Etn) at δ 3.91 and 3.98. In accordance with the 1′→6 linkage and the position of GlcN3N at the “non-reducing end”, the 13C NMR spectrum (Table S1) displayed a typical down-field displacement by ∼10 ppm for C-6 of the 6-substituted GlcN (δ 71.0; compared with δ 60.0 for C-6 of GlcN3N′, which is non-substituted in the free lipid A). The acylation pattern was confirmed by 1H,13C-HSQC spectroscopy (Fig. 1), which showed only one characteristic downfield shift due to a deshielding effect for the i17:0[3-O(i15:0)] R2′ i.e. the H-3/C-3 R2′ cross-peak at δ 4.95/70.7. This finding indicated that only the OH-group of i17-0(3-OH) is acylated giving rise to an acyloxyacyl residue [i17:0-3-O(i15:0)] showing a 3+2 type acyl chain distribution in the penta-acylated lipid A, which is in good agreement with the MS data (Figs. S2 and S3). Taking together the data of the chemical studies defines the structure of the lipid A of C. canimorsus shown in Fig. 2 A. The structure of E. coli hexa-acylated lipid A is depicted for comparison (Fig. 2 B). The E. coli lipid A consists of a β-(1′→6)-linked GlcN disaccharide that is phosphorylated at positions 1 and 4′ and carries four (R)-3-hydroxymyristate chains (at positions 2′, 3′, 2 and 3). The R2′ and R3′ 3-hydroxylated acyl groups in GlcN′ are further esterified with laurate and myristate, respectively [22].
The structure of C. canimorsus LA-core is depicted in Fig. 2 C and its structural analysis will be described elsewhere (Zähringer et al., manuscript in preparation). The C. canimorsus LPS core features only one Kdo, to which a P-Etn is attached in position 4. Usually, mono-Kdo LPS-core have a phosphate attached to the Kdo at that position. Thus, the only net negative charge in this core oligosaccharide originates from the carboxy-group of the Kdo. The inner core continues with two mannoses (Man) to which another P-Etn is attached in position 6 of ManI residue in the core oligosaccharide. The outer core consists of Galactose (Gal) and l-Rhamnose (to which the O-antigen is attached). A positively charged Galactosamine (GalN) residue is linked to the (second) ManII residue in position 6 (U. Zähringer, unpublished results).
E. coli lipid A biosynthesis has been unravelled in detail [22], [33]. Analyzing the genome of C. canimorsus 5 [5], we identified the genes required for the synthesis of lipid A-Kdo [33]. Only lpxA, lpxA′, lpxC and lpxD seem to cluster in one operon, the other genes are dispersed (Fig. 3 A). The difference in acylation of the 3′ and 3 position and the hybrid backbone of the lipid A consisting of a β-1′,6-linked GlcN3N′-GlcN disaccharide, suggests that two lpxA genes might be present in C. canimorsus and indeed two lpxA genes were identified (termed lpxA and lpxA′) in the C. canimorsus 5 genome (Fig. 3 A). In Acidithiobacillus ferrooxidans GnnA and GnnB are responsible for the biosynthesis of GlcN3N [34]. Based on the sequences of A. ferrooxidans, gnnA and gnnB could be identified in the genome of C. canimorsus (Fig. 3 A). In the biosynthetic pathway of E. coli lipid A, enzyme LpxM adds the acyloxyacyl-residue [14:0-3-O(14:0)] representing the 6th acyl chain [22]. In good agreement with the penta-acylation of lipid A in C. canimorsus 5 was our finding that lpxM could not be identified in the genome (Fig. 3 A). C. canimorsus LPS core features only one Kdo, suggesting a mono-functional Kdo transferase (WaaA/KdtA) or a Kdo hydrolase two-protein complex (KdoH1/2) as in Helicobacter pylori or Francisella novicida [35], [36]. Searches with KdoH1/2 did not hit any gene in the C. canimorsus 5 genome. Therefore, C. canimorsus possesses either a mono-functional WaaA or a KdoH1/2 complex without significant sequence similarity to known Kdo hydrolases. We have further investigated the enzymes leading to the addition of an Etn at the 1 phosphate of lipid A. In H. pylori, the addition of a P-Etn at 1 position has been proposed to result from a two-step mechanism [37]. In a first step the 1 phosphate is removed by a phosphatase (LpxE), and subsequently a P-Etn-transferase (EptA or PmrC, YjdB) adds a P-Etn to the 1 position of lipid A [37] (Fig. 3 B). In H. pylori lpxE and eptA are encoded by one operon (Hp0021-Hp0022). C. canimorsus eptA was annotated as Ccan 16950. Search for a lipid A phosphatase were based on lpxE and/or lpxF sequences from P. gingivalis [38], F. novicida [39], Rhizobium etli [40] H. pylori [37], [41] and on all available Bacteroidetes-group pgpB sequences. Three lpxE/F candidates have been found in the C. canimorsus 5 genome (Ccan16960, Ccan14540 and Ccan6070). All candidates were deleted and the mutated bacteria were tested for endotoxicity. Only deletion of Ccan16960 affected endotoxicity (data not shown). Interestingly, Ccan16960 is located within the same operon as eptA and the two genes overlap by 20 bp. Following the operon organisation of H. pylori, Ccan16960 has been annotated as lpxE. The predicted function of lpxE and eptA was validated by KO and analysis of the resulting phenotype (Ittig et al., manuscript in preparation).
The presence of the 4′ kinase LpxK and the absence of a 4′ phosphate leads to the assumption of the presence of a 4′ phosphatase, LpxF. Several candidate genes were identified (besides lpxE: Ccan 14540 and Ccan6070) and deleted but they had to be ruled out, as no deletion did affect the endotoxic activity (data not shown), thus, we lack annotation of lpxF. The proposed complete biosynthesis of C. canimorsus lipid A-Kdo is depicted in Fig. 3 C, starting from UDP-N-acetyl-d-glucosamine and ribulose-5 phosphate.
The endotoxic activity of wt C. canimorsus 5 LPS (S-form) was compared to the endotoxic activity of E. coli O111 LPS using three different approaches: (i) Purified LPS samples were assayed for TLR4 dependent NFκB activation with HEK293 cells overexpressing human TLR4/MD-2/CD14 and a secreted reporter protein (HEKBlue human TLR4 cell line), (ii) purified LPS samples were assayed for induction of TNFα release by human THP-1 macrophages, (iii) purified LPS samples were tested for stimulation of IL-6 release by canine DH82 macrophages. In the two assays involving human TLR4 (Fig. 4 A and Fig. 4 C) C. canimorsus LPS appeared to be about 100 fold less endotoxic than E. coli O111 LPS (both S-form LPS). In contrast to human macrophages, where C. canimorsus LPS was found 10–100 fold less endotoxic than E. coli O111 LPS (Fig. 4 B), for canine macrophages the difference in endotoxicity of the two LPS was around 1000 fold (Fig. 4 E). In addition, lipid IVA seems not to be an agonist of canine TLR4 as is the case for murine TLR4 [42].
Generally, the lipid A part of a LPS is considered as sufficient to trigger full TLR4 activation. Minor differences between lipid A and LPS or LA-core have so far been attributed to differential bioavailability/solubility in water even if solid experimental proof is lacking. We have, therefore, examined the endotoxic activity of C. canimorsus lipid A, LA-core and LPS using the HEKBlue hTLR4 cell line and the TNFα release by human THP-1 macrophages. LPS and LA-core exhibited an endotoxicity in the same range, whereas the LPS was less than 10-fold more endotoxic than the LA-core (Fig. 4 B and Fig. 4 D). In contrast, C. canimorsus lipid A appeared to be absolutely non-stimulatory up to 5 µg/ml (Fig. 4 B and Fig. 4 D), around 20,000-fold less active than the LA-core and 200,000-fold less active than LPS on a weight basis (ng/ml) indicating a even higher difference on a molar basis. As the C. canimorsus LPS and the LA-core showed similar endotoxicity, the increase in endotoxicity in comparison to the lipid A must have been raised by the contribution of the core oligosaccharide. Minor differences in endotoxicity between LPS and LA-core as the 10- to 100-fold difference observed between E. coli lipid A and E. coli O111 LPS (Fig. 4 B and Fig. 4 D) might be explained by differential bioavailability/solubility in water/buffer and by a direct contribution of the core-oligosaccharide in TLR4/MD-2 binding as suggested [21]. However, in the case of C. canimorsus LA-core the direct contribution of the core-oligosaccharide might be far more pronounced as in E. coli, since C. canimorsus has a lipid A lacking a net negative charge. A role of the core-oligosaccharide in providing solubility to lipid A was ruled out by the fact that no increase in endotoxicity was observed by the addition of triethylamine (TEN) or dimethyl sulfoxide (DMSO) to the C. canimorsus lipid A stock solution followed by sonication (see Fig. 4 F).
The increase in endotoxicity of the C. canimorsus LA-core in comparison to the lipid A must have been raised by the contribution of the core oligosaccharide (Fig. 4). The 4′ phosphate of E. coli lipid A is known to interact with Arg264 and Lys362 of TLR4 and Lys58 and Ser118 of MD-2 [21]. C. canimorsus lipid A lacks the 4′ phosphate and features only one net negative charge in the LPS core, namely the carboxylic oxygen of Kdo. Based on the known structure of E. coli LPS bound to TLR4/MD-2 (3FXI, [21]) we measured the interaction distances from the carboxylic oxygen of Kdo to Arg264 and Lys362 of TLR4 and to Lys58 and Ser118 of MD-2. The carboxylic oxygen of Kdo is within close distance to Arg264 and Lys362 of TLR4 and Lys58 and Ser118 of MD-2 and hence could contribute to binding to MD-2 or TLR4.
To assess the ability of C. canimorsus lipid A or LA-core to interact with human MD-2, we monitored their ability to compete with the binding of E. coli LPS-Biotin to MD-2. Culture supernatants of cells producing human MD-2 were incubated with biotinylated E. coli O111 LPS, either alone or in combination with different concentrations of a competitor. As a source of LBP and soluble CD14, 7.5% FCS (v/v) was added. After purification of LPS based on biotin, co-purification of MD-2 was monitored by Western blotting. C. canimorsus LA-core abolished the copurification of MD-2 with the E. coli LPS-Biotin at higher concentration than the positive controls, E. coli O111 LPS and lipid IVA but at lower concentration than unbiotinylated E. coli penta-acyl lipid A (Fig. 5 A and B). Lipid IVA is expected to be a very potent competitor, as it has been shown to bind deeper into the MD-2 pocket and thus likely stronger to MD-2 than E. coli lipid A [21], [24]. These results indicate that C. canimorsus LA-core binds to human MD-2, likely in the same pocket as E. coli LPS. This experiment does not reflect the antagonistic capacity of C. canimorsus LA-core as even native E. coli O111 LPS could prevent the co-purification of human MD-2 (Fig. 5 A and B). In contrast to the LA-core, C. canimorsus lipid A did not significantly affect the copurification of MD-2 with E. coli LPS-Biotin even at high concentration (Fig. 5 A and B). Thus, C. canimorsus lipid A seems not to bind to human MD-2 at all or to bind to MD-2 only very weakly, in contrast to the LA-core. To rule out a major contribution of the core-oligosaccharide by providing solubility to the lipid A, the same MD-2 binding experiment has been performed with C. canimorsus lipid A pre-treated with DMSO or TEN and sonicated to improve solubility. These C. canimorsus lipid A samples did not significantly affect the copurification of MD-2 with E. coli LPS-Biotin even at high concentration (Fig. 5 C). We conclude from this experiment that the C. canimorsus LPS core promotes the interaction and binding of the lipid A to MD-2 either via direct interaction with MD-2 or via binding to LBP or CD14.
In order to assess the contribution of the C. canimorsus LPS core in binding of the lipid A to MD-2, we modelled the binding of C. canimorsus lipid A to human MD-2 (Fig. 6 A) and compared it to the binding of E. coli lipid A. Some differences between the two complexes could be observed at the level of the lipid chains after just few ns of simulation (Fig. 6 A). In both cases the R3′ and R3 chains (see Fig. 2 for nomenclature) were fully stretched and interacted with the same residues. No empty space was left by R3″ (missing in C. canimorsus) because the longer R2′ and R2″ chains filled the void. While in E. coli the R2 chain is stretched toward the inner side of the pocket, in C. canimorsus it was projected toward the pocket exterior, due to both i) its longer size and ii) to the presence of the bifurcated terminus of the close R2″. The R2 chain of C. canimorsus lipid A was thus not completely buried inside the MD-2 pocket and it was even more exposed to the surface than the hydroxymyristate chain at position 2 in E. coli. This probably enables the i17:0(3-OH) chain at position 2 to interact with TLR4, as has been reported for the R2 chain of hexa-acylated E. coli LPS [21]. It should be mentioned here that penta-acylated E. coli lipid A is endotoxically almost inactive [13], and the acyl chains might be completely buried inside MD-2. Thus C. canimorsus penta-acylated lipid A is expected to behave differently from penta-acylated E. coli lipid A due to the extended length of the acyl chains and the bulky iso-groups. Overall the arrangement of the sugar moieties with respect to the MD-2 was similar for both complexes, the only major discrepancies being the orientation of the 1-phosphoryl group (1 phosphate in E. coli, 1 P-Etn in C. canimorsus). The calculated binding energy for the two complexes was very similar when calculated at both MM-GBSA (molecular mechanics, the generalized Born model and solvent accessibility) and MM-PBSA (molecular mechanics, Poisson-Boltzmann solvent accessible surface area) level, being in both cases the MD-2 – E. coli lipid A complex slightly more stable (Fig. 6 C). To understand this trend the total binding free energy was fractionated into a list of interaction energies between each residue of MD-2 and each fragment of lipid A (Fig. 6 B), as coded in Fig. 2. Each pairwise binding free energy value has been further fractioned into its electrostatic, steric (Van der Waals), and solvation (polar and cavitation) components. For each term contributions arising from backbone and sidechain have been singled out. In both cases the GlcN′ (E. coli) or the GlcN3N′ (C. canimorsus) moieties (2′ NH group) interacted with the backbone carbonyl of Ser120 establishing a strong (about 4–5 kcal/mol) and persistent interaction. Favorable interactions were also observed between GlcN and residues Phe121 and Lys122. The side chain of Phe121 established a strong apolar interaction (Van der Waals, non-polar solvation) with the extended R3 acyl chain in both complexes. The hydrogen bond between the NH group of Ser120 and the carbonyl of the R3′ chain was found to be strong and persistent in both cases. Neither the 1 phosphate group (E. coli) nor the 1 P-Etn (C. canimorsus) established favorable interactions with MD-2, whereas the 4′ phosphate group (missing in C. canimorsus) could be accounted for the slightly greater stability of the MD-2 E. coli lipid A complex, due to the strong (about 7.5 kcal/mol) interaction established with both the backbone and the sidechain of Ser118 (see Fig. 6 B). In summary, we found that in the final complex the arrangement of the sugar moieties with respect to the MD-2 and the calculated binding energy for the two complexes was very similar for E. coli lipid A and C. canimorsus lipid A.
C. canimorsus LPS, lipid A or LA-core were further tested for a possible antagonistic activity on the action of E. coli O111 LPS using HEKBlue human TLR4 cells. The cells were preincubated for 3 h with various concentrations of purified C. canimorsus lipid A, LA-core or LPS samples, then stimulated with 5 ng/ml E. coli O111 LPS for further 20–24 h and the TLR4 dependent NFκB activation was measured. C. canimorsus LPS, LA-core and lipid A appeared to be no antagonist of E. coli O111 LPS binding to human TLR4, in contrast to the tetra-acylated antagonist lipid IVA (Fig. 7 A and B). In a second assay, human THP-1 macrophages were preincubated for 3 h with purified C. canimorsus lipid A, LA-core or LPS samples at the concentration indicated. Then the THP-1 cells were stimulated with 1 ng/ml E. coli O111 LPS for further 20 h and TNFα release was measured. C. canimorsus lipid A exhibited no antagonism to E. coli O111 LPS binding to human TLR4 (Fig. 7 D). Again lipid IVA showed the expected antagonism (Fig. 7 C and D). Dependent on the assay no antagonism or a very weak antagonism of C. canimorsus LPS was observed. This is in agreement with the notion of a partial agonist [43], which includes a certain degree of antagonism at sub-agonist concentration.
All tested lipid A and LA-core fractions exhibited no activity towards human TLR2, as tested by HEK293 cells overexpressing human TLR2/MD-2 and a secreted reporter (Fig. S5). This proves that the stimulation of HEKBlue human TLR4 cells with C. canimorsus lipid A-core observed is only due to activation of TLR4.
We showed here that C. canimorsus has a penta-acylated lipid A, a feature often correlated to low endotoxicity [13], [25]. In addition, the ester-bound 4′ phosphate is lacking. This structural feature is known to reduce the endotoxic activity by a factor of ∼100 [13], which can now be better explained based on the recent data obtained with x-ray crystallography on the TLR4/MD-2/LPS complex [21]. In this complex, phosphate groups of lipid A play a crucial role. The 4′ phosphate is thought to bind to positively charged amino acids in the LRR of TLR4 (Arg264, Lys362) as well as to MD-2 (Ser118 and Lys58) in a well-defined manner. This ionic interaction seems to be critical for the ligand affinity of lipid A, enabling formation of a hexameric (TLR4/MD-2/LPS)2 complex necessary for signalling [21]. In the endotoxic lipid A, there is another negatively charged group, 1 phosphate, which binds to positively charged amino acids in the complex, especially in the LRR of both TLR4 and the counter TLR4, called TLR4* (Lys388* of TLR4*, Lys341, Lys362 of TLR4) and also to Arg122 of MD-2. In contrast to the 4′ phosphate which binds to two proteins (TLR4 and MD-2), the 1 phosphate is involved in binding to three proteins in the complex (TLR4, TLR4*, and MD-2), suggesting that this group might be even more important for the formation of a stable hexameric (LPS/TLR4/MD-2)2 complex, as has been reported [44]. We showed in this work that the lipid A of C. canimorsus contains a P-Etn group at position 1, thus neutralizing the negative charge of the 1 phosphate group. Therefore, we propose that such modified phosphorylation may exert a “shielding effect” on the negative charge of the phosphate and, hence, can explain why the lipid A of C. canimorsus is significantly reduced in its endotoxic activity.
The lipid A structure of C. canimorsus is similar to that of the closely genetically related E. meningoseptica with respect to the nature and position of the acyl chains [30]. As reported for E. meningoseptica, we also found some heterogeneity with respect to the nature of the amino sugar at the non-reducing end in the lipid A backbone, but it was significantly lower (2–5% in C. canimorsus compared to ∼30% in E. meningoseptica) [30]. It has to be pointed out that this structural modification has no influence on the biological activity of lipid A, as it was shown for Campylobacter jejuni [45]. The Etn substitution at position 1 of C. canimorsus lipid A is however not present in E. meningoseptica [30]. One might thus expect that the lipid A of C. canimorsus is less endotoxic than that of E. meningoseptica. To confirm this suggestion a comparative study of lipid A of both species must be carried out. Since the genus Capnocytophaga belongs to the Bacteroidetes phylum [46], it is also not surprising that the structure of lipid A from C. canimorsus shares some important traits involved in specific TLR4 and MD-2 binding with the structure of Bacteroides fragilis lipid A, which we determined earlier [47]. In particular, the lipid A from both bacteria are (3+2) penta-acylated, lack the 4′ phosphate and share iso-branched and linear acyl chains, including i15:0, 16:0(3-OH), and i17:0(3-OH).
In agreement with its structural specifics, C. canimorsus lipid A was shown here to exhibit a very low activity towards human TLR4. C. canimorsus LPS and LA-core are 100- respectively 1000- fold less active than E. coli O111 LPS towards human TLR4, which reminds the activity of the closely related lipid A of E. meningoseptica [30].
The data obtained with human TLR4 may seem to contradict previous findings that whole heat killed C. canimorsus bacteria do not stimulate human TLR4 [9]. However, in that early study, only one concentration of bacterial lysate was used and compared to purified E. coli LPS. From the results presented here, we know that below a certain concentration, pure C. canimorsus LPS is weakly active and the threshold concentration for endotoxicity is higher than that of E. coli LPS. Thus, the C. canimorsus extracts used in previous experiments may have contained LPS in insufficient concentrations. In contrast to what was shown in Capnocytophaga ochracea [48], C. canimorsus LPS and lipid A were found not to antagonize the action of E. coli LPS on human TLR4.
The endotoxicity of the C. canimorsus LPS is probably reduced to the level, which is tolerable in the dog's mouth. We found C. canimorsus LPS was even slightly less active towards canine than human TLR4 in comparison to E. coli LPS. This reduced inflammatory potential might benefit colonization of the dog's mouth. This reduced endotoxicity may probably as well explain why the disease in humans often begins with mild symptoms [2], [6], [49] and finally progresses to severe septicemia with shock and intravascular coagulation. The higher threshold concentration for endotoxicity of C. canimorsus LPS is in line with an initial immune evasion. Nevertheless, at high concentrations it reaches an activation comparable to the highly active E. coli LPS, which might contribute substanitally to the septic shock observed in patients suffering from C. canimorsus infections. Features of the LPS could therefore account for initial evasion of C. canimorsus from the host immune system, while the same LPS might later on induce the endotoxic shock when present at higher concentration.
E. coli lipid A and O111 LPS exhibit a 10- to 100-fold difference in endotoxicity and similar findings were made for P. gingivalis or Proteus mirabilis [27], [28]. The lipid A from E. meningoseptica also shows only minor differences in TLR4 activation to its LPS [30]. In contrast, we found that C. canimorsus lipid A was around 20,000 fold less endotoxic than the LA-core, even higher when compared on a molar basis, suggesting an important role of the core-oligosaccharide in TLR4/MD-2 binding and activation. This indicates the importance of the LPS core for TLR4 activation in the case of C. canimorsus, which has a lipid A devoid of a net negative charge. The C. canimorsus LPS core exhibits only one unshielded negative charge, on the carboxylic oxygen of Kdo. The negative charged carboxyl-group of Kdo in the C. canimorsus core could therefore directly participate in TLR4 or MD-2 binding, besides the reported inner core interactions with TLR4/MD-2 [21]. We found that the MD-2 binding ability of C. canimorsus lipid A is strongly reduced compared to the LA-core and we could exclude that changes in solubility were the reason for the differences observed. This finding could explain the difference in endotoxicity, as a lipid A not properly bound to MD-2 cannot activate TLR4. It seems as if the C. canimorsus LPS core interacts with CD14, LBP or MD-2 and thus enables the binding to MD-2. By molecular modeling C. canimorsus lipid A was predicted to bind MD-2 in a very similar way as E. coli lipid A and the calculated binding energy for the two complexes was similar. As the energetic state of the final complex would therefore be stable and favorable in the case of C. canimorsus lipid A, we propose that the interactions of the LPS core with MD-2 (or LBP/CD14) preceed the final lipid A – MD-2 binding, rather than only stabilizing it. In our model, summarized in Fig. 8, we suggest an intermediate state in which the lipid A in the case of E. coli or the core in the case of C. canimorsus form ionic interactions or hydrogen bonds with MD-2 allowing the lipid A – MD-2 complex to form at all. However, we could not rule out a direct role of the LPS-core in binding to CD14 or LBP. To our knowledge, this is the first reported example of the core-oligosaccharide changing dramatically the endotoxicity of lipid A.
13:0(3-OH) was purchased from Larodan, Malmö, Sveden and 2,3 diamino-2,3-dideoxy-d-glucose (2× HCl) from United States Biochemical Corporation, Cleveland, OH, USA. All other chemicals, solvents and reagents were of highest purity commercially available. E. coli O111 LPS was purchased from Sigma-Aldrich, lipidIVA from PeptaNova. E. coli F515 lipid A (hexa- and penta-acyl) was purified as described [50], [51]. The analysis and isolation of C. canimorsus LA-core will be described elsewhere (Zähringer et al., manuscript in preparation). Purchased reagents were resolved according to manufacturer's instructions. Aliquots of lipid IVA were kept at −80°C.
C. canimorsus bacteria were harvested from 600 blood plates in phosphate buffered saline (PBS) and washed with distilled water, ethanol (300 ml) and acetone (300 ml), followed each time by centrifugation at 18,000× g for 30 min. Bacteria were air dried and resuspended in PBS containing 1% phenol for killing and storage in the deep freezer prior to LPS extraction. Cells were washed with ethanol, acetone and diethyl ether (each 1 L) under stirring (1 h, room temperature). After centrifugation cells were dried on air to give 11.2 g. For the isolation of LPS, C. canimorsus 5 bacteria were extracted by phenol-water [52]. The LPS was identified in the water phase, which also contained a large amount of an unknown glucan polymer separated by repeated ultracentrifugation (100,000× g, 4 h, 4°C, 3 times). The glucan was further analyzed (U. Zähringer and S. Ittig, manuscript in preparation) and the LPS identified in the sediment. The crude LPS preparation was further subjected to RNAse/DNAse treatment (30 mg, Sigma) for 24 h at room temperature followed by Proteinase K digestion (30 mg, 16 h, room temp.) and dialysis (2 days, 4°C), and lyophilization. The yield of enzyme-treated LPS related to bacterial dry mass was 70 mg (0.6%).
Lipid A was prepared from C. canimorsus 5 (25 mg) LPS by hydrolysis with 2% AcOH (4 ml) at 100°C until precipitation of lipid A (2–8 h). The sediment was extracted three times with a water-chloroform mixture (10 ml) and the organic phase was concentrated to dryness under a stream of nitrogen to give 17.7 mg of crude lipid A. The lipid A was purified by reversed phase HPLC as described elsewhere [53] with the following modifications: an Abimed-Gilson HPLC system equipped with a Kromasil C18 column (5 µm, 100 Å, 10×250 mm, MZ-Analysentechnik) was used. Crude lipid A samples (2–5 mg) were suspended in 0.4 mL solvent A and the mixture was sonicated. A 0.1 M EDTA-sodium salt solution (100 µl, pH 7.0) was added forming a bi-phasic mixture which was vortexed and injected directly onto the column. Samples were eluted using a gradient that consisted of methanol-chloroform-water (57∶12∶31, v/v/v) with 10 mM NaOAc as mobile phase A and chloroform-methanol (70.2∶29.8, v/v) with 50 mM NaOAc as mobile phase B. The initial solvent consisted of 2% B which was maintained for 20 min after injection, followed by a linear three step gradient raising from 2 to 17% B (20–50 min), 17 to 27% B (50–85 min), and 27 to 100% B (85–165 min). The solvent was held at 100% B for 12 min and re-equilibrated 10 min with 2% B and hold for additional 20 min before the next injection. The flow rate for preparative runs was 2 ml/min (∼80 bar) using a splitter (∼1∶35) between the evaporative light-scattering detector (ELSD) and fraction collector. The smaller part of the eluate was split to a Sedex model 75C ELSD (S.E.D.E.R.E., France) equipped with a low-flow nebulizer. The major part was collected by a fraction collector in 1 min intervals (∼2 ml each). Nitrogen (purity 99.996%) was used as gas to nebulize the post column flow stream at 3.5 bar into the detector at 50°C setting the photomultiplier gain to 9. The detector signal was transferred to the Gilson HPLC Chemstation (Trilution LC, version 2.1, Gilson) for detection and integration of the ELSD signal.
Sugar and fatty acid derivatives were analysed by gas-liquid chromatography (GLC) on a Hewlett-Packard HP 5890 Series chromatograph equipped with a 30-m fused-silica SPB-5 column (Supelco) using a temperature gradient 150°C (3 min)→320°C at 5°/min. GLC-MS was performed on a 5975 inert XL Mass Selective Detector (Agilent Technologies) equipped with a 30-m HP-5MS column (Hewlett-Packard) under the same chromatographic conditions as in GLC.
Analyses of lipid A were performed in negative and positive ion modes on a high resolution Fourier transform ion cyclotron resonance mass spectrometer, FT ICR-MS (Apex Qe, Bruker Daltonics, Billerica, MA, USA), equipped with a 7 T superconducting magnet and an Apollo dual electrospray-ionization (ESI)/Matrix-assisted laser desorption ionization (MALDI) ion source. Data were recorded in broadband mode with 512 K data sampling rate. The mass scale was calibrated externally by using compounds of known structure. For the negative ion mode samples (ca. 10 ng/µl) were dissolved in a 50∶50∶0.001 (v/v/v) mixture of 2-propanol/water/triethylamine (pH∼8.5). For the positive ion mode samples, a 50∶50∶0.03 (v/v/v) mixture of 2-propanol/water/30 mM ammonium acetate adjusted with acetic acid to pH 4.5 was used. The samples were sprayed at a flow rate of 2 µL/min. The capillary entrance voltage was set to 3.8 kV and the drying gas temperature to 150°C. The mass numbers given refer to that of the monoisotopic ion peak. For MS/MS in the positive ion-mode small amounts of TEN were added to the sample preparation to obtain the [M+TEN+H]+ adduct ions [32] which were selected for collision induced decay (CID) in the collision cell infrared multiphoton dissociation (IRMPD) within the ion cycIotron resonance (ICR) cell.
Lipid A samples (1–3 mg) were exchanged twice with deuterated solvents [chloroform-d1/methanol-d4 1∶1 (v/v), Deutero GmbH, Kastellaun, Germany] and evaporated to dryness under a stream of nitrogen. Samples were dissolved in 180 µl chloroform-d1/methanol-d4/D2O 40∶10∶1 (v/v/v, 99.96%) and analyzed in 3 mm NMR tubes (Deutero). 1H-, 13C-, and 31P-NMR spectra were recorded at 700.7 MHz (1H) on an Avance III spectrometer equipped with a QXI-cryoprobe (Bruker, Germany) at 300 K. Determination of NH-proton signals was performed in chloroform-d1(99.96%)/methanol/H2O 40∶10∶1 without exchange in deuterated solvents. Chemical shifts were referenced to internal chloroform (δH 7.260, δC 77.0). 31P NMR spectra were referenced to external aq. 85% H3PO4 (δP 0.0). Bruker software Topspin 3.0 was used to acquire and process the NMR data. A mixing time of 100 ms and 200 ms was used in TOCSY and ROESY experiments, respectively.
Quantification of GlcN, GalN (internal standard) and GlcN3N by GLC and GLC-MS was done after strong acid hydrolysis of 0.5 mg lipid A in 4 M HCl (16 h, 100°C), followed by acetylation (N-acetylation) in pyridine/acetic acid anhydride (10 min, 85°C), reduction (NaBH4) and per-O-acetylation. The response factor of the per-O-acetylated GlcNAc-ol, GalNAc-ol, and GlcNAc3NAc-ol derivatives, necessary for the quantification of GlcN3N by GLC, was determined in addition by external calibration with synthetic reference sugars. Etn, GlcN, GlcN3N and their corresponding phosphates (GlcN-P and Etn-P), were determined from the hydrolysate by reversed phase HPLC using the Pico-tag method and pre-column derivatization with phenylisothiocyanate according to the supplier's instructions (Waters, USA). Quantification of total phosphate was carried out by the ascorbic acid method [54]. For analysis of ester- and amide-linked acyl chains, the lipid A was isolated from LPS (1 mg) by mild acid hydrolysis (0.5 mL, 1% AcOH, 100°C, 2 h), centrifuged and the lipid A sediment was separated into two aliquots and lyophilized. Ester-linked acyl chains were liberated from the first aliquot by treatment with 0.05 M NaOMe in water-free methanol (0.5 mL) at 37°C for 1 h. The mixture was dried under a stream of nitrogen and acidified (M HCl) prior to extraction with chloroform. The free acyl chains were converted into methyl esters by treatment with diazomethane and hydroxylated acyl chains were trimethylsilylated with N,O-bis(trimethylsilyl)trifluoroacetamide for 4 h at 65°C [55]. The acyl chain derivatives were quantified by GLC-MS using the corresponding derivatives of 17:0 (50 µg) and 13:0(3-OH) (50 µg, Larodan, Malmö, Sweden) as internal standards for the calibration of the response factor of non-hydroxylated and hydroxylated acyl chains, respectively. For analysis of total acyl chains, the second aliquot was subjected to a combined acid/alkaline hydrolysis as described [56]. Briefly, acyl chains were liberated from the lipid A by strong acid hydrolysis (4 M HCl, 100°C, 21 h) and extracted three times with water/chloroform (0.5 mL each). The organic phase containing the N- and O-linked acyl chains was treated with diazomethane, trimethylsilylated and quantified as described above.
The strains used in this study are listed in Table S2. E. coli strains were grown in LB broth at 37°C. C. canimorsus 5 [9] was routinely grown on Heart Infusion Agar (HIA; Difco) supplemented with 5% sheep blood (Oxoid) for 2 days at 37°C in presence of 5% CO2. Bacteria were harvested by scraping colonies off the agar surface, washed and resuspended in PBS. Selective agents were added at the following concentrations: erythromycin, 10 mg/ml; cefoxitin, 10 mg/ml; gentamicin, 20 mg/ml; ampicillin, 100 mg/ml.
HEK293 stably expressing human TLR4, MD-2, CD14 and a secreted NFκB dependent reporter were purchased from InvivoGen (HEKBlue hTLR4). Growth conditions and endotoxicity assay were as recommended by InvivoGen. Briefly, desired amount of LPS or lipid A were placed in a total volume of 20 µl (diluted in PBS) an added a flat-bottom 96-well plate (BD Falcon). 25000 HEKBlue hTLR4 cells in 180 µl were then added and the plate was incubated for 20–24 h at 37°C and 5% CO2. If the antagonistic activity of a compound on the action of E. coli O111 LPS was assayed, the compound was added in a total volume of 10 µl (diluted in PBS), 25000 HEKBlue hTLR4 cells in 180 µl were added and the plate was incubated for 3 h at 37°C and 5% CO2. Then the cells were stimulated with 5 ng/ml E. coli O111 LPS and the plate was incubated as above. Detection followed the QUANTI-Blue protocol (InvivoGen). 20 µl of challenged cells were incubated with 180 µl detection reagent (QUANTI-Blue, InvivoGen). Plates were incubated at 37°C and 5% CO2 and colour developed was measured at 655 nm using a spectrophotometer (BioRad). If needed the C. canimorsus lipid A stock solution (1 mg/ml) was supplemented with 0.1% v/v TEN or 50% v/v DMSO and sonicated for some minutes just before the assay. The TEN containing lipid A stock solution was further diluted in a 0.1% TEN solution to keep the TEN concentration constant in all samples. Due to the high concentration of DMSO used, this lipid A stock solution was further diluted with PBS. As a control the same amount of TEN or DMSO has been added to E. coli O111 LPS samples tested in the same assay. DMSO concentration in 50 µg/ml and 5 µg/ml were found to affect physiological test conditions. These data have therefore been excluded from the figure.
Human THP-1 monocytes (ATCC TIB-202) were cultured as recommended by the American Type Culture Collection (RPMI 1640 medium complemented with 10% v/v heat-inactivated fetal bovine serum, 2 mM L-Glutamine). Monocytes were seeded at 1.5×105 cells/ml in 24 well-plates (BD Falcon) in growth medium containing 10−7 M PMA (Sigma-Aldrich). For differentiation and attachment the cells were incubated for 48 h at 37°C and 5% CO2 and then washed with growth medium and fresh PMA-free medium was added. After further incubation for >1 h the cells were challenged with the indicated amount of LPS or lipid A in a total volume of 20 µl (diluted in PBS). After 20 h of incubation the supernatants were harvested and immediately analyzed for TNFα by an ELISA. ELISA was performed in accordance with the manufacturers instructions (BD OptEIA). If an antagonist of E. coli O111 LPS was assayed, the compound was added in a total volume of 10 µl (diluted in PBS) to the THP-1 cells and the plates were incubate for 3 h at 37°C and 5% CO2. Then the cells were stimulated with 1 ng/ml E. coli O111 LPS and the plate was incubated for 20 h at 37°C and 5% CO2.
Canine DH82 macrophages (ATCC CRL-10389) were cultured in DMEM supplemented with 15% v/v heat-inactivated fetal bovine serum, 2 mM L-Glutamine and non-essential amino acids (Sigma-Aldrich) in a humidified incubator at 37°C and 5% CO2. Cells were seeded at 1×105 cells/ml in 24 well-plates (BD Falcon). The cells were incubated for 24 h at 37°C and 5% CO2, before being challenged with the indicated amount of LPS in a total volume of 10 µl (diluted in PBS). After 14 h of incubation the supernatants were harvested and immediately analyzed for content of IL-6 by an ELISA. ELISA was performed in accordance with the manufacturer's instructions (R&D Systems, DY1609).
Biotinylation of E. coli O111 LPS (Sigma-Aldrich) was performed as described previously [57] using biotin-LC-hydrazide (Pierce, Rockford, IL). To verify that the biotinylation did not affect the functionality of the LPS, E. coli LPS-Biotin was assayed for endotoxicity with the HEKBlue human TLR4 cell line (Data not shown). Biotinylation reduced the endotoxic potential at low concentrations, but only slightly at concentrations used in the MD-2 binding assay.
MD-2 binding assays was performed as described [57], [58]. HEK293 cells were transfected using Fugene6 (Roche, 3∶2 protocol) with a plasmid (kind gift of K. Miyake and C. Kirschning) encoding human MD-2 with a C-terminal Flag-His-tag (pEFBOS-hMD2-Flag-His) [15]. The medium was exchanged 3–8 h post transfection with fresh growth medium. The cells were incubated for 48 h and the supernatant was harvested and pooled. Fresh FCS was added to the hMD-2 supernatant (7.5% v/v) as a source of CD14 and LBP. For each binding reaction, 4 ml of hMD-2 supernatant were combined with 250 ng, 500 ng, 1 µg, 2 µg, 5 µg or 10 µg of the competitor, incubated at room temperature and gently rocked for 30 min. If needed the C. canimorsus lipid A stock solution (1 mg/ml) was supplemented with 0.1% v/v TEN or 50% v/v DMSO and sonicated for some minutes just before addition to the hMD-2 supernatant. 1 µg of biotinylated E. coli O111 LPS was added and the supernatant was further incubated for 3–4 h at room temperature. Biotinylated LPS–hMD-2 complexes or single biotinylated LPS were captured by addition of 120 µl (total volume) streptavidin-agarose beads (IBA) per sample. The beads were previously prepared by washing them three times with a buffer (100 mM Tris, 150 mM NaCl, pH 8.0). For binding, the supernatants containing the beads were incubated overnight on a rotator at 4°C. Agarose beads were pelleted by centrifuging for 30 s at 5000× g and 4°C and washed three times with PBS containing 0.5% Tween 20. The beads were finally resuspended in 60 µl SDS-loading dye (without dithiothreitol) and boiled for 5 min at 95°C. The protein content in the sample was analyzed by non-reducing, denaturing 4–12% Tris-glycine Polyacrylamide gels (Invitrogen) or 4–15% Tris-glycine Polyacrylamide gels (BioRad) and then transferred to polyvinylidene fluorid (PVDF) membrane (ImmobilonP, Millipore). Membranes were probed using monoclonal anti-Flag antibody (Sigma-Aldrich) according to the manufacturer's instructions using ECL-Plus reagent (GE Healthcare).
Blast-p search tool [59] against the C. canimorsus 5 genome [5] was used. Search sequences were obtained from the National Center for Biotechnology Information. All available Bacteroidetes-group sequences were used as search if available, but standard E. coli sequences have always been included. The highest scoring subjects over all the searches have been annotated as corresponding enzymes. Difficulties in annotation were only observed for lpxE. lpxE search was based on lpxF and/or lpxE sequences from P. gingivalis [38], F. novicida [60], R. etli [40], H. pylori [37], [41] and on all available Bacteroidetes-group pgpB sequences. Three lpxE/F candidates have been found in the C. canimorsus 5 genome (Ccan 16960, Ccan 14540 and Ccan 6070). All candidates have been deleted and only deletion of Ccan 16960 affected endotoxicity (data not shown). Since this gene is encoded in an operon with the predicted eptA and since the same operon structure (lpxE-eptA) has been identified in H. pylori [37] Ccan16960 was annotated as lpxE.
The MD-2 - E. coli LPS complex (PDB code 3FXI) [21] was used to construct models for the MD-2 - E. coli lipid A and for the MD-2 – C. canimorsus Lipid A. The modeling of the lipid A moieties was performed using the VMD [61] program and the leap module of the AMBER11 [62] suite of programs. To investigate the time-dependent properties of the two MD-2 – lipid A complexes, the constructed systems were subjected to molecular dynamics simulations [63] in the framework of a classical molecular mechanics [64] (MM) description. MM parameters from the Glycam06 [65], [66] force field were adapted to describe the acyl chains and the sugar moieties, while the Amber99SB [67], [68] force field was employed for the MD-2 protein. Advanced methods based on quantum chemistry were employed to obtain the missing parameters of the ester linkages and hydroxyl groups on the acyl chain C2 atoms, the branching at the bottom of the C. canimorsus acyls, the phosphate/P-Etn groups and the GlcN3N′ moiety. Bonding parameters were obtained by performing relaxed potential energy scans [69] (bonds, angles, dihedrals), while charges were calculated on the optimized geometries of selected capped fragments. All the scan and geometry optimizations were conducted at the RI-MP2/def2-TZVP [70], [71], [72] level using the COBRAMM [73] suite of programs efficiently linking the ORCA2.8 [74] (wave-function calculation) and the GAUSSIAN09 [75] (optimization/scan driver) programs. Charges were calculated according to the RESP procedure at the HF/6-31G*//MP2/def2-TZVP. Both MD-2 – lipid A complexes were embedded in a 6.5×6.5×6.5 nm3 box of TIP3P [76] water molecules and the appropriate number of Na+ and Cl− ions were added to neutralize the systems charge. The systems were relaxed (conjugate gradient geometry optimization) to remove clashes before stating molecular dynamics simulations. The systems were both heated to 300 K in the NVT (constant particle number, volume, temperature) ensemble for 500 ps and then equilibrated in the NPT (constant particle number, pressure, temperature) until relevant structural parameters (density, RMSD on the protein Cα) were found to be stable (1 ns). Statistics were then performed on trajectories collected from 10 ns long simulations of the equilibrated systems. All molecular dynamics calculations were performed with the sander module of the AMBER11 package; bonds involving H atoms were constrained using the SHAKE algorithm [77] to allow for using a time step of 2 fs. Pressure was controlled via a simple Berendsen weak coupling approach [78], while a Langevin thermostat (collision frequency set to 3 ps−1) was used to enforce the desired temperature. Molecular dynamics trajectories were analyzed using the VMD software, the ptraj module of the AMBER11 suite and the ProDy [79] package. A set of 300 snapshots of the equilibrated trajectories was subjected to further analysis to quantify the binding energy between MD-2 and each of the two lipid A moieties. Both the MM-PBSA and MM-GBSA approaches [80] were used to calculate the MD-2 – lipid A binding energy, while a full interaction energy decomposition [81], [82] was performed using the cheaper MM-GBSA method; the MMPBSA.MPI module of AMBER11 was used to perform the binding free energy calculations, while a locally developed software was used to process, analyze and plot the results.
Quantification was performed using MultiGauge software (Fujifilm).
|
10.1371/journal.pntd.0003520 | CD8+ T Lymphocyte Expansion, Proliferation and Activation in Dengue Fever | Dengue fever induces a robust immune response, including massive T cell activation. The level of T cell activation may, however, be associated with more severe disease. In this study, we explored the level of CD8+ T lymphocyte activation in the first six days after onset of symptoms during a DENV2 outbreak in early 2010 on the coast of São Paulo State, Brazil. Using flow cytometry we detected a progressive increase in the percentage of CD8+ T cells in 74 dengue fever cases. Peripheral blood mononuclear cells from 30 cases were thawed and evaluated using expanded phenotyping. The expansion of the CD8+ T cells was coupled with increased Ki67 expression. Cell activation was observed later in the course of disease, as determined by the expression of the activation markers CD38 and HLA-DR. This increased CD8+ T lymphocyte activation was observed in all memory subsets, but was more pronounced in the effector memory subset, as defined by higher CD38 expression. Our results show that most CD8+ T cell subsets are expanded during DENV2 infection and that the effector memory subset is the predominantly affected sub population.
| Dengue is a disease affecting approximately 400 million people annually, especially in tropical and subtropical areas of the globe. The immune response against the dengue virus is still under investigation and it is important to understand why the disease can be fatal in a small proportion of cases. In this work, we explored how an important cell type of the immune system, namely the CD8+ T cell, reacts during dengue infection. Using a method known as flow cytometry, we demonstrated that these cells expand and become highly activated, during the days following the onset of dengue fever symptoms. This expansion is associated with a decreased dengue virus load in the patients’ blood, suggesting that CD8+ T cells play an important role in viral control. Interestingly, we found that a subset of CD8+ T cells, called effector memory, is greatly expanded during dengue infection. Our results are important because they might contribute to the understanding of disease mechanisms during dengue infection and may help in the development of a novel vaccine against dengue.
| Dengue is the most prevalent arthropod-born viral disease in tropical and subtropical areas of the globe, affecting approximately 400 million people annually [1]. The World Health Organization estimates that nearly 40% of the world’s population lives in areas at risk for dengue transmission. Dengue cases in Central and Latin America have increased almost five-fold in the last 30 years. During 2008, up to one million cases were reported in Americas, and higher numbers of deaths were documented in the South [2]. In the latest decades, Brazil has been hard hit by the disease, accounting for more than 60% of the total reported cases in the Americas [2]. The continuing occurrence of the disease in resource limited countries and the lack of novel therapeutic approaches or a highly effective vaccine make dengue fever a neglected disease. Surveillance for dengue is absent in most countries, and no existing model for predicting an outbreak in endemic regions is widely available. Therefore, it is important to increase our knowledge of disease pathogenesis, with the goal of developing new strategies to fight the epidemic.
The mechanisms by which the dengue virus (DENV) causes severe illness remain to be elucidated. Both biological properties of the viral isolates and immunogenic host factors seem to contribute to the level of pathogenicity [3,4,5,6]. Whereas immunity induced by natural infection is believed to provide serotype-specific lifelong protection, previous infection by a distinct serotype is considered to increase the risk for the development of dengue hemorrhagic fever (DHF) and dengue shock syndrome (DSS) [5,7]. The immunological processes during dengue infection are not yet completely defined. However, incidence of mild dengue manifestations and occasional progression to the more severe disease likely reflect a complex interplay between host and viral factors including cytokine production by inflammatory cells. Previous studies reported increased levels of circulating cytokines and soluble receptors in DHF patients when compared to those with dengue fever (DF), suggesting that immune activation may be related to disease severity [8]. T cell activation mechanisms are based on the binding of specific T cell receptors (TCRs) to MHC molecules [9]. CD8+ T cells are one of the most important cell types to recognize and eliminate infected cells. Some authors have suggested that high numbers of CD8+ T cells might be protective by reducing viral load [10]. Memory T lymphocytes remain present in the absence of antigenic stimulation and have the capacity to expand rapidly upon secondary challenge. In the last decade, several surface markers have been used to distinguish among effector memory (TEM), central memory (TCM), and terminally differentiated memory cells (TEMRA) [11]. In this work, we explored the state of CD8+ T cell activation in different compartments during the acute phase of dengue fever.
All procedures adopted in this study were performed according to the terms agreed by the Institutional Review Board from the Hospital das Clínicas, University of São Paulo (CAPPesq—Research Projects Ethics Committee). This study was approved by CAPPesq under protocol 0652/09. Written informed consents were obtained from all study volunteers.
Whole-blood samples were collected, using sterile EDTA-treated Vacutainer tubes (BD Brazil), from patients with DENV2 dengue at the Ana Costa Hospital, Santos, State of São Paulo, during the 2010 first semester outbreak [12]. Patients with suspected dengue fever, dengue with warning signs or severe dengue were invited to participate in the study. A rapid rest was performed to confirm the diagnosis of acute dengue disease, followed by the detection of dengue viral load determination (see next sections). Primary dengue infection was considered when dengue IgG-specific antibodies were not detected in the presence of reactive dengue IgM-specific antibody and/or NS1 antigenemia. Secondary dengue infection was considered in the presence of dengue IgG-specific antibodies at acute phase up to 6 days of symptoms.
A commercial Dengue Duo Test Bioeasy (Standard Diagnostic Inc. 575–34, Korea), a rapid test kit, was used for dengue diagnosis, by detection of both dengue virus NS1 antigen and IgM- and IgG-specific antibodies in human blood. Samples were considered positive for acute dengue fever when NS1 or IgM bands were reactive in the testing kit. We also considered an acute case whenever DENV2 RNA was detected.
The IgG avidity test was used to determine if patients presented with a primary or secondary DENV infection [13]. Samples with low avidity IgG antibodies were classified as primary DENV infection, whereas samples with high avidity IgG antibodies were classified as secondary. Samples in which IgG antibodies were not detected could not be classified, although the majority were probably from primary DENV infection. Viral load was determined by an “in-house” real-time polymerase chain reaction (RT-PCR) method. RNA was extracted from 140μL of plasma using the Qiagen Viral RNA kit (Qiagen, USA). All RT-PCR reactions were performed in duplicate. RT-PCR was conducted using the SuperScript III Platinum SYBR Green One-Step qRT-PCR kit with ROX (Invitrogen, USA) and 0.4 μM of primers covering all four DENV serotypes [14]. Cycling conditions were: 10 minutes reverse transcription at 60°C, 1 min Taq polymerase activation at 95°C, followed by 45 cycles consisting of 95°C without holding time, 60°C for 3 seconds, and 72°C for 10 seconds. The reaction was run on an ABI 7300 RT-PCR equipment (Applied Biosystems, Brazil). As an internal control Bovine Diarrhea Virus (BDV) was added to the samples before RNA extraction and also run in a parallel RT-PCR assay. Supernatant from DENV-3 cell cultures was included as an external control. DENV-3 supernatant was previously quantified by a commercial dengue RT-PCR kit (RealArt; artus/QIAGEN, Germany) [15] and used to generate a standard curve. The detection limit of this assay was 100 copies/ml.
Peripheral blood absolute CD4+ and CD8+ T cell counts were assessed using the BD Multitest CD3-FITC/CD8-PE/CD45-PerCP/CD4-APC monoclonal antibody (mAb) cocktail from BD Biosciences (San Diego, CA), according to the manufacturer’s instruction, using a FACSCanto flow cytometer (BD Biosciences). Cell surface staining was routinely performed on 100 μL fresh whole blood.
Peripheral blood mononuclear cells (PBMCs) were isolated from fresh EDTA-treated blood by Ficoll-Hypaque gradient centrifugation and frozen in liquid nitrogen as previously described [16]. PBMC were isolated from volunteers and stored in liquid nitrogen until used in the assays. To characterize the activation profile of CD8+ T lymphocytes, we used the markers HLA DR and CD38. HLA DR is a transmembrane glycoprotein encoded by genes within the Human Leucocyte Antigen (HLA) complex. CD38 is a nonlineage-restricted type II transmembrane glycoprotein that has emerged as a multifunctional protein. Cells expressing both markers are likely to be activated. The following monoclonal antibodies were used in the FACS assays: anti-CD8-peridin chlorophyll protein (PerCP) (clone SK1), CD45RA-fluorescein isothiocyanate (FITC) (clone L48), CD38-phycoerythrin (PE) (clone HB7), from BD Biosciences (San Jose, CA, USA); CCR7-phycoerythrin—cyanine (PE-Cy7) (clone 3D12), HLADR-Alexa 700 (clone L243), CD27-APCH7 (clone M-T271), CD4-Pacific Blue (clone RAPA-T4), from BD Pharmingen (San Jose, CA, USA); CD3-ECD (clone UCHT1), from Beckman Coulter (Marseille, France) and Fixable Aqua dead cell stain kit, from Molecular Probes (Oregon, USA).
After thawing, cells were centrifuged at 1,500 rpm for 5 minutes and transferred into 96 V bottom well plates (Nunc, Denmark) in 100 L of staining buffer (PBS supplemented with 0.1% sodium azide [Sigma] and 1% FBS, pH 7.4–7.6) with the surface monoclonal antibodies panel. Cells were incubated at 4C in the dark for 30 minutes, washed twice, and resuspended in 100 L of fixation buffer (1% paraformaldehyde Polysciences, Warrington, PA in PBS, (pH 7.4–7.6). Fluorescence minus one (FMO) was used for gating strategy [17]. The strategy is shown in S1 Fig.
CD8+ T cell proliferation was assessed using a Ki67 staining protocol. Ki67 is a cell-cycle-associated antigen expressed exclusively in proliferating cells. After staining with surface markers CD3-PERCP (clone SK7), CD8-allophycocyanin cyanine-7 (APC-Cy7) (clone SK7), from BD Biosciences (San Jose, CA, USA); CD4-Alexa 700 (clone RAPA-T4), from BD Pharmingen (San Jose, CA, USA) and Fixable Aqua dead cell stain kit, as described above, cells were fixed with 4% fixation buffer for 10 minutes. Cells were washed with staining buffer once and re-suspended in 100 L of permeabilization buffer from BD Biosciences (San Jose, CA, USA) and incubated for 15 minutes. Cells were washed with staining buffer twice. Ki-67-FITC (clone B-56) was added and cells incubated at 4C in darkness for 30 minutes. Finally, the cells were washed twice, and re-suspended in 100 L of 1% fixation buffer.
Samples were acquired on a FACSFortessa, using FACSDiva software (BD Biosciences), and then analyzed with FlowJo software version 9.4 (Tree Star, San Carlo, CA). Fluorescence voltages were determined using matched unstained cells. Compensation was carried out with CompBeads (BD Biosciences) single-stained with. Samples were acquired until at least 200,000 events were collected in a live lymphocyte gate. The analysis strategy is shown in S2 Fig.
Because most continuous variables presented an overdispersed distribution, results were summarized as medians and 25% to 75% interquartile ranges (IQR) and compared across dengue patient groups and non exposed controls, using nonparametric Kruskal-Wallis or Mann-Whitney tests (continuous variables). When the Kruskal-Wallis test indicated a statistically significant difference (P<0.05) among more than two groups, a Dunn’s multiple comparison post-tests was carried out to determine between which groups the differences were sustained. Potential correlations were explored using Spearman rank correlation tests. The software Prism, version 5.0, was used for analyses (GraphPad Software, San Diego, CA).
Peripheral venous blood was obtained from 74 patients with acute dengue fever and 17 matched donors who were asymptomatic and negative for DENV IgM, NS1, and RNA. The characteristics of the dengue fever patients and healthy controls are depicted in Table 1. No differences were seen in gender and age distribution comparing both groups. As expected, dengue fever patients had lower number of platelets (median 152,000 cells/μl, interquartile range 25%–75% [IQR], 110,000–207,000) when compared to controls (median 226,000 cells/μl, IQR, 166,000–310,000), p<0.0001. Platelets decreased during the first days of disease, with a median of 174,000 cells/μl (IQR, 147,000–232,000) on days 1 and 2, 153,000 cells/μl (IQR, 115,000–206,000) on days 3 and 4, and 94,000 cells/μl (IQR, 28,000–154,000) on days 5 and 6 after the onset of symptoms.
Overall leukocyte counts were also lower (median 4,400 cells/μl, IQR, 3,275–6,400) compared to controls (median 8,100 cells/μl, IQR, 6,140–9,335), p<0.0001. Numbers were lower on days 1 and 2 (median 5,100 cells/μl, IQR, 3,750–6,450) and 3 and 4 (median 3,600 cells/μl, IQR, 3,100–5,100), p<0.001, but recovered on days 5 and 6 to levels of the control group.
A subset of 30 dengue fever patients (Table 2) was selected for expanded immunophenotyping experiments. To be representative of the disease natural history after onset of symptoms, 10 of these patients were at days 1 and 2, 10 patients at days 3 and 4, and 10 patients at days 5 and 6, as detailed in Table 2, along with 17 healthy controls.
We first evaluated the percentages and the absolute numbers of CD8+ T lymphocytes in acute dengue fever patients. As shown in Fig. 1A, the percentage of CD8+ T cells of overall circulating lymphocytes remained constant up to the fourth day after onset of symptoms. However, this percentage increased on days 5 and 6, with an increased median of 38%, IQR, 29–53 (p<0.05), and this was higher than observed in healthy controls.
Absolute numbers of CD8+ T cells were lower from the first to the fourth days after onset of symptoms (median 253 cells/μl, IQR, 151–358 for days 1 and 2; median 201 cells/μl, IQR, 158–345 for days 3 and 4) when comparing dengue patients with healthy controls (median 465 cells/μl, IQR, 329–605). On the fifth and sixth days, we observed a higher number of cells (median 534 cells/μl, IQR, 285–1644), with wider distribution values (Fig. 1B).
Dengue viral load was evaluated in the course of dengue fever. As expected, higher viral loads were observed in the first and second days after onset of symptoms, as shown in Fig. 2A, decreasing thereafter. Remarkably, dengue viral load negatively correlated with the number of circulating CD8+ T lymphocytes. As demonstrated in Fig. 2, we observed that higher viral load was seen only when CD8+ T lymphocytes remained below 450 cells/μl (arbitrary dotted line parallel to the y axis), whereas higher CD8+ T lymphocyte counts were associated with Dengue viral load bellow 1,050 copies/ml (arbitrary dotted line parallel to the x axis). These results imply that these CD8+ T cells may be playing a role in the control of DENV replication in the acute phase of disease. Of note, statistically significant correlations (p<0.05) were seen on days 1 and 2 (r = -0.6) (Fig. 2B) and on days 5 and 6 (r = -0.5) (Fig. 2D). In contrast, no correlation was observed on days 3 and 4 (Fig. 2C).
Samples from 30 patients with dengue fever were selected for the remaining experiments. The aim was to be representative of the disease natural history, up to six days after onset of symptoms. Results are shown for days 1 and 2 (10 patients), 3 and 4 (10 patients), and 5 and 6 (10 patients), as detailed in Table 2. These were compared to the 17 healthy controls.
We observed that more CD8+ T lymphocytes expressed Ki67 in dengue fever cases when compared to controls, either expressed in absolute numbers or percentage of stained cells (median 14 cells/μl, IQR, 7–40 vs. 4 cells/μl, IQR, 3–12, p = 0.002; median 4%, IQR, 2–14 vs. median 1%, IQR, 1–2, p<0.0001). However, this increase in expression was largely seen on days 5 and 6 (median 113 cells/μl, IQR, 21–418 and median 20%, IQR, 10–31), suggesting that these cells proliferate later in the course of the disease (Fig. 3).
We addressed the levels of CD8+ T lymphocytes activation using surface staining for CD38 and HLA-DR. Coinciding with CD8+ T lymphocyte expansion and proliferation, higher cell activation could be detected later in the course of disease, on days 5 and 6, compared to controls either in percentages (median 34%, IQR, 20–59 vs. median 3, IQR, 3–5, p<0.0001) or in absolute numbers (median 114 cells/μl, IQR, 45–1110 vs. median 14 cells/μl, IQR, 12–21, p<0.0001), as depicted in Fig. 4.
The cellular activation profile was different among the subpopulations of CD8+ T lymphocytes. Using comprehensive staining panels, we did not observe statistically significant differences in activation of naïve cells (Fig. 5A and 5B). On the other hand, higher activation was observed on days 5 and 6 in the central memory (TCM), effector memory (TEM), and terminally effector memory (TEMRA) cell percentages (Fig. 5C, 5E, and 5G, respectively). Nevertheless, this effect was only seen in the TEM subpopulation when absolute numbers were evaluated (Fig. 5F), suggesting that TEM cells are largely responsible for this phenomenon. We also observed that the TEM CD8+ T cell subset was negatively correlated with DENV viral load, suggesting that the activation of such particular phenotype may have a central role in controlling virus replication; however, given the post-hoc nature of this analysis this result needs to be interpreted with caution, requiring confirmatory experiments.
Changes in lymphocyte subsets in dengue fever have long been recognized, including an increase in CD8+ T lymphocyte numbers [8,18,19]. Using samples collected in a DENV2 outbreak in the coast of the State of São Paulo, Brazil [12], we were able to demonstrate that the percentage CD8+ T cell count increased later in the course of disease, after onset of symptoms. This was associated with higher Ki67 expression, suggesting a proliferative rebound that follows the peak viremia. This phenomenon may be related to increased cell activation, as has been suggested by others [20]. In this paper we explored in more detail the activation status of different CD8+ T cell subpopulations in adult patients with dengue fever.
During high viral burden, several circulating cells are activated including monocytes [21], NK cells, CD4+ and CD8+ T cells [8,22,23]. This activation seems to be a natural immune response to the pathogen and reflects its efforts to control viral replication. Our results show that the expanding number of CD8+ T cells was associated with lower viremia, especially later in the disease course. Dung et al., also demonstrated that activated CD8+ T cell expansion (evaluated by the expression of HLA-DR and CD38 molecules) was associated with viral control [24]. However, it is possible that high levels of cellular activation may be harmful to the host and may be related to disease severity. A number of studies have found increased markers of immune cell activation in patients with dengue hemorrhagic fever (DHF) compared with patients with classic dengue fever (DF) [25]. Indeed, children who developed DHF had higher percentages of CD8+ T cells and NK cells expressing CD69, an early activation marker than those with DF during the febrile period of illness [8,26]. Also, children admitted with acute dengue fever had increased levels of NK cells and T lymphocyte activation and the severity of disease was associated with higher activation status [23].
In the last few years considerable progress has been made in identifying different T cell memory subsets to dissect the heterogeneity of human immune responses [27]. In this paper, we evaluated in different CD8+ T subpopulations in adults with dengue fever using a comprehensive panel of antibodies. Our current study demonstrates differences in activation status among the various CD8+ T cell subpopulations in dengue fever patients. The percentages and numbers of effector memory (TEM) cells, characterized by the CCR7-CD27+CD45RA+/- phenotype [28], were the most activated in the later phase of the disease, as demonstrated by the expression of HLADR and CD38 molecules. Other subpopulations also exhibited increased activation, including central memory (TCM) and terminally differentiated memory cells (TEMRA). However this finding was restricted to the percentage and not the absolute numbers of these two TCM and TEMRA subsets suggesting that TEM cells are the most activated subset in the later stages of acute dengue fever. TEM cells have immediate effector function, by secreting IL-2, IFNγ, and other cytokines in response to infectious pathogens [29,30,31].
One limitation of our study is the lack of any data regarding antigen-specific responses since the observed expansion of TEM cells was described in CD8+ T cells using only surface staining. Further studies using DENV-derived proteins or peptides for stimulation of PBMC from acute dengue cases are warranted. Antigen-specific CD8+ T cell responses have been recently described in general population from Sri Lanka hyperendemic area, with higher magnitude and more polyfunctional responses for HLA alleles associated with decreased susceptibility to severe disease [32].
DENV-reactive CD8+ T cells are important in the control of viral replication [33] and may have different responses to different epitopes [34]. DENV serotype-cross-reactivity of CD8+ T cells has also been demonstrated after primary infection [35]. The observed expansion of TEM cells, which may contain such cells, should be explored in future studies to verify their antigen-specific characteristics [24]. The analysis of human memory T and B cells has the capacity to identify the antigens that are targeted by effector T cells, thus providing a rational for vaccine design. In fact, many dengue vaccine candidates have been using replicating virus, including chimeric dengue virus [36], which can induce a significant immune reaction against the vaccine [37] [38] [39].
Based on our findings that different CD8+ T cell subpopulation are activated to different levels, it may be important to investigate the status of CD8+ T cell differentiation when analyzing antigen-specific responses. Considering the key role of CD8+ T cell activation and antigen-specific responses in the pathogenesis of dengue fever, further investigation should be conducted to explore the mechanisms of activation pathways in disease pathogenesis.
|
10.1371/journal.ppat.1006980 | Viral targeting of TFIIB impairs de novo polymerase II recruitment and affects antiviral immunity | Viruses have evolved a plethora of mechanisms to target host antiviral responses. Here, we propose a yet uncharacterized mechanism of immune regulation by the orthomyxovirus Thogoto virus (THOV) ML protein through engaging general transcription factor TFIIB. ML generates a TFIIB depleted nuclear environment by re-localizing it into the cytoplasm. Although a broad effect on gene expression would be anticipated, ML expression, delivery of an ML-derived functional domain or experimental depletion of TFIIB only leads to altered expression of a limited number of genes. Our data indicate that TFIIB is critically important for the de novo recruitment of Pol II to promoter start sites and that TFIIB may not be required for regulated gene expression from paused promoters. Since many immune genes require de novo recruitment of Pol II, targeting of TFIIB by THOV represents a neat mechanism to affect immune responses while keeping other cellular transcriptional activities intact. Thus, interference with TFIIB activity may be a favourable site for therapeutic intervention to control undesirable inflammation.
| Viruses target the innate immune system at critical vulnerability points. Here we show that the orthomyxovirus Thogoto virus impairs activity of general transcription factor IIB (TFIIB). Surprisingly, impairment of TFIIB function does not result in a general inhibition of transcription but in a rather specific impairment of selective genes. Transcriptome and functional analyses intersected with published CHIP-Seq datasets suggest that affected genes require de novo recruitment of the polymerase complex. Since the innate immune system heavily relies on genes that require de novo recruitment of the polymerase complex, targeting of TFIIB represents a neat mechanism to broadly affect antiviral immunity. Conversely, therapeutic targeting of TFIIB may represent a mechanism to limit pathological side effects caused by overshooting immune reactions.
| Transcription of DNA by RNA polymerase II (Pol II) is central to gene expression and subject to regulation at multiple levels. A large number of dedicated protein complexes, as well as chromatin remodelling factors, are essential for transcription and regulate distinct phases of the transcriptional process [1]. The first step includes subunits of Pol II and general transcription factors (GTFs), which sequentially assemble into a preinitiation complex (PIC) that recognizes promoter regions on DNA and responds to regulatory signals in order to start mRNA synthesis [2]. GTFs include Transcription Factor (TF) IIA, TFIIB, TFIID, TFIIE, TFIIF, and TFIIH [3]. TFIID nucleates PIC assembly by binding promoter sequences through its TATA binding protein (TBP subunit) [4] and recruits TFIIB, which mediates the association of Pol II with the promoter [5]. Subsequently, TFIIE and TFIIH unwind DNA at the promoter for the initiation of transcription [6]. Pol II is released from the promoter to enable downstream transcription, while most of the GTFs dissociate from the promoter [7].
The recruitment of Pol II to the promoter is considered to be the major controlling step in gene expression [8]. However, an increasing number of genes are discovered to have higher accumulation of Pol II on their promoters without the corresponding enrichment within the gene bodies [9–12]. Such rate-limiting step, known as pausing, offers an additional regulatory switch in metazoans [13–17]. Pausing occurs when Pol II within the early elongation complex is prevented from further elongation through binding of Negative elongation factor (NELF) and DRB-sensitivity-inducing factor (DSIF) [18–20]. Release of paused Pol II is facilitated by positive transcription elongation factor (P-TEFb) enabling productive elongation and mRNA transcription [16,18,21].
Regulated gene expression is crucial for most biological processes. Particularly tight control of gene expression is required for controlling the immune system. Expression of immune-related genes is strictly regulated under non-perturbed conditions, while invasion of pathogens, such as viruses and bacteria, or toxins, induce a plethora of pro-inflammatory cytokines and expression of anti-viral/bacterial effector proteins. Virus-induced expression of the human interferon (IFN)-β gene is one of the best characterized examples for inducible gene expression in higher eukaryotes. IFN-β is not expressed under steady state conditions but is synthesized in response to virus infection. This process relies on the sensing of virus infection or replication products by dedicated pattern recognition receptors (PRRs), which initiate a signalling cascade that leads to local disassembly of nucleosomes and activation of transcription factors including interferon regulatory factors (IRF) 3 and -7 [22,23], which are master regulators of type-I IFNs (IFN-α/β). Secreted IFN-α/β bind type-I IFN receptor [24], which trigger signal transducer and activator of transcription (STAT)-1 and -2 [25], which lead to the transcription of several hundred genes encoding proteins with antiviral properties or regulatory functions mounting an efficient antiviral barrier.
Most viruses evolved mechanisms to target vulnerable points of the antiviral defence system. These mechanisms often include viral perturbations of transcriptional processes. However, viral strategies vary widely: while some large DNA viruses such as pox- and herpesviruses express a number of open reading frames (ORFs) acting in parallel to impair sensing, signalling and activation of transcription factors [26–31], small RNA viruses with limited genetic coding capacity preferentially perturb the antiviral system at critical hubs. The Npro protein encoded by the positive strand RNA virus Bovine viral diarrhoea virus (BVDV), for instance, specifically leads to proteasomal degradation of IRF3 and thus alleviates induction of IFN-α/β [32,33]. Rift Valley fever virus (RVFV), a negative strand RNA virus belonging to the virus family of Bunyaviridae, degrades polymerase II subunit TFIIH through its Non-structural protein Small (NSs) [34,35] resulting in general inhibition of transcription, followed by broad inhibition of protein expression and detrimental effects for cell viability. An orthomyxovirus Thogoto virus (THOV) also targets the polymerase complex through association of its Matrix protein long (ML) with TFIIB [36]. However, in contrast to RVFV NSs, THOV ML expression appears not to lead to general inhibition of gene expression and plasmid-driven ML expression does not show an apparent toxicity effect. Instead, it is believed that ML specifically impairs expression of type-I interferon genes through engaging IRF3 [37–40].
However, several lines of evidence suggest an activity of ML exceeding simple inhibition of IRF3. These include inability of THOV ML to inhibit IRF3 nuclear translocation [38]. Unbiased evaluation of THOV ML interacting proteins by affinity purification coupled with mass spectrometry (AP-MS) as well as in vitro binding assays (data shown here and [40]) did not support direct interaction of THOV ML with IRF3. Furthermore, ML has also been proposed to impair expression of NF-kappa B target genes [40], which can be induced independently of IRF3 [41]. We therefore asked how ML can display an exceptionally specific response, by targeting a very central transcription factor. Moreover, we expected to expand our knowledge on the functionality of TFIIB in higher eukaryotes.
To identify proteins specifically involved in IFN-α/β inhibition by Thogoto virus (THOV), we used affinity purification followed by liquid chromatography and tandem mass spectrometry (AP-LC-MS/MS). To this end, we expressed HA-tagged viral M and ML proteins in HEK293 cells and performed affinity purification followed by LC-MS/MS analysis. In line with the large shared regions between both proteins, the majority of identified proteins were equally well enriched in M and ML precipitates (Fig 1A and 1B, S1A Fig, S1 Table). However, we identified two interactors that were highly significantly enriched in ML samples: the general transcription factor IIB (TFIIB) and the largely uncharacterized CAPN15, a member of the calpain family of proteases (Figs 1A and S1A).
We validated ML-specific binding of GST-tagged M and ML to FLAG-tagged CAPN15 and TFIIB (Fig 1C). In order to refine binding requirements we also used C-terminal point mutants of ML proteins bearing alanine replacements at positions S283A/W284A (SW) and T270A/E271A (TE) (Fig 1B). As published previously[40], ML(SW) lost the ability to interact with TFIIB, whereas ML(TE) bound to TFIIB similarly well to the wild type protein. Conversely, we found that CAPN15 only inefficiently precipitated with exogenously expressed ML(TE) but bound TFIIB comparably well to the wt ML protein (Fig 1B and 1C). To validate these findings in an infection context, we generated recombinant THOV that expressed corresponding mutants of ML. As expected, THOV expressing ML(SW) lost the ability to interact with TFIIB while it retained binding to CAPN15 and virus expressing ML(TE) lost binding to CAPN15 while it bound to TFIIB (S1B Fig). Infection of HeLa cells with recombinant THOV wild type (rTHOV-wt) as well as with the virus bearing TE mutation in ML (rTHOV-TE), which is still capable to bind TFIIB but cannot bind CAPN15, induced only minimal amounts of IFN-α/β (Fig 1D). In contrast, ML deletion mutant rTHOV-ΔML and SW ML (rTHOV-SW), which lost binding to TFIIB but retained CAPN15 affinity, potently stimulated IFN-α/β production (Fig 1D). From these experiments we concluded that the ability of ML to interact with TFIIB is necessary and sufficient to interfere with IFN-α/β induction and that binding to CAPN15 does not affect induction of type-I interferons.
To further understand how ML interferes with the function of TFIIB, we tested distribution of TFIIB on a subcellular level by confocal imaging. In uninfected cells TFIIB showed nuclear localization (Fig 2A), consistent with its function in RNA transcription [42]. Surprisingly, rTHOV-wt infected cells showed a dramatic re-localization of TFIIB from the nucleus into the cytoplasm (Fig 2A). This re-localization was specific to the ability of ML to bind TFIIB, since infection with recombinant THOVs that expressed ML variants with impaired TFIIB binding, such as rTHOV-ΔML and rTHOV-SW, resulted in nuclear TFIIB localization (Fig 2A). To test whether the observed effect was indeed dependent on ML activity, we transiently transfected HA-tagged ML and analyzed subcellular localization of TFIIB. In line with infection experiments, transfection of ML, but not M, led to accumulation of TFIIB in the cytoplasm already at 16 h.p.t. (Fig 2B), where it also remained at 24 h.p.t. (S2A Fig). Additionally, we could confirm TFIIB re-localization by subcellular fractionation of ML transfected cells. In cells expressing M, endogenous (Fig 2C), as well as overexpressed (Fig 2D) TFIIB localized to the nucleus, whereas in cells that received ML TFIIB was almost exclusively detectable in the cytoplasm (Figs 2C and 2D). Of note, ML did not generally affect nuclear/cytoplasmic distribution since histone H3 still localized to the nucleus (Figs 2C and 2D). Altogether, these data suggested that ML interferes with TFIIB function by altering its subcellular distribution and therefore would not allow TFIIB to contribute to transcriptional processes.
It is known that other viruses similarly interact with components of transcription machinery. RVFV NSs protein, for instance, sequesters the p44 subunit of TFIIH [43] and recruits an E3 ligase complex to the p62 subunit leading to its proteasomal degradation [35] resulting in destabilization of TFIIH and general shutdown of mRNA transcription. To test whether ML impairs the stability of TFIIB, we employed a pulse-SILAC LC-MS/MS approach that allowed us to assess proteome-wide protein abundance, stability and translation rates of 5416 proteins (S2B Fig, S2 Table). Compared to mock, infection with both rTHOV-wt and rTHOV-ΔML slightly increased total and newly synthesized TFIIB levels (Fig 2E). However, there was no difference between rTHOV-wt and rTHOV-ΔML (Fig 2E). The expression and stability of other PIC components were not affected by infection with rTHOV-wt either (S2C Fig). Moreover, although the total levels of protein translation were reduced in virus infected cells, no difference between rTHOV-wt or rTHOV-ΔML infection could be seen (Fig 2E). Collectively, these data suggested that ML binds to TFIIB but this interaction does not lead to generally reduced abundance of TFIIB or PIC proteins and does not affect general translation rates in THOV infected cells.
Based on the ability of ML to impair cellular localization of a central component of the PIC, we examined whether ML has a broader effect on transcription than previously anticipated. To this aim, we performed pulse-labelling experiments using radioactively labelled [3H]5-Uridine (Fig 3A), which is incorporated into newly synthesized RNA and therefore allows to assess global transcription rates in a time-dependent manner. In agreement with the levels of translation, total RNA synthesis was similar in rTHOV-wt and rTHOV-ΔML infected cells (Fig 3A).
To test the effect of ML in an unbiased manner, we performed transcriptome analysis of HeLa cells that were left uninfected or infected with rTHOV-wt, rTHOV-ΔML or rTHOV-SW (S3A Fig, S3 Table, GEO GSE105152). The RNA levels of viral NP transcripts were comparable in virus infected cells and not detected in mock (Fig 3B). As expected, infection with rTHOV-ΔML and rTHOV-SW elicited expression of IFIT1 and IFIT3, whereas their expression in rTHOV-wt sample was ~50 fold lower (Fig 3B). Interestingly, expression of housekeeping genes, such as GAPDH, appeared quite stable independently of the virus used (Fig 3B), validating the lack of general inhibition on the transcriptional level. Whole transcriptome analysis using next generation sequencing of cells infected with rTHOV-wt revealed only 63 significantly changed genes (q<0.05; fold change ≥2) (Fig 3C, x-axis, green and orange dots, S3 Table), which is in line with our above findings that targeting of TFIIB by ML does not significantly impact general transcription rates in a negative or positive manner. In contrast, in cells that were infected with rTHOV-ΔML we identified 518 differentially regulated genes compared to mock (q<0.05; fold change ≥2) (Fig 3C, y-axis, blue and orange dots, S3 Table). rTHOV-SW elicited a comparable response to rTHOV-ΔML infection, underlining that changes occurred in a TFIIB-dependent manner (S3B Fig, S3 Table). General properties of genes significantly upregulated by rTHOV-ΔML and rTHOV-SW (S3B Fig) were related to innate antiviral response as determined by GO term over-representation analysis (Fig 3D). Additionally, we screened promoter regions of co-regulated genes for putative transcription factor binding sites [44] to identify gene populations with diverse nature of regulation. This analysis suggested that compared to ML mutant viruses, the wt virus blunted expression of genes that were under control of diverse transcription factors. Most affected were IRFs and STATs, but also RELA, SPI1, and FOS (S3C Fig, S4 Table). In sum, these data suggested that general transcription rates and transcription of constitutively active genes were not affected by THOV infection and that the wild-type virus inhibited a broad array of inducible inflammation-related and -unrelated genes driven by diverse transcription factors.
Transcriptome profiling suggested that cells infected with rTHOV-wt show minimal changes in gene expression compared to mock infected cells. This prompted us to test, whether ML may selectively inhibit any dynamic changes in gene expression without affecting ongoing transcriptional processes. To test this and to delineate whether ML is not just inhibiting upstream processes resulting in broad spectrum of changes, we employed reporter constructs that allow to directly measure the transcriptional activity of selected promoters after stimulation with defined activating ligands. We used type-I IFN driven luciferase reporter constructs such as promoters for interferon stimulated response element (ISRE), the ISG54 promoter or Mx1, all of which contain STAT1 binding sites and thus are directly responsive to IFN-α/β treatment. As expected, co-expression of ML impaired their activity, whereas M or ML(SW) did not show this effect (Figs 4A and S4A).
The inhibitory effect of ML, but not M or ML(SW) could also be seen for other inducible promoters containing NFκB or NFAT sites (Figs 4B and S4B). However, in agreement with minimal influence on constitutive gene expression, the activity of EF1-α-promoter-driven Renilla luciferase was not affected by co-transfection of M, ML or ML(SW) (Figs 4C and S4C). Surprisingly, IRF1 promoter, even though showing a dynamic change after IFN-γ stimulation, was not affected by ML overexpression (Figs 4C and S4C). Similarly, HSP70 promoter (Figs 4C and S4C) and EGF targets cyclin D1 and DUSP6 were induced equally well despite ML presence (Figs 4D and S4D). From these experiments we concluded that ML represses dynamic changes in gene expression, but this inhibitory effect is limited to the activity and regulation of selected promoters.
Since ML affects gene expression by interacting with TFIIB, we hypothesized that TFIIB may be required for expression of selected genes and that depletion of TFIIB would mimic the effect of ML. To test this, we performed siRNA-mediated knockdown of TFIIB and tested whether this influenced gene expression from inducible promoters. Transient depletion of TFIIB did not affect cell viability within the timeframe of this experiment (Fig 5A), which is in line with a recent report[45].
In order to gain insights into genome-wide distribution of TFIIB-dependent genes, we performed RNA-seq analysis after TFIIB depletion followed by stimulation with TNF-α (S5A Fig). This experimental setup allowed us to compare the impact of TFIIB depletion on general gene expression as well as the cellular response to a defined stimulus (TNF-α treatment) (Fig 5B, S5 Table, GEO GSE105153). In control siRNA-treated cells, TNF-α treatment induced expression of 125 genes (Fig 5B, x-axis, green dots). Compared to control knockdown, TFIIB depletion in combination with TNF-α treatment led to a minor difference in the overall gene expression pattern–from 11348 mapped genes only 148 genes showed decreased and 138 genes increased expression (Fig 5B, y-axis, blue dots). Notably, TFIIB depletion resulted in two populations of TNF-α responsive genes: 47 TNF-induced genes were inhibited more than 2-fold in expression by TFIIB depletion (Fig 5B, red squares), while 60 genes were induced similarly in TFIIB depleted and control cells (Fig 5B, black squares). We validated this effect by qPCR for two exemplary targets: IFN-β expression was severely affected by TFIIB depletion while induction of the control gene Ephrin-A1 (EFNA1) was comparable in control and TFIIB depleted cells (S5B Fig). We performed qRT-PCR on independent samples applying three distinct stimuli, TNF-α, IFN-α and EGF, all of which induce a specific gene expression profile (S6 Table). As expected TNF-α led to two distinct populations of genes: genes that showed ≥1.5-log2 fold decreased expression in the absence of TFIIB (Fig 5C, red symbols) and genes, which were equally induced despite TFIIB depletion (Fig 5C, black symbols). In agreement with RNA-seq analysis, TFIIB was required for the activation of inflammatory cytokines IL-6, IL-8, CXCL10 and IFITs (Fig 5C, red symbols), whereas induction of EFNA1, IKBKE, IRF1, A20 and IκBα was not significantly altered (Fig 5C, black symbols). TFIIB was also required for inducible expression of some IFN-α/β stimulated targets: IFN-α/β dependent induction of OAS2, GBP4, CXCL10, CXCL11, ISG15, Mx1 and IFITs was significantly impaired in the absence of TFIIB (Fig 5D, red symbols). However, IFN-α stimulation of other genes such as A20, IRF1, IκBα was not affected by TFIIB depletion (Fig 5D, black symbols). In line with minimal effects of ML on EGF targets (Fig 4D), direct TFIIB knockdown did not significantly affect basal or induced expression of any of the EGF responsive gene tested, besides IL-8 (Fig 5E). Similarly, TFIIB depletion after infection with rTHOV-ΔML (mixed stimulus) led to a decreased IFIT3 expression, but not STAT1, p65 or p38 (S5C Fig). Overall, despite different ligands that had been used for stimulation, components of signalling pathways, were induced comparably well in siTFIIB and siScrambled treated cells. In contrast, expression of inflammatory cytokines and genes encoding antiviral effector proteins was severely impaired (Fig 5C–5E).
In general, two different modes of transcriptional regulation exist: (i) gene expression can be regulated through pausing of already recruited Pol II at the promoter when stimulus-dependent activation releases this pausing activity to drive expression of genes (S6A Fig). Prototypic genes falling in this category of gene regulation are targets of the EGF signalling cascade [8,46,47]. (ii) Another mechanism to regulate gene expression is de novo recruitment of Pol II to transcriptional start sites (S6A Fig). This mode of regulation is best described for a subset, but not all genes associated with inflammatory processes [48]. The fact that EGF targets were not regulated by TFIIB depletion (Figs 4D and 5E) could be attributed to their regulation by polymerase pausing [46], which potentially does not require TFIIB. The diverse behaviour of signalling components and inflammatory cytokines in respect to TFIIB knockdown (Fig 5B–5E) further suggested de novo Pol II recruitment as the main discriminating feature of the affected genes. To corroborate this observation, we performed analysis of publically available chromatin-immunoprecipitation followed by deep sequencing (ChIP-Seq) datasets and calculated the occupancies of Pol II, TFIIB, NELF and DSIF [49,50] at the promoter and gene body regions (Fig 6, S7 Table).
These datasets allowed us to define genes that are actively transcribed (high Pol II in the gene body), paused (high Pol II at the promoter and low in the gene body, high NELF and DSIF at the promoter) or not occupied by polymerase or pausing-associated factors and therefore require de novo recruitment of the pre-initiation complex (low Pol II at the promoter and in the gene body, low NELF and DSIF at the promoter) (S6 Fig). The status of TFIIB during elongation and pausing is not well characterized, thus we speculated that TFIIB levels at the promoters would be reflective of those of Pol II. We then mapped identified occupancies to the genes, found to be influenced by TFIIB depletion in RNA-seq experiments (Figs 5B and 6A red profiles) and the qPCR screen (Figs 5C–5E and 6A red dashed lines), and those genes observed to be non-affected by TFIIB depletion (Figs 5B–5E and 6A black profiles). Remarkably, genes affected by TFIIB depletion showed significantly lower occupancies of Pol II, and TFIIB in the promoter region, as compared to non-affected genes (Fig 6B). In addition, promoters of genes that were not affected by TFIIB depletion, showed high occupancies of NELF and DSIF, which are markers for paused promoters. Indeed, calculating pausing indices (promoter and gene body occupancy ratio > 2) revealed significantly higher pausing rates in non-affected vs affected genes (Fig 6C), demonstrating that non-affected genes are regulated by promoter-proximal pausing and supporting our hypothesis of de novo recruitment as mechanistic requirement for regulation of genes affected by TFIIB depletion. In sum, these data indicate that TFIIB is needed for induced expression of genes that require de novo polymerase II recruitment and that TFIIB absence is better tolerated by genes regulated through polymerase pausing. Furthermore, these data are consistent with the inhibitory activity of ML on reporter constructs and explain the broad, yet relatively selective activity of wild-type THOV.
Selective regulation of gene expression would allow for the possibility to interfere with de novo transcription, which would be of particular interest to modulate inflammatory processes associated with pathogenicity and disease. Compared to M, ML has a 38 amino acids (aa) C-terminal extension (Fig 7A) and we hypothesized that expression of this fragment may be sufficient to regulate TFIIB activity.
We generated GFP-fusion constructs containing C-terminal fragments of ML (Fig 7A) and tested their ability to directly associate with TFIIB in co-precipitation experiments with FLAG-TFIIB (Figs 7B and S7A). The 38 aa bearing GFP-ML (266c) protein failed to bind TFIIB, while GFP-ML (257c; C-terminal 47 aa), GFP-ML (247c; C-terminal 57 aa) and GFP-ML (247–294) successfully precipitated with TFIIB (Figs 7B and S7A). This suggested that binding of TFIIB requires the C-terminal “L” portion of ML as well as additional amino acids in the C-terminus of M. To assess their functionality, we co-transfected the same fragments with ISRE reporter into HEK293 cells (Fig 7B bottom panel). After stimulation with IFN-α, only the 57 aa ML fragment (GFP-ML (247c) was able to inhibit ISRE promoter activation, but not EF1-α promoter (Figs 7B and S7B). ML (257c) and ML (247–294) were able to interact with TFIIB, but failed to inhibit ISRE activation, indicating that the C-terminal 47 to 57 aa are required for the functional inhibition of TFIIB. To further narrow down the active region, we generated 2 aa truncation mutants, fused to mCherry (Fig 7A) and tested their ability to bind TFIIB and to block the activation of ISRE reporter upon IFN-α stimulation (Fig 7C). Only ML fragments (247c) and (249c) were able to inhibit ISRE induction, but not EF1-α promoter activity (Fig 7C bottom panel, S7C Fig). We concluded that the inhibitory domain of ML is thus located to aa 249–304 and consists of the L-region and a minimal part of M (miniM)–thus here called the miniML domain. Mimicking THOV and using its miniML inhibitory domain to target a single transcription factor TFIIB, one could modulate a wide variety of pro-inflammatory cytokines and effector molecules requiring de novo recruitment of the polymerase complex without affecting general gene expression and genes required for cellular homeostasis, growth, maintenance of protein stability, energy metabolism and signalling in particular (Fig 8).
The innate immune system is comprised of a functional network between cytokines and antiviral effector molecules and constitutes an essential antiviral defence barrier. Here, we identified a mechanism that involves viral targeting of a single general transcription factor, TFIIB, and allows perturbance of the innate immune system at a broad scale with minimal influence on other ongoing transcriptional processes. Through alternative splicing, THOV expresses the immunoregulatory Matrix protein variant ML, that bears a 38 amino acid C-terminal extension as compared to M. Unbiased AP-MS analysis shows that ML binds two proteins, TFIIB and CAPN15. However, the inhibitory properties of ML rely solely on the ability to bind TFIIB, while CAPN15 appears to be dispensable for immunoregulation on transcriptional level. In contrast to other viruses that degrade their target proteins [34,35], THOV does not lead to destabilisation of TFIIB or any other protein of the PIC. Rather, ML generates a TFIIB depleted environment in the nuclei of infected cells by re-localizing TFIIB into the cytoplasm. In this way, TFIIB would not be available for transcriptional purposes and won’t be able to undergo regulatory post-translational modifications, such as phosphorylation on Ser65 [51], required for transcription initiation. The ability to translocate TFIIB, may be explained by the fusion of the TFIIB-interaction region (miniML) to the matrix protein M. Orthomyxoviruses replicate and partially assemble in the nucleus before the translocation process into the cytoplasm, in case of THOV initiated by the accumulation of M in the nucleus. Given the limited coding capacity of THOV, fusion of miniML to M may provide the virus with a possibility to shuttle ML between the nucleus and the cytoplasm along with its target TFIIB. In addition, the miniML on its own is hydrophobic, resulting in limited solubility and stability. From a virus perspective, fusion to the M protein may therefore be beneficial to increase functionality of miniML.
While TFIIB is a general transcription factor and considered to be essential for the transcription of all known genes [5], wt THOV infection did not affect transcription of the majority of genes but showed prominent inhibition of IFN and ISG expression. This effect had been previously attributed to specific inhibition of the transcription factor IRF3 [38]. However, transcriptome-wide assessment of gene expression in infected cells, as well as reporter assays studying gene regulation at the promoter level, clearly show that ML has activity on inducible gene expression that goes beyond IRF3 inhibition. Intriguingly, in contrast to inflammation-related promoters, the activity of IRF1 and HSP70 promoters as well as expression of EGF target genes were not affected by ML presence. Remarkably, a large proportion of genes responsive to EGF are regulated by polymerase pausing, i.e. show deposited Pol II at their transcription initiation sites prior to their activation [46,52,53], a property that has also been shown on reporter constructs coding for isolated promoters [17]. Interestingly, IRF3 targets have been reported to mostly require de novo recruitment of Pol II, while RELA targets can be regulated by both, polymerase pausing or de novo recruitment of Pol II [48]. Our experiments suggest that promoters, which are known to be regulated by polymerase pausing are not affected by ML, while promoters that rely on de novo recruitment of polymerase to the transcriptional start site are sensitive to ML presence. The apparent specificity of IFN inhibition in the context of virus infection, is not mediated by ML itself but results from the nature of the given experiment: virus infection activates an IRF3 dependent type-I interferon response that requires de novo recruitment of Pol II to transcriptional start sites. ML inhibits this recruitment, resulting in an apparent block of interferon gene induction. However, combining other stimuli with ML expression would result in a similar specific pattern, given the recruitment of Pol II is required for expression of the induced genes.
These data indicate that ML is inhibiting de novo recruitment of polymerase to transcriptional start sites and that TFIIB plays a predominant role in this process, while it is dispensable to regulate expression from pre-loaded promoters, at least after Pol II has been associated. Experimental depletion of TFIIB phenocopied ML activity. Stimulation experiments using IFN-α, TNF-α or EGF showed that TFIIB depletion allowed to discriminate between two separate populations of response genes: some genes were induced normally despite TFIIB knockdown (non-affected) while other genes were severely impaired in expression in the absence of TFIIB (affected). This behaviour could be explained by Pol II and TFIIB occupancies at their promoters: under steady-state conditions non-affected genes had prominent Pol II and TFIIB levels loaded in the promoter region (paused polymerase). The affected group, however, displayed significantly lower Pol II and TFIIB loading, and their expression would therefore require de novo recruitment of the transcriptional apparatus. It has been reported that TFIIB is dispensable for the activation of some genes [45,54] and that TFIIB might be regulating only specialized versus housekeeping genes during cardiac hypertrophy [55]. It seems surprising that TFIIB depletion in mammalian cells shows no effect on cell viability and general gene or protein expression. However, the strict requirement of TFIIB has mostly been established through the excellent work studying de novo recruitment of Pol II using in vitro reconstitution systems employing recombinant proteins or in lower eukaryotes (Saccharomyces cerevisiae), which do not use polymerase pausing as gene regulation mechanism [56–60] and may thus be more dependent on TFIIB for general transcriptional processes. Regulation of TFIIB levels in higher eukaryotes may allow an additional layer of gene expression control, for instance, during development [61], injury [62], and cancer [63], which is in line with the high mobility and short chromatin association time of TFIIB as compared to Pol II [64–68]. Transient association of TFIIB with chromatin [68] may be sufficient to place Pol II at the transcriptional start site. Even if TFIIB is dissociated, depleted or sequestered by viruses, genes that bear loaded Pol II on their promoters can still be transcribed. In contrast, genes that are not loaded with Pol II require de novo recruitment and are therefore dependent on TFIIB activity.
Among genes that are highly dependent on de novo recruitment of Pol II are many inflammatory cytokines and antiviral effector molecules. In contrast, housekeeping genes and signalling components commonly show paused promoters, which gives them an advantage to bypass complete assembly of the transcriptional apparatus and allows them to react instantly to incoming stimuli. This differential mode of transcriptional regulation appears to be used for expression of immune-related genes: while many cytokines and chemokines mounting a pro-inflammatory environment require de novo recruitment of Pol II, negative regulators commonly show paused promoters [69]. Such regulation may allow the immune system to keep at bay undesired inflammation and have an easy access to negative regulators. Indeed, depletion of TFIIB or delivery of ML impairs expression of essential pro-inflammatory molecules (CXCL10, IL-6, IL-8, IFITs etc.), while signalling components (IRFs, IκBα, SMAD3, BCL6 etc.) and negative regulators of inflammation (A20, TGF-β) or genes required for tissue repair (ITGA1, ITGA2, PLAUR, EPHA2 etc.) are unaffected. Intriguingly, while genes requiring de novo Pol II recruitment show different Pol II occupancy depending on the cell type [70,71], paused genes generally have lower cell-to-cell expression variability [72].
Given its regulatory properties, targeting of TFIIB by viruses is very beneficial for the successful infection: while expression of pro-inflammatory cytokines that are required to mount an antiviral response are inhibited, negative regulators which mediate a tolerogenic environment as well as tissue repair factors are expressed (Fig 8). Importantly, modulation of TFIIB activity provides an exciting opportunity to therapeutically control selected gene expression without massively affecting global transcription. Application of miniML, an active sequence of the ML-TFIIB interacting domain, mimics the viral activity and therefore allows to modulate gene expression depending on the context. Employment of miniML may allow therapeutic taming of overshooting immune responses and could therefore be beneficial to treat diverse immunopathologies, such as septic shock, specific pathogen-induced immunopathologies, autoimmunity as well as antiviral immunity. Furthermore, gene expression of most dsDNA viruses [45,73–75] requires the host apparatus including de novo recruitment of the polymerase complex, thus, interference with TFIIB activity may allow to modulate their viral gene expression and viral spread.
Recombinant human interferon-α (IFN-α) was a kind gift from Peter Stäheli, recombinant human TNF-α and IFN-γ were purchased from PeproTech, recombinant human EGF was a kind gift from Kirti Sharma, PMA from Felix Meissner, NiCl2 was purchased from Sigma-Aldrich. HEK293T transgenic for the Mx1-promoter driven firefly luciferase gene were described previously [76], HeLa FlpIn were a kind gift from Andrea Musacchio, HeLa S3 (CCL-2.2) and Vero E6 (CL-1586) were purchased from ATCC, HEK293 were a kind gift from Andrew Bowie, HeLa Kyoto expressing GFP-tagged TFIIB from BAC transgene were from Ina Poser [77]. All cell lines were maintained in DMEM (GE Healthcare Life Sciences) containing 10% foetal calf serum (GE Healthcare Life Sciences) and antibiotics (100 U/ml penicillin, 100 μg/ml streptomycin). Duplex siRNAs targeting human TFIIB (siGENOME SMARTpool) or Scrambled control were from Dharmacon (GE Healthcare Life Sciences). Recombinant THOVs expressing ML (rTHOV-wt), lacking ML (rTHOV-ΔML) or bearing mutated ML proteins (rTHOV-SW and rTHOV-TE) were described previously [40,78,79]. Virus infections were performed in reduced medium volume for 1 h at 37°C with subsequent exchange of medium and incubation at 37°C for desired periods of time.
Thiazolyl blue tetrazolium bromide (MTT) was from Sigma-Aldrich. Primary antibodies used in this study were the following: GST (Cell Signaling Technology 2624), FLAG M2 (Sigma-Aldrich F3165), β-actin-HRP (Santa Cruz sc-47778), THOV NP and THOV M/ML were described previously [80], HA (Cell Signaling Technology 3724), TFIIB (Cell Signaling Technology 4169), β-tubulin (Sigma-Aldrich), histone H3 (Abcam), GFP (Invitrogen A6455), mCherry (Rockland 600-401-P16), GFP-DyLight-488 (Rockland 600-141-215). Secondary antibodies detecting mouse and rabbit IgG were from Jackson ImmunoResearch and Dako. DAPI and secondary antibodies for immunofluorescence were purchased from Invitrogen.
pCAGGS expression plasmids for HA- and GST-tagged M and ML and FLAG-tagged TFIIB were described previously [40,78]. pCAGGS-eGFP-FLAG was a kind gift from Urs Schneider, pCAGGS-FLAG-CAPN15 was generated by inserting human CAPN15 into pCAGGS. GFP, GFP-ML(266c), GFP-ML(257c), GFP-ML(247c), GFP-ML(247–294) were generated by inserting GFP and ML fragments into pcDNA3. ML fragments 247c to 257c were generated by site-directed mutagenesis and inserted into pmCherry. pGEX-GST-TFIIB was described previously [81]. Primers can be provided upon request. The following reporter constructs were used in this study: pISRE-luc was purchased from Stratagene, pGL3-Mx1-ff-luc was described previously [82], NFkB-luc was a kind gift from Andrew Bowie, NFAT-luc and ISG54-luc from Tilmann Bürckstümmer, EF1-α-ren from Engin Gürlevik, pIRF1-GAS-ff-luc was described previously [40], HSP70-luc was a kind gift from Mark Hipp.
Total amounts of IFN-α/β in cell supernatants were measured by using 293T cells stably expressing the firefly luciferase gene under the control of the mouse Mx1 promoter (Mx1-luc reporter cells) [76]. Briefly, cell supernatants were harvested and virus particles were removed with Amicon spin columns with a cutoff of 100 kDa (Millipore) according to the manufacturer's instructions. Mx1-luc reporter cells were seeded into 96-well plates in sextuplicates and were treated 24 hours later with filtered supernatants diluted 1:10 in DMEM-5% FCS. At 16 hours post incubation, cells were lysed in passive lysis buffer (Promega), and luminescence was measured with a microplate reader (Tecan). The assay sensitivity was determined by a standard curve.
For reporter assays, HEK293 cells were plated in 96-well plates 24 hours prior to transfection. Firefly reporter and Renilla transfection control were transfected using polyethylenimine (PEI, Polysciences) in sextuplicates for untreated and treated conditions. In 24 hours cells were stimulated for 16 hours with corresponding inducer and harvested in passive lysis buffer (Promega). Luminescence of Firefly and Renilla luciferases was measured using dual-luciferase-reporter assay (Promega) according to the manufacturer’s instructions in a microplate reader (Tecan).
For immunofluorescence analysis, HeLa Kyoto cells stably expressing GFP-TFIIB were grown on coverslips, transfected with M/ML proteins or infected with rTHOV, and fixed with 4% /w/v) paraformaldehyde (PFA) for 15 min at room temperature, blocked in blocking buffer (1xPBS containing 0.1% foetal calf serum (w/v) and 0.1% Triton X-100 (v/v)) for 1 hour at room temperature. Stainings were performed for 1 hour at room temperature in blocking buffer. Confocal imaging was performed using an LSM780 confocal laser scanning microscope (ZEISS) equipped with a Plan-APO 63x/NA1.46 immersion oil objective (ZEISS).
For cytoplasmic-nuclear fractionation, HEK293T cells were seeded in 9 cm dishes, transfected with HA-tagged M/ML and FLAG-TFIIB proteins. After 24 hours the cells were washed and fractionated in sucrose-NP-40 buffer supplemented with protease inhibitors (10 mM HEPES pH 7.9, 0.34 M sucrose, 3 mM CaCl2, 3 mM MgAc, 0.1 mM EDTA, 0.5% NP-40). The lysates were incubated 10 min on ice and centrifuged at 3500 rpm for 5 min. Cytoplasm fraction (supernatant) was transferred into a new tube and mixed with 4x-Laemmli buffer. Nuclear fraction (pellet) was washed in sucrose buffer without NP-40 and resuspended in sucrose buffer with benzonase, incubated for 10 min on ice and centrifuged at 3500 rpm for 5 min. Pellets were resuspended in 1x Laemmli buffer. Both fractions were loaded on 12% SDS gel, proteins of interest were detected by western blotting.
GST and GST-TFIIB were purified from BL21 and bound to GST-agarose beads for 2 h at 4°C. Expression and purity was controlled by SDS-PAGE and Coomassie staining. GFP-ML fragments were in vitro transcribed and translated (IVT) using TNT quick coupled transcription/translation kit (Promega) and radioactively labelled with [35S]-Methionine/Cysteine. GST-fusion proteins were incubated with IVT fragments for 2 h at 4°C, washed, separated on 12% SDS-gel and detected by autoradiography (Kodak BiomaxMR).
For affinity purification, cell lysates were prepared by lysing HEK293 cells expressing HA- or GST-tagged M/ML proteins and FLAG-tagged CAPN15 or TFIIB for 30 min on ice in TAP lysis buffer (50 mM Tris pH 7.5, 100 mM NaCl, 5% (v/v) glycerol, 0.2% (v/v) Nonidet-P40, 1.5 mM MgCl2 and protease inhibitor cocktail (EDTA-free, cOmplete; Roche)). For affinity-purification with HA-tagged proteins, HA affinity resin (Sigma-Aldrich) was incubated with cell lysate in TAP lysis buffer for 60 min at 4°C on a rotary wheel. For affinity-purification with Flag-M2-tagged proteins, Flag-M2 affinity resin (Sigma-Aldrich) was incubated with cell lysate and processed as above. Beads were washed three times with TAP lysis buffer, followed by two times with TAP wash buffer [lacking 0.2% (v/v) Nonidet-P40], boiled in 2x Cell Signaling SDS buffer for 5 min at 95°C and subjected to SDS-PAGE and Western Blot analysis.
For quantitative purification of ML-binding proteins, HA affinity resin (Sigma-Aldrich) was incubated with lysates of HEK293 cells expressing HA-tagged M or ML proteins and processed as above. Four independent affinity purifications were performed for each protein. Bound proteins were denatured by incubation in 6 M urea-2 M thiourea with 1 mM DTT (Sigma-Aldrich) for 30 min and alkylated with 5.5 mM iodoacetamide (Sigma-Aldrich) for 20 min. After digestion with 1 μg LysC (WAKO Chemicals USA) at room temperature for 4 h, the suspension was diluted in 50 mM ammonium bicarbonate buffer (pH 8). Protein solution was digested with trypsin (Promega) overnight at room temperature. Peptides were purified on stage tips with three C18 Empore filter discs (3M) and analysed by mass spectrometry as described previously [83]. Briefly, peptides were eluted from stage tips and separated on a C18 reversed-phase column (Reprosil-Pur 120 C18-AQ, 3 μM, 150×0.075 mm; Dr. Maisch) by applying a 5% to 30% acetonitrile gradient in 0.5% acetic acid at a flow rate of 250 nl/min over a period of 95 min, using an EASY-nanoLC system (Proxeon Biosystems). The nanoLC system was directly coupled to the electrospray ion source of an LTQ-Orbitrap XL mass spectrometer (Thermo Fisher Scientific) operated in a data dependent mode with a full scan in the Orbitrap cell at a resolution of 60,000 with concomitant isolation and fragmentation of the ten most abundant ions in the linear ion trap.
HeLa cells were grown in normal medium containing light (L) amino acids and infected with rTHOV viruses for 1 h. After 18 hours the cells were incubated in starvation medium (lacking Lys and Arg) for 30 min. Subsequently, SILAC medium containing heavy (H) labelled amino acids (Lys8, Arg10) was added. 6 hours later total protein lysates were prepared and subjected to LC-MS/MS analysis. Briefly, lysates were prepared in SDS lysis buffer (50 mM Tris pH 7.5, 4% sodium dodecyl sulfate), boiled for 5 min at 95°C, sonicated for 15 min with a Bioruptor (Diagenode) and centrifuged for 5 min at 16,000× g at room temperature. Protein concentration was determined by Lowry assay (DC Protein Assay, BioRAD), and 50-μg aliquots were reduced with 10 mM DTT for 30 min, alkylated with 55 mM IAA for 20 min at room temperature, and precipitated with 80% acetone for 3 h at 20°C. After centrifugation for 15 min at 16,000× g at 4°C, pellets were washed with 80% acetone, dried for 30 min at room temperature and dissolved in 6 M urea-2 M thiourea. Proteins were digested with LysC and trypsin at room temperature and peptides were purified on stage tips and analysed by LC-MS/MS using an Easy nano LC system coupled to a Q Exactive mass spectrometer (Thermo Fisher Scientific). Peptide separation was achieved on a C18-reversed phase column (Reprosil-Pur 120 C18-AQ, 1.9 μM, 200×0.075 mm; Dr. Maisch) using a 95-min linear gradient of 2 to 30% acetonitrile in 0.1% formic acid. The mass spectrometer was set up to run a Top10 method, with a full scan followed by isolation, HCD fragmentation and detection of the ten most abundant ions per scan in the Orbitrap cell.
Vero cells were seeded in 6-well plates, infected with rTHOV viruses at high MOI and subjected to metabolic labelling of newly synthesized RNA at different time points after infection. For labelling, the cells were incubated in [3H]-5-Uridine-containing medium (20 μCi/ml und 1 ml/well) for 1 h and lysed in the lysis buffer from RNA extraction kit with subsequent total RNA extraction (RNeasy, Qiagen).
Cell viability was determined by MTT assay. Briefly, 0.5 mg/mL MTT were added to the cells and incubated for 3 h at 37°C, reaction was stopped by aspirating the medium and solving the crystals in 1:1 mix of DMSO:ethanol for 15 min shaking at room temperature, followed by absorbance (570 nm) measurement using a microplate reader (Tecan).
Duplex siRNAs (100 pmol of siRNA per 1×105 cells) were transfected using Neon Transfection System (Invitrogen) according to the manufacturer´s instructions for HeLa cells.
Cells were lysed in 1x SDS lysis buffer (62.5 mM Tris HCl pH 6.8, 2% SDS, 10% glycerol, 50 mM DTT, 0.01% bromophenol blue) containing protease inhibitors. Protein lysates were boiled at 95°C for 5 min, separated by SDS-PAGE and transferred onto nitrocellulose membrane (GE Healthcare Life Sciences). After blocking in 1xPBS containing 5% nonfat dry milk (Sigma-Aldrich) and 0.05% Tween for 30 min at room temperature, the membrane was first incubated for 1 h at room temperature with primary antibodies and then washed three times in 1x PBS containing 0.05% Tween with subsequent incubation in horseradish peroxidase-conjugated secondary antibodies and three additional washes. Detection was performed with SuperSignal West Femto kit (Pierce).
Total RNA was isolated using the NucleoSpin RNA II kit (Macherey-Nagel), including on-column DNase digestion, and 200 to 500 ng of RNA was reverse transcribed with the PrimeScript RT Master Mix (Takara). RNA levels were then quantified by real-time RT-PCR using the QuantiTect SYBR Green RT-PCR kit (Qiagen) and a CFX96 Touch Real-Time PCR Detection System (BioRad). Each cycle consisted of 15 sec at 95°C, 30 sec at 50°C and 30 sec at 72°C, followed by melting curve analysis. Primer sequences are provided in S8 Table.
Total RNA was isolated as described above. Library preparation and RNA sequencing was performed by Max Planck-Genome-Centre Cologne, Germany (http://mpgc.mpipz.mpg.de/home/). For RNA-Seq analysis trimmed and quality-filtered reads were mapped with Tophat2 [84] to the Ensembl genome annotation (version 70) and the human genome assembly GRCh37. Expression levels and differential gene expression were quantified using the cufflinks2 package [85]. In order to stabilize extreme fold change ratios generated by cuffdiff, we filtered out genes with a maximum per sample read count below 10 and calculated stabilized fold change values as the ratio of sample FPKM values each shifted by +2. The datasets can be found under GEO GSE105154.
Raw mass-spectrometry data were processed with MaxQuant software versions 1.4.1.8 and version 1.5.1.6 [86] using the built-in Andromeda search engine to search against human and mouse proteomes (UniprotKB, release 2012_06) containing forward and reverse sequences, and the label-free quantitation algorithm as described previously [83,87]. In MaxQuant, carbamidomethylation was set as fixed and methionine oxidation and N-acetylation as variable modifications, using an initial mass tolerance of 6 ppm for the precursor ion and 0.5 Da for the fragment ions. For SILAC samples, multiplicity was set to 2 and Arg10 and Lys8 were set as heavy label parameters. Search results were filtered with a false discovery rate (FDR) of 0.01 for peptide and protein identifications. Protein tables were filtered to eliminate the identifications from the reverse database and common contaminants. Data were analysed in Perseus.
In analysing mass spectrometry data from affinity purifications, only proteins identified on the basis of at least two peptides and a minimum of three quantitation events in at least one experimental group were considered. Label-free quantitation (LFQ) protein intensity values were log-transformed and missing values filled by imputation with random numbers drawn from a normal distribution, whose mean and standard deviation were chosen to best simulate low abundance values. Significant interactors of bait proteins were determined by multiple equal variance t-tests with permutation-based false discovery rate statistics. We performed 250 permutations and the FDR threshold was set between 0.02 and 0.1. The parameter S0 was empirically set between 0.2 and 1, to separate background from specifically enriched interactors.
For data analysis from pulsed SILAC experiments, we used log-transformed heavy to light protein ratios. Only proteins with valid values were considered for analysis. Profile plots were generated using LFQ intensities of log-transformed heavy-labelled protein intensities. Missing values were filled by imputation.
For the quantification of occupancies of TFIIB, Pol II, NELF and DSIF, we re-analysed published HeLa ChIP-Seq data from [49,50]. Trimmed and quality-filtered reads were mapped to genome assembly GRCh37 with bowtie2 [88] and filtered by mapping quality score (cutoff 30). A custom genome annotation file was generated based on Ensembl canonical transcripts containing target intervals relative to the transcription start site of -500 to +500bp (promoter), -50 to +300bp (downstream promoter) and +300bp to transcript end (gene body). Transcripts shorter than 1000bp were not considered. Read-coverage for these intervals was quantified using featureCounts [89] (TFIIB and H3K4me3: promoter; Pol II, NELF and DSIF: downstream promoter; Pol2: gene body). Where replicate samples were available, counts were normalized by library size, background-corrected by subtraction of input control and averaged across replicates. Gene pausing indices were calculated based on the Pol II samples from [50] as the length-normalized count ratio between the downstream promoter and the gene body intervals [12,72].
P-values for the significance of differential read coverage between selected sets of genes were calculated using a negative binomial count model with a log link function in R (using the MASS package).
|
10.1371/journal.ppat.1004504 | Plasticity between MyoC- and MyoA-Glideosomes: An Example of Functional Compensation in Toxoplasma gondii Invasion | The glideosome is an actomyosin-based machinery that powers motility in Apicomplexa and participates in host cell invasion and egress from infected cells. The central component of the glideosome, myosin A (MyoA), is a motor recruited at the pellicle by the acylated gliding-associated protein GAP45. In Toxoplasma gondii, GAP45 also contributes to the cohesion of the pellicle, composed of the inner membrane complex (IMC) and the plasma membrane, during motor traction. GAP70 was previously identified as a paralog of GAP45 that is tailored to recruit MyoA at the apical cap in the coccidian subgroup of the Apicomplexa. A third member of this family, GAP80, is demonstrated here to assemble a new glideosome, which recruits the class XIV myosin C (MyoC) at the basal polar ring. MyoC shares the same myosin light chains as MyoA and also interacts with the integral IMC proteins GAP50 and GAP40. Moreover, a central component of this complex, the IMC-associated protein 1 (IAP1), acts as the key determinant for the restricted localization of MyoC to the posterior pole. Deletion of specific components of the MyoC-glideosome underscores the installation of compensatory mechanisms with components of the MyoA-glideosome. Conversely, removal of MyoA leads to the relocalization of MyoC along the pellicle and at the apical cap that accounts for residual invasion. The two glideosomes exhibit a considerable level of plasticity to ensure parasite survival.
| Toxoplasma gondii can infect most warm-blooded animals, and is an important opportunistic pathogen for humans. This obligate intracellular parasite is able to invade virtually all nucleated cells, and as with most parasites of the Apicomplexa phylum, relies on a substrate-dependent gliding motility to actively penetrate into host cells and egress from infected cells. The conserved molecular machine (named glideosome) powering motility is located at the periphery of the parasite and involves the molecular motor, myosin A (MyoA). The glideosome exists in three flavors, exhibiting the same overall organization and sharing some common components while being spatially restricted to the central IMC, the apical cap and the basal pole of the parasite, respectively. The central and apical glideosomes are associated with MyoA (MyoA-glideosome) whereas the basal complex recruits myosin C (MyoC). Deleting components of the MyoC-glideosome uncovers the existence of complementary and compensatory mechanisms that ensure successful establishment of infection. This study highlights a higher degree of complexity and plasticity of the gliding machinery.
| The phylum of Apicomplexa groups numerous important animal and human pathogens. The best-studied members include the medically important Plasmodium species and Toxoplasma gondii for which robust reverse genetic approaches have been developed. T. gondii belongs to the subgroup of Coccidians that comprises other cyst-forming parasites such as Neospora, Eimeria, Cryptosporidium and Sarcocystis species that infect the intestinal tracts of animals and cause foodborne diseases referred to as coccidioses. In humans, Cryptosporidium can cause enteritis while T. gondii infection is usually asymptomatic or causes flu-like symptoms. Up to one third of the world's population is infected by T. gondii, generally without consequence because the immune response constrains the parasite to persist as a dormant encysted form. However, this form lasts for the life span of its host and can reactivate into an invasive and fast-replicating stage in case of immunosuppression [1].
Apicomplexan parasites are surrounded by a three-layered pellicle composed of a classical external plasma membrane (PM) and a double-membranous inner membrane complex (IMC) constituted of flattened vesicles [2]. Freeze fracture analyses of the pellicle of Toxoplasma, Eimeria and Sarcocystis have revealed a structural compartmentalization of the IMC with a cone-shaped plate called apical cap, and the remainder regularly arranged in longitudinal strips joined at the posterior pole [3], [4]. More discrete sub-compartments have been recently visualized in T. gondii through a family of proteins named IMC sub-compartment proteins (ISPs) located either at the apical cap, in the middle part of the IMC or in a basal region lying in the last third of the parasite length [5]. In addition, Coccidians possess a conoid, a motile organelle composed of tubulin fibers arranged in spiral at the apical pole [6], [7]. At the opposite pole, the basal complex remains more enigmatic but is composed of a basal polar ring where the membrane occupation and recognition nexus protein 1 (MORN1) localizes and a posterior cup where centrin 2 is found [8]. In Toxoplasma, both apical and basal complexes originate close to the centrosomes very early in the cell division process, and during the development of the daughter cells their basal complex appears as a ring structure that migrates to the basal pole and constricts in the mature parasite [8].
Most invasive apicomplexan zoites exhibit a unique substrate-dependent motion referred to as gliding motility, which allows parasites to cross non-permissive biological barriers and assists host cell invasion and egress from infected cells. The force generated by the parasite to propel itself inside a target cell originates from a conserved actomyosin machinery termed the glideosome that is located at the pellicle, in the limited space between the PM and the IMC [9]. The glideosome sustains the forward movement of the parasite by rearward translocation of adhesins that are apically secreted by the micronemes and bound to host cell receptors [10]. In Toxoplasma, the molecular motor complex is composed of the myosin heavy chain A (MyoA) and two associated light chains, the myosin light chain 1 (MLC1) and the essential light chain 1 (ELC1) [11]–[13]. This complex is recruited to the IMC via association with the C-terminal domain of the gliding-associated protein GAP45 [14]. While two integral membrane proteins of the IMC, GAP40 and GAP50, secure a firm anchoring of the complex in the outer membrane of the IMC, the N-terminal domain of GAP45 is fluidly inserted into the PM through two lipid modifications, myristoylation and palmitoylation [14]–[16]. The central sequence of GAP45 adopts an extended coiled-coil conformation that critically maintains the cohesion between the PM and the IMC during glideosome function, holding the two membranes at an optimal and constant distance [14]. Recently, MyoA was excised by DiCre recombinase in T. gondii and clones have been obtained establishing that this motor is dispensable for parasite survival, however their invasion rate was reduced by 80% [17]. In contrast MyoA could not be permanently excised in a parasite mutant lacking the gene coding for the two alternative spliced variants of the class XIV myosin B (MyoB) and myosin C (MyoC). Moreover the genes coding for actin (ACT1), MLC1 and GAP45 were conditionally excised but similarly, the parasites failed to be cloned, indicative of their essentiality [18].
GAP70 is a protein closely related to GAP45, which is found only in Coccidians and localizes exclusively to the apical cap [14]. Like GAP45, GAP70 is anchored by its N-terminal acylation to the PM and by its C-terminus to the apical IMC. It recruits MyoA but exhibits a longer coiled-coil domain and only partially complements GAP45 inducible knockout (GAP45-iKO) [14]. GAP70 is presumably tailored to accommodate a longer distance between the PM and the IMC. While GAP45 is essential for the lytic cycle of the parasite, GAP70 can be deleted without noticeable phenotype, likely due to a compensatory effect of the abundant GAP45 [14]. The C-terminal domains of GAP45 and GAP70 are very similar, and hence the specific determinant that targets GAP70 to the restricted area of the IMC remains unknown.
Coccidians possess a third member of this family, GAP80, which is shown here to localize to the posterior pole of T. gondii tachyzoites. GAP80 assembles a new glideosome around MyoC (MyoC-glideosome). Characterization of the partners interacting with GAP80 led to the identification of IMC-associated protein 1 (IAP1), a key determinant for the assembly of the MyoC-glideosome at the posterior polar ring. While this complex is dispensable for parasite survival, disruption of its individual components was strikingly compensated by the assembly of a chimeric glideosome composed of components of the MyoA- and MyoC-glideosomes. These findings shed light on the complexity and versatility of the gliding machine in the coccidian subgroup of Apicomplexa.
TgGAP70 (TGME49_233030) [14] and TgGAP80 (TGME49_246940) code for proteins showing considerable sequence similarity with GAP45 but which are restricted to the coccidian Toxoplasma, Neospora and Sarcocystis, in contrast to GAP45 which is found across the whole Apicomplexa phylum. The amino acid sequence alignment of these family members highlighted a significant conservation in the extreme C-terminus, which has been implicated in the interaction between GAP45 and MLC1-MyoA [14] (Figure S1A). In contrast to GAP45 and GAP70, the central region of GAP80 is not predicted to adopt a coiled-coil conformation probably due to the high content of proline residues (14% versus less than 3% in GAP45 and GAP70) and is instead predicted to fold into several short alpha helices (Figure S1B and file S1).
A knock-in (KI) strategy in Ku80-KO recipient strain [19], [20] was designed to insert a Ty-tag just upstream of the conserved C-terminal region of GAP70 (KI-GAP70Ty) and GAP80 (KI-GAP80Ty), respectively (Figure S1C). Stable parasite lines confirmed that both genes are expressed in the tachyzoite stage (Figure 1A). GAP80 exhibited the same abnormal migration behavior on SDS-PAGE as previously reported for GAP45 and GAP70 with an apparent molecular weight of 80 kDa whereas the predicted size is 45 kDa. Epitope tagging of GAP70 at the endogenous locus confirmed localization to the apical cap of the parasite previously reported based on expression of a second epitope-tagged copy [14]. In sharp contrast, GAP80 localized exclusively to the basal pole of mature parasites and showed a ring-shaped staining corresponding to the posterior polar ring (Figure 1B). To determine if the C-terminus of GAP80 was sufficient to confer the posterior localization, this domain consisting of the last 85 amino acids (aa) of the protein was either fused to GFP (MycGFPCtGAP80) or exchanged with the corresponding C-terminal domain of GAP70 (GAP70TyCtGAP80) and expressed as a second copy (Figure 1 C, E). As a control, expression of a second copy of GAP80Ty was found mainly targeted to the basal pole, opposite to the apical microneme staining of MIC4, and also slightly at the parasite periphery due to overexpression (Figure 1D). Exchange of the C-terminal domain in GAP70TyCtGAP80 conferred a posterior localization to the otherwise apically localized GAP70 (Figure 1D). MycGFPCtGAP80 also targeted to the basal polar ring, confirming that this C-terminal domain was sufficient to act as a targeting determinant (Figure 1F). As previously observed for the C-terminus of GAP45, the C-terminal domain of GAP80 alone was detectable in the nascent IMC of the daughter cells whereas the full-length proteins were found in the mature pellicle only (Figure 1F). This restriction is likely due to the absence of N-terminal acylation (bioinformatically predicted) in the case of MycGFPCtGAP80, which would anchor GAP80 to the PM prior to its association with the basal pole [14].
To ascertain the association of GAP80 with the membrane, fractionation experiments were completed. While KI-GAP80Ty was insoluble in PBS and high salt, it was partially solubilized in carbonate indicative of a peripheral protein that can also be partially extracted in the non-ionic detergent Triton X-100 (Figure 2A). GAP80Ty expressed as a second copy was more readily extracted in the various conditions, likely due to a looser IMC interaction caused by the overexpressed fraction that localizes to the periphery of the parasite and lacks basal-specific anchor(s).
To identify the interacting partners of GAP80, co-immunoprecipitation (co-IP) experiments were performed in the presence of Triton X-100 using anti-Ty antibodies on 35S-methionine and -cysteine metabolically labeled parasites expressing a second copy of either GAP80Ty or GAP45Ty as control. The eluted fractions showed similar profiles for MLC1, MyoA, GAP40 and GAP50 but an additional protein migrating around 130 kDa appeared only in the GAP80Ty co-IP (Figure 2B). The same profiles were also obtained for the co-IPs performed with the KI-GAP80Ty and KI-GAP45Ty strains [21] (Figure 2C) and the presence of MLC1, MyoA and GAP40 were verified by western blot analyses (Figure S2A). However, in the KI-GAP80Ty elution, GAP40 is much less abundant than within the MyoA-glideosome and the absence of GAP45 was confirmed by immunoblot despite the presence of a visible band at a similar size on the autoradiograph (Figures 2C). To validate that most of the glideosome components were shared between the two complexes, a reverse co-IP experiment was carried out on metabolically labeled parasites expressing a second tagged copy of the shared MLC1 (MLC1Ty). All the components of the glideosome were again present in the bound fraction including the protein migrating at around 130 kDa (Figure 2D). Mass spectrometry analyses confirmed that the band around 80 kDa corresponded to GAP80 and identified the protein migrating above 130 kDa (Figure S2B) to be encoded by TGME49_255190 (ToxoDB, [22]). This gene, previously described as the myosin B/C (MyoB/C), gives rise to two alternatively spliced products, MyoB and MyoC that differ in the length of their tail domain and in their localization [23]. MyoB/C was identified from 31 unique peptides covering 33% of the sequence (Table S1) and including one peptide (DVSYLIGMLFQR) specific to MyoC that was previously reported to be the predominant product expressed in the tachyzoite stage [23]. MyoC was C-terminally tagged in the endogenous locus and as previously observed with the expression of a second tagged-copy [23], MyoC was visible to the posterior polar ring of mature parasites and in the “late” stage developed daughter cells once the most basal sub-compartment of the IMC is built (Figure 2E). In contrast to GAP80-Ty, which is able to associate with a limited amount of MyoA, co-IPs performed with KI-MyoC-3Ty established that MyoC only interacts with GAP80, likely GAP50, ELC1 and MLC1 whereas GAP45 and MyoA are absent (Figure 2C). Since the GAPs are migrating in close proximity, western blot analyses on the co-IP materials were performed to confirm the presence of GAP40 and the absence of GAP45 in the MyoC complex. The association of MyoC with GAP80Ty was further confirmed by western blot analysis of the co-IP, using anti-MyoC antibodies (Figure S2C).
To determine if ELC1 was interacting with MyoC in addition to MyoA, a strain in which the endogenous gene was tagged at its C-terminus by knock-in was generated (KI-ELC1-3Ty) and the co-IP experiment performed using anti-Ty antibodies immunoprecipitated the entire glideosome including GAP80 and MyoC (Figure 2F). Finally, localization of endogenous ELC1, MLC1 and GAP40 clearly showed a staining posterior to that of GAP45 and corresponding to the location of the MyoC-glideosome (Figures 2G and S2D). In contrast, the signal for MyoA perfectly co-localized with GAP45 and was absent from the basal end (Figure 2G).
Taken together these findings identified a new coccidian-specific glideosome named the MyoC-glideosome, which shares the anchoring components to the IMC with the MyoA-glideosome, broadly conserved across the phylum of Apicomplexa. While GAP70 is part of the MyoA-glideosome at the apical cap [14], GAP80 belongs to the MyoC complex located at the posterior polar ring. MyoC belongs to the unconventional class XIV and appear to share the same myosin light chains, MLC1 and ELC1, with MyoA.
Given the largely shared composition of the MyoC-glideosome with the MyoA-glideosome, we reasoned that either GAP80 and/or a yet unidentified component should act as trafficking determinant(s) to confine the MyoC-glideosome to the posterior polar ring.
The C-terminal domain of GAP80 was sufficient to target GFP to the posterior pole and likely also sufficient to recruit the MyoC complex, as it is the case for the MyoA complex with GAP45. Toward the identification of a specific component anchoring the MyoC-glideosome to the basal sub-compartment of the IMC, we completed co-IP experiments with anti-Myc antibodies on parasite strains expressing either MycGFPCtGAP80 or the control MycGFPCtGAP45 (Figure 3A). MycGFPCtGAP80 efficiently immunoprecipitated MyoC, MyoA, MLC1, GAP40 and GAP50. Importantly, two additional components associated with MycGFPCtGAP80 became clearly visible when the samples were not boiled prior to loading on SDS-PAGE suggesting proteins with TMD or strongly associated with membranes [24], [25]. Preparative co-IPs were then performed with MycGFPCtGAP80 and MycGFPCtGAP70 as a control in order to identify the putative anchorage(s) (Figure S3A). Two bands, one below 40 kDa (protein 1) and one above 35 kDa (protein 2), were cut out and 8 and 9 proteins were identified by mass spectrometry, respectively (Table S2). Besides obvious contaminants corresponding to the abundant surface protein SAG1, heat shock and ribosomal proteins, peptides corresponding to MLC1 and GAP50 were also found. More interestingly, peptides corresponding to three hypothetical genes present only in Coccidians and exhibiting a similar cell cycle transcription profile as MyoC were identified and investigated further by epitope tag knock-in at the endogenous locus. The TGME49_283510 product was the only candidate localized to the posterior polar ring and the basal sub-compartment of the IMC and was named IAP1 for IMC-associated protein 1 (Figure 3 B, C and table S3). No transmembrane spanning domain was apparent for IAP1 but instead five cysteine residues were predicted to be palmitoylated with a high probability [26], supporting the strong interaction with the IMC (Figures 3D and S3B). In addition, acylation at multiple sites could explain why IAP1 migrated higher than its expected size, a shift that was even more pronounced when one (Figure S3C) or three acidic Ty-tags (Figure 3B) were added.
To demonstrate that IAP1 belongs to the MyoC-glideosome, parasites expressing KI-IAP1-3Ty were used to perform a co-IP together with the IMC-localized protein ILP1 [27] also tagged similarly at the endogenous locus (KI-ILP1-3Ty) and used as a negative control (Figure S3D). Western blot analyses revealed the presence of MLC1 and GAP40 in the bound fraction of KI-IAP1-3Ty but not in the control KI-ILP1-3Ty strain (Figure 3E). The presence of MyoC in this complex was visualized by autoradiography of a co-IP performed on metabolically labeled parasites expressing KI-IAP1-3Ty (Figure S3E).
To unravel how IAP1 associates with the IMC, we examined the contribution of four out of the five predicted palmitoylated cysteine residues lying in the N-terminal part of the protein (Figure 3D). A truncated version of IAP1 encompassing the 113 first residues (KI-Nt-IAP1-3Myc) was generated by knock-in in the Ku80-KO as well as in the KI-GAP80Ty background (Figure 3B, D). KI-Nt-IAP1-3Myc was still anchored to the posterior pole of the parasite but lost the polar ring localization and concomitantly GAP80 relocalized from the polar ring to the broader basal sub-compartment of the IMC (Figure 3F). To more directly analyze the contribution of the N-terminal cysteine residues in IAP1 anchoring, a second copy of IAP1 mutant exhibiting C3, C4 and C7 changed to alanine residues (AAA-IAP1-Ty, Figure 3D), controlled by tubulin promoter, was stably expressed. In contrast to its wild type counterpart that localized to the basal polar ring, AAA-IAP1-Ty was found in the cytoplasm (Figure 3G). In addition, this mutant was completely soluble in PBS while the wild type protein (endogenous or second copy) was fully extracted only in the presence of detergent (Figure 3H). Taken together, these data established that IAP1 is a component of the MyoC-glideosome that contributes to its basal polar ring localization most likely via N-terminal palmitoylation.
To gain insight into the function of the MyoC-glideosome without impacting on the MyoA-glideosome, MyoC, GAP80 and IAP1 were targeted for genetic disruption.
We first generated an N-terminal tagged version of the full-length MyoC by replacing the endogenous promoter by a Tet-inducible one in the TATi strain (MyoC-iKO) and a truncated version lacking the neck and tail domains (KI-MyoC-ΔN&T-Myc) by single homologous recombination in the MyoC locus (Figure S4A, B). In contrast to wild type MyoC that localized to the posterior polar ring of mature parasites and growing daughter cells, MyoC-ΔN&T-Myc was cytosolic (Figure 4A). Deletion of the neck and tail domains of MyoC destabilized GAP80 as shown by the reduced amount of GAP80 detectable by western blot (Figure 4B) but did not impact on GAP80 or IAP1 basal localization (Figure 4A). In addition, no noticeable phenotype has been observed during the lytic cycle by plaque assay (Figure 4C). Given the previous association of MyoB/C with pellicle integrity during cell division [23], we examined the rate of replication by counting the number of parasites per vacuole 24 hours post invasion (Figure S5A). This mutant showed no defect in intracellular growth and no impairment in egress (Figure S5B).
Since MyoC-ΔN&T-Myc showed no loss of fitness, a conventional knockout was produced (MyoC-KO) by double homologous recombination (Figure S5 C, D). Given the position of MyoC at the basal polar ring, we monitored by time-lapse microscopy the ability of MyoC-KO parasites to perform twirling during an induced egress assay and observed no defect compared to the wild type strain (Videos S1 and S2). We finally compared the co-IP of KI-GAP80Ty from a wild type and a MyoC-KO strain and confirmed the absence of MyoC while the rest of the complex was still assembled (Figure 4D). Importantly, the interaction of MyoA with GAP80 suggested that this motor had the potential to substitute for the absence of MyoC at the posterior polar ring. Since GAP80 level is lower in the absence of MyoC, it was not possible to make a quantitative comparison of the co-IPs between the two parasite lines.
To gain further information about the MyoC-glideosome, a conventional knockout of the GAP80 gene was generated in the Ku80-KO strain (Figure S5E, F). The absence of phenotype by plaque assay indicated that GAP80-KO parasites were able to accomplish their lytic cycle normally (Figure 5A) and indeed the individual steps including intracellular growth and egress were not altered (Figure S5 G, H). Surprisingly, upon deletion of GAP80, neither MyoC nor IAP1 showed an altered localization (Figure 5B). Given the homology between GAP80 and GAP45, it appeared plausible that GAP45 could compensate for the deletion of GAP80. We tested this hypothesis by performing a co-IP using anti-GAP45 antibodies on metabolically labeled wild type parasites and GAP80-KO (Figure 5C). In addition to the MyoA-glideosome components precipitated in the Ku80-KO strain, MyoC was precipitated in GAP80-KO parasites only, confirming that in this mutant strain GAP45 was able to interact with MyoC and hence possibly compensates for the absence of GAP80.
Finally, we generated an IAP1-KO strain in the Ku80-KO background (Figure S6A, B) as well as in MyoC-iKO and KI-GAP80Ty backgrounds. Since IAP1 is involved in the recruitment of the MyoC-glideosome to the basal polar ring, it has no known counterpart in the MyoA-glideosome to rescue its deletion. No loss of parasite fitness was monitored in the absence of IAP1 (Figure 5D) or in intracellular growth and egress (Figure S6C, D). However, in the absence of IAP1, GAP80 was no longer detectable at the basal polar ring or elsewhere in the parasite (Figure 5E) but remained detectable by western blot even though it appeared less abundant, likely due to its reduced stability in the absence of the complex (Figure 5F). MyoC was also absent from the basal polar ring and instead localized to the cytoplasm, at the periphery and also concentrated at the apical polar ring (Figure 5E). Strikingly, GAP45, a component of MyoA-glideosome clearly extended its localization to the basal end of GAP80-KO and IAP1-KO mutants as shown by the labeling of GAP45 in the posterior area, which is normally excluded in the Ku80-KO background strain (Figure 5G).
Individual deletion of the components of the MyoC-glideosome for which counterparts exist in the MyoA-glideosome are compensated for by the formation of a chimeric glideosome attesting to the adaptability and versatility of T. gondii. Moreover, deletion of IAP1, the protein responsible for the recruitment of GAP80-MyoC-MLC1-ELC1 to the basal polar ring, leads to the relocalization of at least some components of MyoA-glideosome to possibly compensate for the absence of a motor at the basal pole.
To tackle MyoC function, it was necessary to hamper MyoA incorporation into the basal glideosome to avoid a compensatory effect. To achieve that, we thought of introducing a non-functional point mutation in the ATP-binding site of endogenous MyoC. This parasite line was created using the same double homologous strategy as for the MyoC-iKO strain except that in the N-terminal homology fragment previously used to recombine, the ATP-binding site GESGAGKT was mutated to GESGAGET (Figure S4B, C). This mutation in the P-loop of MyoC (MyoC-K205E) is predicted to interfere with ATP-binding and thus should result in strong actin binding in combination with a lack of motile activity. To ensure the integration of the mutation, the N-terminal MyoC fragment was synthetized with a different codon usage up to and including the mutated ATP-binding site and the homologous region was lying downstream (Figure S4C). Additionally, to enhance the efficiency of the recombination, the locus was targeted with a specific guide RNA (gRNA) CRISPR-CAS9 plasmid [28]. Stable MycMyoC-K205E parasites were obtained (Figure S4D) and two clones from independent transfections were sequenced for the presence of the mutation. Surprisingly MyoC-K205E failed to localize to the basal polar ring of mature parasites as previously described [23] (Figure 6A). Instead, MyoC-K205E was clearly visible in all the late stage developed daughter cells identified with IMC1 (Figure 6A) implying that the protein is expressed and targeted to the basal polar ring during division but is not incorporated in this structure in the mature parasites. MyoC-K205E is not detectable by western blot (Figure 6B) suggesting that it is destabilized when not incorporated into the basal pole of mature parasites. Parasites expressing a non-functional MyoC have no phenotype in intracellular growth, invasion or egress (Figure S4 E-G). These data suggest that functional MyoC might be necessary for its integration in the basal pole. The expression of a non-functional MyoC led to a situation similar to the deletion of MyoC, leaving again physical room for a compensatory mechanism.
Given the overall similarities in the architecture and composition of the two glideosomes, it appeared legitimate to assume that the MyoC-glideosome participates in some aspects of the gliding function possibly exemplified by the stationary twirling where the parasite rotates, contacting the substrate via its posterior pole [29]. To support this notion we first anticipated that GAP80 could functionally complement the depletion of GAP45 when expressed at a suitable level. A second copy of GAP80 (GAP80Ty) was therefore introduced in the inducible knockout of GAP45 (GAP45-iKO) [14] and expressed under the control of the tubulin promoter (Figure 7A). Overexpression of GAP80Ty led to an overflow of the protein to the entire pellicle in addition to the basal polar ring and to a rescue of the recruitment of MLC1-MyoA to the pellicle upon depletion of MycGAP45i in the presence of anhydrotetracycline (ATc) (Figure 7B). This correlated with the partial complementation seen by plaque assay (Figure 7C) and the normal intracellular growth curve (Figure 7D) [14]. Moreover, GAP80Ty was able to complement both the invasion and egress defects caused by GAP45 deletion to levels almost comparable to the controls (Figure 7E, F and Table 1). The ability of GAP80 to restore the motility defect linked to depletion of GAP45 upon ATc treatment was assessed in the gliding trail assay using anti-SAG1 antibodies to detect the trails (Figure 7G). GAP80Ty is predominantly associated with the MyoA-glideosome upon GAP45i depletion while the level of assembly with MyoC remains constant (Figure 7H).
Complementation of GAP45-iKO by overexpression of GAP80Ty demonstrates that GAP80 is able to recruit a MyoA motor complex at the pellicle and to maintain sufficient cohesion between the PM and the IMC.
Ultimately, to circumvent potential compensatory mechanisms due to the plasticity between the MyoA- and MyoC-glideosomes, we opted for the disruption of GAP80 in the GAP45-iKO background (GAP45-iKO/GAP80-KO). When compared to GAP45-iKO, GAP45-iKO/GAP80-KO did not exhibit any defect in intracellular growth in the presence of ATc demonstrating that the MyoC-glideosome does not play a role in cell division (Figure 8A). In the presence of ATc, GAP45-iKO parasites were not able to egress from infected cells in response to calcium ionophore (A23187) stimulation as monitored by time-lapse microscopy (Videos S3, S4). The same phenotype was observed for GAP45-iKO/GAP80-KO parasites depleted in GAP45 (Videos S5, S6). In agreement with these observations, the two mutants showed severe defects in gliding (Figure 8B) and in egress assays following ATc treatment (Figure 8C and Table 1). Depletion of GAP45 in GAP45-iKO was previously reported to exhibit 20% residual invasion [14] that could be attributed either to leakiness of the Tet-inducible system, to a redundant or compensatory effect via the action of a distinct motor, or a distinct motor-independent mechanism of host cell penetration [17]. Invasion assays performed with GAP45-iKO/GAP80-KO in the presence of ATc revealed an enhanced defect with 50% less invasion compared to GAP45-iKO (Figure 8D and Table 1). These results establish that the MyoC-glideosome contributes to an efficient invasion process in T. gondii.
To further investigate the behavior of the MyoC-glideosome in the absence of MyoA, we freshly excised the gene from the loxP-MyoA strain [17] and cloned the parasites. GAP45 and MLC1 were previously shown to remain localized at the periphery of the parasite in the absence of MyoA [17] and hence antibodies raised against these two proteins were used for co-IP experiments to assess the composition of the complex (Figures 8E and S7). The same experiments were performed in parallel with MyoA-iKO parasites depleted in MyoA [12]. In both cases, around 50% more MyoC is found associated with MLC1 and GAP45 in the absence of MyoA (average of 2 and 3 independent experiments for MyoA-KO and MyoA-iKO, respectively, figure S7).
To determine if MyoC could substitute for the absence of MyoA in the peripheral glideosome and be possibly responsible for the residual invasion [17], MyoA was disrupted in the MyoC-iKO background by CRISPR/CAS9 mediated gene disruption [28]. Stable parasites were obtained and 3 independent clones were sequenced for the mutations introduced in the MyoA locus to repair the double-stranded break generated by the CAS9 at the specific target sequence (Figure S7D). In the absence of MyoA, MyoC is localized not only to the basal polar ring of mature parasites but also relocalized peripherally up to the apical polar ring (Figure 8F). In the absence of MyoA, the amount of MLC1 dropped dramatically while concurrently the amount of MyoC slightly increased (Figure 8G). As MyoA-KO has already a very severe phenotype in egress, no further aggravation could be scored in MyoC-iKO/MyoA-KO treated with ATc (Figure 8H). In contrast, the invasion defect that was around 10% in MyoA-KO further dropped to less than 2% upon MyoC depletion (Figure 8I and Table 1).
This study reports the identification and characterization of a new glideosome in T. gondii tachyzoites through the dissection of GAP80, a gliding-associated protein belonging to the GAP45 family and localized to the basal polar ring. The overall arrangement of the three glideosomes is similar and centered around a GAP45 family member that recruits a myosin motor complex to a sub-compartment of the IMC (Figure 9A). GAP45 is conserved across the phylum while GAP70 and GAP80 are restricted to the Coccidian subgroup of Apicomplexa that possess a sub-compartmentalized IMC. The three proteins are predicted to be N-terminally acylated at the plasma membrane and exhibit an extended central region predicted to form a coiled-coil domain or short alpha helices that vary significantly in length. While GAP45 recruits MyoA-MLC1-ELC1 along the central IMC, GAP70 and GAP80 are tailored for the apical cap and the basal complex, respectively. GAP80 recruits MyoC and assembles the MyoC-glideosome that shares GAP50, GAP40, MLC1 and ELC1 with the MyoA-glideosome. Further characterization of the complex identified IAP1, which is the necessary determinant to restrict its localization to the posterior polar ring. Deletion of IAP1 resulted in the loss of MyoC and GAP80 staining at the posterior ring. Given the absence of a TMD, the localization of IAP1 could either be mediated by palmitoylation that would stabilize the protein in the membrane bilayer, or by interaction with an un-identified protein. Alanine substitution of the three N-terminal cysteine residues predicted to be palmitoylated resulted in a cytoplasmic localization of the mutated IAP1. This result strongly suggests that palmitoylation at one or more sites is involved in the attachment of IAP1 to the lowest sub-compartment of the IMC and basal polar ring. In this context, one of the two recently characterized IMC-located protein S-acyl transferases might play an instrumental role in targeting [30]. Interestingly, the truncated version of IAP1 that encompasses 113 aa including three of the four predicted palmitoylated cysteine residues was associated with the basal sub-compartment of the IMC but not anymore to the polar ring. The same relocalization was observed for GAP80 indicating that these two proteins are interacting together and implicates the N-terminus of IAP1. The C-terminus of IAP1 is therefore associated with the polar ring either directly by palmitoylation or possibly by interaction with an integral membrane protein that remains to be identified.
MyoC and MORN1 are the only two proteins identified so far at the posterior polar ring of the growing daughter cells. MORN1 emerges much earlier than MyoC in dividing parasites where is also detected as dots at the extremities of the nascent IMC and at the centrocone [8], [23], [31]. MyoC was anticipated to play a role in cytokinesis by acting in a constrictive ring with MORN1 however no perturbation of MORN1 was observed upon cytochalasin D treatment suggesting already that the driving force of the constriction of the basal ring was unlikely dependent on actin/myosin [31]. Here we show that parasites lacking MyoC still assemble the basal complex as observed by the localization of GAP80 and also divide normally. In contrast, disruption of MORN1 has been achieved using two different strategies that led to a defect in basal complex assembly, cytokinesis and apicoplast segregation [32], [33].
MLC1 and ELC1 are two myosin light chains shared between MyoA and MyoC as shown by their ability to co-immunoprecipitate these two motors (Figure 2). In absence of MyoA, MLC1 remains associated with GAP45 and the pellicle confirming the interaction previously described between the C-terminal part of GAP45 and the N-terminal extension of MLC1 [14] while the signal of ELC1 is largely reduced (Figure 8E).
The fact that GAP70 and GAP45 are assembled with the same components has complicated the assessment of GAP70 function since a compensatory effect via GAP45 could not be excluded [14]. Deletion of GAP80 led to a significant recruitment of MyoC by GAP45. This compensatory mechanism illustrates the versatility of GAP45 and GAP80 in recruiting both motors (Figure 9B). While the three GAP45 family members are expressed in tachyzoites, only GAP45 appears to compensate for the loss of the two others probably due to its high level of expression. Although GAP45 might not be optimally tailored to function at the apical and posterior sub-compartments of the IMC, the compensation in the absence of GAP70 or GAP80 is sufficient to sustain gliding, invasion and egress. In contrast, GAP70 or GAP80 fail to compensate for the absence of GAP45 likely due to structural constraints and low level of their expression. Consistent with this view, the overexpression of a second copy of either GAP70 or GAP80 partially complemented the loss of GAP45 leading to a sub-optimal spacing between the IMC and the PM as shown in the case of GAP70 [14]. To circumvent those compensatory mechanisms, GAP80 was disrupted in GAP45-iKO. While motility is already severely compromised in GAP45-iKO, an additional 50% decrease in invasion was observed in GAP45-iKO/GAP80-KO compared to GAP45-iKO upon ATc treatment. Given that GAP45 fulfills the dual function of recruiting the MyoA motor complex to the pellicle and holding together the two membranes of the pellicle during motility [14], it was not possible to distinguish the impact of each phenomenon on motility, egress and invasion in GAP45-iKO/GAP80-KO.
MyoC-glideosome harbors two other specific components, MyoC and IAP1 that in principle offer an opportunity to address directly the function of the basal glideosome upon individual deletion of these two genes however significant plasticity and compensatory mechanisms were observed as well (Figure 9B). Deletion of IAP1 led to the disassembly of MyoC-glideosome and its replacement at the basal pole by at least one component of MyoA-glideosome, GAP45, based on IFA. Deletion of MyoC showed no significant impact on the parasite lytic cycle, as also recently reported [18]. However, co-IP experiments performed by immunoprecipitating either GAP80 or IAP1 showed that MyoA can be incorporated into the basal glideosome. Importantly, even a non-functional form of MyoC failed to integrate into the glideosome of the mature posterior pole, offering again the possibility for a compensatory effect to replace the defective MyoC.
The ultimate way to chase MyoC function was to disrupt both MyoC and MyoA simultaneously. However the recent report of the inability to clone excised-MyoA/MyoC-KO parasites [18] was indicative of a synthetic lethality between the two genes. While as expected, we also failed to generate parasites lacking both genes, we however succeeded in generating a MyoC-iKO/MyoA-KO. Importantly, in the absence of MyoA, MyoC clearly relocalized to the parasite periphery and at the apical pole, providing evidence that MyoC could partially replace MyoA in the peripheral and apical glideosomes (Figure 9C). Upon depletion of MyoC with ATc, invasion dropped to less than 2% compared to wild type parental strain, which further confirmed the central role played by both motors in invasion.
Genomic DNA has been prepared from tachyzoites (RH strain) using the Wizard SV genomic DNA purification system (Promega). RNA was isolated from tachyzoites using Trizol (Invitrogen). Total cDNA was then generated by RT-PCR performed with the Superscript II reverse transcriptase (Invitrogen) according to the manufacturer's instructions.
All amplifications were performed with the LA or Ex Taq (TaKaRa) polymerases and the primers used are listed in supplementary table S4.
T. gondii tachyzoites (RHhxgprt-, ku80-ko-hxgprt- [20] strains and their derivatives expressing the epitope-tagged proteins) were grown in confluent human foreskin fibroblasts (HFF) or in Vero cells maintained in Dulbecco's Modified Eagle's Medium (DMEM, life technology, Invitrogen) supplemented with 10% fetal calf serum, 2 mM glutamine and 25 µg/ml gentamicin.
Parasite transfections were performed by electroporation as previously described [38]. The hxgprt gene was used as a positive selectable marker in the presence of mycophenolic acid (25 mg/mL) and xanthine (50 mg/mL) for the pTUB8 vectors transfected in RH strains, as described previously [39].
Ku80-KO [19], [20] and derivative strains have been transfected with 20 µg of the knock-in constructs and with 40 and 60 µg of p2854-DHFR-5′3′GAP80, p2854-DHFR-5′3′IAP1 or pTub5-CAT-5′3′MyoC vectors. 1 µg/ml of pyrimethamine and 20 µM of chloramphenicol have been used to select the resistant parasites.
The pT/230-ble-5′3′GAP80 and the pTUB8GAP80Ty vectors were transfected in the ΔGAP45e/GAP45i strain [14]. Knockout parasites transfected with pT/230-ble-5′3′GAP80 were selected with 30 mg/ml of phleomycin and complemented parasites transfected with pTUB8GAP80Ty were selected with 1 µg/ml of anhydrotetracyclin (ATc).
To facilitate insertion by double homologous recombination in the MyoC locus, 70 ug of the 5′MyoC-pTetO7Sag4-MycNtMyoC-mut vector, and 30 ug of the MyoC gRNA-specific CRISPR/CAS9 vector have been transfected. To efficiently disrupt MyoA locus, 30 ug of the MyoA gRNA-specific CRISPR/CAS9 vector have been transfected. In both cases, 24 hours after transfections, parasites were sorted by flow cytometry and cloned into 96-well plates using a Moflo Astrios (Beckman Coulter).
The antibodies used in these study were described previously as follow: polyclonal rabbit: α-catalase [40], α-GAP45, α-PRF [41], α-MyoA [42], α-MLC1 [11], and α-GAP40 [21]; mouse monoclonal, α-ACT [11], α-Ty (BB2), α-Myc (9E10). For Western blot analyses, secondary peroxidase conjugated goat α-rabbit/mouse antibodies (Molecular Probes) were used. For immunofluorescence analyses, the secondary antibodies Alexa Fluor 488 and Alexa Fluor 594-conjugated goat α-mouse/rabbit antibodies (Molecular Probes) were used.
To generate the α-IMC1, a fragment encoding amino acids 140–610 was amplified from RH cDNA with primers IMC1-1/IMC1-2 and cloned into pETHTb (kindly provided by A. Houdusse, Paris) between the BamHI and XhoI sites in frame with 6 N-terminal histidine residues. The fusion protein was expressed into E. coli BL21 strain, affinity purified on Ni-NTA-agarose beads (Qiagen) according to the manufacturer's protocol under nature conditions and used to immunize two rabbits according to the Eurogentec standard protocol.
Parasite-infected HFF cells seeded on cover slips were fixed with 4% paraformaldehyde (PFA) or 4% PFA/0.05% glutaraldehyde (PFA/GA) in PBS, depending of the antigen to be labeled. Fixed cells were then processed as previously described [42]. Confocal images were generated with two laser scanning confocal microscopes: a Leica (TCS-NT DM/IRB and SP2) using a 1003 Plan-Apo objective with NA 1.4 and a Zeiss (LSM700, objective apochromat 63x/1.4 oil) at the Bioimaging core facility of the Faculty of Medicine, University of Geneva. Stacks of sections were processed with ImageJ and projected using the maximum projection tool.
Freshly released tachyzoites were harvested, washed in PBS, and then resuspended in PBS, PBS/1% Triton X-100, PBS/1M NaCl, or PBS/0.1 M Na2CO3 [pH 11.5]. Parasites were lysed by freeze and thaw followed by sonication on ice. Pellet and soluble fractions were separated by centrifugation for 30 minutes at 14,000 rpm at 4°C. The solubility of the catalase (CAT) was checked in the different conditions as control.
Parasites were lysed in PBS or PBS-1% Triton X-100 and mixed with SDS–PAGE loading buffer under reducing conditions. The suspension was either boiled or subjected to sonication on ice. SDS-PAGE was performed using standard methods. Separated proteins were transferred to nitrocellulose membranes and probed with appropriate antibodies in 5% non-fat milk powder in PBS-0.05% Tween20. Bound secondary peroxidase conjugated antibodies were visualized using either the ECL system (GE healthcare) or SuperSignal (Pierce).
Host cells were infected with parasites for 6 or 7 days before fixation with PFA/GA. Giemsa staining was then performed as described in Plattner et al., 2008.
Parasites were grown for 24 hours before fixation with PFA/GA. Double-labelling IFA was performed using α-GAP45 and α-actin antibodies. The number of parasites per vacuole was determined by counting the parasites in 100 vacuoles in duplicate for three independent experiments. For GAP45-iKO and MyoC-iKO containing strains, parasites were pre-treated for 30 hours ± ATc prior to inoculation.
This assay was performed as previously described [43] with the following specificities. Freshly released parasites were then inoculated on new host cells and allowed to invade for 20 minutes before fixation with PFA/GA for 5 minutes. The samples were first incubated with anti-SAG1 antibodies in PBS-2% BSA to reveal extracellular parasites and then, following Triton X-100 permeabilization, they were incubated with anti-IMC1 or anti-GAP45 antibodies to reveal the intracellular parasites. The number of intracellular and extracellular parasites was determined by counting 100 parasites in duplicate for four independent experiments. For GAP45-iKO and MyoC-iKO containing strains, parasites were pre-treated for 42 hours ± ATc.
New host cells were inoculated with freshly released parasites allowed to grow for 30 hours ± ATc. Parasite-infected host cells were then incubated for 5 min at 37°C with DMEM containing 0.06% DMSO or 3 µM of the Ca2+ ionophore A23187 from Streptomyces chartreusensis (calbiochem) before fixation. Double-labelling IFA was performed using α-GRA3 and α-SAG1 antibodies. The average number of egressed vacuoles was determined by counting 100 vacuoles for each condition for four independent experiments. For GAP45-iKO and MyoC-iKO containing strains, parasites were pre-treated for 24 hours ± ATc prior inoculation.
Parasites were grown for 42 hours ± ATc. Freshly released parasites were allow to settle on poly-L-lysine coated coverslips for 8 minutes in DMEM and then incubated for 10 min in an HEPES/calcium-saline solution before fixation with PFA/GA. Anti-SAG1 antibody was used without permeabilization to visualize the trails and the parasites. Three independent experiments have been performed.
HFF cells were heavily infected with freshly egressed parasites and washed several hours later. After 30 hours, cells were incubated in methionine/cysteine-free DMEM (sigma) for 1 hour before incubation in DMEM containing 50 µCi [35S]-labeled methionine/cysteine (Hartmann analytic GmbH) per ml for 4 hours at 37°C. For co-IPs, freshly released tachyzoites were harvested, washed in PBS and lysed in CoIP buffer (1% (v/v) Triton X-100, 50 mM Tris-HCl, pH 8, 150 mM NaCl) in the presence of a protease inhibitor cocktail (Roche). Cells were frozen and thawed five times, sonicated on ice, incubated for 10 min on ice, and centrifuged at 14,000 rpm for 1 hour at 4°C. Supernatants were incubated with monoclonal α-Ty or α-Myc or polyclonal α-GAP45 or α-MLC1 antibodies for 1 hour at 4°C on a rotating wheel. Protein A-Sepharose CL-4B (GE Healthcare Life Sciences) was then added and the incubation continued for 1 hour. Complexes were then washed three times in CoIP buffer. Finally, beads were resuspended in SDS loading buffer under reducing conditions.
Samples obtained after co-IP assays were separated by SDS-PAGE and stained with coomassie blue or silver stain. Bands of interest were excised from the gel and sent to the Proteomics Core Facility (Faculty of Medicine, Geneva, Switzerland) for analysis according to their standard protocols for protein identification. The fragments were generated with trypsin and the peaklist files were searched against the Toxoplasma gondii GT1 database (Toxoplasma Genomics Resource, release 8.2 of 31-May-2013, 8102 entries) using Mascot (Matrix Sciences, London, UK).
Genbank accession numbers: KF897514 for TgGAP80 and KF897515 for TgIAP1
|
10.1371/journal.pbio.0060088 | Reconstitution of DNA Strand Exchange Mediated by Rhp51 Recombinase and Two Mediators | In the fission yeast Schizosaccharomyces pombe, genetic evidence suggests that two mediators, Rad22 (the S. pombe Rad52 homolog) and the Swi5-Sfr1 complex, participate in a common pathway of Rhp51 (the S. pombe Rad51 homolog)–mediated homologous recombination (HR) and HR repair. Here, we have demonstrated an in vitro reconstitution of the central step of DNA strand exchange during HR. Our system consists entirely of homogeneously purified proteins, including Rhp51, the two mediators, and replication protein A (RPA), which reflects genetic requirements in vivo. Using this system, we present the first robust biochemical evidence that concerted action of the two mediators directs the loading of Rhp51 onto single-stranded DNA (ssDNA) precoated with RPA. Dissection of the reaction reveals that Rad22 overcomes the inhibitory effect of RPA on Rhp51-Swi5-Sfr1–mediated strand exchange. In addition, Rad22 negates the requirement for a strict order of protein addition to the in vitro system. However, despite the presence of Rad22, Swi5-Sfr1 is still essential for strand exchange. Importantly, Rhp51, but neither Rad22 nor the Swi5-Sfr1 mediator, is the factor that displaces RPA from ssDNA. Swi5-Sfr1 stabilizes Rhp51-ssDNA filaments in an ATP-dependent manner, and this stabilization is correlated with activation of Rhp51 for the strand exchange reaction. Rad22 alone cannot activate the Rhp51 presynaptic filament. AMP-PNP, a nonhydrolyzable ATP analog, induces a similar stabilization of Rhp51, but this stabilization is independent of Swi5-Sfr1. However, hydrolysis of ATP is required for processive strand transfer, which results in the formation of a long heteroduplex. Our in vitro reconstitution system has revealed that the two mediators have indispensable, but distinct, roles for mediating Rhp51 loading onto RPA-precoated ssDNA.
| Homologous recombination promotes genetic diversity in the next generation and serves as a driving force for evolution. It also provides efficient machinery for repairing DNA damage such as double-strand breaks. Homologous recombination involves DNA exchange between homologous chromosomes, which is mediated by evolutionarily conserved proteins called recombinases. It is thought that a recombinase binds to single-stranded DNA (ssDNA) to form a nucleoprotein filament called the presynaptic filament, and that this higher order structure engages in a search for homologous DNA sequences. Once a homologous duplex is found, the presynaptic filament initiates strand exchange. However, when ssDNA regions are created, they are immediately covered by replication protein A (RPA), thereby inhibiting recombinase filament formation even under conditions in which homologous recombination is appropriate. Previous studies suggested that mediator proteins help load recombinases onto ssDNA, and further studies showed that at least two mediators function together in a single recombination pathway. How these mediators coordinate recombinase loading has been unclear. We have addressed this question by reconstituting an in vitro strand exchange reaction with purified proteins including a fission yeast recombinase, Rhp51, two mediators, Rad22 and the Swi5-Sfr1 complex, and RPA. Our results indicate that Rad22 orchestrates the loading of Rhp51 onto RPA-coated ssDNA by acting as a scaffold for nucleating the recombinase filament, whereas the other mediator, Swi5-Sfr1, stabilizes and activates the filament.
| Homologous recombination (HR) generates genetic diversity by rearranging DNA sequences using homologous DNA information. It is also an important mechanism for repairing DNA double-stranded breaks (DSBs) and restarting stalled DNA replication forks. Accordingly, HR is essential to the preservation of genome integrity; defects in HR result in hypersensitivity to genotoxic agents and chromosomal aberrations. One conspicuous example of the role of HR in genome integrity is cancer prevention via the tumor suppressors BRCA1 and BRCA2, both of which interact with the Rad51 recombinase [1].
Genetic analyses of DSB repair and mitotic and meiotic HR in the budding yeast Saccharomyces cerevisiae have revealed a major pathway for HR that is under the control of proteins in the Rad52 epistasis group [2–4]. The Rad51 recombinase belongs to the Rad52 group and plays a key role in HR: Rad51 forms nucleoprotein filaments with single-stranded DNA (ssDNA), referred to as presynaptic filaments, and promotes strand exchange with donor DNA in an ATP-dependent manner. A series of analyses suggested that the assembly pathway for Rad51 on ssDNA in vivo is spatiotemporally regulated by replication protein A (RPA) and other Rad52 epistasis group proteins, such as Rad52 and Rad55–57 [5,6]. RPA immediately binds to ssDNA regions once they are formed (e.g., by the resection of DSB ends or by stalled replication forks). Rad51 alone cannot bind to RPA-coated ssDNA, as RPA has higher affinity for ssDNA than does the Rad51 recombinase. Rad52 assists in loading Rad51 onto RPA-coated ssDNA and in assembling the Rad51 nucleoprotein filament. The Rad51 paralogs Rad55 and Rad57, which form a heterodimer, assist Rad51-mediated filament assembly and/or stabilize the filament, leading to efficient strand exchange [7,8]. Proteins that facilitate Rad51 loading or filament stabilization are referred to as mediators [3,9]. The basic characteristics of the early steps of HR are widely conserved among eukaryotes; however, multicellular eukaryotes, including humans, have five Rad51 paralogs, XRCC2, XRCC3, RAD51B, RAD51C, and RAD51D. Several Rad51 paralog complexes have been observed, and these are also assumed to function as mediators [10,11]. In addition, BRCA2, a tumor suppressor, has been suggested to act as a recombination mediator [12–15].
The fission yeast Schizosaccharomyces pombe, which is evolutionarily distant from S. cerevisiae, uses an HR pathway very similar to that of budding yeast. However, a notable exception in S. pombe is the Swi5-Sfr1 complex, which functions as an additional mediator in the HR pathway involving Rad22 and Rhp51 (fission yeast Rad52 and Rad51 homologs, respectively) (reviewed in [16]). The Swi5-Sfr1 complex operates in the Rhp51-dependent HR pathway in parallel with another mediator, the Rhp55-Rhp57 complex (fission yeast Rad55 and Rad57 homologs, respectively) in vivo [17,18].
Swi5 is a small protein that is evolutionarily conserved from S. cerevisiae to man, but it has no known protein motifs [17,19,20]. Sfr1 (Swi5-dependent recombination repair protein 1) was identified as a Swi5 interactor that is involved in HR repair [17]. It shares homology with the C-terminal half of Swi2, which overlaps the interaction region for Swi5 and Rhp51 [17], and this region is modestly conserved from S. cerevisiae to man [20]. Sfr1 also lacks known functional motifs.
S. cerevisiae Sae3 and Mei5 are Swi5 and Sfr1 homologs, respectively. However, unlike S. pombe Swi5 and Sfr1, Sae3 and Mei5 are expressed only during meiosis. sae3 and mei5 mutants both show meiotic phenotypes very similar to those of the S. cerevisiae dmc1 mutant. These mutant phenotypes and the cellular localization of the two proteins suggest that they are specific for meiotic recombination associated with the meiosis-specific recombinase Dmc1 [20,21]. On the other hand, swi5 mutants exhibit more severe meiotic defects than do dmc1 mutants. Thus, S. pombe Swi5 clearly has an additional function beyond that involved in Dmc1-dependent activities [19].
We recently purified the Swi5-Sfr1 complex and found that it has ssDNA and double-stranded DNA (dsDNA) binding activities but that it lacks nuclease, helicase, and ATPase activities [22]. Consistent with genetic studies, the purified Swi5-Sfr1 complex stimulates both Rhp51- and Dmc1-mediated strand exchange in vitro [22] (reviewed in [16]). The stimulation of Rhp51-mediated strand exchange is closely related to its ssDNA-dependent ATPase activity, which is enhanced by the Swi5-Sfr1 complex. The Swi5-Sfr1 complex does not enhance the binding of Rhp51 to ssDNA per se. On the other hand, the mediator enhances the binding of Dmc1 to ssDNA. The molecular bases of the different effects of Swi5-Sfr1 on the two recombinases are still unknown.
An important issue emerged from our previous study. The Swi5-Sfr1 complex cannot efficiently overcome the inhibitory effect of RPA when RPA is bound to ssDNA prior to Rhp51 binding. This observation is consistent with the observation that the Swi5-Sfr1 complex does not appreciably affect the ssDNA binding capacity of Rhp51. However, the canonical definition of a recombination mediator is that it is an ancillary factor that overcomes the inhibitory effect of RPA on a recombinase.
S. cerevisiae Rad52 protein has been shown to interact directly with both RPA and Rad51 and to promote Rad51 filament formation by mediating the displacement of prebound RPA from ssDNA, leading to effective strand exchange mediated by Rad51 [23–25]. Therefore, it is possible that Rad22 acts exclusively to overcome the inhibitory effect of RPA and that the Swi5-Sfr1 complex acts exclusively to activate Rhp51 filaments. Thus, the concerted actions of these two mediators, Rad22 and the Swi5-Sfr1 complex, would direct the loading of Rhp51 onto ssDNA, leading to efficient strand exchange.
The work described here addresses this hypothesis with an in vitro system that we have established, which reconstitutes the early central step of homologous recombination. We found that Rad22 overcomes the inhibitory effect of RPA on strand exchange mediated by Rhp51-Swi5-Sfr1, as predicted. However, Swi5-Sfr1 is still essential for strand exchange, and both Rad22 and Swi5-Sfr1 are required for full reaction efficiency. In-depth analysis indicates that the two mediators work concertedly, but not exclusively, by different effects on Rhp51 to form the active filament required for effective DNA strand exchange. In addition, we have shown that the Swi5-Sfr1 mediator stabilizes and activates Rhp51-ssDNA filaments in an ATP-dependent manner, whereas Rad22 is not involved in Rhp51 activation.
We first determined whether purified Rad22 can overcome the inhibitory effect of RPA on strand exchange mediated by Rhp51-Swi5-Sfr1. Figure 1A shows a schematic diagram of the Rhp51-mediated three-strand exchange assay used in this study, in which pairing of a (+) strand DNA (circular ssDNA [css]) with a homologous linear duplex DNA (linear dsDNA [lds]) derived from øX174 phage generates a joint molecule (JM) that is converted to nicked circular DNA (NC) and linear ssDNA products by strand exchange. S. pombe Rad22 was bacterially expressed and purified as described in Materials and Methods. As previously reported [22], when css was first incubated with Rhp51 (and the Swi5-Sfr1 mediator) and then with RPA, large amounts of JMs and NCs were produced (Figure 1B). RPA was essential for strand exchange (compare lanes 1 and 2 in Figure 1B). In contrast, when css was first incubated with RPA and then with Rhp51 and Swi5-Sfr1, JM and NC formation was dramatically reduced (Figure 1B, lane 4), as previously reported [22]. This inhibitory effect of RPA was blocked by the addition of purified Rad22 (Figure 1B, lane5).
A roughly equivalent amount of bacterial ssDNA binding protein (SSB; 2 μM) could be substituted for RPA (1 μM) when the strand exchange reaction was initiated by Rhp51/css complex formation (Figure 1B, lane 3). This result is consistent with RPA acting to prevent reversal of the already formed DNA joints by sequestering the free ssDNA, a reaction in which RPA can be replaced by SSB. However, when css was precoated with SSB (2 μM) before adding Rhp51 and Swi5-Sfr1, JM and NC product formation was severely reduced (Figure 1B, lane 6). More importantly, Rad22 could not overcome the inhibitory effect of SSB (Figure 1B, lane 7), indicating that functional interactions between RPA with Rad22 are important for this step.
RPA-coated ssDNA is assumed to be a natural substrate for the in vivo strand exchange reaction. Therefore, the requirements for reactions initiated with RPA-coated ssDNA were examined (Figure 1C). These reactions were strictly dependent on Rhp51, Swi5-Sfr1, RPA, and ATP. Rad22 was not essential, but in its absence, the levels of JM and NC products were severely reduced. Time-course experiments clearly demonstrated that Rad22 alone stimulated very little Rhp51-dependent strand exchange when RPA-coated ssDNA was used for a substrate (Figure 1D). Swi5-Sfr1 alone stimulated the reaction, and reactions in which Rad22 and Swi5-Sfr1 were coincubated with Rhp51 proceeded with robust efficiency. These results clearly indicate that full reaction efficiency requires the functions of the two mediators, Rad22 and Swi5-Sfr1.
Next, we examined whether the timing of Rad22 addition affects the strand exchange reaction (Figure 2A), since the addition order is critical for the Rhp51-Swi5-Sfr1–mediated reaction [22]. Note that protocol 2 in Figure 2A employs the same addition order as that of the standard reaction (e.g., the reaction shown in Figure 1B). Surprisingly, the time at which Rad22 was added was not crucial (Figure 2A): all protocols were highly efficient, with the exception of reactions lacking Rad22. These results indicate that Rad22 can overcome the inhibitory action of RPA irrespective of when it is added.
These data suggest that Rad22 may coordinate strand exchange in a single mixture that includes all protein components. To address this possibility, we set up the following reactions. We prepared two mixtures, one containing RPA-coated css and the other containing all other reaction constituents. The reactions were started by combining the two mixtures and incubating at 37 °C for 120 min. As shown in Figure 2B, the results clearly indicate that this protocol allows fully efficient reactions initiated from RPA-coated ssDNA and that Rad22 is essential, since reaction efficiency was substantially decreased if it was omitted (compare lanes 2 and 3 in Figure 2B). Furthermore, we found that a mixture containing all protein components (RPA, Rad22, Rhp51, and Swi5-Sfr1) efficiently promoted strand exchange when combined with css (Figure 2B, lane 4). Rad22 is essential for this reaction as well (compare lanes 4 and 5). These results indicate that Rad22 coordinates the functions of all proteins and orchestrates DNA strand exchange in vitro.
The Rad22-dependent reactions required ATP hydrolysis. ADP or ATPγS did not promote strand exchange (Figure 2C), as previously observed for the Swi5-Sfr1–dependent reaction in the absence of Rad22 [22]. Interestingly, AMP-PNP supported a small amount of JM formation, but not NC production. The most effective concentration of Rad22 was approximately one tenth the concentration of Rhp51. Higher concentrations of Rad22 inhibited the reaction (Figure 2D).
The inhibitory effect of RPA may result from its higher affinity for ssDNA compared to that of Rhp51. If this is correct, either or both of the two mediators may function to displace RPA from ssDNA to facilitate the loading of Rhp51, thereby leading to efficient presynaptic filament formation for the strand exchange reaction. In addition, once the correct filament is formed, it should be stabilized to protect against further RPA binding to ssDNA, since the thermodynamic equilibrium favors RPA binding to ssDNA. Either or both of the two mediators may be involved in this stabilization, as well.
To test these hypotheses, we set up pull-down assays in which the conditions were the same as for the strand exchange reaction. We first performed a titration experiment of RPA to css-bound beads (css beads) (10 μM nucleotides). The results of this experiment indicated that approximately 30% of input RPA (1 μM) was excess to RPA bound (∼0.7μM) to css beads (Figure 3A). Next, we analyzed the nucleotide dependency of Rhp51 binding to ssDNA (Figure 3B and 3C). In the absence of adenine nucleotides, Rhp51 did not bind to ssDNA, indicating that the ssDNA binding activity of Rhp51 requires an adenine nucleotide. Titration experiments indicated that both ATP and AMP-PNP were highly efficient cofactors, whereas ADP and ATPγS were slightly less effective (Figure 3B). Note that the amount of Rhp51 (5 μM) used in the standard strand exchange reaction is excessive (about 2.5- to 3-fold) to its binding to css beads in the presence of any adenine nucleotide.
We then performed competition experiments to compare the ssDNA binding activity of Rhp51 to that of RPA. Mixtures containing RPA (1 μM) and Rhp51 (5 μM) were incubated with css beads. Protein–DNA complexes were pulled down, and the amount of Rhp51 bound to ssDNA was analyzed by SDS-PAGE (Figure 3D–3G). In the presence of adenine nucleotides, but in the absence of mediators, Rhp51 was not pulled down with css, indicating that RPA indeed has a higher affinity for ssDNA than does Rhp51 (Figure 3D). A similar result has been reported for Rad51 and RPA from budding yeast [26].
We examined which of the two mediators assists Rhp51 loading onto ssDNA. The Swi5-Sfr1 complex alone promoted Rhp51 loading onto ssDNA in the presence of ATP and AMP-PNP (Figure 3E). About 22% and 16% of the input Rhp51 were pulled down with ssDNA in the presence of ATP and AMP-PNP, respectively. Reactions containing ADP or ATPγS or lacking an adenine nucleotide only weakly supported Rhp51 loading (about 5% of the input Rhp51 was pulled down).
Incubation with Rad22 alone allowed a small amount of Rhp51 to be pulled down (about 10%) in the absence of a nucleotide cofactor or in the presence of ATP, ADP, ATPγS, or AMP-PNP (Figure 3F). The amounts of Rhp51 that were pulled down were not significantly affected by the absence or presence of the nucleotide cofactor, or by the type of nucleotide in this case. Budding yeast Rad52 associates with RPA-ssDNA to accelerate the Rad51-mediated displacement of RPA [27]. However, since the binding of Rhp51 to ssDNA required an adenine nucleotide (Figure 3B and 3C), the detection of Rhp51 in the pulled-down ssDNA protein complexes may reflect Rhp51 bound to Rad22 that is associated with preformed RPA-ssDNA complexes, rather than the direct binding of Rhp51 to ssDNA. The very low level of Rhp51 pulled down in the presence of ADP and ATPγS in the absence of Rad22 (Figure 3D and 3E, lanes 4 and 5) suggests that this basal level of Rhp51 is dependent on Rad22. We also observed Rhp51–Rad22 and Rad22–RPA interactions in the absence of an adenine nucleotide with an immunoprecipitation assay (unpublished data); similarly, tight Rad51–Rad52 and Rad52–RPA interactions have been reported for budding yeast and human cells [23–30]. These observations also support the notion that the small amount of Rhp51 pulled down when Rad22 alone is included is due to the basal level of Rhp51 that binds to Rad22, which in turn is associated with RPA bound to css beads.
When Rad22 and Swi5-Sfr1 were present, Rhp51 was loaded onto ssDNA (39% of input) in an ATP-dependent manner (Figure 3G). The amount of Rhp51 loaded under these conditions was much higher than with Swi5-Sfr1 alone or with Rad22 alone, indicating that the two mediators function synergistically to promote Rhp51 loading onto ssDNA. Interestingly, the nonhydrolyzable ATP analog, AMP-PNP, but not ATPγS, supported Rhp51 loading under these conditions. Small amounts of pulled-down Rhp51 were detected in the absence of nucleotide (Figure 3G, lane 2) and in the presence of ADP or ATPγS (Figure 3G, lanes 4 and 5). These low levels are likely due to Rhp51 associated with ssDNA through the Rad22–RPA interaction, as mentioned above.
We next examined which protein factor is directly involved in displacing RPA from ssDNA (Figure 4). RPA was first incubated with css beads. ATP, Rhp51, and/or the two mediators were then added. The mixtures were pulled down, and proteins in the bound and unbound fractions were analyzed by SDS-PAGE. Neither Rad22 nor Swi5-Sfr1 could displace RPA from ssDNA (Figure 4A–4C). Only high amounts of Rhp51 could displace RPA at a detectable level, indicating that Rhp51 per se, but not the mediators, is an intrinsic factor that displaces RPA from ssDNA (Figure 4D). Swi5-Sfr1 stimulated Rhp51-dependent RPA displacement (Figure 4F), whereas Rad22 alone stimulated displacement only modestly (Figure 4E). See also the graphs of the amounts of displaced RPA (Figure 4H) and bound Rhp51 (Figure 4I). At low protein concentrations, Rad22 appeared to assist the loading of Rhp51 onto ssDNA at a considerably high level, as judged by the amount of Rhp51 in the bound fractions (Figure 4I). However, since the RPA levels in the unbound fractions increased only slightly when Rad22 was also included (Figure 4H), the observed increase of Rhp51 in Figure 4I may be due to Rhp51 bound to RPA-coated css beads via Rad22. It has been reported that Rad22 interacts with both Rhp51 and RPA [31,32], and the ternary complex can also be detected in vitro by coimmunoprecipitation (unpublished data). However, RPA displacement mediated by Rhp51 with Swi5-Sfr1 was further stimulated by coincubation with Rad22 (Figure 4G). This synergistic effect was robust; three independent experiments yielded the same results. Therefore, these results again indicate that the two mediators function coordinately to assist in the displacement of RPA. In the absence of ATP, RPA displacement did not increase from basal levels, even in the presence of all the protein components in a 4-fold excess relative to their concentrations in the standard reaction (unpublished data), indicating that RPA displacement requires ATP. This result is consistent with the hypothesis that Rhp51 per se is a displacing factor, since neither Rad22 (unpublished data) nor Swi5-Sfr1 is an ATP-binding protein [22].
Time-course experiments for RPA displacement from and Rhp51 loading onto ssDNA were also performed (Figure 5). Rhp51 alone was not loaded onto ssDNA and did not promote RPA displacement (Figure 5A). A small amount of Rhp51 in the presence of Rad22 was loaded onto ssDNA, but this increase and the displacement of RPA ceased within 30 min (Figure 5B). Swi5-Sfr1 promoted Rhp51 loading and RPA displacement (Figure 5C), and coincubation of Rad22 and Swi5-Sfr1 strongly enhanced both processes (Figure 5D). A quantitative presentation of these assays is shown in Figure 5E and 5F. Taking these results together, we conclude that the two mediators work concertedly, but not exclusively, to promote Rhp51 loading onto and RPA displacement from ssDNA to form the active presynaptic filament required for effective DNA strand exchange.
The results described above suggest that the ATPase activity of Rhp51 plays an important role in both RPA displacement and Rhp51 filament formation. Therefore, we examined the ATPase activity of Rhp51 under various conditions (Figure 6). It has been reported that Rhp51 has low activity, but considerably higher than that of other Rad51 proteins in the absence of ssDNA and that this basal-level ATPase activity is very slightly enhanced in the presence of ssDNA [22,33]. We show here that neither of the two mediators has an effect on the intrinsic (DNA-free) or dsDNA-dependent ATPase activities of Rhp51 (Figure 6A and 6C). However, the Swi5-Sfr1 complex stimulated the ssDNA-dependent ATPase activity of Rhp51 about 3-fold (Figure 6B), consistent with a previous report [22], but Rad22 had no effect in this respect, regardless of the presence of the Swi5-Sfr1 complex (Figure 6B).
We hypothesized that stimulation of the ssDNA-dependent ATPase activity of Rhp51 by Swi5-Sfr1 alters the Rhp51 filament. An RPA attack experiment supported this hypothesis (Figure 7A–7E). Rhp51 filaments were allowed to form on css beads under various conditions, and RPA was then added. Proteins that remained bound to ssDNA were analyzed by SDS-PAGE. In the presence of ATP, but absence of RPA, Rhp51 was pulled down efficiently (Figure 7A, lanes 1–3, and Figure 3C). Swi5-Sfr1 had no effect on the ssDNA binding capacity of Rhp51, as previously reported [22]. However, when RPA was added to the reaction mixture in the absence of Swi5-Sfr1, it became detectable in the ssDNA fraction, and almost all of the Rhp51 bound to ssDNA disappeared (Figure 7A, lane 4). In contrast, when RPA was added to the reaction mixture in the presence of Swi5-Sfr1, the Rhp51 filament became resistant to RPA attack in a Swi5-Sfr1 concentration-dependent manner (Figure 7A, lanes 5 and 6). The formation of the resistant Rhp51 filament was ATP dependent, and neither ADP nor ATPγS could replace ATP in this reaction (Figure 7B–7D). Interestingly, AMP-PNP made the Rhp51 filament resistant to RPA attack, even in the absence of Swi5-Sfr1 (Figure 7E), consistent with a report that human Rad51 forms more stable filaments with AMP-PNP [34,35].
Rad22 had little detectable effect on the formation of resistant Rhp51 filaments (Figure 7F). However, large amounts of Rad22 modestly facilitated Rhp51 pull down (Figure 7F, lane 6). Unlike what was seen for coincubation with Swi5-Sfr1 (Figure 7A), however, the amount of RPA bound to ssDNA was constant (Figure 7F, lanes 4–6, and the histogram below). In addition, the recovery of Rhp51 was adenine nucleotide-independent (unpublished data). Therefore, Rhp51 recovered with assistance of Rad22 is not the same as the resistant Rhp51 induced by Swi5-Sfr1. Since Rad22 binds strongly to both Rhp51 and RPA (unpublished data, and [31,32]), and the ternary complex is formed in solution in the absence of ATP (unpublished data), these results indicate that an RPA-Rad22-Rhp51 complex bound to ssDNA via RPA is pulled down.
Taken together, these results suggest that Swi5-Sfr1 induces activation of the Rhp51 filament to promote strand exchange in an ATP-dependent manner. Rad22 may not be directly involved in filament activation, but rather, it may recruit Rhp51 to RPA-coated ssDNA.
We measured the ATPase activity of Rhp51 using RPA-coated ssDNA as a cofactor, which more closely approximates physiological conditions. As shown in Figure 6D, Swi5-Sfr1 stimulated the ATPase activity of Rhp51, but this stimulation was less than that observed when RPA-free ssDNA was used as a cofactor (compare Figure 6B and 6D). Interestingly, the level of stimulation of the Rhp51 ATPase activity by Swi5-Sfr1 reached the level of that observed with RPA-free ssDNA when Rad22 was added to the reaction (Figure 6D). This synergistic stimulation of Rhp51 ATPase activity is consistent with the coordinated action of these two mediators in DNA strand exchange. In a control experiment, we constructed and purified an Rhp51 Walker B box mutant protein (D to N alteration; Rhp51D244N). The behavior of Rhp51D244N during purification by chromatography was the same as that of the wild-type protein (unpublished data). The mutant did not produce any detectable levels of phosphate generated by ATP hydrolysis. Coincubation of Swi5-Sfr1 with Rhp51D244N did not increase the level of hydrolyzed phosphate (Figure 6E). This result indicates that the ATPase activity stimulated by Swi5-Sfr1 is indeed that of the wild-type Rhp51 protein.
This study demonstrates an in vitro reconstitution of the central step in eukaryotic HR. Our system consists entirely of purified components, including recombinase, RPA, and the Rad22 and Swi5-Sfr1 mediators, and it reflects the genetic requirements for these components in vivo. Using this system, we present for the first time robust biochemical evidence that the two mediators function in a concerted manner to form the active Rhp51 filament. Dissection of the reaction uncovered several of the molecular details of strand exchange. First, Rad22 overcomes the inhibitory effect of RPA, but not of SSB, in strand exchange mediated by Rhp51-Swi5-Sfr1 (Figure 1B). In addition, Rad22 negates the need for a strict order of addition of protein components, indicating that it coordinates strand exchange (Figure 2). However, even in the presence of Rad22, Swi5-Sfr1 is essential for Rhp51-mediated strand exchange, highlighting the different fundamental properties of the two mediators (Figure 1C). Although the molecular functions of the two mediators are distinct (see below), they function synergistically to promote Rhp51 loading (Figure 3). Importantly, Rhp51, but not Rad22 or the Swi5-Sfr1 mediator, displaces RPA from ssDNA (Figures 4 and 5). We previously showed that Swi5-Sfr1 stimulates the ssDNA-dependent ATPase activity of Rhp51 [22]. Here, we demonstrated that Rad22 does not affect the ATPase activity of Rhp51 (Figure 6). Most important, Swi5-Sfr1 renders Rhp51 resistant to RPA attack in an ATP-dependent manner (Figure 7A), suggesting that Swi5-Sfr1 stabilizes the presynaptic filaments of the recombinase. Since this stabilization is ATP-dependent, Swi5-Sfr1 stimulates the ssDNA-dependent ATPase activity of Rhp51, and the mediators stimulate strand exchange mediated by Rhp51. Thus, these mutual relationships strongly suggest that the induced stabilization of the Rhp51 filament reflects a structural/functional alteration upon activation. In contrast, Rad22 is not involved in Rhp51 activation (Figure 7F). Based on these results, we propose a model for the early step of the strand exchange reaction involving Rhp51 and the two mediators. Rad22 recruits Rhp51 to RPA-coated ssDNA. Rad22 and Swi5-Sfr1 collaborate in displacing RPA and loading Rhp51 onto ssDNA, but the displacing factor itself is Rhp51, and displacement requires the ATP binding activity of Rhp51. Swi5-Sfr1 activates Rhp51 recruited to RPA by Rad22 to form and stabilize/activate the presynaptic filament in an ATP-dependent manner, promoting processive strand exchange.
Localization of budding yeast Rad51 to DSB sites requires Rad52 [5,6,26,36,37]. Although experimental data on whether Rhp51 focus formation is dependent on Rad22 functions are not yet available, at least to our knowledge, this is widely believed to be true as well for fission yeast, based on similarities between the two systems. However, the contribution of Rad22 to our in vitro strand exchange reaction was relatively small. In contrast, the influence of the other mediator, Swi5-Sfr1, is much stronger than that of Rad22. Our results indicate that substoichiometric amounts of Rad22 are optimal to overcome the RPA inhibitory effect on strand exchange (1 to 5∼10 ratio of Rad22 to Rhp51; see Figure 2D). This is similar to what has been seen for budding yeast Rad52 in strand exchange [38]. These results suggest that Rad22 may only transiently interact with the Rhp51 nucleoprotein filament. Such transient interactions likely mediate assembly of the Rhp51 filament. However, Rad22 (and presumably Rad52) may have (an)other uncovered function(s) involved in the overall in vivo strand exchange reaction. Based on chromatin immunoprecipitation analyses, Wolner and coworkers suggested that Rad51 binds first, followed by Rad52 and Rad55, to a single DSB site in budding yeast [26]. This result apparently contradicts the genetic dependency of Rad51 focus formation on Rad52. Wolner et al. proposed that only the stable binding of Rad51 is detectable by chromatin immunoprecipitation, reflecting the assembly of a nucleoprotein filament catalytically competent for strand invasion. Indeed, Arai et al. have demonstrated that a stoichiometric complex of Rad52 with Rad51 is required for the efficient formation of D-loops via strand invasion [39]. The strand exchange system we used in this study does not include a strand invasion step, accounting for the small amounts of optimal concentration and a lower dependency on Rad22. An apparent strong dependency on Swi5-Sfr1 may be valid only for a three-strand exchange reaction using a long DNA substrate, an assay that mimics the strand transfer reaction but which does not include strand invasion. Further study will be needed to reveal functional differences between the two mediators in D-loop formation.
We propose here a two-phase activation mechanism of Rhp51 filament formation. Rhp51 is the only component with ATPase activity among the proteins used in our assay, and this activity enables the protein to bind to ssDNA in an adenine nucleotide-dependent manner. However, Rhp51 alone cannot efficiently promote strand exchange; instead, it requires Swi5-Sfr1 [22]. Thus, the first phase of activation is one in which Rhp51 is activated to bind ssDNA. Swi5-Sfr1 then further activates ATP-bound Rhp51 to make it catalytically competent for strand exchange. In the second phase of activation, the Swi5-Sfr1 mediator also renders the Rhp51 filament resistant to attack by RPA (Figure 7A).
The ATPase activities of Rad51 proteins from budding yeast and human cells are dependent on the presence of ssDNA [40–42]. In contrast, the Rhp51 ATPase is considerably efficient in the absence of ssDNA, and the presence of ssDNA does not enhance its activity. Binding of human Rad51 to DNA can occur in the absence of ATP, but budding yeast Rad51 requires ATP for DNA binding [43]. Rhp51 also requires adenine nucleotides for ssDNA binding (Figure 3). These observations indicate that Rad51 properties relevant to ATPase and DNA binding are not universal, and variations may be related to the apparent differences in activation mechanisms for strand exchange. Budding yeast and human Rad51 proteins can carry out a robust strand exchange reaction in the absence of a Swi5-Sfr1-type mediator in vitro. Human Rad51 is already activated for the first step because it can bind ssDNA without ATP. Once ATP is included in the reaction, human Rad51 may be further activated to the catalytically competent state. Budding yeast Rad51 requires ATP to bind DNA, implying that the first stage is similar to that of Rhp51. Once Rad51 binds to ssDNA, its ATPase activity is stimulated, indicating that the second phase of activation readily occurs without a Swi5-Sfr1–type mediator. In other words, the basal level of activation of the first phase is higher in human Rad51, and the second phase is higher in budding yeast Rad51. This idea can reconcile the apparent differences among the three recombinases.
Interestingly, under biochemical conditions that slow ATPase activity, such as the presence of Ca2+, strand exchange mediated by human Rad51 is enhanced [34,44,45]. However, activation by Ca2+ holds only for human Rad51, not for budding yeast Rad51 [44] or fission yeast Rhp51 (unpublished data). The presence of Ca2+ attenuates the disassembly of Rad51 from ssDNA, resulting in stable filaments on ssDNA [45]. In contrast, Swi5-Sfr1 stimulates the Rhp51 ATPase [22] and stabilizes the filament in an ATP-dependent manner (Figure 7). We cannot explain why these two opposing effects lead to a stimulation of strand exchange, although the differences in the basal-level status of yeast and human Rad51 proteins may be a key to this issue.
AMP-PNP, but not ATPγS, induces a similar stabilization of Rhp51 that is independent of Swi5-Sfr1. Although Rhp51 binding to AMP-PNP is thought to mimic the transient ATP-bound form during ATP hydrolysis, the AMP-PNP–bound and ATP-bound Rhp51 forms are qualitatively different. The AMP-PNP–bound form of Rhp51 must be locked in a unique activated state with Swi5-Sfr1, which presumably represents the second phase of activation, since the ATP-bound Rhp51 filament itself is not sufficient for strand exchange. Importantly, AMP-PNP cannot support processive strand exchange. Both the Swi5-Sfr1–activated ATP-bound form and the AMP-PNP–bound form are competent states that promote homologous pairing and the transient formation of a three-strand intermediate. However, hydrolysis of ATP is required for the subsequent steps of consecutive strand transfer from duplex DNA to ssDNA that result in the formation of a long heteroduplex. Swi5-Sfr1 promotes both of the states that permit homologous pairing and consecutive strand transfer by stimulating the ssDNA-dependent ATPase activity of Rhp51. The enhancement of the Rhp51 ATPase activity by Swi5-Sfr1 is not due to an increased ADP–ATP exchange rate, but rather to an enhanced turnover rate [22]. A rapid turnover between the first and the second stages might promote efficient strand exchange. The precise mechanism by which Swi5-Sfr1 induces activation of the Rhp51 filament remains to be clarified in future studies.
S. pombe Rhp51 was purified as previously described [22]. Alternatively, Rhp51 was expressed in the Escherichia coli strain BL21-CodonPlus(DE3)-RIPL carrying an Rhp51 expression plasmid derivative of pET11b (Novagen). Cells were incubated at 37 °C in LB media containing ampicillin. When cell density reached an optical density at 600 nm (OD600) of approximately 0.5, isopropyl-ß-D-thiogalactopyranoside (IPTG) was added to a final concentration of 0.5 mM, and the culture was further incubated for 18 h at 18 °C. The cells were collected by centrifugation and resuspended in R buffer (20 mM Tris-HCl [pH 8.0], 1 mM EDTA, 1 mM dithiothreitol [DTT], 10% glycerol) containing 300 mM NaCl. The cells were disrupted by sonication, and the lysate was clarified by ultracentrifugation. Proteins in the lysate were precipitated by ammonium sulfate fractionation at 40% saturation and centrifuged 35,000 × g for 30 min. The pellet was resuspended in P buffer (20 mM potassium phosphate [pH 7.5], 0.5 mM EDTA, 10% glycerol, and 0.5 mM DTT) containing 300 mM KCl and diluted by P buffer to a final concentration of KCl 50 mM before being subjected to SP Sepharose (GE Healthcare) chromatography. The pass-through fraction was collected and directly subjected to Q Sepharose (GE Healthcare) chromatography. The proteins were eluted with a linear gradient of 50 mM to 800 mM KCl in P buffer. Rhp51 was eluted at approximately 500 mM KCl. The peak fractions were diluted 5-fold with P buffer and loaded onto a HiTrap Heparin column (GE Healthcare). Rhp51 was eluted at approximately 400 mM KCl in a linear gradient of 100 mM to 700 mM KCl in P buffer. The peak fractions were diluted 4-fold with P buffer and loaded onto a Resource Q column (GE Healthcare). Rhp51 was then eluted at approximately 500 mM of KCl in a linear gradient of 100 mM to 600 mM KCl in P buffer. The Rhp51 preparation obtained by this procedure is indistinguishable from that obtained by a previously described method [22].
An rhp51 derivative with a mutated Walker B box (rhp51 D244N) was constructed by site-directed mutagenesis using a QuickChange kit (Stratagene). The sequences of the primer set are 5′-CATTGTTAGTTGTCaATAGTTGTACTGCC-3′ and 5′-GGCAGTACAACTATtGACAACTAACAATG-3′, where the lowercase letters are mutations that convert D to N. The mutant gene was subcloned into pET11b and expressed in the same manner as wild-type rhp51, and rhp51 D244N was purified as described above. The chromatographic elution patterns of Rhp51D244N were the same as for wild-type Rhp51.
S. pombe Rad22 was also expressed in E. coli BL21-CodonPlus(DE3)-RIPL carrying a Rad22 expression plasmid derivative of pET11b (Novagen). Rad22 expression was induced by 0.2 mM IPTG at 30 °C for 3 h. The induced cell lysate was processed as described above, and the clarified lysate in R buffer containing 500 mM NaCl was precipitated by ammonium sulfate (30% saturation). The pellet was resuspended in R buffer containing 200 mM NaCl and directly loaded onto a Q Sepharose column, and Rad22 was eluted in one step with 600 mM NaCl in R buffer. Rad22 fractions were collected, diluted 3-fold with R buffer and loaded onto a HiTrap Heparin column. Rad22 was eluted at 500 mM NaCl with a linear gradient of 200 mM to 600 mM NaCl in R buffer. Peak fractions were collected and diluted 5-fold with R buffer and loaded onto a HiTrap SP column (GE Healthcare). A linear gradient of 100 mM to 600 mM NaCl in R buffer allowed elution of Rad22 at approximately 300 mM NaCl. Rad22 fractions were applied to a Superdex 16/60 200 pg column (GE Healthcare) and developed in R buffer containing 1 M NaCl. Rad22 was eluted in the void fractions and dialyzed against R buffer containing 100 mM NaCl. The dialyzed sample was applied to a Resource Q column. A linear gradient of 100 mM to 700 mM NaCl in R buffer allowed elution of Rad22 at approximately 250 mM NaCl.
Protein concentrations were determined by measuring absorbance at 280 nm. The following extinction coefficients (ε280) were used: 1.86 × 104 M−1 cm−1 for Rhp51 and Rhp51D244N, 2.93 × 104 M−1 cm−1 for Rad22, 1.44 × 104 M−1 cm−1 for the Swi5-Sfr1 complex, and 9.89 × 104 M−1 cm−1 for RPA.
The purification of S. pombe RPA and the Swi5-Sfr1 complex was previously described [22]. E. coli SSB was purchased from Sigma-Aldrich.
Procedures for the standard reaction protocols were essentially the same as previously described [22], with the exception of Rad22 addition. Briefly, the reactions (10 μl) contained the following components: 10 μM øX174 ssDNA (css), 10 μM ApaLI-linearized øX174 dsDNA (lds) (New England Biolabs), 5 μM Rhp51, 0.5 μM Rad22, 0.5 μM Swi5-Sfr1, 1 μM RPA, 2 mM ATP, and an ATP regeneration system (8 mM creatine phosphate and 8 U/ml creatine kinase) in buffer F (25 mM Tris-OAc [pH 7.5], 1 mM DTT, 5% glycerol, 3 mM Mg(OAc)2, 100 mM KCl). When replacing RPA, SSB was used at 2 μM. Reactions were incubated for 120 min at 37 °C and terminated by adding 1.2 μl of a stop solution containing 8% SDS and 0.6 μl 20 mg/ml Proteinase K, with a final incubation for 30 min at 37 °C. The products were analyzed by 1% agarose gel electrophoresis as previously described [22].
Immobilized øX174 ssDNA beads (css beads) were prepared by annealing a 5′-biotinylated 100-mer oligonucleotide to øX174 ssDNA and capturing the fragments with Dynabeads M-280 Streptavidin (Invitrogen), as previously described [22,46]. To determine the amount of ssDNA immobilized on the beads, an aliquot of the css-beads suspension was denatured by 0.1 M NaOH, and the concentration of the released ssDNA was determined by measuring at A260. About 80% of css was immobilized on the beads. In a standard assay, a bead suspension (2 μl) containing 33 ng of css was mixed with the indicated amounts of each protein in the presence or absence of nucleotide in 10 μl of buffer F containing 0.01% (v/v) NP-40 for 30 min at 37 °C, with constant tapping. The beads were captured with a Magnet Stand Dynal MPC (Invitrogen), and the supernatants (the unbound fraction) and beads (bound fraction) were separated. The bead-bound proteins were eluted with 15 μl of SDS-PAGE loading buffer, and 12 μl of the eluates was analyzed by SDS-PAGE. A 5-fold concentration of SDS loading buffer (3 μl) was added to the supernatants, and 12 μl of each sample was analyzed by SDS-PAGE. The gels were stained with BioSafe CBB G-250 (Bio-Rad), gel images were captured by LAS-4000 (Fuji Photo Film), and protein band densities were quantified with Multi Gauge (Fuji Photo Film) to determine the amounts of bound and unbound proteins.
The procedures were conducted essentially as previously described [22]. Reaction mixtures (13.5 μl) contained 5 μM Rhp51 in buffer F. In some assays, 1 μM RPA, 0.5 μM Swi5-Sfr1, 0.5 μM Rad22, 10 μM øX174 ssDNA, or 10μM ApaL1-linearized øX174 dsDNA were added, as indicated. The reactions were started by adding 1.5 μl of a mixture of [γ-32P]ATP and cold ATP (final concentration, 2 mM) at 37 °C. Aliquots (2 μl) were taken at various time points and immediately mixed with 4 μl of stop solution (0.5 M EDTA). Samples (1 μl) were subjected to thin layer chromatography, as previously described [22]. The amounts of 32Pi and [γ-32P]ATP in each spot were determined using a phosphoimager (Fuji BAS2500).
|
10.1371/journal.pntd.0004384 | Mitochondrial Genome Sequence of the Scabies Mite Provides Insight into the Genetic Diversity of Individual Scabies Infections | The scabies mite, Sarcoptes scabiei, is an obligate parasite of the skin that infects humans and other animal species, causing scabies, a contagious disease characterized by extreme itching. Scabies infections are a major health problem, particularly in remote Indigenous communities in Australia, where co-infection of epidermal scabies lesions by Group A Streptococci or Staphylococcus aureus is thought to be responsible for the high rate of rheumatic heart disease and chronic kidney disease. We collected and separately sequenced mite DNA from several pools of thousands of whole mites from a porcine model of scabies (S. scabiei var. suis) and two human patients (S. scabiei var. hominis) living in different regions of northern Australia. Our sequencing samples the mite and its metagenome, including the mite gut flora and the wound micro-environment. Here, we describe the mitochondrial genome of the scabies mite. We developed a new de novo assembly pipeline based on a bait-and-reassemble strategy, which produced a 14 kilobase mitochondrial genome sequence assembly. We also annotated 35 genes and have compared these to other Acari mites. We identified single nucleotide polymorphisms (SNPs) and used these to infer the presence of six haplogroups in our samples, Remarkably, these fall into two closely-related clades with one clade including both human and pig varieties. This supports earlier findings that only limited genetic differences may separate some human and animal varieties, and raises the possibility of cross-host infections. Finally, we used these mitochondrial haplotypes to show that the genetic diversity of individual infections is typically small with 1–3 distinct haplotypes per infestation.
| The scabies mite is a skin parasite that infects humans and other animal species, causing scabies, a contagious disease characterized by extreme itching. Scabies infections are a major health problem in developing countries and in indigenous Australian populations, where scabies is associated with pyoderma (skin sores) and linked to rheumatic fever and rheumatic heart disease. Little is known about the genetics of the scabies mite. We have assembled the mitochondrial genome of scabies mites obtained from human patients in Australia and from a pig model. While investigating the genetic diversity of each infestation, we found that mitochrondial genomes clustered into two broad clades and showed limited genetic diversity within each infestation. Remarkably, one closely related clade included both human and pig mites, suggesting that mite transmission from pig to human may be possible. This could have major implications in the management of porcine mange and human scabies.
| The scabies mite is an ectoparasitic arachnid that causes an itchy skin infection, known as scabies. Each year around 300 million people worldwide are affected by scabies [1]. Scabies is responsible for a significant disease burden in affected populations through its obligate parasitic lifecycle, which facilitates secondary infections by other pathogens. A severe, but more rare form of scabies, known as crusted scabies, is characterised by hyper-infestation. It generally occurs in immune-compromised individuals [2], although it can occur in patients with no overt immunological deficiency [3]. Cases of crusted scabies can play a significant role in transmission [4]. The mite also infects more than a hundred species of mammals, creating an animal welfare and economic burden in primary industry [5–7].
Scabies represents a major health problem in many remote Indigenous communities in Australia and particularly affects children. Up to 25% of adults and 50% of children acquire scabies infections each year, and 7 out of 10 children under 1 year contract scabies with first presentation peaking at 2 months of age [4, 8]. Scabies is associated with pyoderma (skin sores) in Indigenous communities, but also in many other circumstances of disadvantage globally [9]. In tropical regions, the major pathogens of pyoderma are Group A streptococcus (Streptococcus pyogenes; GAS) and Staphylococcus aureus; with S. pyogenes considered the dominant and usually primary pathogen [4, 10, 11]. This is also the case for remote Indigenous communities in northern and central Australia [4, 10–12]. Sequelae of infection with GAS include acute post-streptococcal glomerulonephritis, which can be clustered (epidemic) or sporadic, and acute rheumatic fever [4, 13, 14]. Rheumatic heart disease, characterised by heart valve damage, occurs as a consequence of acute rheumatic fever and repeated episodes of it can result in cumulative heart valve damage, with consequent heart failure and death [5, 13]. The prevalence of rheumatic heart disease in Indigenous communities is amongst the highest in the world [4, 15]. Several recent studies provide molecular evidence of the scabies mite itself promoting streptococcal growth in pyoderma through complement inhibitors [1, 16, 17]. Association of scabies with these long-term health problems makes it an important factor to consider in Indigenous health. Despite this, there is a relative paucity of genetic and genomic information on scabies [1].
The scabies mite belongs to the subclass Acari (Arthropoda: Chelicerata: Arachnida: Acari), which contains around 48,000 species of ticks and mites [18], and order Sarcoptiformes. Currently, the scabies mite is classified taxonomically as a single species with different varieties based on host specificity [19]. Reportedly, cross infectivity is rare and when it happens it is generally of temporary nature and self-limiting infestation [19, 20]. However, evidence regarding host specificity is conflicting, with some studies suggesting cross infectivity is possible between certain animal and human varieties [21, 22], while others suggest mono-specificity of scabies mite varieties [4, 5, 23–26]. For example, genetic and phylogenetic studies that used hypervariable satellite markers, 16S rRNA and cytochrome oxidase subunit I of mitochondria (cox1), have shown that human and dog mites are genetically distinct in north Australian sympatric populations and that gene flow between the scabies mite population is extremely rare [19, 27]; while another study, using the cox1 gene showed that dog mites from China, USA and Australia are genetically similar to with certain human mites from Australia [21, 22]. Choice of the genetic marker may play a role in conflicting outcomes. Additionally, it has been postulated that each variety, when infesting a non-native host, can acquire morphological and innate characteristics suited to the host through selection if it is allowed to persist due to immunodeficiency or malnutrition of the host [20].
Host specificity has been an area of ongoing investigation because potential cross infectivity, in particular between domestic and companion animals and humans, would have important implications for disease control programs. There are no morphological differences between mites from different host species [20], but multiple failed experimental cross infestation attempts indicated that physiological differences may exist [1, 27, 28]. Mites from dogs have successfully established long term infestations in rabbits [28]. Microsatellite studies in wild animal mite populations also indicated a limited gene flow between mites from sympatric host populations [29], but a further study suggested a prey to predator transfer of mites may be possible [30]. Taken together, available data suggests that a limited gene flow occurs between host-associated populations of scabies mites, however a strict ‘host taxon’ law cannot be assumed.
Here, we describe the in silico isolation, de novo assembly and analysis of the mitochondrial genome of scabies mites from a variety of complex metagenomic samples. The samples consisted of thousands of whole mites collected from two clinical isolates from different regions of Australia and replicates from a laboratory porcine model of scabies [31]. We assembled the mitochondrial genome by applying a bespoke, iterative bait-and-assemble strategy to the massively parallel sequencing data from each mixture—demonstrating the utility of the general approach on complex metagenomics mixtures. Additionally, we identified 6 mitochondrial haplotype clusters in the human (Sarcoptes scabiei var. hominis) and pig (Sarcoptes scabiei var. suis) scabies/mange mite populations sampled. This allowed us to examine the genetic diversity within and between isolates and suggests an extremely low level of diversity overall. To the best of our knowledge, these findings provide the first view of the genetic diversity of individual scabies infestations (intra-host diversity) based on whole mitochondrial genome sequences.
The collection of human patient samples was approved by the Human Research Ethics Committee of the Northern Territory Department of Health and Menzies School of Health Research (approval 13–2027) and informed consent was obtained in writing from each participant. Animal care and handling procedures used in this study followed the Animal Care and Protection Act, in compliance with the Australian code of practice for the care and use of animals for scientific purposes, outlined by the Australian National Health and Medical Research Council. The study was approved by the QASP and the QIMR Berghofer MRI Animal Ethics Committees (DEEDIAEC SA2012/02/381, QIMR A0306-621M).
Skin scrapings were collected from two unrelated patients with severe crusted scabies (patient A and patient B) from different regional areas of Northern Territory, Australia. The collections from individual patients were made 14 months apart and were likely independent. On both occasions, scabies mites (S. scabiei var. hominis) were individually picked from the skin. Each sample contained >1000 mites. Two samples of pig mites (S. scabiei var. suis) were collected from an inbred population of mites from a pig model [31]. The samples were taken from different pigs from consecutive cohorts (where infections are passed on to new piglets from a previous group) at different time points. The first sample consisted of >1000 whole adult mites (pig unwashed). The second sample also consisted of >1000 mites, but was split into three subsamples, which were washed using different protocols to reduce the amount of bacteria present on the surface of the mites due to the wound micro-environment (pig washed 1, 2 and 3 respectively). The protocol entailed: (1) 15 min wash at room temperature in 4% paraformaldehyde in water [32]; (2) 1 hour incubation at 37°C in 150 mM NaCl, 10 mM EDTA, pH8.0, 0.6% SDS, and 0.125 ug/ul lysozyme [adapted from 33] and; (3) 1 hour incubation at 37°C in 1% bleach (Sodium hypochlorite) in water. Mites were subsequently rinsed twice in water. Between wash steps mites were centrifuged at 10000 rpm for 2 min.
Whole mites were crushed and DNA was extracted from each sample using a Blood and Cell culture DNA Kit QIAGEN and a modified procedure, adapted from the manufacturer’s protocol. Washed mites were submerged in 1 ml of ice-cold lysis buffer (20 mM EDTA, 100 mM NaCL, 1% TritonX-100, 500 mM Guanidine-HCl, 10 mM Tris pH7.9) and homogenized with stainless steel beads of 2.8mm diameter at 6800rpm, 3 cycles, 30 sec per cycle, 30 sec between cycles. The suspension of lysed mites was supplemented with DNase free RNase A to 0.2 mg/ml and with Proteinase K to 0.8 mg/ml and incubated at 50°C for 1.5 h. After centrifugation at 4000g for 10 min to pellet insoluble debris the genomic DNA was isolated on the QIAGEN genomic tip as instructed in the manufacturer’s protocol. Six DNA libraries were constructed and 100 nucleotide (nt) long paired-end reads were generated on an Illumina HiSeq 2500. Additionally, 54 S. scabiei var. suis eggs were collected from the pig model. To reduce any surface bacteria, these were washed washed twice in 4% paraformaldyde and rinsed twice in water. DNA was extracted separately from 51 individual eggs, and from 3 pools of 5 eggs and 1 pool of 16 eggs. Sequencing libraries were constructed using the Nugen Ovation SP Ultralow DNA kit for 3 single eggs, and pools of 5 and 16 eggs, and sequenced using an Illumina HiSeq 2500 with 2x100 nt reads. Raw data is available via ENA accession PRJEB12428.
Sequence read quality was assessed using FASTQC [34]. Adapter and quality (Q≥20) trimming was performed using TrimGalore! (v0.3.1) [35]. For the human and unwashed pig samples, reads were de novo assembled using Velvet (v1.2.08) [36]. To establish the best k-mer size, we initially assembled all reads from patient B using k-mer values of k = 61, 63, 65, 67, 69, 71, 73, 75, 79, 85, 89 and 95. Based on the quality of these assemblies, we used k = 69, 75, 77, 79, 81, 83, 85, 89 and 95 for patient A and the unwashed pig samples.
Metagenomic profiling of the patient B mite assembly was carried out using PhymmBL (v4.0) [37]. We augmented the bundled PhymmBL model database with interpolated Markov models trained on the spider mite, Tetranychus urticae (strain London) (available at https://bioinformatics.psb.ugent.be/gdb/tetranychus/) [38] as a proxy for scabies mite. To estimate the species abundance, reads were aligned back to contigs using Bowtie2 (v2.2.3) in local alignment mode [39] and the number of reads aligning to each contig were counted.
For the two human samples and the unwashed pig sample, the mitochondrial genomes were assembled individually using a bait-and-reassemble strategy (see Fig 1 for overview). First, contigs from the whole genome assemblies were aligned to European house dust mite (EHDM), Dermatophagoides pteronyssinus mitochondrial reference genome (NCBI GenBank: EU884425) [40] using LASTZ (v1.02.00) [41] with default settings. For patient B, contigs from the k = 65 assembly were used for alignment; for patient A, contigs from k = 77, 79 and 89 were used; and for the unwashed pig sample, contigs from the k = 69 and 81 assemblies were used. These different k-mer value assemblies varied in their N50 values, largest contigs sizes and median read coverage. The aligned contigs were then filtered against the National Center for Biotechnology Information (NCBI) NT database for scabies mite mitochondrial contigs using BLASTN (v2.2.29) [42] to remove likely false positives due to homology with other mitochondrial genomes (e.g. host mitochondrial genome). After the filtering step, contigs from patient B k = 65 assembly gave the best coverage of EHDM mitochondrial reference genome. Adapter and quality trimmed reads from the patient B sample were then aligned back to the filtered patient B k = 65 contigs using Bowtie2 (v2.2.3) in local alignment mode. Aligned reads were then used to de novo assemble the patient B scabies mite mitochondrial genome using Velvet (v1.2.08). Assemblies were run using k = 69, 79, 87, 89, 91 and 95. A k-mer size of 91 gave the longest single contig and was selected as the reference mitochondrial genome of scabies mite.
To generate the mitochondrial reference for patient A, adapter and quality trimmed reads from the patient A sample were aligned to the patient B mitochondrial reference genome, and matching reads were de novo assembled using Velvet. The final contigs were scaffolded and gaps filled with Ns. This scaffold was then used to realign the patient A reads and reassemble. This process was run iteratively. The second iteration gave the minimal number of Ns in the contig and this was retained as the patient A mite mitochondrial genome.
The same process was used for the unwashed pig sample, but in this case the first iteration gave the best contig with minimal Ns.
Mitochondrial protein-coding genes were predicted on patient B mitochondrial reference genome using the MITOS pipeline (accessed online, August 2014) [43] using the invertebrate mitochondrial genetic code (with default settings). To verify the annotations, open reading frames (ORFs) between stop to stop codons were extracted from the assembled patient B mite genome using the GETORF program from the EMBOSS package (v6.6.0) [44] using the invertebrate mitochondrial codon table. The extracted 774 ORFs were then used to search the NCBI NR protein database using BLASTP (v2.2.29) (E-value threshold = 0.1). For further verification, 13 protein-coding genes from EHDM were aligned to the patient B mite reference genome using TBLASTN (v2.2.29).
12S and 16S ribosomal RNA were annotated using covariance models for those genes using Infernal (v1.1rc4) [45, 46]. The covariance models were built from multiple sequence alignments of 12S and 16S rRNA genes from three sarcoptiformes, Steganacarus magnus (NCBI RefSeq: NC_011574), Dermatophagoides pteronyssinus (NCBI GenBank: EU884425) and Dermatophagoides farinae (NCBI RefSeq: NC_013184) using Clustal-omega (run on the web version on 15/12/2013) [47]. MITFI (within the MITOS pipeline) was used to identify 21 of 22 standard mitochondrial tRNA genes.
For each of the six samples, reads from each sample were aligned to the patient B reference genome using the Bowtie2 aligner in local mode. Pileups were generated using SAMtools (v0.1.19-44428cd) [48] mpileup. Varscan (v2.3.6) [49] was then used to call SNPs with the mpileup2snp command (Min Coverage: 300; Min Variant Frequency: 0.01; otherwise default parameters). SNPs called within the 100 nt boundary at both ends of the genome (positions 1–100 and 13,820–13,919) were manually filtered out due to lack of reliability of alignments at those low complexity regions. For each sample, SNP frequency was estimated using the ratio of reads supporting the variant to total read depth at the SNP.
Sequencing of the two human and four pig samples (unwashed and 3 washed technical replicates) generated 46 (patient B) to 62 (pig washed 1) million read pairs per sample (S1 Table). As expected, the libraries also contained host and microbial DNA. We estimated the level of contamination using two approaches. PhymmBL was applied to the patient B mite mitochondrial assembly to classify contigs into taxa. This revealed that 44% of contigs were from bacteria or other non-arachnid species (S1A Fig). Species abundance was estimated by re-aligning the reads back to the contigs. This revealed that 6% of the reads were derived from contaminants (S1B Fig).
Using our bait-and-assemble strategy, we assembled the mitochondrial genome of each sample (ENA accessions: LN874268-LN874270). The patient B mitochondrial genome assembly comprised a single contig of length 13,919 nt. The patient A and unwashed pig assemblies were 13,902 and 14,044 nt respectively. Both also consisted of single contigs, but each contained a contiguous gap of 118 (patient A) to 255 (unwashed pig) Ns. We were unable to generate a circular mitochondrial genome sequence and found the contig ends were composed of repeat rich sequences. Assuming the scabies mitochondrial genome is circular, we estimate that an AT-rich gap of approximately 300 nt in length is present in the human mite reference assembly. Similar gaps are also present in the other assemblies.
Similar to other sequenced Acari mitochondrial genomes, the scabies mite mitochondrial genome is also highly AT-rich with a 19.32% GC-content. The strand bias is characterized by GC and AT skews, calculated using (G%-C%)/(G%+C%) and (A%-T%)/(A%+T%) respectively. The GC skew for the leading strand is -0.0327 and AT skew is 0.0341. Negative GC skew and positive AT skew are similar to standard metazoan genome strand biases including Acari. Reversal in GC skew in terms of strands has been observed in only two mites so far, Varroa destructor [40, 50] and house dust mite [40].
Thirty-five predicted genes were identified in the scabies mite (var. hominis) mitochondrial genome (Fig 2). These include 13 protein coding genes, two ribosomal RNA genes, and 20 tRNA genes. All of the expected mitochondrial genes for a typical metazoan mitochondrial genome were identified, except for two tRNAs (Alanine and Tyrosine). Search with BLASTP of extracted stop-to-stop ORFs from the mitochondrial genome also verified 12 of the 13 protein-coding genes (except the ATP8 gene) and TBLASTN alignment of 13 protein coding genes from the EHDM mitochondrial genome to the scabies mite mitochondrial genome also confirmed 12 of the 13 protein coding genes (except the ND4L gene) in the scabies mite mitochondrial genome. Two tRNA genes identified by tRNAscan-SE were considered ambiguous due to their overlap with other genes (A is overlapping with C and Y is overlapping with NAD4L). However, the overlap is not on the same strand.
The gene order and strand specification of the protein coding and ribosomal RNA genes are the same as that of EHDM. All identified tRNA genes, except tRNA-C and tRNA-V also maintain the same gene order and strand as EHDM. The tRNA genes that are not syntenic with EHDM (tRNA-C, tRNA-V) are also on the opposite strand to that of EHDM.
To identify genetic polymorphisms in the sequenced mite pools, reads were mapped back to the reference Mt genome (patient B). The average depth of coverage was 2914 across samples (average per sample ranging 1698–4299). A total of 665 single nucleotide polymorphisms (SNPs) were identified across all samples relative to the reference Mt genome (patient B assembly): 601 SNPs in the patient A sample, 598 SNPs in the patient B sample, and 102 SNPs in the unwashed pig sample, while the washed pig samples (w1, w2, w3) contained 102, 100, 102 SNPs respectively (S2 Table). The four pig mite samples are effectively replicates (biological and technical) and were highly concordant.
Within each sample, SNP allele frequencies, estimated from the ratio of reads supporting the variant to the total coverage, were tightly grouped into a small number of clusters (Fig 3A), suggestive of the presence of just a few mitochondrial haplotypes in each sample. To estimate the number and frequencies of haplotypes, and to infer their sequences, we used k-means clustering (S3 Table). SNPs common to all haplotypes within a sample have a frequency of 1. Additional haplotypes are defined by the presence of extra SNPs with mean frequency less than 1. We also examined reads supporting SNPs located close to each other (<100 nt) for evidence that SNPs were on the same haplotype (supported by the same reads) or distinct haplotypes (never co-occurring on one read). Only 4 pairs of SNP were located within 100 nucleotides of each other and in two different frequency clusters. In each sample, we gave the haplotype with the highest frequency the prefix H1. Haplotypes with successively lower frequencies are labeled H2 and H3, in that order.
In the patient B sample, a single clear cluster in the SNP frequencies was observed (Fig 3A). This implies the presence of two haplotypes. The dominant haplotype, H1_B_REF, has an estimated frequency of 0.98 and corresponds to the consensus obtained from de novo assembly. The second haplotype (H2_B) is defined by the presence of 598 SNPs and has an estimated frequency of 0.02.
The patient A sample contains a single clear cluster, containing 593 SNPs with a frequency of 1.0. An additional 8 SNPs do not cluster well with the dominant group. These fall into four other clusters identified by k-means. We chose to ignore the three smallest clusters, resulting in just two closely-related haplotypes: H1_A, which has a frequency of 0.98, and H2_A, which has a frequency of 0.02 and is separated from H1_A by just 4 SNPs. The lack of replication of these SNPs makes the significance of these closely related haplotypes unclear.
All four pig samples contain three clear clusters, forming three haplotypes in each sample. The haplotypes with the highest frequencies (H1_u, H1_w1, H1_w2, H1_w3) have estimated frequencies of 0.89, 0.91, 0.89 and 0.90 respectively. Additional haplotypes are composed of combinations of the clusters (S3 Table) and are concordant between samples. The second haplotypes have estimated frequencies of 0.10, 0.07, 0.09 and 0.09 for the unwashed and three washed samples respectively and we label these H2_u, H2_w1, H2_w2, and H2_w3. The third haplotypes, which we label H3_u, H3_w1, H3_w2 and H3_w3 for the unwashed and three washed samples respectively, have estimated frequencies 0.02, 0.01, 0.02 and 0.01 respectively. The H1_u, H1_w1, H1_w2, H1_w3 haplotypes have 82, 82, 81, 81 SNPs; H2_u, H2_w1, H2_w2, H2_w3 have 82, 82, 81, 81 SNPs and H3_u, H3_w1, H3_w2, H3_w3 have 87, 87, 85, 87 SNPs respectively.
Similar analyses of DNA sequencing data from a limited number of individual scabies mite eggs or small pools did not identify clusters in the SNP allele frequencies (S2 Fig), suggesting that multiple haplotypes provide evidence for genetic diversity rather than heteroplasmy.
To understand the relationship between the 16 inferred haplotypes, we constructed a phylogenetic tree based on the SNPs present in each haplotype sequence using MEGA5 (v5.2.2) [51] with a distance measure based on the number of differences between haplotype sequences (S4 Table). The tree shows that the haplotypes fall into two broad clades and six haplogroups (Fig 3B). Three haplogroups are found in clinical isolates (human mite haplogroups 1–3), while the pig mites comprise three haplogroups (pig mite haplogroups 1–3). The average difference between haplotypes within each of the pig mite haplogroups is 1–2 SNPs. The two haplotypes in human mite haplogroup 3 are almost identical and very similar to human mite haplogroup 2, forming one of the clades, while human mite haplogroup 1 is distinct from the other human mite haplogroups, and more similar to the pig mite haplogroups, forming the second clade.
We de novo assembled the mitochondrial genome of the scabies mite using massively parallel sequencing data from thousands of pooled whole mites obtained from two clinical isolates from different parts of northern Australia and from a laboratory pig model. Our approach was to initially perform a full metagenomic de novo assembly. As expected, the samples were contaminated by bacterial reads presumably from the scabies mite gut and the scabies lesion micro-environment. We then iteratively selected Mt contigs and used these as bait to recruit and assemble genuine mite reads. Our bespoke de novo assembly approach has some similarity to an existing bait-and-assembly method called MITObim [52]. MITObim first directly baits the short reads using a closely related mitochondrial genome, then iteratively maps reads and performs contig extension using MIRA. De novo assembly is also supported, but only for “well behaved” data. In contrast, our approach performs full de novo assembly of the metagenomic mixture, baiting of contigs using the house dust mite genome and filtering of host mitochondrial contigs, then 1–2 iterations of read alignment and de novo assembly using velvet. Our results substantially extend the limited case studies and simulations used to validate MITObim and demonstrate that the general approach of mitochondrial genome bait-and-assembly works on real examples that are highly complex metagenomic mixtures involving both genetic heterogeneity and host/bacterial contamination.
Mitochondrial genomes are generally circular, but rare occurrences of linear mitochondrial genomes have been reported [53]. While we were unable to circularise the scabies mite mitochondrial genome assemblies, it seems unlikely that it is linear; a simpler explanation is that the region contains highly repetitive AT-rich sequences (which we observe on the flanking regions of the break) and these are difficult to map reads to and assemble across. The same region in EHDM mitochondrial genome contains the origin of replication site. This might also be a factor for the missing coverage in this region.
The gene content and organisation of the Sarcoptes scabiei var. hominis mitochondrial genome is the same as the recently published var. canis mitochondrial genome contig (NCBI GenBank: CM003133 JXLN01000000) [54]. The gene order is also similar to the rabbit ear mite, Psoroptes cuniculi, mitochondrial genome (NCBI Refseq: NC_024675) [55] with the exception of the two additional tRNA genes found in the latter (Fig 4) and the house dust mite (NCBI GenBank: EU884425), suggesting a close evolutionary relationship between these species. These observations further support the hypothesized close evolutionary relationship between the parasitic scabies mite and free-living house dust mite [56, 57]. However, it is distinct from most other Acari mitochondrial genomes sequenced to date [40], including Tetranychus urticae (NCBI GenBank: EU345430). It is also quite distinct from Limulus polyphemus (NCBI GenBank: NC_003057), which represents the arthropod ground pattern of gene arrangement for chelicerates [58].
We found variations in the diversity of infestations between individuals (clonal or heterogeneous). Mites from skin scrapings from one patient with crusted scabies (patient A) appeared to be essentially clonal. This lack of diversity suggests that either a single female mite may initiate infection, or selection by the host immune system may operate. Mites from the second patient (patient B), who had more severe crusted scabies than patient A, showed two highly divergent haplotypes. Remarkably, one of the haplotypes in the patient B sample was much more similar to the pig mite haplotypes than to the other human mite haplotypes. This may reflect an actual cross-species infection, highlight the possibility of this, or could merely suggest that the genetic differences between varieties are limited.
The population-level genetic diversity of scabies mites has been studied extensively using single or small numbers of gene sequences. One recent study based on cox1 gene sequences showed that scabies infestations of humans, dogs and other animals fall into 3 clades [21]. Two clades contained only geographically isolated var. hominis mites, while the third contained closely related human and animal mites. Another recent study, which used 3 gene sequences, reported 5 clades—4 distinct var. hominis clades (one for each geographical region studied) and one clade containing closely-related var. hominis and other animal scabies mite varieties [21, 22]. These findings are consistent with our observation that one of the patient B mite haplotypes was more similar to the pig mite haplotypes than to the other human mite haplotypes. Since we used the entire mitochondrial genome, we expect our results to be robust. Moreover, our intra-sample inference of haplotypes has revealed that patients and animals can harbour multiple genetically divergent mites, or near clonal infestations. To the best of our knowledge, this is the first view of intra-host genetic diversity for scabies mites.
One issue with our analysis of haplotypes is that we are unable to utilize a few unclustered and unreplicated SNPs, in particular, from the patient B sample. These may define additional haplotypes or represent genetic drift. Notwithstanding, these haplotypes would be extremely closely related. Our justification is that these are very small differences between haplotypes and may be due to sequencing errors and thus erroneously detected variants, whereas large numbers of SNPs with clustered frequency are more likely to be real. Regardless, our analysis provides a first look at the genetic diversity of scabies infections within patients. However, further study with broader population sample is required to refine the intra- and inter-host population diversity structure.
Scabies is responsible for significant morbidity in Indigenous Australians in many remote communities of northern and central Australia. Effective control and prevention of scabies epidemics in those communities is of paramount importance as scabies has a long-term effect on the life expectancy and quality. We have sequenced, assembled and annotated the mitochondrial genome of the scabies mite using a bespoke bait-and-assembly approach; we identified SNPs in multiple isolates from patients and a laboratory pig model, and inferred the haplotype structure and diversity of individual infections. We used these tools to investigate the genetic diversity within individual infestations. Larger samples are now needed.
The development of genomics resources for studying the scabies mite will accelerate research into this parasite, just as genome sequences have for other neglected parasitic diseases. For example, in malaria genomic resources provided means to identify drug resistance causing mutations [59] or in schistosomiasis it helped suggesting new approaches to preventions and strategies for control [60]. The scabies mite mitochondrial genome sequence will facilitate further population genetics research in this area.
|
10.1371/journal.pntd.0002170 | Phylogeographical Studies of Ascaris spp. Based on Ribosomal and Mitochondrial DNA Sequences | The taxonomic distinctiveness of Ascaris lumbricoides and A. suum, two of the world's most significant nematodes, still represents a much-debated scientific issue. Previous studies have described two different scenarios in transmission patterns, explained by two hypotheses: (1) separated host-specific transmission cycles in highly endemic regions, (2) a single pool of infection shared by humans and pigs in non-endemic regions. Recently, A. suum has been suggested as an important cause of human ascariasis in endemic areas such as China, where cross-infections and hybridization have also been reported. The main aims of the present study were to investigate the molecular epidemiology of human and pig Ascaris from non-endemic regions and, with reference to existing data, to infer the phylogenetic and phylogeographic relationships among the samples.
151 Ascaris worms from pigs and humans were characterized using PCR-RFLP on nuclear ITS rDNA. Representative geographical sub-samples were also analysed by sequencing a portion of the mitochondrial cox1 gene, to infer the extent of variability at population level. Sequence data were compared to GenBank sequences from endemic and non-endemic regions.
No fixed differences between human and pig Ascaris were evident, with the exception of the Slovak population, which displays significant genetic differentiation. The RFLP analysis confirmed pig as a source of human infection in non-endemic regions and as a corridor for the promulgation of hybrid genotypes. Epidemiology and host-affiliation seem not to be relevant in shaping molecular variance. Phylogenetic and phylogeographical analyses described a complex scenario, involving multiple hosts, sporadic contact between forms and an ancestral taxon referable to A. suum.
These results suggest the existence of homogenizing gene flow between the two taxa, which appear to be variants of a single polytypic species. This conclusion has implications on the systematics, transmission and control programs relating to ascariasis.
| Ascaris lumbricoides, the world's most common human nematode, and A. suum, the pig roundworm, are two of the most important soil-transmitted helminthes of public health and socio-economic concern. However, previously documented similarities at the morphological and genetic level, coupled with evidence for hybridization and gene flow, have clouded the taxonomic distinctiveness of these two nematodes. To date, molecular epidemiological studies have been carried out, mostly in highly endemic regions, where two different transmission cycles have been described. Recently, pigs have been recognized as an important source of human ascariasis in China, opening questions about the zoonotic potential and the efficiency of control programs. Here, samples from non-endemic regions have been analysed using a nuclear marker to identify nematodes to species level plus a mitochondrial marker to investigate the phylogeographic relationships among individuals of the two species from both endemic and non-endemic regions. Results obtained suggested that A. suum and A. lumbricoides may be variants of the same species, with the lack of fixed genetic differences and considerable phylogeographic admixture confirming an extremely close evolutionary relationship among these nematodes. This study highlights the need to further explore the evolutionary affinities of the two taxa to help shed light on the epidemiology of ascariasis.
| Ascariasis in pigs and in humans is caused by two of the most socio-economically important nematodes: Ascaris suum Goeze, 1782 and Ascaris lumbricoides Linneaus, 1758, respectively. Human ascariasis is a soil-transmitted helminthiasis (STH), included in the WHO list of neglected tropical diseases (NTD), infecting more than one billion people [1]. Even if the majority of infections are asymptomatic, clinical manifestations of human ascariasis typically involve acute and chronic symptoms (lung inflammation and fever due to larval migration; abdominal pain, nausea, retarded growth in children and intestinal obstruction due to the massive presence of adult worms) [1]. Ascariasis in pigs is frequent in both intensive and extensive breeding systems, being a source of substantial economic losses [2].
Due to their morphological and biological similarities, the taxonomic distinctiveness of A. lumbricoides and A. suum still represents a debated scientific issue. Importantly, this issue is of great relevance for both systematists and epidemiologists alike, given its implications on parasite transmission, zoonotic potential, and the establishment of control programs [3], [4], [5]. Several hypotheses have been proposed to explain the origin of the two ascarid taxa in their respective hosts and their taxonomic status [3], namely: a) A. suum and A. lumbricoides are two valid species; b) A. suum is the ancestor of A. lumbricoides, originated by an allopatric event of host-switching; c) A. lumbricoides is the ancestor of A. suum; d) A. suum and A. lumbricoides are conspecific and therefore occur as variants of a single polytypic species.
Previous molecular epidemiological studies have described two different scenarios in transmission patterns that could be explained by two different hypotheses. First, distinct, host-specific transmission cycles have been observed in highly endemic regions as Guatemala and China [4], [5], [6], [7]. Second, a single pool of infection, shared by humans and pigs, has been observed in non-endemic regions, as Denmark and North America [8], [9]. Conversely, recent results strongly suggest that A. suum acts as an important source of human ascariasis in endemic area such as China, where both Ascaris spp. co-occur. Here, the authors observed cross-infections and hybridization of human and pig Ascaris, thus supporting the second hypothesis on transmission cycles [10].
Considering the uncertain epidemiological picture, the main aim of the present study was to investigate genetic variation in two nuclear and mitochondrial target regions (ITS and cox1, respectively) within and among Ascaris populations of human and pig origin, collected from a range of non-endemic regions. These molecular data, along with other published sequences available at both local and global scales, were then used to infer the evolutionary, phylogenetic and phylogeographic relationships among samples. The nuclear ribosomal marker (ITS) was chosen to distinguish A. suum, A. lumbricoides and the hybrid form of the two taxa. Meanwhile, mitochondrial DNA is the most frequently used molecular marker in this kind of studies, due to desirable biological features such as maternal inheritance, high mutation rate, very low recombination rate, haploidy, and putative selective neutrality, making mtDNA markers particularly suitable as barcoding tools to identify sibling and cryptic species [11], [12].
Studies aimed at investigating the molecular epidemiology of ascariasis are important not only to clarify the transmission patterns of the two roundworms, but also to better quantify the level of gene introgression between host-associated populations [10]. Such knowledge is important, given that introgression often results in the selection of novel genes, the promotion of rapid adaptive diversification, and homogenization across the genomes of the interbreeding populations [13], [14]. Additional sources of information are now available from the recently published draft genome of A. suum [15].
A total of 151 adult nematodes belonging to Ascaris spp. were collected from pig (n = 143) and human (n = 8) hosts. Nematodes collected were repeatedly washed in saline and stored in 70% ethanol. Collection data including collecting sites, hosts, number of parasites specimens analysed and identification codes are summarised in Table 1.
DNA was isolated using the Wizard Genomic DNA purification kit (Promega) according to the manufacturer's protocol.
All samples, from human and animal origin, were obtained from existing collections. Samples from human origin were obtained from existing collections at Tor Vergata and Sant'Andrea Polyclinics in Rome. Data collection includes only the geographical origin of patients and no reference to personal data was recorded, thus guaranteeing the absolute anonymity of these specimens.
Sample collection at the Polyclinics that provided the nematodes from humans was performed in concordance with the WMA Helsinki Declaration (Edinburgh 2000) and its subsequent modification, as well as with the Italian National Law n. 675/1996 on the protection of personal data.
The entire ITS nuclear region (ITS1, 5.8S, ITS2) was amplified using 5.0 µl of template DNA (20–40 ng), 10 mM Tris-HCl (pH 8.3), 1.5 mM MgCl2 (Bioline), 40 mM of a nucleotide mix (Bioline), 50 pmol/µl each of the forward primer NC5 (5′-GTAGGTGAACCTGCGGAAGGATCAT-3′) and the reverse primer NC2 (5′-TTAGTTTCTTCCTCCGCT-3′) described by Zhu et al.[16] and 1.0 U of BIOTAQ DNA Polymerase (Bioline) in a final volume of 50 µl. The PCR was performed in a GenePro Eurocycler Dual Block (Bioer) under the following conditions: 10 min at 95°C (initial denaturation), 30 cycles of 30 sec at 95°C (denaturation), 40 sec at 52°C (annealing) and 75 sec at 72°C (extension), and a final elongation step of 7 min at 72°C. A negative control (without genomic DNA) was included in each set of amplification reactions.
A representative subset of specimens (Table 2) was also analysed by sequencing a portion of the mitochondrial cytochrome oxidase I gene (cox1), after amplification using the forward primer As-Co1F (5′-TTTTTTGGTCATCCTGAGGTTTAT- 3′) and the reverse primer As-Co1R (5′-ACATAATGAAAATGACTAACAAC- 3′), as described by Peng et al. [6], under the following conditions: 5 min at 94°C, followed by 35 cycles of 94°C for 30 s; 45 s at 55°C; 90 s at 72°C, followed by 5 min at 72°C. Aliquots (5 µl) of individual PCR products were separated by electrophoresis using agarose gels (1%), stained with ethidium bromide (0.4 µg/ml) and detected using ultraviolet trans-illumination.
Positive ITS amplicons were digested with the restriction endonuclease HaeIII, as the resulting patterns have been previously proved useful for the identification of human and pig Ascaris species [8]. Digests were resolved by electrophoresis in 2% agarose gels, stained with ethidium bromide (0.4 µg/ml), detected under UV trans-illumination, and the fragments sizes determined by comparison with a 100 bp DNA ladder (Promega). Information on geographical origin, hosts, codes, number of parasites successfully genotyped, and genotypes recovered using PCR-RFLP are available in Table 1.
Positive amplicons were purified by SureClean (Bioline), following the manufacturer's instructions, and then sequenced by MWG Eurofins DNA. Two different datasets were created, each representing different partial cox1 alignments: the first including only samples analysed in the present paper (Dataset1), with the exclusion of two human nematodes due to small sample size (single specimens from Pakistan and Romanian human patients), and the second including all GenBank retrieved sequences of specimens collected from endemic and non-endemic regions (Dataset2). Information about specimens sequenced for cox1, identification codes and accession numbers, also of GenBank retrieved sequences are available in Table 2.
Nucleotide sequences were aligned using Clustal X implemented in MEGA 5 [17] and then analysed using DnaSP v5 [18] to infer haplotype composition. In addition, sequences were analysed using Arlequin 3.11 [19] to estimate several variability indexes: the relative frequencies of haplotypes; population differentiation (FST) among samples for Dataset1; hierarchical analyses of molecular variance (AMOVA) to evaluate the amount of population genetic structure for Dataset2, using information on the allelic content of haplotypes, as well as their frequencies. The significance of the covariance components associated with the different levels of genetic structure (within individuals of populations, among populations and among groups) was tested using non-parametric permutation procedures [20]. The AMOVA was undertaken twice, using two different criteria to define groups and population structure: geographical origin (endemic and non-endemic regions) and host affiliation (pig and human).
Both Dataset1 and 2 were also analysed using a phylogenetic approach based on Bayesian reconstruction method. The program JModeltest [21] was used to compare the fit of nucleotide substitution models using the Akaike Information Criterion (AIC), under a total of 83 models, corresponding to 11 different schemes; the best-fit model and parameters determined for both cox1 datasets were then used for the Bayesian analyses. The Bayesian analyses were performed using the HKY+I model for both datasets (as selected by ModelTest), using BEAST software [22]; datasets were run twice for 106 generations. Posterior probability values (BPP) shown in the Bayesian consensus trees were determined after discarding trees from the burn-in period. For each dataset, burn-in was estimated to include the first 104 generations. A second phylogenetic method was performed only on Dataset 2 using MEGA5 [23]: the evolutionary distances were computed using the Tamura-Nei [24] with Neighbor joining method (NJ) and statistical support at nodes was evaluated using 1000 pseudoreplication bootstrap [25]. Phylogenetic trees included Anisakis Dujardin 1845 as outgroup (GenBank accession number: JN102304).
Moreover, statistic parsimony networks [26] using TCS software [27] were inferred for both datasets in order to determine the phylogeographic distribution and genealogy of the Ascaris specimens analysed, running the network at a 95% connection limit, which is the maximum number of mutational connections between pairs of sequences justified by the parsimony criterion.
A PCR product of around 1000 bp was obtained for 137 of the 151 specimens analysed. Amplicons were subsequently digested using the HaeIII restriction enzyme. This approach yielded the identification of three genetically distinct banding patterns belonging to the genus Ascaris: the “lumbricoides” genotype displays two bands of about 610 bp and 370 bp, the “suum” genotype shows three bands of about 610 bp, 230 bp and 140 bp, and the “hybrid” genotype displays all the four bands mentioned above (Figure 1).
While the proportion of each genotype varied somewhat across the various localities sampled, all regions revealed instances of discordance between the expected genotype and host of origin (Table 1). For Italy, although 49 of 60 positive samples from pigs displayed the expected “suum” genotype, nine displayed the “hybrid” genotype and two displayed the “lumbricoides” genotype. In contrast, neither of the two positive human isolates displayed the expected “lumbricoides” pattern, instead revealing one “suum” and one “hybrid” genotype. Positive samples obtained from nematodes collected in other countries included four specimens from humans and 71 from pigs. Of the human nematodes, three specimens (Syrian, Pakistan and Romanian patients) showed the typical “lumbricoides” genotype and one (another Romanian patient) displayed the “suum” genotype. Among Slovak pigs (n = 44), 36 showed the “suum” genotype, four the “lumbricoides” genotype, and four the “hybrid” pattern, while Hungarian pigs (n = 27) included 19 specimens and eight specimens displaying the “suum” genotype and “hybrid” genotypes, respectively. Overall, the “hybrid” genotype was encountered in specimens from both pig and human hosts, at a frequency of 16%.
A PCR product of around 400 bp was obtained for 62 specimens amplified. The alignments of Dataset1 (62 sequences) and Dataset2 (120 sequences) yielded a usable alignment of 327 bp. Representative sequences for each haplotype recovered in the course of the present study are available in GenBank under the following Accession Numbers: Hap1: KC455923, Hap2: KC455924, Hap3: KC455925, Hap4: KC455926, Hap5: KC455927, Hap6: KC455928, Hap7: KC455929, Hap8: KC455930, Hap9: KC455931, Hap10: KC455932, Hap11: KC455933, Hap12: KC455934, Hap46: KC455935.
Twelve haplotypes were identified in Dataset1 (Hap1-12), with a total haplotype diversity (Hd) of 0.70 (haplotypes recovered were deposited in GenBank, see Table 2 for accession numbers). Five haplotypes were observed in Slovak sample, with Hd = 0.71; three haplotypes were observed in Hungarian sample, with Hd = 0.24 and seven haplotypes were observed in Italian sample, with Hd = 0.62. The most frequent haplotype was Hap5, shared among the Italian (frequency of 61.5%), Hungarian (87.5%) and Slovak samples (5.5%). Hap1 was the most frequent haplotype in the Slovak population (44.4%) and it has been less frequently reported also in Italian specimens (7.7%). Results from FST analysis showed significant differences between Slovak sample and the Italian (0.29) and Hungarian samples (0.49), and little differentiation between Italian and Hungarian samples (0.05). Considering Dataset2, forty-five haplotypes were identified, with Hd = 0.89; Hap5 was observed also in the Chinese pig sample. The Italian and Slovak samples showed haplotype Hap7 in common with endemic (Brazil, Zanzibar and China) and non-endemic regions (Japan); the Italian sample showed also haplotype Hap12 in common with endemic regions. Information about haplotypes recovered in the partial cox1 sequences analyses, haplotype affiliation to phylogenetic clusters A(A1, A2)-B-C, GenBank accession numbers, codes, correspondences to genotypes identified using RFLP approach on ITS, hosts, endemic and non-endemic origin of samples and haplotypes relative frequencies for populations of Dataset1 are available in Table 2.
AMOVA analysis suggested a higher influence of the epidemiological (endemic/non-endemic origin) criterion in modulating the accumulation of variability with respect to host affiliation, even if the percentage of variation at group level was not significant (3.83% and 0.10%; p = 0.38 and 0.61, respectively). Significant values (p≤0.05) were obtained for the variation observed among populations within groups and among individuals within populations in both analyses, but with an opposite trend: percentage of variation within population was higher than among populations of the same group if the endemic/non-endemic criterion is considered as feature to group samples.
Bayesian and NJ phylogenetic analyses, based on Dataset1 and Dataset2, described similar topologies, with three main clusters (Figure 2), analogous to the clusters named A, B and C in Anderson and Jaenike [28] and Snabel et al. [29] studies. Clusters A and B have been recently reported also by Iniguez et al. [30].
Cluster A includes samples from both pigs and humans collected from endemic and non-endemic zones; it showed further slight internal subdivision according to host affiliation and epidemiological features, although no statistical support for this partitioning was found. Sub-cluster A1 contains mainly specimens from pigs and few from humans, collected from non-endemic zones. It is important to underline that the specimens of human origin (ASR_H and ASI12_H) included in this group showed the typical “suum” genotype for PCR-RFLP analysis of the ITS region. Sub-cluster A2 includes mainly specimens from humans collected from endemic areas, except for one human sample (ASI13 corresponding to Hap12) collected from non-endemic regions, although the country origin of the patient is unknown. Cluster B is also characterized by the presence of specimens from both pigs and humans collected from endemic (Brazil, China, Zanzibar, Pakistan) and non-endemic zones (Japan, Italy). Cluster C comprises only specimens from pig collected from non-endemic regions (Italy and Slovakia). It appears to be well separated from clusters A and B that are more closely related to each other. The existence of the three clusters is well supported by very high posterior probability values (BPP ranging from 92 to 98 for Dataset1 and from 90 to 100 for Dataset2); NJ tree bootstrap values show high statistical support for cluster C (93) and lower values for cluster A (51) and B (38), nevertheless the value supporting the distinctiveness of cluster C from A and B together is fairly high (77).
Results obtained from parsimony network analysis on Dataset2 (Figure 3) describes a very complex scenario where the three clusters observed in phylogenetic analysis are recognized and the slight subdivision inside cluster A is still evident. The main haplogroup, where Hap5 is the more frequent and typically associated to A. suum, corresponds to cluster A1 with several haplotypes branching around. The star-like distribution of haplotypes is also evident in the other haplogroups, represented by Hap12 for cluster A2 and Hap7 for cluster A. Cluster A2 is mainly represented by haplotypes from endemic regions, typically associated to A. lumbricoides, with the exception of Italian and Japanese human cases; while cluster B includes both pig and human specimens from endemic and non-endemic regions. The Slovak haplogroup appears completely separated from the other haplotypes. These results confirm the relationships observed in the Bayesian phylogenetic trees.
Human and pig Ascaris spp. are two of the world's most common soil-transmitted parasites and together cause serious health and socio-economic problems. Ascariasis is considered a NTD as it occurs commonly in rural and poor urban areas and promotes poverty due to its high impact on child health and development, pregnancy and worker productivity. Similarities in the morphology and biology of these two nematodes entail ongoing ambiguity concerning their taxonomic status and argue for the need to delve deeper into their comparative molecular epidemiology.
The present paper provides additional information on the molecular epidemiology of ascariasis in non-endemic regions, such as Italy and Eastern Europe. Molecular characterization using a PCR-RFLP approach on a nuclear marker has confirmed that most pig nematodes sampled herein displayed the typical A. suum pattern, corresponding to the genotype G3, while the two human nematodes from endemic regions such as Pakistan and Syria showed the typical A. lumbricoides pattern, corresponding to the genotype G1 [31]. Cross-infection is confirmed in both hosts by instances of A. suum genotypes in human nematodes and A. lumbricoides in pigs. Moreover, a significant percentage of nematodes displaying the “hybrid” pattern, corresponding to the G2 genotype [31], has been observed in both human and pig nematodes, strongly inferring the presence of gene flow between the two taxa. This combined evidence suggests that Ascaris suum can function as a relevant agent of human infection in non-endemic areas. These data are in agreement with recent results described firstly by Betson et al. in Zanzibar [32] and then by Zhou et al. in China [10], where zoonotic transmission of A. suum is suggested to occur also in these endemic areas. The zoonotic potential of A. suum therefore needs to be reevaluated in order to plan more efficient control programs.
Phylogenetic analyses revealed the homology to the clusters previously observed in Anderson and Jaenike [28] and in Snabel et al. [29], confirming that geographical origin plays an important role in structure of cluster A, where endemic and non-endemic samples split in two sub-clades, but not in cluster B, which contains specimens from both epidemiologically classified regions. Finally, significant values on population differentiation analysis and high haplotype diversity confirm the genuine separation of cluster C. As these parameters are important indexes for evaluating genetic diversity and differentiation, further analysis will be required to understand the significance of this pronounced genetic dissimilarity.
Phylogeographic analyses are helpful in understanding population differentiation, species formation and ecological adaptation [33]. Results obtained from the haplotype network analysis have revealed a very complex scenario: the typical A. suum haplotype is the most frequent among samples from non-endemic regions plus is observed also in human patients (circle A1); moreover, this haplogroup is closely related to the haplogroup including the distinctive A. lumbricoides haplotypes found in endemic regions (circle A2), which is related in turn to a mixed group homologous to cluster B obtained in phylogenetic inferences (circle B). The picture described a cross-linked relationships among haplotypes, where no clear geographical or host-affiliation criteria seem to be relevant in shaping haplogroups. Shared haplotypes between pig and human Ascaris spp. could be explained by evolutionary processes such as introgression and/or retention of ancestral polymorphisms, as suggested previously [9], [34]. In addition, molecular variance analysis underlined that accumulation of genetic variability is observed at the individual and population level rather than at the level of groups defined on geography or host-affiliation.
The overall results showed no fixed differences between human and pig Ascaris, describing two taxonomic entities intimately interconnected and therefore likely to experience gene flow. These data strongly infer the absence of a major genetic barrier between the two taxa and therefore suggest that A. suum and A. lumbricoides may be variants of the same species, as suggested by Leles et al. [3] and Liu et al. [35], and more recently by Iniguez et al. [30]. Together all four studies have found no evidence of diagnostic genetic heterogeneity between human and pig Ascaris, plus an absence of genetic clusters discriminating each host.
|
10.1371/journal.pgen.1005768 | Prp4 Kinase Grants the License to Splice: Control of Weak Splice Sites during Spliceosome Activation | The genome of the fission yeast Schizosaccharomyces pombe encodes 17 kinases that are essential for cell growth. These include the cell-cycle regulator Cdc2, as well as several kinases that coordinate cell growth, polarity, and morphogenesis during the cell cycle. In this study, we further characterized another of these essential kinases, Prp4, and showed that the splicing of many introns is dependent on Prp4 kinase activity. For detailed characterization, we chose the genes res1 and ppk8, each of which contains one intron of typical size and position. Splicing of the res1 intron was dependent on Prp4 kinase activity, whereas splicing of the ppk8 intron was not. Extensive mutational analyses of the 5’ splice site of both genes revealed that proper transient interaction with the 5’ end of snRNA U1 governs the dependence of splicing on Prp4 kinase activity. Proper transient interaction between the branch sequence and snRNA U2 was also important. Therefore, the Prp4 kinase is required for recognition and efficient splicing of introns displaying weak exon1/5’ splice sites and weak branch sequences.
| Prp4 is an essential protein kinase that is involved in the splicing of some introns. Using a conditional mutant of Prp4, we showed that a subset of genes, including several cell cycle–regulatory genes, are dependent on Prp4 for splicing. Furthermore, we could convert genes between Prp4-dependent and -independent states by introducing single-nucleotide mutations in the exon1/5’ splice sites and branch sequence of introns. This work shows that Prp4 activity is required for splicing surveillance in a subset of mRNAs.
| Introns are removed from pre-mRNAs by spliceosomes, which are highly dynamic macromolecular complexes consisting of five small nuclear RNAs (snRNAs; U1, U2, U5, and U4/U6) associated with specific proteins in subcomplexes called snRNPs. In vitro, spliceosomal subcomplexes assemble on pre-mRNAs in a time-dependent manner. The intron to be removed is defined during the formation of spliceosomal complex B, which contains the base-paired U4/U6 snRNP as well as the other three snRNPs. ATP-dependent helicases control the specific rearrangement of the B complex so that catalysis of the transesterification reactions can occur in spliceosomal complex C, formed by the departure of snRNPs U1 and U4 and rearrangement of snRNPs U2, U6, and U5, culminating in the splicing reaction [1–3]. Although pre-mRNA splicing is an important part of regulated gene expression, little is known about the assembly and activation of spliceosomes in vivo. Introns are presumably recognized and removed by the spliceosome during or shortly after transcription, suggesting that parts of the spliceosomal complex must be recruited to transcribed chromatin areas, installed at introns, and then activated for catalysis. Several lines of evidence suggest that the 5’ splice site (SS) is defined by direct interactions with snRNA U1, and that definition of the 3’ SS also involves the interaction of the branch sequence (bs) with snRNA U2 [4,5].
In the fission yeast Schizosaccharomyces pombe, approximately 45% of genes contain at least one intron, with some genes containing as many as 15. Compared with the introns in budding yeast Saccharomyces cerevisiae, fission yeast introns are relatively small: the average intron sizes in S. cerevisiae and S. pombe are 256 nt and 83 nt, respectively. In S. cerevisiae, only 5% of genes contain introns, and the 5’ SS and bs at the 3’ region of the intron are highly conserved; in contrast, in S. pombe, the 5’ SS and bs are as variable as the corresponding sequences in mammals [6–8]. Regulated alternative splicing has not been observed in mitotically active fission yeast cells [9]; however, certain genes appear to be regulated by splicing during sexual differentiation. The 80 nt intron in the pre-mRNA encoding the meiotic cyclin Rem1, for example, is spliced after the initiation of meiosis when the gene is transcribed by the Forkhead family transcription factor Mei4 [10,11].
We identified and characterized the Prp4 kinase, which phosphorylates the spliceosomal protein Prp1 (scPrp6/hsU5-102K) in vitro and in vivo. Prp4 also phosphorylates Srp2, one of the two typical SR (serine/arginine-rich) protein family members present in fungi [12,13]. Prp1 contains 16 direct C-terminal tetratricopeptide repeats (TPRs), preceded by an N-terminal domain of approximately 30 kDa that contains no known motifs. The C-terminal region containing the TPRs is highly conserved across a wide range of organisms, whereas the N-terminal region is not [13]. Prp4 phosphorylates Prp1 at sites in the N-terminal region [14,15]. In fission yeast, the structural integrity of the N-terminal domain is essential for pre-mRNA splicing. Short deletions throughout the N-terminus of Prp1 do not prevent spliceosome assembly but lead to the formation of stalled precatalytic spliceosomes that contain unspliced pre-mRNA and the U1, U2, U5, and U4/U6 snRNAs, indicating that the N-terminus of Prp1 is involved in early spliceosome activation [14]. In mammals, the Prp4 ortholog Prp4K phosphorylates equivalent sites in the N-terminal region of Prp1, and also phosphorylates SR proteins [13,16,17].
SR proteins contain one or two N-terminal RNA recognition motifs (RRMs) and a C-terminal RS domain enriched in arginine–serine dipeptides [18]. In general, SR proteins control a network of RNA processing events, including the regulation of SS selection [19]. In mammals, they can act as splicing enhancer or silencer depending on their position of binding [20]. It is known that they play an important role not only in alternative splicing but also in constitutive splicing [21,22]. In case of constitutive splicing they take part in intron recognition at the exon1/5´ SS by interacting with hsU1-70K (spUsp101) as well as at the 3´ SS by interacting with hsU2AF1 (spUaf2) [23–26]. In S. pombe there are two SR proteins known, Srp1 (hsSRSF2) and Srp2 (hsSRSF4/5/6) [27,28]. They were found as single proteins but also as a complex depending on their phosphorylation states [29]. Srp2 is known to be phosphorylated by Prp4 kinase while Srp1 is not [13]. However, it was shown that overexpression of Srp1 can suppress the splicing phenotype of the mutant allele prp4-73ts [30]. The temperature-sensitive allele prp4ts caused at the restrictive temperature of 36°C a cell-cycle arrest in the G1 and G2 phases. This phenotype was also observed at the permissive temperature of 25°C when the mutant allele prp4-73ts was expressed from a multicopy plasmid [15].
Cell-cycle arrest in the G1 and G2 phase has been observed primarily in mutants of genes involved in cell-cycle regulation. Therefore, in our attempt to elucidate the mechanism underlying this phenotype, we first examined the splicing of cell-cycle regulatory genes that contain introns, e.g., cdc2, res1, and res2. We performed these experiments in cells expressing the analogue-sensitive (as) mutant prp4-as2, which encodes a kinase that can be chemically inhibited. This analysis revealed two classes of introns in fission yeast: those whose splicing is dependent on Prp4 kinase activity (Prp4-dependent) and those whose splicing is independent of Prp4 kinase activity (Prp4-independent). This finding was confirmed and extended by a genome-wide search for Prp4-dependent and -independent introns, which demonstrated that both intron classes are sometimes present within the same gene. For detailed characterization, we focused on two genes, res1 and ppk8; the single res1 intron is Prp4-dependent, whereas the single ppk8 intron is Prp4-independent. These two intron classes are affected by mutations in the exon1/5’ SS region or the bs of the same intron. The potential interactions between the exon1/5’ SS region and snRNP U1, and between the bs and snRNP U2, determine the Prp4 dependency of the intron. Taking into consideration the results of this and previous studies, we propose that phosphorylation of different substrates by Prp4 kinase helps the spliceosome to recognize and splice efficiently introns with weak SSs that differ from the consensus sequence.
For these experiments, we constructed a conditional analogue-sensitive allele, prp4-as2, which allows reversible inhibition of Prp4 kinase using an ATP analogue. To inhibit Prp4, 10 μM of 1NM-PP1 was added to growing cultures of cells expressing the prp4-as2 allele. Addition of inhibitor caused growth arrest and a concomitant decrease in the number of septated cells (Fig 1A and 1B). Cells arrested in the G1 and G2 phases of the cell cycle, as observed for the prp4ts strain. Indeed, fluorescence-activated cell sorting (FACS) analysis revealed cells with both 1C and 2C DNA content after 60 min inhibition of Prp4 (Fig 1C). This growth arrest was transient, and cells resumed growing after approximately 180 min, as indicated by the disappearance of the 1C peak and the reappearance of septated cells after 240 min (Fig 1).
The transient arrest in the G1 and G2 phases of the cell cycle led us to hypothesize that the genes whose splicing was affected by Prp4 inhibition were directly or indirectly involved in mitotic cell-cycle progression. For example, the gene encoding the cell-cycle regulator Cdc2 contains four introns that interrupt its open reading frame. Oscillating Cdc2 kinase activity levels regulate cell-cycle progression in G1 and G2 [31]. The decision point at which cells either make the transition from G1 phase to DNA synthesis or exit the cell cycle for conjugation is known as START. Cdc2 kinase and components of the Mlu1-binding factor (MBF) are involved in control of START by regulating the expression of genes required for DNA replication. The multimeric MBF complex consists of Cdc10, Res1, and Res2 [32,33]. The cdc10 gene is intronless, whereas the res1 and res2 genes each contain a single intron (of 127 nt and 164 nt, respectively) located in the 5’ region of the open reading frame [34,35].
Thus, we focused on the splicing of intron-containing genes that regulate the transition from G1 phase to DNA synthesis. Semiquantitative reverse transcription polymerase chain reaction (RT-PCR) analyses were performed to measure the mRNA and pre-mRNA levels of res1, res2, cdc2, and two control genes, rpl29 and tbp1 (Fig 1D). rpl29 contains one intron and encodes large ribosomal subunit protein 29; tbp1 contains three introns and encodes the TATA-binding protein [36]. The highly efficient splicing of res2 and rpl29 were barely affected by inhibition of Prp4 kinase. By contrast, for res1 and tbp1, unspliced pre-mRNAs accumulated and almost no mRNA was detected, as little as 10 min after the addition of inhibitor. This strong inhibition of splicing was transient, and mature spliced mRNA transcripts of both genes were observed again after 60 min. After 180 min, spliced mRNA levels were similar to those observed in the absence of inhibition (Fig 1D). Splicing of all four introns of the cdc2 transcript was only slightly affected by inhibition of Prp4 kinase, and mature cdc2 mRNA was detected at all time points. Remarkably, the splicing pattern of res1 and tbp1 remained basically the same throughout the time course (Fig 1D).
Collectively, these results indicated that the introns of the five genes we investigated could be categorized into two classes: Prp4-dependent and Prp4-independent. Subsequent experiments showed that the res1 intron was primarily responsible for the cell-cycle arrest in G1 following Prp4 inhibition: replacing the wild-type res1 gene with an intronless copy (res1Δintron) led to a similar growth delay, but the cells now primarily arrested in G2 phase (S1 Fig).
To further examine these two classes of introns, we performed a genome-wide search for additional Prp4-dependent and–independent introns. RNA prepared from the prp4-as2 strain grown in the presence (30 min and 60 min exposure, +Inh) or absence (-Inh) of the 1NM-PP1 inhibitor were subjected to RNA sequencing (RNA-seq), and the resultant sequence reads were aligned to the spliced or unspliced fission yeast genome reference sequence. To examine global changes in splicing efficiency, the Relative Splicing Efficiency Index (RSEI) of annotated fission yeast introns was calculated for the untreated and treated datasets. Fission yeast introns were divided into two classes. The first class contained the 72% of all introns with a positive RSEI in the presence and absence of 1NM-PP1, indicating that splicing of these introns was Prp4-independent. This class included res2, rpl29, and cdc2 (Fig 2A). The second class contained the 28% of introns for which RSEI was positive in untreated cells but negative in cells treated with 1NM-PP1, indicating that splicing of these introns was Prp4-dependent. This class included res1 and tbp1 (Fig 2A). Notably, we were unable to identify any gross sequence features that differentiated these two classes of introns. For example, there were no significant differences in intron size or obvious additional sequence motifs that were specific to either class (Fig 2B). Comparison of the SS sequences of Prp4-independent and -dependent introns revealed only slight differences between the two classes especially at position -1 in the exon 1 and positions +4 to +6 in the 5´ SS. The Prp4-dependent introns differed more frequently at these positions from the consensus sequence compared to Prp4-independent ones (Fig 2C). Moreover, in genes containing several introns, not all introns necessarily behaved in the same way upon inhibition of Prp4. This different behaviour of introns within one gene was also observed for temperature-sensitive alleles of other splicing factors [37,38]. As shown in Fig 2D, the mrp17 gene, encoding mitochondrial ribosomal subunit Mrp17, has one Prp4-dependent and one Prp4-independent intron; by contrast, rpb5, encoding a DNA-directed RNA polymerase subunit, contains two Prp4-independent introns, and tbp1 contains three Prp4-dependent introns.
We wished to characterize the changes in splicing efficiency when the exon1/5’ SS region and bs of these two intron types were mutated. For these experiments, we selected two genes, each containing one intron of similar size and structure, but with RSEI of opposite sign following inhibition of Prp4 kinase. The Prp4-dependent gene was res1, which contains a 127 nt intron, has an RSEI of -1.36, and is essential for mitotic growth (Fig 3A). The Prp4-independent gene was ppk8, which contains a 117 nt intron and has an RSEI of +1.89 (Fig 4A). This gene is non-essential for mitotic growth; it encodes a putative serine/threonine kinase potentially involved in signal transduction [39]. Because we wanted to compare the pre-mRNA splicing of these functionally very different genes following mutation, we constructed two reporter genes, called res1’ and ppk8’, respectively. Both reporter genes are driven by the nmt1-8 promoter and contain a frameshift mutation early in exon 1, a HindIII restriction site upstream of the 5’ SS, and the nmt1 termination region for 3’ end processing. These manipulated genes were introduced into the leu1 locus by homologous recombination (Figs 3A and 4A), and semiquantitative RT-PCR analyses were performed using the appropriate primers. As shown in Fig 3B–3C, the res1’ intron at the leu1 locus was spliced in a Prp4-dependent manner, like endogenous res1+, whereas the intron of ppk8’ was Prp4-independent like endogenous ppk8+ (Fig 4B–4C). This is consistent with the notion that the Prp4 dependency of a gene is not governed by its genomic context: neither chromosomal location nor the identity of the promoter and 3’ termination region determined whether an intron was spliced in a Prp4-dependent manner. Therefore, the differences in Prp4 dependency must be due to subtle differences in or around the intronic sequences.
Base-pairing interactions between pre-mRNA and snRNA U1 play a role in establishing and determining the 5’ SS [40–42]. This recognition process is more complicated in mammals than in yeast, because alternative pre-mRNA splicing requires selection of one out of several possible 5’ SSs [43]. In fungi, however, and particularly in fission yeast, there is little or no alternative splicing [9]. In fission yeast, it has been suggested that nine nucleotides from the 5’ end of the U1 snRNA interact with the pre-mRNA, base-pairing with six nucleotides of the 5’ SS and three nucleotides of exon 1, to determine the 5’ SS [44]. In addition, the 5’ end of snRNA U1 in fission yeast becomes pseudouridinylated (Ψ). This mechanism is also conserved in eucaryotes. However, in mammalian U1 snRNA two adjacent nucleotides in this region are pseudouridinylated, whereas in fission yeast, only nucleotide number 3 from the 5’ end of snRNA U1 is pseudouridinylated. This nucleotide in the U1 snRNA interacts with the 5’ SS nucleotide +4 of the pre-mRNA (Figs 3B and 4B). In general, pseudouridinylated nucleotides base-pair with A, C, G and U in an A-form RNA duplex, but the highest thermal stability can be found between A-Ψ and G-Ψ [45–48].
To analyse if an increased base-pairing potential leads to a Prp4-independent intron and vice versa, several mutations were introduced into the exon1/5´ SS of the reporter genes res1´ and ppk8´ (Figs 3 and 4). We mutated positions -3, +3 and +4 (Fig 3D, res1´-1) or only positions +3 and +4 (Fig 3D, res1´-2) of the res1’ intron. These changes, which increased the potential interactions between U1 snRNA and the pre-mRNA by at least four hydrogen bonds, converted the res1’ intron into a Prp4-independent intron. A time-course RT-PCR experiment revealed that these mutations allowed efficient splicing in the presence (+Inh) or absence (-Inh) of inhibitor (Fig 3D). As a control, we also mutagenized position +1 or +2 of the res1’ intron to a C or A, respectively; the resultant mutants were not recognized efficiently as introns independent of Prp4 activity (Fig 3E and S2 Fig, res1´-14). At the time that introns were discovered, it was shown that the GU at the 5’ end is necessary for recognition of an intron [49]. Therefore, this control experiment confirms our interpretation that the transient interaction of the exon1/5’ SS region with snRNA U1, including nine nucleotides from the 5’ end of U1, allows efficient splicing in the absence of Prp4 kinase activity (compare Fig 3D, res1’-1 and res1’-2 with Fig 3E). To test whether only one additional interaction at position +3 or +4 is already sufficient for Prp4 independency, both mutants were constructed and resulted in Prp4-independent introns. While an interaction at position +3 lead to an efficiently spliced intron at all time points, an interaction at position +4 caused a slightly decreased splicing efficiency after inhibition (Fig 3F). For all further experiments the Prp4-independent res1´-2 exon1/5´ SS was used to analyse which additional mutations lead to Prp4 dependency again. When these continuous interactions were interrupted by mutating position +5 in the intron from a G to an A, the intron became Prp4-dependent once again (Fig 3G). All mutations at position +5 that did not allow Watson–Crick hydrogen bonding with a C at position 2 of snRNA U1 caused the res1’ intron to be Prp4-dependent (Fig 3G and S2 Fig, res1’-15 and res1’-16); therefore, these experiments also prove that proper interaction between U1 and the exon1/5’ SS region is established by formation of hydrogen bonds between the two opposing bases. By contrast, the Prp4 independence of this intron was unaffected by all mutations at position +6 (the last 5’ SS nucleotide) that did not allow hydrogen bonding with the A at position 1 of the U1 snRNA (Fig 3H, res1´-7 and S2 Fig, res1’-17 and res1’-18). But if there is no interaction at positions +3, +4 and +6 the former Prp4-dependent intron (Fig 3C, res1´) is not recognized anymore even in presence of Prp4 kinase activity (Fig 3H, res1´-8). Thus, increasing the base-pairing potential between U1 snRNA and pre-mRNA within the 5´ SS results in Prp4-independently spliced introns and vice versa. Similar, rules seemed to apply for mutations at positions -1, -2 and -3 of exon 1 of the res1´ gene (Fig 3I). If only position -1 or position -2 can form a Watson–Crick hydrogen bond, therefore creating a weaker interaction in the exon 1 compared to res1´-2, splicing efficiency decreased after inhibition of Prp4, but the intron was still spliced independently (Fig 3I, res1’-10 and res1’-11). However, completely absent or very weak hydrogen bonding with the nucleotide at position -3 of exon 1 caused the intron to be Prp4-dependent (Fig 3I, res1’-12 and res1’-13).
Like ppk8, the intron of the ppk8’ gene integrated into the leu1 locus was also Prp4-independent (Fig 4A–4C). First, the 5’ SS nucleotide +1 G was mutated to a C, the intron was no longer recognized regardless of Prp4 kinase activity, as demonstrated by the exclusive presence of pre-mRNA and absence of mature mRNA (Fig 4D). Then position +3 was mutated from an A to a U, preventing Watson–Crick hydrogen bonding with the snRNA U1, and position +4 from a U to an A, creating an A-Ψ interaction (Fig 4E, ppk8’-2). This shows that changing the position of two hydrogen bonds of the 5´ SS converted a Prp4-independent intron into a Prp4-dependent one (compare Fig 4C, ppk8´ with 4E, ppk8´-2). When the interactions at both positions, +3 and +4, were absent, the resultant intron was no longer recognized (Fig 4E, ppk8’-3). However, changing the interactions within the exon 1 creating a ppk8’ gene containing the exon1/5’ SS sequence of the res1+ gene was Prp4-dependent (Fig 4F, ppk8’-4).
Next, we introduced mutations at the end of exon 1 to determine the effect of the pairing of these nucleotides on Prp4 dependency. If the nucleotides at positions -1, -2, and -3 of exon 1 in ppk8’ were unable to form hydrogen bonds with the appropriate positions in snRNA U1, the intron was spliced in a Prp4-dependent manner (Fig 4F, ppk8’-5). The same rule applied if hydrogen bonding only occurred at position -3 (Fig 4F, ppk8’-6). However, if a potential hydrogen bond could be formed with the nucleotide at position -1 or -2, stabilizing the interactions between exon 1 and snRNA U1, the intron was spliced efficiently in a Prp4-independent manner (Fig 4F, ppk8’-7 and ppk8´-8). Taken together, these results clearly demonstrate that the number and position of the potential hydrogen bonds between the nucleotides of the exon1/5´ SS region of an intron and the 5’ end of the U1 snRNA is one reason whether an intron is spliced in a Prp4-independent or -dependent manner.
As discussed above, we did not find any obvious differences in the bs consensus between +RSEI and–RSEI introns (Fig 2C). The consensus branch sequence of S. pombe is 1.C/U 2.U 3.A/G/U/C 4.A 5.C/U; the most frequent bs sequences are CUAAC (42%) and CUAAU (23%) [39,50]. The branch point A is the fourth nucleotide in this sequence [51–53]. The 5 nt degenerate bs of fission yeast is similar to that of mammals [6,8]. Recognition of the bs via base-pairing of snRNA U2 is also conserved, and the nucleotide at position 3 in the bs of S. pombe is opposite a pseudouridine in snRNA U2 [45,47,54]. The bs of the res1 intron has the most common sequence (CUAAC), and the pseudouridine in snRNA U2 is at position 39 from the 5’ end (Fig 5A). To study the influence of mutations in the bs on Prp4 dependency, we used res1´ transcripts with exon1/5´ SS regions AAG/GUAAGU (Prp4-independent) and AAG/GUUUGU (Prp4-dependent), respectively. We combined mutations in the 5´ SS with mutations in the bs. As expected, mutation of the branch point in position 4 (from A to U) combined with the Prp4-dependent 5’ SS prevented recognition of the intron, regardless of the presence or absence of inhibitor (Fig 5B, res1’-A). The reporter gene carrying the Prp4-independent 5’ SS sequence was spliced extremely inefficient, and it was further inhibited by inactivation of Prp4 (+Inh); consequently, only a small amount of mRNA was detected at the end of the time course (Fig 5B, res1’-2A).
The third nucleotide in the bs, which is degenerate for A, G and U, is supposed to interact with the pseudouridine at position 39 in snRNA U2. When we mutated this nucleotide from A to U, we converted the Prp4-independent intron into a Prp4-dependent one (compare Fig 5C, res1´-2B with Fig 3D, res1’-2). In contrast, the res1’-B gene, which already contains the Prp4-dependent exon1/5’ SS region, was inefficiently spliced even in the absence of Prp4 inhibitor. After addition of kinase inhibitor pre-mRNA accumulated completely (Fig 5C, res1’-B). These findings are in accordance with the notion that mutations in the third position of the bs lead to a Prp4-dependent intron when there is no Ψ-A interaction. In addition, if the intron was already Prp4 kinase-dependent because of a weak interaction between the exon1/5’ SS region and snRNA U1, the mutation in the bs intensifies the effect (Fig 5C, res1´-B).
The second nucleotide (U) in the bs is 100% conserved in all S. pombe introns. When we changed this U to a G, the intron was no longer recognized, neither without nor with Prp4 kinase inhibitor (Fig 5D). Furthermore, mutations in the first or the last nucleotide of the bs combined with the Prp4-dependent exon1/5´ SS lead to an extremely inefficient splicing event which was further intensified after inhibition of the kinase (Fig 5E, res1’-D and Fig 5F, res1’-E). However, when combined with the Prp4-independent exon1/5’ SS these mutations were spliced independently, but after inhibition of the kinase the splicing efficiency decrease slightly (Fig 5E, res1’-2D and Fig 5F, res1’-2E). To summarize the influence of mutations within the bs it is obvious that in case of a weak exon1/5´ SS the negative effect on intron recognition is intensified which leads mostly to intron retention. In contrast mutations in the bs combined with a strong exon1/5´ SS show different results depending on the position of the mutations. Except position 2 all others show an improvement of intron recognition as long as Prp4 kinase is active.
The 5’ SS and the bs of the introns in S. cerevisiae are highly conserved (GUAUGU and UACUAAC, respectively, in almost all introns) [53,55,56], whereas the corresponding sequences in S. pombe are much more degenerate; GUAAGU and GUAUGU are the most frequent 5’ SSs (29% and 21% of all introns, respectively), and CUAAC is the most frequent bs (42%) [39]. In this context we have shown that Prp4 kinase is one of the major components to facilitate proper recognition of introns with weak SSs. This kinase is involved in the process that helps to sense and influence proper base-pairing between the exon 1/5’ SS region and snRNA U1 and between the bs and snRNA U2; in this background, proper base-pairing refers to an interaction that accurately determines the 5’ and 3’ SS, leading to an efficient splicing event.
In this study we have shown by mutating the exon1/5’ SS and the bs of reporter genes that a Prp4-dependent intron can be changed into an–independent one and vice versa (Fig 3C and 3D and Fig 4C and 4E). The experimental set up also demonstrates that the information for Prp4 dependency resides in the region of the SSs of an intron and has been confirmed by inserting introns with weak and strong SSs into an intronless gene (S3 Fig). Regarding Prp4 dependency the positions for proper base-pairing between the exon1/5’ SS region and snRNA U1 play different roles. For example, positions +1 and +2 are invariable as it was known before [49]. Mutations of these positions lead to accumulation of unspliced pre-mRNA in the presence and absence of Prp4 kinase activity (Figs 3E, 4D and S2, res1´-14). This is different, if one considers position +3 and +4. Position +4 in the 5’ SS is the most degenerate, and can be occupied by any of the four nucleotides (Fig 2C). This position has been suggested to appear opposite a pseudouridinylated nucleotide of snRNA U1 (Fig 3B). Pseudouridine can base-pair with all nucleotides, although the thermodynamic parameters of these wobble base-pair interactions are different from those of Watson–Crick interactions [46,48,54]. If there is only one Watson-Crick interaction present at position -1 in exon 1 and the base-pairing at position +4 in the 5´ SS is interrupted, the intron is spliced Prp4-independently (Fig 4C). In contrast, an interruption at position +3 leads to a Prp4-dependent intron (Fig 4E, ppk8´-2). However, if there are two Watson-Crick interactions in exon 1, at positions -1 and -2, it does not matter which position, +3 or +4, is mutated (Fig 3F). The intron is spliced Prp4-independently. These results show that a stable interaction between the exon1/5´ SS and snRNA U1 within this region is needed for Prp4 independence. It also indicates the influence of the interaction with the pseudouridine which changes the helical structure at this position leading to a different binding affinity at position +4 [57,58]. Furthermore, interruption of the Watson-Crick interaction at position +5 leads to Prp4 dependency (Fig 3G and S2 Fig, res1´-15 and res1´-16). However, interruption of base-pairing at position +6 only becomes relevant, if there are no Watson-Crick interactions at positions +3 and +4 of the 5’ SS. In this case the intron is not recognized and therefore retained even in presence of Prp4 kinase activity. This shows that only three and intermittent base-pair interactions in the 5´ SS with snRNA U1 are insufficient for intron recognition (Fig 3H).
The snRNA U1 not only interacts with the 5´ SS of the intron, but also with the last three nucleotides of the exon 1 [44]. In general, the 5’ SS consensus sequence differs from the exon 1 sequences in that the 5’ intron sequences are much more highly conserved, whereas the three nucleotides of the exon 1 sequences are much more variable (Fig 2C). For example, for three introns in the same gene, it would be uncommon for the three nucleotides upstream of the 5’ SSs to be identical. Therefore, this region was also examined in this study regarding Prp4 dependency. As we have shown, if there is no interaction within exon 1 these introns are spliced Prp4-dependently (Fig 3I, res1’-13 and Fig 4F, ppk8’-5). The same results were obtained when an interaction only at position -3 was present (Fig 3I, res1’-12 and Fig 4F, ppk8’-6). Interestingly, formation of hydrogen bonds at positions -1 or -2 could stabilise the interaction between snRNA U1 and exon1/5´ SS, leading to Prp4 independence (Fig 3I, res1’-10 and res1’-11 and Fig 4F, ppk8’-7 and ppk8’-8). This rule could also be confirmed by the two introns of the wildtype gene mrp17 (Fig 2D). In this case, intron I has only one possible Watson-Crick base-pairing at position -3 (CCA/GUAAGU) and is Prp4-dependent (compare with Fig 3I, res1´-13). On the contrary, intron II displays one Watson-Crick interaction at position -2 and one wobble base-pairing at position -3 (UAA/GUAUGU) which leads to Prp4 independence (compare with Fig 4F, ppk8´-7). Probably, stabilising this interaction within the exon 1 helps to determine the proper site where the first transesterification reaction occurs.
Additionally, mutations in the bs were combined with a strong or weak exon1/5´ SS which had different effects on intron recognition and splicing efficiency (Fig 5). When combined with a weak exon1/5´ SS, mutations within the bs lead to intron retention in nearly all cases even without inhibition of Prp4 kinase. Therefore, the accumulation time course of pre-mRNA after kinase inhibition reflects an additive effect (Fig 5B–5F, res1´-A-E). In combination with a strong exon1/5´ SS the mutations in the bs showed different effects on splicing. The change of the branch point is almost invariable and therefore leads to intron retention even if no kinase inhibitor was added (Fig 5B). The mutation at position 2 also resulted in complete intron retention even if no kinase inhibitor was added (Fig 5D, res1´-2C). Most interestingly, this position is 100% conserved in all S. pombe introns (Fig 2C). The third position in the bs is the most degenerate, and can be occupied by any nucleotide (Fig 2C); this position interacts with a pseudouridinylated nucleotide at the end of snRNA U2 which is responsible for bulging out the branch point [54,59] (Fig 5A). The mutation of this position leads to a clear Prp4-dependent intron (Fig 5C, res1´-2B). On the contrary, the interruption of possible base-pairing at positions 1 and 5 seems to play a minor role since splicing is still independent on Prp4 kinase activity, although after inhibition of the kinase splicing efficiency decreased slightly in both cases (Fig 5E, res1’-2D and 5F, res1’-2E).
Although the underlying rules seem to be very complex, it is obvious that Prp4-dependent introns are distinguished from Prp4-independent introns by their reduced potential for hydrogen bonding between the exon1/5’ SS region and snRNA U1 or between the bs and snRNA U2. A complementary interaction between the exon1/5´ SS region and snRNA U1 serves as a default state, marking the structure for the first transesterification reaction in the pre-spliceosome. A similar structural marking for proper hydrogen bonding is also associated with the bs-U2 interaction thereby determining the nucleotides where the second transesterification reaction will occur. So far, we can only speculate here about the consequences of the phosphorylation of Prp1 and Srp2 by Prp4 kinase and propose that the phosphorylation by Prp4 might play a role in stabilising the interaction between the SSs and the snRNAs allowing time in concert with the other proteins to display the proper intron borders for the transesterification reactions. Prp1 is a spliceosomal protein operating at the level of precatalytic spliceosomes; therefore, it is reasonable to speculate that phosphorylation of Prp1 could be involved in adjusting a precatalytic spliceosome on introns displaying weak SSs until proper hydrogen bonds between the pre-mRNA and snRNA U1 and U2 are formed. Phosphorylation of Srp2 by Prp4 might play a similar role, helping precatalytic spliceosomes with the recognition of introns and thereby stabilising their interaction with weak exon1/5’ SSs and weak branch sequences. Indeed, it is known for mammalia that the SR protein hsSRSF1 binds to the pre-mRNA and subsequent phosphorylation affects its interaction with a protein of the U1 particle [23]. In S. pombe it has been shown that Srp2 interacts with spUaf2 which binds to the 3´ SS [60]. It seems likely that it also takes part in exon1/5´ SS recognition. However, there are still open several questions regarding this mechanism. Particularly, we hope to identify further components involved and thereby advance our knowledge about the function of Prp4 kinase in the splicing of introns displaying weak SSs.
The standard genetic and molecular techniques used in this study were described previously [61,62]. All strains used in this study are listed in S1 Table.
The analogue-sensitive prp4-as2 kinase allele was generated by introducing a point mutation into the kinase domain at position 238; this mutation changes the gatekeeper phenylalanine residue to alanine [63,64]. The prp4-as2 allele was fused to the kanMxR gene using the pRS426 vector in yeast recombinational cloning [65,66]. The resultant construct was used to produce a PCR fragment, containing the prp4-as2 kinase allele and the resistance marker, which was transformed into the wild-type strain L972. Growing colonies were selected on plates containing geneticin. Proper replacement of the prp4 locus on chromosome III was confirmed in geneticin-resistant transformants by PCR using the appropriate primers [67,68].
To construct reporter genes, res1´ and ppk8´ were fused to the thiamine-repressible nmt1-8 promoter by cloning the open reading frames into vector pML81HA [69,70]. Both open reading frames contain a frameshift between the HA-tag and the ATG, creating a stop codon at the beginning of the gene; this is intended to exclude the influence of additional Res1 molecules on cell-cycle regulation. To distinguish between the original res1 and ppk8 transcripts and the res1´ and ppk8´ transcripts at the leu1 locus by RT-PCR, a HindIII restriction site was introduced into exon 1 of both genes (Figs 3A and 4A). The forward primers in each pair (res1_Mut_F, ppk8_Mut_F) detect the HindIII site.
To determine the pre-mRNA splicing patterns of intron-containing genes, RNA was extracted from whole-cell extracts. For RT-PCR, RNA was treated with RQ1 RNase-free DNase (Promega) to eliminate possible DNA contaminants. Five micrograms of RNA was treated with 2.5 U RQ1 DNase in reaction buffer containing 40 mM Tris–HCl (pH 8.0), 10 mM MgSO4, and 10 mM CaCl2. After incubation for 10 min at 37°C, RQ1 DNase was inactivated by the addition of 2 mM EGTA (pH 8.0) and heating to 80°C for 10 min. RNA was reverse transcribed and cDNA was amplified using Tth reverse transcriptase (Roboklon, EURX). RNA was incubated for 3 min at the calculated annealing temperature in the presence of 1× Tth RT Buffer, 0.25 mM dNTP mix, 20 pmol reverse primer, 2 mM MnCl2, and 0.25 U/ml Tth RT, followed by incubation at 70°C for 25 min. PCR mix containing 1× PCR-Buffer Pol A, 80 pmol reverse primer, 100 pmol forward primer, and 2 mM MgCl2 was then added. cDNA was amplified with 28−45 cycles of 94°C for 30 s, 53−60°C for 30 s, and at 72°C for 60 s. The primer sequences are provided in S2 Table. PCR products were resolved on 2% agarose gels.
Directional mRNA sequencing libraries were prepared by combining the Illumina TruSeq mRNA-Seq and Illumina TruSeq small RNA protocols. Briefly, mRNA selection was performed using 4 μg of total RNA and oligo-dT beads, as described in the low throughput protocol for Illumina TruSeq RNA sample preparation. The mRNA was subjected to fragmentation at 94°C, treated with Antarctic Phosphatase (NEB) and T4 polynucleotide kinase (NEB), and then purified using RNeasy MinElute spin columns (Qiagen). TruSeq indexed RNA adapters were ligated to the RNA and further processing, including 11 cycles of PCR for library amplification, was performed as described in the Illumina v1.5 small RNA protocol. Finally, fragments corresponding to an insert size of 250–500 nt were selected on a 6% Novex TBE gel (Invitrogen). After elution from the gel slice, library quality was confirmed using a DNA 1000 Bioanalyzer chip on an Agilent 2100 Bioanalyzer and. Sensitive quantitation was performed using a KAPA Library Quantification Kit (Kapa Biosystems). Five indexed libraries were pooled and run in each HiSeq lane using Illumina HiSeq v3 sequencing chemistry. Base calling was performed using the Illumina pipeline software version 1.8.1 (within HCS 1.4.8). Adapters used during library preparation were removed from reads using the TagDust tool [71]; approximately 1% of the initial reads were removed in this way. Reads were mapped to the S. pombe genome using the TopHat software (v2.0.5) [72]. The TopHat alignment was performed using the annotation defined in the Ensembl database (version ASM294 v1.15), taking into account the orientation of the reads. The percentage of mapped reads was approximately 95% for all samples. Finally, the unequivocally mapped reads (85–89% of the initial reads) were selected for further analysis. The average coverages for the exonic and intronic spaces were determined considering only genomic elements longer than 30 bp. This information was then plotted as three different graphs showing the coverage of 5’ exons, introns, and 3’ exons. Exons that were defined simultaneously as the 5’ exon for one intron and the 3’ exon for another intron were considered twice. Finally, the coverage was calculated by normalization to the total number of mapped reads using BEDTools [73]. Plots were generated with custom R scripts.
The RNA-seq data have been deposited in NCBI's Gene Expression Omnibus and are accessible through GEO Series accession number GSE75517 (https://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?acc=GSE75517)
For each intron, the number of split and unsplit reads that mapped to the splice junction (defined as the three bases around the 5’ exon junction) were counted, and the base-2 logarithm of the ratio between the two values was calculated. This value is called the Relative Splicing Efficiency Index (RSEI). If the RSEI was positive, then more reads indicated spliced mRNA than unspliced pre-mRNA. On the other hand, a negative RSEI indicates more unspliced than spliced RNA. Only intron sequences with more than ten reads for each sample were used for further analysis. This approach identified 2557 Prp4-independent introns (i.e., those with a positive RSEI irrespective of Prp4 inhibition) and 1008 Prp4-dependent introns (i.e., those with a negative RSEI in the presence of kinase inhibitor).
|
10.1371/journal.ppat.1001098 | Steric Shielding of Surface Epitopes and Impaired Immune Recognition Induced by the Ebola Virus Glycoprotein | Many viruses alter expression of proteins on the surface of infected cells including molecules important for immune recognition, such as the major histocompatibility complex (MHC) class I and II molecules. Virus-induced downregulation of surface proteins has been observed to occur by a variety of mechanisms including impaired transcription, blocks to synthesis, and increased turnover. Viral infection or transient expression of the Ebola virus (EBOV) glycoprotein (GP) was previously shown to result in loss of staining of various host cell surface proteins including MHC1 and β1 integrin; however, the mechanism responsible for this effect has not been delineated. In the present study we demonstrate that EBOV GP does not decrease surface levels of β1 integrin or MHC1, but rather impedes recognition by steric occlusion of these proteins on the cell surface. Furthermore, steric occlusion also occurs for epitopes on the EBOV glycoprotein itself. The occluded epitopes in host proteins and EBOV GP can be revealed by removal of the surface subunit of GP or by removal of surface N- and O- linked glycans, resulting in increased surface staining by flow cytometry. Importantly, expression of EBOV GP impairs CD8 T-cell recognition of MHC1 on antigen presenting cells. Glycan-mediated steric shielding of host cell surface proteins by EBOV GP represents a novel mechanism for a virus to affect host cell function, thereby escaping immune detection.
| The Ebola virus (EBOV) is a highly pathogenic virus that infects humans and non-human primates, causing severe disease or death in the majority of these cases. The interaction of this virus with its host on a cellular level is only just beginning to be understood. EBOV, like many viruses, affects the expression or function of several cell surface proteins, including adhesion factors and protein complexes responsible for allowing the immune system to recognize infected cells. Our group and others have previously shown that expression of the main viral glycoprotein of EBOV in cultured cells is sufficient to cause this disruption. Here we have identified the mechanism by which this disruption occurs. Heavily glycosylated domains of the EBOV glycoprotein form a steric shield over proteins at the cell surface. This steric interference blocks the detection of affected surface proteins using antibody reagents, but also has the functional effect of abrogating cell adhesion and preventing interactions with CD8 T cells. The results from this study highlight a novel mechanism for viral disruption of host cell surface protein functions and give insight to interactions between the Ebola virus and its host.
| EBOV is an enveloped, negative-stranded RNA virus, a member of the family Filoviridae, and the causative agent of Ebola viral hemorrhagic fever. To date, five subtypes of EBOV have been identified: Zaire, Sudan, Côte d'Ivoire, Reston and Bundibugyo. EBOV Zaire is the most pathogenic subtype in humans, with mortality rates reaching 90% [1]. The basis for the high pathogenicity of EBOV is unclear, however immune dysregulation has been hypothesized to play a role [2]. Similarly to many other viral systems, EBOV infection appears to downmodulate the expression of host surface proteins involved in cellular recognition, most notably major histocompatibility complex (MHC) molecules and integrins [3].
EBOV encodes two forms of its glycoprotein. One is a dimeric, secreted form (sGP), which is transcribed directly from the viral RNA [4], [5] and whose function remains unclear. A second glycoprotein species results from transcriptional editing of the glycoprotein ORF and encodes a trimeric, membrane-bound form (GP). This form is expressed at the cell surface and is incorporated into the virion [4] and drives viral attachment and membrane fusion. GP is initially translated as a precursor (GP0), which is then cleaved by furin in the Golgi into two subunits, a surface subunit, GP1 and a membrane-spanning subunit, GP2 [6]. These subunits remain covalently connected through a single intermolecular cysteine bond [7]. Expression of the main viral glycoprotein, GP, has been shown to cause effects in cell culture on host surface proteins similar to those observed during viral infection, and so is proposed to be an important determinant of viral pathogenesis [8], [9], [10], [11]. Because sGP is the predominant form transcribed, it has been postulated that the balance between sGP and GP serves to regulate the cellular effects of GP [11].
Expression of high levels of EBOV GP in cultured cells disrupts cell adhesion resulting in loss of cell-cell contacts as well as cell rounding and loss of attachment to the culture substrate [8], [10], [12]. This can be observed in a variety of cell lines and primary cell types [12]. Interestingly, while transient GP expression does not cause death in human embryonic kidney 293T cells, primary human cardiac microvascular endothelial cells have been reported to undergo anoikis, or detachment-mediated apoptosis, upon transduction of GP [12], [13]. By flow cytometry, cells expressing GP display dramatically lowered levels of various surface proteins, including several members of the integrin family and MHC class I (MHC1); however, the exact complement of surface proteins affected by GP appears to differ by cell type [10], [12], [14]. Importantly, EBOV infection of 293T cells was observed to cause similar reduction of β1 integrin and MHC1 staining by flow cytometry, confirming that observations from transient GP expression are not simply artifacts of overexpression [3]. The effects of EBOV GP are caused by a highly glycosylated region in GP1, the mucin domain [8], [12], [14]. This domain encompasses approximately 150 amino acids, contains numerous N- and O- linked glycosylation sites, and is a distinctive feature of filoviral GPs. The mucin domain is not only necessary, but also sufficient for the observed EBOV GP-mediated effects upon surface protein expression and cellular adhesion [8], [15].
Few studies have been undertaken to investigate the mechanism by which EBOV GP disrupts adhesion and causes surface protein downmodulation. Our recent analysis concluded that the cellular endocytic factor dynamin does not play a role in surface protein downmodulation, suggesting the process may not involve cycling of proteins from the cell surface [15]. In contrast, Sullivan and colleagues have reported that this process requires dynamin [14]. Additionally, it has been reported that the extracellular signal-regulated kinases (ERK 1/2) play a role in downmodulation [16] suggesting an active process. In the present study, we provide direct evidence that EBOV GP-mediated loss of surface protein recognition occurs via steric shielding of surface epitopes, not by protein removal from the cell surface. Moreover, we demonstrate that EBOV GP expression blocks MHC1-mediated stimulation of T cells. Based upon these findings, we present a model in which the heavily glycosylated EBOV glycoprotein acts as a “glycan umbrella” to physically occlude access to host proteins, and GP itself, thereby impairing host protein function. EBOV GP-mediated steric occlusion represents a unique viral mechanism to interfere with the function of host proteins.
EBOV GP expression can dramatically reduce the observed levels of numerous host cell surface proteins including factors involved in immune recognition and cellular adhesion [10], [12], [14]. This effect can be seen by analysis of MHC1 or β1 integrin by flow cytometry staining in HEK293T cells transiently expressing Zaire EBOV GP (Figure 1A). Overall, a 10- to 50-fold reduction in surface levels of these host markers is observed in cells transfected with an EBOV GP cDNA. Additionally, there appears to be a critical threshold of EBOV GP expression required to induce surface protein downmodulation [15]. In parallel with the decrease in staining for host proteins, EBOV GP expression also appears to be reduced, resulting in a distinctive horseshoe-shaped flow cytometry profile (Figure 1A and [14], [15], [16]). Despite this apparent decrease in surface protein levels observed by flow cytometry, there were no consistent, significant changes in total protein levels for the EBOV glycoprotein upon analysis by Western blot in either adherent or non-adherent EBOV GP transfected cells (data not shown). To look directly at host protein expression in cells expressing EBOV GP, nonadherent, GP-transfected 293T cells were collected and analyzed by flow cytometry for expression of β1 integrin ([15] and Figure 1B, left panel). As previously described [15], these nonadherent cells represent the lower two quadrants of the “horseshoe” and appear to have reduced levels of both β1 integrin and EBOV GP. In contrast to the flow cytometry results, analysis of EBOV GP in these cells by immunofluorescence microscopy after fixation and permeabilization reveals extensive staining at the plasma membrane (Figure 1B, right panel). Similar to these results, previously published microscopic analysis of cells expressing EBOV GP also shows extensive plasma membrane staining with little evidence of significant accumulation of GP in internal vesicles [15], [17].
To evaluate steady state levels of host proteins and EBOV GP in cells transiently expressing the viral glycoprotein, the transfected cells were fixed, permeabilized and analyzed by flow cytometry. In vector-transfected cells, the permeabilization treatment had little effect upon staining for β1 integrin or MHC1 (Figure 2A). However, in cells transiently expressing EBOV GP, which displayed dramatically reduced levels of β1 integrin and MHC1 by surface staining (Figure 2B, left column), fixation and permeabilization reveals no decrease in either of these host proteins (Figure 2B, right column). Similarly, the apparent loss of EBOV GP staining is reversed by this treatment. These effects are best illustrated by comparison of the lower two panels in Figure 2B where without treatment, 9.3% of the cells displayed low MHC1 and EBOV GP levels, however after fixation and permeabilization the number of double negative cells was reduced to background levels and these now appear as MHC+, GP+ cells in the upper right quadrant. As expected, the untransfected cell population of 32–34% remains unaltered by this treatment (Figure 2B, upper left quadrants). Overall, this analysis suggests that the apparent downmodulation observed is not due to reduced steady-state levels of protein. Rather these transfected cells express unaltered levels of EBOV GP and MHC1, however these proteins are inaccessible for surface staining.
Recent structural analysis of EBOV GP suggests that the recognition site for the monoclonal antibody, KZ52, employed in the flow cytometry analysis resides near the base of the protein [18] below the globular GP1 and heavily glycosylated mucin domains in GP. This finding, coupled with our results suggesting that downmodulation in these cells was not accompanied by a reduction in steady-state levels of β1 integrin or MHC1, or a significant re-localization of EBOV GP, prompted us to consider the hypothesis that EBOV GP mediates its effects by blocking access to epitopes of proteins on the cell surface including epitopes within GP. Additionally, this hypothesis is consistent with the apparent threshold of GP expression required for downmodulation as well as the lack of a dynamin requirement [15].
To test this hypothesis, we engineered epitopes within EBOV GP at locations which, based on their position relative to the mucin domain and the globular region of GP, are predicted to be more accessible than the KZ52 epitope. Two constructs were created with an AU1 antibody epitope tag at the N or C terminus of the mucin-like domain, termed NmucAU1 GP and CmucAU1 GP, respectively. Cartoon depictions of each construct are shown in Figure 3C and D. These constructs were well expressed, as judged by Western blot analysis for EBOV GP and the AU1 tag (Figure 3A). The sub-cellular localization of these constructs was also evaluated in HeLa cells by immunofluorescence microscopy and was found to be indistinguishable from wt GP (Figure 3B).
Although the structure of the mucin domain is unknown, its mucin-like O glycosylation may force the domain into an extended conformation as has been suggested for cellular mucin proteins [19]. This would likely position the C terminus of the mucin domain so that it is more exposed compared to the GP core (Figure 3C and D). Based upon the proposed steric occlusion model, we hypothesized that the AU1 epitope of CmucAU1 would be most accessible to antibody staining. In contrast, the AU1 epitope in NmucAU1 might be less accessible than the epitope in CmucAU1 because of its location at the base of the mucin domain. Cells expressing wt GP, CmucAU1 GP, and NmucAU1 GP were analyzed by flow cytometry. When stained with the GP-specific KZ52 antibody, the epitope-tagged mutants displayed the characteristic comma-shaped flow cytometry plot seen with wt GP (Figure 3C and D; top rows). In contrast to the reduced KZ52 staining observed, the AU1 epitope in CmucAU1 was highly visible by flow cytometry (Figure 3C and D; bottom middle panels). Staining of the AU1 epitope on NmucAU1 GP was intermediate relative to CmucAU1 GP and wt GP KZ52 staining (Figure 3C and D; bottom right panels). In support of the shielding model, these data demonstrate that cells exhibiting reduced levels of β1 integrin and MHC1 have high surface levels of GP as revealed by AU1 staining, not reduced levels as indicated by KZ52 staining. Furthermore, these data suggest that antibody accessibility to epitopes in GP differs based on the epitope position relative to the mucin domain and the globular regions of GP1.
The data presented above are consistent with EBOV GP affecting recognition of epitopes within GP by shielding, however we wished to address if a similar mechanism was responsible for the apparent downmodulation of host surface proteins. To directly address whether EBOV GP sterically occludes host surface protein epitopes, we sought to unmask MHC1 and β1 integrin staining. We hypothesized that dissociation of the GP1 subunit, which includes the mucin domain and globular “head” region of EBOV GP, from GP2 at the cell surface should relieve the shielding of previously occluded epitopes. The GP1 subunit is covalently linked to GP2 via a single sulfahydryl bridge between residues C53 and C609 [7]. We have previously demonstrated that this bond can be reduced by incubation with DTT, allowing for dissociation of the EBOV GP1 subunit from the surface of virions [20]. To confirm that DTT is able to effectively remove GP1 from the cell surface, cells expressing GP were incubated with DTT then the supernatant was analyzed for GP by Western blot. Figure 4A reveals that GP1 was readily detected in the supernatant of cells incubated with DTT compared to mock treated cells. Control experiments also demonstrated that the DTT treatment did not significantly alter surface expression of β1 integrin or MHC1 in mock-transfected cells (Figure 4B). Moreover, this treatment did not result in permeabilization of the cells (Figure 4C) which, as shown above (Figure 2), could also rescue β1 integrin and MHC1 staining. In addition, these and the following experiments were carried out in the presence of azide and 2-deoxy glucose to ensure that the trafficking of nascent or recycled protein did not complicate the interpretation of this assay.
We next examined the effect of DTT treatment on surface staining of β1 integrin and MHC1 in cells expressing EBOV GP. Flow cytometry analysis of the DTT-treated, GP-expressing cells indicates that GP-induced loss of staining of β1 integrin and MHC1 is reversed by DTT treatment and subsequent dissociation of GP1 from the cells: upon DTT treatment, staining of β1 integrin and MHC1 is restored to nearly control levels (Figure 4D). Interestingly, staining for GP was also rescued, resulting in cells that stained positively for both GP and β1 integrin or MHC1. This is somewhat counter-intuitive, as KZ52 makes critical contacts with residues on GP1 [18], which is removed from the cell surface by DTT. These data suggest that DTT treatment removes a significant amount of GP1 from the cell surface – enough to reverse the steric occlusion of β1 integrin and MHC1 epitopes, as well as the KZ52 epitope. However, sufficient GP1 remains on the cell surface to allow for staining of GP by flow cytometry. Supporting this hypothesis, there appears to be a very modest downmodulation of integrin and MHC1 on the cells that appear to have the highest levels of GP (Figure 4D, upper right quadrant of the +DTT flow cytometry plots). This finding agrees with our previously published study that suggests a threshold level of EBOV GP is needed to downmodulate β1 integrin, MHC1 or GP [15].
Removal of surface GP1 by DTT reverses the apparent downmodulation of surface proteins induced by EBOV GP. To ensure this effect could be directly attributed to the EBOV glycoprotein we tested the effect of DTT on cells expressing a mutant form of GP lacking the endoproteolytic site required for processing GP0 into GP1 and GP2 subunits. Previous analysis demonstrated that this mutant EBOV glycoprotein, GP cl(-), retains normal viral entry function [20], [21] and is therefore likely folded similarly to wt EBOV GP. As shown in Figure 4F, GP cl(-) also downmodulates MHC1 similarly to wt EBOV GP. However in contrast to wt GP, DTT treatment of cells expressing this uncleaved form of GP does not relieve the observed downmodulation of MHC1 (Figure 4F). As anticipated, DTT treatment of cells expressing GP cl(-) produced no increase in GP release compared to untreated cells (Figure 4E). The EBOV glycoprotein found in the supernatant from the GP cl(-) expressing cells likely represents trimeric GP released by the cellular enzyme TACE [22]. Overall, these data strongly support the model proposed for EBOV GP mediated occlusion of host surface proteins.
EBOV GP is a heavily glycosylated protein, and we have previously shown the mucin domain to be sufficient to induce loss of staining of host surface proteins by flow cytometry [15]. Therefore, we directly addressed whether GP glycosylation plays a role in the shielding of surface epitopes. GP-expressing cells were treated with several glycosidases or pre-treated with a small molecule inhibitor of mucin synthesis, benzyl-α-GalNAc, then assayed for β1 integrin staining by flow cytometry. Importantly, none of the glycan-interfering treatments used here increased the staining for β1 integrin in cells transfected with empty vector (Figure 5A). Also, these treatments did not cause the permeabilization of cells, allowing us to attribute changes in staining to alterations at the cell surface (Figure 5B). Staining for β1 integrin on GP-expressing cells was increased by incubation with PNGaseF, an endoglycosidase that cleaves N-linked sugar moieties (Figure 5D left). Similarly, staining for β1 integrin was increased by incubation with neuraminidase, an exoglycosidase that cleaves sialic acid, which is a common component of mucin sugars. (Figure 5D, middle). When GP-expressing cells were incubated with both PNGaseF and neuraminidase, an additive effect was seen and β1 integrin staining was further increased (Figure 5D, right). The effect of glycosidase treatment on cellular GP was also analyzed by Western blot (Figure 5C, left). PNGaseF treatment results in loss of the top band of GP1, which is the maturely-glycosylated form and the appearance of bands which co-migrate with GP1 that has been PNGaseF treated under denaturing conditions, but which still contains O glycosylation. Treatment with neuraminidase did not result in a perceivable shift in migration of GP1; this is likely due to the small mass of these glycans and the resolution of the gel. These data indicate a direct role for N-linked glycans in GP-mediated loss of β1 integrin staining.
To directly address the role of O glycosylation in host protein downmodulation by EBOV GP, O glycosylation was perturbed by pre-incubating cells with benzyl-α-GalNAc or the control vehicle DMSO. This compound is a competitive inhibitor of β1,3-galactosyltransferase, which prevents the modification of core O glycan structures, resulting in shorter O-linked glycans and reduced sialyation [23], [24], [25]. Cells pre-treated with benzyl-α-GalNAc, then transfected with vector encoding GP showed increased staining for β1 integrin compared to DMSO treated cells, consistent with a role for O glycoslyation in the shielding of epitopes by the GP mucin domain (Figure 5E, left plot). In cells pre-treated with benzyl-α-GalNAc and expressing GP, incubation with PNGaseF further increased staining for β1 integrin (Figure 5E, right plot). Cells pre-treated with benzyl-α-GalNAc were also incubated with O-glycosidase, which can cleave unmodified core GalNAc structures; however, no further increase in β1 integrin was observed (data not shown). This is perhaps due to remaining modification of the core O glycans. The effect of these treatments on GP glycosylation was analyzed by Western blot (Figure 5C, right blot). Treatment with benzyl-α-GalNAc results in a modest increase in mobility for bands corresponding to GP containing O glycosylation, which are most easily seen in samples that have been PNGase-treated after cell lysis. Our data here suggest that the mass of O glycosylation is reduced, but not fully eliminated. This is expected, as benzyl-α-GalNAc only reduces mucin modification, but does not prevent the synthesis of initial core glycans. Taken together, these data demonstrate that surface N- and O- linked glycans, presumably on EBOV GP, contribute to the ability of GP to mask surface β1 integrin epitopes.
The data presented above demonstrate that epitopes at different locations within EBOV GP are differentially occluded in GP expressing cells (Figure 3). To determine if a similar situation occurs for cellular proteins, four monoclonal antibodies that recognize distinct regions of MHC1 were analyzed. In cells expressing GP, staining for MHC1 is blocked regardless of the epitope examined (Supplemental Figure S1). Given the ability of EBOV GP to effectively mask all of the epitopes examined on MHC1, we wanted to address whether this had functional consequences for MHC1. Human OV79 cells expressing the HIV Gag-derived peptide SLYNTVATL (SL9) were used to test the effect of EBOV GP on MHC1 antigen presentation. These cells present the SL9 antigen using a stably expressed MHC1, HLA-A2. The OV79- SL9 cells were mock transduced or transduced with adenoviral vectors encoding GFP (AdGFP) or GFP and EBOV GP (AdGP), which resulted in nearly 100% of cells expressing GFP (Figure 6A). Expression of EBOV GP dramatically reduced MHC1 levels in these cells whereas the control AdGFP vector had no effect on MHC1 expression (Figure 6B). Primary human CD8 T-cells transduced with a lentiviral vector expressing a T-cell receptor (868TCRwt) specific for SL9 were used to assess antigen presentation by GP-expressing OV79 cells. T-cell activation was measured by intracellular staining for production of the cytokine MIP-1β in CD8+ 868TCRwt+ expressing cells (Figure 6C). Production of MIP-1β has been shown to be the most sensitive indicator of HIV-specific CD8 T-cell activation [26]. Quantification of the CD8 activation results demonstrates that expression of EBOV GP had a profound effect on antigen presentation by the target cells, reducing T-cell responses to nearly background levels (Figure 6D). In contrast, the AdGFP control cells only modestly reduced the number of responding T-cells. Similar results were obtained using 293T target cells (data not shown). The ability of EBOV GP to interfere with T-cell function does not appear to be caused by inhibitory signaling from the GP-expressing antigen presenting cells (APCs) since T-cells were fully activated when anti-CD3/CD28-coated beads were added to a mixture of GP-expressing OV79 and CD8+ 868TCRwt+ cells (data not shown). Given the observations that all of the queried epitopes on MHC1 were occluded by GP expression and that T-cells are not stimulated by these APCs, the most straightforward explanation is that EBOV GP expression impedes T-cell recognition of antigen presenting cells. Overall these data support a model in which EBOV GP not only masks MHC and other surface proteins from antibody recognition, it also functionally inactivates them.
An important component of the virus host interaction is viral modulation of host functions. Many viruses alter expression and/or function of host surface proteins to affect signaling, immune surveillance, or viral superinfection. EBOV GP expression in cell culture has been observed by several groups to cause dramatic changes in cell adhesion and reduction in surface protein staining by flow cytometry [10], [12], [14], [16]. EBOV infection causes a similar reduction of β1 integrin and MHC1 staining by flow cytometry, suggesting that observations from transient GP expression are not simply artifacts of overexpression [3]. EBOV GP-induced effects have previously been assumed to result from removal of surface proteins from the plasma membrane. In this study we analyzed the mechanism of downmodulation of host surface proteins by the Ebola viral glycoprotein, GP. We show that reduction in surface staining for the host proteins MHC1 and β1 integrin is not accompanied by decreases in the total cellular levels of these proteins. Moreover, the observed self downmodulation of EBOV GP does not result in relocalization of GP away from the plasma membrane. Using epitopes placed at various locations in EBOV GP we find that the observed GP surface levels appear to differ based on epitope position relative to the mucin domain and the globular regions of EBOV GP. A similar observation has been made using a series of monoclonal antibodies to EBOV GP [27]. Additionally, the apparent downmodulation of surface proteins is reversed by removal of the EBOV GP1 subunit by reduction or by enzymatic digestion of the carbohydrate modification on EBOV GP. Finally, our data demonstrate that EBOV GP expression dramatically impairs antigen presentation by host cells. Taken together these data support a model in which EBOV GP utilizes a steric occlusion mechanism to downmodulate accessibility and function of host surface proteins.
The ability of viruses to affect host surface proteins has been well documented. For example, viruses may downregulate their cellular receptor, as in the case of HIV downregulation of CD4 and measles virus downregulation of the complement regulatory protein [28], [29]. Other common targets for virus mediated downmodulation are surface proteins related to immune surveillance. MHC1 is known to be downregulated from the cell surface by many viral proteins: HIV nef, Adenovirus E19, and KSHV K3 and K5, to name a few [30], [31], [32]. Activating ligands for natural killer (NK) cells have also been shown to be actively downregulated by KSHV and Hepatitis C virus [33], [34]. Multiple mechanisms and cellular pathways have been implicated in viral dysregulation of the various host surface molecules (reviewed for MHC1 in [35]). The model demonstrated here of glycan mediated steric occlusion by EBOV GP represents, to our knowledge, a distinctive mechanism for viral regulation of host surface proteins. Indeed, a similar steric masking model has recently been proposed for EBOV GP [27]. The polydnavirus, Microplitis demolitor bracovirus expresses a mucin domain-containing glycoprotein which can abrogate cell adhesion and thus may utilize a mechanism similar to that proposed here for EBOV [36].
Our observation that enzymatic removal of carbohydrate modification can relieve downmodulation, coupled with prior observations that the mucin domain of EBOV GP is sufficient for downregulation [8], [15], suggests that the steric occlusion observed is mediated, at least in part, by N- and O-linked modification of EBOV GP. A similar glycan mediated steric hindrance model has been proposed for cellular mucin proteins, which can disrupt a variety of cell-cell interactions at the plasma membrane [37], [38], [39], [40], [41]. For the cellular mucin proteins, densely-arrayed O-linked glycans are critical for disruption of cell adhesion, with different core glycan structure and subsequent modifications influencing the function and anti-adhesive properties of the protein [42]. Additionally, the number of mucin tandem repeats positively correlates with the anti-adhesive properties of Muc1 [41]. Similarly, we have shown that sequential removal of glycosylation sites in the mucin domain of EBOV GP led to a step-wise reduction in cell detachment suggesting that such modifications within GP are involved in downmodulation [12]. The O-linked glycosylation found on the EBOV GP mucin domain may promote an extended conformation as is seen for cellular mucin proteins [19] allowing this domain in GP to act as an approximately 150 residue long flexible rod that can protrude and mask epitopes in the immediate vicinity.
The ability of carbohydrate modification to protect epitopes on the surface of a viral glycoprotein is well established. Indeed, a glycan shield model has been proposed for other viral glycoproteins, most notably HIV, as a mechanism to avoid host immune recognition [43]. An extended glycosylated protrusion provided by the mucin domain may be a characteristic feature that distinguishes the “glycan umbrella” of EBOV GP from other viral glycoproteins where the glycan shield does not cause steric occlusion of host factors. Another feature of the proposed model is that EBOV GP must localize in close proximity to the affected proteins; perhaps within plasma membrane microdomains inhabited by the host proteins. This requirement may explain the critical threshold for the observed GP effects as well as the variety of proteins regulated by EBOV GP. It may be that the ability to occupy these microdomains is, in addition to the extensive carbohydrate modification, a characteristic feature of EBOV GP. Based upon our results it appears likely, therefore, that the heavily glycosylated EBOV GP acts as a glycan umbrella to physically occlude access to nearby host proteins, and GP itself, thereby impairing host protein function.
It is intriguing to consider the role in EBOV replication or pathogenesis of GP-induced steric occlusion of surface proteins. Based upon our observations of proteins at the plasma membrane it is plausible that EBOV GP functions to shield epitopes on the surface of virions thereby contributing to infection and/or persistence in the natural reservoir. Notably the KZ52 monoclonal antibody employed in these studies is neutralizing but fails to protect nonhuman primates from EBOV infection [44], [45]. Perhaps variation in GP density on virions produced in vivo differentially affects the neutralization sensitivity of viruses in nonhuman primates. Additionally, the ability of GP to mask MHC1 and other molecules on the cell surface, coupled with the inhibitory effect of GP on cell-cell adhesion, may be a strategy for avoiding CD8 T cell-mediated killing of EBOV infected cells. Our data demonstrating that GP-expressing cells do not effectively activate CD8 T cells supports this hypothesis. Interestingly, this mechanism is proposed for adenocarcinomas, in which cellular mucin protein overexpression can result in metastasis due to loss of adhesion, and has been shown to prevent recognition and killing by NK and cytotoxic T cells [38], [46], [47]. However, the rapid time course of EBOV infection and its impairment of adaptive responses may render escape from CD8 cells unnecessary in humans. Instead, protection from NK cells may be more important and the ability of EBOV GP to effect NK cell recognition should be explored. Alternatively, the ability to mask MHC1 may be more critical for viral infection or persistence in the natural reservoir for EBOV. Finally, it is known that the interface between the innate and adaptive immune response is affected during EBOV infection (reviewed in [2]). We have previously shown that EBOV GP causes rounding in macrophages [12]. It is possible that EBOV GP shielding and inhibition of adhesion molecules or other immune regulatory proteins on professional antigen presenting cells such as macrophage or dendritic cells plays a role in the immune dysfunction characteristic of EBOV infection.
For GP studies, cDNA encoding the membrane-anchored form of Zaire EBOV GP (Mayinga strain, accession number U23187) was used. For AU1 tagged GPs, the amino acids, DTYRYI were added using linker insertion into GP that had been engineered to have a unique XhoI site at position 312 encoding the amino acids LE (NmucAU1) and a unique NotI site replacing amino acid 463 with the amino acids KRPL (CmucAU1). EBOV GP harboring mutations in the endoproteolytic site, GP cl(-), has been previously described [20]. All constructs were cloned into the pCAGGS expression vector.
293T and HeLa cells were cultured in DMEM (Gibco) with 10% fetal bovine serum (HyClone) and penicillin/streptomycin (Gibco) at 37°C with 5% CO2. For flow cytometry and Western blotting, 293T cells were plated in 10 cm or 6-well plates one day prior to transfection. Cells were transiently transfected by Lipofectamine 2000 according to manufacturer's directions with 30 µg or 4 µg DNA per 10 cm plate or 6-well, respectively. Immunofluorescence microscopy was performed using HeLa cells that were plated on glass coverslips in 24-well plates and transfected with 1.5 µg DNA as above.
Purified CD8 T cells from normal donors were obtained from the University of Pennsylvania Center for AIDS Research Immunology Core under a University of Pennsylvania IRB approved protocol. The human ovarian adenocarcinoma line OV79 has been described previously [48]. To create the OV79-SL9 antigen-presentig cells, OV79 cells were sequentially transduced to express HLA-A*02 [49] and a construct of GFP fused to a codon-optimized sequence of HIV-1 p17 Gag50–102. High titer lentiviral vectors were produced as described previously [50].
Primary human CD8 T cells were cultured in X-Vivo 15 (Lonza) supplemented with 5% HABS (Valley Biomedical, Winchester, VA), 2 mM GlutaMax and 25 mM HEPES (Invitrogen). CD8 T-cells were transduced to express the SL9-specific HLA-A2 restricted 869TCR as described previously [51]. Transduction efficiencies were assessed by flow cytometric analysis of TRBV5-6 staining (anti-Vbeta5a, Thermo-Fisher) or HLA-A*02- SL9 tetramer stain (Beckman Coulter Immunomics).
OV79-SL9 cells were plated at 16,000 cells/well on 48 well plates. After an overnight incubation cells were transduced with adenovirus expressing GFP (Ad GFP) or GFP and the EBOV Zaire glycoprotein (Ad GP) as described previously [12]. Briefly, adenoviruses were diluted in media and applied to cells at an MOI of 300. Media alone was used as a control. 48 h after transduction, target cells were analyzed for GFP and HLA expression. Floating and adherent cells, lifted by incubation with versene, were combined and stained for HLA-ABC or isotype control with APC-conjugated antibodies (BD-Biosciences). Alternatively, cells were stained for different MHC1 epitopes with W6/32 (eBiosciences), YTH862.2 (Santa Cruz Biotechnology), BB7.2 (BD Pharmingen), or GJ14 (Chemicon) primary antibodies, flowed by Alexa Fluor 647 (Invitrogen) secondary antibodies. 10,000 viable (forward scatter versus side scatter) events were collected on an LSR-II flow cytometer running BD FACSDiva-6 (BD-Biosciences), and analyzed in FlowJo (Tree Star Inc.).
SL9-specific TCR–transduced CD8 T cells were mixed with unmodified or adenovirally transduced OV79-SL9 target cells at a 2∶1 ratio for 1 h, followed by 4 h in the presence of brefeldin-A (Golgiplug, BD Biosciences). Stimulation with TPA (3 mg/ml, Sigma-Aldrich) and ionomycin (1 mg/ml; Calbiochem) with brefeldin-A was used as positive control. Cells were washed in PBS and surface-stained using CD8 conjugated to APC-H7, and then fixed and permeabilized with the Caltag Fix & Perm kit (Invitrogen) and stained using anti-TRBV5-6 FITC and macrophage inflammatory protein-1b (MIP-1b, CCL4)-PE. Sequential gates of 10,000 viable (forward scatter versus side scatter), CD8 positive events were acquired for all conditions on an LSR-II flow cytometer running BD FACSDiva-6 (BD-Biosciences). Data were analyzed for cytokine production in FlowJo (Tree Star Inc.).
Transfected cells were removed by resuspension in the culturing media. Cells were pelleted at 4°C for 3 min at 1300×g. Pellets were resuspended in 1% Triton X-100 or RIPA buffer with complete protease inhibitor cocktail (Roche) for 5 minutes. Lysates were cleared by centrifugation at 4°C at 20,800×g. 30 µl samples were mixed with reducing SDS buffer, boiled for 5 minutes, and separated on a 4–15% Criterion PAGE gel (Bio-Rad). Proteins were transferred to PVDF (Millipore) at a 400 mA constant current. Membranes were blocked in 5% milk in TBS. Membranes were probed with rabbit polyclonal anti-GP sera which recognizes the GP1 subunit [52], rabbit anti-AU1 antibodies (Bethyl labs), or anti-GAPDH monoclonal antibodies (Calbiochem) in blocking buffer. Protein was detected with stabilized goat anti- rabbit or mouse HRP conjugated antibodies (Pierce) in blocking buffer. Membranes were visualized with SuperSignal Femto substrate (Pierce).
293T cells were detached from the plate 24 hours post transfection with PBS lacking Ca++ and Mg++ (−/−), 0.5 mM EDTA and combined with floating cells in culture media. Alternatively, floating cells in cluture media were removed and used exclusively (where indicated). Cells were pelleted at 4°C at 250×g, then resuspended in flow wash buffer (PBS −/− with 1% bovine calf serum and 0.05% NaAzide) and aliquoted for staining. For detection of EBOV GP, cells were stained with the human MAb, KZ52 [45] and detected with FITC anti-human IgG (PharMingen). For detection of AU1 epitopes, cells were stained with rabbit polyclonal anti-AU1 antibodies (Bethyl labs) and detected with FITC goat anti-rabbit IgG (Rockland). For detection of β1 integrin, cells were stained with anti-human CD29 PE-Cy5 conjugate (eBioscience); for detection of MHC1, cells were stained with anti- HLA-ABC PE-Cy5 conjugate (eBioscience). For intracellular staining, cells were permeabilized using Cytofix/Cytoperm (BD Biosciences) for 20 min on ice, followed by washing with Permwash (BD Biosciences). Antibodies where then diluted in Permwash buffer. For detection of GM130 and calnexin, mouse monoclonal FITC-conjugated antibodies were used (BD Transduction Labs). All staining was performed on ice, followed by washing. Live cell gates were drawn based on forward and side scatter. For each sample, 10,000 or 20,000 events in the live cell gate were collected and analyzed. Data were collected on a Becton Dickinson FACSCalibur and analyzed using FlowJo software (Tree Star, Inc.).
For HeLa cells, media was removed at 24 hours post-transfection, cells were washed with PBS and fixed with 3% PFA in PBS for 20 minutes. For non-adherent 293T cells, media containing floating cells was removed from plate, then centrifuged onto poly-D-lysine coated coverslips (BD Biosciences), then fixed. All samples were then washed with PBS, then permeabilized with 0.2% saponin, 1% goat serum in PBS for 5 minutes, then washed with PBS. Cells were blocked with 10% goat serum, 0.1% Tween-20 in PBS for 2 hours. For GP staining, coverslips were incubated with mouse anti-EBOV GP MAb 42/3.7 (gift from Yoshihiro Kawaoka) and detected with goat anti-rabbit Alexa Fluor 594 antibodies (Invitrogen). For AU1 staining, coverslips were incubated with rabbit anti-AU1 antibodies (Bethyl labs) and detected with anti-rabbit Alexa Fluor 488 antibodies (Invitrogen). Cells were washed with PBS after each staining step. Coverslips were mounted on glass slides with mounting medium containing DAPI (Vectasheild). Z-section images were collected on a Leica DMRE fluorescence microscope using Open Lab software (Improvision). Thirty z-sections per image were collected at 0.2 µm intervals. Z-section data were deconvoluted using Velocity software (Improvision) to a 98% confidence level or 15 iterations. Images shown are single, deconvoluted, z-sections.
At 24 hours post-transfection, sodium azide was added to 0.1% and 2-deoxy glucose was added to 10 mM. Cells were incubated an additional 30 min. Cells were then harvested and resuspended in flow wash buffer supplemented with 0.1% azide and 10 mM 2-deoxy glucose. DTT was then added to 150 mM and cells were incubated at 37°C for 20 minutes. Cells were then pelleted at room temperature and the supernatant was removed and blotted for GP as described above. Cells were then washed twice in flow wash and stained for flow cytometry as described above.
At 24 hours post-transfection, floating cells were harvested and resuspended in 100 µl flow wash buffer. 100 U of neuraminidase (NEB) and/or 1000 U of PNGaseF (NEB) was then added. Cells were then incubated at 37°C for 20 minutes. Cells were then washed twice and aliquoted for flow cytometry or Western blotting as described above. Alternatively, cells were incubated with 2 mM benzyl-α-GalNAc (Sigma) or DMSO at 31°C for 48 hours. Cells were then given fresh media with 2 mM benzyl-α-GalNAc or DMSO and cultured at 37°C for 1 hour. Cells were then transfected as described above. At 24 hours post-transfection, floating and adherent cells were harvested and resuspended in 100 µl flow wash buffer. 1000 U of PNGaseF (NEB) or 2.5 mU of O-glycosidase (Sigma) was then added. Cells were then incubated and analyzed as above. For PNGaseF treatment of cell lysates, 30 µl of lysate was incubated with glycoprotein denaturing buffer (NEB) for 10 minutes at 60°C. Samples were then incubated with G7 buffer, NP40, and 500 U PNGase F (NEB) for 2 hours at 37°C, then blotted for GP as described above.
The authors would like to thank Christian Fuchs for technical assistance, Erica Ollmann Saphire and Dennis Burton for providing the KZ52 antibody, Yoshihiro Kawaoka for providing anti-GP monoclonal antibodies, and Andrew Rennekamp and Rachel Kaletsky for helpful discussion.
|
10.1371/journal.ppat.1002362 | The Human Herpesvirus-7 (HHV-7) U21 Immunoevasin Subverts NK-Mediated Cytoxicity through Modulation of MICA and MICB | Herpesviruses have evolved numerous immune evasion strategies to facilitate establishment of lifelong persistent infections. Many herpesviruses encode gene products devoted to preventing viral antigen presentation as a means of escaping detection by cytotoxic T lymphocytes. The human herpesvirus-7 (HHV-7) U21 gene product, for example, is an immunoevasin that binds to class I major histocompatibility complex molecules and redirects them to the lysosomal compartment. Virus infection can also induce the upregulation of surface ligands that activate NK cells. Accordingly, the herpesviruses have evolved a diverse array of mechanisms to prevent NK cell engagement of NK-activating ligands on virus-infected cells. Here we demonstrate that the HHV-7 U21 gene product interferes with NK recognition. U21 can bind to the NK activating ligand ULBP1 and reroute it to the lysosomal compartment. In addition, U21 downregulates the surface expression of the NK activating ligands MICA and MICB, resulting in a reduction in NK-mediated cytotoxicity. These results suggest that this single viral protein may interfere both with CTL-mediated recognition through the downregulation of class I MHC molecules as well as NK-mediated recognition through downregulation of NK activating ligands.
| The long coevolution of herpesviruses with their hosts has resulted in the development of a diverse array of viral immune evasion strategies and host counter-strategies. The identification of viral proteins that impair the function of cellular immune-recognition receptors has proven fertile ground for the discovery of fundamental concepts in immunology and cell biology. While the cytomegaloviruses have demonstrated an extraordinary array of immunoevasive tactics, little is known about the immunoevasive strategies of the closely-related human herpesvirus-7 (HHV-7). We have previously demonstrated that the U21 gene product from HHV-7 likely interferes with viral antigen presentation to cytotoxic T cells by rerouting class I major histocompatibility molecules to lysosomes for degradation. In addition to the host's cytotoxic T cell response, virus infection also induces the expression of Natural-Killer (NK) activating ligands, alerting cytotoxic NK cells to identify and kill virus-infected cells. Here we describe a novel function for the same viral protein - U21 - in interfering with NK cell recognition. Our findings provide the first indication that HHV-7, too, may have found it necessary to strategize mechanisms of NK escape.
| Human herpesvirus-7 (HHV-7) is a T-lymphotrophic beta-herpesvirus, most closely related to human herpesvirus-6 (HHV-6) and human cytomegalovirus (HCMV). HHV-6 and -7 share many biological properties: HHV-6 and -7 possess genomes that are almost entirely colinear, and both HHV-6 and -7 can cause the formation of giant multinucleated cells in culture, features reminiscent of those seen in HCMV infection in vitro. Primary infection with either of these viruses results in a short febrile illness, and more than 90% of adults are seropositive for both HHV-6 and HHV-7 [1].
Like all herpesviruses, HHV-7 remains latent or establishes persistent lifelong infections in its host. In so doing, herpesviruses have evolved numerous strategies to evade immune detection. Most herpesviruses, including HHV-7, have evolved mechanisms to interfere with viral antigen presentation by class I MHC molecules (for review see [2]–[4]). Although preventing surface expression of class I MHC molecules may be an effective means of escaping CTL detection, the absence of class I products from the cell surface may render the host cell susceptible to Natural Killer (NK) cell attack (for review, see [5]).
Activation of NK cells is regulated by the balance of inhibitory and activating signals received through cell surface NK receptors (for review, see [6], [7]). NK inhibitory receptors bind to classical and non-classical class I MHC molecules. NK activating receptors bind to NK activating ligands, some of which are structurally similar to class I MHC molecules. When an NK cell encounters a potential target cell, it is thought to integrate the activating and inhibitory signals it receives; if activating signals prevail, the NK cell can then directly kill its target. In response to microbial infection or other cell stressors, cells can increase the surface expression of NK activating ligands, improving the likelihood that NK cells recognize and kill cells that become harmful. Viral strategies to remove inhibitory ligands (class I MHC molecules) from the cell surface of an infected cell might further skew the balance in favor of NK killing. Not surprisingly, viruses have also evolved counter-strategies to interfere with NK engagement (for review, see [8], [9]). For example, presumably to escape NK detection, several viruses selectively downregulate HLA-A and HLA-B locus products, while leaving HLA-C, -E, and other non-classical class I MHC molecules at the plasma membrane as inhibitory ligands for NK cell receptors (for review, see [5], [10]).
In addition to the selective downregulation of NK-inhibitory HLA molecules, another strategy employed by viruses to escape NK engagement is the downregulation of NK activating ligands from the cell surface. For example, the HCMV immunoevasin UL16 was found to bind to two members of a family of cellular proteins termed UL16-binding proteins, or ULBPs [11]. UL16 was also found to associate with a protein called MICB, for MHC class I chain-related protein family [11]. Both the MICs and the ULBPs are activating ligands for the same NK activating receptor, NKG2D, and both MICs and ULBPs share structural similarity with class I MHC molecules (see schematic depicting the general structure of these ligands, Figure 1, panel A) [12]–[14]. HMCV UL16 binds to ULBP1, ULBP2, ULBP6 (RAET1L) and MICB and retains these activating ligands intracellularly, reducing NK recognition of HCMV-infected cells [15]–[18]. Indeed, the NKG2D ligands are frequent targets of viral immunoevasins, underscoring the importance of these ligands in anti-viral immunity. Both murine and human cytomegaloviruses encode multiple immunoevasins that affect NK activating ligands. In addition to UL16, HCMV UL142 retains MICA in the Golgi [19], and MCMV m145, m152, and m155 impair the cell surface expression of murine NKG2D ligands [20]–[23]. Adenovirus E3/19K can also sequester MICA and MICB intracellularly [24], and the KSHV K5 ubiquitin ligase promotes the degradation of MICA and MICB [25], [26]. In addition, HCMV, KSHV, and EBV all encode microRNAs that target MICB mRNA, reducing its expression [27].
We have shown previously that HHV-7 encodes a gene product, U21, that can associate with and affect the surface expression of all classical class I MHC gene products (HLA-A, -B, and -C) by rerouting them to lysosomes, where they are degraded [28]. U21 can also bind to and downregulate the non-classical class I MHC gene products (HLA-E and -G)[28]. Such a comprehensive downregulation of NK-inhibitory class I molecules should shift the balance of ligands toward one that would activate NK cells, alerting them to HHV-7 infection. For HHV-7 to succeed, then, it must also encode a means to escape engagement by NK cells.
Because U21 can associate with and affect the surface expression of such a wide variety of class I MHC gene products, we hypothesized that U21 might also affect the structurally-related activating NKG2D ligands. Class I MHC molecules are composed of three domains, α1, α2, and α3, and they assemble with a light chain, β2-microglobulin (β2m), and peptide. Like class I MHC molecules, MICA and MICB are also type I membrane proteins that contain α1, α2, and α3 domains, but they do not associate with β2m or peptide (see schematic, Figure 1, panel A) [29], [30]. The NKG2D ligands share structural homology, but little amino acid identity with class I MHC molecules; MICA, for example, shares an average of only ∼29% amino acid identity with the HLA-A2 class I heavy chain [11]. The ULBP proteins possess α1 and α2 domains, but lack an α3 domain, and rather than transmembrane domains, ULBP1–3 are GPI-anchored proteins [11].
Here we show that U21 can associate with the NKG2D ligand ULBP1 and redirect it to lysosomes for degradation. U21 can also reduce the surface expression of two other NKG2D ligands, MICA and MICB, resulting in protection from NK cytotoxicity.
Constitutive levels of NKG2D ligands are low in most cells, thus, to examine the effect of U21 upon the surface expression of the NK ligands, we generated U373 astrocytoma cell lines individually stably expressing ULBP1, ULBP2, ULBP3, MICA, or MICB, using retrovirus-mediated gene transfer. We then stably expressed U21 in each of the NK-ligand-expressing cell lines. Each cell line expressed similar levels of U21, as assessed by immunoblot (Figure 1, panel B). We next investigated the effect of U21 expression on the surface expression of each NKG2D ligand using flow cytometry. U21 expression resulted in a slight downregulation of ULBP1 and ULBP3 from the cell surface, while surface levels of ULBP2 were essentially unchanged (Figure 1, panels C - E, compare light gray shaded traces to black traces). The NK ligands MICA and MICB, however, were more markedly reduced by U21 (Figure 1, panels F and G). Similar results were seen in U21-expressing K562 cells, as discussed later.
U21 also binds to and reroutes class I MHC molecules to lysosomes, reducing their presence at the cell surface. When we examined U21's effect upon the surface expression of endogenous class I MHC molecules in each of the NK ligand-expressing cell lines, endogenous class I molecules were more effectively downregulated by U21 than were any of the NK ligands (Figure 1, panels H-L, compare light gray shaded traces to black traces). Of note, the surface expression of endogenous class I MHC molecules is similarly affected in each of the cell lines, with the majority of U21-expressing cells showing reduced surface class I levels, and a smaller population of cells exhibiting normal surface levels of class I molecules. This broad distribution reflects the variable levels of U21 expression among individual cells; in general, we find that cells expressing higher levels of U21 possess less class I MHC, ULBP1, or MIC on their cell surface (Figure S1). The similar pattern of surface class I MHC downregulation within each population of cells serves as a second, indirect measure of the similar levels of U21 expression in each of the stable cell lines. These results suggest that, in addition to class I MHC molecules, U21 may also impair the surface expression of the NKG2D ligands MICA and MICB. To a far lesser extent, U21 affects the surface expression of the ULBPs.
U21 expression results in a dramatic steady-state redistribution of class I MHC molecules from the cell surface to the lysosomal compartment (Figure 2, panels A and B). Although U21 accompanies class I MHC molecules to lysosomes, and both molecules are degraded there, U21 does not colocalize with class I MHC molecules in lysosomes [30]. Instead, U21 is localized in the ER/Golgi (Figure 2, panel C, [31], [32]). Explanations for this phenomenon include the possibility that U21 may be more sensitive to lysosomal proteases, such that we are not able to visualize a concentration of U21 in lysosomes. However, incubation in lysosomal protease inhibitors does not significantly alter the steady-state distribution of U21. It is possible the epitopes recognized by our antibodies, all located within the cytoplasmic tail of U21, may become masked upon U21's arrival in lysosomes, accounting for our inability to visualize U21 in lysosomes. Perhaps most likely is the possibility that U21 exists in far greater abundance in the ER and Golgi, such that we cannot see U21 in lysosomes over the brilliant ER/Golgi labeling. We have previously discussed this differential localization of U21 with its client class I MHC molecules in greater detail [30].
Because U21 reduced the surface expression of ULBP1, ULBP3, MICA, and MICB, we next assessed whether expression of U21 might lead to relocalization of the NK ligands within the cell, just as it does for class I MHC molecules. Like endogenous class I MHC molecules, in the absence of U21, the NK-ligands were localized primarily on the plasma membrane, with some labeling of the Golgi, as the molecules traverse the biosynthetic pathway (Figure 2, left panels). U21 expression did not result in appreciable relocalization of ULBP2 and ULBP3, (Figure 2, panels G-L), even though U21 expression resulted in slight reduction of ULBP3 surface expression (Figure 1, panel E). MICA and MICB, on the other hand, seemed to disappear in cells expressing U21 (Figure 2, panels N,O and Q,R asterisks). In cells expressing U21 (Figure 2, panel F), ULBP1 was relocalized to a punctate compartment resembling U21-relocalized class I MHC molecules (Figure 2, compare panels E (asterisks) to B). Redistribution of ULBP1 molecules correlated with the level of U21 expression in the cell; in Figure 2, panels E and F show a population of cells exhibiting heterogeneous expression of U21. In cells with intense U21 labeling (asterisks), ULBP1 punctae are more readily apparent. We therefore examined whether ULBP1 in U21-expressing cells was localized in a lysosomal compartment. Double-label immunofluorescence microscopy showed colocalization between ULBP1 and the lysosomal membrane protein lamp2 in U21-expressing cells (Figure 3).
In cells expressing U21, class I MHC molecules are degraded in a lysosomal compartment [32]. Since the relocalization of ULBP1 resembled the relocalization of class I MHC molecules, we next asked whether U21 expression also resulted in the lysosomal degradation of ULBP1, using pulse-chase analysis of ULBP1 in the presence of lysosomal protease inhibitors. Stabilization of a protein in the presence of lysosomal protease inhibitors would suggest a role for lysosomal proteases in the turnover of that protein.
We first performed pulse-chase analysis of ULBP1 in the absence of U21. After a 15-minute pulse label of ULBP1-expressing U373 cells, ULBP1 migrated at approximately 31 kDa (Figure 4, panel A, lane 1). At the 2- and 6-hour chase points, ULBP1 migrated more slowly, at ∼37 kDa (Figure 4, panel A, lanes 2 and 3). ULBP1 possesses a single N-linked glycosylation consensus site, thus the ∼6 kDa increase in ULBP1 at the later chase points is either the result of additional modifications to its single predicted N-linked glycan, or possibly O-linked glycosylation.
Surprisingly, we recovered ∼5-fold less labeled ULBP1 immediately following the pulse label (0 hr chase) than after the 2- and 6-hour chase periods (Figure 4, panel A, compare lanes 1 to 2 and 3). One possible explanation for the difference in recovery of ULBP1 is that the ULBP1 monoclonal antibody can more easily recognize ULBP1 after it has moved beyond the ER and Golgi; the immunoprecipitations were performed under non-denaturing conditions, thus it is possible that the epitope recognized by the anti-ULBP1 antibody was partially masked while ULBP1 was in the ER, perhaps as a result of protein complex formation during addition of the GPI anchor. Protein complex formation during addition of the GPI anchor might also explain why we observe no Endo H-sensitive GPI-linked UBLP1 after the pulse label. Alternatively, antibody recognition may become enhanced by modifications to the N-linked glycan that occur in the Golgi compartment or structural changes that occur after addition of the GPI anchor.
To determine whether the size difference between the two species of ULBP1 seen in lanes 1 and 2 was solely attributable to modifications to its predicted N-linked glycan, we digested the immunoprecipitated proteins with either endoglycosidase H (Endo H) or peptide N-glycosidase:F (PNGase:F). Endo H cleaves N-linked glycans found on proteins in the ER and early Golgi. Glycoproteins that have progressed beyond the medial Golgi become resistant to digestion with Endo-H, thus resistance to Endo H can be used to follow the movement of a protein through the secretory pathway. After the 15 minute pulse label, Endo H digestion of ULBP1 resulted in a more rapidly migrating polypeptide of ∼28 kDa, consistent with its predicted molecular weight in the absence of N-linked glycans (Figure 4, panel A, compare lanes 1 and 4). This Endo H-sensitive form represents newly-synthesized ULBP1 present in the ER or early Golgi. At the later chase points, the 37 kDa form of ULBP1 was resistant to Endo H digestion, suggesting that this form of ULBP1 has progressed beyond the medial Golgi (Figure 4, panel A, lanes 5 and 6).
PNGase:F cleaves N-linked glycans from all polypeptide chains, regardless of oligosaccharide processing within the secretory pathway. After the 15 minute pulse label, PNGase:F digestion of ULBP1 resulted in a polypeptide of ∼28 kDa, co-migrating with the 28 kDa Endo H-digested form (Figure 4, panel A, compare lanes 7 and 4). If the slower migration of the 37 kDa form of ULBP1 were solely the result of N-glycosylation, the mobility of the products of Endo H and PNGase:F digestion should be the same – migrating at 28 kDa. Instead, the 37 kDa Endo H-resistant form of ULBP1 was reduced to a polypeptide of ∼31 kDa after PNGase:F digestion, suggesting a separate post-translational modification to ULBP1 (Figure 4, panel A, lanes 8 and 9). ULBP1 is a GPI-anchored protein [11]. It is possible that the slower mobility of ULBP1 may reflect the addition of the GPI anchor. However, since glypiation occurs in the ER, before glycoproteins become Endo H-resistant, we should observe the 31 kDa Endo H-sensitive GPI-linked ULBP1 after the initial pulse label, and we do not. It is also possible that the PNGase:F-resistant form of ULBP1 is O-glycosylated.
Having established the normal trafficking of ULBP1 in the absence of U21, we next performed pulse-chase experiments in U21-expressing cells. To investigate whether U21 expression resulted in the destabilization of ULBP1, we performed pulse-chase experiments in the presence of the lysosomal protease inhibitor leupeptin, and the vacuolar H+-ATPase inhibitor folimycin. In cells lacking U21, ULBP1 appeared stable throughout the 6 hr chase period, and lysosomal inhibitors did not significantly increase its stability (Figure 4, panel B, compare lanes 2 and 3 to 5 and 6). In cells expressing U21, however, we recovered fewer labeled ULBP1 molecules after the 6-hour chase period, and in the presence of lysosomal inhibitors, ULBP1 was stabilized (Figure 4, panel B, compare lanes 8 and 9 to 11 and 12). Quantification of the stabilization of ULBP1 in the presence of lysosomal protease inhibitors is shown in Figure 4, panel C. We have shown previously that U21 and class I MHC molecules are also stabilized in the presence of leupeptin and folimycin [32], [33]. To ensure that the effectiveness of our lysosomal protease inhibitors was comparable to inhibition observed in prior experiments, we immunoprecipitated U21 molecules from the same cells, in parallel (Figure 4, panel B, lanes 13–16, compare lanes 14 and 16). We note that the stabilization of ULBP1 resulting from U21 expression is not as dramatic as U21-mediated stabilization of class I MHC molecules. Nonetheless, these results, and the relocalization of ULBP1 to lysosomes in U21 expressing cells (Figure 3), suggest that U21 can also reroute ULBP1 to lysosomes for degradation.
When we recovered U21 from metabolically-labeled ULBP1-expressing cells, we observed labeled co-precipitating class I heavy chains, but we did not observe co-precipitating ULBP1 molecules (Figure 4, panel B, lanes 13–16). Likewise, when we recovered ULBP1, we observed no labeled coprecipitating U21 molecules (Figure 4, panel B, lanes 7–12, migration position of U21 is shown in lanes 13–16). We therefore performed coimmunoprecipitation experiments using lysis buffer containing digitonin rather than TritonX-100. Under these conditions, when we recovered U21, we observed coprecipitation of two polypeptides identical in size to ULBP1 (Figure 5, panel A, lane 4). In the reciprocal immunoprecipitation, however, when we recovered ULBP1 from digitonin lysates, we failed to observe co-precipitation of U21. It is possible that the anti-ULBP1 antibody precludes co-precipitation of U21 with ULBP1.
To confirm the identity of the polypeptides coimmunoprecipitating with U21 as ULBP1 molecules, we recovered U21 molecules from cells expressing HA-tagged ULBP1 (Figure S2), and immunoblotted the immunoprecipitates with an antibody directed against the HA epitope tag. Because steady-state levels of proteins destined for the lysosomal compartment can be difficult to detect, we sought to minimize the lysosomal degradation of U21 and HA-ULBP1 by preincubating the cells in the presence of lysosomal protease inhibitors. In lysates from digitonin-lysed HA-ULBP1-expressing cells, we observed both the ER-resident, non-GPI-linked form of HA-ULBP1, as well as the GPI-linked mature form, with the mature form predominant (Figure 5, panel B, lanes 1 and 2). When we immunoprecipitated U21 from HA-ULBP1 cells expressing U21, the HA antibody recognized both forms of HA-ULBP1 in the anti-U21 immunoprecipitation (Figure 5, panel B, lane 4). Thus, U21 associates with both the ER- and mature forms of ULBP1 molecules. These results suggest that U21 binds to ULBP1 in the ER and maintains its association with ULBP1 through the secretory pathway en route to lysosomes.
The localization of the MICA and MICB in U21-expressing cells is unusual; rather than punctate lysosomal localization, cells expressing U21 exhibit very little MIC labeling at all (Figure 2, panels N, O, Q, and R, asterisks). Given the structural similarity between class I MHC molecules, ULBPs, and the MIC proteins, we thought it likely that MICA and MICB would also be rerouted to lysosomes for degradation, and surmised that perhaps the half-life of the MICs in the lysosomal compartment might be too short to allow visualization of the MIC proteins in punctae.
We therefore examined the turnover of MICB in the presence of lysosomal protease inhibitors. Because MICA and MICB share 83% similarity, and because both proteins are similarly affected by U21, we chose to examine MICB rather than MICA, since in general, MICB-expressing cells exhibited more dim, yet visible, punctae than MICA-expressing cells (Figure 2, compare panels N and Q).
In the absence of U21, after a 15 minute pulse-label, MICB migrated at a molecular mass of ∼60 kDa (Figure 6, panel A, lane 1). MICB became heterogeneously glycosylated (∼65–75 kDa) following a 2-hour chase period (Figure 6, panel A, lane 2), and no detectable MICB remained after 6 hours of chase (Figure 6, panel A, lane 3). Unlike class I MHC molecules, the cell surface expression of MICA and MICB is regulated by metalloproteinase cleavage, which results in shedding of the soluble extracellular domains from the cell surface, thus the half-life of MICB is likely the combined result of cleavage and release of soluble MICB by metalloproteinases [34]–[37] and of routine protein turnover [38]. Incubation of the MICB-expressing cells in lysosomal protease inhibitors resulted in slight stabilization of the MICB molecules (Figure 6, panel A, lanes 4–6, panel B, MICB), suggesting that some fraction of MICB is degraded in the lysosomal compartment.
In cells expressing U21, the turnover of MICB molecules was accelerated (Figure 6, panel A, lanes 7–9, and panel C), suggesting that, as for ULBP1 and class I MHC molecules, U21 might function to reroute MICB to the lysosomal compartment for degradation. However, lysosomal protease inhibitors stabilized MICB in U21 cells to approximately the same degree as in cells expressing MICB alone (Figure 6, panel A, lanes 10–12, and panel B), suggesting that U21 does not enhance the lysosomal degradation of MICB. To ensure that our lysosomal protease inhibitors were active, we recovered U21 from the same MICB-U21-expressing lysates, and found the lysosomal protease inhibitors were successful at stabilizing U21 and class I MHC molecules (Figure 6, panel A, lanes 13–18).
Because the half-life of MICB was reduced in U21-expressing cells, we performed immunoblot analysis to examine the steady-state levels of MICB. In the absence of U21, we detected three polypeptides with an anti-MICB antibody (Figure 7, panel A, lane 1). The uppermost band corresponded in size to the mature form of MICB, and is the predominant form. The middle band corresponded in size to the immature ER form, and the lower band corresponded in size to soluble “shed” MICB. Expression of U21 resulted in a reduction in the steady-state level of MICB, primarily of the mature form (Figure 7, panel A, lane 2, top bands), suggesting that U21 expression results in the degradation of MICB after it has acquired Endo H-resistance, later in the secretory pathway.
We envisioned two possible explanations for the reduction in steady-state levels of MICB: U21 might enhance MICB degradation by either lysosomal or proteasomal proteases. However, neither lysosomal hydrolases nor proteasomal proteases appeared to participate appreciably in U21-mediated degradation of MICB (Figure 6, panels A and B, data not shown). Alternatively, we hypothesized that U21 might accelerate MICB's cleavage and release from the cell, resulting in the appearance of reduced steady-state levels of MICB. To examine the effect of U21 expression on the amount of secreted MICB in the supernatants, we examined the steady-state levels of MICB in supernatants of MICB- and MICB-U21-expressing cells. Rather than elevated levels of secreted MICB, we found a reduction in the amount of MICB present in the supernatant of U21-expressing cells, instead suggesting that U21 impairs the release of MICB (Figure 7, panel A, lanes 3 and 4).
To further evaluate the effect of U21 on MICB secretion, we performed a pulse-chase experiment over a shorter 4-hr chase period, recovering MICB from both lysates and supernatants after 0, 0.5, 1, 2, and 4 hours of chase (Figure 7B). The turnover of MICB was rapid; after 4 hours, there was very little labeled MICB recoverable from either control or U21-expressing cell lysates (Figure 7B, panel i, lanes 5 and 10). To examine shedding of MICB into the medium, we recovered MICB from the medium by immunoprecipitation and subjected it to SDS-PAGE. While we detected labeled secreted MICB in the medium of MICB-expressing cells (Figure 7B, panel ii, lanes 1–5), we recovered almost no MICB in the medium of U21-expressing MICB cells (Figure 7, panel ii, lanes 6–10), further suggesting that U21 impairs the shedding of MICB into the medium.
In cells expressing U21, the heterogeneously glycosylated MICB migrated slightly faster than MICB from control cells (Figure 7B, panel i, compare lanes 4 and 7), suggesting that U21 expression affected the trimming of either the MICB core polypeptide or its N-linked sugars. To determine whether U21 expression induced a reduction in the core polypeptide size of MICB, we digested MICB immunoprecipitates with PNGase:F (Figure S3). PNGase:F digestion resulted in polypeptides of identical mobility, suggesting that the core polypeptide remained unchanged, and that the U21-induced increase in MICB mobility must be the result of altered post-translational modifications to its N-linked glycans.
To examine whether the U21-induced downregulation of surface MICA and MICB could protect cells from NK recognition, we next performed NK cytotoxicity assays. NK-mediated cell lysis depends on the integrated response of NK cells to both inhibitory and activating ligands. Since U21 affects both class I MHC molecules (NK inhibitory ligands) and the MIC proteins (NK activating ligands), analysis of NK cytoxicity toward U21-expressing cells is complicated. To simplify the assessment of U21's effect upon MICA and MICB, we expressed U21 in the erythroleukemic cell line K562, which lack class I MHC molecules, thus any effect of U21 on NK cytotoxicity toward K562 cells should be independent of U21's ability to downregulate the surface expression of class I MHC molecules.
For these cytotoxicity assays, we generated a population of K562 cells stably expressing U21 (See Figure S4). When we examined the surface expression of the endogenous NKG2D ligands in the U21-expressing K562 cells, similar to our results in U373 cells, we observed a very slight decrease in the surface expression of endogenous ULBP1 in the U21-expressing K562 cells (Figure 8, panel A), and an even more prominent decrease in the surface expression of MICA and MICB (Figure 8, panels D and E). Surface levels of endogenous ULBP2 also appeared slightly reduced in the K562 cells, while the endogenous surface ULBP3 appeared to be unaffected (Figure 8, panels B and C), as was the surface expression of ICAM-1, an adhesion molecule critical for synapse formation between the NK and target cell (Figure 8, panel F) [39] (for review see [40]).
To further investigate whether the effects of U21 expression upon MICB were similar in both K562 and U373 cells, we also examined the half-life of MICB in the U21-expressing K562 cells. Because MICB expression in K562 cells is inherently low, we were unable to detect endogenous MICB by immunoblotting. We therefore performed intracellular labeling of MICA or MICB, comparing MIC expression levels using flow cytometry. Steady-state levels of MICA and MICB were reduced in K562 cells expressing U21 (Figure 8, panels G and H), suggesting that U21 expression can also result in the destabilization of MICA and MICB in K562 cells. In contrast, steady state levels of ICAM-1 (data not shown) or the transferrin receptor remained unaffected by U21 expression in these cells (Figure 8, panel I).
To evaluate the sensitivity of the U21-expressing K562 cells to NK cytotoxicity, we incubated the target K562 cells in the presence of effector NKL cells, and determined NK cell cytotoxicity using a flow cytometric assay. While control K562 cells were sensitive to NK cytotoxicity, cells expressing U21 were resistant to NK lysis (Figure 9, panel A). U21-expressing cells were also resistant to NK cytotoxicity from peripheral blood mononuclear NK cells (Figure S5).
To further define the mechanism of U21-mediated protection of K562 cells from NKL cytotoxicity, we performed NK cytotoxicity assays in the presence of blocking antibodies directed against MICA, MICB, and ULBP1. In control K562 cells, NK cytotoxicity was reduced by 15% in the presence of an antibody directed against MICA, by 25% in the presence of an antibody directed against MICB, and by 80% in the presence of both antibodies (Figure 9, panel B and D). This synergistic effect in the presence of blocking antibodies to both MICA and MICB suggest that the signal for NKL activation can be mediated through either ligand when one is unavailable, but when both are blocked, target cells are not able to activate the NKL cells. In contrast, pre-incubation of the K562 cells with an antibody directed against ULBP1 or an isotype control had no effect on cytotoxicity (Figure 9, panel D). Thus, the majority of NKL cytotoxicity between NKLs and K562 cells is likely mediated through engagement of MIC proteins on the surface of the target K562 cell with NKG2D on the NKL cells.
Blocking of both MICA and MICB on control K562 cells reduced the cytotoxicity to a level similar to that seen for U21-expressing K562 cells (Figure 9, panel C). Unlike control K562 cells, the cytotoxicity toward U21-expressing K562 cells was not further reduced by pre-incubation with antibodies directed against MICA or MICB (Figure 9, panel C). Since NKL-mediated cytotoxicity of K562 cells is largely mediated through MICA and MICB, the ability of U21 to reduce the surface expression of MICA and MICB likely contributes significantly to the mechanism by which U21 protects K562 cells from NK cytotoxicity.
If U21 can bind to and reroute ULBP1 molecules to the lysosomal compartment in U373 cells overexpressing ULBP1, why are endogenous ULBP1 surface levels in K562 cells only minimally reduced by U21, and why does ULBP1 seem to play no appreciable role in NK cytotoxicity toward K562 target cells? One possibility to explain the lack of NKL-mediated cytotoxicity through ULBP1 in the K562 cells may be the relative level of surface-expressed ULBP1 and its affinity for NKG2D. To explore the possibility that the surface expression of ULBP1 is insufficient to provide ligand for NKG2D, we overexpressed the ULBP1 activating ligand in K562 target cells. When ULBP1 was overexpressed, we observed an increase in NK cytotoxicity toward ULBP1 expressing cells (data not shown). We conclude that the constitutively-expressed ULBP1 on the surface of K562 cells may not be sufficient to induce NKL-mediated killing through engagement of NKG2D. ULBP1-expressing U373 cells, on the other hand, express abundant ULBP1, yet U21 has little effect upon the surface expression of ULBP1 in these cells. We therefore also think it likely that the affinity of U21 is lower for ULBP1 than for MICA and MICB. U21 does not act upon ULBP1 as effectively as it acts upon the MICs and class I MHC molecules.
HHV-7 U21 has long been known to bind to and reroute class I MHC molecules to the lysosomal compartment, likely providing HHV-7 a means of escaping detection by CTLs (2001). We have now demonstrated that U21 can reduce NK-mediated cytoxicity as well, by affecting surface expression of the NKG2D ligands MICA and MICB.
The MIC proteins are structurally similar to class I MHC molecules, and U21 expression resulted in the reduced surface expression of both the MICs and class I MHC molecules, thus we surmised that U21 acts upon the MICs and class I MHC molecules in a similar manner. However, we were unable to demonstrate association between U21 and the MIC molecules. Moreover, the MICs do not seem to be intensely localized to lysosomes in U21-expressing cells, nor is their stabilization by lysosomal protease inhibitors increased upon U21 expression. Why this difference? It is possible that U21 acts differently upon MICs than upon class I MHC molecules. Alternatively, the reagents available to detect the MIC molecules may not be sufficient to follow the MIC molecules as they traffic to lysosomes. The MIC family of NKG2D ligands is highly regulated. MIC expression at the transcriptional level is upregulated in response to stressors such DNA damage, autoimmunity, and infection (For review, see [41]). In cells that express MIC proteins, the trafficking of the MICs is also complex; MIC expression at the cell surface is regulated both by internalization and recycling, as well as by shedding of the MIC molecules into the medium [34], [37], [38]. Perhaps because of the relative complexity of the trafficking of MIC molecules, U21's molecular impact upon the MIC molecules is less straightforward.
Unlike class I molecules, even in the absence of U21, MICB is partially stabilized by lysosomal protease inhibitors. If U21 functioned to enhance the routing of MICB to lysosomes, we would have expected to observe enhanced stabilization of MICB by lysosomal protease inhibitors, but we did not (Figure 6). We therefore reasoned that if the MICB that disappears during the chase period is not degraded in lysosomes, then it must be shed or released from the cell. In control MICB-expressing cells, this indeed appears to be the case. In cells expressing U21, however, MICB appears to be neither degraded in lysosomes nor released into the supernatant (Figure 7). We can think of two possibilities to explain these observations: either MICB is shed, but the MICB shed from U21-expressing cells is less stable, explaining why we are unable to recover it from the medium, or, perhaps MICB interacts with U21 within the cell, rendering it undetectable by MICB antiserum. We also found that U21 expression resulted in altered glycosylation of MICB, thus perhaps this altered glycosylation of MICB may affect its stability upon release into the extracellular environment. Alternatively, it is possible that interaction of U21 with MICB precludes its shedding by the metalloproteases that cleave it [37].
Several similar trafficking studies examining the mechanism for HCMV immunoevasin UL16-mediated effect upon MICB have been performed [15], [42]. Like U21, UL16 localizes to the ER and Golgi region [15], [42], reduces the cell-surface expression of MICB [15], and impairs the release of soluble MICB into the medium [43]. But, as for U21, the complexity of MIC regulation and trafficking has impeded a clear understanding of how UL16 affects MICB trafficking [15], [42], [43].
The experiments described herein also illuminate some of the mechanistic and structural aspects of the interaction between U21 and class I MHC molecules: the discovery that U21 binds to the class I MHC-like ULBP1 contributes to our understanding of U21's interaction with class I MHC and class I-like molecules. U21 can associate with a wide range of class I MHC molecules, including HLA-A, -B, -C, -E, -G, and even murine class I MHC molecules [28], [32]. The greatest degree of conservation among all of these class I molecules exists within the α3 domain of these proteins, thus we originally hypothesized that the α3 domain was an important structural feature for association of U21 with these molecules [28] (see schematic, Figure 1, panel A). Interestingly, ULBP1 possesses α1 and α2 domains, but lacks an α3 domain. The structural similarities between class I MHC molecules and ULBP1 would therefore suggest that the binding of U21 to class I molecules involves the α1 and/or α2 domains.
U21 expression also results in the relocalization of ULBP1 to lysosomes. However, when U21 is expressed in a cell line that offers a choice of substrates - ULBP1 or class I MHC molecules - the choice is clear: U21 has a more striking affect upon class I MHC molecules than upon ULBP1, suggesting that U21 may possess greater affinity for class I MHC molecules than for ULBP1. It is important to note, however, that our experiments were performed in cells exogenously overexpressing ULBP1, or in K562 cells, which express low constitutive levels of ULBP1. In the context of an HHV-7 infection, it is possible that the relative expression of ULBP1 and class I MHC molecules may be such that U21 can act effectively upon ULBP1 molecules. However, when comparing the effect of U21 upon all of the NKG2D ligands, it is clear that U21 expression causes a much greater reduction in the cell surface expression of the MIC molecules than of the ULBPs, suggesting that U21's primary effect may be upon the MIC molecules.
U21's true utility as an immunoevasin during HHV-7 infection is, as yet, difficult to assess, because a bacterial artificial chromosomal system to facilitate genetic manipulation of the viral genome has not yet been established, and HHV-7 is a human herpesvirus for which there is no animal model. Additionally, although HHV-7 is known to infect T cells, the site of infection where immune escape is most critical for the virus is not certain; HHV-7 is shed in the saliva of healthy individuals, thus salivary glands may be a site of persistent infection where evasion of NK cytotoxicity is essential. It is also possible that, as for closely-related rhesus CMV, immunoevasin involvement in the escape of immune detection is important not during primary infection, but during superinfection with other strains [44].
HHV-7 U21 can bind to and affect the surface expression of many different HLA class I alleles, including the NK-inhibitory ligands HLA-C and HLA-E, downregulation of which, in principle, should render the cell susceptible to NK attack. We had therefore hypothesized that HHV-7 must encode other means of NK cell evasion [28]. We now demonstrate that HHV-7 U21 also reduces the surface expression of the NK activating ligands MICA and MICB, thereby preventing NK cytoxicity toward U21-expressing cells. Thus HHV-7, through a single viral protein, encodes a means to escape both CTL and NK cell detection. Interestingly, like U21, at least two other viral proteins can affect both CTL and NK cell recognition. The murine CMV m152 gene product (gp40) causes retention of both class I MHC molecules and NKG2D-ligands in the endoplasmic reticulum-Golgi intermediate compartment [21], [22], [45]. Additionally, unlike MCMV, which encodes multiple means of affecting both CTL and NK detection, Adenovirus is known to encode only one polypeptide, E3/19K, which can influence both CTL and NK recognition. E3/19K binds to class I MHC molecules and to MICA and MICB molecules and retains them in the ER [24], [46], [47]. Thus, MCMV gp40, Adenovirus E3/19K, and HHV-7 U21 all recognize multiple structurally-similar class I MHC and class I MHC-like molecules and may possess dual function during viral infection.
U373 and HEK293T were cultured in Dulbecco's modified Eagle medium (DMEM) supplemented with 5% newborn calf serum (NCS) and 5% fetal bovine serum (FBS) in the presence of puromycin (375 ng/ml) (Sigma-Aldrich, St. Louis, MO) or geneticin (G418) (500 ng/ml)(Invitrogen, Carlsbad, CA), as needed. K562 cells were cultured in RPMI supplemented with 10% FBS. NKL cells (generously provided by Dr. M. J. Robertson, Indiana University) were cultured in RPMI supplemented with 10% heat-inactivated FBS, 1 mM sodium pyruvate and 50–100 U/ml IL-2. U373 cells stably expressing the NKG2D ligands ULBP1, ULBP2, or ULBP3, or MICA or MICB were generated by retroviral transduction using the vector, pLNCX (Clontech, Mountain View, CA). In some cases, clones were isolated to generate cell lines with homogeneous expression of the NK activating ligand. These cell lines were then transduced with a lentiviral vector, pHAGE-puro-MCS (PPM)-U21, in which U21 was expressed under the control of a CMV promoter, and an IRES-driven puromycin N-acetyl transferase gene (Pac) allowed for puromycin selection [48]. K562 cells stably expressing U21 were generated by lentiviral transduction using a vector identical to PPM but containing the gene for Zs-Green instead of the Pac gene (PMG) [48]. The multiple cloning site (MCS) and puromycin cassette were modifications made to the pHAGE vectors in our laboratory, where we excised the gene for ZsGreen and replaced it with the puro cassette. The MCS was inserted in place of the gene for Ds-Red.
Monoclonal antibodies to ULBP1 (m295, IgG1), ULBP2 (m311, IgG1), ULBP3 (m551, IgG1) ULBP4 (m479, IgG1), MICA (m673, IgG1), and MICB (m360, IgG1) were generously provided by Amgen (Thousand Oaks, CA). Affinity purified goat polyclonal antibodies directed against ULBP1 (AF1380) and MICB (AF1599) were purchased from R&D systems (Minneapolis, MN). BMO2, a monoclonal antibody directed against MICB, was purchased from Axxora (San Diego, CA). The anti-lamp2 monoclonal antibody H4B4 was generously provided by Dr. T. August (Johns Hopkins Medical School, Baltimore, MD). The transferrin receptor (TfR) monoclonal antibody, (anti-CD71) (clone H68.4) was purchased from Zymed Laboratories (San Francisco, CA). Monoclonal anti-glyceraldehyde-3-phosphate dehydrogenase (GAPDH) was purchased from Imgenex (San Diego, CA). The intercellular adhesion molecule-1 (ICAM) monoclonal antibody (anti-CD54) was purchased from BD Biosciences. W6/32 is a monoclonal antibody that recognizes properly-folded class I MHC molecules [49]. HC10 is a monoclonal antibody that recognizes free class I MHC heavy chains [50]. Fluorescein isothiocyanate (FITC)-conjugated W6/32 was purchased from eBiosciences (San Diego, CA). HA.11 is a monoclonal antibody directed against hemagglutinin (HA), and was purchased from Covance (Princeton, NJ). A polyclonal antibody (MCW50) directed against the cytoplasmic tail of U21 was generated in our laboratory [31].
ULBP1–3, MICA, and MICB were amplified from plasmids provided by Dr. D. Cosman and cloned in to LNCX using the primers below. Constructs were verified by DNA sequencing. U21 was subcloned from PPM-U21 into PMG using XhoI and BamHI.
ULBP1:5'-ATACTCGAGGCCACCATGGCAGCGGCC GCCAG
3'-GTCAGGCTTTCATCTGCCAGCTAGAAT GAAG
ULBP2:5'-AGTCTCGAGGCCACCATGGCAGCAGCC GCCGC
3'-GTCAAGCTTTCAGATGCCAGGGAGGATG
ULBP3: 5'-AGTCTCGAGGCCACCATGGAGACAG
3'- GTCAAGCTTTCAGATGCCAGGGAGGATG
MICA: 5'-AGTCTCGAGGCCACCATGGGGCTGGGC CCGG
3'-GTCAAGCTTCTAGGCGCCCTCAGTGG
MICB: 5'-AGTCTCGAGGCCACCATGGGGCTGGGC CGGG
3'-GTCAAGCTTCTAGGTGCCCTCAGTGG
Packaging, envelope, and vector plasmids were cotransfected into HEK293T cells using TransIT-293 (Mirus Bio, Madison, WI). Viral supernatants were harvested at 48–72 hrs, filtered and either used to infect desired cell lines directly (retroviral transductions) or concentrated prior to infections (lentiviral transductions). Lentiviruses were concentrated by ultracentrifugation for 3 hrs at 4°C at 50,000xg. K562 cells were infected twice by spinoculation (1000 X g for 2 hr at 30°C) with concentrated lentiviruses. For G418- and puromycin-resistant constructs, cells were cultured in selection medium for at least 10 days.
Cells grown on glass coverslips were washed with PBS, fixed with 4% paraformaldehyde in PBS, permeabilized with 0.5% saponin in PBS and 3% BSA, incubated with primary antibodies, washed and incubated with Alexa488- or 594-conjugated secondary antibodies (Invitrogen). Colocalization studies (Figure 3) were performed using a Zenon Alexafluor 488 kit (Invitrogen) to label anti-lamp2 and the images were deconvoluted using Auto Quant 3D deconvolution software (Media Cybernetics, Bethesda, MD).
Adherent cells were detached with trypsin or 5 mM EDTA in PBS prior to labeling. Cells were washed in ice-cold PBS, and incubated with primary antibodies in 1% bovine serum albumin/phosphate buffered saline (BSA/PBS) for 30 min on ice. For nonconjugated primary antibodies the cells were then washed with 1% BSA/PBS and incubated with Goat F(ab)2 Anti-Mouse IgG (H+L) Phycoerythrin (PE)(R&D Systems, Minneapolis, MN). For intracellular staining, cells were fixed with 1% paraformaldehyde (PFA) and permeabilized with 0.1% saponin prior to staining. Flow cytometry was performed on either a FACSCalibur or FACSAria III (BD Biosciences), or Guava Easycyte mini (Millipore, Billerica, MA) and the data was analyzed using FlowJo software (Treestar, Ashland, OR).
Cells were detached with trypsin and incubated in methionine- and cysteine-free DMEM (Invitrogen) supplemented with 2%FBS for 30 min at 37°C (starve). The cells were labeled with 700 µCi/ml of [35S]-Express label (1100 Ci/mmol; PerkinElmer, Boston, MA) at 37°C and chased with complete DMEM supplemented with 1 mM non-radioactive methionine and cysteine for indicated times at 37°C. Cells were washed with PBS then lysed in Triton X-100 lysis buffer (10 mM Tris-HCl [pH 7.4], 150 mM NaCl, 1%Triton X-100, 0.1 mM phenylmethylsufonyl fluoride (PMSF), and 5 mM N-ethyl-maleimide (NEM)) or digitonin lysis buffer (1% digitonin, 150 mM NaCl, 50 mM Tris-HCl [pH 7.4], 5 mM NEM, 0.1 mM PMSF) for 5 min at 37°C to solubilize lipid rafts, followed by rocking for 15 min at 4°C. Lysates were centrifuged for 10 min at 16,000 X g at 4°C to pellet nuclei and debris. Clarified lysates were incubated overnight at 4°C with designated antibodies and Protein A agarose (Repligen Corporation, Waltham, MA) or protein G agarose (Invitrogen). Immunoprecipitates were washed four times with Triton X-100 wash buffer (10 mM Tris 7.4, 150 mM NaCl, 1%Triton X-100) or digitonin wash buffer (0.1% digitonin, 150 mM NaCl, 50 mM Tris pH 7.4), and subjected to SDS-PAGE gel electrophoresis. When indicated, lysosomal inhibitors leupeptin (Sigma) and folimycin (EMD, San Diego, CA) were added at 200 µM and 20 nM, respectively, during the starve, pulse, and chase.
Quantification was performed from phosphorimages generated on a Storm 820 (GE Healthcare, Piscataway, NJ) using ImageQuantTL software. In general, when measuring the half-life of a protein by pulse-chase analysis, the amount of protein recovered at each time point is calculated as a percent of the protein recovered immediately following the pulse. We used this method to quantify the stability of MICB (Figure 8 panels B and C). However, since immunoprecipitation of immature form of ULBP1 immediately following the pulse is inefficient, we felt it appropriate to normalize to the amount of mature ULBP1 recovered after the 2 hr chase (Figure 5 panel B). All values are also corrected for background. Calculations were performed as follows: For Figure 5, the % ULBP1 = (ULBP16hr - bkgd)/(ULBP12hr - bkgd)*100. For Figure 8, panel B %MICB = (MICB2hr - bkgd)/(MICB0hr - bkgd)*100. For Figure 8, panel C all points are normalized to the 0 hr chase point as in B). For Figure 9, panel C, the %MICB secreted at each time point = (MICBsupernatant - bkgd)/(MICBlysate 0hr - bkgd)*100.
Total cell lysates were prepared in 1% Triton X-100 lysis buffer supplemented with 50 U/ml benzonase (Sigma), followed by the addition of an equal volume of 2% SDS and 100 mM Tris-HCl [pH 7.4] and continued rocking at room temperature for 15 min. Lysates were normalized to total protein concentration as determined by BCA assay (Pierce, Rockford, IL). For immunoblot analysis of secreted MICB, supernatants were collected and concentrated 10 fold using a micron 30 filter (Millipore, Billerica, MA). For immunoprecipitation-immunoblot experiments, cells were treated with 200 µM leupeptin and 20 nM folimycin for 14 hours then lysed in digitonin lysis buffer. Immunoprecipitions were performed, washed 4 times with digitonin wash buffer, and eluted with Laemmli buffer. Lysates and immunoprecipitates were resolved by SDS-PAGE electrophoresis, transferred to BA-85 nitrocellulose membrane (Whatman, Florham Park, NJ) and probed with designated primary antibodies followed by an appropriate HRP conjugated secondary antibody (BioRad, Hercules, CA). Bands were visualized using SuperSignal reagent (Pierce) and quantified with an Alpha Imager (AlphaInnotech, San Leandro, CA).
Target cells (2.5×104) cultured in the presence of IL-2 (50 U/ml) were mixed with various ratios of NKL cells in V-bottom 96 well plates. When indicated, human PBMCs were isolated from blood on Ficoll-Paque (GE Healthcare) according to manufacturer's instruction and used in place of NKL cells. Plates were centrifuged for 5 min at 125 X g to pellet cells, and incubated for 3 hr at 37°C. Cells were then stained with 4 µg/ml 7-aminoactinomycin D (7-AAD)(Sigma) for 5 min and analyzed by flow cytometry. The percentage of target cell death was calculated as the %7-AAD-positive target cells (ZsGreen-positive) at each effector:target ratio minus the %7-AAD-positive target cells in the absence of NKL cells. When indicated, target cells were incubated with 10 µg/ml blocking antibodies for 15 minutes at 37°C prior to the addition of NKL cells. All assays were performed in triplicate and, unless noted, the experiments shown were the average of three independent experiments.
|
10.1371/journal.pntd.0005095 | The Epidemiological Characteristics and Dynamic Transmission of Dengue in China, 2013 | There was a dengue epidemic in several regions of China in 2013. No study has explored the dynamics of dengue transmission between different geographical locations with dengue outbreaks in China. The purpose of the study is to analyze the epidemiological characteristics and to explore the dynamic transmission of dengue in China, 2013.
Records of dengue cases of 2013 were obtained from the China Notifiable Disease Surveillance System. Full E-gene sequences of dengue virus detected from the outbreak regions of China were download from GenBank. Geographical Information System and heatmaps were used to describe the epidemiological characteristics. Maximum Likelihood phylogenetic and Bayesian phylogeographic analyses were conducted to explore the dengue dynamic transmission. Yunnan Province and Guangdong Province had the highest imported cases in the 2013 epidemic. In the locations with local dengue transmission, most of imported cases occurred from June to November 2013 while local dengue cases developed from July to December, 2013. There were significant variations for the incidences of dengue, in terms of age distributions, among different geographic locations. However, gender differences were identified in Guangzhou, Foshan and Xishuangbanna. DENV 1–3 were detected in all locations with the disease outbreaks. Some genotypes were detected in more than one locations and more than one genotypes have been detected in several locations. The dengue viruses introduced to outbreak areas were predominantly from Southeast Asia. In Guangdong Province, the phylogeographical results indicated that dengue viruses of DENV 1 were transmitted to neighboring cities Foshan and Zhongshan from Guangzhou city, and then transmitted to Jiangmen city. The virus in DENV 3 was introduced to Guangzhou city, Guangdong Province from Xishuangbanna prefecture, Yunnan Province.
Repeated dengue virus introductions from Southeast Asia and subsequent domestic dengue transmission within different regions may have contributed to the dengue epidemics in China, 2013.
| Dengue is the most prevalent and rapidly spreading mosquito-borne viral disease. As an imported disease in China, the imported cases play a vital role for the local dengue transmission. There were dengue outbreaks in three Provinces (covering nine Cities/Prefectures) of China in 2013, with several regions had their first dengue outbreak in history including the one from central China. There has been no study so far to explore the dengue transmission dynamics between different regions in China. The purpose of the study is to describe the 2013 dengue epidemiological characteristics and to explore the transmission dynamics of dengue viruses between epidemic focus. The study results indicated that repeated dengue virus introductions from Southeast Asia and subsequent domestic dengue transmission within different regions may have contributed to the dengue epidemics in China, 2013. Population movement could have played a critical role in dengue dynamic transmission, which introduced dengue viruses to non-epidemic areas at broad or finer spatial scales. Therefore, it should be considered in the design of mosquito eradication campaign for dengue control and prevention.
| Dengue is a mosquito-borne viral infectious disease caused by the four antigently distinct serotypes (DENV 1–4), which are mainly transmitted by Aedes aegypti and Aedes albopicuts. Dengue is endemic in more than 100 countries in tropical and subtropical areas, especially in Southeast Asia, the Americas, the Western Pacific, Africa and Eastern Mediterranean regions [1]. Because of unprecedented population growth, uncontrolled urbanization, spread of the mosquito vectors and the population movement, the incidence of dengue has increased dramatically in the past 50 years [2]. It is estimated that 390 (95% CI: 284–528) million people have dengue virus infections with 96 (95% CI: 67–136) million cases annually worldwide [3].
The dynamics of dengue transmission depends on the interactions among hosts, viruses, vectors and environmental factors. Given the restricted range of mosquito flying distance [4, 5], population movement may play a critical role for dengue transmission. At broad spatial scales (e.g., national, international), human movements may make dengue virus being introduced and reintroduced into a region with lower herd immunity [6]. Travel acquired cases were repeatedly imported to Europe from Africa, Southeast Asia and the Americas [7–9]. Local dengue transmission has occurred in Europe for the first time in many decades, with indigenous cases reported in France and Croatia in 2010 [10, 11]. In addition to sporadic cases, dengue outbreak also occurred in Europe. For example, an outbreak with more than 2,000 cases happened in Madeira, Portugal in 2012, which was most probable origin of Venezuela [12]. Aware of the importance of air travel in dengue transmission, researchers developed simple models to estimate the importing risk of dengue or even the possible origin of importation in Europe [13, 14]. In Asia, because of the reintroductions of dengue viruses from Southern Vietnam where dengue is endemic, Northern Vietnam had dengue epidemics occurred frequently [15]. At finer spatial scales (regional, intra-urban, neighborhood), population movements associated with work and recreation are important for dengue transmission [6], and house-to-house human movements may shape spatial patterns of dengue incidence, causing significant heterogeneity in dengue incidence [16].
There was no dengue case notified from 1949 to 1977 in China until an outbreak occurred in Guangdong Province in 1978. Since then, dengue has been detected for nearly forty years in China. It was prevalent in Southern China including Guangdong Province, Hainan Province and Guangxi Province in 1980s. Since 1990, dengue was predominantly occurred in Guangdong Province. Geographically, the dengue outbreaks have expanded gradually from Guangdong, Hainan and Guangxi in Southern coastal regions of China to the relatively Northern regions including Fujian, Zhejiang Provinces and to the relatively Western region Yunnan Province [17]. Neighboring to Myanmar, Laos and Vietnam, Yunnan Province had its first dengue outbreak with 56 cases reported in 2008, of which most were imported cases from Myanmar [17]. In 2013, Guangdong, Yunnan and Henan Provinces had dengue outbreaks. This was the first dengue outbreak in Henan Province, the most northern Province with dengue local transmission. The 2013 outbreak in Yunnan was the second outbreak and also the first severe dengue outbreak in the Province. Guangdong Province has the highest dengue incidence with cases reported every year since 1997, with the most prevalent in 2013.
Our previous study has proven that dengue was still an imported disease in China [18], and the epidemics were probable to be originated from overseas. Three individual studies have reported the dengue outbreaks occurred in Yunnan and Henan Provinces in 2013 [19–21]. However, no study tried to explore the transmission dynamics between locations. The purpose of the study is to describe the 2013 epidemiological characteristics and to explore the possible origins of the epidemics, and the dynamics of dengue viruses between epidemic focus, which are important to dengue control and prevention.
Ethical approval for the study was obtained from the Chinese Center for Control and Prevention Ethical Committee (No.201214) and patient data in the study were de-identified and analyzed in aggregated format.
Records of notified dengue cases of 2013 were obtained from the China Notifiable Disease Surveillance System, including age, gender, occupation, date of onset, type of diagnosis, local case or not. At the study areas, a dengue case is defined as an imported case for which the patient had traveling history to a dengue affected area and reported being bitten by mosquitoes within 15 days of the onset of illness. In some cases, importation is defined based on laboratory results showing that the infecting dengue virus had a high sequence similarity in the preM/E region compared with viruses isolated from the putative source region where the patient had traveled to. Otherwise, the dengue case is considered to be a local case [22].
Henan, Yunnan and Guangdong Provinces had dengue outbreaks in 2013. Full E-gene sequences of dengue virus detected in these Provinces in 2013 were downloaded from GenBank (As of August 25th 2015) (S1 Table). The sequences detected in China were compared with published sequences by using the nucleotide blast program in the NCBI. S2–S4 Tables were the references downloaded with the accession number, collection date and geographical region.
The population data were from the Sixth National Population Census of China conducted by the National Bureau of Statistics of the People’s Republic of China in 2010. The information of epidemiological investigation was downloaded from China Public Health Emergency Management Information System.
There were 4,779 dengue cases reported in the China Notifiable Disease Surveillance System in 2013, including 543 imported cases and 4,236 local cases. No dengue case was notified in Tibet, Qinghai, Ningxia and Shanxi Provinces. Imported dengue cases were reported in other Provinces with the highest numbers in Yunnan and Guangdong Provinces (Fig 1). In Henan Province, outbreak occurred in central China, Xuchang city with the incidence of 0.70 per 100,000. In Southwestern China, dengue outbreaks occurred in Dehong prefecture and Xishuangbanna prefecture locating in west and south of Yunnan Province with incidence of local cases 11.97 and 112.13 per 100,000, respectively. Dengue outbreaks occurred in Central and South of Guangdong Province (Southern China) including Guangzhou city, Foshan city, Dongguan city, Zhongshan city, Zhuhai city and Jiangmen city. The incidence ranged from 0.12 to 25.92 per 100,000, with Zhongshan city the highest and Guangzhou the second highest incidence (Fig 2).
In the regions with local dengue transmission, imported dengue cases occurred almost all year around, with the most cases happening from June to November—accounting for 93.89% of cases (215/229) (the number of imported cases in the regions with local dengue transmission was 229). Dehong and Xishuangbanna had the most imported cases, accounting for 68.12% (156/229) (S1 Fig). Local dengue cases occurred in July to December. The first local dengue case occurred in Zhongshan city, Guangdong Province (S2A Fig). Gradually, the cities around Zhongshan city all had dengue outbreaks. Meanwhile, dengue outbreaks also hit Xuchang city, Henan Province and Dehong prefecture and Xishuangbanna prefecture, Yunnan Province (S2B Fig).
The local incidence among different age groups showed significant differences in each outbreak area and the overall incidence in different age groups had a gradual increase with the age (Fig 3A). The local incidence between male and female was significantly different in Xishuangbanna, Guangzhou and Foshan, and the overall local incidence between male and female had significant difference, with more cases in females than in males (Fig 3B).
DENV 1–3 were detected in all regions with dengue outbreaks in 2013 (S3–S5 Figs). Among these, DENV 1-I, DENV 1-V, DENV 3-II and DENV 3-III were detected in more than one outbreak regions. More than one genotypes have been detected in several locations (Fig 4).
The dengue viruses detected in Guangdong Province in Clade 1 showed that Thailand strains were introduced to Guangzhou first, and then transmitted into its neighboring cities: Foshan and Zhongshan, and were then transmitted to Jiangmen from Zhongshan (Figs 5A and 6A). The phylogeographical results indicated that dengue viruses detected in Guangzhou, located in Clade 2, 3, 11, 12, were probably imported from Thailand, Singapore, Malaysia, Indonesia (Figs 5A, 7B, 6A and 6B), while Clade 13 showed that some Guangzhou strain was transmitted from Xishuangbanna, Yunnan Province (Figs 8A and 6C). Dengue viruses detected in Dehong, Yunnan Province (Clade 4, 9) were probably introduced from its neighbor, Myanmar (Figs 5A, 7A, 6A and 6B). Some strains detected in Jiangmen, the strain in Dongguan and some strains in Zhongshan, Guangdong Province, were probably imported from Singapore (Clade 5, 7, 16; Figs 5A, 5B, 8B, 6A and 6C). The outbreak occurred in Zhuhai, Guangdong Province was probably caused by introduced viruses from Thailand (Clade 6, Figs 5A and 6A). The results indicated that some strains detected in Foshan, Guangdong Province were introduced from India, Cambodia and Laos (Clade 8, 10, 14; Figs 5B, 7A, 8A and 6). Fig 8A indicated that the dengue viruses in Xishuangbanna were probably introduced from Laos (Clade 13; Fig 6C). In addition, the viruses detected in Henan Province were also probably imported from Laos (Clade 15; Figs 8A and 6C)
Dengue has been detected in China for nearly 40 years, and has become more serious in these years with increased incidence and expanded outbreak regions. Our previous study has proven that dengue is still an imported disease in China [18], and the results from this study suggested that the introduced dengue cases from overseas may have contributed to the dengue outbreaks. The 2013 outbreaks occurred in different locations of China were mainly due to the introduced viruses from Southeast Asia and domestic dengue transmission within different regions of China.
The demographic characteristics of dengue cases showed that dengue mainly affects adults in China, which were different from that in dengue endemic countries, where adult acquired lifelong immunity [28]. Adults were vulnerable in China given they have more opportunities to expose to mosquitoes when there is lower herd immunity. The sex distribution of dengue varied in different countries, and even in different locations or different epidemic time periods within a country [29–32]. The sex distribution of dengue may be attributed to the relevant demand on the health services of different sexes.
Located in Central China, Henan Province has a temperate climate. This was the first dengue outbreak in the Province. Geographically, it was the northernmost place with dengue outbreak in China, which is almost on the same latitude as Madeira, Portugal where dengue outbreak occurred in Europe. Generally, dengue occurs mainly in tropical and subtropical areas, where suitable climatic conditions play great roles in dengue transmission. Liu-Helmersson found that dengue epidemic potential in temperate was increased with increasing diurnal temperature range [33]. In addition, human movement plays an important role in dengue outbreak in temperate climate region. The dengue outbreak in Henan occurred in a small village with six porcelain factories. The epidemiological investigation showed that all dengue cases had neither travel history, nor collective activities. However, five people from the village worked in Laos where dengue was endemic. These five people returned to their home village in Henan for their vocational leave in May 2013. Although none of them were reported having dengue, two family members of these five people reporting dengue infections, indicating that they may have picked up the virus from the Laos and transmitted to their family members. In addition, the products of porcelain factories were usually sold to Guangzhou and Yunnan Province where dengue was prevalent in 2013 [21]. While phylogenetic and phylogeograhical analyses showed that the Henan strains and Yunnan strains (Xishuangbanna) were separated into different clades with high posterior probability support (Clade 13, 15; Fig 8A). Dengue was epidemic in Laos in 2012–2013 and the epidemic was severe in 2013, which might be attributed to the serotype switch from DENV 1 to DENV 3, and genotype II of DENV 3 was the predominant genotype [34, 35]. The strains detected in Henan Province were genotype II of DENV 3 (Clade 15) and the phylogeographical analysis indicated that the dengue viruses were probably introduced from the Laos with high ancestral location state probability.
Xishuangbanna prefecture, Yunnan Province, neighboring to Myanmar and Laos, had its first dengue outbreak in 2013. The phylogenetic results showed genotype II of DENV 3 caused the outbreak (Clade 13). As mentioned before, the same genotype was prevalent in Laos in 2012–2013. The epidemiological field investigation showed that there were imported cases from Laos and Myanmar at the beginning of the epidemic, and some businessmen has travel history to Laos. The phylogeographical analysis indicated that the viruses detected in Xishuangbanna were probably introduced from Laos. Therefore, because of the repeated introductions and lack of local herd immunity, the viruses imported from Laos probably contributed to the dengue outbreak in Xishuangbanna, Yunnan Province in 2013.
Dehong prefecture, Yunnan Province, bordering Myanmar also had a large dengue outbreak in 2013. The epidemiological analysis showed that there were 245 dengue cases in Dehong, of which 101 cases were imported cases from Myanmar. Most local dengue cases engaged in trade activities with Burmese around. The strains detected in Dehong were from local dengue cases and imported cases, and the imported cases were all from Myanmar [20]. The phylogenetic analysis indicated that the dengue viruses detected in Dehong were classified into genotype I in DENV 1 and Asian I genotype in DENV 2. The ML tree and MCC tree suggested the local cases and some imported cases from Myanmar were clustered together and the phylogeographical analysis indicated that the dengue outbreak occurred in Dehong were caused by two different genotypes, which were both introduced from Myanmar. Therefore, two genotypes co-circulated in Dehong invading from neighboring Myanmar and contributed to the dengue transmission.
Dengue epidemics occurred in Dehong and Xishuangbanna, Yunnan Province were attributed to the dengue viruses transmission across the border, and the main dengue transmission vector Aede aegypti existed in these two areas played a critical role in local dengue transmission [36].
In Guangdong Province, local dengue cases occurred in Pearl River Delta of Guangdong (PRD) in 2013, where is one of the most densely urbanized regions in the world [37]. The area is one of the homelands of overseas Chinese, especially to the those live in Southeast Asia, which brings about many travels from there each year. Dengue is endemic in Southeast Asia, with severe dengue outbreaks in Indonesia, Thailand, Singapore and Malaysia in 2013 [38–40]. These countries are tourism resorts and there have been intensive trade activities between these countries and the PRD. The dengue viruses may have imported to PRD frequently from these endemic areas because of the frequent population movement and probably contributed to the dengue outbreaks in these PRD areas. The phylogenetic and phylogeographical results indicated that the strains detected in the PRD areas, especially in Guangzhou showed diversity, and most strains were probably introduced from Singapore, Thailand, Malaysia, Indonesia, Cambodia and Laos (Fig 6).
Neighboring to Guangzhou, Dongguan is characterized by its fast urbanization and manufacturers in southern China. Although dengue infections have been reported in Guangzhou almost every year with four serotypes being isolated, Dongguan was free of dengue until its first dengue outbreak in 2010, which might be caused by viruses imported from Malaysia [41]. For the second dengue outbreak in Dongguan in 2013, although the Dongguan strain, Singapore and Malaysia strain were clustered together, the phylogeographical result indicated that Singapore strain caused the dengue outbreak with high location probability.
Because of the multiple introductions, dengue outbreaks occurred in epidemic seasons in Southern China in the context of suitable weather conditions [42]. Population movement at finer spatial scales contributed to the epidemic foci expansion, and may contribute dengue transmission from one epidemic city to another one within China [6]. The phylogenetic analyses indicated that the dengue epidemic transmitted to neighboring Foshan city southwestward and Zhongshan city southward from Guangzhou city, and then to Jiangmen city from Zhongshan city (Fig 6A). Given the increased economic link and population movement, there existed the probability dengue virus dispersion to Foshan from Guangzhou, because the outbreak in Guangzhou was earlier than that in Foshan. The phylogenetic analysis showed that the outbreak in Zhongshan was caused by two serotypes, with epidemic two peaks. The cases in Zhongshan were located on the border with Guangzhou and the dengue outbreak in Zhongshan was earlier than that in Guangzhou. Therefore, there existed the probability that the first peak occurring earlier than that in Guangzhou was caused by introduced viruses from Singapore (Fig 8B; Clade 16) and the second was probably attributed to the viruses transmitted from Guangzhou. The cases in Jiangmen city mainly occurred in the districts adjacent to Zhongshan city—this could be because that some of the residents working in Zhongshan while living in Jiangmen due to cheaper property price in Jiangmen and higher salary in Zhongshan, which made the dengue viruses transmitted to Jiangmen from Zhongshan feasible.
The phylogeographical analyses showed that Guangzhou not only transmitted viruses to adjacent areas, but also received from other area nationally. The result showed that there existed the probability that dengue virus transmitted to Guangzhou from Xishuangbanna, Yunnan Province. The ML tree and MCC tree both showed the Guangzhou strain was embraced in Xishuangbanna clade. Being the centre for industry, finance, transportation and trade in southern China, there are many migrant workers in Guangzhou, nationally and internationally. Xishuangbanna is a tourism resort, which attracted many people including these from Guangzhou. Given dengue outbreak occurred earlier in Xishaungbanna than that in Guangzhou, the dengue virus in Xishuangbanna probably contributed to dengue transmission in Guangzhou.
This is the first study exploring the dengue dynamics in China, which is critical to understand dengue transmission and will be helpful to prevent and control dengue occurrence in China. Because the E-gene data were downloaded from GenBank, we are unable to obtain the exact epidemiological information of each single sequence. Among the three Provinces with dengue outbreaks in 2013, we tried to assess whether the cases in Yunnan Province were local cases or not from published papers regarding its outbreak [19, 20]. For Henan and Guangdong Provinces, we contacted the Provincial CDCs, and confirmed that the sequences deposited to GenBank were all from local cases. However, we are not able to get the exact onset time for each case.
Dengue is still an imported disease in China. Because of population movement and the close connections between Southern China and Southeast Asia, Southeast Asia is still the main sources of dengue viruses introduced to China. At the background of climate change and the existence of dengue vector, dengue epidemic is not restricted in Southern China any more. The Pearl River Delta of Guangdong are merging more and more close in recent years,which makes dengue virus disperse easily. Therefore, population movement plays a critical role in dengue dynamic transmission, which introduces dengue viruses to non-epidemic areas at broad or finer spatial scales. Apparently relevant dengue control and prevention strategies should be updated.
|
10.1371/journal.pgen.1004004 | PP2A/B55 and Fcp1 Regulate Greatwall and Ensa Dephosphorylation during Mitotic Exit | Entry into mitosis is triggered by activation of Cdk1 and inactivation of its counteracting phosphatase PP2A/B55. Greatwall kinase inactivates PP2A/B55 via its substrates Ensa and ARPP19. Both Greatwall and Ensa/ARPP19 are regulated by phosphorylation, but the dynamic regulation of Greatwall activity and the phosphatases that control Greatwall kinase and its substrates are poorly understood. To address these questions we applied a combination of mathematical modelling and experiments using phospho-specific antibodies to monitor Greatwall, Ensa/ARPP19 and Cdk substrate phosphorylation during mitotic entry and exit. We demonstrate that PP2A/B55 is required for Gwl dephosphorylation at the essential Cdk site Thr194. Ensa/ARPP19 dephosphorylation is mediated by the RNA Polymerase II carboxy terminal domain phosphatase Fcp1. Surprisingly, inhibition or depletion of neither Fcp1 nor PP2A appears to block dephosphorylation of the bulk of mitotic Cdk1 substrates during mitotic exit. Taken together our results suggest a hierarchy of phosphatases coordinating Greatwall, Ensa/ARPP19 and Cdk substrate dephosphorylation during mitotic exit.
| Greatwall kinase regulates a switch between kinase and phosphatase activity during mitotic entry and constitutes an essential element of mitotic control. This control system is further complicated by the fact that Greatwall itself is regulated via phosphorylation and acts by phosphorylating its substrates ENSA and ARPP19. A missing link in this central mitotic switch is the phosphatase that counteracts Greatwall and its target Ensa. We have used mathematical modeling and experimental validation to identify these phosphatases. We demonstrate genetic evidence that Greatwall phosphorylation is counteracted by PP2A/B55, while Fcp1 regulates ENSA dephosphorylation. Based on these findings we present a new model for the regulation of mammalian cell division.
| Phosphorylation of more than thousand proteins by Cdk1 and other mitotic kinases drives entry into mitosis [1], [2]. As cells exit mitosis, these post-translational modifications have to be removed by phosphatases. Mitotic kinase and phosphatase activity appears to be inversely regulated to avoid futile cycles of phosphorylation and dephosphorylation [3], [4]. Moreover, mitotic kinases themselves are regulated by phosphorylation and dephosphorylation resulting in a complex feedback system of cell cycle control [5]. Cdk1 is negatively regulated by phosphorylation at Thr14/Tyr15 by Wee1 and Myt1 kinases and dephosphorylation of this site by the Cdc25 phosphatase constitutes the decision point to enter mitosis [6]. Cdk1 actively participates in its own activation by negatively regulating its inhibitor Wee1 [7]–[9] and positively regulating its activator Cdc25 [10]. In Xenopus egg extracts this switch is counteracted by the phosphatase PP2A/B55δ [11], [12] suggesting that inhibition of PP2A/B55δ is an intrinsic element of the G2/M transition. This is achieved by Greatwall kinase (Gwl) [13]–[15] that phosphorylates and activates the PP2A/B55δ inhibitors Endosulfin (Ensa) and ARPP19 [16]–[19]. The Gwl phosphorylation motive FDSGDY is identical in Ensa and ARPP19 and thus detectable with the same phospho-specific antibody. For simplicity we will refer in our analysis to Ensa/ARPP19, because it is impossible to distinguish between the phosphorylation of the two proteins with specific antibodies. Depletion of Gwl kinase in Xenopus mitotic extracts results in rapid Cdk1 inactivation and exit from mitosis, while Gwl depletion in interphase extracts blocks Cdk1 Thr14/Tyr15 dephosphorylation and mitotic entry [14], [15]. In human cells Gwl kinase depletion causes a delay in mitotic entry, reduces Cdk substrate phosphorylation and results in chromosome alignment defects and aberrant mitotic exit [20], [21]. Cdk1 phosphorylates Gwl at multiple sites and is required for its activation [22]. Thus, Cdk1 and Gwl activation are locked in a complex feedback loop at the G2/M transition.
To gain a more precise understanding of this switch-like transition, the phosphatases that target Gwl itself and Ensa/ARPP19 have to be identified. Inactivation of these phosphatases is likely to play a major role in the initiation of the Cdk1 activation loop. Moreover, the reactivation of these phosphatases is likely to be a crucial element during mitotic exit. The identity of the major Cdk1 counteracting phosphatase in mammalian cells is also still under debate. Current models propose that PP2A/B55δ is not only required for the Cdk1 activation loop, but also to directly dephosphorylate mitotic substrates during mitotic exit [3], [23]. Thus, PP2A/B55δ has been proposed to be the major Cdk1 counteracting phosphatase during mitotic exit in Xenopus egg extracts, equivalent to the function of Cdc14 phosphatase in budding yeast. This hypothesis is based on the observation that co-depletion of Wee1, Myt1 and Gwl causes mitotic exit from Xenopus egg extracts despite persistent high Cdk1 activity [24]. Conversely, B55δ depletion does not block Cdk1 substrate dephosphorylation, when mitotic exit is triggered by Cdk1 inhibition in Xenopus egg extracts [12] and PP1 has also been implicated to act as a major mitotic exit phosphatase [25]. In human cells PP2A/B55α has been implicated in regulating mitotic exit, but B55α depletion shows only a delay but not a block of Cdk substrate dephosphorylation following Cdk inactivation [26]. Thus, the network of phosphatases that counteract Gwl, Ensa/ARPP19 and Cdk phosphorylation during mitotic exit remains to be determined.
To gain insight into the dynamics of the Gwl/Cdk1 feedback loop during mitotic entry and exit we performed mathematical modelling (see Material and Methods) of the Cdk1 regulatory network shown on Fig. 1A. This network has multiple positive circuits (positive and double-negative feedback loops) including the antagonism between Cdk1 and PP2A/B55 that plays a key role in the switch-like transitions at G2/M and mitotic exit. PP2A/B55 inhibits Cdk1 through the Tyr-modifying enzymes while Cdk1 down-regulates PP2A/B55 activity through Gwl and Ensa/ARPP19 activation. As a consequence of these feedback loops the network has two qualitatively different states corresponding to G2 and M phases. In G2, both Gwl and Cdk1 are inactive while PP2A/B55 is active, but in M phase the opposite is true. Based on this information, we built mathematical models to analyse the impact of different Gwl phosphatases on the dynamics of mitotic entry and exit (Fig. S1). Inactivation of Cdk1 (by chemical inhibition) stabilizes the G2 phase (high Cdk1 Tyr15 phosphorylation, inactive Gwl and active PP2A/B55), allowing Cyclin B to accumulate above the threshold normally required for G2/M transition. Mitotic entry can be triggered by either terminating Cdk1 kinase inhibition, or by inactivation of PP2A by Okadaic acid (OA). In this scenario Cdk1 re-activation destabilizes the G2 state and initiates rapid transition into M phase characterized by Tyr15 dephosphorylation, Gwl activation and PP2A/B55 inhibition (Fig. 1B–D). Conversely, PP2A inhibition causes only a slow decrease in Cdk1 Tyr15 phosphorylation because phosphorylation of Tyr-modifying enzymes is compromised by low Cdk1 activity (Fig. 1E–G). If the Gwl phosphatase is insensitive to OA, PP2A inhibition should not induce phosphorylation of these proteins, because Cdk1 activity remains low and the counteracting phosphatases are active (Fig. 1E). In contrast, if Gwl is targeted by an OA sensitive phosphatase, Cdk1 Tyr-dephosphorylation should be accompanied by Gwl phosphorylation (Fig. 1F and G). For simplicity we presume constitutive activity of the Ensa/ARPP19 phosphatase. Activation of Gwl overcomes this activity and results in Ser67 phosphorylation. Furthermore, the model predicts that Gwl phosphorylation precedes Tyr15-dephosphorylation (Fig. 1F and G) because the latter requires prior phosphorylation of Cdc25 and Wee1 to remove the inhibitory Tyr15 phosphorylation. To obtain further information about Gwl phosphatases we also simulated mitotic exit triggered by Cdk1 inactivation (Fig. 1). The model suggests that Gwl will be instantaneously dephosphorylated after Cdk1 inactivation, if a constitutive phosphatase is responsible for its inactivation (Fig. 1H and I). If the Gwl phosphatase is directly participating in a double negative feedback loop (such as the M-phase inactive phosphatase PP2A/B55) then Gwl inactivation is a slow process to allow for time of phosphatase reactivation (Fig. 1J). Thus, our model of the kinetics of mitotic entry and exit potentially revealed important information about the nature of the Gwl phosphatase, which can be tested experimentally.
To test these predictions it is necessary to monitor Gwl phosphorylation during mitotic entry and exit using a phospho-specific antibody. In order to study potential Gwl activation by phosphorylation, we scanned for Cdk consensus sites following the universal DFG motif in Gwl that marks the beginning of the T-loop/activation segment. Fig. 2A shows that there is a conserved Threonine (Thr194 in human Gwl) followed by a Proline about 20 residues downstream of the DFG motif. When mutated to Alanine, only Thr194, but not Thr193 results in a significant loss of Gwl activity (Fig. 2B and C) whereas a Thr194Ser mutation does not affect Gwl kinase activity (Fig. S2A). The equivalent of human Gwl Thr194 in Xenopus is also required for kinase activity and mutants lacking this phosphorylation site are unable to reconstitute Gwl depleted egg extracts, suggesting that phosphorylation of this residue is essential for survival [22]. We therefore raised a phospho-specific pThr194 antibody to monitor Gwl phosphorylation at this residue. Fig. 2D shows that the antibody detects Flag-Gwl immuno-precipitated from nocodazole-arrested cells. We detected only weak signal in Gwl purified from asynchronous, and no signal in Gwl from Cdk1 inhibited cells, and also in a mitotic Thr194Ala mutant of Gwl. The antibody also detected Gwl specifically in mitotically enriched cells following a double Thymidine block release synchronisation (Fig. 2E and Fig. S2B). To determine, if Thr194 was directly phosphorylated by Cdk, we incubated WT and Thr194Ala Flag-Gwl that was immuno-precipitated from asynchronous human cells with recombinant cycA/Cdk2 in the presence of ATP (Fig. 2F). Addition of recombinant CycA/Cdk2 resulted in strong Thr194 phosphorylation. Applying alkaline phosphatase to the reaction significantly reduced the signal and the Thr194Ala mutant did not cross-react with the phospho-specific antibody after incubation with cycA/Cdk2. These data demonstrate that Thr194 is a Cdk site that is phosphorylated during mitosis in human cells and that the Gwl pThr194 antibody specifically cross-reacts with this phosphorylated residue.
We also determined the precise localization and timing of Gwl Thr194 phosphorylation by immunofluorescence (Fig. S2C–D). The pThr194 antibody strongly stained mitotically arrested cells and this signal was absent in most mitotic cells after Gwl siRNA depletion (Fig. S2C). There was little background signal detectable in interphase cells, but Gwl Thr194 phosphorylation occurred in the nucleus of G2 cells with separated centrosomes, increased in metaphase cells on the mitotic spindle and decreased to background levels in telophase cells (Fig. S2D). In prophase cells Gwl Thr194 phosphorylation occurred both in the nucleus and at the centrosomes (Fig. 2G). Mitotic Gwl was specifically enriched at the spindle poles and also showed distinctive foci in the metaphase plate that overlap with CenpA, suggestive of centromeric localization (Fig. 2H). This centromeric and polar phosphorylation was absent in early anaphase cells, while cytoplasmic Gwl appeared to remain phosphorylated, indicating a localised phosphatase activation in the early anaphase spindle.
The availability of antibodies to monitor Cdk, Gwl and Ensa/ARPP19 phosphorylation allowed us to experimentally test the mathematical models described in Fig. 1. This required a system to specifically inactivate and rapidly re-activate Cdk1. For this purpose we used the previously published cdk1as DT40 cell-line [27]. These chicken lymphocytes have the advantage of a rapid and complete re-activation of Cdk1 following removal of the ATP analogue inhibitor 1NMPP1. Thr194 was not phosphorylated in 1NMPP1 inhibited cdk1as cells, suggesting that in vivo Cdk1 is required for this phosphorylation. Release from Cdk1 inhibition by 1NMPP1 resulted in rapid simultaneous Cdk1 Tyr15 de-phosphorylation and Gwl Thr194 phosphorylation within 5 minutes (Fig. 3A and B). As predicted by the models presented in Fig. 1E–G, treatment of the 1NMPP1 inhibited cells with OA triggered a much slower G2/M switch and caused Cdk1 Tyr15 dephosphorylation after 60 minutes of phosphatase inhibition and a steady increase in Cdk substrate phosphorylation (Fig. 3C and D). Gwl Thr194 and Ensa/ARPP19 phosphorylation occurred rapidly within 30 minutes of OA addition (Fig. 3C and D). These data correlate with the models in Fig. 1F and G demonstrating that Gwl phosphatase is OA sensitive. The experiments also verify the prediction that Gwl phosphorylation precedes Cdk1 dephosphorylation.
Our model was built on the assumption that Ensa/ARPP19 is targeted by a constitutive phosphatase. Therefore, PP2A inhibition induced Gwl activation should result in a steady increase of phosphorylation of Ensa/ARPP19. However, Ensa/ARPP19 phosphorylation initially increased at 30 and 60 minutes after OA, but was lost at the 120 minutes timepoint, when Cdk1 was fully dephosphorylated (Fig. 3D). This prompted us to hypothesize that perhaps another OA insensitive phosphatase is activated at this late time point that can target Ensa/ARPP19, but not Gwl Thr194 phosphorylation. To test, if this activity was sensitive to other phosphatase inhibitors we repeated the experiment in the presence of the PP1 inhibitor tautomycin (TC). Fig. 3D (OA+TC) shows that this alternative inhibitor is indeed sufficient to suppress Ensa/ARPP19 dephosphorylation at the 120 minute timepoint. These data suggest that Gwl Thr194 phosphorylation is opposed by an OA sensitive phosphatase and that this phosphorylation occurs before Tyr15 dephosphorylation. Our data also suggest that Ensa/ARPP19 phosphorylation is targeted by a different phosphatase that is sensitive to TC, or a combination of OA and TC.
To gain further insight into the dephosphorylation dynamics of Gwl-Ensa/ARPP19 and Cdk substrates, we experimentally tested the mitotic exit models in Fig. 1H–J. For this purpose we arrested cells in mitosis and then triggered mitotic exit by Cdk1 inhibition. The effects of Cdk inhibitors on metaphase arrested cells are somewhat contentious [28]–[30]. To try and address this problem, we performed the mitotic exit experiment with three different Cdk inhibitors in the presence of proteasome inhibition and OA (Fig. S3A and S3B). Cdk substrate dephosphorylation was triggered by Roscovitine, Flavopiridol and RO3306 and progressed with comparable dynamics in the presence of proteasome and PP2A inhibition. However, OA appeared to cause an increase in Cdk phosphorylation activity in the metaphase arrested cells suggesting a role of OA sensitive phosphatases such as PP2A on Cdk substrate phosphorylation at this cell cycle stage. The effect of the Cdk inhibitors was not a result of off-target effects on Gwl, because none of the compounds affected Gwl in vitro activity (Fig. S3C).
For the purpose of testing the model, we chose the Cdk1 inhibitor RO3306. We blocked cells in mitosis using the Eg5 inhibitor STLC, and triggered mitotic exit by Cdk1 inhibition with RO3306. Our model predicts that Cdk1 inactivation is sufficient to trigger mitotic exit and that Gwl and Ensa/ARPP19 dephosphorylation would progress slowly, if the counteracting phosphatase is locked in a double negative feedback loop with Gwl, which is indeed what was observed. Gwl, Ensa/ARPP19 and Cdk substrate dephosphorylation all occur with a 15–30 minutes delay (Fig. 4A left most panel). We then tested the sensitivity of these dephosphorylation events to OA and TC. In accordance with our results presented in Fig. 3C & D, this experiment suggests that Gwl Thr194 dephosphorylation is OA sensitive, while Ensa/ARPP19 dephosphorylation is only inhibited in the presence of both OA and TC that inhibit both PP2A and PP1 (Fig. 4A). Accordingly, a higher dose OA concentration of 5 µM can also inhibit Ensa dephoshorylation (Fig. S3D). Both OA and TC did cause an increase in Cdk phosphorylation in the mitotic cells. However, Cdk substrate dephosphorylation did appear to proceed with unperturbed dynamics in the OA+TC treated cells following Cdk1 inhibition and was only blocked in the presence of a more global phosphatase inhibitor Calyculin A (CalA) (Fig. 4E).
We also observed that Gwl still shifts towards lower electrophoretic mobility, suggesting additional dephosphorylation of other residues, despite the presence of OA and TC and the block in Thr194 dephosphorylation (Fig. 4A). This suggests that other phosphatases may also contribute to Gwl dephosphorylation at residues different from Thr194. This notion prompted us to analyse the effect of phosphatase inhibitors on Gwl activity. For this purpose we expressed Flag-Gwl in HeLa cells synchronised in mitosis by STLC, and measured the in vitro kinase activity of immuno-precipitated Gwl before and after Cdk1 inhibition in the presence of different phosphatase inhibitors. Cdk1 inactivation caused a reduction to 10% of mitotic Gwl activity as well as Thr194 dephosphorylation and disappearance of the Gwl bandshifts (Fig. 4C and D). After OA addition, Gwl Thr194 dephosphorylation remained unchanged, and the protein remained active to a level of 60% relative to the mitotic control. Surprisingly, addition of TC caused a small increase in Thr194 dephosphorylation, but did not significantly affect Gwl activity. In both cases (OA and OA+TC) the high bandshift Gwl bands, were only gradually shifted down compared to the untreated control. Calyculin A led to a much more significant change in electrophoretic mobility of Gwl and blocked the inactivation of Gwl following Cdk1 inhibition to give about 80% activity, relative to the mitotic control. This suggests that Thr194 phosphorylation, as well as the bulk activity of Gwl, is counteracted by an OA sensitive phosphatase. However, other phosphorylation events in the protein are controlled by OA resistant phosphatases that have an additional but smaller impact on kinase activity.
PP2A/B55 itself is a good candidate for the Gwl Thr194 phosphatase, but the experiments with OA do not allow discriminating between different PP2A complexes and other OA sensitive phosphatases. This is suggested by the slow dephosphorylation following Cdk1 inactivation (Fig. 4A) fitting to a model of double negative feedback between Gwl and the Gwl phosphatase (Fig. 1J). To test this hypothesis we performed siRNA depletion of B55α and δ subunits and monitored Gwl dephosphorylation following Cdk1 inactivation of STLC synchronized mitotic cells (Fig. 5A). Depletion of B55α and δ had a synergistic effect and significantly blocked Gwl Thr194 dephosphorylation. This demonstrates that both PP2A/B55α and δ contribute to this event. Similar to the OA experiments B55α and δ depletion did cause an increase in metaphase Cdk substrate phosphorylation but did not affect the relative amount of Cdk1 substrate dephosphorylation following Cdk1 inhibition.
The results presented in Fig. 3 and 4 demonstrate that the identity of the Ensa/ARPP19 phosphatase is different from the ones targeting Gwl and other Cdk substrates. However, these data do not allow us to conclude on the nature of this phosphatase other than its sensitivity to a combination of OA and TC. The RNA polymerase II C-terminal tail domain phosphatase Fcp1 has recently been shown to play an important role during mitotic exit [31], [32]. We hypothesised that Fcp1 may be required for Ensa/ARPP19 dephosphorylation and tested this idea using Fcp1 siRNA. Fig. 5B shows that depletion of Fcp1 had no effect on Gwl Thr194 and Cdk substrate dephosphorylation following Cdk1 inactivation, but blocked dephosphorylation of Ser67 in Ensa/ARPP19.
To confirm that PP2A/B55 and Fcp1 can directly dephosphorylate Gwl and Ensa, we set up in vitro phosphatase assays using recombinant substrates [12] (see Materials and Methods). We purified transiently expressed Flag-B55α and GFP-Fcp1 from nocodazole treated mitotic Hek293T cells before or after Cdk1 inhibition by Flag and GFP immuno-precipitation (Fig. 5C and D). In the case of Gwl we observed a significant increase in phosphatase activity with B55α purified from mitotic cells before, or after Cdk1 inhibition (Fig. 5C). This could mean that PP2A/B55 is already active in a mitotic arrest, or that we activated it during the purification process. GFP-Fcp1, on the other hand, appeared to dephosphorylate Ensa only when purified from mitotic extracts after Cdk1 inhibition (Fig. 5D), suggesting a more stable (potentially post-translational) regulation of this phosphatase during mitotic exit. We also tested the sensitivity profile of Fcp1 against phosphatase inhibitors to determine whether OA and TC synergistically inactivate it, as suggested by the results in Fig. 4A. In vitro, both 1 µM OA and 10 µM TC were sufficient to inhibit this Ensa phosphatase (Figure S3E). It is difficult to relate the in vitro and in vivo concentrations of the phosphatase inhibitors, because of transport into and out of the cell and competitive binding between different phosphatases. Most likely, the combination of OA and TC has a synergistic effect in inhibiting Fcp1, but we cannot rule out indirect effects from inhibition of other phosphatases by combining OA and TC.
The results presented here suggest a complex phosphatase network counteracting the Cdk1 activation loop. These data allow us to update the model of the mammalian mitotic switch by including the new information on PP2A/B55 and Fcp1 (Fig. 6). The inclusion of an additional layer of phosphatase regulation does not alter the fundamental characteristic of the mitotic switch model, but it influences the temporal dynamics of mitotic entry and exit. Our mathematical modeling together with experiments indicates that mutual antagonism between Gwl and PP2A/B55 accelerates both mitotic entry and exit with synthesis and degradation of CycB, respectively. Our data also indicate that while PP2A is principally responsible for the inactivation step of Gwl, phosphatases other than PP2A regulate additional phosphorylation sites and also contribute to Gwl inactivation (see Fig. 4C and D). We have also identified Fcp1 as the phosphatase counteracting the Gwl phosphorylation site in Ensa/ARPP19. This implicates that Fcp1 is yet another intrinsic element of the mitotic switch and helps explain the observation that this phosphatase is required for mitotic exit [32]. Fcp1 appears to be subjected to further regulation by Cdk1 and activated during mitotic exit. These novel elements of the mitotic switch remain to be investigated.
Strikingly neither PP2A, PP1 nor Fcp1 appear to be sufficient to dephosphorylate the bulk of mitotic Cdk1 substrates, if the Gwl feedback loop is bypassed by Cdk1 inhibition. Dephosphorylation of the majority of Cdk substrates proceeds despite the presence of inhibitors (Fig. 4A), depletion of B55 subunits (Fig. 5A) and Fcp1 (Fig. 5B) and persistent activation of Gwl and Ensa (Fig. 4A and 5B), which additionally inhibit PP2A/B55. These results suggest that PP2A/B55 is not the only mitotic exit phosphatase that removes the bulk of mitotic phosphorylation. This notion is also supported by experiments in Xenopus egg extracts, in which Cdk1 inhibition by p27 can trigger mitotic dephosphorylation even in the absence of B55δ [12]. This is further supported by the observation that B55 depleted cells do not show a prolonged arrest in mitosis, as would be expected in the absence of the major Cdk1 antagonizing phosphatase [26]. Thus, the nature of the phosphatases that directly counteracts Cdk1 during the mitotic exit in mammalian cells remains elusive. In yeast Cdc14 covers this role, but single deletions of the mammalian equivalents Cdc14 A and B do not have the expected mitotic phenotypes [33], [34]. However, a double deletion of both orthologues has, to our knowledge, not been reported. Further experiments will be necessary to identify the mammalian version of this crucial mitotic exit phosphatase, and to analyse how it is integrated in the mitotic entry and exit switches.
Entry into mitosis is triggered by the activation of Cdk1/CycB. The network that controls Cdk1/CycB activity involves regulation of inhibitory kinase, Wee1 and activatory phosphatase, Cdc25. In interphase, Wee1 dependent phosphorylation keeps Cdk1 inactive while Cdc25 dependent dephosphorylation activates Cdk1 and promotes entry into mitosis. Both Wee1 and Cdc25 are subjected to Cdk1/CycB dependent phosphorylations which lead to the inactivation of Wee1 and activation of Cdc25. As a consequence, Cdk1 activity is controlled by feedback loops involving Wee1 (Cdk1—|Wee1—|Cdk1) and Cdc25 (Cdk1→Cdc25→Cdk1). Cdk1 subunit is in excess compared to its activating partner Cyclin B. Therefore, it is assumed that all Cyclin B is in the complex such that total CyclinB (CycBT) is sum of active Cdk1/CycB and inactive Cdk1/CycB dimers.
The network also includes the regulation of Cdk1/CycB counteracting phosphatase, PP2A/B55. We consider that the dephosphorylation of Wee1 and Cdc25 is controlled by PP2A/B55. Cdk1 indirectly inhibits PP2A/B55 via Gwl-Ensa/ARPP19 pathway. Thus, Cdk1 promote Wee1 and Cdc25 phosphorylations by directly phosphorylating them and by indirectly inhibiting their phosphatase forming coherent feed-forward loops. Gwl-Ensa/ARPP19 pathway involves Cdk1/CycB dependent phosphorylation of Gwl (Gwl) which in turn phosphorylate the phosphatase inhibitor Ensa/ARPP19. The phosphorylated Ensa/ARPP19 binds directly to PP2A/B55 to form an inhibitory complex. The phosphatases that dephosphorylate Gwl and Ensa/ARPP19 are unknown. Therefore, we consider three different scenarios. Gwl is dephosphorylated by (A) OA insensitive phosphatase (B) OA sensitive phosphatase (PP2A) and (c) PP2A/B55. We also assume that ENSAPt (sum of phosphorylated forms of Ensa/ARPP19) is dephosphoryated by OA insensitive phosphatase. We predict using the model the effect of different phosphatases on both the steady state of the system and dynamics of chemical inhibitor induced mitotic entry and exit.
The temporal dynamics of each component in the network is described by non-linear ordinary differential equation (ODE). All the individual biochemical reactions are approximated by mass action kinetics. The set of ODEs are integrated numerically using stiff solver in XPPAUT, a freely available program from G. Bard Ermentrout, University of Pittsburgh, PA, USA; http://www.math.pitt.edu/~bard/xpp/xpponw95.html).
One parameter bifurcation diagrams (Suppl. Fig. S1) were computed using the program AUTO. Bifurcation diagram is used to illustrate how steady state of the dynamical system changes as a function of control parameter. We compute the diagram with respect to CycBT as a parameter since its level is unaffected by the dynamics of the system. We present the XPPAUT code of model that can be used to re-produce all the simulations (Fig. 1 and Fig. S1) in the manuscript. The comparative analysis of the three models is carried out by using the same set of parameter values except for the changes corresponding to Gwl phosphatase (see below). The total concentrations of the proteins are assumed to be equal to one except for PP2A (PP2t = 0.5) whose concentration is kept below its stoichiometric inhibitor, ENSA concentration. The choice of kinetic parameter values is based on the criteria that G2/M transition should exhibit bistable characteristic [4][6]. Two parameter bifurcation diagrams were used to study the dependence of bistability on kinetic parameter values. We observed bistable behaviour over a wide range of kinetic parameters values (kagwl>1; kigwl′ = 0–10; kaensa>1; kiensa = 0–2.5) and within this range the dynamical characteristics of the three models as shown in Figure 1B–J are not affected. These characteristics include rapid Tyr15 dephosphorylation (in all the three models) during mitotic entry with Cdk1 inhibition and release in comparison with the Cdk1 inhibition + OA and delay in Gwl inactivation during mitotic exit only when Gwl is dephosphorylated by PP2A/B55. The inclusion of PP2A/B55 dependent regulation of Gwl also makes the system bistable over wider range of total Cyclin B levels, because of extra double–negative feedback loop between Gwl and PP2A/B55.
The parameter values in XPPAUT code correspond to the situation of Gwl dephosphorylation by PP2A/B55 and the other two models can be simulated with following changes: Gwl dephosphorylation by OA insensitive phosphatase: kigwl′ = 2; kigwl = 0.
Gwl dephosphorylation by OA sensitive phosphatase (PP2A): kigwl″ = 2; kigwl = 0.
The chemical inhibition of Cdk1 and PP2A are modelled as a reversible inhibitor binding reaction. The mitotic entry simulation in the presence of inhibitor is done with parameter value RO = 25 (for Cdk1 inhibition), OA = 100 (for PP2A inhibition).The mitotic exit simulation in the presence of inhibitor is done with parameter value RO = 100 (for Cdk1 inhibition).
# Initial conditions
init MPF = 0, Cdc25 = 0, Wee1 = 1, Gwl = 0, ENSAPt = 0, PP2 = 0.5
# Use the following initial conditions for mitotic exit
# MPF = 0.96, Cdc25 = 0.97, Wee1 = 0.03, Gwl = 0.9, ENSAPt = 0.75, PP2 = 0.027
#Differential equations
MPF′ = k25*(CycT-MPF) - kwee*MPF
Cdc25′ = Va25*MPFa*(Cdc25T-Cdc25) - Vi25*PP2a*Cdc25
Wee1′ = Vawee*PP2a*(Wee1T-Wee1) - Viwee*MPFa*Wee1
Gwl′ = kagwl*MPFa*(GwlT-Gwl) - (kigwl′+ kigwl″*PP2T/(1+OA)+ kigwl*PP2a)*Gwl
ENSAPt′ = kaensa*Gwl*(ENSAT - ENSAPt) - kiensa*ENSAPt
PP2′ = -kas*(ENSAPt - (PP2T-PP2))*PP2 + (kdis + kiensa)*(PP2T-PP2)
MPFa = MPF/(1+RO)
PP2a = PP2/(1+OA)
aux preMPF = CycT-MPF
aux PP2a = PP2/(1+OA)
aux MPFa = MPF/(1+RO)
k25 = k25′*(Cdc25T-Cdc25) + k25″*Cdc25
kwee = kwee′*(Wee1T -Wee1) + kwee″*Wee1
#parameters
par CycT = 1, Va25 = 2, Vi25 = 2, Vawee = 2, Viwee = 2
par kagwl = 10, kigwl′ = 0.02, kigwl″ = 0, kigwl = 2, kaensa = 2, kiensa = 0.6
par kas = 100, kdis = 1
par Cdc25T = 1, Wee1T = 1, GwlT = 1, ENSAT = 1, PP2T = 0.5,
par k25′ = 0.01, k25″ = 1, kwee′ = 0.01, kwee″ = 1
par RO = 0, OA = 0
@ total = 30,dt = 0.5,meth = STIFF
@ xlo = 0,xhi = 30,ylo = 0,yhi = 1
@ NPLOT = 6, yp1 = preMPF, yp2 = Cdc25, yp3 = ENSAPt, yp4 = PP2a, yp5 = Gwl
@ NTST = 15,NMAX = 20000,NPR = 1000,DS = 0.02
@ DSMAX = 0.05,DSMIN = 0.001,PARMIN = -1,PARMAX = 1
@ AUTOXMIN = 0,AUTOXMAX = 1,AUTOYMIN = 0,AUTOYMAX = 1
done
Gwl cDNA was cloned by RT-PCR from mRNA isolated from HeLa cells using RNeasy mini kit (Quiagen). Reverse-transcriptase PCRs (RT PCRs) were performed using the SuperScript III One-Step RT-PCR System from Invitrogen. Gwl cDNA was amplified from HeLa cell messenger RNA (mRNA) (using primers 5′GGG GAC AAG TTT GTA CAA AAA AGC AGG CTT AAT GGA TCC CAC CGC GGG AAG c3′; 5′GGG GAC CAC TTT GTA CAA GAA AGC TGG GTC CTA CAG ACT AAA TCC AGA TAC GG3′) and cloned into the pCR II-TOPO cloning vector using the TOPO-TA Cloning Kit from Invitrogen. In all cases the manufacturer's recommended protocols were followed and the presence of the expected DNA sequences confirmed by restriction digestion and sequencing. The cDNA was then cloned into an N-terminal Flag tag mammalian expression destination vector (a gift from Dr. Stephan Geley, University of Innsbruck, Austria). Site directed mutagenesis was carried out using the QuickChange XL Site-Directed Mutagenesis Kit from Stratagene (California, USA) following the manufacturer's protocol. Human B55α was cloned by RT-PCR as above using the primers 5′GGG GAC AAG TTT GTA CAA AAA AGC AGG CTT A ATG GCA GGA GCT GGA GGA GG3′ and 5′GGG GAC CAC TTT GTA CAA GAA AGC TGG GTC CTA ATT CAC TTT GTC TTG AA3′ cloned into a Gateway Flag-tag expression vector. Fcp1 cDNA was synthesized by Genscript and cloned into the EGFP-C1 vector via Bgl2 and EcoR1 sites. Human Ensa was cloned by RT-PCR using the primers 5′ GGG G ACA AGT TTG TAC AAA AAA GCA GGC TTA ATG TCC CAG AAA CAA GAA GA 3′ and 5′ GGG GAC CAC TTT GTA CAA GAA AGC TGG GTC TCA TTC AAC TTG GCC ACC CG3′ and further cloned into a his-tag Gateway bacterial expression vector (a gift from Dr. Stephan Geley, University of Innsbruck, Austria).
Rabbit phospho-(Ser) CDKs substrate antibody was purchased from Cell Signaling. Peptides and polyclonal rabbit Phospho-Thr194 Gwl were generated and purified by Eurogentec, Belgium. Peptides and polyclonal rabbit phospho-Ser67 Ensa/ARPP19 was generated by Generon and we also obtained a phosphor-specific P-Ser67 Ensa antibody from Dr. Satoru Mochida (Kumamoto University, Japan). To make phospho-specific monoclonal antibody (mAB) to phospho threonine and tyrosine on cdk1, a phospho-peptide EKIGEGpTpYGVVYKGC was coupled to KLH and injected into Balb/c mice. Hybridoma cells produced from the spleen cells from a hyperimmune mouse were fused with Sp/0 myeloma cells using standard procedures. Positive clones were identified using immunoadsorbent assays and immunoblotting of Xenopus egg extracts. mAb CP3.2 gave the strongest signal in these assays.; α -tubulin (clone DM1A), myc (clone 9E10) and CENPA (clone 3–19) mouse monoclonal antibodies were purchased from Abcam (Cambridge, UK), PP2A/B55α (clone 2G9) and PP2A/B55 (clone D-10) mouse monoclonal antibodies from Santa Cruz Biotechnology (Heidelberg, Germany),PP2AC from Sigma (SAB4200266), histone H3 mouse monoclonal antibody (clone 6.6.2) from Millipore (Watford, UK) and mouse monoclonal anti-MASTL from Sigma (HPA027175). HRP-conjugated and polyclonal goat anti-rabbit or mouse antibodies were from Dakocytomation (UK). Secondary antibodies for immunofluorescence were from Invitrogen.
HeLa cells were cultured in Dulbecco's modified Eagle Medium (DMEM) supplemented with 10% FBS, 2 mM L-glutamine, 100 U/mL penicillin and 0.1 mg/mL streptomycin in a 37°C, 5% CO2 incubator. Chicken DT40 cells including cdk1as cells were cultured as previously described (24).
Cdk1as cells were synchronized in G2 with 10 µM 1NMPP1 for 6 h, then were released in mitosis after 1NMPP1 wash out and collected 5 min later for immunofluorescence assays. Alternatively, G2 cells were incubated with 1 µM of OA or/and 10 µM TC for indicated times and harvested for immuno-fluorescence and immuno-blotting analysis.
HeLa cells were synchronized with 5 µM of S-trityl-L-cysteine (STLC; Tocris, Bristol, UK) for ∼18 h, then were harvested by mitotic shake-off. The cells were then incubated in STLC-containing media supplemented with 1 µM of okadaic acid (OA; Calbiochem, Merck Chemicals, Nottingham, UK) or/and 10 µM of tautomycetin (TC; Tocris), or 100 nM of calyculin A (Calbiochem) for 1 h. Next, Cdk1 activity was inhibited with 10 µM of RO3306 (Calbiochem) for 0, 15 or 30 min and the cells were collected for immunoblotting analysis.
To knock-down PP2A/B55α and δ subunits, 40 nM of each siRNA B55 (5′-GCAGAUGAUUUGCGGAUUA-3′ for B55α and 5′-CAUCCAUAUCCGAUGUAAA-3′ for B55δ, Qiagen, Manchester, UK) were reverse transfected using Lipofectamine RNAiMAX reagent (Invitrogen) two consecutive days according to the provider's instructions. The cells were treated with STLC for mitotic synchronization 48 h after the 2nd transfection and mitotic exit experiment was carried out the following day.
Cells were lysed in 40 µL of EBC buffer (50 mM Tris pH 7.5, 120 mM NaCl, 0.5% NP40, 1 mM EDTA, 1 mM DTT, Protease and Phosphatase inhibitors (Complete and PhosStop; Roche Diagnostics, West Sussex, U.K)) and 10 µl 5× sample buffer (0.01% bromophenol blue, 62.5 mM Tris-HCl pH 6.8, 7% SDS, 20% sucrose and 5% β-mercaptoethanol). The samples were sonicated then boiled at 95°C for 5 min. Samples were then analysed by western blotting and the signal was using Immobilon Western Chemiluminescent HRP substrate (Millipore). Image J was used to quantify the intensity of phospho-(Ser) CDKs substrate signal. α-tubulin was used to normalize the samples.
HeLa cells were transfected with 10 µg of Flag-Gwl plasmid (using GeneJuice transfection reagent (Novagen, Merck Chemicals) following the manufacture's guidelines. Hek293T cells were transfected using CaCl2. 24 h later, 5 µM of STLC, 75 ng/mL Nocodazole or 10 µM RO3306 was added to the media and incubated for 16–18 h. Indicated cells were then treated with 1 µM of OA or/and 10 µM of TC or 100 nM of calyculin A for 1 h. Mitotic exit was induced by adding 10 µM of RO3306 for 30 min. Cells were lysed on ice for 20 min in IP buffer (20 mM TrisHCl ph 7.5, 137 mM NaCl, 10% glycerol, 0.5% NP-40, 2 mM EDTA, Protease and Phosphatase inhibitors (Complete and PhosStop)) supplemented with the appropriate phosphatase inhibitors (OA and/or TC or Calyculin A). Protein samples were clarified by centrifugation at 13 000 rpm for 30 min at 4°C. Flag-Gwl protein was captured on anti-flag M2 magnetic beads (Sigma) for at least 2 h at 4°C. After 2 washes with IP buffer and 2 with kinase buffer (50 mM MOPS pH 7.5, 5 mM MgCl2, 0.4 mM EDTA and 0.4 mM EGTA), the beads were incubated with 10 µg of myelin basic protein (MBP, Millipore) diluted in kinase buffer supplemented with 25 µM of ATP and 0.003 MBq of γ-32P-ATP for 20 min at 37°C. The reaction was stopped by adding the sample buffer and was analysed by western blot and autoradiography. Recombinant Cdk2/cycA was a gift from Dr. Julian Gannon (CRUK Clare Hall, UK).
HeLa cells were grown on coverslip, then were fixed with 3.7% of formaldehyde (Sigma-Aldrich, Dorset, UK) in PBS for 10 min (Fig. 2B) or mitotic cells were collected and were spun onto slides at 500 r.p.m. for 5 min then fixed (Fig. 2C). DT-40 cells were cyotspun at 1000 r.p.m. for 3 min then fixed. After several PBS washes, cells were permeabilized with 0.1% NP-40 in PBS fro 10–20 min then stained with indicated antibodies and mounted with Prolong Gold DAPI solution (Invitrogen, Paisley, UK).
Images were acquired on DeltaVision microscope equipped with a UPLS Apo, N.A. 1.40, 100× oil immersion objective (Olympus), standard filter sets (Excitation 360/40; 490/20; 555/28; Emission 457/50; 528/38; 617/40) and a CoolSNAP_HQ2 camera (Photometrics). Images were exported to Omero software (Version 4.4.4). Deconvolution were performed using SVI Huygens Professional Deconvolution Software (Version 3.5) and images were converted into Adobe Photoshop files.
Recombinant His-tagged Ensa was purified using Ni-NTA agarose (Qiagen) from BL21 E.coli following the manufacturer's protocol. 1 mg of purified protein was then phosphorylated for 2 h at 37°C by recombinant engineered Gwl kinase (Hochegger lab, unpublished results) in kinase buffer (20 mM Hepes-HCl pH 7.8, 10 mM MgCl2, 15 mM KCl, 1 mM EGTA, 5 mM NaF, 20 mM β-glycerophosphate, 1 mM DTT and 1.85 MBq γ-32P-ATP (PerkinElmer)). To eliminate unincorporated ATP and the kinase, the phosphor-Ensa protein was captured on Ni-NTA beads. After elution, the phospho substrate was concentrated by spinning on ultrafiltration columns (Vivaspin 500, 3 000 MWCO, Fisher) in buffer containing 20 mM TrisHCl pH 7.5, 150 mM NaCl and 0.01% Tween 20. Hek293 cells were transfected with GFP-FCP1 using a calcium phosphate transfection protocol. 24 h later, cells were treated with 100 ng/mL of nocodazole for 16–18 h and mitotic cells were incubated in DMSO or 10 µM of RO3306 for 30 min at 37°C. GFP-FCP1 was then immunoprecipitated from cell extracts using magnetic GFP-Trap_M beads (Chromotek, Germany) according to the provider's instructions. After washes, beads were resuspended in phosphatase buffer (20 mM HepesNaOH pH 7.9, 10 mM MgCl2, 20 mM KCl, 10% glycerol, 1 mM DTT and, 0.2 mM PMSF). Phospho-Ensa was added to the mixture and samples were incubated for 90 min at 30°C. The reaction was stopped by TCA protein precipitation then the samples were treated as described in [12]. The radioactivity was measured with a liquid-scintillation counter.
Recombinant His-tagged Gwl was produced in insect cells and purified on Ni-Agarose beads. and concentrated by spinning on ultrafiltration columns (Vivaspin 2, 10 000 MWCO) in 40% glycerol/PBS buffer. His-tagged Gwl was phosphorylated in vitro by the recombinant Cdk2/cyclinA complex (a gift from Tim Hunt). The kinase reaction and the purification of the phosphorylated substrate were similar to the preparation of the phosphorylated recombinant ENSA. Flag-B55α was transfected in HEK293 cells. The phosphatase assay with B55α was carried out in a similar way as FCP1 phosphatase assay except the Flag-B55α was captured on anti-flag M2 magnetic beads (Sigma) and phosphorylated his-Gwl was used as a substrate.
|
10.1371/journal.ppat.1003508 | Structural Basis for Native Agonist and Synthetic Inhibitor Recognition by the Pseudomonas aeruginosa Quorum Sensing Regulator PqsR (MvfR) | Bacterial populations co-ordinate gene expression collectively through quorum sensing (QS), a cell-to-cell communication mechanism employing diffusible signal molecules. The LysR-type transcriptional regulator (LTTR) protein PqsR (MvfR) is a key component of alkyl-quinolone (AQ)-dependent QS in Pseudomonas aeruginosa. PqsR is activated by 2-alkyl-4-quinolones including the Pseudomonas quinolone signal (PQS; 2-heptyl-3-hydroxy-4(1H)-quinolone), its precursor 2-heptyl-4-hydroxyquinoline (HHQ) and their C9 congeners, 2-nonyl-3-hydroxy-4(1H)-quinolone (C9-PQS) and 2-nonyl-4-hydroxyquinoline (NHQ). These drive the autoinduction of AQ biosynthesis and the up-regulation of key virulence determinants as a function of bacterial population density. Consequently, PqsR constitutes a potential target for novel antibacterial agents which attenuate infection through the blockade of virulence. Here we present the crystal structures of the PqsR co-inducer binding domain (CBD) and a complex with the native agonist NHQ. We show that the structure of the PqsR CBD has an unusually large ligand-binding pocket in which a native AQ agonist is stabilized entirely by hydrophobic interactions. Through a ligand-based design strategy we synthesized and evaluated a series of 50 AQ and novel quinazolinone (QZN) analogues and measured the impact on AQ biosynthesis, virulence gene expression and biofilm development. The simple exchange of two isosteres (OH for NH2) switches a QZN agonist to an antagonist with a concomitant impact on the induction of bacterial virulence factor production. We also determined the complex crystal structure of a QZN antagonist bound to PqsR revealing a similar orientation in the ligand binding pocket to the native agonist NHQ. This structure represents the first description of an LTTR-antagonist complex. Overall these studies present novel insights into LTTR ligand binding and ligand-based drug design and provide a chemical scaffold for further anti-P. aeruginosa virulence drug development by targeting the AQ receptor PqsR.
| Populations of bacterial cells collectively co-ordinate their activities through cell-to-cell communication via the production and sensing of signal molecules. This is called quorum sensing (QS) and in many bacteria, QS controls the expression of virulence genes, the products of which damage host tissues. Consequently, QS systems are potential targets for antimicrobial agents which do not kill bacteria but instead block their ability to cause disease. Pseudomonas aeruginosa causes a wide range of human infections and produces an armoury of virulence factors. Since many of these are controlled by alkylquinolone (AQ)-dependent QS, we determined the crystal structure of the AQ receptor (PqsR) in order to visualize the shape of the AQ-binding site and better design PqsR inhibitors which compete for the AQ binding site and so block QS. This work in conjunction with the chemical synthesis of AQ analogues resulted in the discovery of potent quinazolinone inhibitors of PqsR. These blocked AQ and virulence factor production in P. aeruginosa as well as biofilm development. Our studies present novel insights into the structure of PqsR and create further opportunities for target-based antibacterial drug development.
| Bacterial cells communicate with each other through quorum sensing (QS), a mechanism for co-ordinating gene expression at the population level via the release and detection of self-generated signalling molecules [1]. Once a critical threshold concentration of QS signal has been attained, a change in collective behavior ensues through the activation of a sensor or regulator protein. In general, QS facilitates the coordination of population behavior to enhance access to nutrients, provide collective defense against other competitor organisms or to encourage community escape where population survival is at risk [1]. QS signal molecules are chemically diverse and include both small peptides and organic molecules such as the N-acylhomoserine lactones (AHLs) and 2-alkyl-4(1H)-quinolones (AQs). In addition, many bacteria possess several interacting QS modules organized into regulatory hierarchies employing multiple signal molecules from the same or different chemical classes. Such QS hierarchies regulate motility and biofilm development, secondary metabolite production, bioluminescence and virulence [1]. With respect to the latter, the global emergence of multi-antibiotic resistant bacteria and the paucity of new clinically effective antibiotics have renewed interest in the development of agents which control infection through the attenuation of bacterial virulence rather than inhibition of growth [2], [3]. In this context, the QS-dependent regulation of virulence offers an attractive suite of potential targets which include the QS signal synthase, the response regulator and the QS signal molecule itself [4], [3].
While there have been extensive attempts to unravel the molecular basis for AHL-dependent QS and to develop inhibitors directed against LuxR-type transcriptional regulators, there is relatively little structural information on the recognition and mechanism of action or inhibition of AQ-type QS signals. AQs are produced by pathogens such as Burkholderia pseudomallei and Pseudomonas aeruginosa [5], [6]. P. aeruginosa thrives in diverse ecological niches and causes both acute and chronic infections in humans, animals, plants and insects. Multi-antibiotic resistant strains have emerged globally as a major cause of hospital-acquired infections for which current therapeutic options are very limited [7]. P. aeruginosa produces diverse exotoxin virulence determinants and secondary metabolites including cyanide, readily forms biofilms and is naturally resistant to many antimicrobial agents. Since many of these virulence genes are controlled by QS [8], P. aeruginosa has emerged as a paradigm pathogen since it employs a sophisticated multi-signal QS system incorporating both AHL/LuxR type and AQ-dependent gene regulatory systems [8] (Figure 1). With respect to the AQs, P. aeruginosa produces over 50 different congeners which were originally identified via their antimicrobial properties but are now known to possess QS, immune modulatory, cytochrome inhibitory, metal chelating, membrane vesicle-stimulating and oxidant activities (reviewed in [9]).
2-Heptyl-3-hydroxy-4(1H)-quinolone (the ‘Pseudomonas Quinolone Signal’, PQS) and its immediate precursor, 2-heptyl-4-hydroxyquinoline (HHQ) are frequently considered to be the primary AQs involved in QS although other active AQ analogues, notably the C9 congeners, 2-nonyl-3-hydroxy-4(1H)-quinolone (C9-PQS) and 2-nonyl-4-hydroxyquinoline (NHQ) are produced by P. aeruginosa in similar concentrations [10], [11]. The synthesis and action of PQS and HHQ and related congeners depends on the pqsABCDE operon, which is positively controlled by the transcriptional regulator PqsR (MvfR) [12], [13]. The first four gene products of this operon are required for AQ biosynthesis [9]. HHQ is released into the extracellular milieu where it is internalized via adjacent cells [14] and oxidized to PQS via the action of the mono-oxygenase PqsH [5], [13], [15]. The function of the pqsE gene product, a putative metallohydrolase, is not currently understood. Although it does not contribute to AQ biosynthesis, it is required for swarming motility biofilm development and virulence and is involved in the negative regulation of the pqsABCDE operon [15], [16]. Strains with mutations in pqsR and pqsA are severely attenuated in experimental animal infection models highlighting the important contribution made by AQ signalling to pathogenicity [16], [12]. Furthermore the presence of AQs in the sputum and broncho-alveolar lavage fluid of cystic fibrosis patients chronically infected with P. aeruginosa provides evidence of their importance in human infection [17], [18].
AQ synthesis and pqsE expression are subject to a positive feedback loop which involves the activation of PqsR by HHQ and PQS and their C9 congeners to drive the expression of the pqsABCDE operon [14], [19], [20], [21], [22]. In whole P. aeruginosa cell assays, HHQ and PQS exhibited EC50s in the low micromolar range for the PqsR-dependent activation of pqsA [23]. Activation of PqsR depends on the AQ alkyl chain length [23], PQS congeners with C1, C3, C5 or C11 alkyl chains exhibit only weak activities compared with the C7 compounds. However, the C9 congeners, 3-hydroxy-2-nonyl-4(1H)-quinolone (C9-PQS) and 2-nonyl(1H)-4-hydroxyquinoline (NHQ) are highly active [23]. Although both HHQ and PQS can activate PqsR, the PqsH-mediated introduction of a hydroxyl group at the 3 position of the quinolone ring confers additional physicochemical functionality to PQS over HHQ including iron chelation [14] and microvesicle formation [24].
Attempts to attenuate AQ signaling and the virulence of P. aeruginosa without perturbing bacterial growth have so far mainly focused on enzymes which inactivate PQS [25] and methylated or halogenated derivatives of the AQ precursor anthranilate such as 2-amino-4-chorobenzoic acid (4-CABA) which inhibits AQ biosynthesis probably at the level of PqsA by competing with anthranilate for the enzyme active site [26], [27]. This approach has recently shown promising results in limiting the systemic proliferation of P. aeruginosa infection in mice although the concentrations required to inhibit AQ production were high (millimolar range; [27]) and so unlikely to be clinically useful. More recently a number of PqsD inhibitors have been identified [28].
An alternative approach for blocking AQ biosynthesis and signalling and hence virulence would be to target the response regulator PqsR. Although the structures of the PqsR-activating ligands PQS and HHQ are known (Figure 2), there is no information on the PqsR ligand binding site. PqsR belongs to the LysR family of transcriptional regulators (LTTRs) which are widespread in bacteria [29] among which P. aeruginosa appears to have one of the largest repertoires [30]. LTTR proteins generally possess a highly conserved N-terminal helix-turn-helix (HTH) DNA-binding domain but a poorly conserved C-terminal ligand-binding domain usually termed the co-inducer binding domain (CBD) where the co-inducer is a low molecular weight ligand [29]. LTTRs function as either activators or repressors in feedback loops in which the co-inducer ligand is required for transcriptional control. LTTR proteins are generally thought to function as tetramers recognizing multiple binding sites within the promoter/operator region of the target gene(s) which include a regulatory binding site (RBS) incorporating the LTTR box (general consensus T-N11-A) and an activation binding site (ABS). Occupation of these sites results in DNA bending and contact with RNA polymerase to initiate transcription [29]. Structural studies of LTTRs have been mostly restricted to the C-terminal CBD because of the insolubility associated with the HTH domain [29]. LTTR CBDs typically have two α/β sub-domains (CBDI and CBDII) connected by an anti-parallel β-sheet known as the hinge region [31], [32]. Well characterised ligands for LTTRs include catechols and chlorinated aromatics which are co-inducers for CatM and BenM respectively binding into a pocket between the CBD subdomains [29].
With respect to PqsR, how this unusual LTTR [30] recognizes and responds to the larger hydrophobic ligands associated with AQ-dependent QS is not known. The structural basis for the recognition of AQs by PqsR has not been elucidated and there is consequently a lack of molecular detail to facilitate the development of PqsR inhibitors as novel therapeutics. Through a multidisciplinary effort we provide new insights into the structure of the PqsR co-inducer binding domain (PqsRCBD) in complex with both a native AQ agonist and a potent quinazolinone (QZN) antagonist. The QZN scaffold was evolved through ‘ligand based design’ and we show that in P. aeruginosa a very simple isosteric replacement is sufficient to switch a QZN from potent agonist to potent antagonist turning QS-dependent virulence gene expression on or off respectively. Although other LTTR agonist complex structures have been described this, to our knowledge, is the first description of an LTTR-antagonist complex crystal structure.
To investigate the molecular basis of PqsR ligand recognition we first sought to determine the crystal structure. The full length PqsR receptor containing both the DNA-binding domain and CBD was insoluble when expressed in Escherichia coli. We thus focused on a construct spanning C-terminal residues 94-332 incorporating the CBD (Figure 3A). This was soluble and was utilised for initial crystallisations together with a truncated version which removed the 23 amino acid C-terminal tail. Crystals were obtained only in conditions where the precipitant, 2-methyl-2,4-pentanediol (MPD) was present and the structure was determined using SAD and SIR phases in spacegroup P6522 to 2.5 Å resolution (Table 1). Crystallisation of the PqsRCBD has been reported previously by two groups at 5 Å [33] and 3.25 Å resolution respectively [34]. The latter P6522 crystal form has similar cell dimensions to the crystal form reported here although the crystallisation conditions differ and MPD is not present. The construct reported in Xu et al. [34] has different domain boundaries spanning residues 91–319 compared to residues 94–294 and 94–332 reported here.
The PqsRCBD structure has one molecule in the crystallographic asymmetric unit. Analytic gel filtration was performed with a calibrated Superdex 75 10/300 column revealing the recombinant PqsRCBD eluted close to the 43 kDa marker (ovalbumin) and above the 29 kDa marker (carbonic anhydrase) (Figure S1A). As the calculated monomer molecular weight for PqsRCBD is 22.7 kDa this indicates a dimeric species is present in solution. This is in agreement with the gel filtration data for the PqsRCBD reported by Xu et al. [34].
The topology of the PqsRCBD structure is illustrated in Figure 3B showing two subdomains (CBDs I and II) connected by an antiparallel β-sheet termed the hinge region as observed in related LTTR structures of BenM and CatM [35], [36]. Overall, CBDI is similar to other LTTRs with the exception of helix α2 which is shorter in PqsR. In CBDII a number of large changes are observed as a helix present in BenM between strands β5 and β6 is absent and replaced by an irregular loop structure (L1) lying at the junction of the two subdomains. The hinge region also adopts a different conformation in PqsRCBD and the β4, β8 antiparallel strands are repositioned closer to the C-terminal helix. These two changes of a helix removal and the repositioning of the hinge region effectively open up the CBDII structure to form a large hydrophobic pocket occupied by two MPD molecules (Figures 3C and 3D).
MPD molecule 1 (MPD1) is highly buried, surrounded by aliphatic residues (Leu, Ile, Ala) from both sub-domains and a water mediated interaction with the carbonyl group of Ile236 at the bottom of the pocket (termed the B pocket, Figure 3D). MPD2 is predisposed closer to the surface in CBDII interacting with the side chains of Tyr258 and Val211 (the A pocket, Figures 3C and 3D).
We examined the crystal lattice and a large dimer interface was observed with a buried surface area of 1232 Å2 (topology shown in Figure 3E). A feature of the dimer is that it presents the two CBDII pockets on the same surface by aligning the two hinge regions which interact at either end of strands β4 and β8 forming contacts. At the centre, the β4 side chain Lys266 is fully extended forming a salt bridge to Glu259 close to the dimer axis. Further interfacial contacts occur between helices α3 and α6 at either end of the dimer (Figure 3E). A second feature is the two interlocking loops which form a lid partially covering the MPD molecules (lid loops L1 and L2 shown in red and blue respectively in Figure 3F). The pocket in each monomer connects across the dimer interface via a narrow channel shown as mesh in Figure 3E.
A second crystallographic dimer interface occurs in the PqsR lattice which buries a smaller surface area (614 Å2). This involves an antiparallel β-sheet formed by strand β2 as well as antiparallel packing of the two N-terminal α1 helices (Figure S1B). One side of the interface consists of hydrogen bonds from the β-sheet and sidechain-sidechain interactions from symmetry related Ser123 side chains (Figure S1 C and D). The residue Ser112 side chains from helix α1 form a similar side chain hydrogen bond close to the dimer axis. In the centre, hydrophobic contacts come from residues above and below the β-sheet (Figure S1 C and D).
We next sought to determine how the PqsRCBD recognises a naturally occurring AQ co-inducer. As the PqsRCBD lattice has a high solvent content (72%) we performed crystal soaking experiments with the most active native ligands produced by P. aeruginosa namely HHQ, NHQ (C9-HHQ), PQS and C9-PQS (Figures 2 and 4A) as well as shorter carbon chain length derivatives. For the majority of these experiments, a calculation of difference (Fo-Fc) maps resulted in the characteristic electron density for two MPD molecules occupying the PqsRCBD pocket and no evidence of a bound ligand (Figure S2A). By contrast the soaking of NHQ for 24 h resulted in elongated and connected electron density spanning the two sub-pockets (Figure S2B). The electron density in the deeper B pocket was observed to be planar in shape and model building allowed fitting of the quinolone moiety of NHQ in one unique orientation. The alkyl chain can be modelled extending into the remaining cigar shaped electron density to occupy the A pocket. The Tyr258 side chain pins the alkyl chain down against side chains from residues Ile186, Val170, Leu189, Ile236 from the bottom and sides of the A pocket forming a comfortable fit (Figures 4B and 4C). The bicyclic ring structure is enclosed on either side in the B pocket by contacts from Leu207, Leu208 and Ile236. From above, contacts come from Ile149 and Ala168 and from below the sidechain of Phe221 (Figure 4B).
The fit of the bicyclic ring into the B pocket is not precise and beneath the carbonyl and in front of the aliphatic ring two vacant sub-pockets are present. A surprising feature of the structure is that the interactions are all hydrophobic with the absence of any hydrogen bonds or electrostatic interactions to the NHQ carbonyl oxygen or NH group of the bicyclic ring (Figure 4D). Although the interactions with NHQ are exclusively hydrophobic, modelling of the additional OH present in the alternative co-inducer, PQS, reveals this group could potentially form an additional contact through a hydrogen bonding interaction with the carbonyl group of Leu207.
Superposition of the PqsRCBD-NHQ with the PqsRCBD-MPD structure reveals subtle local conformational changes in the binding pocket (Figure 4E). The Thr265 side-chain rotates by 180° to make a direct contact with NHQ and a concomitant 0.7 Å movement of the main chain affects a similar movement in the position of the adjacent residue Asp264 in the hinge region affecting strands β4 and β8 (Figure 4E).
To probe the contribution of the key PqsRCBD amino acid residues identified above (Figure 4B) to AQ binding, 13 site-specific substitutions were introduced into a 6His-tagged PqsR which retains activity in P. aeruginosa (Figure 5A). PqsR mutant functionality was evaluated by (i) Western blot to confirm PqsR expression (Figure 5C) and (ii) the ability to restore pqsA expression in the P. aeruginosa ΔpqsR miniCTX::pqsA'-lux reporter strain (Figure 5B). Each of the pqsR mutations altered pqsA promoter activity with mild reductions observed for three mutations Ile186Ala, Ile236Phe and Leu207Glu which exhibited 44.1%, 20.5% and 13.7% of the wild-type PqsR-6His, respectively (Figure 5B). In the PqsRCBD structure, Ile186 is positioned at the far end of the A pocket and substitution by Ala removes the side chain atoms which contact the end of the alkyl chain; a loss of this contact would be predicted to reduce the interaction with the A pocket. Ile236 is positioned at the bottom of the A pocket lying at a boundary with the B pocket. It is fully buried by bound NHQ in the complex structure making contacts between its side chain atoms and the planar surface of the bicyclic ring. Mutation to Phe would be predicted to disrupt binding by introducing extra volume and changing the shape of the pocket. Leu207 is positioned on the side of the B pocket and interacts with NHQ at the junction between the alkyl chain and the bicyclic ring through its terminal side chain atom. The Leu207Ala and Leu207Glu mutations both result in a ∼10% response compared with wild type indicating that the altered size and charge similarly affect optimal ligand contacts. An almost complete loss of activity (<2% of that of the wild type) is observed for four mutations (Ile149, Phe221, Tyr258, Ile263). As we are unable to purify a soluble form of the full length wild type or mutant PqsR variants we could not rule out whether these four mutations have an effect by disrupting protein folding rather than ligand binding. Furthermore, we also observed that the PqsRCBD protein precipitates in the presence of its highly hydrophobic ligands even at dilute protein concentrations (a finding also noted by Xiao et al., [21]) making alternative biophysical approaches difficult.
Using a P. aeruginosa ΔpqsA ΔpqsH ΔpqsR triple mutant which does not produce endogenous AQs and containing a chromosomally-integrated miniCTX::pqsA'-lux reporter, we examined the response of the PqsR-6His variants to PQS and HHQ. The data presented in Figure 5D shows that the variants exhibited reduced responses to PQS and HHQ, consistent with the data in Figure 5B. However, the degree of response was significantly different for the two co-inducers, with PQS giving a consistently higher response, e.g. the Leu207Glu mutation responded much more poorly to HHQ (5%) than to PQS (39%). This observation is consistent with the notion that the additional 3-OH group of PQS can potentially form a hydrogen bond to the main chain carbonyl of Leu207 and thus it may not rely as heavily as HHQ/NHQ on the side chain interaction with Leu207 which is altered in this mutant.
To conserve the steric requirements for optimal ligand/receptor interactions, antagonists have frequently been discovered through structural modification of native agonists. Hence we used the closely related 2-alkyl-4(3H)-quinazolinone (QZN) system as a template to probe structure-activity relationships (SARs). We focused on QZN analogues with C7 or C9 alkyl side chains as AQ congeners with C5 or C11 have little activity ([23]). A series of 42 variously substituted QZNs (Figure 6) was synthesized and characterized as described in Text S1 and the corresponding EC50s and IC50s for each compound determined via dose–response curves generated using PqsR-dependent P. aeruginosa miniCTX::pqsA'-lux reporter gene fusion assays. For comparative purposes, Figure 2 summarizes the EC50 data obtained for PQS, HHQ and their corresponding C9 congeners, NHQ and C9-PQS in both P. aeruginosa ΔpqsA and P. aeruginosa ΔpqsAH mutant backgrounds since both HHQ and NHQ can be converted to the corresponding 3-hydroxy compound by the mono-oxygenase, PqsH [15]. Figure 2 shows that the 4 co-inducers have similar EC50s in P. aeruginosa.
The SAR data for the QZNs is summarized in Figure 6. C7-QZN and C9-QZN (Figure 6, 7 and 8) which are 3-aza analogues of HHQ and NHQ respectively, and 7F-C9-QZN (Figure 6, 8), all of which lack a substitution in the 3-position, were devoid of agonist or antagonist activity. Hydroxylation at the 3-position of C7-QZN and C9-QZN gave compounds (Figure 6, 9 and 10) which were substantially weaker agonists than PQS. These agonist properties were substantially improved by the introduction of a halogen substituent at the 7-position of the carbocyclic ring as in 3-OH-7Cl-C9-QZN and 3-OH-7F-C9-QZN (Figure 6, 11 and 12). However the 3-methoxy variants, 3-OMe-C7-QZN and 3-OMe-C9-QZN (Figure 6, 13 and 14) were inactive unless halogenated at C-7 (Figure 6, 15 and 16). In this QZN series, 3-OMe-7F-C9-QZN (Figure 6, 16) was as potent an agonist as PQS. Introduction of a second fluorine to give 3-OMe-6F,7F-C9-QZN reduced potency by ∼25-fold (Figure 6, 17)
To explore the QZN SAR further and to identify an essential pharmacophore for antagonist activity, we replaced the 3-OH group with the isosteric 3-NH2 group in the above derivatives and synthesized 3NH2-C7-QZN (Figure 6, 19) and its alkyl chain altered variants (Figure 6, 20, 21, 29–32). None of these compounds were agonists. This was particularly interesting given that replacement of the 3-OH in PQS with 3-NH2 (Figure 2, 1 and 5a) results in a compound which is a more potent agonist than the natural ligand (EC50 0.4±0.15 µM).
QZNs with branched chains (Figure 6, 29 and 30), unsaturation (Figure 6, 31) or with increased hydrophilicity (Figure 6, 32) were all inactive while compounds 19, 20 and 21 (Figure 6) were antagonists, the most potent being 3NH2-C7-QZN (Figure 6, 19; IC50 54±15.5 µM). An attempt to further improve the activity via the introduction of a 6-Cl substituent in 3NH2-C9-QZN to yield 3NH2-6Cl-C9-QZN (Figure 6, 22) resulted in the complete loss of antagonist activity but gratifyingly, potency was greatly increased by a 7-Cl substituent (3-NH2-7Cl-C9-QZN; Figure 6, 23; IC50 5±1.6 µM). Fluorine substituted derivatives (Figure 6, 25 and 26) were also antagonists, the most potent compound being 3-NH2-6F,7F-C9-QZN (IC50 1.2±0.4 µM).
Introduction of an electron withdrawing group CF3 at C-7 as in 3-NH2-7CF3-C9-QZN or 8-aza as in 3-NH2-8aza-C9-QZN (Figure 6, 27 and 47 respectively) resulted in the complete loss of activity. The presence of electron donating methoxy substituents as in 3-NH2-6OMe,7OMe-C9-QZN (Figure 6, 28) also rendered the compound inactive. Further modification of the 3-NH2 group of the C9-QZNs by acetylation, dimethylation, or 2-aminoethylation (Figure 6, 33, 34 and 35 respectively) yielded only inactive or weak antagonists, the potency of which could be increased as before by the presence of a Cl at the 6 or 7 position of the carbocyclic ring (Figure 6, 36 and 37). The 3-(3-aminopropyl) compound (Figure 6, 38) only showed marginal improvement in activity. However these compounds (Figure 6, 35–38) also exhibited growth inhibitory activity and were therefore excluded from further work given that a primary objective was to obtain PqsR inhibitors which attenuate P. aeruginosa virulence without inhibiting growth.
Compounds 39–46 (Figure 6) represent further attempts to increase antagonist potency through modification of the alkyl side chain by synthesising derivatives where the alkyl chain is terminally substituted with aryl (Figure 6, 39–42), heteroaryl, biaryl or cyclohexyl groups (Figure 6 compounds 43,44,45 and 46) respectively. These QZNs were all inactive apart from the phenyl substituted compounds 3-C2NH2-7Cl-PhC3-QZN and 3-NH2-7Cl-PhC3-QZN (Figure 6, 39 and 41) which were weak antagonists (IC50s 79.2±2.7 µM and 39.6±11 µM respectively).
To determine whether the QZN antagonists are competitive inhibitors which interact with the AQ-binding pocket, we first investigated whether inhibition of PqsR by 3-NH2-7Cl-C9-QZN could be overcome by increasing concentrations of PQS. The data are shown in Figure 7A which reveals that PQS above 25 µM competitively overcomes QZN-mediated PqsR inhibition. To investigate how the 3-NH2-7Cl-C9-QZN interacts with PqsRCBD crystals were soaked in a solution of the compound. The resulting structure revealed the quinazolinone moiety is buried in the B pocket with the alkyl chain extending into the A pocket. The QZN molecule forms very similar hydrophobic interactions with the pocket noted for NHQ (Figure 7B). In addition, the Cl atom of the 3-NH2-7Cl-C9-QZN occupies the vacant sub pocket present in front of the aliphatic ring and forms a hydrogen bond with the side chain of Thr265 (Figure 7C). The 3-NH2 substituent forms a hydrogen bond to the main chain carbonyl oxygen of Leu207 (Figure 7 B and D). Superposition of this structure with the agonist NHQ-PqsRCBD complex reveals differences resulting from the additional contacts made by the QZN (Figure 7D). This tilts the QZN bicyclic ring relative to NHQ affecting a subtle repositioning of the alkyl chain. The QZN interactions affect small changes in the main chain of the L1 loop and the 7-Cl atom induces a rotation of the Thr265 side chain rotamer by 90° to NHQ. In addition, the introduction of the 7Cl substituent into PQS to generate 7Cl-PQS resulted in an agonist which is ∼135 times more potent than PQS itself (Figure 2) providing further confirmation of the importance of the vacant sub-pocket adjacent to the Thr265 residue.
The QZN antagonists of PqsR were identified on the basis of their inhibition of pqsA transcription through the competitive antagonism of the AQ-dependent activation of PqsR. To determine whether the QZNs could also inhibit the expression of target virulence genes such as lecA (which codes for the cytotoxic galactophilic lectin protein LecA, which also contributes to biofilm development; [37]) and phzA1 (which codes for an enzyme involved in the biosynthesis of the redox-reactive phenazine pigment pyocyanin; [38]), we constructed lecA-lux and phzA1-lux reporter gene fusions integrated in the chromosome of wild type P. aeruginosa PAO1. Since lectin A and pyocyanin production depend on pqsE expression [16], which in turn requires the PqsR-dependent activation of the pqsABCDE operon, inhibitors of PqsR should result in the down-regulation of lecA and phzA expression. Compared with the control, the expression of lecA is reduced by 3-NH2-7-Cl-C9-QZN (Figure 8A). Similar results were obtained with the phzA1 promoter for which 12.5 µM 3-NH2-7Cl-C9-QZN reduced activity by ∼50% (data not shown). In agreement with the findings for the phzA1 promoter, pyocyanin levels were also substantially reduced in P. aeruginosa cultures treated with the QZN (Figure 8B). These data are consistent with a reduction in AQ levels and LC-MS/MS analysis of P. aeruginosa cultures grown in the presence of 3-NH2-7Cl-C9-QZN shows that the production of HHQ, NHQ, their corresponding N-oxides as well as PQS are reduced to very low levels (Figure 8C). Since AQ-dependent QS also contributes to biofilm maturation we examined the impact of 3-NH2-7Cl-C9-QZN on biofilm development under flow conditions using a microfluidics device. Representative confocal microscope images of the green fluorescent protein (GFP)-tagged P. aeruginosa wild type grown in the presence or in the absence of 3-NH2-7Cl-C9-QZN, and of the P. aeruginosa ΔpqsA mutant strain are shown in Figure 8D. In common with the ΔpqsA mutant, 3-NH2-7Cl-C9-QZN-treated wild type biofilms exhibited reduced surface area coverage. This is consistent with previous reports on reduced biofilm formation in P. aeruginosa pqsA mutants, primarily as a result of a reduction in the release of extracellular DNA, an important constituent of the extracellular matrix which is released via a process that requires PQS [39], [40].
Understanding the molecular recognition of AQs has important implications for gaining insight into the molecular basis of the PqsR receptor ligand and inhibitor interactions. We determined the crystal structure complex of the PqsRCBD domain with native agonist and synthetic antagonist ligands. This revealed a core structure similar to that of other LTTR proteins incorporating two sub-domains (CBDI and CBDII). Among LTTR proteins, the CBD domains have the same overall topology despite little sequence similarity [36]. However, in contrast to other LTTRs, where a small primary ligand-binding pocket is located in the cleft between the two sub-domains [35], the PqsR ligand binding site is larger extending into CBDII as well as occupying a large B pocket between CBDI and CBDII (Figure S3). This pocket is partially covered by lid loops upon formation of a large dimer interface. A central antiparallel dimer organisation is common among LTTR crystal structures, however these are formed between the ‘cleft’ side of the monomer whereas the PqsRCBD dimer is formed from the hinge region β-strands.
Crystal soaking experiments with NHQ revealed that this hydrophobic cavity within the PqsRCBD constitutes the AQ-ligand binding site where the quinolone moiety is buried within the B pocket, the alkyl chain extending into the surface crevice of the CBDII A pocket. The PqsRCBD-NHQ complex is stabilised entirely by hydrophobic interactions and no electrostatic interactions are involved.
In P. aeruginosa reporter gene fusion assays, HHQ, PQS and their C9 congeners (NHQ and C9-PQS) each have similar EC50s (Figure 2; [23]) whereas the C11 congeners are inactive, consistent with the space constraints noted from the crystal structure. The importance of the amino acids observed to form the PqsRCBD ligand binding site in the crystal structure was investigated using site-directed mutagenesis. Of the 13 residues mutated, all resulted in a major reduction in activity except for the I186A mutation which is located at the edge of the A pocket and retained ∼44% activity. Interestingly the increased A pocket space available as a consequence of the I186A replacement did not increase the activity of the C9 (i.e. NHQ) or the C11 congeners of HHQ (data not shown).
PQS and its biosynthetic precursor HHQ (as well as their C9 congeners) can both act as activating PqsR co-inducers. Using a P. aeruginosa ΔpqsA ΔpqsH double mutant which cannot convert exogenously supplied HHQ to PQS, we observed a ∼3-fold higher induction of the pqsA promoter by PQS when compared with HHQ (Figure 2). This is much lower than the 100-fold higher induction reported by Xiao et al. [21] using a different P. aeruginosa strain and reporter assay. PQS has also been reported to potentiate the binding of recombinant PqsR in crude E. coli lysates to DNA more effectively than HHQ [20], [21]. The greater efficiency of PQS over HHQ may be a consequence of the increased H-bonding opportunities with the Leu207 carbonyl afforded by the presence of the 3-OH substituent. In this context it is perhaps noteworthy that the L207A and L207E PqsR mutants are significantly more responsive to PQS than to HHQ. Although we were unable to obtain a structure for the PQS complex with PqsRCBD, it is anticipated that as both the QZN and NHQ PqsRCBD structures superpose accurately, that their similarity in chemical structure with PQS will result in a similar binding. Nevertheless, the primary advantage of introducing a 3-OH substituent in HHQ is probably not enhancement of PqsR activation but to confer additional functionalities since PQS, unlike HHQ, induces outer membrane microvesicle formation [24] and is a potent iron chelator [14].
LTTRs assemble into oligomers (tetramers, and in one case an octamer) which is the functional DNA-binding structure affecting DNA bending and the recruitment of RNA polymerase [41], [35], [29]. The degree of DNA bending is determined by the presence or absence of the co-inducer which causes a conformation change in the LTTR resulting in a relaxation of the degree of DNA bending [29], [42]. In the redox switch LTTR protein OxyR, the reduced form is a tetramer and the activated, oxidised form undergoes a conformational change in CBDII affecting a CBDI interface within the tetramer. This is thought to position the DNA binding domains appropriately for interaction with DNA and the RNA polymerase [41]. The nature of the quaternary arrangement formed by PqsR in complex with DNA and co-inducer has yet to been elucidated. Progress towards this goal is hampered by the inability to prepare full-length recombinant PqsR receptor heterologously expressed in E. coli, an important goal for any future studies in this area.
A comparison of the PqsRCBD structure with other LTTRs reveals that the L1 loop in CBDII occupies the same region of the topology as key residues required for co-inducer mediated conformational changes [41], [35], [32]. In OxyR, the region equivalent to the PqsR L1 loop switches conformation upon oxidation or reduction and this is linked to a re-organisation of the tetramer interface [41]. In TsaR and CatM, co-inducer interactions and conformational switching are mediated by an α-helix occupying the same position as the PqsR CBDII L1 loop [35], [32]. Thus co-inducer affected changes in the region of the L1 loop may be the first steps on the pathway to activation of the receptor and hence gene expression.
In our search for potent PqsR antagonists as novel therapeutics, we focused on the QZN system since sterically it is closely related to the natural AQ ligands. We systematically varied the nature and size of the substituents at the 2 and 3 positions in the heterocyclic ring as well as positions 6 and 7 in the carbocyclic ring of the QZN structure to deliver a range of analogues with pharmacophores that may have the desired stereo-electronic properties for antagonist activity. Thus a total of 46 QZNs (Figure 6) were synthesised, characterised (see supplemental Text S1) and assayed for their agonist and antagonist activities in a whole P. aeruginosa bacterial cell assay. The QZN analogues lacking substitution at C-3 (Figure 6, 5–7) were inactive although their corresponding AQs, HHQ and NHQ (Figure 2, 2 and 3) displayed strong agonist activity. Substitution at C-3 with OH gave analogues 8 and 9 (Figure 6) which like the corresponding PQS and C9-PQS (Figure 2, 1 and 4) were partial agonists. The 3-methoxy derivatives 12 and 13 (Figure 6) were weak antagonists but were totally devoid of agonist activity. Surprisingly, substitution with a halogen in the C-7 position in the carbocyclic ring reversed the activity and indeed 3-OMe-7F-C9-QZN (Figure 6, 15) had potent agonist activity (EC50 2.2 µM) comparable with that of the natural ligands. The introduction of a second halogen at C-6 has the opposite effect and consequently 3-OMe-7F-C9-QZN (Figure 6, 16) is a much weaker agonist.
The QZN compounds synthesized fell in two distinct groups with 3-OH (and OMe) QZNs generally behaving as agonists and the 3-NH2 QZNs as competitive antagonists in whole bacterial cell assays of PqsR activity. Additionally, an alkyl chain of 9 carbons at C-2 in the heterocyclic ring and a halogen at C-7 in the carbocyclic ring are essential for optimum activity in both series. 3-NH2-7Cl-C9-QZN (IC50 5.0 µM), 3-NH2-7F-C9-QZN (IC50 4.3 µM) and 3-NH2-6F,7F-C9-QZN (IC50 1.7 µM) were the most potent antagonists discovered in our studies. Of these, 3-NH2-7Cl-C9-QZN was shown to be an effective inhibitor of AQ signaling by antagonizing AQ biosynthesis, virulence gene expression, pyocyanin production and biofilm development. Furthermore, the preference for a halogen at C-7 over C-6 is clearly apparent from the PqsRCBD/3-NH2-7Cl-C9-QZN complex structure (Figure 7C) where additional H-bonding opportunities for the 7Cl substituent are afforded by the pocket formed by the Thr265. Thus it is probable that 3-NH2-7Cl-C9-QZN binds more strongly to the PqsRCBD than the native ligands via the strengthened electrostatic interactions between 3-NH2 substituent, the water molecule and Leu207 backbone carbonyl in conjunction with additional H-bonding between the 7-Cl and Thr265. However this will require experimental verification.
Remarkably, the simple replacement of the 3-OH with 3-NH2 in the 7Cl-substituted QZNs converts the compound from a potent agonist to a potent antagonist (Figure 6, 11 and 23). The importance of this small but significant finding is that the replacement of the PQS 3-OH with 3-NH2 does not affect the switch from agonist to antagonist (as both compounds are strong agonists with similar EC50s (Figure 2; 1.9 µM compared with 0.4 µM). This indicates that for the QZNs compared with the AQs, the stereo-electronic consequences of the additional QZN ring nitrogen are profound in terms of PqsR activation.
In addition to the preliminary SAR studies for PQS agonists [23], [43], PqsR antagonists have recently been described by Klein et al. [44] and by Lu et al. [45]. The former identified substituted benzamides lacking extended alkyl chains which bind to the PqsRCBD and exhibit relatively weak agonist or antagonist activities by using (±)-trans-U50488 as a template for rational design since this κ-opioid receptor agonist was reported to stimulate pqsA transcription [46]. The biological evaluation of these compounds and those of Lu et al. [45] have mostly been undertaken using a heterologous E. coli-based PqsR-dependent transcriptional reporter which is more sensitive to PQS than P. aeruginosa [23], [43]) probably as a consequence of the numerous efflux pumps present in the latter. HHQ analogues with electron-withdrawing C6 substituents (nitrile (–CN), triofluoromethyl (–CF3) and nitro (–NO2)) were potent antagonists in E. coli whole cell assays with EC50s in the nanomolar range [45]. However, at the concentrations tested, they failed to reduce PQS production in P. aeruginosa although pyocyanin levels were lower after treatment with the 6-CF3 analogue. Interestingly, the 7-CF3-substituted HHQ, in contrast to the 6-CF3 analogue, was devoid of antagonist activity and retained agonist activity at about 50% that of HHQ [45]. In the E. coli-based assay, PQS analogues with Cl substitutions at 5, 6, 7 or 8 all exhibited similar activities to PQS [43]. However, in P. aeruginosa, 7-Cl-PQS is ∼135× more potent than PQS (Figure 2), a finding consistent with the PqsRCBD/3-NH2-7Cl-C9-QZN complex structure which revealed that a 7-Cl substituent can occupy a pocket and form an H-bond with the side chain of Thr265.
Competitive PqsR antagonists such as 3-NH2-7Cl-C9-QZN bind within the PqsRCBD ligand binding pocket in the same orientation as agonists such as NHQ and form additional hydrogen bonds to the side chain OH of Thr265 and the main chain carbonyl of Leu207. Since LTTR agonists stimulate transcription of target genes through changes in the orientation of the DNA and ligand-binding domains [32] this would suggest that the binding of 3-NH2-7Cl-C9-QZN, although likely to be tighter than an agonist, is not productive and either maintains the PqsR conformation in the same state as the unbound protein or drives the formation of a different, but inactive, conformation. The latter mechanism has been reported for an antagonist that binds in place of the native N-acylhomoserine lactone ligand to the LuxR family protein CviR and forces the transcriptional regulator to adopt a conformation incompatible with high affinity DNA operator binding [47]. Here we have found that superposition of the PqsRCBD-3-NH2-7Cl-C9-QZN and PqsRCBD-NHQ complexes results in subtle changes which tilt the bicyclic ring of the QZN relative to that of NHQ so the QZN interaction is not productive for PqsR activation.
Taken together this work has increased our knowledge of the molecular recognition of ligands by PqsR, demonstrated how the simple and subtle exchange of two isosteres (OH for NH2) within a co-inducer molecule can effectively switch virulence gene expression on or off and provided a template structure for the development of QZNs as novel therapeutics which control infection through attenuation of P. aeruginosa virulence.
The E. coli and P. aeruginosa strains used in this study (Table S1) were grown in Lysogeny broth (LB) at 37°C. For AQ quantification, P. aeruginosa was grown in minimal medium [48]. When required for plasmid maintenance in E. coli (pET28a derivatives) or P. aeruginosa (pME6032 derivatives), kanamycin (50 µg/ml) or tetracycline (125 µg/ml) were respectively added to the growth medium.
The P. aeruginosa ΔpqsR in-frame deletion mutant and the triple ΔpqsA ΔpqsH ΔpqsR mutant were constructed in the PAO1 parent and ΔpqsA ΔpqsH mutant [14] respectively using the pDM4ΔpqsR plasmid. The upstream and downstream fragments of pqsR were amplified by PCR from PAO1 chromosomal DNA using the primers pairs FWpqsRUp-RVpqsRUp and FWpqsRDown-RVpqsRDown, respectively (Table S2), introduced into pDM4 [49] and the resulting ΔpqsR mutants obtained by allelic exchange [50]. The pqsA promoter fused to the luxCDABE reporter operon was introduced into both the ΔpqsR and ΔpqsA ΔpqsH ΔpqsR mutants using the miniCTXpqsA-lux plasmid as described previously [23].
For complementation assays, the pqsR gene with or without a C-terminus 6xHis coding sequence (pqsR-6H) was amplified by PCR from chromosomal DNA with primer pairs FWpqsR-RVpqsR or FWpqsR-RVpqsR-6H, respectively (Table S2), and cloned by EcoRI-SacI digestion into pME6032 [51]. Site-directed mutations were generated in pqsR-6H using the splicing by overlap extension PCR method [52]. Briefly, in the first step two distinct PCRs (PCR-1 and PCR-2) were carried out for each pqsR-6his derivative mutant, using chromosomal P. aeruginosa DNA as a template. Each PCR-1 was performed with the forward primer FWpqsR and with a mutagenic reverse primer carrying the mutation in the desired codon (Table S2). Each PCR-2 was performed with a mutagenic forward primer complementary to the reverse primer utilized in the corresponding PCR-1 (Table S2) and with primer RVpqsR-6H. In the second PCR step products obtained from PCR-1 and PCR-2 for each mutation were spliced together using the FWpqsR and RVpqsR-6H primers. The mutated pqsR-6H variants were cloned by EcoRI/SacI digestion in pME6032 and verified by DNA sequencing. The expression of each of the PqsR variants was confirmed by Western blot analysis using a mouse anti-6xHis antibody (1∶1,000; Sigma-Aldrich, St. Louis, MO, USA).
For overexpression of the PqsRCBD, the regions corresponding to the PqsR C94-332 and PqsR C94-294 CBD were amplified by PCR and cloned into pET28a. The recombinant plasmids were introduced by transformation into E. coli Rosetta 2 (DE3) and grown at 37°C to an OD600 0.8 prior to induction with IPTG (1 mM) at 20°C for 16 h. After harvesting by centrifugation, the bacteria were lysed by sonication, centrifuged and filtered to remove cellular debris prior to nickel affinity column purification and elution with an imidazole gradient (0 to 1 M). The 6xHis tag which included the thrombin cleavage sequence was removed from the PqsRCBD proteins using thrombin (Novagen; enzyme/substrate ratio 1∶1000 for 24 h). Final purification was achieved by gel filtration using a Superdex 75 16/60 gel filtration column, with a mobile phase consisting of 20 mM Tris-HCl, 150 mM NaCl, pH 7.4. Both constructs yielded approximately 25 mg PqsRCBD/litre of culture as confirmed by SDS-PAGE. The same strategy was used to obtain the PqsR94-332 selenomethionine-labelled protein after transforming the E. coli methionine auxotroph strain B834 (DE3) with pPqsR94-332 and growing the recombinant strain in selenomethionine medium. A protein concentration of 25 mg/ml was used for 96-well crystal screening (Qiagen kits) and crystals were obtained for PqsR94-332 and from several conditions using only the MPD suite (Qiagen) after 24 h at 19°C. Optimized conditions in 24-well sitting drop plates containing a reservoir of 100 mM trisodium citrate pH 6.0, 200 mM ammonium acetate and 3% v/v MPD and identical crystals grew with the shortened PqsR94-309 construct.
Analytical gel filtration was performed with a Superdex 75 10/300 column equilibrated with running buffer of 20 mM Tris-HCl, 150 mM NaCl, pH 7.4. The standards used were ovalbumin (43 kDa), carbonic anhydrase (29 kDa) and ribonuclease A (13.7 kDa).
Crystals were transferred to a solution with cryoprotectant of 25% MPD and cryo-cooled for collection of diffraction data. The detailed method for PqsRCBD-MPD structure determination is outlined in Text S1. Native and derivative datasets were collected at beamline IO4 of the Diamond synchrotron and data was processed using XDS and reduced with the CCP4 suite (statistics are shown in Table 1 together with the description of the SIRAS structure determination for the PqsRCBD-MPD structure). PqsR ligands were dissolved in 100% MPD or in a 1∶1 mixture of MPD and isopropanol to give a concentration of 20 mM. When added to recombinant PqsRCBD even low concentrations of these compounds resulted in heavy precipitation. The soaking of PqsR94-309 crystals was carried out for 24–48 h with ligands at 5–10 mM. Soaking experiments with HHQ, NHQ, PQS, C9-PQS and shorter chain analogues were carried out and in each case crystals were transferred to a solution with cryoprotectant of 25% MPD and cryo-cooled for collection of diffraction data. Datasets were collected at beamline IO2 of the Diamond synchrotron and data were processed using XDS and reduced with the CCP4 suite (Table 1). Rigid body refinement (REFMAC) was carried out to adjust for small changes in cell dimension and 2Fo-Fc and Fo-Fc electron density maps were calculated using the CCP4 suite. Additional electron density was observed for NHQ and 3-NH2-7Cl-C9-QZN soaked crystals. Model building was carried out using COOT and refinement with REFMAC.
The AQs and QZNs listed in Figures 2 and 6 were synthesised and characterised as described in the supplemental information provided in Text S1.
The impact of the AQs and QZNs on PqsR-dependent gene expression in P. aeruginosa was evaluated using lux-based pqsA, lecA and phzA1 promoter fusions (Table S1) in 96-well microtiter plates as described before [14],[23]. Bioluminescence and bacterial growth were quantified using a combined luminometer-spectrometer (Tecan GENios Pro). Where required, AQs or QZNs were added to reporter strains and EC50 or IC50 values were extracted from the sigmoidal dose–response curves obtained using Prism2 (Graphpad, San Diego, USA).
The impact of 3-NH2-7Cl-C9-QZN on (a) AQ production was assayed by LC MS/MS after extracting bacterial cultures with acidified ethyl acetate [11]; (b) pyocyanin was quantified spectrophotometrically [16] and (c) biofilm development was examined using a Bioflux 200 microfluidics device (Fluxion Biosciences; in conjunction with GFP-labelled P. aeruginosa strains, [53]). All assays were performed in triplicate at least twice.
|
10.1371/journal.pcbi.1002967 | Differential Expression Analysis for Pathways | Life science technologies generate a deluge of data that hold the keys to unlocking the secrets of important biological functions and disease mechanisms. We present DEAP, Differential Expression Analysis for Pathways, which capitalizes on information about biological pathways to identify important regulatory patterns from differential expression data. DEAP makes significant improvements over existing approaches by including information about pathway structure and discovering the most differentially expressed portion of the pathway. On simulated data, DEAP significantly outperformed traditional methods: with high differential expression, DEAP increased power by two orders of magnitude; with very low differential expression, DEAP doubled the power. DEAP performance was illustrated on two different gene and protein expression studies. DEAP discovered fourteen important pathways related to chronic obstructive pulmonary disease and interferon treatment that existing approaches omitted. On the interferon study, DEAP guided focus towards a four protein path within the 26 protein Notch signalling pathway.
| The data deluge represents a growing challenge for life sciences. Within this sea of data surely lie many secrets to understanding important biological and medical systems. To quantify important patterns in this data, we present DEAP (Differential Expression Analysis for Pathways). DEAP amalgamates information about biological pathway structure and differential expression to identify important patterns of regulation. On both simulated and biological data, we show that DEAP is able to identify key mechanisms while making significant improvements over existing methodologies. For example, on the interferon study, DEAP uniquely identified both the interferon gamma signalling pathway and the JAK STAT signalling pathway.
| High throughput technologies, such as next generation sequencing, microarrays, mass spectrometry proteomics, and metabolomics, are capable of evaluating the expression levels of thousands of genes, proteins, or metabolites in an individual run. As a result, the life sciences are experiencing a massive influx of data, exponentially increasing the size of databases [1]–[3]. Currently, databases contain millions of data sets from transcriptomics and thousands of from proteomics [4]–[10]. Differential expression analysis, the comparison of expression across conditions, has become the primary tool for finding biomarkers, drug targets, and candidates for further research. Typically, gene expression data have been analyzed on a gene-by-gene basis, without regard for complex interactions and association mechanisms. Ignoring the underlying biological structure diminishes the power of analysis, obscuring the presence of important biological signals.
Genes and proteins can be grouped into different categories on the basis of many traits: sequence, function, interactions, etc.. Grouping genes by biological pathway is often the most relevant approach to biologists. For this study, we represent biological pathways as directed graphs, where the nodes are biological compounds and the edges represent their regulatory relationships, either catalytic or inhibitory. A catalytic edge exists when expression of the parent node increases expression of the child node (i.e. A3 is a parent to child A4 with a catalytic edge, Figure 1). In an inhibitory relationship, expression of the parent node decreases expression of the child node (i.e. A1 is a parent to child A4 with an inhibitory edge, Figure 1). Further, we define a path as a connected subset of the pathway (i.e. A3A4A7 is a path, A1A2A3 is not, Figure 1). We use the term path to signify either a simple path or a simple cycle, where the term simple implies no repeated nodes.
While biological pathways have long been known, recent experimental data and computational advances have elucidated many previously uncharacterized mechanisms. Repositories contain information about thousands of biological pathways, with each pathway containing up to several hundred proteins [11]–[14]. Identifying the handful of pathways most relevant to a particular data set is an important challenge. The primary assumption of this paper is that biologically relevant pathways are characterized by co-regulated differential expression of their paths.
Currently, the most popular approach to connect expression data to pathways is through gene set analysis. Gene set analysis methods consider sets of genes simultaneously as opposed to the gene-by-gene basis commonly used in differential expression analysis. One of the most prominent set-based methods is Gene Set Enrichment Analysis (GSEA), where the identified genes are ranked based on expression values [15], [16]. Significance of enriched gene sets is determined from a maximum running sum, which is calculated for each gene set by simultaneously walking down the ranked gene list and incrementing or decrementing the score on the basis of set membership. Other approaches calculate set based scores through different metrics and distributions [17]–[21]. Some of these methods compare gene sets relative to others (known as enrichment analysis or competitive approaches) while others compare individual gene sets across conditions without regard for other sets (known as self -contained approaches) [22].
The major limitation of set-based approaches in their application to pathway datasets is that they neglect the graph structure of the pathway. For example, in Figure 1, sporadic patterns of expression in nodes A1..A8 would prevent identification of significant differential expression by set analysis. Considering the additional information contained in the edges, it becomes clear that A3A4A7 represents a path with similar differential expression from reactants to products. Consequently, A3A4A7 represents a differentially expressed path and may possess biological significance, but is unlikely to be identified as such by set based approaches.
We define pathway analysis as any approach which identifies patterns of differential expression in a data set by considering pathway structure. In pathway analysis, researchers are generally interested in identifying pathways associated with a biological condition and determining the components of those pathways that explain the association. Thus, hypothesis testing can be viewed as a two-step procedure: first, test an entire pathway for differential expression; second, identify the path providing the greatest contribution to that differential expression. Recent approaches to pathway analysis test the generic hypothesis of a pathway differential expression without identifying specific paths [23]–[31].
One of the most popular methods for pathway analysis, signalling pathway impact analysis (SPIA, Table 1) combines a set analysis score with a cumulative pathway score [23], [24]. The pathway score is calculated by summing all edges in the graph. Catalytic and inhibitory relationships are considered by using a multiplier on the expression values. While this score takes into consideration the graph structure of pathways, it includes all possible paths, rather than just differentially expressed paths. For example, in Figure 1, the SPIA score would be based on the combination of path scores for A3A4A7, A1A4A7, A2A5A7, A3A6A8, and A3A6A7 and the set score for A1..A8.
Protein interaction permutation analysis, designed for siRNA experiments, calculates the significance of the number of interactions in a network for which both genes are “hits” [25]. Recently, Zhao et al. introduced an approach that includes pathway structure in the analysis of genome wide association studies [26]. However, neither of these methods are directly applicable to expression data. Other pathway analysis approaches calculate set enrichment scores, but weight gene products based on their correlation with neighboring genes in the pathway [27], [28]. Alternatively, other approaches integrate omics data over pathways, but encode all expression data as −1, 0, or 1, limiting the information utilized from experimentation [29]. A mixed linear model presents an advanced approach to the hypothesis test, but is limited to acyclic models and implementation remains complex [30]. Like SPIA, all of these approaches account for the pathway structure as a whole, rather than identifying differentially expressed paths. To our knowledge, popular commercial pathway tools (i.e. Ingenuity Pathway Analysis, BioBase, GeneGo, Metacore, Ariadne) currently offer no methods that directly incorporate pathway analysis.
High-throughput data analysis typically falls into the category of p>>n problems, where the number of genes or proteins, p, is considerably larger than the number of samples, n. Pathway and gene set analysis methods have the added complexity that gene expression within pathways is often highly correlated. Therefore, the statistical analysis approaches described above typically rely on random permutations of biological replicates in order to preserve expression correlation structure. However, the small sample size limits the number of possible permutations and, hence, the precision of p-value estimates. In addition, permutation tests are only applicable to simple experimental designs. Utilizing a random rotation approach circumvents these issues [32]–[34].
In this study, we present a new pathway analysis method, Differential Expression Analysis for Pathways (DEAP). The primary assumption of DEAP is that patterns of differential expression in paths within a pathway are biologically meaningful. DEAP calculates the path within each pathway with the maximum absolute running sum score where catalytic/inhibitory edges are taken as positive/negative summands. To assess the statistical significance, we use a random rotation. Similar to other pathway analysis methods, DEAP tests a generic hypothesis of overall pathway differential expression. Contrary to current methods, DEAP identifies the most differentially expressed path to provide a refined focus for further biological exploration.
As illustrated in Figure 2, the DEAP algorithm begins by overlaying expression data onto the pathway graph (Figure 2.1). Every possible path from the graph is independently examined (Figure 2.2). A recursive function calculates the differential expression for each path by adding or subtracting all downstream nodes with catalytic or inhibitory relationships, respectively (Figure 2.3).
As an example, the score for the path containing all nodes in the inhibitory string in Figure 3 (left), where green = +1 and red = −1, is calculated as:(1)The absolute value of the expression level is utilized as the DEAP score (Figure 2.4) to determine the path with maximal differential expression (Figure 2.5). The DEAP algorithm returns both the maximum absolute value and the path associated with that maximum value. The algorithm is formalized in Methods: DEAP Algorithm.
DEAP scores for different pathways are not directly comparable due to size and structure differences among pathways. Thus, we employ a self-contained approach which individually assesses the significance of each pathway. Generating a null distribution is complicated by the low number of samples relative to gene identifications and the correlation of gene expression within pathways. Most existing approaches use permutation tests to preserve the correlation between genes; however, small sample size limits their effectiveness. We use random rotation to circumvent these issues [32]–[34]. Our random rotation implementation is applicable to a wide range of complex experimental designs with multiple conditions and replicates. The significance levels are adjusted for multiple comparisons using the false discovery rate method of Storey and Tibshirani [35].
For each pathway in the analysis, DEAP outputs its score, the corresponding p-value, and the path with the maximum absolute score (see examples in Files S1, S2). The open source implementation (licensed under the GNU Lesser General Public License v3.0) of this algorithm is available in Supplemental Materials (File S3).
Data from the five pathways illustrated in Figure 3 were simulated as described in Methods. Algorithmic performance was measured in terms of power, the percentage of times each differentially expressed pathway was identified as significant (p<0.05), which is equivalent to one minus the type II error rate. The power of DEAP was compared to GSEA and SPIA, the two most popular gene set and pathway analysis methods, respectively. Comparative analysis of these methods included four key parameters: the overall effect (mean of ‘on’ genes, μ), variation in individual gene effects (σ2g), sample size (n), and type I error rate.
Regardless of the level of differential expression, DEAP was consistently more powerful than were other approaches (Figure 4). For small μ values (low differential expression), the power of DEAP was approximately twice that of GSEA and SPIA, demonstrating improved sensitivity. For μ = 1 (high differential expression), DEAP had an increase in power over both GSEA and SPIA of two orders of magnitude. At μ = 1.25, the performance of SPIA improved substantially, approaching that of DEAP on all pathways except the long alternate route where SPIA was confounded by noise (Figure S1). Across the board, GSEA performed poorly because GSEA did not consider pathway structure and is dependent on comparisons to other pathways.
Sample size and within-gene variance also have significant effects on the performance of the algorithms. As sample size (n) grew, the power of DEAP relative to other approaches increased, particularly in pathways containing inhibitory edges (Figure 5). As variance (σ2g) increased, DEAP exhibited minor increases in power (Figure 6). Further, DEAP consistently outperformed GSEA and SPIA as variance increased.
To estimate the type I error rate, we simulated random data under the null hypothesis (μ = 0, σ2g = 0, n = 10). The plots in Figure 7 displays type I error rates with respect to the nominal values. SPIA was notably more conservative for every pathway structure. The performance of both GSEA and DEAP was on target; however, DEAP was more conservative on pathways with inhibitory edges (Figure S4).
An additional advantage of DEAP is the ability to identify the maximally differentially expressed path of the pathway. For the simulated data with μ = 1 and μ = 2, DEAP identified the entire differentially expressed path 99% and 100% of the time, respectively. For example, the long alternate route contains 14 proteins, but DEAP identified the differentially expressed region that contains only four, substantially reducing the search space.
In addition to comparing DEAP to GSEA and SPIA, we compared DEAP to several modifications of the DEAP algorithm, which were altered as follows: scores normalized by pathway length; all weights set to +1; and sum taken across the entire pathway. We also compared DEAP to a set-based implementation with rotation. DEAP had substantially higher power than all four approaches (Table S1 and Figures S1, S2, S3, S4).
While simulated pathways provide easily controllable examples to validate DEAP as an appropriate test of the hypothesis, biological pathways bring increased complexity from which the signal must be detected. To validate DEAP on more realistic pathway structures, we simulated activity on biological pathways from the KEGG and Reactome databases [13], [14].
In the case of KEGG [13], we simulated data on the TGF-ß signaling pathway to indicate activity in the TGF- ß receptors leading to cell cycle arrest (Figure 8). In terms of sensitivity to the pathway effect (μ), variance (σ2g), and sample size, DEAP outperformed both GSEA and SPIA on the TGF-ß signaling pathway (Figure 8). Notably, increased variance diminishes the power of SPIA, but does not affect DEAP, reflecting its ability to identify signal in the noisy environments common in biological experimentation.
In the case of Reactome [14], we simulated data on the post-transcriptional silencing by small RNAs pathway from to indicate RNA cleavage (Figure 9). DEAP had superior performance over GSEA and SPIA in terms of all tested variables: pathway effect (μ), variance (σ2g), and sample size (Figure 9).
In both sets of simulated data on real biological pathways, the type I error estimate was conservative for DEAP, GSEA, and SPIA (Figure S5). In addition to DEAP, GSEA, and SPIA, we applied the four alternative formulations of DEAP to both sets of biological pathways and noted the consistently strong performance of DEAP (Figures S6, S7).
To verify that the simulated data effects are biologically relevant, we also applied DEAP to two sets of biological data on biological pathways. The experimental data are from a transcriptomic study of interferon [36], [37] and a proteomic study of chronic obstructive pulmonary disease (COPD). We applied DEAP, GSEA, and SPIA to identify differentially expressed pathways from the PANTHER database [11]. Pathway associations with the phenotypes were determined based on a literature review using Google Scholar (details in Methods: Biological Data Validation).
We analyzed a microarray expression data of cells of radio-insensitive tumors that had been treated with interferon [36], [37]. DEAP identified six pathways with known literature associations with interferon while GSEA identified five and SPIA identified none (Table 2). The two most clearly relevant pathways for this transcriptomics data set were interferon gamma signalling, as the cells had been stimulated with interferon; and JAK STAT signalling, the pathway being studied by the authors of the microarray study [36], [37]. Unlike GSEA and SPIA, DEAP identified these pathways as significantly differentially expressed. The lack of overlap between the pathways identified by GSEA and DEAP is indicative of the different hypotheses being tested by these two approaches, with GSEA focusing on non-specific differential expression among pathway genes and DEAP focusing on differential expression among pathway connected genes. As such, these two approaches should be viewed as complementary approaches that can be simultaneously utilized to augment biological discovery.
Additionally, DEAP analysis of the interferon transcriptomics data uses path identification to reduce the search space for future experimentation. Consider the Notch signalling pathway, which contains 26 proteins and is known to be activated by interferon treatment [38]. GSEA and SPIA both did not identify Notch signalling as significantly differentially expressed due to generally sporadic expression patterns. However, DEAP analysis focused on consistent differential expression of 4 connected nodes and labelled Notch signalling as significantly differentially expressed (Figure 10). Without identifying the maximally differentially expressed path, the Notch signalling pathway would have been overlooked. Further, future experimentation can now focus on those four proteins exhibiting the most significant differential expression.
In order to illustrate DEAP on a different data type, we also analyzed a proteomics study which compared healthy smokers with patients diagnosed with COPD (Methods: Biological data, Table 3). On this data set, GSEA identified nine pathways, four of which had apparent associations with COPD. SPIA identified only one pathway with significant differential expression. DEAP identified 12 pathways and eight had literature-verified implications with COPD. Of notable clinical relevance to COPD is the inflammation mediated by chemokine and cytokine signalling pathway, which was identified only by DEAP [39].
DEAP takes into account the graph structure of a pathway and determines the maximally expressed path. Pathway-centric analysis by DEAP is complementary to set-based analysis of other functional categories, as seen in both biological examples (Tables 2–3). Application of the random rotation approach allows for accurate assessment of statistical significance of the DEAP scores. On simulated data for simulated pathways, DEAP both increased power over existing approaches and accurately controlled the false positive rate. With high differential expression, this translated to a two-fold increase in the power of DEAP over GSEA and SPIA. On simulated data applied to real biological pathways, DEAP showed the strongest performance for all levels of pathway effect, variance, and sample size. Analysis of experimental transcriptomic and proteomic data indicates that DEAP identified important pathways related to a particular disease or condition where other approaches failed, specifically identifying six pathways related to interferon and eight related to COPD. Further, DEAP uniquely identified the most expressed path of the pathway with 100% accuracy in simulated data.
Though we demonstrated DEAP on transcriptomics and proteomics studies, DEAP is widely applicable to other omics research areas (metabolomics, lipidomics, etc.) and expression technologies (next generation sequencing, RNAseq, etc.). This broad applicability extends from the flexible design of DEAP: the only required inputs are expression levels of biomolecules and corresponding pathways. Appropriate scaling of the expression levels is defined by the user. For instance, RNAseq data is very similar to spectral count proteomics data in that they are both count-based. Thus, RNAseq read counts can be used as input for DEAP in the same manner as peptide spectral counts. Further, RNA transcripts can be used in place of proteins.
To identify the most important pathways for further study, pathways can be ranked based on DEAP score significance. Specifically, future studies can be focused on the most differentially expressed paths within the pathways with the lowest false discovery rate, which can be especially beneficial when studying pathways that contain hundreds of biological compounds. Currently, DEAP is being integrated with our proteomics analysis pipeline SPIRE (http://proteinspire.org) and expression database MOPED (http://moped.proteinspire.org) [10], [40] (Table 1). Application of DEAP to existing and future studies has the potential to discover meaningful biological patterns.
Expression data (presumably on a log scale) for each gene in a pathway was simulated using a multivariate normal distribution defined in Equation 2:(2)In this equation d is the indicator of whether a gene is ‘on’ or ‘off’. The value of d is 0 if the gene is ‘off’ and +1 if the gene is up-regulated and ‘on’ and −1 if the gene is down-regulated and ‘on’. The value of d is determined by the predefined pathways. The variable μ is the mean of the absolute value of expression for ‘on’ genes and, therefore, represents the ‘pathway effect’. The value of μ is held constant for each gene in the pathway and across replicate samples. The variable g is assumed to come from a normal distribution with mean 0 and variance σ2g. The variance σ2g measures how much individual gene expression deviates from the overall ‘pathway effect’, μ.
The value of g is randomly generated (although in many of the simulations is set to 0) for each gene in the pathway, but the same value is used for replicate samples. The variable e is assumed to come from a normal distribution with mean 0 and variance 1 and represent random variation in gene expression. The value of e is randomly generated for each combination of gene and sample. The simulations varied the values μ, σ2g, and the sample size (number of independent samples of pathway data). R scripts were used to generate the simulated data [41].
Five diverse pathways were specifically created to test the efficacy of identification by different scoring methods (Figure 3). Gray colored nodes had unaltered values from a standard normal distribution. Nodes labelled as green and red were sampled with μ values of +X and −X, respectively, where X was a positive number. Simulated data and pathways are available on Dryad: doi:10.5061/dryad.qh1pg.
Microarray data from a study of cells treated with interferon were acquired from the Gene Expression Omnibus (GDS3126) [6]. The sample was taken from radio-resistant tumors following treatment with a mixture of interferons [36], [37]. It was hypothesized that interferon and biochemically-related pathways would be stimulated in this data set. The expression value was the logarithm of the case/control ratio. Though microarrays measure mRNA expression, the pathways represent information in terms of proteins. Therefore, the gene identifiers in the microarray data were mapped to UniProt protein identifiers using the UniProt website [42]. Handling the one-to-many relationship of genes and proteins is discussed below (see Methods: DEAP). When duplicate probes existed for the same gene, the expression value utilized for the gene was the arithmetic mean of these probes.
The COPD proteomics data can be found at PeptideAtlas (raw data) [7] and MOPED (processed data) [10] (moped.proteinspire.org). We analyzed data from CD4 and CD8 T-lymphocytes. The control patients were healthy smokers, with an average FEV1/FVC of 82.5%. Case patients had been medically diagnosed with COPD and had an average FEV1/FVC of 42.0%. A total of 10 cases and 10 controls were utilized in this analysis. Additional experimental details can be found associated with the PeptideAtlas accession numbers in Table S3. On MOPED, data is stored under the experimental name “steffan_copd.” The tandem mass spectrometry data were analyzed through SPIRE with the parameters in Table S4 [40]. Protein expression was measured by the number of peptide spectral matches identified for each protein normalized by the total number of spectra in the sample. For pathway analysis, we used the difference between the log normalized expression values.
Pathway data were downloaded from the PANTHER database [11]. A total of 165 pathways downloaded in SBML format from PANTHER pathway version 3.01. PANTHER pathways contain information about proteins, biochemicals, and other substrates. For the purposes of data interpretation, the pathways were broken into their protein components using an internally developed python script where connections of proteins through biochemical substrates were maintained as protein-protein interactions PANTHER's internal identifiers were mapped to UniProt identifiers. Ultimately, parsing of the PANTHER pathway database resulted in a graph structure in which each node represented a set of proteins that act as a set of reactants and/or products. Inhibitory or catalytic edges between two sets of proteins were determined as detailed in PANTHER.
We used random rotation approach to estimate the null distribution of the test statistics and compute the p-values [32]. Rotation testing has been used recently in gene set analysis as an alternative to permutation and parametric tests [33], [34]. Rotation tests have an advantage over permutation tests in that they produce reasonable results for small sample sizes and complex experimental designs. Rotation testing assumes that pathway and set data come from independent random samples of a multivariate normal distribution with mean zero under the null hypothesis. A rotation test is carried out by multiplying the original data by a random rotation matrix, calculating the test statistic, and repeating the procedure to generate a null distribution. Adjustments for an overall mean, covariates, or blocking factors are handled by performing the rotations of an orthogonal projection of the original data on to the residual space from a linear model and then transforming the rotated data back. A random rotation matrix was generated by first generating a matrix X of standard normal random variables and then taking the rotation matrix to be the orthogonal matrix Q from the QR decomposition of X. Scripts to carry out rotation testing were written using the R programming language and are available in File S3, released under the GNU Lesser General Public License v3.0. The user is able to input a custom design matrix which accounts for complex experimental designs with multiple conditions and replicates.
Given: a current edge, all other edges in graph, expression values for all proteins:
For single channel (unpaired) data, define E(x) to be the difference between the logarithm of the arithmetic mean of expression values associated with protein x in the two conditions.
For two channel (paired) data, define E(x) to be the arithmetic mean of the log expression ratio(s) associated with protein x.
The recursive function operates as follows:
In DEAP, the maximum order (by absolute value) path is used to test the null hypothesis about the expression of the entire pathway. This claim, that the expression of one path answers questions about the expression of the pathway, is justified on two levels.
On a biological level, significant fluctuations in activity do not require differential expression of an entire pathway. For example, in Figure 1, A3A4A7 represents a path with similar expression levels that proceeds all the way from reactants to products, a pattern that seems to be significant.
From a logical perspective, consider a pathway, P, as the union of all paths of the pathway, P1, P2, …, PK. Each path is completely defined by its set of edges. Note that the k-paths are not entirely disjoint in the sense that some paths might share the nodes and the edges. However, we require each path to have a distinct set of edges. To test the hypothesis of a differentially expressed pathway requires testing whether any of the constituent paths is differentially expressed. This corresponds to testing the family of k-null hypothesis. To control the family wise error rate, we use a maximum order statistic, since the probability of making at least one incorrect decision under the null is equivalent to the probability of the maximum order statistic exceeding the threshold.
To approximate a null distribution of the test statistic, s*, we performed n rotations of the data. For each rotation sample, we recompute the DEAP score, si. The p-value is calculated as a proportion of scores that are at least as extreme as the observed score, the proportion of simulated DEAP scores whose value are greater than or equal to the observed DEAP score:
The DEAP algorithm was implemented to allow for efficient computation.
By maintaining global maximum and minimum values and updating their values as the recursive function proceeds, it is not necessary to examine all paths of the graph independently. Rather, we can initialize DEAP score calculations only at leaf edges, which have no upstream edges pointing to any proteins in their reactant set. To ensure that closed cycles are not missed, we track the edges which have been visited and examine additional edges until the difference of the complete edge set and the already visited edge set is empty. This greatly reduces the number of calculations per graph.
Once the recursive function has returned a maximum and minimum score for a particular edge, that score will remain constant regardless of the preceding edge except in the case of cycles (see paragraph below). Therefore, we use a dictionary mapping edges to maximum and minimum scores to prevent duplicative score calculations. After this implementation, score calculations that took several hours on particularly complex pathway structures completed in seconds.
In the case of cycles, scores may be dependent on the node of the cycle which is examined first. For these cycles, our current implementation represents a heuristic estimator rather than the exact optimal solution. Bidirectional edges are subject to this same limitation as they are equivalent to a two node cycle. Implementations that determined the exact optimal solution were prohibitively slow for practical application. Except in edge cases, the heuristic implementation will provide approximations of sufficient quality to identify significant patterns of differential expression.
Every DEAP score calculation is independent of other DEAP score calculations, so we set up processing for multi-threading. For example, on a 64-bit Intel Core i7-2720QM CPU with 8GB RAM, speed improvements of approximately 4-fold were noted for the score calculation process. Specific running time is highly dependent on expression data set size, experimental design, pathway complexity, and number of rotation testing iterations. Running DEAP on 90 simulated data files each with 10 samples, 1000 proteins, 1000 pathways, and performing 100 data rotations took 72 minutes when multi-threaded and 260 minutes when performed on a single thread.
The function tracks edges that have already been examined in a particular recursive cycle to prevent entrance into infinite loops in cyclical pathways. To control for duplicate protein identifiers, summations over the products and reactants were performed on the set of unique expression values rather than for every identifier. For example, if protein A and protein B both had expression levels of 1.743 and were both in the same protein set, then it was assumed they were the result of data duplication and 1.743 was only added to the score once. This duplication elimination was implemented primarily due to issues arising from redundant protein identifiers and potential mRNA translation into multiple proteins. For instance, the five UniProt identifiers for variants of Histone H3 (Q6NXT2, P68431, Q16695, Q71DI3, and P84243) are included in the same PANTHER pathway unit and share near identical protein sequences, so their proteomic and transcriptomic identification will be duplicated.
The algorithm was implemented in Python and is available in File S3, released under the GNU Lesser General Public License v3.0.
Accuracy of pathway associations with experimental conditions were validated using a Google Scholar literature search. The literature search was performed by searching Google Scholar (http://scholar.google.com) for a combination of the pathway name and details of the experimental condition. We continued searching Google Scholar until satisfied that the association was confirmed or felt reasonably certain that there was not yet a literature confirmed association. Once a literature association was confirmed, the most pertinent reference was retained and cited in this manuscript.
The DEAP approach is based on the following fundamental assumptions:
The GSEAlm package for the R Project, available through BioConductor, was utilized to perform GSEA analysis [44]. Pathways were transformed into a gene set matrix and multi-sample expression data were loaded appropriately. Since GSEA performs test for up- and down-regulation independently, the minimum of these two values was taken and multiplied by two to adjust for a two-tail test.
SPIA analysis was performed using the SPIA package for the R Project, available through BioConductor [45]. To convert the pathways into the SPIA format, inhibitory and catalytic relationships were formatted into the inhibition and activation matrices, respectively. Since the SPIA implementation only allowed input of single expression ratios, the arithmetic mean of expression values for each protein was input into SPIA.
|
10.1371/journal.pntd.0002614 | Favipiravir (T-705) Inhibits Junín Virus Infection and Reduces Mortality in a Guinea Pig Model of Argentine Hemorrhagic Fever | Junín virus (JUNV), the etiologic agent of Argentine hemorrhagic fever (AHF), is classified by the NIAID and CDC as a Category A priority pathogen. Presently, antiviral therapy for AHF is limited to immune plasma, which is readily available only in the endemic regions of Argentina. T-705 (favipiravir) is a broadly active small molecule RNA-dependent RNA polymerase inhibitor presently in clinical evaluation for the treatment of influenza. We have previously reported on the in vitro activity of favipiravir against several strains of JUNV and other pathogenic New World arenaviruses.
To evaluate the efficacy of favipiravir in vivo, guinea pigs were challenged with the pathogenic Romero strain of JUNV, and then treated twice daily for two weeks with oral or intraperitoneal (i.p.) favipiravir (300 mg/kg/day) starting 1–2 days post-infection. Although only 20% of animals treated orally with favipiravir survived the lethal challenge dose, those that succumbed survived considerably longer than guinea pigs treated with placebo. Consistent with pharmacokinetic analysis that showed greater plasma levels of favipiravir in animals dosed by i.p. injection, i.p. treatment resulted in a substantially higher level of protection (78% survival). Survival in guinea pigs treated with ribavirin was in the range of 33–40%. Favipiravir treatment resulted in undetectable levels of serum and tissue viral titers and prevented the prominent thrombocytopenia and leucopenia observed in placebo-treated animals during the acute phase of infection.
The remarkable protection afforded by i.p. favipiravir intervention beginning 2 days after challenge is the highest ever reported for a small molecule antiviral in the difficult to treat guinea pig JUNV challenge model. These findings support the continued development of favipiravir as a promising antiviral against JUNV and other related arenaviruses.
| Argentine hemorrhagic fever (AHF) is a severe and often-fatal disease caused by infection with Junín virus (JUNV). Presently, there is an unmet need to develop new therapeutics to address current medical, public health and national security concerns, as JUNV is considered a potential bioterror agent amenable to aerosolization and intentional release. In the present study, favirpiravir, a promising anti-JUNV drug in clinical development for the treatment of influenza, was evaluated in an experimental small animal model of AHF. Guinea pigs challenged with JUNV were treated with favipiravir twice daily for two weeks starting 1–2 days after infection. Consistent with pharmacokinetic analysis that showed greater plasma levels of favipiravir in animals dosed by intraperitoneal injection, administration by this route resulted in a dramatic protective effect as 78% animals survived the infection compared to 11% in the placebo-treated group. Favipiravir treatment inhibited JUNV replication and prevented the development of disease observed in animals receiving placebo during the acute stage of infection. The high level efficacy observed following post-exposure prophylaxis with favipiravir is the highest ever reported for a small molecule antiviral in the guinea pig JUNV challenge model and thus supports its continued development as a promising antiviral therapy for the treatment of AHF.
| Several New World (Junín, Machupo, Guanarito, Sabia, and Chapare) and Old World (Lassa and Lujo) arenaviruses cause viral hemorrhagic fever (HF) with case fatality rates generally in the range of 15–30% [1]. They are rodent-borne viruses that can be transmitted via the aerosol route therefore making them potential bioterror agents. Exposure to Lassa virus (LASV) and mortality associated with Lassa fever (LF) in hyperendemic areas of West Africa are estimated to be as high as 300,000 infections and 10,000 deaths annually [2]. Of the New World arenaviral HFs endemic in different regions of the South America, Junín virus (JUNV), the etiologic agent of Argentine HF (AHF), causes the greatest morbidity and mortality. AHF cases, although reduced in number, continue to be reported despite the vaccination of individuals with the greatest risk of exposure [3].
Immune plasma has proven to be an effective treatment for AHF if administered within 8 days of initial disease symptoms, but has been associated with a late neurological syndrome and is not readily available outside of Argentina [4]. Ribavirin is the only licensed antiviral known to offer protection in cases of LF [5], but remains largely unproven with only limited data in cases of AHF and Bolivian HF due to Machupo virus infection [6], [7]. Adverse effects primarily in the form of hemolytic anemia are generally considered to be reversible with cessation of ribavirin treatment [8]–[10]; however, teratogenicity and embryotoxicity are of concern [11], [12]. One must also consider the added risk of using a drug that can cause anemia to treat arenaviral HF diseases, which have a propensity for bleeding. At present, there are very few promising antivirals which have demonstrated anti-arenavirus activity in vivo [13].
Favipiravir (T-705; 6-fluoro-3-hydroxy-2-pyrazinecarboxamide) is a novel antiviral compound developed by the Toyama Chemical Co., which selectively and potently inhibits the RNA-dependent RNA polymerase (RdRP) of influenza [14]. It has been found to inhibit all serotypes and strains of influenza A, B and C viruses against which it has been tested [15], [16], including those resistant to currently approved neuraminidase inhibitors [17]. Remarkably, it is also active against alpha-, arena-, bunya-, and flaviviruses, both in cell culture and rodent models [16], [18]–[20], and it has shown in vitro activity against members of the paramyxo-, picorna-, and calicivirus families [21], [22].
To date, oral favipiravir treatment has been shown to be active in lethal hamster and guinea pig arenaviral HF models based on challenge with Pichindé virus (PICV), a New World arenavirus that produces in these species many of the clinical disease manifestations associated with AHF, including vascular leak and thrombocytopenia [23]–[25]. Importantly, favipiravir was efficacious in treating advanced disease even when initiating treatment one week after challenge when clinical disease was clearly apparent in guinea pigs [25]. In vitro studies have demonstrated favipiravir activity against pathogenic strains of JUNV, Machupo, and Guanarito viruses [19]; however, its activity against a bona-fide HF arenavirus in vivo has not been explored until now. In the present study, the antiviral activity of favipiravir against JUNV infection modeled in guinea pigs was investigated.
All animal procedures complied with USDA guidelines and were conducted at the AAALAC-accredited Robert E. Shope, M.D. Laboratory at The University of Texas Medical Branch (UTMB; Galveston, TX) under protocol # 0903023 approved by the UTMB Institutional Animal Care and Use Committee.
Outbred male Hartley strain guinea pigs (300–350 g) were obtained from Charles River (Wilmington, MA) and acclimated for 1 week prior to challenge. Animals were sorted prior to the start of both experiments so that the average group weights were similar. IPTT-300 electronic transponders were subcutaneously implanted for identification and temperature measurement in conjunction with the DAS 6002 scanner (BMDS, Seaford, DE).
The Romero strain of JUNV was kindly provided by Thomas Ksiazek (UTMB). The virus stock was grown in Vero cells and the titer determined by plaque assay. The virus was prepared in minimal essential medium (MEM) for intraperitoneal (i.p.) challenge with approximately 1000 plaque-forming units (PFU). The actual challenge dose for each experiment was determined by plaque titration of the inoculation medium. All live virus work was performed in BSL-4 containment at UTMB in accordance with institutional health and safety standard operating procedures.
Favipiravir was provided by the Toyama Chemical Company, Ltd. (Toyama, Japan). Ribavirin was supplied by ICN Pharmaceuticals, Inc. (Costa Mesa, CA). Compounds were suspended in GERBER NatureSelect 1st FOODS carrot food (ingredients: carrots and water) for oral administration in the first efficacy study and 2.9% sodium bicarbonate solution (Sigma-Aldrich, St. Louis, MO) for i.p. dosing in the second study.
For virus titration of organs and serum collected from infected guinea pigs, approximately 0.2–0.5 g of each organ was homogenized in 0.5 ml PBS. Serum was separated from whole blood by centrifugation. The homogenates and sera were held at −80°C until titration. The homogenates were centrifuged to remove cellular debris and the cleared homogenate for each organ and the serum were titrated by a focus-forming unit (FFU) assay as follows.
Vero E6 monolayers were infected with serial 10-fold dilutions of serum or tissue homogenate for one hour at 37°C. For the second experiment, serum was also tested undiluted. Following infection, cells were overlayed with 0.8% Tragacanth (Sigma-Aldrich) in MEM supplemented with 2% fetal bovine serum and 1% penicillin and streptomycin. After 7 days in culture, the overlay was removed, the cells fixed with 10% buffered formalin for 30 min at room temperature followed by overnight refrigeration. Fixed cells were permeabilized in 70% ethanol for 20 min and washed with phosphate buffered saline (PBS) prior to overnight staining with primary antibody (antisera to JUNV Candid #1 kindly provided by Dr. Robert Tesh, World Reference Collection for Emerging Viruses and Arboviruses, UTMB) diluted 1∶1000 in PBS with 5% milk and 1% Tween 20. The cells were then washed with PBS and the secondary antibody, goat anti-mouse IgG labeled with horseradish peroxidase (HRP; DakoCytomation, Carpinteria, CA) diluted 1∶500 in PBS with 1% bovine growth serum (BGS), was added to plates and incubated for 4–5 h. After washing with PBS, AEC Substrate Chromagen (DakoCytomation) was added for 15 min and the reaction was stopped with distilled water prior to counting of FFUs.
Heat-inactivated guinea pig sera from survivors were diluted 1∶10 in MEM supplemented with 1% FBS, and titrated in two-fold serial dilution steps. Equal volumes (150 µl) of JUNV Romero strain containing approximately 1000 PFU/mL and serum dilutions were mixed and incubated for 1 h at 37°C and 5% CO2. Confluent monolayers of Vero E6 cells (seeded in 12-well plates) were infected with 100 µl of the virus–serum mixtures. After 1 h incubation at 37°C and 5% CO2, the wells were overlaid with 0.5% agarose MEM with 1% FBS. The plates were incubated at 37°C and 5% CO2 for 7 days, and then stained with 0.25% crystal violet in 10% buffered formalin. The plates were washed and the plaques enumerated. The neutralizing antibody titer of a serum was considered positive at the highest initial serum dilution that resulted in >80% (PRNT80) reduction of the number of plaques as compared to guinea pig serum from the mock-infection.
All cell counts were quantified using a Hemavet Mascot hematology analyzer (Drew Scientific, Dallas, TX) equipped with veterinary software to measure white blood cell count, red blood cell count, hemoglobin concentration, hematocrit, mean corpuscular volume, mean corpuscular hemoglobin, mean corpuscular hemoglobin concentration, neutrophils, lymphocytes, monocytes, eosinophils, basophils, leucocytes, reticulocytes, and platelet count. Blood chemistry was performed using a VetScan2 Chemistry Analyzer (Abaxis, Inc., Sunnyvale, CA), which provides a diagnostic panel that includes albumin, alkaline phosphatase, alanine aminotransferase, amylase, total bilirubin, blood urea nitrogen, calcium, creatinine, glucose, potassium, total protein and globulin.
Guinea pigs were treated with 100 mg/kg of favipiravir administered by placement of drug prepared in carrot baby food vehicle in the back of the oral cavity with a tuberculin syringe or by i.p. injection of drug prepared in a 2.9% sodium bicarbonate solution. Plasma was obtained from 3 animals per group by saphenous vein puncture at 0.25, 0.5, 1, 2 or 4 h after treatment. Samples were deproteinized and analyzed by HPLC for quantitation of favipiravir as previously described [25].
The Mantel-Cox log-rank test was performed to analyze Kaplan-Meier survival plots. Hematology and blood chemistry were analyzed by the Student's two-tailed t-test. Virus titers were analyzed using one-way analysis of variance (ANOVA) followed by Bonferroni multiple comparisons test. All statistical evaluations were done using Prism (GraphPad Software).
Based on studies modeling oral favipiravir therapy in PICV-infected guinea pigs [25], a dose of 300 mg/kg/day was selected for the initial efficacy trial in the JUNV infection model. Animals were dosed twice daily by oral installation for a duration of 14 days, starting 1 day after challenge with 1300 PFU of the Romero strain of JUNV. As shown in Figure 1A, favipiravir treatment improved survival outcome compared to guinea pigs treated with placebo. Two of the 10 guinea pigs treated with favipiravir survived the infection, and those that succumbed survived, on average, >4 days longer than the animals treated with placebo (20.3±2.6 days and 16.2±1.2 days, respectively).
Weight change over the course of the study was used to assess the effect of favipiravir treatment on the condition of the animals. The mean weights of the placebo-treated guinea pigs began to drop on day 10 post-infection, and were below the initial starting weights by day 12 (Figure 1B). In contrast, mean weights in animals treated with favipiravir did not fall below initial starting weights until day 18. Favipiravir treatment also delayed the onset of fever (defined as a body temperature of ≥39.8°C) from 8.8±2.3 days in the placebo group to 16.5±3.3 days. Ribavirin, included as a positive control [26], performed similarly to favipiravir as the survival, weight, and temperature curves did not differ significantly between the two drug treatment groups.
Blood samples were collected from all animals on day 9 post-infection for analysis of virus titers. The early time point was selected because of concerns that the deep anesthesia required for blood collection from the cranial vena cava may have resulted in the loss of sick animals if the procedure would have been delayed beyond day 9. Unfortunately, viral titers in the placebo-treated animals were not well developed, as only 3 of 10 animals had detectable levels of virus in the range of 200–1300 FFU/ml. Nevertheless, virus could not be detected in favipiravir- or ribavirin-treated guinea pigs. Viral RNA was not detected in the serum or brain, liver or spleen tissues collected from the surviving favipiravir- and ribavirin-treated animals at the conclusion of the study (data not shown). Moreover all survivors appeared to be in good health by physical examination and all the aforementioned tissues were histologically normal.
Despite proven effectiveness in the PICV guinea pig infection model by oral treatment with favipiravir suspended in carrot baby food vehicle [25], the lower than expected efficacy observed in the first study prompted a PK analysis to compare plasma levels of favipiravir following administration by i.p. and oral treatments. Compared to oral instillation of the compound, i.p. injection of an equivalent dose of 100 mg/kg resulted in higher plasma concentrations of favipiravir within the first hour, and similar levels at the later time points (Figure 2). The area under the curves (AUC)s of the drug concentrations were 56.1 by the oral instillation method and 103.4 by i.p. injection. Peak favipiravir plasma concentrations approximate 80 µg/ml (500 µM) for i.p. treatment, with >10 µg/ml (60 µM) present at 2 h. These levels are well above the reported 50% effective concentration (EC50) of <20 µM for JUNV in cell culture [19]. Thus, the single dose PK data suggest that favipiravir is more bioavailable through i.p. dosing.
A second experiment was conducted wherein favipiravir was dosed by i.p. injection starting 48 h post-challenge with 750 PFU of JUNV. Favipiravir treatment started 2 days post-infection provided a highly significant level of protection (7/9 survivors; 78%) compared to the placebo group (1/9 survivors; 11%), and performed better than ribavirin (3/9 survivors; 33%) (Figure 3A). The weight data were consistent with the survival curves with the favipiravir-treated animals mirroring the weight gain of the sham-infected animals through day 14 before leveling out temporarily as a few animals became ill (Figure 3B). Guinea pigs in the placebo group began to develop fevers as early as day 6 post-infection, while temperatures were in the normal range in the drug-treated animals through the first two weeks (Figure 3C). Notably, the surviving placebo-treated animal had undetectable (PRNT80<20) levels of JUNV neutralizing antibodies at the conclusion of the 42-day study. In contrast, all guinea pigs treated with favipiravir or ribavirin developed substantial neutralizing antibody titers of 320. The placebo group survivor also gained weight and maintained normal body temperature throughout the entire experiment suggesting that the virus challenge did not produce the desired infection.
In a subset of animals sacrificed on day 14 of infection, favipiravir treatment prevented the thrombocytopenia and leucopenia commonly associated with severe disease and also maintained a number of hematologic and blood chemistry parameters at normal baseline levels (Table 1). Moreover, we were unable to detect virus in the serum, brain, heart, kidney, liver, lung, and spleen of favipiravir-treated animals (Figure 4). With the exception of a single animal that had >5 logs of virus per gram of spleen, guinea pigs treated with ribavirin were also free of virus. All surviving guinea pigs were rapidly gaining weight and observed to be in good health by physical examination at the termination of the study. Taken together, the results demonstrate robust inhibition of viral replication and a high level of protection by i.p favipiravir treatment in the difficult to treat JUNV guinea pig infection model.
In the present study, the efficacy of oral and i.p. favipiravir treatment was evaluated in the guinea pig JUNV challenge model. Based on previous success in treating PICV infection in guinea pigs [25], favipiravir was initially dosed by orally feeding the animals the drug suspended in carrot baby food. The results from this trial did demonstrate a significant protective effect that was comparable to ribavirin, the only small molecule antiviral that has demonstrated activity against severe JUNV infection, which initially showed promise over 25 years ago [27], [28]. Ribavirin has been evaluated in a small-scale clinical trial in patients with advanced cases of Argentine HF, and did produce some signs of efficacy [3]; however, because it is associated with toxicity primarily in the form of hemolytic anemia, safer and more effective options are needed. The results from the initial guinea pig study were encouraging and supported further consideration of favipiravir through a second experiment designed to improve upon the limited success of the first study.
The fact that oral favipiravir provided complete protection against lethal PICV infection in guinea pigs underscores the greater challenge of treating the more virulent JUNV infection, which was also less responsive to ribavirin in the present and past studies [25]–[27]. It is likely that the higher plasma concentrations achieved when administering favipiravir by the i.p. route resulted in greater levels of drug in target organs, which may have contributed to the remarkable efficacy observed in the second experiment. Because the animals generally did not like the taste of favipiravir suspended in the baby food vehicle, it was difficult to accurately deliver the doses in BSL-4 containment. This may have led to diminished amounts of drug actually making it into the gut for subsequent absorption into the circulation.
It is likely that higher doses of the well-tolerated oral favipiravir [25] would improve survival outcome in guinea pigs challenged with JUNV; however, the increased viscosity of the favipiravir suspension at higher drug concentrations results in a paste-like consistency that is increasingly difficult to administer by mouth. It is also possible that delaying the initiation of treatment from 24 h in the first study to 48 h for the second study may have elicited a better immune response to the additional 24 h of viral replication, which combined with the inhibitory effects of favipiravir may have facilitated the clearance of the virus and afforded the greater level of protection observed. Notably, the clinical laboratory findings and lack of virus on day 14 correlated well with the survival data.
The principal mechanism of action of favipiravir against influenza A virus was shown to be direct inhibition of the viral polymerase [14]. Although direct evidence to support this claim is lacking for other viruses sensitive to the action of favipiravir, findings from arenavirus and norovirus studies are consistent with the RdRP serving as the main target [19], [21]. A recent report suggests that favipiravir induces lethal mutagenesis in influenza A viruses through selective pressure applied in cell culture [21]. However, it is uncertain whether this mechanism plays any role in vivo. Collectively, the evidence suggests that favipiravir selectively inhibits RNA virus RdRP, with only limited toxicity to cells. This specificity makes favipiravir an attractive candidate for a broadly active therapeutic with potential to treat multiple viral diseases. Our findings with i.p. favipiravir treatment represent the most significant level of protection ever reported for an antiviral drug intervention in the difficult to treat JUNV guinea pig infection model [26]. A study to define the therapeutic window in guinea pigs and efficacy studies in a nonhuman primate model are planned.
|
10.1371/journal.pcbi.1002907 | Structure-based Molecular Simulations Reveal the Enhancement of Biased Brownian Motions in Single-headed Kinesin | Kinesin is a family of molecular motors that move unidirectionally along microtubules (MT) using ATP hydrolysis free energy. In the family, the conventional two-headed kinesin was experimentally characterized to move unidirectionally through “walking” in a hand-over-hand fashion by coordinated motions of the two heads. Interestingly a single-headed kinesin, a truncated KIF1A, still can generate a biased Brownian movement along MT, as observed by in vitro single molecule experiments. Thus, KIF1A must use a different mechanism from the conventional kinesin to achieve the unidirectional motions. Based on the energy landscape view of proteins, for the first time, we conducted a set of molecular simulations of the truncated KIF1A movements over an ATP hydrolysis cycle and found a mechanism exhibiting and enhancing stochastic forward-biased movements in a similar way to those in experiments. First, simulating stand-alone KIF1A, we did not find any biased movements, while we found that KIF1A with a large friction cargo-analog attached to the C-terminus can generate clearly biased Brownian movements upon an ATP hydrolysis cycle. The linked cargo-analog enhanced the detachment of the KIF1A from MT. Once detached, diffusion of the KIF1A head was restricted around the large cargo which was located in front of the head at the time of detachment, thus generating a forward bias of the diffusion. The cargo plays the role of a diffusional anchor, or cane, in KIF1A “walking.”
| It is one of the major issues in biophysics how molecular motors such as conventional two-headed kinesin convert the chemical energy released at ATP hydrolysis into mechanical work. While most molecular motors move with more than one catalytic domain working in coordinated fashions, there are some motors that can move with only a single catalytic domain, which provides us a possibly simpler case to understand. A single-headed kinesin, KIF1A, with only one catalytic domain, has been characterized by in vitro single-molecule assay to generate a biased Brownian movement along the microtubule. Here, we conducted a set of structure-based coarse-grained molecular simulations for KIF1A system over an ATP hydrolysis cycle for the first time to our knowledge. Without cargo the simulated stand-alone KIF1A could not generate any directional movement, while a large-friction cargo-analog linked to the C-terminus of KIF1A clearly enhanced the forward-biased Brownian movement of KIF1A significantly. Interestingly, the cargo-analog here is not merely load but an important promoter to enable efficient movements of KIF1A.
| Time dependent structural information is of central importance to understand detailed mechanisms of biomolecular systems. In particular, biomolecular machines dynamically transit many structurally and chemically distinct states making cycles in state space, by which they fulfill their functions. Unfortunately, no single experimental technique provides sufficient spatio-temporal resolution for them. X-ray crystallography and others provide structural information at high resolution, but this is primarily static. Biochemical and single molecular experiments tell us kinetic and dynamic behaviors, but their spatial resolution is limited. To fill the gap among them, molecular dynamics (MD) simulations have been playing important roles. Yet, due to their size and long time scale involved, atomistic MD cannot cover an entire cycle of molecular machines at the moment [1]. To overcome this limitation, recently, we initiated to use structure-based coarse grained MD (CGMD) methods [2], [3] to mimic the cycle of machines for the case of F1-ATPase and others [4], [5]. Notably, most of these machines contain more than one ATPase domains and their coordinated dynamics are crucial to understand the mechanisms [6], [7], [8], [9]. This is an interesting issue, but at the same time, makes the cycle unavoidably complicated. Thus, for the simplicity and clarity, it is good to study those that contain only one ATPase domain and that have much of crystallographic information. In this sense, a single-headed kinesin, KIF1A, is an ideal target system, for which here we performed CGMD simulations mimicking an entire ATP hydrolysis cycle.
Kinesin is a family of molecular motors that move unidirectionally along microtubule (MT) using ATP hydrolysis free energy [10]. In the family, the conventional kinesin, kinesin-1, was experimentally characterized to move toward the plus ends of MT processively with discrete 8-nm steps per one ATP hydrolysis reaction, where the coupling between ATP hydrolysis reactions and 8-nm steps is rather tight [11], . The conventional kinesin is a two-headed motor and has been shown to “walk” in a hand-over-hand fashion by coordinated motions of the two heads [6],[9],[17],[18]. In this sense, it is a surprise that even though KIF1A, a member of kinesin family, is a single-head motor, it still can move processively and directionally along MT, as observed by single molecule experiments [19], [20], [21]. In particular, the mechano-chemical coupling of KIF1A is loose: KIF1A can move back and forth stochastically with an average biased towards the forward direction, with step sizes in multiples of 8-nm. This is in contrast to conventional kinesin that seldom shows backward steps without a large load and that shows a uniform step size of 8-nm per one ATP hydrolysis [19], [20], [21]. Thus, KIF1A must use a different mechanism from the conventional kinesin to achieve the overall unidirectional motions. How KIF1A, with only one head, can generate the unidirectional movements driven by ATP-hydrolysis reaction is unclear in terms of structural dynamics, which we address in this paper by structure-based CGMD. Various molecular simulations have been applied to kinesin as well as other molecular motors [4], [8], [22], [23], [24].
For the molecular simulations of KIF1A movements, structural information on nucleotide-dependent conformational change is indispensible. X-ray crystallography provides KIF1A structures in two major conformations; ATP and ADP bound forms [25], [26], [27]. The two forms share the overall fold of the head domain with some changes. One crucial change is in the helix α4; its orientation relative to the rest of the head is rotated by about 20 degrees between the two forms (Fig. 1A, blue for ATP-form and red for ADP-form). Another major change is in the so-called neck-linker region, which is the C-terminus of the head domain: the neck-linker is ordered and tightly docked to the core of the head in the ATP form (magenta in Fig. 1B), while it is disordered and thus invisible in the ADP form. This neck-linker docking/undocking has been implicated as a source of the power-stroke in the kinesin family [28], [29]. The K-loop (L12-loop) and L11-loop, which flank the α4 helix, also show some changes between the two forms (Fig. 1A). Cryo-electron microscopy (cryo-EM) of the KIF1A-MT complex together with X-ray structures of building blocks led to structural models for the KIF1A-MT complex in the two major forms of KIF1A (the ATP and ADP forms) [30], [31]. The modeled complexes show that, in both of the forms, the key interaction sites of KIF1A with MT is the α4 helix, which fits to a groove located between α-tubulin and β-tubulin. In the two forms of KIF1A, the orientation of the α4 helix relative to MT is mostly unchanged, which leads to the 20-degrees rotation of the core domain relative to the long axis (z-direction in this article) of MT depending on the bound nucleotide states (Fig. 1C): In the ATP-form, the core adopts the “upright” docking (blue in Fig. 1C left), while in the ADP form the core is rotated about 20 degrees and adopts the “tilted” docking (red in Fig. 1C right) [31]. This core rotation has been suggested to be important for KIF1A movement [26], [31].
The KIF1A-MT complex models provide a clue for the processivity of KIF1A. Regardless of the nucleotide states, the positively-charged K-loop of KIF1A is close to the negatively charged E-hooks, disordered C-terminus regions of α/β-tubluin [26], [31]. Thus, the long-ranged electrostatic attractions between K-loop and E-hooks are assumed to prohibit the KIF1A from completely leaving from MT. This idea is supported by a mutational experiment, in which charge reduction of K-loop decreased the processivity [19].
The ATP hydrolysis cycle and its correlation with KIF1A head motion have been investigated previously [19], [20], [21], [26] (see Fig. 2A). The ATP form of the KIF1A head binds strongly to MT (T-phase in Fig. 2A), whereas the direct contact of the ADP-form of KIF1A to MT is weak. Thus, after the ATP hydrolysis and Pi release, the KIF1A head can detach from MT (still loosely bound to MT via electrostatic interactions of the K-loop and E-hooks). The detached head starts diffusion along MT under the constraints generated by the interaction of the K-loop and E-hooks. After the long one-dimensional diffusion along MT, KIF1A can finally find a binding site located at the groove between α- and β-tubulins (D-phase in Fig. 2A). The contact between tubulin and KIF1A induces ADP dissociation from KIF1A leading to the nucleotide free state. In this state, KIF1A binds MT tightly (Φ-phase in Fig. 2A). At the final stage, ATP binding induces neck-linker docking and the rotation of the core (T-phase in Fig. 2A). The above knowledge, however, does not tell us the mechanism of how KIF1A can generate directional movement towards the plus end of MT.
To address the mechanism of directional movement, we designed and conducted a series of molecular simulations employing structure-based CG protein models. The structure-based CG protein models have proven to be useful to study mechanical aspects of kinesin [8], and other molecular motors [4], . Based on the energy landscape view of proteins [32], [33] and structural data for the two forms of KIF1A, we set up single and/or two-basin energy landscapes of KIF1A for every phase of an ATP cycle [34], [35]. Then, the ATP hydrolysis cycle was mimicked by dynamically switching the energy functions of KIF1A in different phases of the cycle (Fig. 2B) [4], [36]. While the full-length KIF1A has a rather long C-terminal tail, we here concentrate on a truncated KIF1A (C351) that was used in in vitro single-molecule assays [19],[20],[21]. We note that, by employing the structure-based CG simulations, our purpose here is not to conduct a single simulation that most accurately approximate the real molecular system, as some parameters in the CG simulations are not accurately derived from atomic interactions. Instead, taking advantage of the speed of the structure-based CG simulations, we systematically conduct a series of simulations for a broad range of these parameter values. These comparative computer experiments are useful for a mechanistic understanding.
We designed a simulation system for one ATP hydrolysis cycle of KIF1A that induces KIF1A motions along MT. The simulation system contains 7 protein subunits: a KIF1A molecule that moves dynamically and three copies of tubulin αβ dimers that were fixed in the form of a segment of single protofilament of MT (Fig. 1D). All the proteins were modeled at a one-bead-per-residue resolution (each amino acid was represented by a bead located at the Cα position).
For KIF1A, we employed structure-based CG models that concisely represent the energy landscape, which is a globally funnel-like shape where the bottom of the funnel can have more than one basin [33]. We focused on a truncated KIF1A C351 (unless otherwise mentioned) since the motility of this type KIF1A is intensively investigated in the single-molecular assay of Okada et al [19], [20], [21]. Conformational changes of KIF1A upon chemical reactions were simulated by the multiple-basin model [35], while the long-time dynamics that do not involve chemical reactions, such as diffusion process, were simulated by a single-basin perfect-funnel (i.e., Go) model [34], [37] (see below and Materials and Methods for more details). Protein dynamics was simulated by stochastic differential equation, i.e., the Langevin equation (see below and Materials and Methods for more details). The crystal structures of ATP-bound KIF1A (designated as KIF1A(T) hereafter) and ADP-bound KIF1A (KIF1A(D)) are available from the Protein Data Bank and were used in the CG models as reference structures of the corresponding states. For the KIF1A-MT complex structures, the cryo-EM-based models for the ATP- and ADP-bound KIF1A-MT complexes are also available and were used (we designate XT and XD respectively). These models explain the high and low affinities in ATP-bound and ADP-bound forms of KIF1A, respectively, by the number of direct contacts. The structure for nucleotide-free KIF1A (KIF1A(Φ)) is currently unavailable; we assumed that the neck linker is disordered based on experiments, and that the KIF1A(Φ)-MT complex structure XΦ except the neck linker to be the same as that of XT because both states have a similarly high affinity to MT. Using these complexes, we modeled the interactions between KIF1A and MT as a Go-like pair potential (unless otherwise mentioned).
In the current CG model, the interaction strength between KIF1A and MT is a key parameter. First, the interaction strength parameter had to be tuned so that KIF1A(T) can stably bind to MT while KIF1A(D) can detach from MT during the affordable simulation time. This tuning was easy because, as mentioned above, the modeled complex structures of KIF1A-MT have more residue-contacts in the ATP form than in the ADP-form. A more delicate tuning was necessary for the affinity of KIF1A(D) to MT because KIF1A(D) is expected to detach from MT and later reattach. Obviously, a too weak interaction does not lead to attachment of KIF1A to MT, whereas a too strong interaction does not allow the detachment from MT. Via many preliminary runs, we found a certain range of the interaction strength parameters that satisfy these conditions (described in the next subsection).
Our simulation started from the XT. KIF1A was bound to the central tubulin αβ dimer (Fig. 1D). We simulated the KIF1A(T) state for 5×105 τ, where τ is the unit of time in CG-simulation, using the multiple-basin potential with two basins: a stable basin at XT and a meta-stable basin at XD structures (Fig. 2B top). The unit of time τ can be mapped to ∼0.128 ps in real time scale based on the diffusion constant of the KIF1A head (see Materials and Methods for the detail information). Then, we induced the conformational change to the ADP-bound form by switching the potential so that the XD structure becomes more stable than XT (see the second row and left cartoon of Fig. 2B). With this setting, we simulated the system for 4×106 τ, which is long enough to complete the conformational change to ADP-form. For many samples, KIF1A(D) detached from MT during this period. We note that, throughout the simulations, a constraint potential was applied that represents long-range loose interactions between the K-loop and E-hooks, by which KIF1A cannot move far away from MT (see Materials and Methods for details). Then, we conducted a long simulation (2×108τ) with the single-basin Go potential for the XD (the second row and central cartoon in Fig. 2B). The switch from the multiple-basin potential to the single-basin Go potential saves computer time and is done solely for technical reasons. During this period, many trajectories showed KIF1A re-attachment to MT. Once KIF1A attached to MT, we continued the run for another ∼1×107τ and then moved to the next stage. The next stage is a preparation to the subsequent conformational change to the nucleotide-free (Φ) state and uses the multiple-basin model with the stable basin at XD and the meta-stable basin at XΦ for 5×105τ. After that, corresponding to the release of ADP, we induced the conformational change to the nucleotide free form by switching the potential so that the XΦ structure is more stable than XD (the third row right in Fig. 2B). Subsequently, for a long time dynamics, we used the single-basin potential for the Φ state for 1×107τ. Finally, ATP-binding is mimicked by switching the potential to the single potential for XT. We simulated the T state for ∼2×108τ, which completes the XTXDXΦXT cycle.
We now analyze KIF1A movement during one ATP cycle. As in Fig. 2A, it is expected that KIF1A detaches from MT and attaches to MT both in the D-phase. Thus, modeling of the interaction between KIF1A(D) and tubulin is very delicate. Since the CG modeling is unavoidably less accurate, instead of deciding one “correct” interaction strength, we scanned the strength over a certain range.
In a strong interaction case (designated as [stand-alone/strong], εgoKIF1A-MT = 0.225) (Throughout the paper, the energy unit corresponds to kcal/mol (∼1.7 kBT = ∼6.95 pN.nm) although the mapping is rather approximate), we saw KIF1A cannot detach from MT for 99 of 100 trajectories (Fig. 3A) within the simulated time. Whereas, with a weak interaction ([stand-alone/weak], εgoKIF1A-MT = 0.153) that was carefully tuned after trial-and-errors, we found that KIF1A can detach from MT and attach to MT (the first three cases in Fig. 3B) for 186 of 235 samples (∼80%). The rest 49 samples did not show detachment (bottom in Fig. 3B). The first, second, and third cases in Fig. 3B illustrate the one forward step (+8 nm), the zero-step (0 nm), and the one backward step (−8 nm) within one ATP hydrolysis cycle, respectively (For an example of stand-alone KIF1A movements for [stand-alone/weak], see Supporting Information Video S1). Of the 186 cases that KIF1A detached from and attached to MT within one ATP chemical cycle (TDΦT), the positions of KIF1A at the end of simulations were +8 nm (the forward step) for 44 cases, 0 nm (zero-step) for 92 cases, and −8 nm (the backward step) for 50 cases (For statistics, Table 1). We note that the system contained only 3 pairs of tubulin αβ's that correspond to kinesin binding sites of +8 nm, 0 nm, and −8 nm so that possibilities of two steps were out of the scope here. On average, no significantly biased move was observed. Apparently, this does not explain the in vitro single molecule experiments that found forward biased moves.
The simulations above did not consider electrostatic interactions at all, which may have affected the results. Indeed, recent work by Grant et al reported forward bias of two-headed kinesin landing due to electrostatic interactions [22]. We thus added the electrostatic interactions between KIF1A and MT by the Debye-Huckel formula and repeated the same set of simulations for 80 samples for the case of [stand-alone/weak/DH]. We set the salt concentration of 50 mM, and put +1 charges to all the Lys, Arg, and His residues and −1 charges to all the Asp and Glu residues in the simulated system. Of 80, 6 samples did not show detachment, 12 samples showed one forward-step (8 nm), 16 samples showed one backward step, and 46 samples returned to the original site (see Fig. S1). Thus, inclusion of the simple electrostatic interactions did not produce forward-biased movements although it changed the trajectories to some extents (see Figs S2, S3, S4, S5, S6). We further tried simulations with many different sets of parameters never finding biased motions.
Our results is apparently inconsistent with the biased binding mechanism proposed in [21]. There can be two possibilities. 1) Some fine effect which is not included in our CG simulations, such as more accurate electrostatic treatment by Grant et al, is responsible for the forward biased binding. 2) The forward-biased binding is not realized. Further work is necessary to solve the issue.
In struggling for search of models/situations that exhibit the forward biased move of KIF1A, we came up with a situation that a large cargo-analog is attached to the C-terminus of the neck-linker of KIF1A. The cargo-analog is modeled as a large sphere of ∼1 µm radius, and thus has very small diffusion constant. There are some in vitro experiments for myosin, as well as another kinesin mutant, that suggest the importance of diffusion anchor linked at the end of motor proteins for processive and directional movements [38], [39]. Technically, we added a mass point with large friction coefficient to the C-terminus of the neck linker.
With a large cargo-analog, we first used a strong interaction between KIF1A and MT ([cargo/strong], εgoKIF1A-MT = 0.225, the same strength as the case of [stand-alone/strong]), and simulated one ATP cycle for 109 samples. We modeled the cargo as a sphere of radius 3000 times as large as the radius of an amino acid, which is ∼1 µm. Assuming the same density as amino acids, the mass of the cargo scales as 30003 times as large as that of an amino acid. The Stokes-Einstein law D = kBT/6πηr, where η is water viscosity: ∼0.8 m [Pa s] and r is the radius of the particle, gives that the diffusion constants Dcargo for the cargo is 3000 times smaller than the diffusion constant of an amino acid. (See Materials and Methods for the detailed information).
In the simulations, we found most samples either moved one-step forward (52 of 109 cases, an example in the upper panel of Fig. 4A top and Video S2) or re-bound to the original site (56 of 109 cases, the upper panel of Fig. 4A bottom), while almost no case of the backward step was found (Table 1). In ATP-bound state (t<5105τ), KIF1A head kept binding to MT firmly and the cargo-analog did not move significantly at 4 nm in front of the head corresponding to the length of the neck-linker (a snapshot in Fig. 4B top, a histogram in Fig. 5 left). Immediately after the ATP hydrolysis, KIF1A head detached from MT quickly. After the detachment, KIF1A head exhibited quasi-one dimensional diffusion along MT, while, due to the large friction, the cargo-analog did not move significantly. Thus, the fluctuation of the KIF1A head was restricted around the almost-fixed cargo located 4 nm in front (Fig. 4B and Fig. 5 left). The cargo-analog played a role of an anchor (or a cane). After some diffusion, the detached head finally re-bound to MT. Because of the limited range of diffusion, the re-binding site was either the forward site (+8 nm) or the original site (0 nm). After the attachment on MT, we changed the state of the system from ADP-state to Φ-state, which did not lead to any marked difference in the movement of the cargo or the head. After that, ATP binding to KIF1A induced the neck-linker docking that moved the position of the cargo-analog, which is about 8 nm in case of the forward step (Fig. 5 left). Thus, after one ATP cycle (TDΦT), the 8-nm or 0-nm displacements of the cargo-analog as well as the head were realized stochastically.
With a weak interaction between KIF1A and MT ([cargo/weak], εgoKIF1A-MT = 0.153), we still found clear forward bias (the bottom panel of Fig. 4A) although the details were different. In particular, due to a weaker interaction, the average time for the head diffusion increased, which resulted in larger exploration by one-dimensional diffusion and appearance of the one backward step (26 of 150 samples) as well as the one forward (43 of 150) step, and the zero step (81 of 150) (Fig. 5 middle). As noted before, our simulation system included only the three binding sites and thus two forward or backward steps were not realized by design. For comparison, Fig. 5 right shows the histogram of the move for the case of [stand-alone/weak], confirming that no significant bias is observed.
We now focus on the detachment process of the KIF1A head from MT after the ATP hydrolysis and Pi release. Upon the potential switch from ATP- to ADP-state at t = 5105 τ (Fig. 2B top to the second row left), the decrease in the number of residue-contacts between KIF1A head and MT led to the reduced binding energy, which could induce the detachment of KIF1A head.
Interestingly, with the strong interaction, the stand-alone KIF1A simulation showed the KIF1A head detachment with the probability 1%, whilst the simulation with the cargo-analog promptly induced the head detachment with the probability 100% (Fig. 6A). Thus, clearly, the cargo-analog enhanced the KIF1A head detachment from MT. Even with the weak interaction between KIF1A and MT, the detachment probability was 79% for the stand-alone KIF1A (Fig. 6A).
With the strong interaction, we tested the detachment process with three cargo sizes (and thus three frictions and masses) (the inset in Fig. 6A); the small (light-green), the middle-size (red, the default one) and the large (purple) cargoes correspond to the radii of 2000, 3000, and 4000 times of one amino acid, respectively. Technically, for given radii, masses and frictions were scaled according to Stokes-Einstein law in the same way as described. We see that KIF1A did not detach from MT with the probability 11% for the case of the small cargo, while the detachments probabilities were 100% for the system with the middle or the large cargo. Thus, the relatively large friction/mass cargo promoted the detachment of the KIF1A head.
In the complex of KIF1A-MT, the α4 helix of KIF1A fits into a groove of MT. When the ATP hydrolysis occurs in the KIF1A head bound to MT, the head tends to make conformational change from ATP-form to ADP-form. With the constraint on the α4 helix, the conformational change would induce about 20 degree clockwise rotation of the head relative to the microtubule (viewed from the top as shown in Fig. 1C left to right), which increases the distance between C-terminal of the head and the cargo rapidly. Then, a tag-of-war between the head and the cargo takes place. When the cargo is sufficiently large, the cargo is less mobile and wins the tag-of-war, thus finally pulling the KIF1A head out of MT.
Fig. 6B illustrates time series of the binding energy for the strong interaction case. For the case of [stand-alone/strong] (orange), the binding energy was weakened from ∼−30 kcal/mol in T-state to ∼-20 kcal/mol in D-state, but the latter was strong enough to hold the KIF1A head stably. For the large cargo case (purple), upon ATP hydrolysis, KIF1A promptly detached from MT. For the cases of small (light-green) and the middle-size (red) cargoes, TD switch immediately weakened the binding energy to ∼−12.5 kcal/mol, which were followed either by detachment or by the relaxing to the binding energy ∼−20 kcal/mol in D-state (light-green). This transient intermediate state with the binding energy ∼−12.5 kcal/mol corresponds to the frustration imposed by the immobile cargo. Similar behavior was seen in the case of the weak interaction (Fig. 6C).
We found it interesting to plot the trajectories in the plane (zrelative,EB) [zrelative: the relative position of the cargo (zcargo-zhead), EB: the binding-energy] both for the cases with and without the cargo-analog (Fig. 6D). Trajectories start from the right-lower area in (EB, zrelative) plane. With the large cargo (red and blue), after the relaxation of the binding energy from the initial condition to 0 kBT (the detachment), the cargo-analog moved. Whereas, without the cargo-analog, the C-terminus fluctuation occurred first and then KIF1A head may or may not detach from MT (orange and dark-green). The difference comes from the different time scale for the mobility of the cargo-analog.
Experimentally, the binding free energy of KIF1A head with MT was estimated from the dissociation constant as ∼−20 kBT in the ADP bound state [19]. In the current simulations, the binding energies in the D-phase are −35 kBT for the strong interaction case (see Fig. 6B) and −17 kBT for the weak interaction case (Fig. 6C). Note that the experimental estimate is the free energy about the standard state, while the estimates from simulations are merely interaction energies. Thus these numbers should not be quantitatively compared. With the uncertainty in mind, perhaps, the real binding strength may fall in between the strong and the weak interaction cases.
Next, we analyze the diffusion and the attachment processes of KIF1A head after the detachment in ADP-state (Fig. 7). The attachment rate for [cargo/strong] is larger than that for [cargo/weak], as expected. Interestingly, the attachment rate for [cargo/weak] was much smaller than that for [stand-alone/weak], probably due to the restricted motions anchored by the large cargo-analog. Thus, the large cargo-analog enhanced the detachment, but retarded the attachment.
Fig. 7B shows a transient histogram for the z-coordinates of the KIF1A head and of the cargo-analog soon after the detachment from MT. With the cargo-analog (Fig. 7B left and middle), its positions were nearly fixed, whereas the head fluctuated broadly (∼4 nm in both directions), which coincides with the length of neck linker. We note that, since we measure the diffusion after the detachment from MT, the histograms for [cargo/strong] and for [cargo/weak] are nearly the same: The diffusion process itself (up to the attachment) was not affected by the interaction strength. As the diffusion time increases, the cargo-analog slowly moves, which enables the head to reach the backward site, as well as the forward site. For the stand-alone case (Fig. 7B right), C-terminus position diffused quickly after the detachment from MT, and the distributions of the C-terminus and the head were nearly symmetric about the starting point (0 nm).
In the simulations, the average times τattachment for attachment of the KIF1A head to MT for the system cargo/strong and cargo/weak were ∼0.2×108 τ (∼2.5 µs) and ∼0.5×108 τ (∼6.4 µs), respectively. A rough estimate of the diffusion length in this time scale is ∼1.2–1.9 nm, which is small.
After ATP binding, the neck-linker docked to the head core. The neck-linker docking moves the cargo-analog by about +8 nm when the head landed to the forward site (Fig. 8). The docking rate depends on the size of the cargo-analog, as expected.
Only in the cases of the weak interaction, after the attachment of the head onto MT, occasionally the head re-detached from and then re-attached to MT (Fig. 9). This extra processes, being not coupled with ATP cycle, did not produce significant bias in the KIF1A move.
In the above simulations, the cargo was always placed at z∼4.25 nm based on the ATP-form reference structure, which may raise a concern that this specific initial positioning may affect the stepping statics. To address this concern, we performed the same type of one-ATP cycle simulations with various initial cargo positions z; z = 4.75, 4.25, 3.75, 3.25, 2.75, 2.25, 1.75, and 1.25 nm especially for the cargo/strong case. This range corresponds to the range of cargo found at the end of original simulations (see Fig. S7 A the upper panel which shows the distribution of the probability density for the relative position of the cargo at the end of original simulation). From each of these cargo (initial) positions, 10 simulations were conducted. From the initial position of the cargo: z>3 nm, we found clear forward-biased moves, whereas z<3 nm, the head seldom detached from MT (the lower panel of Fig. S7A). Overall, by distributing the initial cargo positions, the forward bias is somewhat reduced on average. Importantly, however, we still clearly see, on average, forward-biased moves of KIF1A with the cargo.
In this paper, we primarily focused on the specified molecular construct (C351) used in in-vitro motility assay experiments [19], [20], [21], in which the length of neck-linker except for His-tag is 22-residues. Here, to test robustness of our results, we investigated the stepping statics of another construct that has 5-residue longer neck-linker. The 5-residue segment is modeled as a flexible chain (by Modeller). In a similar way to the above sub-section, we estimated the range of the cargo position (Fig. S7B upper panel which shows the distribution of the probability density for the position of the cargo), and repeated simulations (10 runs each) with the initial cargo position at z = 6.25, 5.75, 5.25, 4.75, 4.25, 3.75, 3.25, 2.75, 2.25, 1.75, and 1.25 nm. From the initial position of the cargo: z>4.5 nm, we found clear forward-biased moves, whereas z<4.5 nm, the head seldom detached from MT (Fig. S7B lower panel). Thus, although the bias is weakened, we still see clear forward-biased moves of KIF1A with the cargo linked by 5-residue longer linker.
Conventional kinesin is dimeric and “walks” in a hand-over-hand fashion, akin to human walking by two legs. Extending the analogy to human walking, the current simulations suggest that the large cargo-analog can play the role of a cane for the walk of single-headed kinesin; with a cane, we can walk even with one leg.
Although in this work, we focused on a truncated KIF1A which is used in the single-molecular assay of Okada et al [19], [20], [21], we should note that the cellular function of KIF1A in vivo is markedly more complicated than the situation we considered here. Several experiments [40], [41] showed that KIF1A may be dimerized by virtue of being bound to a single cargo-analog in some case. Our model does not straightforwardly apply to the dimeric KIF1A system in vivo.
Next, we discuss in vitro experiments of related systems. First of all, forward-biased movements were observed for single-head kinesins, both KIF1A and a single-head construct of conventional kinesin mutant, with latex beads linked to C-terminus, where the size of beads are sub-µm to µm [13], [21]. Thus the current simulations are perfectly consistent with these results. In a study of myosin VI, single-molecule experiments reported very similar phenomenon to our simulations [38]. A single-head construct of myosin VI did not show directional movements without beads/cargo. When a bead was attached to an end of the myosin head, it exhibited directional movements. It was argued that the bead played the role of diffusion anchor.
Some other experiments in vitro are subtle. While a truncated single-headed kinesin (K351) did not show marked processive movements, it exhibited processive and directional movements when fused with BDTC (1.3-S subunit of propionibacterium shermanii transcarboxylase) [39]. Here, BDTC is much smaller than beads/vesicles and thus does not apparently correspond to the current simulations. Yet, the linked BDTC increased K351-BDTC affinity to MT which implies that BDTC attractively interacts with tubulin. Thus, the interaction of BDTC with MT may provide additional friction to the C-terminus of K351, which is qualitatively the same as the role of the cargo-analog in our simulations.
Next, we discuss the dynamics of KIF1A head and cargo-analog. In Video S2 and Fig. 4, the cargo looks almost fixed at the beginning. First, we note that, although we used the cargo-analog of ∼1 µm-size, Video. S2 and the snapshot in Fig. 4B drew much smaller ball to “visualize” the cargo position. Thus a small ball is purely for graphics. Taking into account the ∼1 µm-sized cargo together with the Stokes-Einstein law and Maxwell-Boltzmann distribution, we see that the cargo does not move much: Based on the Stokes-Einstein law DKIF1A = 1.2×108 [nm2/s], and Dcargo = 2.9×105 [nm2/s], respectively, we estimate that, while the KIF1A head diffuses for ∼8 nm distance, the cargo diffuses only ∼0.4 nm which is rather small. Thus, it is physically reasonable that the cargo looks immobile in Video. S2, and Fig. 4. Furthermore, in Video S2 and Fig. 4A, the motions of KIF1A head look like Brownian dynamics with little effect of inertia, whereas the motions of the cargo-analog look under-damped oscillation. This difference can be understood by estimating the lifetime of the corresponding velocity correlations. We estimated that the lifetime of velocity correlation for the KIF1A head is ∼10 τ, and that for the cargo-analog is ∼108 τ. A characteristic time scale of the detached KIF1A head to find the adjacent binding site (z = L or -L) from the dissociation was ∼106 τ (see Material and Methods). Therefore, within this time scale, the motions of KIF1A should be diffusive, while the motion of the cargo-analog is damped oscillation.
Related to these arguments, we also note time scale for KIF1A head diffusion. As in Result, our estimate in simulations was τattachment∼2.5 µs–6.4 µs. Experimentally, the mean duration time of weak-binding state was estimated as τw = 7.5 ms [19]. However, τw obtained via an indirect estimate seems to include not only the diffusional searching time, but also the ADP release time. Since ADP release is a slow process, we do not know much on the diffusional search time.
We next discuss the effect of neck-linker length on the enhanced forward-biased motions. After the neck-linker docking, the average positions of cargo for C351 and a 5-residue longer variant C351+5 were similar, while the distribution of the positions was broader for the C351+5 case than that for the C351 case. (Fig. S7 A upper panel). Based on Fig. S7, we can estimate that the average steps per one ATP cycle are about 2.7 nm for C351 and about 2.0 nm for C351+5. Thus, a longer neck linker contributes to broadening of the distribution of the cargo position at the time of ATP hydrolysis, which results is gradual decrease in the forward biased movement of KIF1A. The full-length KIF1A has much longer neck linker. In the recent simulation study for the dimeric kinesin with a long tail domain and cargo [42], the neck-linker docking itself did not bring the cargo to the forward position significantly. On top, the full length KIF1A tends to dimerize. So, our analysis focuses on the truncated KIF1A construct such as C351 and argument for the full-length system needs further analysis.
Importantly, the current simulations showed that the linked cargo-analog is sufficient to induce the biased Brownian movement of KIF1A, but whether the linked diffusion anchor is necessary or not was not investigated. In vitro experiments, KIF1A exhibited forward-biased Brownian movements even when only a chromophore was attached [19], [20], [21]. Since the chromophore is much smaller than the cargo/bead, this does not correspond to the current simulations. Note that His-tags attached to C-terminus of KIF1A in vitro constructs may also contribute to additional interactions with MT since tubulin contains negatively charged C-terminus tails (E-hooks) on the surface. The result that some constructs with too short neck-linker did not exhibit directional movements suggests importance of a certain length of neck-linker between the head and the His-tag. Yet, we do not exclude the possibility of other mechanisms that induce the directional movements. In particular, recent computational work reported that electrostatic interactions between two-head kinesin heads and MT can provide modest bias to the forward direction [22]. Thus, the linked cargo-analog can be to enhance the forward bias.
In the end, we briefly mention about possible experiments which can test the current simulation results. The direct test of the proposed mechanism is to perform a motility assay with a bead or dye attached to the core of the KIF1A head which is far from N- or C- termini and which is located on the surface opposite to the MT binding orientation. Even better way is to introduce two imaging probes; a bead in the C-terminus and a dye to the core of KIF1A head. We expect to see movements in asymmetric hand-over-hand fashion. Another easier but more indirect test is to investigate ion-strength dependence of the stepping statics for C351. If the interaction between KIF1A and MT is weakened by a higher ion-strength, the re-binding time of KIF1A becomes longer, which results in weakening of the forward-bias.
To simulate an ATP cycle by the structure-based CG model, we need reference KIF1A-MT complex structures Xν for every states ν ( = T, D, Φ) in the cycle (T, D, and Φ correspond to ATP, ADP and nucleotide free state, respectively). The crystal structures of ATP-bound KIF1A (KIF1A(T)) and ADP-bound KIF1A (KIF1A(D)) are available from the Protein Data Bank and were used in the CG models as references. For the KIF1A-MT complex structures, we used the model structures 2HXF (the pdb id) for the KIF1A(T)/tubulin αβ complex XT, and 2HXH for the KIF1A(D)/tubulin αβ complex XD [30], [31] (see Fig. 1A, 1B, and 1C). Since motility-assay experiments heavily used a chimera protein C351 where the catalytic core of KIF1A was fused to the neck linker of conventional kinesin (KIF5C) [19], [20], [21], we employed the same chimera C351 (KIF1A-KIF5C), except for N-terminal T7-tag and C-terminal His-tag and Cys. Missing residues in the loop of KIF1A, including a part of the neck-linker region, were modeled by Modeller [43]: We constructed 200-samples and chose the model with the best Modeller score. These modeled loops were treated as flexible regions with reduced force constants (see Coarse-grained model in Supporting Information Text S1).
For the nucleotide free state, no structure is available. Since KIF1A (Φ) constitutes the strong-binding state in a similar way to KIF1A (T), we decided to use the same structure as the XT excluding the neck linker, which is known to be disordered in Φ-state [28], [44], [45], [46]. We treated the neck liner in Φ-state as flexible regions.
Both 2HXF and 2HXH contain missing residues in tubulin αβ at the so-called E-hook region, which were not explicitly modeled. Instead, we included the effect of E-hook in a simpler way (see Coarse-grained model in Supporting Information Text S1).
The simulation system here contains a KIF1A molecule and three copies of tubulin αβ dimers (Fig. 1D) where all the tubulin molecules were fixed throughout simulations. The initial structure of simulations contained XT structure of KIF1A attached to the central tubulin αβ dimer at the form of 2HXF. The coordinates were set so that the MT protofilament is along the z-axis with the plus end being positive z, and KIF1A-binding surface of tubulin is roughly perpendicular to y-axis (see Fig. 1D). The origin was defined by the position of Cα-atom of Phe94 of KIF1A at the initial structure (roughly the center of mass of KIF1A). The period of the MT is about 8 nm along z-axis.
We applied the structure-based CG models for the KIF1A-MT system [34], [35]. KIF1A and three tubulin αβ dimers were represented by a set of beads, where each bead placed at the position of Cα atom represents one amino acid. (See Supporting Information Text S1 for detail).
To mimic an ATP hydrolysis cycle (Fig. 2A), we employed a simulation protocol summarized in Fig. 2B. The dynamics of the KIF1A protein were simulated by the underdamped Langevin equation at a constant temperature T = 290.0 K with CafeMol [47]. The step size dt of the time integration is dt = 0.1 τ, where τ∼0.128 (ps) is the unit of time in CG-simulation.where vi is the velocity of the i-th bead and a dot represents the derivative with respect to time: t (thus, vi = . ξi is a Gaussian white random force, which satisfies <ξi> = 0 and <ξi(t) ξj(t′)> = 2mi γi kBT δij δ(t -t′) 1, where the bracket denotes the ensemble average and 1 is a 33 unit matrix. kB, is the Boltzmann constant, γi and mi are the friction coefficient and the mass for the one residue. (See Supporting Information Text S1 for more detail).
The units of CG-simulations are given as follows: The length unit is 0.1 nm. The energy unit is kcal/mol (∼6.95 pN.nm). The unit of mass can be defined as setting mi, the mass for an amino acid. We set mi = 10, which is just a convention in CafeMol, which leads to the mass unit as 2.27510−26 kg.
The friction coefficient γi for a residue is decided so that the diffusion constant of the KIF1A head in simulations roughly agrees with that (1.2×108 [nm2/s]) by the Stokes-Einstein law: By setting γi = 0.1 for an amino acid, we obtained a reasonable diffusion coefficient of KIF1A head (DKIF1A = 1.58×10−5 [nm2/τ]) in simulation and the time unit in CG-simulation τ ∼0.128 (ps).
As for the default-sized cargo, we modeled its radius 3000 times as large as the radius of an amino acid, which is ∼1 µm. So, the mass mcargo of the cargo scales as mi×30003, which gives 2.7×1011. The friction coefficient γcargo for the cargo is 0.1×3000/30003 = 1.1×10−8 [1/τ].
In this paper, we used the underdamped Langevin dynamics as the equation of motions. We note that, even when we use the underdamped Langevin equation, it gives us overdamped motions when we investigate motions in time scales longer than the velocity correlation time. For an amino acid, the velocity correlation is given by 1/γi = 10 τ (where the simulation time unit τ corresponds to ∼0.128 ps in real time scale). For the KIF1A head, we computed the lifetime of the velocity correlation of the center of mass, which was ∼10 τ, while that for the cargo was ∼108 τ. Thus, for the characteristic time scale tc ∼L2/2DKIF1A ∼106 τ of the KIF1A head to find the adjacent binding site (z = L or -L) from the dissociation, an amino acid and the KIF1A head behave as overdamped, while the cargo motion is underdamped.
|
10.1371/journal.pcbi.0030226 | Molecular Basis for Evolving Modularity in the Yeast Protein Interaction Network | Scale-free networks are generically defined by a power-law distribution of node connectivities. Vastly different graph topologies fit this law, ranging from the assortative, with frequent similar-degree node connections, to a modular structure. Using a metric to determine the extent of modularity, we examined the yeast protein network and found it to be significantly self-dissimilar. By orthologous node categorization, we established the evolutionary trend in the network, from an “emerging” assortative network to a present-day modular topology. The evolving topology fits a generic connectivity distribution but with a progressive enrichment in intramodule hubs that avoid each other. Primeval tolerance to random node failure is shown to evolve toward resilience to hub failure, thus removing the fragility often ascribed to scale-free networks. This trend is algorithmically reproduced by adopting a connectivity accretion law that disfavors like-degree connections for large-degree nodes. The selective advantage of this trend relates to the need to prevent a failed hub from inducing failure in an adjacent hub. The molecular basis for the evolutionary trend is likely rooted in the high-entropy penalty entailed in the association of two intramodular hubs.
| The protein interaction network or interactome emerged as a powerful descriptor in the large-scale phenotypic studies of the post-genomic era. A major concern in such analysis is the integration of interactomic information with other phenotypic descriptors such as expression profile, co-localization, developmental phase, and large-scale protein–structure data. The latter aspect of the integration is the focus of this contribution. We investigate the molecular basis of network robustness to node failure in the most thoroughly characterized interactome, the yeast network. Node failure is by no means a random occurrence across the network as often claimed, but likely to arise in the node-proteins which are structurally the most vulnerable, that is, the ones most prone to misfolding and to form aberrant associations, including aggregates. Thus, network robustness mandates that such nodes not be directly connected, as failure in one hub is likely to induce failure in an adjacent hub. This observation led us to investigate the molecular basis for the avoidance of connections between highly central proteins and to delineate the graph topology resulting thereof. We show how this topology arose in present-day networks and how it differs from the more generic emerging topology of the ancestral network.
| Scale-free networks have been proposed as universal models to describe diverse complex systems such as the Internet, social interactions, and metabolic and proteomic networks [1,2]. The scale-free “topology” is defined by a power-law distribution: A(n) ∝ n−γ, where A(n) is the abundance of n-degree nodes and γ is a positive exponent. It has been recently noted that such a generic definition does not determine a unique graph topology [3,4]. Rather, topologies ranging from the assortative [3,5], with frequent like-degree node connections, to the highly dis-assortative [5], with like-degree nodes avoiding each other, may fit the same connectivity scaling law [3]. In a purely operational sense, a highly self-dissimilar network is hereby regarded as modular in the sense that high-degree nodes tend to avoid each other [6], and, thus, highly interconnected regions are loosely connected to each other. The definition hinges on the assumption that highly interconnected regions are organized around hubs (the nodes with high degree of connectivity) which would be then characterized as intramodular [3,4].
To determine the graph topology of the yeast protein network [6–10] beyond the power-law distribution and its evolution from a primeval network, we make use of a metric indicative of the degree of graph modularity [3]. The metric is informative of network structure because it increases with the frequency of like-degree connections, and decreases as the graph topology approaches a modular organization in the sense defined above. It should be noted that there is no inherent contradiction in having a scale-free network endowed with a modular topology that reflects a self-dissimilar or dis-assortive structure, since the characterization of scale-free network is solely based on degree distribution [3,6,9].
We found that the present-day network is actually a self-dissimilar graph, most often linking nodes of dissimilar degrees, thus revealing a marked avoidance of intramodular hub connections in accordance with previous observations [6]. By contrast, ancestors of the network obtained through orthologous categorization of the yeast open reading frames (ORFs) [8] are progressively more assortative as we regress toward the network of ancient proteins. The assortative topology brings the ancient network closer to a physical system, where assortativity becomes a generic attribute of the statistical mechanics of phase transitions, and thus an emerging property more readily attainable than modularity [11].
The robustness of the present-day network is found to differ from typical scale-free attributes, since it minimizes its vulnerability to hub failure and not to random node failure [2], with the former being more likely in protein interaction networks, as shown below. The evolution toward self-dissimilarity is shown to be reproducible through propagation laws of connectivity accretion that promote progressive increase in modularity. Finally, the molecular basis for the observed trend toward a scarcity of like-degree node connections is delineated.
The metric S(G) (0 ≤ S(G) ≤ 1), for a graph G with scale-free degree distribution is defined by [3]:
where E(G) is the set of graph edges, (i, j) is a generic edge linking nodes i and j, Xi, Xj are the respective node degrees (connectivities), and smax(G) is the maximum over all s(H)-values, where H is a graph with the same connectivity distribution as G obtained by connectivity rewiring. This distribution-preserving rewiring is constructed following [3,6].
For a given scaling degree distribution, the metric is informative of the graph structure, reaching its maximum value (S(G) = 1) in the case where edges are most frequently connecting similar-degree nodes and decreases as the frequency of dissimilar-degree connections increases [3,6]. Thus, a low S(G)-value is indicative of graph modularity in the sense defined above, because the expected frequency of hub–hub connections is low and because connections involving hubs are always dominant contributors to the sum defining S(G) (Equation 1).
Using this metric, we determined the modularity along the natural evolution of the yeast protein interaction network. Node ancestry classes are defined through orthologous representativity in other genomes informative of the yeast evolution (Methods). Ancestry classes are labeled using binary vectors [8] and defined based on the existence of orthologs in other fungi (00011) (36% of yeast proteome), in all other eukaryotes diverging earlier than fungi (00111) (19%), in eubacteria (01111) (9.5%), in archaea but not in eubacteria (10111) (8%), in all ancestral groups (11111) (3.5%), and exclusively in yeast (00001) (24%). Thus, a binary vector denotes an ancestry class of proteins. The ancestry is given by the extent of ortholog representativity. Thus, the binary vector indicates from the right entry (yeast) to the left (progressively more distant life domains) the ortholog representativity of the proteins, with nth entry = 1 if an ortholog of the protein exists in life domain n, and = 0 otherwise. Thus, the network evolution from the ancient-protein (11111) network is retraced by trimming the present-day network through progressive removal of ancestry classes, starting with the most recent (00001). Although the network still contains false-positive and false-negative data in spite of state-of-the-art curation (Methods), the impact of these factors is likely randomly distributed across classes [8] and thus will not significantly affect our conclusions.
The trimming of the present-day network following the schedule imposed by ancestry is based on the assumption that a gene arising at a certain point in evolutionary time in an ancestral organism will be detectable in all species diverging thereafter. The ancestry of a yeast protein is thus defined by the number of orthologous ORFs [8,12]. Thus, no effort is placed in our study in reconstructing the ancestral sequence, a daunting task at the proteomic scale, but rather in assessing its ancestry by genomic comparison. Gene loss or interaction loss due to deleterious evolutionary pressure is possible after speciation, although very difficult to assess and typically neglected in related evolutionary models [8,12].
The present-day and ancestral networks all fit the scale-free connectivity scaling (Figure 1A). However, their graph topologies are radically different. The ancient protein network possesses a high probability of connection between similar-degree nodes, as indicated by the large S(G)-value, and thus, it is significantly scale-free and assortative. This topology evolved into the scale-rich self-dissimilar graph (S(G) = 0.32) found at the present time (Figure 1B). In contrast with its ancestors, the present-time network tends to connect higher-degree nodes to lower-degree ones, as revealed by the low S(G)-value. Thus, while the ancestral network is actually endowed with the “emergent” properties commonly ascribed to scale-freeness [1,2], such as robustness to random failure, assortativity, and hub-like core, the present-day network is far less generic, more modular [9], and more robust to hub failures. This is evidenced by the dearth of inter-hub edges subsumed in its lower S(G)-value. The selective advantage of this trend relates to the need to prevent a failed hub from inducing failure in an adjacent hub, as shown below.
There are 319 nodes with a present-day degree X > 8 incorporated along the evolution of the network that starts at the ancient network (cf. [8]). All such nodes may be characterized as intramodular hubs [13] that avoid each other and make up for the increased level of scale-freeness in the network topology (Figure 1B). The molecular basis for this like-degree avoidance is described below.
We tested the sensitivity of the results to persistent noise in interactomic data (see Methods for curation details). Thus, in Figure 1B, we contrasted the previously reported behavior of the scale-free metric against the results from progressive trimming of a comprehensive interactome of protein complexes in which ephemeral interactions and high-throughput artifacts have been filtered out [14]. The S-values differ by less than 9% along the entire evolutionary span. Furthermore, the trend toward higher modularity (lower S-value) appears to be commensurate with organismal complexity (Figure 1B), as we incorporate the S(G)-values calculated for the interactomes of Caernohabditis elegans (worm) [15] and drosophila (fruit fly) [16].
The dynamics of node removal associated to the evolutionary regression is indicated in Figure 1C, where the percentage of node removal associated with each of the four successive trimming iterations is computed for each node connectivity class in the present-day network. The node removal becomes more severe for the nodes of low connectivity and less pronounced as we approach a higher degree of centrality, in accord with the likely higher level of ancestry of high-degree nodes [17].
The trend toward increasing modularity associated with evolutionary change was further validated by disproving the null hypothesis that this trend holds irrespective of network topology. Thus, in several computer experiments (cf. [3,6]) we randomly rewired the present-day network while preserving the present-day node-degree distribution indicated in Figure 1A. We then successively trimmed the rewired networks following the orthologous classification scheme and computed S(G)-values corresponding to the successive trimmings. The results are shown in Figure 1D. We clearly see that the monotonic and dramatic increase in modularity observed for the real yeast network along the ancient → present-day evolution is not a generic network property, but very much depends on the specifics of the network topology that subsume the biological information. Alternatively, we also randomly rewired the present-day network this time without preserving the degree distribution and randomly and successively trimmed it, removing an equal number of nodes as in the orthologous classification procedure. Again, no trend toward decreasing modularity could be associated with the trimming or, conversely, no clear trend toward increasing modularity is found upon network growth.
An alternative indicator of modularity put forth by Newman [18] has been also utilized to better describe the evolutionary trend. Newman's approach not only provides a measure of topological dissimilarity but also identifies or separates the dominant or tightest module, and ultimately—through iteration of the separation procedure—provides a modular partition of the network. The initial modular partition of the network is dictated by the spectrum of a symmetric graph-related matrix. Thus, the dominant moduleM℘ is associated with the largest positive eigenvalue, λ1, of the symmetric matrix B defined as:
where A is the adjacency matrix describing the edge set E(G) (Aij = 1 if nodes i and j are connected, Aij = 0 otherwise) and m = ½ΣjXj is total number of edges in the network. The dominant moduleM℘ is univocally defined by the characteristic function χM℘(j) = ½(sj(u1)+1), where u1 is the eigenvector of B associated with λ1 and sj(u1) = 1 if the j-th coordinate of u1 is positive and = −1 otherwise. In set-theory notation: χM℘−1({1}) =M℘. This constructive procedure reveals the most densely connected group of nodes with only sparser connections to the rest of the graph and may be further iterated on G\M℘, etc., until a full modular partition of G is achieved. A similar definition of the module is provided in [10].
A modularity parameter Q is then defined as an indicator of the number of nodes falling within modules minus the expected number for a random rewiring of the network, normalized to the total number of nodes in the network. Thus, Q is given by:
where the dummy index n ranges over all eigenvalues, unT is the transposed eigenvector of B associated with eigenvalue λn, and s = (sj(u1)).
The trend toward increasing modularity associated with evolutionary change in the yeast network evolution is then verified adopting the Q-measure, as shown in Figure 1E: in the ancient network, 39% of the nodes were contained in a module and this number increases to 54% in the present-day network. The dominant module in the ancient network comprises all its 19 ribosomal proteins (see also Protocol S1). This network prevails until class 00111 is incorporated, at which time the signaling module dominates and prevails as dominant in the present-day topology.
The topological differentiation resulting from connectivity accretion concurrent with progressive incorporation of node classes in the order (11111) → (01111) → (00111) → (00011) → (00001) may be algorithmically reproduced. Thus, the primeval network of ancient nodes–proteins may be abstractly developed, i.e., without reference to concrete molecular features of the node, in a manner entirely consistent with the S(G) behavior shown in Figure 1B.
The algorithmic behavior of network evolution is determined by the probability P(Xn) = G(n)p(Xn) that node n with degree Xn would acquire a new connection. The p-factor is associated with the rate of connectivity development, while G penalizes like-degree connections that would increase assortativity. The p-factor relates to a preferential attachment law [1,17] in the sense that the probability that a node develops a new connection depends on the number of its pre-existing connections, satisfying:
Two accretion laws have been investigated. While heuristic in nature, their accurate reproduction of the evolving network topology makes them worthy of examination:
Both laws have optimized parameters (Figure 2) and satisfy the limit Equation 4.
To prevent similar-degree node connections, nodes are “tagged for kinship” at every stage of network propagation taking into account the order assigned at that stage. This order is obtained by preserving the order arbitrarily assigned in the primeval network while incorporating new nodes in consecutive order.
To define the accretion rules algorithmically, let n1 < n2 < … be an ordered set of nodes at a specific time in the network development; Gn denote the n-centered subgraph, that is, a subgraph containing node n, all nodes connected to n, and the connecting edges; C(n) = {nodes connected to n}; and {Gn} is a minimal covering of G satisfying G = ∪nGn. Then, we may define ξn = Minimumn′∈C(n) |Xn − Xn′|. Node n is “tagged for kinship” with probability exp(−ξn) provided no node n′ ∈ C(n) with n′ < n has been tagged for kinship. A node n tagged for kinship at a particular stage of network development is assigned the kinship penalty factor
In case of close kinship (ξn = 0), we get G(n) = 0. The creation of an internal connection linking node n with another node already tagged to develop a connection is governed by probability
where Ln = Maximumn′∈A(G) |Xn − Xn′|, and A(G) = nodes tagged to develop a connection at the particular stage of network development. If node n is tagged to develop a connection, and an internal connection develops, then the new edge connects n to existing node n*, with the latter satisfying: n*∈A(G); Ln= |Xn−Xn*|.
The algorithmic network development that best fits natural evolution (Figure 2) is given by accretion law (I) modulated by precluding kinship connections according to Equations 4 and 5. While law (II) also produces a good fit, it does not portray the sigmoidal behavior of S(G) followed by natural evolution. Network development with an accretion law reflecting preferential attachment (G(n) ≡ 1, law (I)) does not significantly increase its self-dissimilarity relative to the differentiating algorithms that enhance modularity.
What sort of selective advantage is associated with evolving toward higher self-dissimilarity or dis-assortativity? We shall show that this trend increases resilience to node failure which is not random, contrary to general assumption [2]. We first note that node failure may result from a loss of the functionally competent structure in favor of a misfolded state. The latter tends to aggregate into a generic aberrant state dominated by the backbone generic information, rather than by the side-chain information that encodes for the native state [19,20]. We cannot assert that misfolding is the sole reason for node failure but it certainly appears to be the dominant one in the light of the results presented below.
Soluble proteins with high levels of backbone exposure are prone to aberrant aggregation [20], and thus likely to “fail” since they would be removed from their normal interactive context by relinquishing their native fold. Since, as shown in Figure 3A, intramodular hubs possess a higher extent of backbone exposure in their native soluble structure (the extreme case of this exposure is represented by native disorder) [16,20,21], we may conclude that failure propensity likely correlates with centrality, at least in intramodular organization.
This finding prompts us to ask the question: Why would the avoidance of hub–hub connections bring about resilience to hub failure? Since hubs are characterized by their extent of backbone exposure, they are highly reliant on binding partnerships to preserve their structural integrity [16]. Thus, by distorting its protein–protein interface, a misfolded binding partner is likely to promote the hub failure. Hence, to prevent a failed hub from inducing failure in another hub, it becomes necessary to minimize the probability that the binding partner of a hub is also a hub. This is precisely the trend reported in Figure 1B.
Thus, we showed that, unlike robustness to random failure, present-day resilience to hub failure is a non-emergent evolutionary trend achieved by enhancing the dis-assortativity of the graph under the generic scale-free degree distribution (Figure 1A and 1B). Hence, the widespread notion that scale-free networks are vulnerable in this sense does not hold in this particular case.
The lower level of connectivity among nodes of similar degree in the present-day network [6] has a molecular basis that may be delineated and prompts us to invoke conformational entropy penalties. As indicated previously, there are 319 present-day hubs incorporated along the evolution of the network. Of such nodes, 37 are represented in PDB complexes (Protocol S1) and shown to contain an extent of backbone exposure in over 50% of the molecule (Methods). Typically, high intramodular centrality implies that protein associations entail considerable induced fit, since the extent of backbone exposure of such hub proteins is significant and thus so is their conformational plasticity [16,21]. To quantify this trend, we established a correlation between present-day connectivity and extent of backbone exposure on PDB-reported proteins incorporated to the ancient network (Figure 3A, Pearson correlation coefficient r = 0.78). This class of nodes is the complement in yeast proteome of class (11111), and thus it is denoted “\(11111)”. We now examine the molecular characteristics of the associations involving proteins in class \(11111), that is, in the complement of the set of oldest proteins, or in the set of proteins incorporated to the ancestral network. This analysis is needed to rationalize the topological difference between the ancient and present-day network.
Induced fit entails a considerable entropic cost associated with the structural adaptation, decreasing the stability of the protein complexes [19]. Thus, induced fits form in the ephemeral complexes typically found in signal-transduction events. On the other hand, a prohibitively high entropic cost would make it unlikely that protein associations would occur if both partners must undergo induced fit. This is reflected in the probability distribution f(Y, Y′) of binding partnerships between pairs of proteins in class \(11111) with backbone exposures Y and Y′ (f(Y, Y′)dY′ = probability of connections between proteins with backbone exposure Y and proteins in the range [Y′, Y′ + dY′]). Proteins with high backbone exposure typically associate with those with low backbone exposure, in an anticorrelated manner (Figure 3B and 3C). Thus, direct comparison of Figures 2 and 3B–3C reveals that high degree nodes in class \(11111) are unlikely to connect with nodes of comparable degree because of the high entropic cost associated with two concurrent induced fits. This anticorrelation (Pearson coefficient r = −0.69) provides a molecular basis for the modularity and self-dissimiliarity of the present-day network.
To extend the validity of the anticorrelation to the full class \(11111), we also adopted a sequence-based predictor of backbone exposure, taking advantage of a tight correlation [16] between extent of backbone exposure and native disorder content, and of the fact that the latter may be predicted directly from sequence [21] (Methods). As backbone exposure in hubs from class \(11111) increases to accommodate interaction partnerships in the evolving network (Figure 3A), their likelihood of mutual interaction decreases. This trend is reflected in the present-day Y-Y′ anticorrelation (r = −0.72) for class \(11111), which evolved from a Y-Y′ correlation (r = +0.66) in the ancient network (Figure 4). This qualitative change reflects the increasing entropy cost of the reciprocal induced fits required to establish hub–hub associations in the proteins incorporated to the ancient class. Thus, the qualitative evolutionary change described at the molecular level (Figure 4) fits the network's seemingly algorithmic progression toward modularity.
Using a metric to quantify the extent of modularity, we examined the evolution of the yeast protein network and found significant topological differences along evolutionary time that reflect a considerable increase in modularity concurrent with evolutionary change. Thus, aided by orthologous node categorization to trace network evolution [8], we established a trend from an “emerging” assortative network [5] to the present-day modular topology [3]. This evolution implies a progressive enrichment in intramodular hubs that avoid each other (cf. [6]), thus increasing resilience to hub failure. This trend is algorithmically reproducible through a network-growth law that disfavors like-degree connections.
The molecular basis for the evolutionary trend toward higher modularity is rooted in the high-entropy cost of the reciprocal induced fits arising from the association of any two intramodular hubs, an event likely to entail structural adaptation in both proteins. Thus, the avoidance of like-degree of nodes of high connectivity is directly related to the extent of backbone exposure and conformational plasticity of hubs, making it entropically costly for them to adapt to binding partners.
This molecular justification of modularity may be complemented by an evolutionary observation. As shown in [8], proteins tend to interact with partners with the same level of ancestry more frequently than with those outside their ancestry class. Thus, the probability that an ancient hub from class (11111) interacts with another hub from the same class is higher than the probability that it would interact with a more recent hub. This effect may in part account for the higher assortativity of the primeval network and for the evolutionary trend toward higher modularity reported in this work. However, a countereffect is also apparent since, by the same token, the probability that a hub from class (11111) interacts with a low-degree node in the same class is also higher than the probability that it interacts with a low-degree node from a more recent class. The relative contribution of each effect is actually subsumed in the computation of evolving modularity reported in this work.
In an alternative molecular approach [22], it was proposed that the number of interactions of a protein is proportional to the number of exposed hydrophobic residues on its surface. This finding would imply that hubs would need to be so hydrophobic that they would hardly qualify as soluble proteins or they would need to be enormous to accommodate all of their binding partners. Furthermore, if this were the case, hub–hub connections would be highly favored through hydrophobic associations, while in known networks this is clearly not the case [6]. Rather, the structural or molecular characteristic of intramodular hubs [17,21] and the attribute that enables them to avoid each other in the network is their likelihood of conformational plasticity and—in the extreme case—native disorder, as demonstrated in this work.
Lacking expression, localization, and developmental coordinates, the protein interaction network provides an incomplete large-scale description of protein–protein associations. Such a study would likely require integration of the interactome and the transcriptome. Thus, the avoidance of like-degree hub connections shown in this work may often materialize in a lack of spatial or temporal correlation between the nodes, a subject of forthcoming work.
Ancestors of the present-day yeast network were obtained by progressive trimming realized through exclusion of node ancestry classes [8]. Node ancestry classes were determined based on across-species ortholog grouping of yeast proteins. Thus, the primeval network is restricted to nodes with orthologs in all domains of life, while the present-day network incorporates all yeast proteins regardless of their level of ancestry. In a preliminary network curation, connections in the present-day network were only included if independently identified in two sources: Comprehensive Yeast Genome Database from the Munich Information Center of Protein Sequences (http://mips.gsf.de/proj/yeast/CYGD/db/index.html) [23], and reliable subsets of high-throughput screening data [24]. In a second level of curation, the data collected was cross-validated using the APID database that integrates five different repositories for protein interactions including more up-to-date two-hybrid high-throughput data [25]. Finally, the interactomic data was filtered through iPfam representativity (homologous PDB interactivity) [26]. We used iPfam as a database of structurally reported interactions and mapped all interacting Pfam domains onto yeast ORFs using the HMM (hidden Markov model)-profile based mapping available from the Pfam MySQL database. We then retained only the interactions between two ORFs whenever both ORFs contained Pfam domains that are seen to interact in iPfam. The resulting dataset comprises an intersection of iPfam and the APID-curated interactome. The annotation with Pfam domains entails a substantial filtering (from 14,437 APID-based interactions to 6,971 interactions) and hence represents a high-confidence network.
Orthologous classification and grouping of the annotated yeast ORFs (http://www.yeastgenome.org/) were determined from the clusters of ortholog groups [27]. Network representations were performed using standard routines from the program PAJEK [28].
Backbone exposure for node n, denoted Yn, is given as a percentage of contour length of the protein corresponding to under-protected residues, as defined below. The data were obtained from 488 yeast proteins (out of 6,199) reported in PDB complexes and four natively disordered yeast proteins [21]. The extent of backbone exposure at a particular residue was determined by counting the number of nonpolar carbonaceous side-chain groups contained within a 6.2 Å radius sphere (∼thickness of three water layers) centered at the α-carbon [17]. The extent of backbone shielding, η, within a structured region averaged over a nonredundant curated PDB database (1,662 proteins, free from redundancy and homology) is η = 14.2, with Gaussian dispersion = 7.2. Thus, a residue or backbone site with η < 7 is regarded as exposed. The statistics vary as other desolvation radii in the range 6Å < r < 7Å are adopted, but the tails of the distribution identify the same exposed residues. The structural integrity of soluble proteins requires that most backbone amides and carbonyls be protected from hydration. Thus, residues with absent backbone coordinates in a PDB entry (natively disordered [21,29]) are regarded as exposed and so are residues from entirely disordered proteins.
We adopt an established relationship between backbone exposure, η, and a structural parameter, λD, that can be reliably determined from sequence: the propensity for inherent structural disorder in a region of a protein domain [17,29]. The latter parameter is assessed with a high degree of accuracy by the program PONDR-VLXT, a neural-network predictor of native disorder [29]. Thus, a disorder score λD (0 ≤ λD ≤ 1) is assigned to each residue within a sliding window. This value represents the predicted propensity of the residue to be in a disordered region (λD = 1 indicates full certainty). Only 6% of >1,100 nonhomologous PDB proteins give false positive predictions of disorder [17,29]. The correlation between propensity for disorder and wrapping implies that it is possible to predict backbone exposure directly from sequence. The correlation was originally established between the PONDR-VLXT score at a particular residue site and the extent of intramolecular protection, ρ, of the backbone hydrogen bond engaging that residue (if any). The latter quantity is operationally defined as ρ = η + η′, where η and η′ correspond to the two residues paired by the hydrogen bond. The strong correlation implies that we can infer the existence of residues with backbone exposure from the PONDR-VLXT score with 94% accuracy for regions with λD > 0.35. The correlation implies that the propensity to adopt a natively disordered state becomes pronounced for proteins that, because of their chain composition, cannot fulfill a minimal protection of their backbone hydrogen bonds.
The SwissProt (http://www.pir.uniprot.org/) numbers for the following yeast proteins/domains are in parentheses: SH3 Domain (P32790), Cytochrome c (Q753F4), Actin (P60010), Myosin V (Q04439), Ubiquitin (P61864), Calmodulin (P06787) and Rad14(P28519). |
10.1371/journal.pcbi.1005901 | A master equation approach to actin polymerization applied to endocytosis in yeast | We present a Master Equation approach to calculating polymerization dynamics and force generation by branched actin networks at membranes. The method treats the time evolution of the F-actin distribution in three dimensions, with branching included as a directional spreading term. It is validated by comparison with stochastic simulations of force generation by actin polymerization at obstacles coated with actin “nucleation promoting factors” (NPFs). The method is then used to treat the dynamics of actin polymerization and force generation during endocytosis in yeast, using a model in which NPFs form a ring around the endocytic site, centered by a spot of molecules attaching the actin network strongly to the membrane. We find that a spontaneous actin filament nucleation mechanism is required for adequate forces to drive the process, that partial inhibition of branching and polymerization lead to different characteristic responses, and that a limited range of polymerization-rate values provide effective invagination and obtain correct predictions for the effects of mutations in the active regions of the NPFs.
| Endocytosis is a dynamic process by which cells internalize substances from outside the cell. Especially in yeast, endocytosis is mechanically demanding due to the high pressure difference across the cell membrane, or turgor pressure. Polymerization of a branched actin network is the major process providing the mechanical force to overcome the turgor pressure. Understanding the kinetics of the actin network, and the mechanical interaction between the actin network and the cell membrane, is thus crucial for the study of endocytosis. We develop an efficient mathematical framework for actin dynamics that can realistically incorporate these two features, thus providing a practical method for quantitatively modeling actin dynamics during endocytosis. The resulting model mechanistically reveals that spontaneous nucleation at the center of the endocytic site is required for successful endocytosis, distinguishes the roles of branching and polymerization, and predicts several other experimentally testable outcomes. The accuracy and efficiency of the method, in describing both mechanics and chemistry, render it applicable to a broad field of membrane-bending processes.
| Forces exerted by polymerization of monomeric actin (G-actin) into filamentous actin (F-actin) are crucial for bending the cell membrane in many important cellular processes, including cytokinesis, cell migration, and, under some conditions, endocytosis [1]. Specifically, actin is required for yeast clathrin-mediated endocytosis (CME), a central mechanism that controls cellular signaling, nutrient uptake and membrane recycling [2]. CME is driven by a transient protein patch, in which different proteins appear in a well-defined sequence [3], including actin and its nucleators. The actin patch bends a small portion of the cell membrane into a highly curved invagination that encloses extracellular substances. The invagination is later severed and its contents, as well as lipids and membrane proteins, are released into the cytoplasm.
For this membrane-bending process, the actin network needs to exert both pulling forces and pushing forces (see Fig 1). The required pushing forces are several pN per filament [4, 5], mainly to overcome the large (∼0.2MPa) osmotic pressure difference [6] (turgor pressure) across the membrane, because of the small number (∼102) [7] of actin filaments at each endocytic site. The machinery driving CME constitutes a coupled mechanochemical network [8]. Force regulates protein dynamics via processes such as the slowing of actin polymerization by opposing force; conversely polymerization of actin and assembly of curvature-generating proteins generate force. We are only beginning to understand the basic properties of this network.
Protein dynamics during CME have been extensively studied via fluorescence imaging methods. Assembly of endocytic proteins (EPs), including F-actin, was first quantified [3, 9] using relative fluorescence intensities. Later, Ref. [10] developed a method for measurement of the absolute counts of the EPs in fission yeast (Schizosaccharomyces pombe). In Ref. [7], and later in Ref. [11], absolute counts were measured in budding yeast (Saccharomyces cerevisiae). These studies have suggested a count of about 6000 polymerized-actin subunits at the endocytic site [2], with the counts of other proteins typically in the range of 50 to 300. Actin nucleators, or “NPFs”, precede actin polymerization, which proceeds over a period of about 15 seconds. These quantitative measurements have inspired several quantitative modeling studies of dynamics of the EPs [4, 8, 12, 13].
The mechanical aspects of CME are less well understood due to the difficulties of measuring forces on a scale of tens of nanometers in vivo. Balance of forces on the actin network requires that inward pulling forces at the center of the endocytic site are opposed by equal pushing forces from the outer regions of the actin network [4]. The mechanics of bending the cell membrane in CME were studied in detail in a recent model [5] based on the “Helfrich” free energy density [14]. The authors calculated the energy-minimizing shape functions of the membrane during endocytosis, using parameters fitted to electron microscopy tomography data [15]. However, the dynamics of the actin force were not included when obtaining the shape functions, so it was not possible to calculate a time-dependent shape nor to include the mechanochemical feedbacks driving the protein and shape dynamics. In Ref. [4], one of us treated actin as an actively growing gel simulated using a finite element method (FEM), and thus predicted endocytic invagination dynamics. However, the actin growth was modeled with a simple phenomenological description, which was not quantitatively compared to experiment. It is thus unclear how well the growing gel represents the actual dendritic structure of the network [16]. The mechanics of stationary CME membrane profiles were also treated in Refs. [17] and [18], but again neither treated the protein dynamics in the process.
A mechanochemical model of CME was proposed in Ref. [8]. It contained several types of feedback interactions, which indirectly impacted actin polymerization. This model explained several traits of endocytic mutants. However, the treatment of actin polymerization was highly simplified, and the model used an extremely low value of the turgor pressure.
A major challenge in developing a complete description of the mechanochemical network driving CME is to accurately model the actin network and its interaction with the cell membrane. In Ref. [12], we and others developed a stochastic model of actin polymerization during CME. The F-actin in the model was modeled via a stochastic-growth method that gave an explicit three-dimensional actin network, with parameters fitted to experimental data. The force opposing actin polymerization was assumed to “kick in” when the network reached a certain size. This work revealed some important feedback mechanisms between actin and its nucleators, required for CME. But the membrane mechanics were oversimplifed by using a step-function force opposing actin polymerization, and the membrane profile was not obtained explicitly. Additionally, this stochastic model required a large ensemble of repeated calculations to produce meaningful results. It was thus difficult to quantitatively fit the model, for use in other potential applications. Furthermore, stochastic models are difficult to use in studying possible oscillatory behaviors and bifurcations of the actin network that require the recognition of subtle changes in the F-actin count.
Deterministic rate-equation approaches, as in Ref. [19] and parts of Ref. [12], would thus be more convenient for experimental fitting and capturing subtle effects. However, such rate equations cannot treat the mechanics and geometry of the actin network because they describe only the average F-actin count rather than the spatial distribution of F-actin. A number of deterministic reaction-diffusion approaches have improved on rate-equation approaches by modeling the spatial distribution of F-actin explicitly, using various assumptions about dynamics and the interaction of F-actin with the cell membrane [20–28]. However, such methods have not explicitly included the oblique branched geometry of the network, and their force-generation component has not been validated by comparison to stochastic-simulation results.
Ref. [29] developed a two-dimensional treatment of the actin network from a spatially dependent rate equation, explicitly treating branching angles. The mesh size was small in comparison with the size of the cell, but still large enough to treat the coarse-grained density of F-actin. This method was shown to give promising results for global cell properties such as migration. However, it is not clear to what extent it can be applied to processes such as CME, which are fully three-dimensional and have significant structure at very small length scales.
In this paper, we propose a Master Equation (ME) method to treat the reciprocal interactions of polymerizing actin and its nucleators with a bending membrane, and apply it to CME. The ME method describes the spatial distribution of F-actin using a single simulation for a given set of parameter values, while having nearly the realism of the stochastic-growth approach implemented with a large numbers of runs. The methodology explicitly includes the branching geometry, and is validated by comparison with stochastic simulations for the case of an actin network pushing an obstacle. It uses a mesh size smaller than the characteristic size of the actin network, to calculate a probability distribution function (pdf) of actin subunits at given points in time and space. It builds on the work of Ref. [29] by treating a three-dimensional geometry, using a smaller mesh size (about 2 nm vs. 100 nm) that allows better treatment of actin-based forces, and calculating the pdf rather than the coarse-grained actin density. It differs from the reaction-diffusion approaches above in its more complete description of both the orientation and length of new branches in three dimensions, in its more accurate treatment of force generation, and in the use of a pdf. We apply the ME method to a mechanochemical model of CME in budding yeast that treats the time courses of F-actin, its nucleator Las17, and the deformation of the membrane. The model integrates the chemical variables F-actin and Las17 (slow), and the membrane shape variables (fast) into one dynamically interacting system, and shows how the actin network bends the cell membrane in real time.
The model accomplishes several important goals: 1) a theory of dendritic actin polymerization that is mechanistically realistic, numerically accurate and computationally efficient, 2) a mechanochemical model of the dynamics of the cell membrane driven by the actin network during CME, and 3) a more accurate model of CME that quantitatively determines several core parameters that were “floating” (not determined by experimental data) in the previous model [12]. The results of the ME model are consistent with experimental data [7, 12] for protein dynamics and the effects of mutations. New predictions from the model include the following: i) a spontaneous nucleation mechanism is required in the central portion of the endocytic site, ii) controlled inhibition of branching and/or polymerization lead to characteristic behaviors of the peak counts of actin and its main nucleator, and iii) a certain range of polymerization rates is required for robust invagination, and correct prediction of the peak counts of actin and its main nucleator in mutants.
Our model treats dendritic actin network growth in the presence of capping, with new filaments created as branches induced by a planar distribution of nucleation-promoting factors (NPFs), as on a membrane or hard substrate. The model describes polymerization in three dimensions, but we introduce it in two dimensions first, for the sake of clarity. The dynamics of the F-actin probability distribution function ρ in the network are treated by a master equation including branching, spontaneous nucleation and severing:
∂ ρ ( x , y , t ) ∂ t = ∫ 0 l ( y , t ) k b r ( x , y + y ′ ) 2 N ( t ) [ ρ ( x - y ′ , y + y ′ , t ) + ρ ( x + y ′ , y + y ′ , t ) ] d y ′ ︸ branching + k n u c ( x , y , t ) ︸ nucleation - k s e v ρ ( x , y , t ) ︸ severing . (1)
Here kbr(x, y) is the branching rate constant, which gives the rate of branching per unit length of F-actin, per molecule of NPF in the membrane. The rate knuc(r, y, t) describes the amount of actin generated per unit time by spontaneous nucleation, while the decay rate ksev is assumed to be controlled by cofilin-driven severing; l(y, t) is the projection onto the y-axis of the length of the filaments added at each time step. The coordinate x is in the plane of the membrane, y is perpendicular to the membrane, and t is the time variable. We use a frame of reference in which the existing actin filaments are stationary, so convective terms are not required. N(t) is the number of molecules of the NPF on the membrane. We assume that the NPFs are uniformly distributed over the membrane, either because they diffuse rapidly in the membrane, or because their initial distribution is uniform. We do not explicitly treat the assembly of curvature-generating proteins. Rather, they are included as in Ref. [5], as a contribution to the forces acting on the membrane.
In the model, we take all the filaments to have either a 45° or −45° angle with respect to the y direction, in line with the oblique alignment generally found in dendritic networks. New filaments instantly polymerize to a final length, whose projection on the y-axis is l(y, t) (the filament length multiplied by cos45 ° = 1 / 2). The length is determined by the force-dependent polymerization rate and the capping rate. We assume that capping (and thus the growth of a filament to its final length) occurs on time scales faster than the evolution of the invagination. The validity of this approximation is discussed below. At a given time, only new filaments that branch from certain F-actin subunits can increase ρ(x, y, t). These subunits are included in the integral in Eq (1). The two branching directions correspond to the x ± y′ and y + y′ terms in the integral in Eq (1). In practice, considering Fig 2 as an example, we use the dimensionless length l ¯ ( y , t ) = l ( y , t ) / ( a / 2 ) normalized by the length of the actin monomer (a) projected onto the y-axis; l ¯ ( y , t ) is thus the number of subunits in a given new filament. We discretize Eq 1 accordingly (see Eq. S4) to describe the discrete spatial distribution of F-actin in the network. Fig 2 illustrates the case l ¯ ( y , t ) = 4 from Eq. S4. This two-dimensional version of the model could be applied to actin networks growing on a long strip of NPF, as in Ref [30].
We extend Eq (1) into three dimensions using a cylindrical coordinate system (see S1 Fig) The spatial coordinates become r, θ and y, and Eq (1) becomes
∂ ρ ( r , y , t ) ∂ t = 1 2 π ∫ 0 l ( y , t ) d y ′ k b r ( r , y + y ′ ) N ( t ) ∫ 0 2 π ρ [ R ( r , θ ) , y + y ′ , t ] d θ - k s e v ρ ( r , y , t ) + k n u c ( r , y , t ) , (2)
where R = y ′ 2 + r 2 - 2 r y ′ cos θ is the radial coordinate of the base of a branch having its end at radius r. Again, we discretize the time and spatial dependence, as described in Eq. S5.
More details of the simulation procedure are described in the Supplementary Material. This azimuthally symmetric ME approach is directly applicable to templated-nucleation experiments of the type described in Ref. [31].
We assess the validity of the ME by examining how it treats a basic mechanochemical problem: an actin network pushing an obstacle that exerts a constant force opposing polymerization. The obstacle is coated with a ring-like NPF region, mimicking the Las17 ring in Ref. [12]; similar results are obtained for other NPF distributions, including a rectangle and a complete circle. In order to focus on the treatment of branched actin network growth, we ignore negative-feedback effects [12] of actin onto the NPFs. We compare ME results with stochastic simulation results. We start the calculation with 400 filaments, each of which has 50 subunits. For each force value, we run calculations for 200 seconds, and for the stochastic case run an ensemble of 100 simulation runs. The results, shown in Fig 3, show that the ME method agrees quantitatively with the stochastic simulations. We find that F-actin count depends linearly on the external force, and that the velocity is independent on the external force. Both findings are also consistent with the previous stochastic study in Ref. [32].
As mentioned above, key approximation of the ME is that filaments are assumed to grow instantaneously to their final lengths l(y, t), controlled by capping and force. This approximation is valid if the time scale of capping is much shorter than the characteristic time over which the actin count varies in the process of interest. The capping rate is on the order of 1s−1 in budding yeast [33], so the approximation should hold reasonably well for processes occurring on time scales of several seconds or more. This holds for the endocytosis system and model studied in Ref. [12], since the time scale of invagination is on the order of ten seconds. In addition, in our previous study, the assumption of instantaneous polymerization/capping gave results very similar to those of the explicit-polymerization methods when the same parameters were used, as shown by the “Four Variable” model in the Supplemental Material of [12]. Thus the range of validity of the ME approach includes the endocytosis model studied below, but otherwise will vary from case to case.
We apply the ME method to CME by solving the actin network dynamics described by Eq (2), while simultaneously calculating membrane shape dynamics using the analysis of Ref. [5]. First, many possible mechanical equilibria corresponding to a range of force values are obtained as in Ref. [5]. These equilibria are given as a shape function for each value of the force. Then, the actin polymerization dynamics are calculated using the ME according to the shape function, and the process is repeated.
We focus on branching induced by the NPF Las17, which is the strongest one in budding yeast. Thus N = Las17 count. Recent superresolution data indicate that the NPF Las17 accumulates in a ring-shaped region, while the protein Sla2, which links F-actin to the cell membrane, accumulates at a spot inside the ring [34]. Therefore we divide the membrane into a ring of pushing forces corresponding to Las17 and a spot of pulling force corresponding to Sla2. The pushing forces are generated by the growth of branched filaments in the network. The pulling forces act on spontaneously nucleated filaments, which are assumed to form a passive layer attached to both Sla2 and the branched network. Sla2, with the membrane attached to it, is thus pulled back with the retrograde flow of the network (see Fig 1), as suggested in Ref. [11].
Actin polymerization also occurs mainly near the membrane [35]. We thus use the following form for the branching rate constant:
kbr(r,y)={ kbrmaxexp[ −(y−yL)22σbr2 ]ifyL<y<yL+ybrandrLin<r<rLout0otherwise (3)
where k b r m a x is the maximum value of kbr, σbr is the width of the branching region (a precise definition is given in the Supplemental Material), ybr is a cutoff imposed for numerical convenience, and r L i n and r L o u t are the inner and outer radii of the Las17 ring. Eq. (S33) gives the corresponding formula for knuc. In practice, we use dimensionless rates k ¯ b r m a x and k ¯ n u c m a x, as in Eqs. S4 and S5. The spatial branching and nucleation functions are shown in Fig 4. Note that even though kbr cuts off sharply at y = yL, the dynamics of Eq 2 will result in a small component of ρ penetrating past yL. This portion of the ρ is used to calculate the pushing force of the actin onto the membrane below.
In our previous work [12], we demonstrated a crucial negative-feedback effect of actin branching on the Las17 count, and here we treat the Las17 dynamics using a similar rate equation:
dN(t)dt=N(t)2[ Nfull−N(t) ]−αNFbr(t), (4)
where Nfull is the maximum possible count of Las17 (from 2-d packing considerations), α is the probability that a branching event will cause Las17 to dissociate from the membrane (thus being inactivated), and
F b r ( t ) = 2 π ∫ y L y L + y b r d y ∫ r L i n r L o u t ( a 2 ) k b r ( r , y ) ρ ( r , y , t ) r d r , (5)
is the number of new branches created per unit time per Las17 molecule. The probability α reflects the strength of binding of the Las17 to the membrane, and a = 2.7 nm is the step size per added subunit. The factor of a / 2 is the projection of a onto the y-axis.
The forces from actin polymerization deform the membrane from one mechanical equilibrium to another, as indicated in Fig 5. The use of equilibrium shape functions is justified, because the kinetics of the actin-membrane system are determined by the slowly varying actin network shape (timescale > 1s), but relaxation of the membrane occurs much faster, at the speed of sound (timescale ∼ 0.001s).
Because the net force on the actin network is exceedingly small [4], the pushing force from the actin network must balance the pulling force, as shown in S2 Fig. The pushing force is generated by network growth in the outer region comprising the Las17 ring, and is thus denoted fout. It is calculated by allowing the actin network to protrude slightly into the membrane, according to Eq 2, and imposing a linear repulsive force between this portion of the actin and the cell membrane (see Supplemental Material for details). The pulling force is exerted in the inner region corresponding to the Sla2 spot, and is thus denoted fin. The balancing forces fin and fout produce a deformation described by the membrane shape function ym[rm(s)], where ym is membrane height and rm is the radial coordinate in the membrane.
During each time step, actin first polymerizes according to Eq 2. The membrane deformation, or invagination depth yI is then determined from the extent of actin polymerization by a procedure implementing a “molecular clutch” based on the amount of F-actin. The possibility of such a mechanism is supported by findings [36] that a clutch transmits forces from the actin cytoskeleton to the extracellular matrix or other cells. A clutch should also be present in CME because there must be a transition in mechanical behavior with increasing F-actin count. When the F-actin count is small, there is very little actin material in the central region of the endocytic patch. Therefore there is almost nothing for the outer filaments, which are moving backwards in retrograde flow, to “grab” onto. This makes it impossible for the growing network to exert a pulling force. On the other hand, when the F-actin count is larger, there is enough material at the center to transmit the force generated by the growing filament to the endocytic coat proteins We implement the clutch as follows. Up to a certain minimum value of the F-actin count, Fmin, yI is taken to vanish; for larger values of F, the actin network is assumed to be completely rigid, and yI is driven by the difference in polymerization rates between the outside and the inside. Details are given in the Supporting Information.
Given yI, the deformation profile is chosen from the pretabulated set (see S3 Fig) as the one with invagination y I p r e closest to yI. The force is chosen by linear interpolation between the force of the profile with y I p r e and that of another profile with y I p r e ± 0 . 1 R Π, so that y I p r e - 0 . 1 R Π < y I < y I p r e + 0 . 1 R Π. The updated shape function and force are used in the next step to determine the branching region, spontaneous-nucleation region and dynamics of actin polymerization (see Fig 5). This approach should describe the dynamics of the membrane deformation well, since the differences between successive membrane shape functions are relatively small.
The initial actin distribution is a ring of filaments represented by ρ ( r , y , 0 ) = ( 1 / 2 π r ) δ ( r - r L i n ) θ ( y - y L ) θ ( y L + l m a x - y ), where r L i n is the inner limit of the Las17 ring. Further, at the beginning of the simulation L(0) = 20, yS = yL = 0, and all forces vanish.
Here we describe our procedure for fitting the model to measured properties of endocytosis in budding yeast. Then we present several experimentally testable predictions: First, a spontaneous nucleation mechanism with a specific spatial location is required for adequate force generation. A substantial fraction of the nucleated filaments must be near the middle of network in order to exert sufficient pulling force to overcome the turgor pressure, being dragged along by the rest of the network as indicated in Fig 1. Second, we quantitatively predict the response of actin and NPF assembly to the drugs CK-666 and Latrunculin A (LatA), which suggests a new direction for quantitative experiments. Third, we constrain the values of key parameters. In our previous model [12], we found that changes in some key parameters can be compensated by changes in other parameters. For instance, a broad range of values of the polymerization rate gave results consistent with experiments; a lower polymerization rate could be compensated by higher branching rate, and vice versa. However, after including the membrane more completely in the new model, we find such compensation to be less effective, limiting the range of parameter values. A polymerization rate within a narrow range is required to sufficiently deform the membrane into “Ω” shapes and to correctly obtain the effects of NPF mutations.
We use the experimentally measured time courses of F-actin (F) and Las17 number (N) [12] as our fitting targets. We use the four quantities (k0, k ¯ b r m a x, ksev, and α) (see Table 1) as our fitting parameters, and regard the rest of parameters as “fixed” (see Table 1 and S1 Table). The values of the fitting parameters are obtained by minimizing the mean-square difference between the measured F and N time courses on one hand, and the model and the experimental data on the other hand:
ϵ=1nN∑i=1nN[ Nmod(ti)−Nexp(ti) ]2[ max(Nexp) ]2+1nF∑i=1nF[ Fmod(ti)−Fexp(ti) ]2[ max(Fexp) ]2, (6)
while keeping the “fixed” parameters unchanged. Here nN and nF are the number of experimental data points of Las17 and F-actin in the time courses. Fexp is obtained from measurements of the time course of Abp1, as in Ref. [12]. At each step of the the fitting process, we randomly vary (k0, k ¯ b r m a x, ksev, α) and calculate ϵ. The new values are accepted if ϵ is lower. The above computation is repeated until ϵ does not decrease despite a large number of attempts (typically about 300). Then the values of the fitting parameters are found for a given set of “fixed” parameters. This process requires a large number of trial calculations (about 300) for one set of the “fixed” parameters, which is nevertheless manageable within the ME method. In the stochastic simulations, one needs to repeat the calculation about 1000 times for the same fitting parameter values to obtain adequate statistics. Thus, a total in the range of 300,000 runs are needed for one set of “fixed” parameters, which is a very demanding computational load. In Ref. [12], we estimated the fitting parameters by first pre-fitting a simplified four-variable rate-equation model to the experimental time courses, and then fitting the stochastic model to experimental maxima and lifetimes starting with the pre-fitted parameter values. This process is less efficient and accurate than the automatic fitting process used here, which is difficult to incorporate into the stochastic model.
There are two additional “fixed” parameters that are estimated via either other experimental data or physical constraints (see Table 1). The zero-force dimensionless filament length l ¯ m a x (the maximal number of subunits) in Table 1 is estimated as l ¯ m a x = k o n G / k c a p, where kon = 11.6μM−1 s−1 [37] is the on-rate constant, G = 5.3μM [33] is the free-actin concentration, and the capping rate kcap is taken to be 1 s−1 [33]. The spontaneous nucleation rate parameter k n u c m a x is fixed by a combination of two constraints: i) that adequate pulling forces can be generated, and ii) that sufficient invagination can be obtained.
Additional parameters, including those describing the geometry, are given in S1 Table.
Fig 6 shows how the actin network invaginates the membrane over time. The membrane forms an ‘Ω’ shaped invagination after 16 seconds into the simulation or about 5 seconds after actin polymerization starts, consistent with observations in electron micrographs [15]. The time courses of the F-actin count F and Las17 count N also reveal a good fit to the experimental data in Ref [12], shown in Fig 7A. Note that the actin distribution extends slightly below the plane of the membrane, and outside the Las17 ring. This occurs because of the nonlocal dynamics of Eq 2. The spreading outside the Las17 ring is physically expected because of the nonzero filament length. The portion below the membrane plane is an approximation used to calculate the pushing force generated by the actin, as described in the Supplemental Material.
The three-dimensional distributions predicted here could be tested by superresolution microscopy methods with resolution on the scale of tens of nanometers. Such methods [34] have found that F-actin forms a hemispherical shape, and Las17 forms a ring, as in the present model. Electron microscopy data in the literature [15, 38, 39] are also consistent with an F-actin hemisphere.
The “acidic” regions of Las17 and other yeast NPFs are believed to control their binding to Arp2/3 complex, and therefore their NPF activity. The mutant containing mutations of both the Las17 and the NPF Pan1, las17 pan1Δacidic (abbreviated as LPΔA) should have a strong reduction in the Las17 branching activity. We choose this mutant to avoid possible compensatory effects from the nucleation-promoting activity of Pan1. As in Ref. [12], we model the mutation via a 40% reduction in k b r m a x. Fig 7B shows that the model matches the measured LPΔA phenotype [7] well, with an accuracy comparable to that of our previous model [12]. Remarkably, the F-actin count is actually increased by reducing kbr. This counter-intuitive phenotype results from a competition between a direct effect and an indirect effect. The direct effect is the reduction in branching rate per molecule of Las17 caused by the mutation. The indirect effect is the resulting increase in Las17 caused by the reduced branching rate, due to the negative-feedback effect described in Eq (4). This increase will tend to increase the F-actin count. For the conditions considered here, the indirect effect outweighs the direct one.
As indicated in Fig 1, actin filaments in the central region are required to exert pulling forces. We assume that these filaments arise from spontaneous nucleation (not requiring NPFs at the endocytic site). Possible sources of the spontaneously nucleated actin filaments can be severed filament fragments [40], or nucleation via Dip1, which is independent of NPFs [41]. A minimum value of k ¯ n u c m a x is required to exert adequate pulling force, since reducing k ¯ n u c m a x reduces the number of filaments in the central region (see Fig 8B–8D). This increases the pulling force per filament, and eventually causes them to detach from the membrane.
The maximum pulling force that a membrane-attached filament can sustain is not known. But Ref. [42] gives a quantitative measurement (> 40pN) of the rupture force between a single actin filament and the crosslinking proteins filamin and α-actinin in vitro, at low loading rates. In red blood cells [43], the interaction force between the actin cytoskeleton and the membrane was found to be ∼10pN per filament in a model fitted to experimental data. However the cytoskeleton-membrane interactions in yeast could be very different from those in red blood cells. We thus based our estimate of the maximum actin filament pulling force on the measured rupture force. To estimate the pulling force per filament in our model, we divided the total maximum pulling force of 725 pN obtained from the membrane-energetics analysis (Supplementary Material) by the number of pulling filaments, estimated as the total F-actin count inside r L i n divided by 2 y n u c / a, a dimensionless measure of the height of the nucleation layer. In the default model, this procedure gave a pulling force of ∼30pN per filament, below the rupture force [42]. When spontaneous nucleation was suppressed by reducing k ¯ n u c m a x, the pulling force exceeded the rupture force, as shown in Fig 8A. Thus a minimum rate of spontaneous filament nucleation in the central region is required for the actin network to pull the membrane without rupture of the actin-membrane interactions.
This prediction might be tested by deletion of the protein Dip1, which could participate in an NPF-independent actin polymerization pathway [41]. If Dip1 nucleates filaments in the central region, its deletion should have two main effects. First, the actin hemisphere should be sparser in the middle. as shown in Fig 8B. This could be verified by superresolution images of dip1Δ cells. Second, in the dip1Δ cells, reduced nucleation should lower the number of pulling filaments (see Fig 8A), causing rupture if the force per filament exceeds the rupture force. This would reduce the efficiency of invagination. Ref. [41] found that in dip1Δ cells, only 40% of the total patches were internalized, compared to ∼90% in WT cells. Internalization of 30% of the patches was also delayed delayed by over 20 seconds. Both of these effects could be due to the reduced number of filaments in the central region.
On the other hand, we find that too large a magnitude of k ¯ n u c m a x also disrupts invagination. We increased k ¯ n u c m a x by a factor of 2 and refitted. In comparison with the default invagination y l m a x = 54 nm, we obtained a maximum invagination y l m a x = 30 nm, which we consider to be a failed event. Therefore, possible values of k ¯ n u c m a x are in a range limited by the constraints of i) adequate pulling force and ii) adequate invagination.
We investigate the responses of the actin-membrane system to the drugs CK-666, which inhibits branching, and Latrunculin A (LatA), which inhibits polymerization and to some extent branching. S4 Fig summarizes the effects on Fmax and y I m a x of a broad range of combinations of parameter changes. Note than in S4 Fig frame D, breaking of pulling filaments bearing very large loads is not included. Inclusion of this effect will lead to failure of invagination at small k ¯ n u c m a x, as discussed above.
The ME enables a “one-shot” approach to stochastic actin network dynamics by treating the pdf.
It represents a statistical average of many ensembles of the same actin network calculated by the equivalent stochastic simulation, which is referred as “the infinite population limit” in Ref. [47]. We feel that the use of a statistical average is legitimate, because endocytosis in yeast is highly stereotypical. The behavior of the actin network typically displays a single mode value [7] (the most likely value) plus a range of fluctuations, instead of having multiple mode values.
The “one-shot” nature of the ME simplifies the process of making modifications and refitting the model to experimental data. For both ME and stochastic methods, calculation of branching processes causes the largest computational load. The ME approach is most efficient when the branching region is small because it scans the physical space for possible branching events. The stochastic simulation, on the other hand, is most efficient when the total number of filaments is small, because it scans the subunits for possible branching sites. Thus, the ME is computationally most powerful for treating dense actin networks with relatively small regions of NPF, especially with spatial symmetry.
The ME approach could be applied to many other problems involving branched network growth, as long as the branching is generated by a flat NPF region at a membrane or surface. For cell migration, the cytoskeleton can sometimes be simplified to a two dimensional network [48] or even a one dimensional network [49] with NPFs close to the cell membrane. The ME method is well suited to this type of problem. Notice that because the ME method is mechanochemical, it could be used to describe mechanical feedback effects on cell migration, if combined with a “Helfrich”-type calculation of the cell membrane forces. The ME method is also well-suited for treating filopodium and lamellipodium geometries where the membrane forces can be calculated straightforwardly from the membrane geometry.
In order to develop a practical method for treating a problem as complex as endocytosis, we have made several simplifying assumptions.
We have developed a computationally efficient master equation (ME) approach for calculating the spatial distribution of F-actin branched networks growing in the presence of mechanical forces. The approach was validated by comparison with stochastic-simulation results. It was then used to develop a mechanochemical model of clathrin-mediated endocytosis in yeast (CME), which treats both the actin network and the cell membrane realistically. The mechanochemical model was used to reveal the time evolution of the actin-membrane system during CME, to quantitatively estimate unknown parameter values, and to predict several important mechanisms in CME that are unseen or omitted in previous models. These predictions provide possible directions for experiments in CME, especially for superresolution microscopy and drug treatments. Beyond CME, the new ME approach provides possible applications to a wide range of problems involving the spatial distribution of branched actin polymerization.
|
10.1371/journal.ppat.1000194 | Recruitment of the Complete hTREX Complex Is Required for Kaposi's Sarcoma–Associated Herpesvirus Intronless mRNA Nuclear Export and Virus Replication | A cellular pre-mRNA undergoes various post-transcriptional processing events, including capping, splicing and polyadenylation prior to nuclear export. Splicing is particularly important for mRNA nuclear export as two distinct multi-protein complexes, known as human TREX (hTREX) and the exon-junction complex (EJC), are recruited to the mRNA in a splicing-dependent manner. In contrast, a number of Kaposi's sarcoma–associated herpesvirus (KSHV) lytic mRNAs lack introns and are exported by the virus-encoded ORF57 protein. Herein we show that ORF57 binds to intronless viral mRNAs and functions to recruit the complete hTREX complex, but not the EJC, in order assemble an export component viral ribonucleoprotein particle (vRNP). The formation of this vRNP is mediated by a direct interaction between ORF57 and the hTREX export adapter protein, Aly. Aly in turn interacts directly with the DEAD-box protein UAP56, which functions as a bridge to recruit the remaining hTREX proteins to the complex. Moreover, we show that a point mutation in ORF57 which disrupts the ORF57-Aly interaction leads to a failure in the ORF57-mediated recruitment of the entire hTREX complex to the intronless viral mRNA and inhibits the mRNAs subsequent nuclear export and virus replication. Furthermore, we have utilised a trans-dominant Aly mutant to prevent the assembly of the complete ORF57-hTREX complex; this results in a vRNP consisting of viral mRNA bound to ORF57, Aly and the nuclear export factor, TAP. Strikingly, although both the export adapter Aly and the export factor TAP were present on the viral mRNP, a dramatic decrease in intronless viral mRNA export and virus replication was observed in the absence of the remaining hTREX components (UAP56 and hTHO-complex). Together, these data provide the first direct evidence that the complete hTREX complex is essential for the export of KSHV intronless mRNAs and infectious virus production.
| Following gene expression in the nucleus, newly transcribed messenger RNA (mRNA) is exported to the cytoplasm, where it is translated into protein. In mammals the vast majority of mRNAs contain introns that must be removed by the spliceosome prior to nuclear export. In addition to excising introns, splicing is also essential for the recruitment of a several protein complexes to mRNA, one example being the human transcription/export complex, which is required for mRNA export. Herpesviruses, such as Kaposi's sarcoma–associated herpesvirus, replicate by hijacking components of the host cells biological machinery, including those proteins necessary for mRNA export. An intriguing caveat in herpesvirology is that herpesviruses, such as Kaposi's sarcoma–associated herpesvirus, produce some mRNAs that lack introns and do not undergo splicing. How then are these intronless mRNAs exported to the cytoplasm? The answer lies in a virus protein called ORF57 that is able to bind to the intronless mRNA and then export them to the cytoplasm. ORF57 achieves this function by mimicking splicing and recruiting the human transcription/export complex to the intronless viral mRNA, thus facilitating its export into the cytoplasm.
| The nuclear export of mRNA composes one part of a larger network of molecular events that begin with transcription of the mRNA in the nucleus and end with its translation and degradation in the cytoplasm. During trafficking to the cytoplasm, a nascent mRNA undergoes numerous co-transcriptional processing steps, including 5′ capping, splicing to remove introns and 3′ polyadenylation [1]–[3]. Of these events it has become clear that splicing is particularly important for mRNA nuclear export [4]. The question of exactly which proteins regulate mRNA nuclear export has been the focus of several recent reviews [5]–[8].
Two distinct multi-protein complexes are recruited to cellular mRNAs as a consequence of splicing, namely the human transcription/export complex (hTREX) and the exon-junction complex (EJC). The hTREX complex contains the proteins Aly (a NXF/TAP-adapter), UAP56 (a RNA-helicase) and the hTHO-complex (a stable complex composed of hHpr1, hTho2, fSAP79, fSAP35 and fSAP24) [9]. A second multi-protein complex, termed the exon-junction complex (EJC) is deposited 20–24 nucleotides upstream of the exon-exon boundary during splicing. Until recently it was believed that Aly and UAP56 were components of the EJC [7], [10]–[12], however, new evidence suggests that Aly and UAP56 are associated exclusively with hTREX and not with the EJC. Therefore, these results suggest that hTREX and EJC are distinct complexes, bind at separate locations on the spliced mRNA [13] and have separate functions, where hTREX directs nuclear export of mRNA and the EJC may instead monitor mRNA fidelity and function during translation [14]–[16].
At present, it is not fully understood what regulates hTREX assembly on the mRNA but in addition to splicing the 5′ cap is also essential for its recruitment [9],[13]. Specifically, an interaction between Aly and the cap-binding complex protein, CBP80 appears to be critical for assembly. Indeed, the 5′ cap has been shown to be required for mRNA export in Xenopus oocytes [13]. In contrast to the EJC which binds near each exon-exon boundary, hTREX is recruited exclusively to the 5′ end of the first exon, presumably regulated in part by the reported interaction between CBP80 and Aly [13]. It has been suggested that localising the export proteins at its 5′ end affords the mRNA polarity when exiting the nuclear pore. Therefore, a current model for mRNA export favours a situation where hTREX is recruited to the 5′ cap of spliced mRNA and once bound Aly stimulates the recruitment of the export factor, TAP. TAP then interacts with p15 and the nucleoporins, providing the connection between the ribonucleoprotein (RNP) and the nuclear pore [17]. The functional roles, if any, played by UAP56 and hTHO-complex in this process remain poorly characterised.
Kaposi's sarcoma-associated herpesvirus (KSHV)/Human herpesvirus 8 (HHV8) is a γ-2 herpesvirus associated with a number of AIDS-related malignancies including Kaposi's Sarcoma (KS), primary effusion lymphoma (PEL) and multicentric Castleman's disease [18]–[21]. In contrast to the majority of mammalian genes, a property shared amongst all herpesviruses is that a proportion of lytically expressed viral genes lack introns. Although, KSHV expresses a higher proportion of spliced genes than other herpesviruses, it still encodes a significant proportion of lytically expressed late structural genes which lack introns. KSHV replicates in the nucleus of the host mammalian cell, and therefore requires its intronless mRNAs to be exported out of the nucleus to allow viral mRNA translation in the cytoplasm. This raises an intriguing question concerning the mechanism by which the viral intronless mRNAs are exported out of the nucleus in the absence of splicing. To circumvent this problem, and to facilitate viral mRNA export, herpesviruses of all subfamilies encode a functionally conserved phosphoprotein which has an essential role in viral lytic replication [22]. In KSHV this protein is encoded by the intron-containing open reading frame 57 (ORF57) and has been the subject of several recent reviews [23]–[26]. The ORF57 gene product interacts with Aly, binds viral mRNA, shuttles between the nucleus and the cytoplasm and promotes the nuclear export of viral mRNA transcripts [27]–[31]. These properties are also conserved in ORF57 homologues such as ICP27 from Herpes simplex virus type-1 (HSV-1), SM protein from Epstein Barr virus (EBV) [32]–[35] and the Herpesvirus saimiri (HVS) ORF57 protein [27], [31], [36]–[38].
Here we show that KSHV ORF57 interacts during viral replication with CBP80 and hTREX, but not the EJC. We further show that ORF57 orchestrates the assembly of hTREX onto an intronless viral mRNA. The ORF57-mediated recruitment of hTREX is achieved via a direct interaction between ORF57 and Aly. Furthermore, in vitro data showed that UAP56 acts as a bridge between Aly and the hTHO-complex protein hHpr1, thereby facilitating the formation of the complete hTREX complex. When we prevented the recruitment of Aly onto intronless viral mRNA using an ORF57 Aly-binding mutant, this resulted in a failure of ORF57-mediated viral mRNA export and significantly reduced virus replication. Strikingly, expression of a dominant negative Aly mutant that prevented the recruitment of UAP56 and hTHO-complex onto intronless viral mRNA resulted in a dramatic reduction in intronless viral mRNA export and infectious virus production. We therefore propose that the entire hTREX complex must be recruited to intronless viral mRNA by ORF57 in order for efficient intronless mRNA nuclear export and KSHV replication to occur.
The hTREX complex contains several nuclear export proteins. Given that KSHV ORF57's primary role is attributed to the nuclear export of intronless viral mRNA, we first assessed if ORF57 interacted with hTREX components using co-immunoprecipitation assays. Moreover, as hTREX forms a complex with the 5′-cap protein CBP80 [13], we were interested if ORF57 also interacted with CBP80. 293T cells were transfected with pGFP or pORF57GFP and untreated or RNase treated total cell lysate was used in co-immunoprecipitation experiments with CBP80-, Aly-, UAP56-, fSAP79- and hHpr1- specific antibodies in addition to an unrelated antibody control (a p53-specific antibody). Each of the hTREX proteins and CBP80 co-precipitated with ORF57, in an RNA-independent manner (Fig. 1A). Moreover, indirect immunofluorescence showed that a proportion of ORF57GFP co-localised with hTREX proteins (Fig. S1).
To assess whether ORF57 also interacts with the EJC, co-immunoprecipitation assays were repeated using an antibody specific for eIF4A3, a core EJC component [39] and a hHpr1-specific antibody, serving as a positive control. No interaction was observed with the EJC core component, eIF4A3, in contrast, ORF57 was readily detectable in the hHpr1 immunoprecipitation (Fig. 1B). A control immunoprecipitation was performed to confirm that the eIF4A3 antibody precipitated EJC components (Y14) in this assay (data not shown).
In order to address potential overexpression artefacts and to assess whether ORF57 interacts with hTREX core components during lytic replication, KSHV-latently infected BCBL-1 cells were reactivated using the phorbol-ester, TPA, and lytic gene expression confirmed by detection of the ORF57 protein in TPA-treated cells by western blot analysis (Fig. 1C (i)). Reactivated BCBL-1 cell lysate remained untreated or was treated with RNase and co-immunoprecipitations performed using an ORF57-specific antibody. Western blot analysis using CBP80- and hHpr1- specific antibodies revealed that ORF57 interacts with CBP80 and hHpr1 during lytic replication, however ORF57 did not precipitate with either eIF4A3 (the EJC core component) or the cellular intronless mRNA-export protein, SRp20 (Fig. 1Cii). Moreover, to confirm that ORF57 failed to interact with additional components of the EJC, co-immunoprecipitations were repeated using reactivated BCBL-1 cell lysates and Y14- and Magoh-specific antibodies. Results demonstrate that ORF57 did not precipitate with these additional EJC components (Fig. 1Ciii). A control immunoprecipitation was also performed to confirm that the Y14- and Magoh-specific antibodies precipitated eIF4A3 in this assay (Fig. S2). Therefore, these data provide the first direct evidence of a viral protein associating with CBP80 and all the core components of the hTREX complex.
One possible explanation for how herpesvirus intronless mRNAs undergo nuclear export is that ORF57 mimics splicing by loading key mRNA export proteins, such as hTREX, onto the intronless viral mRNA. In order to test if intronless KSHV transcripts were associated with hTREX proteins and if ORF57 was necessary for this interaction, RNA-immunoprecipitation (RNA-IP) assays were performed. We chose to perform this assay using 2 intronless KSHV mRNAs, specifically ORF47 and gB. RT-PCR and sequence analysis confirmed that both of these ORFs do not contain introns (data not shown). To perform the RNA-IPs, a vector expressing KSHV ORF47 (a late structural intronless gene) was transfected into 293T cells either alone or in the presence of pORF57GFP. Total cell lysates were then used in immunoprecipitations performed with either CBP80-, Aly-, UAP56- or hHpr1-specific antibodies. RNA-IPs performed on cell extracts transfected with ORF47 alone failed to show an interaction between Aly, UAP56 or hHpr1 and the viral ORF47 mRNA (Fig. 2A). In contrast, extracts from cells transfected with both pORF47 and pORF57GFP displayed a clear interaction between Aly, UAP56 and hHpr1 and the intronless viral ORF47 mRNA (Fig. 2A). CBP80 was found to bind to the intronless ORF47 viral mRNA independently of ORF57 (Fig. 2A). Moreover, this analysis was repeated with a second intronless KSHV mRNA, namely the late structural glycoprotein gB, and similar results were observed (Fig. 2C). These data show that ORF57 is required for the recruitment of core components of hTREX onto intronless viral mRNA.
To determine whether EJC components are recruited to intronless viral transcripts prior to export, RNA-IP assays were also performed using eIF4A3-, Y14- and Magoh-specific antibodies. Results failed to show any interaction between the EJC core components and viral intronless ORF47 and gB mRNAs in the absence or presence of ORF57 (Fig. 2B and 2C). These results show that the EJC is not recruited to intronless viral transcripts by ORF57 and suggests that the EJC is not required for KSHV intronless viral mRNA nuclear export.
To determine whether the hTREX and EJC components were recruited to a spliced viral transcript, RNA-IPs were also performed using a vector expressing the genomic (intron-containing) KSHV ORF50 gene. 293T cells were transfected with pORF50 in the absence or presence of ORF57. Total cell lysates were then used in immunoprecipitations performed with either CBP80-, Aly-, UAP56-, hHpr1-, eIF4A3-, Y14- or Magoh-specific antibodies. Results demonstrated that CBP80, hTREX and EJC components were recruited to the spliced ORF50 mRNA in an ORF57 independent manner (Fig 2D). This suggests that splicing of a viral transcript is sufficient to recruit the cellular proteins necessary for nuclear export. In contrast, ORF57 is required for the recruitment of the hTREX proteins to an intronless viral transcript.
Currently, while it is known that hTREX recruitment to a mammalian mRNA is both 5′-cap- and splicing-dependent, the protein-protein interactions that govern assembly of the hTREX complex itself are not fully understood. As ORF57 functions to recruit hTREX onto the intronless viral mRNA in a splicing independent manner we assessed whether this viral-system could be used to investigate hTREX assembly in more detail. To this end, we sought to determine if any hTREX proteins directly interacted with ORF57. Radio-labelled ORF57 was generated by in vitro coupled transcription/translation (ITT), RNase treated, and used in GST pull-down experiments using constructs expressing GST-, GST-Aly, GST-UAP56 and GST-hHpr1 fusion proteins. Equal amounts of each expressed protein were used in each pulldown experiment (Fig. 3A). Analysis showed that ORF57 bound directly to GST-Aly but not to any other hTREX component (Fig. 3B). Due to the instability of GST-CBP80, a reverse pulldown experiment was performed using GST-ORF57 (Fig. 3C) and radio-labelled ITT CBP80, a GST-Aly pulldown with ITT CBP80 served as a positive control [13]. Results also revealed a direct interaction between CBP80 and KSHV ORF57 (Fig. 3D).
These data suggest that ORF57 only interacts directly with Aly and CBP80, therefore the question remains how the complete hTREX complex associates with ORF57. It has previously been suggested that the hTREX complex is formed by UAP56 bridging the interaction between Aly and the hTHO-complex [9]. Therefore, to further investigate ORF57-hTREX assembly, we assessed which hTREX components were required to reconstitute the ORF57-hHpr1 interaction. GST pulldown experiments were performed using GST-hHpr1 and ITT ORF57 alone or combinations with ITT Aly or recombinant UAP56. When the GST-hHpr1 ITT ORF57 pulldown was repeated in the presence of both ITT Aly and purified UAP56, analysis revealed a clear interaction between hHpr1 and ORF57 (Fig. 3E), suggesting that ORF57 requires both Aly and UAP56 to recruit the hTHO-complex, thus facilitating formation of the ORF57-hTREX complex. These findings provide the first direct evidence that UAP56 functions as a bridge between Aly and the hTHO-complex component hHpr1 to facilitate assembly of hTREX. However, at present we cannot exclude the possibility that ORF57 interacts directly with other hTHO-complex components.
To assess whether hTREX is essential for viral mRNA nuclear export we produced an ORF57 mutant protein which was unable to interact with Aly and as such would be predicted to prevent the recruitment of the complete hTREX complex onto intronless viral mRNA. A minimal region responsible for Aly-binding has been identified in ORF57 and spans 35aa between residues 181 and 215 [28]. Upon closer examination of this sequence, we identified a PxxP-polyproline motif. To assess whether this motif was important for Aly-binding, both proline residues were substituted with alanine residues by site-directed mutagenesis to generate pORF57PmutGFP. To determine if mutating the PxxP-motif in ORF57 led to a loss of Aly binding, GST-Aly pulldown assays were performed using ITT ORF57 or ITT ORF57Pmut. Results demonstrated that the mutant ORF57 protein was unable to interact with GST-Aly, in contrast to the wild type protein (Fig. 4A). Moreover, similar results were observed using pull-down assays with pGFP-, pORF57GFP- or pORF57PmutGFP-transfected 293T cell lysates (Fig. 4B). These data demonstrate that the ORF57 PxxP-motif is required for the direct interaction with Aly. To confirm that the mutagenesis of the PxxP motif had no effect on ORF57 protein stability or other reported functions, several independent experiments were performed to assess the ability of ORF57PmutGFP to localise to nuclear speckles, homodimerise, directly interact with ORF50 and bind viral intronless mRNA (Fig. S3), all of which are features of the wild type ORF57 protein. In each case the ORF57PmutGFP phenotype was indistinguishable from that of wild type ORF57.
Having established that ORF57PmutGFP is unable to interact with Aly and that the mutation does not affect other ORF57 functions, we then asked if, in the absence of Aly-binding, ORF57 was still able to complex with CBP80 and hTREX components. 293T cells were transfected with pGFP, pORF57GFP or pORF57PmutGFP and total cell lysates were used in co-immunoprecipitation experiments, using CBP80-, Aly-, UAP56-, and hHpr1-specific antibodies. In each case the hTREX antibody immunoprecipitated ORF57GFP but not ORF57PmutGFP, demonstrating that in the absence of the Aly-interaction ORF57 was unable to form a complex with hTREX (Fig. 4C). In addition, the ORF57PmutGFP exhibited a reduced but specific binding to CBP80 (Fig. 4C). This reduced binding may be due to the mutation of the PxxP-polyproline motif either affecting CBP80 binding directly or the loss of hTREX binding affects the stability of the CBP80-ORF57 complex. To further investigate whether the mutation of the PxxP-polyproline motif affected direct binding to CBP80, GST pulldown assays were performed using GST-ORF57 and GST-ORF57PmutGFP. Equal amounts of each expressed protein was incubated with radio-labelled ITT CBP80. Results demonstrated that ORF57 and ORF57PmutGFP bound to CBP80 with similar affinity (Fig. S4). This suggests that the reduced binding observed between ORF57PmutGFP and CBP80 may be due to the loss of hTREX, which is possibly required to stabilise the export competent vRNP.
To determine if ORF57PmutGFP was unable to recruit hTREX proteins to KSHV intronless mRNA transcripts in the absence of Aly binding, RNA-IP assays were performed using CBP80-, Aly-, UAP56- or hHpr1-specific antibodies. These data demonstrate that in contrast to pORF57GFP, pORF57PmutGFP is unable to recruit hTREX components to intronless viral mRNA (Fig. 4D). This suggests that a direct interaction between Aly and ORF57 is required for hTREX recruitment onto intronless viral transcripts.
To test if a failure in ORF57-mediated recruitment of hTREX to the intronless ORF47 mRNA prevented nuclear export of intronless KSHV transcripts, two independent mRNA export assays were performed. Firstly, northern blotting was used to detect if intronless ORF47 mRNA was present in the nuclear or cytoplasmic fraction of transfected cells. Very little ORF47 mRNA was detected in the cytoplasmic RNA fraction of cells transfected with pORF47 alone (9.9±4.9%), whereas cells co-transfected with pORF47 and pORF57GFP displayed a clear shift in ORF47 mRNA from the nuclear to the cytoplasmic fraction (81.5±1.0%), indicative of ORF57-mediated viral mRNA nuclear export. However, upon co-transfection with pORF47 and pORF57PmutGFP, the majority of ORF47 mRNA was no longer found in the cytoplasmic fraction (21.3±3.8%), instead it was retained in the nuclear pool at similar levels to those seen for the negative control, symptomatic of a failure in ORF57-mediated viral mRNA nuclear export (Fig. 5A). To confirm that the ORF57 mutant did not affect mRNA stability, total RNA levels were assessed by northern blot analysis. No significant difference in ORF47 mRNA levels was observed between cells expressing wild type or mutant ORF57 proteins (Fig. 5A, right panel). However, a slight decrease in total mRNA levels is seen in the presence of both the ORF57 or ORF57PmutGFP compared to the GFP control. At present, the reason for this is unknown, however, it could be due to the overexpression of the ORF57 protein.
To confirm the above result, a fluorescent in situ hybridisation assay was utilised. 293T cells were transfected with pORF47, in addition to either pGFP, pORF57GFP or pORF57PmutGFP. 24 h post-transfection cells were fixed, permeabilised and incubated with a biotin-labelled oligonucleotide specific for the KSHV ORF47 mRNA. After a 4 hr hybridisation cells were washed and ORF47 mRNA subcellular localisation was visualised using Cy5-streptavidin. Cells transfected with pORF47 and GFP retained the ORF47 mRNA in the nucleus, whereas ORF47 mRNA was clearly visualised in the cytoplasm of cells transfected with pORF47 and pORF57GFP. However, upon transfection with pORF57PmutGFP, ORF47 mRNA was only observed in the nucleus, symptomatic of a failure in ORF57-mediated viral mRNA nuclear export (Fig. 5B). Together, these two independent assays demonstrate that the ORF57-dependent recruitment of hTREX to intronless viral transcripts is essential for their efficient nuclear export.
We were also interested to determine whether the recruitment of the complete hTREX complex is required for virus replication and infectious virion production. To this end, we utilised a 293T cell line harbouring a recombinant KSHV BAC36-GFP genome [40]. This KSHV-latently infected cell line can be reactivated releasing infectious virus particles in the supernatant which can subsequently be harvested and used to infect 293T cells [41]. The 293T-BAC36 cell line was transfected with pGFP, pORF57GFP or pORF57PmutGFP and concurrently reactivated using TPA and incubated for 72 hours. The supernatants from each flask were then harvested and used to re-infect 293T cells and GFP positive cells were scored 48 h post-infection, as described above. Results revealed similar levels of lytic replication and virus production from cells expressing pGFP or pORF57GFP. However, virus production was significant reduced (P = 0.018) upon the expression of the ORF57PmutGFP (Fig. S5). Therefore, these results demonstrate that the ORF57-dependent recruitment of the complete hTREX complex to intronless viral transcripts is essential for efficient virus lytic replication and infectious virion production.
The above data show that ORF57 binds viral intronless mRNA and directly interacts with Aly. Given that Aly is able to recruit the export factor TAP directly, it was of interest to determine if UAP56 and the hTHO-complex are required for viral mRNA export. In contrast to the cellular mRNA model, a major advantage of our viral system is that hTREX assembly on the viral mRNA is dependent upon an interaction with a virus-encoded protein, not splicing. Specifically, ORF57 binds viral mRNA, directly interacts with and recruits Aly which in turn then interacts with and uses UAP56 to bridge an interaction with the hTHO-complex. This ordered recruitment allows us to specifically disrupt the viral mRNA-ORF57-hTREX complex at different points and assess the functional significance on nuclear export. Furthermore, rather than using an artificial in vitro assay to investigate the functional significance of hTREX, we assessed this in the context of the virus replication cycle using the 293T-BAC36 assay described above.
The trans-dominant mutant, pAlyΔC-myc, which has 20 residues deleted from the carboxy-terminus of Aly, is unable to interact with UAP56 [42]. We were interested in establishing if this mutant could be used to disrupt the assembly of UAP56 and hTHO-complex on an intronless viral mRNA and as such provide insights into whether these proteins are essential for nuclear export. However, prior to its use in the replication assay it was essential to confirm that AlyΔC-myc is still recruited by ORF57 to intronless viral mRNA and is able to interact with TAP. To this end, ORF57, UAP56 and TAP were expressed as GST fusion proteins and incubated with either pmyc, pAly-myc or pAlyΔC-myc transfected cell lysates and pulldown analysis performed. Western blotting using a myc-specific antibody demonstrated that Aly-myc interacted with ORF57, TAP and UAP56. In contrast, AlyΔC-myc is unable to associate with UAP56 but retains the ability to interact with both ORF57 and TAP (Fig. 6A). These results suggest that AlyΔC-myc is an ideal mutant to inhibit the recruitment of UAP56 and hTHO-complex on the viral intronless mRNA. However, one caveat to this system is that expression of pAlyΔC-myc may also act in a dominant negative capacity to inhibit spliced mRNA nuclear export [42]. Therefore it was important to allow expression of the spliced ORF57 protein prior to accumulation of pAlyΔC-myc. To this end, transient transfection of pAlyΔC-myc was performed concurrent with reactivation of the KSHV lytic replication cycle, and ORF57 protein levels assessed 24 h later. Results show that comparable amounts of ORF57 were expressed in untransfected, pmyc, pAly-myc and pAlyΔC-myc transfected cell lysates (Fig. 6B).
To test if AlyΔC-myc inhibited the recruitment of UAP56 and the hTHO-complex onto KSHV intronless RNAs, RNA-IPs were performed on reactivated KSHV-infected 293T cells transfected with pmyc, pAly-myc or pAlyΔC-myc. We obtained similar results for pmyc and pAly-myc to those shown in Fig. 2A, where recruitment of hTREX components onto the viral RNA was readily detected 48 h-post reactivation. However, RNA-IPs using cell extracts transfected with AlyΔC-myc showed a dramatic decrease in the recruitment of UAP56 and hHpr1 to viral mRNA (Fig. 6C). RNA-IPs performed using a TAP-specific antibody showed that TAP is recruited to the intronless viral mRNA, irrespective of Aly status. Critically, RNA-IPs using an ORF57-specific antibody produced ORF47 RT-PCR products of a similar intensity, suggesting that ORF57 was not limiting in this assay (Fig. 6C). It should also be noted that we observed a decrease in TAP recruitment to the viral mRNA in the presence of both pAly-myc and pAlyΔC-myc, compared to pmyc control. At present, we are unsure why TAP recruitment is reduced, however, no difference in mRNA nuclear export is observed between pmyc and pAly-myc transfected cells, suggesting that this reduction in TAP recruitment does not impede the nuclear export of intronless viral mRNAs.
To assess if the AlyΔC-myc mutant affected intronless viral mRNA export during replication, northern blot analysis was performed as described above. Results demonstrated that ORF47 mRNA nuclear export is impaired in reactivated cells that expressed AlyΔC-myc, but not in cells expressing myc or Aly-myc (Fig. 6D). Moreover, to determine if expression of AlyΔC-myc had any effect on virus replication, the KSHV-latently infected 293T BAC36-GFP cell line was transfected with pmyc, pAly-myc or pAlyΔC-myc and concurrently reactivated using TPA and incubated for 72 hours. The supernatants from each flask were then harvested and used to re-infect 293T cells. The level of virus replication was determined by scoring the percentage of GFP positive cells 48 h post-infection, as previously described [41]. Similar levels of lytic replication and virus production were observed from pmyc and pAly-myc pre-transfected cells. Strikingly, virus production from pAlyΔC-myc pre-transfected cells was reduced by approximately 10 fold (Fig. 6E). These data demonstrate that ORF57-mediated recruitment of Aly and TAP to an intronless viral mRNA is insufficient for its nuclear export and that a lack of UAP56 and hTHO-complex on an intronless viral mRNA has a profound effect on intronless nuclear export and KSHV lytic replication.
Two distinct multi-protein complexes have been reported to contain export adapter proteins and both are recruited to pre-mRNAs during splicing, namely hTREX and the EJC [6],[8]. A recent report showed that hTREX is recruited exclusively to the 5′ end of the first exon in a splicing- and 5′ cap-dependent manner [13]. In contrast to higher eukaryotes, analysis of herpesvirus genomes has highlighted that a proportion of lytically expressed viral genes lack introns. Herpesviruses replicate in the host cell nucleus and therefore require their intronless mRNAs to be exported out of the nucleus via cellular export pathways. How exactly herpesviruses assemble an export competent intronless mRNA is poorly understood. Here we show that a KSHV encoded protein, ORF57, specifically binds, and subsequently recruits, the hTREX complex, but not the EJC, to intronless viral mRNA. Specific disruption of the ORF57 interaction with hTREX abolishes efficient viral mRNA export. Furthermore, uncoupling of hTREX assembly, demonstrates that recruitment of Aly and TAP alone is not sufficient for intronless viral mRNA nuclear export and virus replication.
Co-immunoprecipitation data show that ORF57 readily associates with components of hTREX, however, no such interaction was observed between ORF57 and the EJC proteins; eIF4A3, Y14 and Magoh. This result suggests that the EJC is not recruited to intronless viral transcripts and this was confirmed using RNA-IP assays. In contrast, hTREX proteins readily precipitated with intronless viral mRNA, in the presence of ORF57, which presumably functions as a linker between hTREX and the viral mRNA. These findings suggest that the essential export adapter complex for intronless KSHV nuclear export is hTREX and not the EJC. It should be noted that these findings are in contrast to previous observations using a homologue of KSHV ORF57 from the prototype γ-2 herpesvirus, Herpesvirus saimiri [29]. One possible explanation for these contrasting data is that co-immunoprecipitations from Williams et al. were performed by over-expressing myc-tagged EJC components, whereas this in study, endogenous EJC proteins was precipitated using an eIF4A3-, Y14 and Magoh-specific antibodies. To test this, we have performed co-immunopreciptations with EJC specific-antibodies using HVS-infected cell lysates. No interactions were observed between HVS ORF57 and the endogenous EJC proteins (Fig. S6), suggesting the previously observed interactions may have been due to the overexpression of the EJC components. In addition to splicing dependency, the cap-binding complex protein, CBP80, is required to recruit hTREX to human pre-mRNA, via a direct interaction with Aly. Interestingly, we detected a direct interaction between ORF57 and CBP80, implying that the 5′ cap may also function in intronless KSHV mRNA export. However, upon disrupting the ORF57 and Aly interaction (via mutation of the PxxP motif), we also observed a reduction of the ORF57-CBP80 interaction. Analysis suggests that this reduction maybe due to the loss of hTREX affecting the stability of the export competent viral RNP. This suggests that although ORF57 interacts directly with Aly and CBP80, these interactions may not overlap and more detailed analysis of the interacting domains for both proteins is required. It is also worth noting however, that in the absence of ORF57, CBP80 did not recruit Aly to the intronless viral transcripts, suggesting that ORF57 is essential for the loading of hTREX on viral mRNA. The lack of EJC recruitment to intronless viral mRNA may have ramifications beyond those of nuclear export, for example, the EJC has been suggested to function in translational efficiency [16],[43]. Intriguingly, the herpes simplex virus type-1 (HSV-1) ORF57 homologue, ICP27, has been implicated in increased translation efficiency [44],[45], we are currently investigating whether ORF57 increases translation of KSHV transcripts during virus replication.
The current model for hTREX assembly on a spliced mRNA describes UAP56 and Aly associating with the mRNA in a 5′ cap- and splicing-dependent manner. Moreover, as shown in Fig. 7A, it has been suggested that UAP56 may bridge an interaction between Aly and the hTHO-complex [9],[42]. In contrast, during KSHV replication hTREX appears to be tethered to an intronless KSHV mRNA via an exclusive interaction with ORF57. Taking advantage of this, we used the ORF57-hTREX complex to gain insight into how individual components of hTREX interact with one another. Our data show that ORF57 interacts exclusively with Aly, which then binds directly to UAP56 and this in turn functions as a bridge to recruit hHpr1 and presumably the complete hTHO-complex (Fig. 7B). This order of hTREX assembly is in broad agreement with the model proposed by Cheng et al who showed using RNase H digestion analysis that Aly was the most 5′ of the hTREX components, with UAP56 and hTHO-complex binding further downstream. Interestingly, the direct interaction observed between ORF57 and CBP80 suggest that ORF57 may recruit hTREX to the 5′ end of the intronless mRNA, perhaps to provide directionality to nuclear export as is the proposed case for spliced human mRNA [13].
The functional significance of hTREX recruitment to intronless viral mRNA is substantiated using an ORF57 point mutant and a dominant-negative Aly mutant. Specifically, we were able to disrupt the direct interaction between ORF57 and Aly by mutating two proline residues within a region of ORF57's Aly-binding domain [28]. This ORF57Pmut was still able to recognise and bind intronless viral mRNA, however, it lacked the ability to recruit hTREX to these transcripts. A failure to recruit hTREX rendered the ORF57Pmut non-functional as a viral mRNA export protein and provides direct evidence that the hTREX complex is essential for the efficient export of intronless viral mRNA and virus replication. The export adapter Aly is able to interact directly with the export factor complex TAP/p15 [46], therefore, we were interested in assessing whether Aly-TAP/p15 recruitment produced an export-competent intronless viral mRNP or if UAP56 and the hTHO-complex were also required for nuclear export. This is of particular importance as a number of ORF57 homologues, such as Herpes simplex virus-type 1 ICP27, have been shown to interact with Aly and TAP, but it is unknown whether they also recruit UAP56 and the hTHO-complex [32],[33]. One major advantage of using the KSHV system to study hTREX assembly in contrast to analysing hTREX recruitment to a spliced human mRNA is that recruitment of hTREX on an intronless viral transcript is mediated via a direct interaction between ORF57 and Aly, which serves to target the remainder of hTREX to the intronless viral mRNA. This facilitated the use of a trans-dominant Aly mutant, termed AlyΔC, which retains a direct interaction with ORF57 yet fails to interact with UAP56. The AlyΔC mutant has limited use as a tool for dissecting hTREX recruitment to spliced human mRNA as it does not bind to spliced mRNA [42]. The introduction of the dominant-negative AlyΔC mutant into a KSHV virus replication system dramatically reduced the amount of UAP56 and hHpr1 recruited to intronless viral transcripts and this in turn led to a striking reduction in intronless viral mRNA nuclear export and significantly, virus replication. Importantly, RNA-IP analysis of intronless viral mRNPs from cells expressing AlyΔC revealed that ORF57, Aly and TAP where all present on intronless viral mRNA, suggesting that UAP56 and perhaps the hTHO-complex possess an unidentified, yet essential role in mRNA nuclear export.
These data place hTREX at the hub of human mRNA nuclear export. However, RNAi studies in Drosophila melanogaster and Caenorhabditis elegans have shown Aly to be non-essential for mRNA export in these systems [47],[48]. In addition, a genome-wide RNAi study in D. melanogaster reported that the conserved THO-complex was only required by a subset of transcripts for nuclear export [49]. Interestingly, D. melanogaster and C. elegans do require UAP56 both for viability and for bulk mRNA nuclear export [10],[50]. This suggests that while there may be redundancy in eukaryotic systems for certain TREX components, others remain essential. Similar controversy surrounds the role of Aly in herpesvirus mRNA export. In contrast to our data, which show the ORF57-Aly interaction to be essential for efficient intronless viral mRNA nuclear export, a study reported that depletion of Aly using RNAi had little effect on ORF57-mediated transactivation [51]. In addition, a second manuscript reports that differences in the ORF57-Aly binding affinity does not effect ORF57 export function [52]. One possible explanation for these discrepancies is that only partial depletion of Aly was achieved by RNAi, and that such a small reduction in total Aly protein (less than 25% compared to control) may not be functionally significant. Likewise, the mutant ORF57 proteins described by Nekorchuk et al (2007) also failed to interact with viral mRNA, which makes it difficult to interpret their significance with regards to ORF57-mediated nuclear export of viral mRNA.
Our findings using a naturally occurring intronless viral mRNA may provide some insight to the nuclear export of cellular intronless mRNAs, which are often studied using in vitro-transcribed cDNAs. The H2A intronless mRNA is exported by SRp20/9G8 that recognise and bind a specific sequence in the target intronless mRNA and subsequently promote export via a direct interaction with TAP (Fig. 7C i) [53],[54]. Our data suggest that ORF57 may function in a similar manner to cellular SR-proteins by binding to a sequence-specific region of the intronless viral mRNA. Work is currently underway in our laboratory to identify potential ORF57-target sequences in intronless viral mRNA. Conversely, it will be of interest to determine whether hTREX is recruited to H2A mRNA by SR-proteins.
More recently, Aly was shown to be recruited to an in vitro transcribed intronless β-globin construct independently of splicing, via a direct interaction with the CBC protein, CBP20 [55]. In contrast, a second publication suggested that Aly is recruited to in vitro transcribed intronless mRNA by UAP56 in ATP-dependent manner [56]. Interestingly, while there are some disparities, the observations made by both groups generally support a model whereby the 5′ cap, Aly and UAP56 are involved in intronless mammalian mRNA nuclear export (Fig. 7C ii and iii). Once again, it will be interesting to see if further analysis reveals the presence of an entire hTREX complex on these mRNAs.
In summary, these data highlight that a complete hTREX complex is required for efficient KSHV intronless mRNA export and replication. Importantly, data herein demonstrate that recruitment of the nuclear export factor TAP and its adapter protein, Aly, are not sufficient to promote nuclear export. These data suggests that UAP56 and the hTHO-complex must be recruited in order to form an export competent KSHV intronless mRNP.
Oligonucleotides used in cloning, RT-PCR analysis and mutagenesis can be found in Table S1. To generate pORF57GFP, ORF57 cDNA was amplified by PCR and cloned into pEGFP-N1 (BD Biosciences Ltd). pORF57GFPpmut was generated using the QuickChange II site-directed mutagenesis kit (Stratagene). pORF47 and pORFgB were cloned into pCDNA3.1+ (Invitrogen). pETORF57 and pETORF57pmut were cloned in pET21b (Novogen). The genomic ORF50 gene was cloned into pCS2MT+ [57]. To generate pGST-ORF57pmut, the NcoI/HindIII fragment of pGST-ORF57 [28] was replaced with the NcoI/HindIII fragment from pORF57GFPpmut. pAlymyc and pAlyΔmyc [42], pGST-Aly, pGST-UAP56, pGST-hHpr1 and pUAP56-His [9] and pET-CBP80 [13] have all been described elsewhere.
SRp20, Y14, and Magoh (Santa Cruz Biotech), p53 (Pharmagen Inc), GFP mAb and GFP pAb (BD Biosciences) and Myc, SC-35, B-actin and GAPDH (Sigma) were purchased from the respective suppliers. Specific antibodies to CBP80, Aly, UAP56, hHpr1, fSAP79, hTho2 and eIF4A3 were previously described [9],[13]. Unless stated all antibodies were used at a dilution of 1:1000 for western blot analysis.
HEK-293T, HeLa cells and 293T BAC36 cells harbouring a recombinant KSHV BAC36 genome [41] were cultured in Dulbecco's modified Eagle medium (DMEM, Invitrogen, Paisley, UK) supplemented with 10% foetal calf serum (FCS, Invitrogen), glutamine and penicillin-streptomycin. KSHV-infected BCBL-1 cells were cultured in RPMI medium (Invitrogen, Paisley, UK) supplemented with 10% foetal calf serum (FCS, Invitrogen), glutamine and penicillin-streptomycin. 293T BAC36 cells and BCBL-1 cells were reactivated using TPA (20 ng/ml) for 24 h. Plasmid transfections were carried out using Lipofectamine™ 2000 (Invitrogen, Paisley, UK), as per the manufacturer's instructions.
Glutathione S-transferase (GST) pulls downs and co-immunoprecipitations, in addition to subsequent protein analysis by SDS-PAGE and western blot, were performed as previously described [29]. Confirmation of successful RNase was carried out as described by Carlile et al (Fig. S7) [55],[58]. A polyclonal antibody to p53 served as an unrelated antibody control throughout. This antibody precipitated the cognate p53 protein (Fig. S8). RNA-immunoprecipitations were performed as previously described [59].
Nuclear and cytoplasmic RNA for northern blots was extracted from transiently transfected cells using the PARIS™ kit (Ambion Inc., Warrington, UK) as per the manufacturer's instructions. Northern blots were carried out as described previously [60]. Membrane bound RNA was hybridised with 32P-radiolabelled random-primed probes specific for ORF47 and 18S rRNA. The blots were then analysed using a FUJIX BAS1000 Bio-Imaging Analyser (Fuji Photo Film Co. Ltd) and data quantified using the AIDA (Advanced Image Data Analyser) version 2.31 software.
In situ hydridisation was performed as previously described [61]. 25 ng of biotin-labelled probes specific for KSHV ORF47 mRNA were denatured and hybridised at 37°C for 4 hrs. For detection, cells were incubated for 30 min with 7 µl of 12.5 µg/ml of Cy5-streptavidin (Molecular probes). Coverslips were mounted in Vectorshield® mounting medium (Vector Laboratories, CA) and staining visualised on an Upright LSM 510 META Axioplan 2 confocal microscope (Zeiss) using the LSM Imaging software (Zeiss).
293-T BAC36 cells harbouring KSHV BAC36 under hygromycin selection were reactivated using 20 ng/ml TPA. At 72 h post reactivation, filtered tissue culture supernatants were used to spinoculate 1×105 HEK 293-T cells in the presence of 5 µg/ml polybrene. Infected EGFP-positive cells were quantified at 48 h post-infection by fluorescence microscopy.
|
10.1371/journal.pcbi.1006569 | Modeling craniofacial development reveals spatiotemporal constraints on robust patterning of the mandibular arch | How does pattern formation occur accurately when confronted with tissue growth and stochastic fluctuations (noise) in gene expression? Dorso-ventral (D-V) patterning of the mandibular arch specifies upper versus lower jaw skeletal elements through a combination of Bone morphogenetic protein (Bmp), Endothelin-1 (Edn1), and Notch signaling, and this system is highly robust. We combine NanoString experiments of early D-V gene expression with live imaging of arch development in zebrafish to construct a computational model of the D-V mandibular patterning network. The model recapitulates published genetic perturbations in arch development. Patterning is most sensitive to changes in Bmp signaling, and the temporal order of gene expression modulates the response of the patterning network to noise. Thus, our integrated systems biology approach reveals non-intuitive features of the complex signaling system crucial for craniofacial development, including novel insights into roles of gene expression timing and stochasticity in signaling and gene regulation.
| Proper development of the body requires boundaries to form between regions in which cells will form different structures, and these boundaries need to be properly organized in space. This must occur accurately even in moving, dividing cells and in the presence of the noise that is inherent in all biochemical processes. We use development of the upper and lower jaw as a model to study boundary formation. In this work, we combine detailed experimental measurements with computational modeling to investigate the role the timing of gene expression plays in organizing spatial boundaries, and find that the different orders of gene expression navigate a tradeoff between precision and accuracy in boundary positioning.
| A fundamental question in developmental biology is pattern formation, i.e. the acquisition of positional identity in cells resulting in spatially organized domains of gene expression. Computational analyses have long sought to address how patterning occurs in growing tissues that change their size and shape by modeling morphogen gradients, signaling between cells, geometric transformations and other mathematically-amenable aspects of development [1–4]. More recently, computational modeling has revealed how signaling networks integrate with one another and the importance of feedback loops in precise regulation of early developmental patterning systems [5–11]. Models for more complex developing structures such as vertebrate limb buds [12,13], hair follicles [14,15], pigment cells in the skin [16], the spinal cord [17,18] or the palate [19,20] require integrating multiple signals within rapidly expanding three-dimensional (3D) tissues.
Pharyngeal arches are bilateral, segmentally-repeated structures that form in the ventral head of vertebrate embryos and give rise to skeletal, muscle and connective tissues of the face and neck, including the upper and lower jaws. Arches are complex both in their 3D morphologies and in their embryonic cellular origins. Streams of cranial neural crest (NC) cells migrate into each arch segment and surround cores of myogenic/vasculogenic mesoderm. The surrounding ectodermal and endodermal epithelia produce signals that subsequently pattern the arch along its dorso-ventral (D-V) axis, resulting in at least three early domains: ventral (V), intermediate (I), and dorsal (D) [21–23]. D-V arch patterning involves a highly-conserved signaling network consisting of the Bone morphogenetic protein 2/4/7 (Bmp) and Endothelin-1 (Edn1) signaling pathways, secreted by ventral arch epithelia [24–28], and dorsal Jagged1 (Jag1)/Notch signaling [29,30]. Errors in these signals can lead to craniofacial birth defects, such as auriculocondylar syndrome in humans, in which Edn1 signal transduction is disrupted leading to partial homeotic transformation of ventral skeletal elements to a dorsal fate [31,32]. Understanding how D-V domains arise in the midst of NC migration and arch growth is both an experimental and a computational challenge.
Both Edn1 and Bmp are crucial for ventral and intermediate arch development, but with distinct effects on gene expression [24,27,30,33–35]. Bmp, which acts as a morphogen in many contexts [36–38], primarily induces and maintains genes expressed ventrally such as Hand2, a critical transcription factor for ventral mandibular identity. In contrast, while Edn1 also induces ventral genes initially in a concentration-dependent manner [26,39] it later becomes primarily required for expression of intermediate genes such as Dlx5/6, which are required for ventral/intermediate mandibular (lower jaw and jaw joint) development, and less dependent on Bmp [24,29,30]. Craniofacial patterning defects in edn1-/- mutants can be largely rescued by injection of Edn1 protein throughout the arch [27], and recent work suggests Edn1 plays a more permissive than instructive role [40].
Given their common targets, what are the advantages of having these two ventral morphogens acting in parallel during early D-V arch patterning? In addition, how does patterning occur robustly in the face of continuous cell divisions, rearrangements, and noise in both the signaling molecules and their downstream gene regulatory networks (GRNs)? To address these questions, we have developed the first computational model of arch D-V patterning that incorporates growth, migration, gene expression and different sources of noise. We represent the known components of the arch GRN with a system of ordinary differential equations (ODEs) that accurately reproduces published genetic perturbations of arch D-V patterning. We establish the model using measurements of spatiotemporal patterns of gene expression and 3D NC cell movements obtained from time-lapse movies of live zebrafish embryos. Quantitative temporal gene expression data reveal that intermediate domain genes are expressed before genes marking the ventral domain and dorsal genes are expressed last. The model confirms that this temporal order of intermediate-ventral-dorsal (IVD) patterning improves some aspects of robustness of D-V patterning (precision, referring to consistency across simulations/embryos), while making other aspects (accuracy, referring to closeness to the theoretical ideal pattern) more sensitive. The model further suggests that Bmp signaling primarily establishes the sizes and positions of patterning domains, while Edn1 plays a permissive role, and that noise in the GRN and each of the signaling pathways affects patterning differently. Our model reveals novel features of the early spatiotemporal dynamics of gene expression that are critical for patterning the complex 3D structure of the craniofacial skeleton during embryogenesis.
Institutional Animal Care and Use Committee protocol #2000–2149.
Neural crest (NC)-derived ectomesenchymal cells in pharyngeal arches 1 (mandibular) and 2 (hyoid) in zebrafish are patterned into three D-V domains between 14–36 hpf, which give rise to distinct skeletal elements in the adult (Fig 2A–2C). Arch D-V length roughly doubles over this period (from 30 to 60 μm). Previous in situ hybridization (ISH) studies have shown that dlx3/4/5/6 are expressed together in an early ventral-intermediate (V-I) domain that later separates into V and I [24,29,30,42]. hand2, the homolog of which is induced by Dlx5/6 in mice [34,60,61], marks the new V domain and represses Dlx genes ventrally, restricting their expression to the I domain [42,62]. However, the precise timing of gene expression between 14–20 hpf, when these domains first appear, remains unclear.
To address this, we have measured transcript levels of seven D-V patterning genes in FAC-sorted arch cells using NanoString analysis (dorsal: jag1b, hey1; intermediate: dlx3b, dlx4b, dlx5a, dlx6a; ventral: hand2) (Fig 2D). Surprisingly, expression of dlx3b peaks early at 20 hpf, followed six hours later by three other intermediate genes (dlx4b, dlx5a, dlx6a), the ventral gene (hand2), and the dorsal gene jag1b at 26 hpf. Similarly with hybridization chain reaction (HCR) in situs, to facilitate co-localization and quantitation of expression (Fig 2E–2T) we find that within the domain of dlx2a expression, which marks NC cells in the entire arch, strong dlx3b expression is detected at ~17 hpf, at least an hour before dlx5a expression appears faintly at ~18 hpf. Meanwhile, hand2 expression is not detected until ~20 hpf. Interestingly expression of hand2 arises abruptly, while other genes such as dlx5a appear more slowly, yet both peak at a similar time point in the NanoString analysis. Thus, intermediate genes are the first to be expressed in the D-V sequence of arch patterning followed by ventral and finally dorsal genes.
The 1D model (Fig 3A–3E) recapitulates the relative timing and sizes of D-V domains in the mandibular arch. Initially intermediate gene expression extends from the ventral end to approximately halfway up the D-V axis. Subsequently, dorsal genes are expressed at the dorsal end of the arch, leaving a section of “unpatterned” cells (white regions, in which gene expression is below the arbitrary 20% threshold) between I and D domains, which gradually diminishes. Initiation of ventral gene expression at 28 hpf creates a narrow V domain, which moves the I domain dorsally. The 2D model (Fig 3F–3J) captures these spatiotemporal dynamics of D-V domain formation. Here, individual cells are also defined as patterned if their gene expression exceeds 20%, and they are gradually colored correspondingly, such that grey indicates cells not yet expressing genes above the cut-off and more deeply colored cells indicate higher and higher levels of gene expression. Initially none of the cells express any of the genes above the 20% cut-off (grey cells). Between 22–35 hpf the arch elongates anteriorly and ventrally. During this tissue deformation the cells acquire D, I and V fates, (yellow, blue and pink, respectively) and form domains of the correct size and shape. The simulation results agree with live imaging of hand2:GFP:sox10:lyn-tdTomato double transgenics (Fig 3K–3O) or dlx5a:GFP;sox10:lyn-tdTomato double transgenics (Fig 3P–3T). We note that the boundaries of gene expression as shown by transgene reporter intensity are sharp (S2 Fig).
To generate a 2D cell-center based model that reflects arch morphogenesis as accurately as possible we have measured mandibular arch deformation in images of 6 sox10:lyn-tdTomato transgenic zebrafish embryos, the average of which is used to generate an arch outline (Fig 3K–3T). By further analyzing time-lapsed images of sox10:nEOS transgenics we find that: 1) cell number roughly doubles between 22–36 hpf, 2) ~90% of this increase in cell number is due to cell division and 3) only ~10% is due to cells migrating into the arch dorsally. These parameters are incorporated into the model to compute the time intervals of cell division and cell migration.
The 2D model also reproduces previously reported phenotypes of genetic or pharmacological perturbations that disrupt D-V patterning (Fig 4). V/I domains do not form and the D domain expands in embryos lacking Bmp or Edn1 signaling (reduced in the modeling simulations to 1% of wild-type expression [24,27] (Fig 4A and 4D). The I domain does not form and D expands ventrally in embryos overexpressing the dorsal factor Jag1 (500% of wild-type production) (Fig 4C). Conversely, the I domain expands to replace D in a jag1-/- mutant (1% of wild-type expression) or when Edn1 is overexpressed (500% of wild-type gradient maximum) (Fig 4B and 4F). Importantly, the model also recapitulates the normal D-V patterning observed experimentally with moderate, uniform Edn1 expression (50% of wild-type gradient maximum), achieved with Edn1 protein injections [24,27] (Fig 4E). Thus, despite the minimal GRN on which it was based, the model captures many aspects of patterning observed experimentally in vivo.
For any three sets of D-V patterning factors, there are six possible different temporal orders of gene expression (VID, VDI, IVD, IDV, DIV, DVI). Our gene expression studies indicate that dlx3b and its associated I domain appear first, so we asked how this order fares in our model as compared with other possible orders. By varying the production and degradation parameters for each gene group we simulate all six temporal orders (S1 and S2 Tables) and examine if any one is more robust than another. Changing the temporal order does not affect the final pattern at 35 hpf (S3 Fig). To compare sensitivity between the three temporal orders of gene expression we compute si(t) (Eq 9), evaluated at 10 equidistant time points between 22–35 hpf. We summarize the 10 resulting data points in box plots (Fig 5, S4–S6 Figs), where the line in the middle denotes the median, the bottom and top edges of the box the 25th and 75th percentile, respectively, and the whiskers extend to the most extreme data points. We normalize the data with respect to the highest value of si(t) for both parameters p1 and p5 at all time points (for unprocessed, time-dependent data see S7 Fig). The distance between the median line and zero (dashed black lines) indicates how sensitive gene expression is to perturbations in the control parameters p1-p7, and the size of the box and length of the whiskers indicate how much the sensitivity changes with time. In general, all model simulations are relatively robust to parameter variations, indicating that the results are not due to the specific choice of parameters.
When cells are exposed to morphogen concentrations typical for the V and I domains (high Bmp and Edn1 concentrations) the GRN is most sensitive to perturbations in the BMP signaling parameters p1 (activation of V genes) and p5 (activation of I genes) (Fig 5; S4 and S5 Figs), and less so to variations in Edn1 signaling parameters p2 and p7. This is particularly true in cases where I genes are expressed last (VDI and DVI), since median sensitivity levels of dorsal genes deviate further from zero (dashed black lines; Fig 5 and S4 and S5 Figs). However, when cells are exposed to low Bmp and Edn1 concentrations (S6 Fig), typical for the D domain, perturbations in the Bmp gradient only affect the ventral genes, while intermediate and dorsal genes are sensitive to perturbations in the Edn1 parameters p2 and p7. This agrees with published evidence that Edn1 plays a primarily permissive role in ventral and intermediate gene expression [27,40]. A permissive role for Edn1 is further evident when we compute the sensitivity to parameter variations in the GRN if only one of the two morphogen gradients is present (S8 Fig). The results are similar to those for two morphogens in all three V, I and D domains. In all cases, the most crucial parameter is the one controlling the effect of the morphogen on the I domain, p2.
To simulate the stochasticity that may occur in vivo, we have added Gaussian noise to the morphogen gradients and to the GRN (Eqs 11–14, Figs 6 and 7, S9–S11 Figs and S6 and S7 Movies). To compute a large number of stochastic simulations for statistics, and for better visualization, we investigate the effects of noise first in the 1D model. We plot the mean (thick lines) and ±σ (shaded regions) over 100 simulations (Fig 6B–6E, 6A for comparison without noise). For the 2D model, we show end-states for single simulations (Fig 6B’–6E’, 6A’ for comparison without noise). The simulations reveal that noise in Bmp signaling is transmitted differently than noise in Edn1 into the GRN. While intermediate gene expression is affected by both signals, ventral gene expression is not affected by noise in the Edn1 gradient, since the only input into the V domain is the Bmp gradient. Dorsal genes are most robust to morphogen fluctuations (Fig 6C–6D’), since they are mostly controlled indirectly through the ventral genes, but they are the most sensitive to gene regulation noise (Fig 6B and 6B’). Since Edn1 is required for intermediate gene expression, but increasing Edn1 signaling does not expand the I domain (Fig 4E and 4F), fluctuations in Edn1 only act in one direction, meaning that lower Edn1 levels due to noise reduce the I domain but higher levels have no effect. As a result the I domain is reduced with fluctuations in Edn1 (Fig 6D and 6D’), when compared to the deterministic case (Fig 6A and 6A’, S9 Fig) in both 1D and 2D models. This means that Edn1 acts “unidirectionally” as a permissive factor. Gene expression noise appears to be the strongest driver of fluctuations in patterning, since even relatively small fluctuations (ν = 0.05) in the GRN alter gene expression profiles substantially (Fig 6B and 6B’). The effects of individual fluctuations are additive when all sources of noise are combined (Fig 6E and 6E’, S11 Fig), i.e. both increased fluctuations in dorsal gene expression due to GRN noise, and expansion of the V domain due to noise in the Edn1 gradient.
Precision refers to the level of variation in boundary positions in stochastic simulations. Similar to our analysis of sensitivity to different parameters in the GRN, we examine if the sequence of D-V domain formation influences precision, as a measure of robustness in response to noise. When fluctuations are limited to the Bmp gradient the genes expressed first absorb most of the noise (Fig 7). The exception is when dorsal genes are expressed earliest, which are in general the most robust to Bmp noise, such that the DIV and DVI sequences are the least susceptible to Bmp fluctuations (Fig 7E and 7F).
This is in contrast to noise in the GRN (v = 0.05), where genes expressed earliest are the most robust and genes expressed later are more susceptible to noise (S9 Fig). This is particularly true when genes expressed in the I domain are first in the D-V sequence. This indicates that patterning I first is beneficial since fluctuations in intermediate gene expression affect both the precision of the V-I and I-D boundaries. With fluctuations in the Edn1 gradient, the I domain is severely reduced when intermediate genes are expressed last, while there is still a distinct I domain in the case of either V or I being expressed first (S10 Fig), further indicating that an early expression of intermediate genes is beneficial for boundary accuracy. Noise in the Edn1 gradient has the most similar effect across the 6 possible temporal orders. Early expression of intermediate genes, however, leads to slightly stronger fluctuations in their expression as a result of Edn1 noise, while still preserving a distinct I domain. Due to the different effects of the three sources of noise in the context of different temporal orders, combining all sources of noise indicates that a later onset of intermediate gene expression leads to the strongest fluctuations and higher sensitivity (S11 Fig).
When differences in the temporal order are enforced explicitly in the simulations (S12 Fig), the responses to the different sources of noise are similar to the case of intrinsic regulation, only with a more severe loss of the I domain due to noise in Edn1 (S13 Fig). However we do not see any distinction between the temporal orders in the extrinsic model with noise in the Bmp gradient (S14 Fig) in contrast to the intrinsic model with Bmp noise (Fig 7).
Accuracy refers to boundary positions relative to the measured wild-type positions (Fig 8). We simulate 10 runs of the 2D model, compute statistics of the boundary error (E) in Eq 8, normalize according to the highest value and plot the mean (line) and ±σ (error bars). Accuracy depends strongly on which gene group is expressed last, especially for the I-D boundary. When either Bmp or Edn1 are noisy, the I-D boundary is positioned most accurately when I is last, slightly less accurate when D is last and inaccurate when V is last (Fig 8B and 8D). In contrast, when Bmp is noisy the V-I boundary shows no clustering of temporal orders of patterning (Fig 8A), but when Edn1 is noisy accuracy increases when I is last (Fig 8C). The boundary accuracy depends less distinctly on the temporal order of patterning when the gene regulation is noisy (Fig 8E and 8F). When we combine all sources of noise a late appearance of the I domain clearly improves positioning of the V-I boundary and slightly improves the I-D boundary (Fig 8G and 8H).
In general the domain boundaries are more sensitive to fluctuations in Bmp than in Edn1, especially at the V-I boundary. Presumably the mutual inhibition between intermediate and dorsal genes makes the I-D boundary more robust to stochastic fluctuations. When the sequence of patterning is controlled by extrinsic factors not inherent in our minimal GRN (S12 Fig), the accuracy of boundary positioning is more sensitive to individual sources of noise in all cases except for positioning the V-I boundary with fluctuations in only Edn1 (S15 Fig). However, with the external control of gene expression timing the noise effects appear to be less additive and the extrinsic model positions the boundaries more accurately.
From the above results, different temporal orders lead to different degrees of noise in gene expression profiles (Figs 6, 7 and S9–S11), while there are observed differences in accuracy of domain boundary positioning (Fig 8). We now quantify these effects to investigate how accuracy and precision relate to each other at both domain boundaries, and with the different sources of noise (Fig 9). Precision is measured as the maximal standard deviation (Figs 7 and S9–S11), corresponding to the maximal width of the shaded regions. Accuracy is the mean value of the boundary error (Fig 8). Hence for both, a low value indicates higher accuracy or higher precision. When all sources of noise are included in the simulation, an anti-diagonal trend is observed, indicating a trade-off between precision and accuracy, where temporal patterns with more precise boundary positioning have less accuracy, and vice versa. The observed IVD pattern favors precision over accuracy at both boundaries, suggesting that the patterning system has evolved to maximize precision in gene expression domain boundaries.
We have analyzed spatiotemporal patterns of expression of D-V patterning genes during pharyngeal arch morphogenesis in zebrafish embryos and combined our experimental observations with published data to generate the first computational model of the developing mandibular arch. Previous efforts have compiled information from imaging and gene expression databases, such as the FaceBase consortium [63], and analyzed cellular behavior during craniofacial morphogenesis [20], but ours is the first to integrate spatiotemporal gene expression patterns with morphogen gradients, tissue measurements, and known mutant phenotypes. Our model captures many of the spatial and temporal features of arch development and recapitulates genetic perturbations. It also provides novel insights into developmental constraints on the system, including: 1) Bmp is responsible for providing positional information to the cells, while Edn1 is permissive, 2) the temporal order of patterning is important for the system’s capacity to account for noise, and 3) the temporal order favors precision over accuracy in boundary positioning.
Many developmental processes that involve periodic patterning reflect underlying reaction-diffusion systems that deal efficiently with noise through their intrinsic feedback loops [12,16,19]. We find that the temporal order of gene expression provides a previously unappreciated factor in improving responses to noise. Surprisingly in our experiments it is the intermediate gene dlx3b that is the first detected at 16–17 hpf by HCR and slightly later (20 hpf) in our NanoString analyses, which may reflect the fact that sorted cells used for NanoString are derived from multiple arches (mandibular, hyoid, branchial) while with HCR we image the first arch directly. dlx3b might serve as an early response factor, possibly integrating input from multiple signals and priming other patterning genes in the GRN. The early appearance of the I domain means that the arch is not patterned consecutively from one end to the other (i.e. the VID or DIV orders), but rather in the peculiar sequence of establishing the middle first. Intuitively one can assume that patterning the central domain first already establishes both boundaries. However, that is not the case here either, since expression of the intermediate genes initially extends to the ventral end of the arch and is only later replaced by the ventral genes, such that both boundaries are established independently and non-simultaneously. We have investigated if this temporal order is beneficial for the response of the system to stochastic fluctuations. Our model simulations suggest that avoiding having intermediate genes expressed last improves the robustness to perturbations in Bmp signaling parameters and precision in the positioning of D-V domain boundaries. While dlx3b knockdown does not cause significant patterning defects, this could reflect compensation by dlx4b or dlx5a [42]. The unique function of dlx3b in the GRN is supported by its distinct spatial expression pattern from the dlx5/6 pair and also from its neighboring dlx4 cluster counterpart [42,64–66].
Our results suggest that altering the temporal order of D-V gene expression, (e.g. using optogenetic approaches to induce expression of hand2 in its normal spatial domain but prior to onset of intermediate gene expression and thus creating a VID sequence) will disrupt the accuracy and robustness of D-V domain boundaries. Future studies are also needed to determine if the temporal order of gene expression in this system is controlled by gene-intrinsic differences in sensitivity to signals (as is the case in our minimal model), or if trans-acting factors play a greater role.
We also note that the precision and accuracy of D-V domain boundary formation are distinctly susceptible to noise, with the boundary between V and I domains especially sensitive to noise in Bmp signaling or in the downstream GRN. Precision and accuracy in the boundaries of gene expression domains are both goals for patterning systems, but our model reveals divergent paths to reach each of those goals in the mandibular arch. We find that the most precise gene expression profiles in response to Bmp signaling are achieved when dorsal genes are expressed first (Fig 7), while the most accurate boundaries were obtained when I genes were expressed last (Fig 8). Instead, our data in zebrafish indicate that I genes are expressed first, reflecting a trade-off that seems to favor precision over accuracy (Fig 9). It is important to note here that the model does not achieve high boundary positioning accuracy for the I-D boundary with any of the temporal patterns even in the absence of noise (S1A Fig). This might be a result of measurement inaccuracy and due to the projection of a 3D geometry into 2D, or that other factors not included in our minimal GRN are responsible for pushing the I-D boundary more towards the ventral end. Such trade-offs between precision and accuracy in other systems typically involve negative feedback [12,18,67], while it is temporal gene expression control that navigates this tradeoff in our model. Future analysis will reveal if feedback loops also operate at arch domain boundaries.
Our modeling results for precision and accuracy derive from combinations of 10–100 simulations, somewhat analogous to variability in patterning that can occur between individual embryos. We speculate that developmentally this suggests that patterning will favor precision (i.e. show a narrow range of variability between individuals), even if that range differs from the species ideal. Evolutionarily, this might permit rapid variation in craniofacial morphology during radiation events while still maintaining intra-population characteristics. Changes in craniofacial structures are one of the most striking adaptations in rapidly evolving vertebrate populations such as African rift lake cichlids, and it would be interesting to examine the parameters of patterning in these species [68,69].
Bmp and Edn1 signaling have overlapping but distinct functions in D-V patterning of the mandibular arch, with Edn1 primarily maintaining intermediate gene expression and playing a permissive rather than instructive role [24,27,30,40]. Our modeling results are consistent with Bmp signaling acting as the principal instructive ventralizing signal, at least for inducing V and I domains. They also suggest that moderate perturbations or noise in Bmp signaling, at levels that do not completely eliminate expression of various factors or formation of later cartilage structures, will still have effects on D-V arch patterning. In several reports, changes in levels of patterning signals result in a moderate-to-severe spectrum of phenotypes. Reduction in Bmp signaling by heat-shock induction of a dominant-negative Bmp receptor can lead to varying degrees of loss of ventral structures, depending on the timing of heat-shock and the dose of the dominant-negative transgene [24,30,70]. Increasing Bmp levels moderately by overexpression of Bmp ligand or injection/bead insertion of Bmp proteins causes homeotic transformations, such as the more dorsal palatoquadrate cartilage acquiring characteristics of the ventral Meckel’s cartilage, while more severe changes in signaling levels leads to significant losses throughout the jaw cartilage [24,30,71]. Modulation of Edn1 signaling in either direction similarly results in varying degrees of patterning changes, although consistent with its permissive role even very high levels of Edn1 do not expand the I domain, while reducing the D domain [24,26,27,30,39].
These divergent phenotypes, along with our analyses of parameter sensitivity and the effects of noise, suggest a region of stability where the patterning network can compensate for changes in signals, with moderate phenotypes appearing at the edges of this region, and severe phenotypes occurring when signaling falls outside the stable region altogether. Our computational model suggests that further experiments up- and down-regulating signaling will reveal transition points where stability breaks down, and those data will in turn improve the quality of our model. We can also extend our framework in the future to incorporate other signals that have been implicated in D-V patterning and growth, including Wnts [72] and Fgfs [73] as well as other GRN components [40,42].
Boundaries between different D-V domains in the mandibular arch are sharp, which is especially prominent in our live imaging (S2 Fig). How is this sharpness achieved? Our model simulates sharp domain boundaries in the absence of noise. However, when stochastic fluctuations are present the domain boundaries are no longer sharp since all cells are responding independently to the noisy signals. Computationally, we can extend our model to investigate if cell-cell communication and cohesion between cells of the same identity will improve boundary sharpening. Experimentally, we can examine if changes in signaling too small to disrupt overall patterning can nevertheless degrade boundary sharpness, which would suggest that this sharpness comes from the patterning pathways themselves. Although our current analysis of live cell dynamics does not suggest a significant degree of NC cell rearrangements once they have reached the pharyngeal arches, cell sorting contributes to domain boundary sharpening in some contexts, such as the neural tube [74], and automated tracking of large numbers of arch NC cells in the future might reveal subtle but important cell movements in arches as they are patterned.
D-V arch patterning of the mandibular arch and its associated GRN are largely conserved among vertebrates [21,23,75]. However, differences in the resulting anatomy of the jaw and skull in the adult, as well as the size, shape, and growth of the mandibular arch primordium make it difficult in some cases to draw clear homologies across species e.g. zebrafish and mouse or human [21,23]. In addition, differences in the timing of gene expression and our ability to identify clear functional homologues across species have made comparisons of the genetic pathways involved challenging. Having made a computational model for the zebrafish jaw that incorporates these spatiotemporal features we can now extrapolate to other species to determine if similar constraints apply. In addition, we can begin to ask questions at a more integrated, systems biology level, about how such changes in size, shape and timing may have arisen and how the GRN has adjusted to compensate for changes in such features as signal propagation and noise.
|
10.1371/journal.pgen.1005735 | Loss of the Yeast SR Protein Npl3 Alters Gene Expression Due to Transcription Readthrough | Yeast Npl3 is a highly abundant, nuclear-cytoplasmic shuttling, RNA-binding protein, related to metazoan SR proteins. Reported functions of Npl3 include transcription elongation, splicing and RNA 3’ end processing. We used UV crosslinking and analysis of cDNA (CRAC) to map precise RNA binding sites, and strand-specific tiling arrays to look at the effects of loss of Npl3 on all transcripts across the genome. We found that Npl3 binds diverse RNA species, both coding and non-coding, at sites indicative of roles in both early pre-mRNA processing and 3’ end formation. Tiling arrays and RNAPII mapping data revealed 3’ extended RNAPII-transcribed RNAs in the absence of Npl3, suggesting that defects in pre-mRNA packaging events result in termination readthrough. Transcription readthrough was widespread and frequently resulted in down-regulation of neighboring genes. We conclude that the absence of Npl3 results in widespread 3' extension of transcripts with pervasive effects on gene expression.
| Npl3 is a yeast mRNA binding protein with many reported functions in RNA processing. We wanted to identify direct targets and therefore combined analyses of the transcriptome-wide effects of the loss of Npl3 on gene expression with UV crosslinking and bioinformatics to identify RNA-binding sites for Npl3. We found that Npl3 binds diverse sites on large numbers of transcripts, and that the loss of Npl3 results in transcriptional readthrough on many genes. One effect of this transcription readthrough is that the expression of numerous flanking genes is strongly down regulated. This underlines the importance of faithful termination for the correct regulation of gene expression. The effects of the loss of Npl3 are seen on both mRNAs and non-protein coding RNAs. These have distinct but overlapping termination mechanisms, with both classes requiring Npl3 for correct RNA packaging.
| Budding yeast Npl3 comprises two RNA binding domains (RBDs) and a C-terminal domain that is rich is Arg, Gly, Ser and Tyr residues. This structure shows similarities to the SR (Ser-Arg rich) class of metazoan pre-mRNA binding proteins [1,2]. Genetic and biochemical analyses have implicated Npl3 in many processes, including pre-mRNA splicing, polyadenylation, mRNA export and cytoplasmic translation [3–7], as well as R-loop prevention and chromatin modification [6,8].
Transcription termination of RNA polymerase II (RNAPII) occurs by polyadenylation-dependent and polyadenylation-independent pathways, correlated with whether the transcript is coding or non-coding (reviewed in [9,10]). Termination of mRNAs, requires two complexes termed cleavage and polyadenylation factor (CPF) and cleavage factor (CF). Together, the CPF and CF complexes facilitate cleavage of the nascent RNA strand and removal of the elongating polymerase, resulting in a polyadenylated RNA product. Two mechanisms have been reported for these processes, which are likely to occur in combination. In the ‘torpedo’ pathway, the nascent RNA molecule is cleaved at the polyA site and the released 3’ fragment of the transcript still bound by RNAPII is degraded by the 5’-3’ exonuclease Rat1. This is proposed to then destabilize the polymerase complex. A second “allosteric” mechanism leads to the elongating polymerase being disengaged from the nascent transcript downstream of the polyA site due to, poorly understood, conformational changes concomitant with assembly of the CPF-CF complex. Notably, analyses on reporter constructs indicated that Npl3 can act as an anti-terminator, by antagonizing cleavage factor 1 (CF1) binding and thus restricting the use of cryptic poly(A) sites [4,7,11].
In addition to mRNAs, RNAPII also transcribes several classes of non-protein coding RNAs (ncRNAs) and the majority of these terminate by polyadenylation-independent pathways. These ncRNAs include the small nucleolar RNAs (snoRNAs), 73 of which function in yeast ribosome synthesis, four small nuclear RNAs (snRNAs) that form the core of the pre-mRNA spliceosome, as well as diverse long ncRNAs (lncRNAs) such as the cryptic unstable trancripts (CUTs). The snoRNAs are processed from pre-snoRNAs that can be independently transcribed, cleaved from polycistronic transcripts, or excised from pre-mRNA introns. Independently transcribed snoRNAs, snRNAs and CUTs are all thought to predominately terminate via a pathway that requires RNA-binding by Nrd1-Nab3 complex and the Sen1 helicase (together termed the NNS complex) [12–19]. Termination of snoRNAs and CUTs by the NNS complex is associated with recruitment of the TRAMP and exosome complexes to the nascent RNA [14,20–22]. The TRAMP complex tags RNAs by the addition of a short 3' oligo(A) tail, and directs target RNAs to the nuclear exosome for degradation [23–26]. This can result in either complete degradation of the RNA, in the case of CUTs, or the processing of long precursor snoRNAs to the shorter, mature form [27]. However, some snoRNAs can also be terminated by mRNA 3’ cleavage factors, with [20,28] or without [29] subsequent polyadenylation. In addition, surveillance factors can influence termination, since loss of exosome activity leads to defects in NNS termination [30–33]. Moreover, gene-length correlates with the termination pathway used, probably via changes in the phosphorylation state of RNAPII [34,35] and/or histone H3, lysine 4 trimethylation [36], both of which can promote NSS termination. Prior data indicate that a proportion of RNAPII transcription events terminate early on protein-coding genes [37–40]. These promoter proximal ncRNAs or “sCUTs” [39] are oligoadenylated, presumably by the TRAMP complex [37], and targeted for turnover by the nuclear surveillance machinery.
To better understand the in vivo functions of Npl3, we determined its RNA binding profile, and identified changes in RNA abundance and RNAPII association when the NPL3 gene is deleted. The absence of Npl3 resulted in transcriptional termination defects at diverse RNAs, with readthrough observed on large subsets of both mRNAs and ncRNAs. These termination defects appear to cause widespread changes in gene expression, both through inappropriate termination and through transcriptional interference at neighboring genes.
To identify direct RNA targets of Npl3 binding, we performed in vivo UV cross-linking and analysis of cDNAs (CRAC) [41]. The endogenous NPL3 gene was tagged with an N-terminal ProteinA-TEV-His6 (PTH) tag, retaining the intact, endogenous NPL3 promoter. This construct supported wild-type growth as the sole source of Npl3 (S1A Fig), indicating that the fusion protein is functional. Yeast cells were UV irradiated while actively growing and PTH-Npl3 was isolated, Npl3-bound RNA fragments were purified, converted to a cDNA library and sequenced by next generation sequencing (all sequence data are available from GEO under accession number GSE70191). S1B Fig shows expression of the tagged protein and an autoradiogram of labeled, associated RNAs. Npl3 binding sites were most frequent on mRNAs, consistent with previous studies [40,42,43], but were also identified on several classes of ncRNA, including rRNAs, tRNAs, snRNAs and snoRNAs, as well as lncRNAs, including CUTs, stable unannotated transcripts (SUTs) and other unannotated transcripts apparently derived from intergenic regions or antisense transcription. The distribution of Npl3 across RNA classes is shown for two independent CRAC experiments in Fig 1A. In both datasets, Npl3 binding was predominately on RNAPII transcripts.
The distribution of Npl3 along transcripts showed distinct patterns for different classes of RNA. On mRNAs, Npl3 binding was highest in the 5’ end region (Fig 1B), consistent with other recent RNA-crosslinking data [40]. A previous ChIP analysis, in which Npl3 is crosslinked to chromatin, found Npl3 enriched at 3' ends [6]. This apparent discrepancy may reflect differences in Npl3 binding at the 5' and 3' ends of genes, with direct RNA binding occurring predominantly at the 5' end, and stronger association with the transcription and processing complexes at the 3' end. Similar 5’ enrichment was reported for nuclear surveillance factors including Nrd1, Nab3 and Mtr4 as well as for RNAPII, and has been proposed to reflect a substantial level of premature transcription termination [37,38,40]. As described above, these promoter proximal lncRNAs are oligoadenylated by the TRAMP complex, and we therefore mapped the association of Npl3 with RNAs carrying non-encoded oligo(A) tails [37]. Among RNA fragments recovered in association with Npl3, 24–28% carried oligo(A) tails, depending on the individual CRAC experiment, indicating that Npl3 frequently binds across the junction between truncated mRNAs and oligoA tails. Note that the total fraction of Npl3 target RNAs that are oligoadenylated is likely to be higher, as only a small region of each transcript is sequenced. In contrast, only 4–4.7% of RNAs bound by RNAPII were oligoadenylated (see below). Fig 1C shows the distribution of Npl3 bound hits containing oligo(A) tails across different RNA classes for two independent CRAC experiments. The distribution of oligo(A) tails in Npl3 target sequences was similar to the overall distribution of hits on mRNAs (Fig 1D). This indicated that Npl3 is frequently bound to degradation substrates or intermediates, including prematurely terminated mRNAs, and suggests that it may function with surveillance factors to mediate early transcription termination and/or RNA degradation.
Npl3 is known to be required for efficient splicing of ribosomal protein gene (RPG) pre-mRNAs [3]. Consistent with this, we found that Npl3 strongly accumulated on introns of these pre-mRNAs relative to other intron-containing pre-mRNAs (S1C Fig). On non-RPG, intron-containing pre-mRNAs, the binding of Npl3 dropped sharply at the 5’ end of the intron. The lower recovery of introns relative to mature message indicates that Npl3 remains bound to mRNAs after splicing (S1D Fig).
The distribution of Npl3 over the CUT class of lncRNAs was similar to that observed for mRNAs with strong enrichment towards the 5' end (Fig 1E), consistent with the proposal that initial cotranscriptional packaging of pre-mRNAs and lncRNAs is similar [37]. In marked contrast, Npl3 binding was enriched towards the 3' end of snoRNAs (Fig 1F), suggesting a role in transcription termination and/or 3' end processing of these ncRNAs. Overall, our RNA binding site data suggest that Npl3 is involved in surveillance and/or transcription termination of both mRNAs and ncRNAs.
Motif analysis did not identify a specific Npl3 binding site. We note, however, that the four most overrepresented 4-nucleotide motifs each contain a U-G sequence (S1E Fig). Npl3, and particularly RRM2, was reported to show strong in vitro binding to U+G rich sequences including U-G dinucleotides [46].
To identify functional targets of Npl3, we assessed the transcriptome-wide effects of the loss of the protein on steady-state RNA levels, using strand-specific tiling arrays. Npl3 was reported to be highly abundant (78,700 copies per cell) [47] and has many different targets, which might show differential binding to residual Npl3 following depletion or relocation. We therefore analyzed the effects of deletion of the NPL3 gene. Tiling array analyses and RNAPII crosslinking were determined using two independent strains in which NPL3 was deleted immediately prior to the commencement of the experiments. Wild-type (WT) and npl3Δ strains were grown to logarithmic phase, RNA was extracted and reverse transcribed to make cDNA, which was then hybridized to tiling arrays. Normalized probe intensity data for all detected transcripts can be found in S1 Table.
Total RNA was extracted from WT and npl3Δ yeast strains, and equal amounts of cDNA were hybridized to strand-specific tiling arrays. Differential expression analysis identified 1391 mRNAs with significantly altered expression (adjusted p-value <0.05), of which 1229 were decreased and 162 were increased (Fig 2A and 2B). S2 Table shows differential expression analysis for all mRNAs, snoRNAs, CUTs and SUTs. The opposite effect was observed for CUTs, with 410 showing significantly increased expression, and only 8 showing significantly decreased expression (Fig 2C and 2D and S2 Table). Increased expression was also observed for snoRNAs; 33 showed significantly altered expression, 31 of which were increased in the mutant strain (Fig 2E and 2F and S2 Table).
To gain an understanding of how lack of Npl3 might lead to a global decrease in mRNA abundance, we ranked all mRNAs by log2 fold change in the mutant compared to the WT strain, according to the differential expression analysis (S2 Table). We then focused our analyses on the 30 most down-regulated genes in the npl3Δ strain, and examined their genomic environment (Table 1). As expected, the most down-regulated gene was NPL3, which is absent from the genome and was discounted from the analysis. We found that 15/30 (50%) of down-regulated genes reside in a convergent orientation with an expressed protein coding gene. A previous analysis found that only 6% of all yeast genes reside in convergent orientations in which both genes are expressed [48]. The proportion of convergent mRNAs with reduced expression in npl3Δ strains was therefore unexpectedly high. At 11 of the 15 convergent mRNA loci (73%), the down-regulated gene is adjacent to a gene that showed clear transcription readthrough, suggesting that their expression is blocked by transcriptional interference. An additional nine down-regulated mRNAs are convergent with an ncRNA that showed transcription readthrough. A further four down-regulated mRNAs are located in tandem with an upstream gene that shows readthrough, while seven mRNAs are apparently down-regulated by both tandem and convergent readthrough. Three of the 30 most down-regulated genes do not appear to be inhibited by convergent or tandem readthrough, or by intergenic transcription. Of these, YJR015W is seemingly down-regulated due to transcription changes over a local chromosome domain, since both upstream tandem genes are also down-regulated, while FMP48 and TPO4 are down-regulated by unknown mechanisms.
Although mRNA expression was most frequently decreased in npl3Δ strains, several mRNAs were up-regulated. We examined the genomic environment for the top 30 up-regulated genes (S3 Table). Eleven of these correspond to spliced ribosomal protein genes, and increased intron signal in the npl3Δ strain accounts for the differential expression. A further eleven up-regulated genes showed increased readthrough from upstream mRNAs or ncRNAs, suggesting that apparent increased expression is due to readthrough signal from the neighboring gene rather than specific up-regulation. The remaining eight genes (HSP12, DDR2, HES1, YDR124W, YML007C-A, ALP1, PUG1 and YCL049C) are apparently specifically up-regulated in npl3Δ strains.
To investigate whether the gene expression changes observed are indeed due to transcriptional interference, we more closely analyzed two strongly down-regulated mRNAs: THO1 and PTC7 (Figs 3 and S2, respectively). In Fig 3A, panels II and III show tiling array expression data for two biological replicates of the npl3Δ strain (upper) and WT (lower) strains in a genome viewer format. Panels I and IV show corresponding data for the association of RNAPII with the nascent transcript as determined by UV-crosslinking and analysis of cDNAs (CRAC; see below). Features on the Watson strand are shown above the chromosomal nucleotide numbers and features on the Crick strand are shown below. Apparent readthrough from the VHR2 gene is associated with strong down-regulation of THO1, which encodes a nuclear pre-mRNA binding protein (Fig 3A). Strand-specific reverse transcription (RT), followed by qPCR confirmed that the VHR2 gene was indeed extended, and that the increased downstream expression was not a distinct transcription product (Fig 3B). Quantification by RT-qPCR indicated that 3’ extended VHR2 is elevated ~5 fold, whereas THO1 expression is reduced ~5 fold. The approximate positions of RT primers and qPCR amplicons are shown by green arrows and red lines, respectively, in Fig 3A.
Similar analysis of the UPF2-PTC7 region revealed that apparent readthrough from the UPF2 gene is associated with strongly reduced expression of PTC7, encoding a Type 2C serine/threonine protein phosphatase (PP2C) (S2A Fig). In this case, RT-qPCR quantification revealed ~10 fold elevated readthrough from UPF2, associated with ~6 fold suppression of PTC7 expression (S2B Fig). This suggests that transcription termination defects in the npl3Δ strain lead to changes in expression of surrounding genes.
It remained possible that changes in RNA abundance for the npl3Δ strain observed in tiling array and RT-qPCR data might reflect reduced pre-mRNA surveillance and degradation rather than altered transcription. To discriminate between increased readthrough and RNA stabilization, we assessed changes in RNAPII occupancy following loss of Npl3. To do this, we used CRAC to crosslink RNAPII to the nascent transcript, which provides genome-wide, strand-specific, nucleotide resolution mapping data in vivo in growing cells. The CRAC technique was applied using strains in which the largest subunit of RNAPII, Rpo21, carried a C-terminal, His6-TEV-Protein A (HTP) tag, as recently described (Milligan et al., submitted). Tagged Rpo21 was well expressed in WT and npl3Δ strains, and was shown to crosslink efficiently to RNA (S3A Fig). Total RNAPII occupancy across different classes of RNA was largely unchanged between the WT and npl3Δ strains (S3B Fig). However, significant differences in the location of RNAPII were observed for individual genes. In Figs 3A and S2A, blue plots show Rpo21 occupancy in WT yeast and red plots show occupancy in npl3Δ. The density of RNAPII was highest at the 5’ ends of most protein-coding genes, consistent with published NET-seq data that maps the transcribing polymerase by sequencing 3' ends of associated nascent transcripts [38], and with the distribution of pre-mRNA binding factors, including Npl3 ([37,40] and Fig 1). Differences in RNAPII occupancy at the two convergent loci are summarized in Figs 3C and S2C. RNAPII occupancy within the VHR2 ORF was comparable between the two strains, and RNA accumulation was very similar in the mutant and WT strains (Fig 3A and 3C). However, in the CUT557 region immediately downstream, RNA accumulation was increased 4.7 fold while polymerase occupancy was increased 1.8 fold in npl3Δ. RNAPII crosslinking in the region between CUT557 and the downstream gene HOR2 was also elevated by 2.1 fold in the mutant, indicating that transcriptional readthrough extends into this region. HOR2 itself appears to be inhibited by transcriptional interference acting in tandem, as shown by decreased RNA accumulation (to 30% of WT), and polymerase occupancy (decreased to 50% of WT). The THO1 transcript is greatly reduced in npl3Δ (to 10% of WT), with polymerase occupancy reduced to 20% of WT.
Analysis of expression and RNAPII occupancy over the UPF2-PTC7 locus also confirmed UPF2 readthrough and PTC7 down-regulation (S2A and S2C Fig). In addition, RNAPII density was decreased over the downstream PPE1 gene. This indicates that the transcriptional readthrough from UPF2 also inhibits expression of this tandem, flanking gene. Down-regulation of PPE1 can only be determined from the RNAPII occupancy data and is not evident from tiling array data as the PPE1 signal is obscured by the UPF2 readthrough signal. This demonstrates the difficulty in discriminating down-regulation due to readthrough in tandem. We conclude that transcriptional readthrough of multiple mRNA genes results in down-regulation of downstream convergent and tandem genes.
To determine whether correctly processed and polyadenylated mRNAs are also produced from genes showing transcriptional readthrough, we analyzed the 3' end of UPF2 in WT and npl3Δ by cleavage with RNase H using an oligo hybridizing ~250 nt upstream of the UPF2 annotated 3’ end. Cleavage reactions were performed with the gene-specific oligo, with and without the addition of oligo(dT) to deadenylate the cleavage product (S2D Fig). We observed substantially less mature polyadenylated UPF2 mRNA in the mutant (lanes 1 and 2, compared to 4 and 5), but the adenylation pattern was apparently the same (lane 2 compared to 5). This indicates that cleavage and polyadenylation of UPF2 mRNA is reduced in the npl3Δ strain, but the location of the residual activity is unaltered.
The tiling array data indicate that expression of the CYC1 gene is down-regulated in npl3Δ due to transcriptional readthrough from the convergent gene UTR1 (Table 1). CYC1 encodes cytochrome C and transcription is up-regulated on glycerol medium. WT and npl3Δ strains were grown in either glucose or glycerol medium and the level of CYC1 mRNA was quantified by RT-qPCR (Fig 3D). On glucose medium CYC1 was reduced ~5.9 fold in npl3Δ relative to WT, validating the findings of the tiling array. However, CYC1 abundance was increased 4.1 fold when the npl3Δ strains were transferred to glycerol medium, resulting in an expression level close to WT. In contrast, the level of THO1 was not increased by transfer of the npl3Δ strain to glycerol medium (Fig 3D). This demonstrates that CYC1 expression remains subject to specific transcription regulation in the absence of Npl3.
Npl3 was crosslinked to ncRNAs (Fig 1) and the npl3Δ mutation altered the expression of ncRNAs including CUTs and snoRNAs (Fig 2), suggesting that the loss of Npl3 might also affect transcription termination on ncRNA genes.
Previous work identified genes that are regulated by upstream CUTs, which inhibit transcription of the downstream mRNA, including the nucleotide biosynthesis factors ADE12 and URA2 [49]. In npl3Δ strains, CUT680 upstream of URA2 and CUT324/325 upstream of ADE12 were accumulated, accompanied by reduced expression of the downstream protein-coding gene (Fig 4A–4C). Metagene analyses show increased polymerase density at the 3' ends of CUTs, and immediately downstream, in npl3Δ compared to WT (Fig 4D). These data suggest that that Npl3 is required for normal termination of CUTs, and that without proper termination these normally unstable transcripts are not efficiently turned over by the nuclear RNA surveillance machinery.
Inspection of microarray data revealed 3’ extensions for many snoRNAs in npl3Δ strains. All H/ACA and C/D box snoRNAs were included in the analysis and, strikingly, we observed extended 3’ ends for 46 of the 51 RNAPII transcribed, monocistronic snoRNA genes, and for all five polycistronic pre-snoRNA transcripts. One gene (SNR13) could not be interpreted due to missing probes (Tables 2 and S4). Another, SNR52, is the sole snoRNA transcribed by polymerase III, and is therefore terminated through a different pathway. This leaves just three RNAPII transcribed snoRNAs that do not show readthrough: U3B (SNR17B), SNR63 and SNR85. Metagene analyses of the Rpo21 CRAC data showed increased RNAPII association towards the 3' ends of all snoRNAs in npl3Δ strains (Fig 5A).
Examples of extended snoRNAs are shown in Figs 5 and S4. The box C/D snoRNA snR60 is extended approximately 500 nt in npl3Δ and appears to terminate about 100 nt into the downstream UBX6 gene (Fig 5B). The presence of extended snR60 was confirmed by northern blot (Fig 5C). S4 Fig shows extension of the box H/ACA snoRNA snR3, determined by tiling array, and RNAPII occupancy data (S4A Fig) and confirmed by RT-qPCR (S4B Fig). Comparison of expression and RNAPII occupancy at this locus is shown in S4C Fig. The snR3 transcript appears to be extended greater than 1000 nt downstream with transcription proceeding through downstream, annotated CUT genes (CUT221/222/223).
In some cases, extension of snoRNA genes was associated with strongly reduced expression of neighboring genes. As an example, SNR3 readthrough correlates with reduced expression of EFM3 (S4A–S4C Fig). Some snoRNAs appear to be extended many kilobases, apparently utilizing the termination site of the next downstream protein gene. To confirm that snoRNA 3’ extensions result from transcriptional readthrough, we calculated “readthrough scores” for three snoRNAs (SNR11, SNR30 and SNR60) that appeared to be extended based on tiling array data, as well as SNR17B that did not appear to be extended. We calculated the sum of all RNAPII hits in the 500 nt 3’ flanking region, relative to the sum of all hits within the snoRNA sequence, and compared this ratio for the WT and npl3Δ strains. For the extended snoRNAs, Rpo21 hits in the 3’ flanking region hits were elevated 1.16 to 2.17 fold in npl3Δ, but reduced to 0.84 fold of the WT for SNR17B (Fig 5D). Overall, the magnitude of RNAPII occupancy changes downstream of snoRNAs in npl3Δ relative to WT is much less than changes in expression. We suggest that the extended snoRNA transcripts predominately reflect defects in RNA surveillance rather than processing/maturation, as we found the abundance of mature snoRNAs to be comparable in the npl3Δ mutant and WT strains (S4D Fig).
Many snoRNAs harbor a cleavage site for the endonuclease Rnt1 (RNase III) positioned downstream of the mature 3’ end (reviewed in [50]). Cotranscriptional cleavage by Rnt1 provides an entry site for 3’-exonuclease processing back to the mature 3’ end of the snoRNA, and also allows the 5’ exonuclease Rat1 to degrade the nascent transcript and terminate the transcribing polymerase [51–58]. We therefore predicted that snoRNAs possessing 3' Rnt1 cleavage sites would not exhibit readthrough in npl3Δ strains. Unexpectedly, however, there was no apparent correlation between readthrough transcription in the npl3Δ strain and the presence or absence of reported Rnt1 cleavage (S4 Table).
No extension was seen on any of the RNAPII transcribed snRNAs (U1, U2, U4 or U5) in the npl3Δ strain (Table 2). It had appeared that snRNAs and snoRNAs utilize related termination pathways [59] and a recent study found extended forms of both snoRNAs and snRNAs in strains lacking Rrp6 [31]. Furthermore, as for snoRNAs, Rnt1 cleavage sites flank the U1, U2, U4 and U5 genes [50]. However, despite these apparent similarities, there are clear differences in their requirement for Npl3.
Strains lacking Npl3 show transcription readthrough on protein coding genes, on which termination generally requires the cleavage and polyadenylation machinery, and on ncRNA genes that are terminated by the Nrd1-Nab3-Sen1 (NNS) complex. The NNS complex is implicated in termination of CUTs, snoRNAs and some mRNAs and physical interactions have been reported between Npl3 and the NNS components [60,61]. We therefore investigated whether this complex is properly recruited in npl3Δ. RNA crosslinking by Nab3 was more efficient than by Nrd1, so we focused our analyses on this protein.
To assess recruitment of the NNS complex we applied the CRAC approach to Nab3-HTP. The npl3Δ strain expressing tagged Nab3 grows very slowly (doubling time 6h), indicating a negative genetic interaction. However, Nab3-HTP was well expressed in npl3Δ and crosslinked to RNA with even greater efficiency than in the WT (S5A Fig). Crosslinking of Nab3 to different RNA classes was similar in npl3Δ and WT strains (S5B Fig). Nab3, like Npl3, binds strongly at the 5' ends of mRNA transcripts (S5C Fig) and showed a substantial frequency of non-templated oligo(A) tails (36% in two experiments) consistent with active surveillance in this region.
Inspection of the VHR2-THO1 convergent gene locus (Figs 3 and 6A) revealed strong peaks of Nab3 binding at the 5’ ends of VHR2 and THO1, reflecting the role of NNS in early termination on protein coding genes. In the npl3Δ strain the peak at the 5’ end of VHR2 was unaltered, whereas the peak on THO1 was lost due to transcription interference. A peak of Nab3 towards the 3’ end of CUT557 presumably reflects the known role of NNS in CUT termination. Notably, this peak was increased when Npl3 is absent, corresponding with the increased CUT557 expression. We conclude that the VHR2-CUT557 readthrough transcripts are likely to be terminated by the NNS pathway rather than by the CPF-CF pathway. Nab3 binding across CUTs was strongly increased in npl3Δ, particularly around the 3' ends of these transcripts and at downstream sites (Fig 6B). The increased binding of CUTs by Nab3 in npl3Δ was greater than the increased RNAPII association we observe in the mutant strain (Fig 4D) suggesting that it reflects not only increased expression of these ncRNAs, but additional non-productive recruitment of this surveillance factor to normal degradation substrates. On snoRNAs we observe a contrasting phenotype, with reduced Nab3 binding across the length of the transcript in npl3Δ strains (Fig 6C). Decreased Nab3 association with snoRNAs may be related to the apparent processing defect, since the NNS complex helps promote 3’ maturation by recruitment of the exosome [27].
Overall our Nab3 binding data suggest that readthrough transcripts are targets of the NNS complex, demonstrated by increased binding of Nab3 in the extended region in npl3Δ compared to WT. In the mutant strain we see a shift in Nab3 binding away from processing targets (snoRNAs) onto surveillance targets (CUTs and extended mRNAs). This might explain why the npl3Δ/Nab3-HTP strain displays a synergistic growth defect. Efficient recruitment of Nab3 is likely to be more critical in an npl3Δ strain, in which many surveillance targets are produced. Mild interference with recruitment due to the tag might therefore have a negative effect on growth in the npl3Δ background, despite giving no clear phenotype in the WT.
Widespread termination defects result in genome-wide expression changes. We next used individual probe intensity data from the tiling arrays to calculate the level of readthrough genome-wide. Three windows were defined for each transcript: DN100 (100 nt immediately downstream of the transcript 3’ end), DN200 (200 nt, starting immediately downstream of DN100), and TRAN (spanning the entire transcript, except for the first and last 50 nt). Median expression values (normalized probe intensities) were calculated for each and a “readthrough score” equal to DN200 / TRAN was obtained for each gene in WT and npl3Δ strains. The readthrough scores obtained for the two strains were then used to calculate readthrough ratios, comparing readthrough in the npl3Δ mutant strain to that in WT yeast (S5 Table). A ratio greater than 1 indicates higher readthrough in the npl3Δ mutant strain. All mRNAs, snoRNAs, CUTs and SUTs were considered, with the exclusion of transcripts less than 200 nt in length, or closer than 400 nt to an annotated Ensembl feature on the same strand. S6A Fig shows the distribution of readthrough across all genes in the npl3Δ strain. The dark and light blue lines show the distribution of readthrough ratios for two replicate experiments, alongside the null ratio where WT is compared to WT (red). Strikingly, most genes show some level of readthrough in the npl3Δ strain. The number of genes showing significant readthrough (false discovery rate = 0.05) ranged from 29% (1165/3961) to 37% (1468/3961), depending on the experiment.
We applied stringent filters (see Bioinformatics section in Experimental Procedures) and plotted readthrough ratios for genes passing all filters (2234) against gene expression (Fig 7A); 32% of genes showed significant readthrough (marked red; FDR = 0.05), demonstrating a requirement for Npl3 in the termination of a substantial proportion of all RNAPII genes. We observed no clear correlation between readthrough ratio and expression level. We ranked all 2234 genes by readthrough ratio (S5 Table) and compared polymerase occupancy around the 3' ends of genes with the highest readthrough rank (top 200) and the control group with a low readthrough rank (1200 genes). We found that polymerase occupancy downstream of the 3' end is higher in high readthrough genes than low readthrough genes in WT yeast (Fig 7B). This suggests that these genes show a tendency towards readthrough, even in the presence of Npl3. This effect is more pronounced in the absence of Npl3 (Fig 7C), with a greater accumulation of polymerase downstream of the 3' end of high readthrough genes.
We next sought to identify factors that might discriminate high readthrough genes from low readthrough genes. We found that readthrough correlated weakly with gene length. Longer genes were more likely to show readthrough (S6B Fig), consistent with a report showing preferential binding of Npl3 to longer genes [6]. To identify potential motifs, we compared the 3’ regions from all genes in the top and bottom groups based on the readthrough ranking. This identified UAUAUA and UAAAUA motif as strongly over-represented in low readthrough genes (Fig 7D). UAUAUA is the binding site for the pre-mRNA 3’-end processing factor Hrp1 [62] and comparison of the locations of the UAUAUA motifs showed enrichment at the expected location upstream of the pA site in low readthrough genes (Fig 7E). The enrichment of Hrp1 binding sites in genes that do not show readthrough in the absence of Npl3 strongly suggests that direct, efficient recruitment of Hrp1 can bypass the requirement for Npl3 in termination.
Gene ontology analysis showed that genes with higher readthrough were enriched for plasma membrane proteins and functions in localization and/or transmembrane transport (S6 Table). This suggests these genes are potentially co-regulated through transcription termination.
Npl3 is bound to all classes of RNAPII transcripts, with enrichment for oligoadenylated RNAs characteristic of nuclear surveillance targets. Deletion of NPL3 revealed its involvement in termination on diverse transcripts that had not appeared to share termination systems. These included many mRNAs and ncRNAs including the CUT class of lncRNAs and most snoRNAs. In contrast, no defects were seen for snRNAs, which have 3’ processing and termination pathways that appeared to closely resemble snoRNAs.
Significant transcription termination defects were seen on approximately 30% of protein coding genes in npl3Δ strains. Readthrough was associated with widespread gene expression changes due to transcriptional interference at downstream genes. This likely reflects the disruption of nucleosome positioning and/or transcription factor binding caused by passage of RNAPII through the nucleosome free regions characteristic of yeast promoters. The precise number of genes that are inhibited by this mechanism is difficult to determine accurately. In the cases of the convergent genes highlighted in the text, the phenotype is clear because the transcripts lie on opposite strands. However, actively transcribed, convergent genes are quite rare in yeast, and transcriptional interference on tandem genes may be less evident. Downstream gene expression may appear unaffected on microarrays, despite generating little functional mRNA, with downstream signal representing extended upstream gene products. From the RNAPII CRAC data it appears that sense-orientated genes some distance from a site of readthrough can display the hallmarks of decreased expression. This was shown, for example, by the decreased RNAPII peak at the 5’ end of the HOR2 gene, located downstream of the extended VHR2 transcript (Fig 3).
The widespread interference seen in the absence of Npl3 highlights the necessity for very efficient release of RNAPII at the 3’ ends of genes. In general, fold changes in RNAPII occupancy were less marked than changes in downstream transcript levels. This indicates that readthrough by a small number of polymerases can drastically alter the regulation of gene-expression. In the case of snoRNAs, it appears that low levels of transcription readthrough, as determined by accumulation of downstream RNAPII, result in high levels of extended transcripts. Normal snoRNA termination and processing require the NSS complex, which stimulates exosome recruitment [27], and Nab3 association with snoRNAs was reduced in npl3Δ strains. These observations strongly indicate that loss of Npl3 also leads to defects in snoRNA 3’ processing and/or surveillance of 3’ extended species. The relative contributions of impaired snoRNA processing versus impaired surveillance in npl3Δ mutants is difficult to assess—as is the case for many substrates for nuclear surveillance/processing factors. Distinguishing the contributions of processing and surveillance is not generally feasible when the phenotype is accumulation of extended species at steady state, and will require the development of very fast, in vivo kinetic analyses.
Termination defects seen in the absence of Npl3 were restricted to RNAPII. However, while diverse classes of RNAPII transcripts are affected, this was not the case for all transcripts of any class. To try to understand what determines this apparent variability in the requirement for Npl3, we ranked protein-coding genes by their degree of readthrough (readthrough ratio) in the absence of Npl3, and sought correlated features in protein coding genes.
A notable correlation was with the elevated presence of consensus, UAUAUA binding sites for the mRNA 3’ cleavage factor Hrp1 in the 3’ regions of transcripts with low readthrough scores (i.e. with low dependence on Npl3 for termination). We postulate that association of Hrp1 and/or other cleavage factors with the pre-mRNA is normally promoted by Npl3-mediated packaging, but this requirement can be alleviated by the presence of high-affinity RNA-binding sites. In contrast, competition between binding of Npl3 and pre-mRNA cleavage and polyadenylation factors including Hrp1 was previously reported for GAL reporter constructs [7,11]. This apparent anti-termination activity of Npl3 is the opposite of our general findings. However, it could readily be envisaged that on individual genes, Npl3 binding sites conflict with the association of specific factors. The GAL genes are not expressed under the conditions used in our analyses, making it difficult to determine whether these effects are also seen on the endogenous genes.
Readthrough ratio was weakly correlated with gene length, with longer genes more likely to exhibit termination defects when Npl3 was absent. Preferential association of Npl3 with longer transcripts as been reported [6], suggesting that these may show greater changes in pre-mRNA packaging in its absence. However, we saw no clear length dependence for Npl3 in termination on ncRNAs, which are generally shorter than mRNAs.
Several distinct, but overlapping pathways for RNAPII termination are normally used by transcripts that are extended in the absence of Npl3. On pre-mRNAs, recognition of the cleavage and polyadenylation site is linked to changes in the transcribing polymerase that make it prone to termination at downstream pause sites. This may involve Tyr1 dephosphorylation in the CTD by the Glc7 phosphatase that associates with the CPF-CF [63]. Loss of Tyr1P promotes binding of the cleavage factor Pcf11, as well as Rtt103, which in turn recruits the Rai1/Rat1 complex for the “torpedo” termination pathway. In contrast, termination of a wide range of ncRNA transcripts involves the Nrd1/Nab3/Sen1 (NNS) complex, which binds to the nascent transcript and to the RNAPII CTD with Ser5P modification, as well as the TRAMP nuclear surveillance complex and promoter proximal nucleosomes with H3K4 trimethylation ([18,22,34,36,59] reviewed in [64]). Other termination mechanisms are initiated by co-transcriptional cleavage by the RNase III homologue Rnt1 [52,65] and by formation of a transcription elongation “roadblock” due to Reb1 binding on the DNA [66].
We found no correlation between known Rnt1 or Reb1 targets and transcription readthrough in npl3Δ strains. Binding of Nab3 to the CUT lncRNAs was increased in npl3Δ strains. A simple, potential explanation might be that the absence of the, normally very abundant, Npl3 protein frees binding sites that can be occupied by other factors, including Nrd1-Nab3. However, the abundance and readthrough of CUTs were also increased in the absence of Npl3, and this may contribute to the apparent changes in Nab3 association. We propose that loss of Npl3 results in aberrant RNP formation that still permits Nab3 recruitment, but binding may be non-productive.
Npl3 was reported to directly stimulate RNAPII elongation and a mutant that disrupts this function, npl3-120, resulted in improved termination. The slower RNAPII elongation rate in npl3-120 strains may enhance termination by increasing the time available for recruitment of 3' end processing factors such as Hrp1. In contrast, an Npl3 mutant (S411A) that blocks a phosphorylation site was associated with impaired transcription termination [67]. This defect was proposed to arise from retention of the mutant Npl3 in association with the RNAPII CTD and the mRNA. However, the list of genes showing 3’ extension in Npl3S411A strains overlaps substantially with the genes showing RT in npl3Δ, indicating that Npl3 retention is not solely responsible for this phenotype. Of the 818 genes showing 3’ extension in Npl3S411A, 143 overlap with the 614 genes showing significant readthrough in npl3Δ (p-value 7.1e-16, Fisher’s exact test).
Npl3 is a highly abundant RNA binding protein that participates in many processing events and associates with all nascent RNAPII transcripts. It seems probable that its absence will result in substantial changes in the nascent RNP structure. We speculate that such inappropriately packaged RNA is associated with downstream defects in transcription termination, reflected by changes in binding by the termination factor Nab3, consequently impairing a remodeling event that promotes removal of the polymerase from the nascent transcript.
Yeast were grown in standard SD medium at 30°C unless otherwise stated. Strains and plasmids used are listed in S7 Table.
All oligonucleotides used are listed in S8 Table.
All yeast analyses were performed in strains derived from BY4741 (MATa; his3Δ1; leu2Δ0; met15Δ0; ura3Δ0) or, in the case of N-PTH-NPL3, BY4727 (MATalpha; his3Δ200; leu2Δ0; lys2Δ0; met15Δ0; trp1Δ63; ura3Δ0). N-PTH-NPL3 is a strain in which a sequence encoding a PTH (proteinA-TEV-His) tag was integrated at the 5' end of NPL3, resulting in the formation of an N-terminally tagged protein utilizing the endogenous NPL3 promoter. As the protein is N-terminally tagged in this strain, the orientation of the tag is reversed, allowing the order of protein purification steps to be retained. Generation of this strain involved inserting a URA3 marker between the NPL3 promoter and the NPL3 ORF, and then replacing the URA3 marker with a sequence encoding the PTH tag. The second PCR, amplifying the PTH tag, was performed on a plasmid expressing N-PTH-NPL3 (pRS415-NPL3-PTH), and amplified a region running from the start of the PTH tag to ~600 nt into the NPL3 ORF to increase integration efficiency.
The CRAC procedure involves purifying protein/RNA complexes, where the RNA has been covalently UV crosslinked to the protein [41]. RNA-protein complexes are purified, and RNAs are partially digested to leave only the 'footprint' bound by the protein. Linkers are then ligated to both ends and the protein is removed by proteinase K digestion. RNAs are reverse transcribed and resulting cDNAs subjected to next generation sequencing using the Illumina platform (Edinburgh Genomics).
WT and npl3Δ yeast were grown to mid-log phase (OD600 ~0.5) and cells were collected by brief centrifugation (3000 xg, for 5 min). Total RNA was isolated by a standard acidic hot phenol method and DNA was removed by treating with RNase-free DNaseI (Turbo DNA-free kit; Ambion). Reverse transcription and array hybridizations were carried out as previously described [68].
Yeast cultures were grown to mid-log phase (OD600 ~0.5) and cells were collected by brief centrifugation (3000 xg, for 5 minutes). Total RNA was isolated by a standard acidic hot phenol method and DNA was removed by treating with RNase-free DNaseI (Turbo DNA-free kit; Ambion). Single stranded cDNA was generated using gene specific primers, designed to prime from the 3' end of the transcript (to measure expression) or from ~500 nt downstream (to measure transcriptional readthrough). Reverse transcription reactions were performed using Superscript III (Invitrogen). The expression level of individual transcripts was determined by quantitative PCR using SYBR green fluorescence for detection. Relative quantities were calculated using a standard curve made with known concentrations of genomic DNA, and were normalized to levels of ACT1 in each RNA sample.
Total RNA was isolated by a standard acidic hot phenol method. For SNR60 readthrough analysis, equal amounts of RNA (10 μg) were resolved on a 1.2% agarose gel in TBE buffer and transferred onto Hybond N+ nitrocellulose membrane overnight in 6x SSC. For detection of mature snoRNAs and RNase H cleavage assay products, samples (4 μg total RNA for snoRNA detection) were resolved on an 8% acrylamide gel containing 8.3 M urea, in TBE buffer and transferred onto Hybond N+ nitrocellulose overnight in 0.5x TBE. Oligo probes were end labeled with [γ-32P] ATP and hybridized to the membrane overnight at 37°C in ULTRAhyb-Oligo (Ambion). Signals were detected using a Fuji FLA-5100.
Samples (30 μg) of RNA were annealed with 750 ng oligo-dT and/or 10 pMoles gene-specific oligo, heated to 65°C and allowed to cool slowly to 30°C. Samples were then incubated with 1 unit RNase H (Roche) at 30°C for 1 hour.
Total extract from crosslinked CRAC samples were loaded onto 4–12% NuPAGE gels and Transferred onto Hybond C nitrocellulose membrane. Following blocking in 5% milk, the membrane was incubated first inn anti-TAP primary antibody (1:5000 overnight) and then anti-rabbit secondary (1:10000 for 1 hour). Signal was visualized using the Licor Odyssey system.
All sequence data are available from GEO under accession number GSE70191.
http://www.ncbi.nlm.nih.gov/geo/query/acc.cgi?token=gdivgqmivxcpzkp&acc=GSE70191
Sequencing data were processed and quality filtered using the fastx toolkit as previously described [37]. Processed reads were mapped to the Saccharomyces cerevisiae genome (SGD v64) using Novoalign (Novocraft) with genome annotation from Ensembl (EF4.74), supplemented with non-coding sequences as previously described [37]. Reads mapping to different transcript RNA classes were determined using the pyCRAC package [17] (Figs 1A, 1C, S3B and S5B). All analyses were performed using genome SGD v64 unless otherwise stated. The distribution of hits across transcripts of different classes was determined in several ways. Firstly, to examine the distribution of proteins at the 5' and 3' ends of mRNAs, hits within 300–900 nt windows aligned to the start (TSS) and end (pA) were plotted using published scripts [37]. The top 2000 bound mRNAs for each protein were included in the analysis and average distribution was plotted (Fig 1B and 1D). A similar analysis was performed to assess binding at snoRNAs and CUTs (Figs 1E, 1F, 4D, 5A, 6B and 6C). In this instance smaller windows were used and hits per million mapped reads were plotted, rather than average distribution. Reads were aligned to the TSS or 3' ends, with flanking regions included as shown. We included all snoRNAs in the analysis, but only included CUTs > 150 nt in length. Hits at introns were also plotted using this approach (S1C and S1D Fig). As an alternative way to assess binding across transcripts, we used pyBinCollector from the pyCRAC package, which normalizes transcripts by length, dividing hits into a given number of bins (S1E, S1F, S3C, S3D, S5C and S5D Figs). Rpo21 occupancy was calculated to determine transcriptional readthrough (Figs 3C, 4C, 5D, S2C and S4C) using pyPileup from the pyCRAC package, with default settings. Hits containing unencoded 3' oligoA tails of 2 of more were determined using a reported pipeline [37,69]. These hits were then mapped to transcript groups and plotted across RNA classes as described above (Fig 1C and 1D).
All microarray data are available in the ArrayExpress database (http://www.ebi.ac.uk/arrayexpress), under accession number E-MTAB-3642. Array data can also be visualized in a genome browser heat map format (http://steinmetzlab.embl.de/tollerveyLabArray). Microarray data were aligned to SGD S.cerevisiae genome version (SGD v57). Normalization of microarray hybridizations was performed as previously described [70] and transcript boundaries shown are as published [71]. Differential expression analyses were carried out using the R-package, Limma [72], controlling for the false discovery rate arising from multiple testing [73]. Five snoRNAs were not included in the differential expression analyses due to lack of transcript boundary information (Fig 2 and S5 Table). These can, however, be viewed in the genome browser heat map.
CRAC hit data were aligned to SGD S.cerevisiae genome version (SGD v57) alongside tiling array expression data at individual loci (Figs 3A, 4A, 4B, 5B, S2A and S4A). Hits were normalized for library size by plotting hits per million mapped reads at each nucleotide. Fig 6A shows CRAC data aligned to SGD v57 without array data.
Readthrough scores were calculated for mono-exonic snoRNAs, mRNAs, CUTs and SUTs with coordinates previously defined [71]. Transcripts that are < 200 nt were excluded, as were transcripts < 400 bp upstream of another annotated transcript [71] or in Ensembl release 68. The exception to this is when an mRNA has an annotated CUT or SUT immediately downstream of the 3' end. In some instances, these annotated ncRNAs appear to correspond to upstream mRNA readthrough, and therefore these mRNAs were not filtered out. Three windows were defined for each transcript: DN100 (100 nt immediately downstream of the transcript 3’ end), DN200 (200 nt, starting immediately downstream of DN100), and TRAN (spanning the entire transcript, except for the first and last 50 nt). The median normalized probe intensities (in log2 space) for each microarray sample were calculated for each window, although windows with < 8 probes were excluded.
The readthrough score was then defined for each gene and each sample (wild-type replicate 1, wild-type replicate 2, npl3-delta replicate 1, and npl3-delta replicate 2) as the median intensity for DN200, minus the median intensity for TRAN. The difference in npl3-delta and wild-type transcriptional readthrough was determined by calculating a readthrough ratio for each gene, defined as the readthrough score for npl3-delta minus the readthrough score for wild-type. Readthrough ratios were also calculated for wild-type replicate 2 versus wild-type replicate 1, to provide an empirical null distribution and enable transcripts with a significant increase in readthrough for npl3-delta versus wild-type to be identified. The Benjamini–Hochberg procedure was used to control the false discovery rate at 0.05. For this step, the two replicate experiments were treated separately, then a stringent list of genes with elevated readthrough obtained by intersecting the results from both replicates.
A series of filters was used to exclude transcripts for which readthrough ratios may be inaccurate, either due to low expression or because of evidence of independent transcription initiation downstream. The following criteria were used: (i) there must be < 10 Cbc1 (cap-binding complex protein 1) CRAC reads in the DN100 window, (ii) TRAN median probe intensity must be > -4.88 for wild-type and npl3-delta, (iii) for npl3-delta, the median probe intensity in the DN100 window must be > 70% that of the TRAN window, and (iv) the median probe intensity in the TRAN window for npl3-delta must be at least 70% that of the same window in the wild-type sample. For filters (ii)-(iv), the mean of the two replicates was used.
Plots of Pol II distribution in regions centered on transcript 3’ ends were obtained by taking the individual Pol II CRAC read distributions for each gene, linearly transforming each gene so that its maximum value was equal to 1, and then summing at each nucleotide for the indicated set of genes (either high or low readthrough groups). We observed that genes with the very lowest readthrough ranks had a spurious negative readthrough ratio due to having increased expression in the npl3Δ strain relative to WT. To limit the contribution of these genes, we took a larger number of genes for the low readthrough group (1200 compared to 200).
Npl3 binding sites were analyzed for enriched motifs by first filtering total reads to exclude low complexity sequences, as previously described [37]. The pyCRAC package [17] was used to calculate statistical overrepresentation scores for every possible k-mer (S1D Fig) using a previously described algorithm [69]. We used pyCRAC to calculate False Discovery rates (FDRs) and selected only reads forming clusters of 5 reads or more with an FDR < 0.05 for further analysis. Reads were further filtered to include only those with one or more T-C substitution, representing a site of crosslinking, and therefore predicted to indicate genuine binding sites with greater stringency.
To identify sequence motifs that differentiate high- and low-readthrough genes, we considered the 1822 genes for which reliable readthrough scores could be established, and separated these genes into quartiles by their readthrough scores. 2234 genes were included in the genome-wide readthrough analysis, but the bottom 250 were excluded from the motif analyses as these were found to have spuriously low readthrough ratios resulting from increased expression in the npl3Δ mutant. Of the remaining 1984 genes, only those with well-defined polyA sites (1822) were included in the motif analysis. For each 6-mer nucleotide motif, we calculated the numbers of genes in each quartile that contained the motif within the region (-80 to -20 nucleotides) from the polyadenylation site (polyA site). The polyA site was defined from Pab1 CRAC data as described [37]. We then identified the motifs that were significantly enriched in the low-readthrough genes, relative to high-readthrough genes, by calculating Z-scores as described [69]. To illustrate the localization of motifs relative the polyA site, we plotted the total coverage of UAUAUA motifs as a function of distance from the polyA site, separately for the top and bottom quartile of genes ranked by readthrough scores.
|
10.1371/journal.pcbi.1006207 | Network supporting contextual fear learning after dorsal hippocampal damage has increased dependence on retrosplenial cortex | Hippocampal damage results in profound retrograde, but no anterograde amnesia in contextual fear conditioning (CFC). Although the content learned in the latter have been discussed, alternative regions supporting CFC learning were seldom proposed and never empirically addressed. Here, we employed network analysis of pCREB expression quantified from brain slices of rats with dorsal hippocampal lesion (dHPC) after undergoing CFC session. Using inter-regional correlations of pCREB-positive nuclei between brain regions, we modelled functional networks using different thresholds. The dHPC network showed small-world topology, equivalent to SHAM (control) network. However, diverging hubs were identified in each network. In a direct comparison, hubs in both networks showed consistently higher centrality values compared to the other network. Further, the distribution of correlation coefficients was different between the groups, with most significantly stronger correlation coefficients belonging to the SHAM network. These results suggest that dHPC network engaged in CFC learning is partially different, and engage alternative hubs. We next tested if pre-training lesions of dHPC and one of the new dHPC network hubs (perirhinal, Per; or disgranular retrosplenial, RSC, cortices) would impair CFC. Only dHPC-RSC, but not dHPC-Per, impaired CFC. Interestingly, only RSC showed a consistently higher centrality in the dHPC network, suggesting that the increased centrality reflects an increased functional dependence on RSC. Our results provide evidence that, without hippocampus, the RSC, an anatomically central region in the medial temporal lobe memory system might support CFC learning and memory.
| When determined cognitive performances are not affected by brain lesions of regions generally involved in that performance, the interpretation is that the remaining regions can promote function despite of (or compensate) the damaged one. In contextual fear conditioning, a memory model largely used in laboratory rodents, hippocampal lesions produce amnesia for events occurred before, but not after the lesion, although the hippocampus is known to be important for new learning. Addressing how the brain can overcome lesion in animal models has always been challenging as it requires large-scale brain mapping. Here, we quantified 30 brain regions and used mathematical tools to model how a brain network can support contextual fear learning after hippocampal loss. We described that the damaged network preserved general interactivity characteristics, although different brain regions were identified as highly important for the network (e.g. highly connected). Further, we empirically validated our network model by performing double lesions of the hippocampus and the alternative hubs observed in the network models. We verified that double lesion of the hippocampus and retrosplenial cortex, one of the hubs, impaired contextual fear learning. We provide evidence that without hippocampus, the remaining network relies on alternative important regions from the memory system to coordinate contextual fear learning.
| Brain lesions provide evidence primarily about the extent to which a brain function can persevere in the absence of the damaged region. A preserved cognitive function or behavior after lesion is generally interpreted as alternative ‘brain routes’ being still able to meet the cognitive demands [1, 2]. In contextual fear conditioning (CFC), hippocampal lesions reveal a complex relation to contextual memory, as they result in profound retrograde amnesia (of pre-lesion events), but often result in no anterograde amnesia [post-lesion events; 3, 4, 5] under defined circumstances [6], suggesting that new learning can be supported by the reminiscent regions. Evidence for hippocampal participation in CFC acquisition have been provided by manipulations ranging from pharmacological injections such as muscarinic [7] and NMDA receptors blockade [8], to optogenetic approaches [9]. Thus, although hippocampus participates in CFC learning if it is functional during acquisition, CFC learning can occur after hippocampal loss.
The CFC learning after hippocampal lesion inspired cognitively-oriented hypotheses about the content learned by non-hippocampal regions. Although these accounts differ on whether the contextual representation without hippocampus is fragmented [elemental; 10, 11] or is—still—a configural representation [reviewed in 12], it is well accepted that learning under hippocampal loss is likely to be 1) different in terms of content learned, 2) less efficient and 3) more prone to generalization and decay over time [12–14]. It is also accepted that the hippocampus has preference over the non-hippocampal regions. This accounts for the impaired CFC observed after temporary manipulations, during which hippocampus inhibits the non-hippocampal regions while unable form long-term memory [12]. However, little attention has been given to the remaining regions supporting CFC learning. Although parahippocampal cortices were pointed out as putative candidates [11], the regions supporting CFC learning after hippocampal lesion have not been empirically addressed. Investigating how these regions learn and store CFC information can help to understand the dynamics of hippocampal function and its interactions within the memory systems.
There is evidence for a large number of regions to compose the neural circuits involved in CFC [15, 16] and spatial/contextual memory [17]. Understanding how a function/behavior can still be supported after lesion requires assessing complex interactions among the remaining regions and the changes in their engagement. Network approaches assess complex brain interactions based on the representation of elements (i.e. brain regions, neurons) and connection concepts (i.e. projections, functional connectivity), and offer quantitative tools for a data-driven assessment of network characteristics related to brain structure and function [18].
Large-scale network studies based on structural and functional MRI data have been paving a solid ground in cognitive neuroscience [19, 20]. They have explored functional network topology in the brain [21] and its importance to learning [22] and emotion [23]. Network studies have also been useful in identifying crucial brain regions (hubs) for network function [24], and to identify functional network changes after traumatic brain injuries [25] and in psychiatric disorders [26, 27]. Some studies took advantage of rodent models and employed network analysis in the expression of the activity-dependent gene c-fos after remote CFC retrieval [28] and later empirically interrogated the network hubs given by the model [29]. Here, we used a similar rationale to investigate how the brain support CFC learning after hippocampal lesion. We used the phosphorylated cAMP response element binding (pCREB, active form of CREB), which is critical to learning-induced synaptic plasticity [30], as our marker of brain region engagement; and examined activation and co-activation of brain regions of hippocampectomized rats after a CFC session. Using network analysis, we examined how the post-lesion network might support CFC learning and memory. We hypothesized that different network attributes in the ‘damaged network’ could be underlying CFC learning after hippocampal lesion. Further, we performed double lesions to empirically validate group differences found in the network analysis. These double lesions tested whether the group differences revealed network changes supporting CFC learning after hippocampal loss (Fig 1).
Experiment 1 aimed to explore how CFC learning under dHPC damage changes other brain regions activity and interactivity compared to CFC learning in control rats. We compared pCREB expression levels between the groups in 30 brain regions (Fig 2) and modelled functional networks based on pCREB expression correlations. Then we employed network tools to explore differences between damaged and control groups. In Experiment 1, the rats initially underwent bilateral electrolytic lesions in the dHPC or SHAM surgery. After surgical recovery, the rats underwent a CFC training session. Half the cohort was perfused 3 h after the training session, and their brains processed for pCREB immunolabelling. CREB is a transcription factor involved in neuronal plasticity [30]. CREB is phosphorylated into pCREB after neuronal activity, which is taken in this study as a proxy of brain region engagement. The other half of the cohort was returned to the homecage and tested for contextual fear memory 48 h later. A group of immediate shock controls (Imm) was added to the cohort that was tested for contextual fear memory.
The histological examination of dHPC lesions revealed that the cellular loss was overall confined to the dorsal part of the hippocampus, with occasional lesion to the overlaying cortex due to the electrode insertion (Fig 3A). The cohort tested for contextual memory had the freezing behavior measured as memory index, and was compared among the groups. The sample size in the memory test cohort was 32 (SHAM: N = 12; dHPC: N = 12; Imm: N = 8). A bootstrapped one-way ANOVA showed a significant group effect (F2,29 = 8.822, p = 0.0011). Multiple comparisons performed with p-corrected t-tests showed higher freezing time in both SHAM (p = 0.0001) and dHPC (p = 0.0178) groups compared to Imm group, but not statistically different from one another (p = 0.4044; Fig 3B). A KS test confirmed these results, showing no difference between SHAM and dHPC samples (D24 = 0.3333, p = 0.2212) and both different from Imm group sample (SHAM: D20 = 0.917, p = 0.0001; dHPC: D20 = 0.667, p = 0.0070; Fig 3C). A Cohen’s d showed a medium effect size between SHAM and dHPC means (d = 0.630) and large effect sizes between these two groups and Imm (SHAM: d = 2.226; dHPC: d = 1.275). These results show no effect of dHPC lesion in CFC learning and are in agreement with past studies [6]. To ensure the consistency of this lack of effect, we performed two additional experiments in which we test dHPC lesions in two paradigms that are known to be impaired by dHPC lesions (i.e. water maze and CFC with post-training dHPC lesions) followed by CFC with pre-training dHPC lesion (S1 Text). These experiments, which are replications of past studies [6, 31], confirm the consistency of this (lack of) effect.
In the pCREB immunolabelling cohort, we tested whether the dHPC lesion altered pCREB expression after CFC learning in any of the studied regions by comparing the pCREB expression in each region between dHPC and SHAM groups. The Fig 3D shows the pCREB expression in each region and each group. The sample size in the pCREB expression cohort was 19 (SHAM: N = 9; dHPC: N = 10). We analyzed the pCREB-positive nuclei density by comparing each region between the groups using t-tests with bootstrap resampling. There was only one marginally significant difference showing a higher level in the SHAM group in the vSub (t = 3.699, fdr-corrected p = 0.053). All other regions did not present a significant difference. This result indicates that dHPC damage diminish the pCREB expression in the vSUB, but otherwise does not alter the overall pCREB-positive nuclei density compared to the SHAM group.
We used the pCREB data to generate correlation-based networks for the SHAM and dHPC groups. As the SHAM groups has three regions absent in the dHPC group (dCA1, dCA3 and dDG), a third network was generated as “SHAM with no dorsal hippocampus”, SHAM-nH, to allow for direct comparisons between the networks (Fig 4). For each matrix, three networks were generated considering correlations with p-values under the thresholds of 0.05, 0.025 or 0.01, respectively. The networks had a very similar connectivity density in all thresholds (SHAM networks had 92, 64 and 40 edges respectively, SHAM-nH had 74, 53 and 35 edges, and dHPC had 77, 53 and 32 edges). All networks had initially one bigger component with occasional disconnected regions and fragmented across the thresholds. As the threshold increased in rigor, the SHAM network remained with one component in the 0.05 and 0.025 thresholds, and fragmented into three components and 5 disconnected regions in the 0.01 threshold. The SHAM-nH network presented the same behavior seen in the SHAM network. The dHPC network had a bigger component in the 0.05 threshold, but fragmented into four components in the 0.025 and 0.01 thresholds (S1 Fig). Although in our study negative correlations were included as absolute values in the edge weights, no negative correlations survived the thresholds. Overall, the networks presented some visual differences in their pattern of connectivity, which we formally tested in the analyses that follow.
We first tested whether the empirical networks (SHAM, SHAM-nH and dHPC) were small-world by comparing their global (Geff) and local (Leff) efficiencies to those of randomized null hypothesis networks. We also tested if the dHPC lesion changed any of the efficiencies or the network small-worldness. The Fig 5 depicts the distribution of the empirical/randomized ratios of Geff and mean Leff for all networks and thresholds. In all cases, Geff ratios are roughly around 1, with a slight decay on the 0.01 threshold. Similarly, the mean Leff ratios are consistently above 1, with the mean and upper range of ratios increasing as the threshold increased in rigor. Equivalent integration (Geff) and robustly higher segregation (Leff) values in empirical networks compared to randomized networks is consistent with small-world networks accounts [32, 33]. These results suggest that the networks engaged in CFC learning are small-world, which is in agreement with a previous work showing small-world organization in CFC retrieval networks [28]. Further, dHPC lesion did not seem to change the dHPC network small-worldness or its levels of Geff and mean Leff compared to the SHAM and SHAM-nH networks, suggesting that the overall characteristic interactivity in the dHPC network still benefit from small-world architecture.
Hubs are defined as nodes positioned to confer the largest contributions to global network function, and are usually identified using multiple centrality metrics [24]. We considered as hub any region among the 25% most central regions in at least three of the four centrality metrics used (weighted degree, Wdg; eigenvector, Evc; closeness, Clo; and betweenness, Bet). Regions that were hubs across all thresholds were considered stable hubs. The Fig 6A and 6B shows the ranked centralities for the dHPC network and the metric intersection in the threshold 0.05 (S2–S4 Figs show all networks and thresholds). In this threshold, the SHAM network showed the regions IL and BLV as hub, whereas in the SHAM-nH the BLV and Por were hubs, and the dHPC network the hubs were the Per_36, Per_35, RSC and LAVL. The Fig 6C shows which regions were considered stable hubs across the thresholds, in each network. In the SHAM network, the IL was the only region identified as a stable hub across all thresholds. In the dHPC network, the RSC, and the Per_36 were stable hubs across all thresholds, and in the SHAM-nH network, no hub was stable across the three thresholds, but the IL was the most stable region (hub in the 0.025 and 0.01 thresholds), similar to the SHAM network. Employing connection-based and distance-based metrics to identify a hub makes more likely that the identified well-connected regions are also inter-region or inter-modular connectors. Noticeably, the dCA1 was in the upper quartile of both connection-based metrics, but not the distance-based ones, across the all thresholds (S2 Fig). These results suggest that different hubs emerged in the dHPC network. However, as the identification was descriptive, with no hypothesis test, it does not allow a priori interpretations regarding differences in the hub score between the networks. However, they are a first indication that there might be differences in the connectivity patterns between the SHAM and dHPC networks, as different regions emerged as hubs in these networks.
We addressed the hub score differences more formally and quantitatively by directly comparing the centralities between the groups in each region and each threshold using permutation test. The Fig 7 resumes the results of the permutation tests for each region, metric and threshold. Most importantly, we observed that the identified stable hubs were overall associated with significantly higher centrality levels in some metrics, comparing the dHPC SHAM-nH networks. In the dHPC network, the RSC showed significantly higher Wdg and Evc in all thresholds, and the Per_36 showed higher Evc levels in the 0.025 and 0.01 thresholds, compared to SHAM-nH network. In the SHAM-nH network, the IL showed higher Evc levels in the 0.025 and 0.01 thresholds, compared to the dHPC network (S5 Fig). Besides the stable hubs, some of the single-threshold or two-threshold hubs were also associated to significantly different centrality levels between the networks. In the dHPC network, the RSGd presented a higher Evc across all thresholds and a higher Wdg in the 0.025 and 0.01 thresholds. The LAVL had a higher Evc in the 0.025 threshold. In the SHAM-nH network, the BLV presented a higher Wdg across all thresholds, higher Bet in the 0.05 and the 0.01 thresholds, and higher Evc in the 0.05 threshold. Further, the CeM and PrL showed higher Evc, and the RSGv showed higher Bet, all in the 0.01 threshold. Some significant differences were present in non-hub regions such as BLP, vCA1, DLE and Por (higher metrics in dHPC network), and LAVM, BLA, and Por (higher metrics in the SHAM-nH network; S5 Fig). Lastly, some single-threshold hubs did not show significantly different centrality metrics in the thresholds they were considered hubs, such as LAVL, Per_35, Por and Cg1 (dHPC network) and CeL, Por (SHAM-nH network). These results provide evidence that when comparing SHAM-nH and dHPC networks, stable hubs in one network were associated to higher centrality levels relative to the other, and vice-versa. These results suggest that the CFC learning network under dHPC lesion has an increased dependence on its new hubs.
The analysis so far focused on the nodes. We next examined edge (correlation coefficients) differences between the dHPC and SHAM-nH networks. First, we compared the distribution of correlations of each matrix between groups using a two-sample KS test. We observed significantly different correlation coefficient distributions between dHPC and SHAM-nH networks in all thresholds (threshold 0.05: D151 = 0.2527, p = 0.0125; 0.025: D106 = 0.3396, p = 0.0042; 0.01: D67 = 4795, p = 0.0005; Fig 8A and 8B and S6 Fig). Next, we compared each correlation coefficient between the groups. We computed the Z-score of the group difference for each correlation coefficient and considered a score of |2| to be significant within the distribution. We observed 21 correlation differences with Z-scores above |2| (Fig 8B). In nearly 2/3 of the significant differences (15 out of 21), the stronger correlation coefficients belonged to the SHAM-nH network, and 9 of them belonged to SHAM-NH hubs in that threshold; whereas only 6 differences the stronger correlation coefficient belonged to the dHPC network, one of which belonged to a hub (Fig 8C). These results were similar across thresholds. In the 0.025 threshold, 19 out of 26 differences were higher in the SHAM-nH network (3 belonging to SHAM-nH hubs; S6 Fig), and in the 0.01 threshold, 20 out of 28 differences were higher in SHAM-nH network (9 belonging to SHAM-nH hubs). Overall, these results show that the SHAM-nH network presented a higher number of significantly stronger correlations compared to the dHPC network, many of which belonged to SHAM-nH hubs for that threshold.
The different correlation distributions and the differences in correlation strengths between the networks add support to the hypothesis of different connectivity patterns in the dHPC network. Further, it suggests that dHPC indirectly influences interactions between other regions, most of which were observed to be weakened.
The network analysis revealed some differences between the dHPC and the SHAM (or SHAM-nH) networks. Particularly, the alternative hubs emerging in the dHPC network (Per_36 and RSC) and their statistically higher centralities compared to the SHAM-nH network suggest that these regions may increase in their importance to CFC learning in the absence of hippocampus. We empirically tested this hypothesis in the next two experiments by damaging both the dHPC and one of these hubs pre-training to CFC. Our hypothesis is whether further insult to the network would compromise the necessary structure of the network to promote CFC learning.
In Experiment 2, because it was technically difficult to damage specifically the Per_36 and most animals had a significant part of the Per_35 damaged, we considered animals with lesions extending to both Per_36 and Per_35, denominating it Per. Henceforth, Per will be mentioned when Per_35 and Per_36 are considered together. During histological analysis, we excluded four rats from the dHPC-Per, two from the Per and one from the dHPC groups due to either extensive bilateral lesions to the regions surrounding Per (Temporal, Auditory, Parietal, Visual cortices, ventral CA1 or Lateral Amygdala), or no detectable dHPC and/or Per cellular loss in most slices examined. The final sample in this experiment was 38 (SHAM, dHPC and Per: N = 10/each; dHPC-Per: N = 8). In the remaining sample, cellular loss was mostly confined to the Per_36, Per_35 and to dHPC. In the dHPC and dHPC-Per groups, slight occasional damage was observed in the secondary Visual and Medial Parietal cortices overlying dHPC due to needle insertion (Fig 9A). In the behavioral analysis, the bootstrapped ANOVA showed no group difference (F = 0.842, p = 0.479; Fig 9B and 9C). The KS test showed no significant differences among groups’ distributions and the Cohen’s d values did not show any considerable effect size (Fig 9 bottom). These results indicate that neither Per or dHPC-Per lesions affect CFC learning and memory.
Previous studies observed no pre-training Per lesion effect on CFC [34, 35], despite some contradictory evidence [36]. Our results support the hypothesis that pre-training Per and dHPC-Per lesions do not affect CFC learning and memory.
During histological analysis, three rats from the RSC and one from the dHPC-RSC group were excluded from the analysis due to non-detectable cellular loss in most slices. The final sample in this experiment was 39 (SHAM: N = 10, dHPC and RSC: N = 9/each, dHPC-RSC: N = 11). The lesions affected mainly the dHPC and RSC, with frequent lesions to RSGd and occasional minor unilateral lesions of RSGv and secondary visual cortex. In the behavior analysis, the bootstrapped ANOVA revealed a main effect of group (F3,35 = 3.691, p = 0.01975), which the p-corrected t tests showed to be due to a lower freezing in the dHPC-RSC compared to that of the SHAM group (t20 = 3.315, p = 0.0270; Fig 10B and 10C). No other significant differences were observed. This result was further confirmed by the KS test, which revealed significantly different distributions between the dHPC-RSC and the SHAM samples (D = 0.609, p = 0.0303). No other differences were observed (SHAM vs dHPC: D = 0.378, p = 0.330; SHAM vs RSC: D = 0.367, p = 0.377; dHPC vs RSC: D = 0.333, p = 0.316; dHPC vs dHPC-Per: D = 0.485, p = 0.098; Per vs dHPC-Per: D = 0.374, p = 0.289). The Cohen’s d values also confirmed the above results showing a large effect size between SHAM and dHPC-RSC means (d = 1.469). Lesser effect size values were observed in the other comparisons (SHAM vs dHPC: d = 0.463; SHAM vs RSC: d = 0.75; dHPC vs Per: d = 0.338; dHPC vs dHPC-Per: d = 1.056; Per vs dHPC-Per: d = 0.598; Fig 10 bottom), although the effect size between dHPC and dHPC-RSC was somewhat large. These results show that both dHPC and RSC contribute to CFC learning, although single lesion of these regions was not sufficient to impair CFC. Further, it supports the network analysis in Experiment 1 that RSC becomes a key region in the dHPC network engaged in CFC learning.
However, a careful analysis of the lesion extensions raises two possible alternative explanations for the observed results. First, the RSC lesion extension seems to be larger in the dHPC_RSC group than in the RSC group, which raises the possibility of the unimpaired behavior in the RSC be due to a smaller lesion. Second, as lesion frequently extended to RSGd in the dHPC_RSC group, there is the possibility that RSGd lesion had an effect in the impaired behavior observed in this group.
To test the first alternative possibility, we measured the percentage of lesion in the dHPC, RSC and RSGd regions in the RSC and dHPC_RSC groups, and compared them between these groups. A t test with bootstrap resampling showed that the RSC and RSGd lesion extensions were, in fact, larger in the dHPC_RSC group than in RSC group (RSC: t = 2.252, p = 0.042; RSGd: t = 2.582, p = 0.021; S7A Fig). Next, we tested whether this difference in the lesion extension had any influence in the observed behavior. We compared the freezing among the groups just as above using the percentage of damage as co-variables by an ANCOVA. The ANCOVA still showed the group main effect (F3, 32 = 3.777, p = 0.0199), but none of the lesion extensions showed any co-varying effect (RSC: F1,32 = 0.872, p = 0.3575; RSGd: F1,32 = 0.001, p = 0.9728; dHPC: F1,32 = 2.937, p = 0.0962). These results suggest that the lesion extension of these regions had no effect on the observed behavior. One thing to be considered however, is that both RSC and dHPC had a minimum of 50% of damage because of our exclusionary data. We do not rule out the possibility that, in an unbiased sample, there could be a relation between lesion extension and behavior, but this relation seems not to be evident when lesion extension exceeds 50% of the target regions. Therefore, it is unlikely that the lack of effect in the RSC group could have been due to the observed smaller RSC lesion observed in the group in comparison to the dHPC_RSC group.
To address the second possibility, we selected dHPC_RSC individuals that had their lesions more constrained to the RSC, with minor to undetectable damage to RSGd (less than 30%, N = 4), and compared them to the other groups. If the dHPC and RSC double lesions have an effect that is independent from the RSGd lesions, the differences observed should still be observable. Bootstrapped t tests showed a lower freezing time in the strict dHPC_RSC subgroup compared to the SHAM group (t13 = 2.532, p = 0.049), whereas it did not differ from any of the other groups (dHPC: t12 = 1.717, p = 0.135; RSC: t12 = 0.992, p = 0.349; dHPC_RSC: t14 = 0.024, p = 0.971; S7B Fig). These results replicate the results above where samples included significant RSGd damage, suggesting that double lesions of the dHPC and RSC impair CFC learning and memory independently of the RSGd damage. It is important to note, however, that the RSGd presented almost as many higher centrality values as the RSC in the dHPC network compared to the SHAM-nH network. Thus, we do not rule out the possibility of RSGd also possesses some increased influence in the dHPC network and in the behavior.
The present study employed network science to investigate CFC learning in dHPC-damaged rats. A fair amount of studies have observed CFC learning in absence of dHPC [6, 12, 13], but no evidence had been provided regarding how the remaining brain systems can support CFC learning without hippocampus. Our study shows four main findings. First, we found that the CFC learning network under dHPC damage did not affect the small-worldness observed in the SHAM and SHAM-nH networks, and presented comparable levels of global and local efficiencies to the SHAM network. Second, we identified different hubs in each network, which were associated with different centrality levels between the dHPC and SHAM-nH networks. Third, differences in correlation coefficients distribution and strength suggest that dHPC indirectly influence interactions throughout the network. Fourth, by damaging the regions identified as hubs in the dHPC network, we showed that double lesion of dHPC and RSC, but not dHPC and Per, disrupt CFC learning and memory. Overall, despite the unaltered topology, dHPC network was sufficiently different such that alternative hubs emerged.
Many studies have observed small-world architecture in both anatomical and functional brain networks [37, 38]. Small-world architecture is proposed to confer optimized cost-efficiency of connections for information flow [39]; as well as protection to central regions to targeted attack, when compared to other topologies [i.e. scale free networks; 21]. Further, computational studies suggest that the concomitant high local processing and integration across distributed clusters provided by small-world networks can support information processing [40]. This is in accordance with empirical studies in somatosensory neuronal networks, which maintain their small-worldness both at ‘rest’ and after stimulation, despite the enhanced interactivity among clusters [41]. Conserved small-worldness with increased levels of integration was also observed in human studies investigating network reconfigurations during motor learning [42], working memory [43], successful visual discrimination [44], recollection [45], and emotional and motivational experiences [23]. In our study, showing that dHPC lesion did not change Geff, Leff levels nor small-worldness was informative as it is in accordance with the observed performance. The unimpaired behavior and unchanged small-world and efficiencies in the dHPC group suggest that the network still maintain an interactivity capable of supporting CFC learning.
In the present study, the RSC and Per_36 showed stable hubness in the dHPC network and presented higher centrality levels compared to SHAM-nH network. These regions are deemed as central components of the proposed antero-temporal (AT, Per_36) and postero-medial (PM, RSC) memory systems that converge to the hippocampus [46], suggesting that dHPC damage increases the importance of the ‘upstream’ regions. Albeit the validation experiments showed impaired CFC memory only in the dHPC-RSC double lesions, but not dHPC-Per, the centrality comparison supports this double lesion data. The RSC displayed more robust centrality differences, with significance in more metrics and in all thresholds. These stable centrality differences may be reflecting an increased demand over—and dependence on—the RSC in the dHPC network.
Our data also corroborates the current framework of Per and retrosplenial cortex (RSG) functions. The Per is related to recognition, affective processing and associative memory of non-spatially referenced cues [47–49], whereas the RSG is important for processing spatial, contextual information and episodic memory [46, 50]. Therefore, it is parsimonious that the CFC network under dHPC damage be more dependent on RSG than on Per.
The RSG has been considered an anatomical connector of the diencephalon, medial temporal lobe and cortices implicated in anterograde amnesia [2, 51]. A recent re-emerged interest in the RSG provided a diverse number of evidences highlighting its function. For instance, studies in humans showed increased activity in RSG for stable landmarks when navigating in virtual reality environments [52–54]. Studies in animal models provided evidence that RSG integrates, encodes and stores spatial information [55–57], that it is necessary during spatial navigation [58, 59] and context fear learning and memory [60–62]. This framework suggests that the RSG is an important component of spatial learning and memory systems. Furthermore, RSG is highly interactive with regions known to be involved in spatial and contextual learning such as hippocampus [63] and Por [64]. The present results are in line with these findings and suggest that in dHPC absence, contextual learning networks might increase their dependence over the RSG.
On a different perspective, increased activation and functional connectivity are hallmarks in patients with traumatic brain injury [TBI; 65]. In resting-state fMRI, regions exhibiting increased connectivity generally compose network rich clubs, which include the RSG and Per [reigons defined as PCC, and ParaHipp, respectively; 65, 66]. Although our data is specific to CFC learning after dHPC lesion, our findings are in line with the human evidence of increased connectivity of RSG and Per after brain injury. There has been some evidence of a positive relationship between the increased functional connectivity in the pre-frontal cortex and behavior (i.e. working memory), which has generally been interpreted as an adaptive compensation [67–69]. However, this positive relationship was also observed to not be maintained after sustained working memory practice, despite the maintained increased functional connectivity during the task [70]. In fact, increased functional connectivity and inefficient behavioral performance is very often observed [71, 72], conflicting with the compensation account. Another account proposes that the ‘hyperconnectivity’ observed after brain injury expresses an overload in the alternative reminiscent pathways still capable of supplying the cognitive demand under a “challenged” circumstance [25, 73]. According to the hyperconnectivity account, performance can be impaired or not depending on the cognitive demand of the task and the level of overload in the reminiscent pathways. The relationship between behavior and increased functional connectivity (i.e. centrality levels) in RSG in our double lesion experiments could be interpreted as evidence for a compensatory account. However, CFC is a very brief experience and, although we did not evaluate dHPC networks over time or after multiple experiences, other studies observed decrease of memory across time [13] and certain conditions that yield an impaired behavior, interpreted as a less efficient learning [6]. Our results complement the existing evidence of CFC learning after dHPC loss and help shape a framework that is in accordance with the hyperconnectivity account.
We also observed an indirect influence of dHPC lesion on interactions among other regions, which is consistent with both simulation of functional brain activity under brain damage [74], and studies on unilateral focal brain lesions [75, 76]. This non-local alteration in connectivity was associated with behavioral impairments in patients. Although we did not observe a contextual fear memory impairment, the altered pattern of connectivity observed gives support to a partially different CFC learning network under dHPC damage, and suggests that what is learned (associated to the shock) might be different under dHPC lesion.
Importantly, the lack of effect on pre-training lesions involving Per should not be taken as evidence against its involvement in CFC. As RSC and Per in this study, pre-training hippocampal lesions do not impair CFC either, under most conditions [6]. Further, post-training lesions to all these regions resulted in impaired CFC memory [6, 36, 61, 77] and after pharmacological manipulations [8, 78, 79], evidencing that these regions do play a role in CFC. Our hypothesis was focused on whether CFC learning would still be supported after further targeted network damage.
Moreover, previous studies employing pre-training single lesions on both Per and RSG have reported conflicting results regarding their effect on CFC. On Per lesions, one study reported impaired CFC memory in Per-damaged animals [36], whereas other reports did not find impairment [34, 35]. These studies employed different lesion methods and behavioral parameters, rendering it difficult to point a source of the discrepancy. Although the present study employed methods closer to that of Bucci and colleagues (2000), the conflicting results remain. Regarding RSG, Keene and Bucci [60, 80] have consistently observed impaired CFC memory in pre-training RSG lesions, whereas another study did not find such impairment [81]. Our procedures were as similar as possible to that of Keene and Bucci [60], however, we aimed for the RSC instead of the whole RSG. Although we did damage portions of RSGd in some animals it is possible that our lack of effect on RSC single lesions was due to not damaging the whole RSG. Alternatively, it is possible that Per and RSG single lesions may be at least partially compensated just as dHPC lesions, resulting in higher rates of mixed results due to a less effective learning [12].
Despite the unimpaired behavior in dHPC-damaged animals, it is very likely that the contextual information learned is different [3, 14]. Some authors discussed about the complexity of the CS under hippocampal damage [11, 12], however, clearly assessing the content learned as CS in CFC preparations remains as a limitation. Findings from tasks that allow a better assessment of the learned content strongly suggest that both Per and RSG support configural learning—defined as complex stimuli bound together in a stimulus-stimulus manner. For instance, Per-damaged rodents have impaired complex visual discrimination tasks [82, 83], and RSG-damaged rodents have impaired spatial memory in tasks in which spatial cues moved between trials [84, 85]. Further, RSG was shown to integrate distributed spatial information across delimiting marks [55]. These data suggest that RSC and Per can support some configural learning in dHPC-damaged animals. This is supported by studies employing whole-hippocampus damage and complex maze tasks [86].
There are some points about the present study that need attention when interpreting the results. First, we do not show both behavioral and brain activation data in the same subjects in Experiment 1, therefore not directly showing that the altered brain networks are linked to preserved CFC learning. The purpose of our study was to investigate how learning and memory formation about CFC occurs in the absence of the dHPC, however, a limitation of the imaging methods used in rodents to assess experience-driven proxies of brain activation (i.e. IEGs such as c-fos and arc) is that they are acquired post-mortem and, thus, require to be the last step in the experimental design. Previous studies used post-retrieval IEG acquisition in order to acquire both brain and behavior data from the same subjects [28]. However, such design would need us to assume that brain activation during CFC learning and retrieval are equivalent, which there is evidence showing the contrary [87, 88]. Further, IEGs and pCREB expressions are primarily related to neuronal plasticity processes [30], and acquiring post-retrieval pCREB expression would not necessarily reflect a measure brain activation during retrieval, but possibly processes that could induce ‘post-retrieval plasticity’ [i.e. new encoding, reconsolidation; 89]. Given this limitation, we ensured that our brain network and behavior results are not product of variability by doing additional experiments challenging this hypothesis. Our Experiments 2 and 3 test whether the network analysis could be a product of chance (variability included), and we ran two additional experiments that replicate previous studies [5, 6] showing paradigms impaired by dHPC lesion followed by unimpaired CFC in rats with pre-training dHPC lesions (S1 Text). Despite this caveat, the confirmation experiments show the consistency of our observations and the relation between the altered brain networks and preserved CFC learning in dHPC damaged rats.
Second, when comparing networks, one seeks differences attributable solely to the network structure, but network attributes such as number of nodes, edge density and mean degree might add confounding effects if they are not equivalent between the networks. However, altering networks so that they match in these attributes (ex. proportional thresholding, node removal) might change the network topology and drive false positive and/or negative differences [90]. As no formal solution exists for comparing networks of differing sizes, in our study, we removed nodes from the SHAM network (producing SHAM-nH) in order to compare it to dHPC network. This node removal did not affect the overall network topology or its small-worldness (Fig 5), minimizing possible biases from this procedure. Additionally, the thresholds applied in the present study intended to remove correlations that were no different from chance, which depended solely on the pCREB co-variation among the regions within each condition. Therefore, there was no proportional thresholding to control edge density and mean degree across conditions. The fact that we did observe equivalent edge densities across conditions was due to their similar co-variations. This characteristic reflected in our analysis (similar efficiencies) and can be observed when plotting the networks based on Euclidean distances (S1 Fig). Importantly, the thresholding could have produced networks with differing edge densities or mean degree, which would require alternative approaches to the analysis.
Third, the lesion method used in Experiment 1 (electrolytic lesion) does not spare fibers of passage, which may have affected connections between other regions. Whilst this could have altered the network more than intended, the behavior data suggests that the network is likely to contain the elements required in CFC learning and memory since no impairment was observed. Furthermore, the networks studied here, which are based on pCREB expression, identified similar hubs to recent anatomical studies based on larger tract-tracing databases [91, 92], making a confounding effect of fiber lesion unlikely.
Fourth, the Experiment 1 differs from Experiments 2 and 3 in number of shocks during the training session. Single shock CFC sessions is generally a weaker experience and tend to yield more variable levels of behavior. We used the three shocks procedure to ensure a robust performance level in Experiments 2–3 such that impairments would be more detectable. Additionally, the performances of SHAM controls and dHPC groups were very similar, ruling out the possibility of a ‘hidden’ memory impairment in the dHPC group in Experiment 1.
There is growing interest in the use of network approaches to predict cognitive performance from brain imaging data [22, 44, 45]. However, formally testing predictions in human experimentation is still a challenge called for attention [93]. We applied network analysis in rodent models and were able to empirically test the validity of these models subsequently. We found that CFC network under dHPC damage increase its dependence on new hubs, and further damaging these new hubs may compromise the formation of the functional network necessary for CFC learning and memory. Future employment of finer techniques (i.e. optogenetics, transgenic animals) may provide sophisticated ways to test network predictions.
A hundred and thirty nine male Wistar rats weighting 300-370g were obtained from the university vivarium (CEDEME, SP). They were housed in groups of 4–5 and maintained on a 12h light/dark cycle, room temperature of 22 ± 2°C, with free access to food and water.
This research made use of one hundred and thirty nine male Wistar rats obtained from the university vivarium (CEDEME, SP). All experiments were approved by the University Committee of Ethics in Animal Research (CEUA, approval numbers #0392/10, #409649 and #7683270116). The guidelines used by CEUA are in accordance with National Institutes of Health Guide for the Care and Use of Laboratory Animals in the USA.
The present study involved stereotaxic surgeries. In these procedures, the rats were anesthetized with Ketamine (90mg/kg, Ceva, Paulínia, Brazil) and Xilazine (50mg/kg, Ceva, Paulínia, Brazil) given in intraperitoneal injections. In the end of the behavioral procedures, the rats were anesthesized with 10% chloral hydrate and perfused transcardially with 4% paraformaldehyde.
The rats were anesthetized with Ketamine (90mg/kg, Ceva, Paulínia, Brazil) and Xilazine (50mg/kg, Ceva, Paulínia, Brazil), and mounted into a stereotaxic frame (David Kopf Instruments, Tujunga, CA). Each animal had their scalp incised, retracted and the bregma and lambda horizontally adjusted to the same plane. Small holes were drilled in the skull in the appropriate sites. The rats received bilateral electrolytic lesions in the dHPC by an anodic current (2 mA, 20 s) passed through a stainless steel electrode insulated except for about 0.7 mm at the tip. The following coordinates were used: - 4.0 mm from bregma (AP), ± 2.0 and ± 4.0 mm from the midline (ML) and -3.6 mm from the skull surface (DV). Control (SHAM) animals underwent the same procedure except that they did not receive currents. After the surgery, the rats received antibiotic and diclofenac intramuscularly (3mg/kg, Zoetis, Madison, NJ) and were allowed to recover for 15 days. To avoid corneal lesions associated to the anesthetic used, the rats had their eyes hydrated with ophthalmic gel (Bausch & Lomb, Rochester, NY) and received a post-surgery injection of yohimbine (2mg/kg, Sigma, St. Louis, MO).
In Experiment 2 the surgeries were performed as above, but the rats received bilateral neurotoxic lesions in the dHPC, Perirhinal cortex (Per), both (dHPC-Per) or SHAMs. The lesions were made by N-methyl-D-aspartic acid (NMDA, 20 mg/ml in 0.1 M phosphate buffered saline, pH 7.4; Sigma, St. Louis, MO) injected by a 10 μl syringe held by a microinjector (Insight, Ribeirão Preto, Brazil) and connected to 27 gauge injecting needles by polyethylene tubes. In the dHPC, 0.45 μl of NMDA was injected at a rate of 15 μl/min in each of the following coordinates: (1) AP: - 2.8 mm, ML: ± 1.5 mm and DV: - 3.6 mm; (2) AP: - 4.2 mm, ML: ± 1.5 and ± 4.0 mm and DV: - 4.0 mm. In the Per, 0.1 μl of NMDA was injected (0.1 μl/min) in each of the following coordinates: AP: - 2.6, - 3.5, - 4.4, - 5.4 and - 6.5 mm, ML: ± 5.9, ± 6.1, ± 6.1, ± 6.5 and ± 6.4 mm, DV: - 7.4, - 7.4, - 7.4, - 7.2 and - 7.0. The needle remained in place for an additional 3 min. The post-surgical procedures were identical to those in Experiment 1.
In Experiment 3, surgeries were performed as in Experiment 2, but for lesions of the dHPC, disgranular retrosplenial (RSC), both (dHPC-RSC) or SHAMs. In the RSC, 0.2 μl of 20 mg/ml NMDA was injected (0.1 μl/min) in the following coordinates: AP: - 3.0, -4.0, - 5.0, - 6.0 and - 7.3 mm, ML: ± 0.4, ± 0.4, ± 0.5, ± 0.7 and ± 0.8 mm, DV: - 0.8, - 1.0, - 1.0, - 1.1 and - 1.5 mm.
We used a fear conditioning chamber (32 x 25 x 25 cm, Med Associates, St. Albans, VT) equipped with Video Freeze System. The chamber was composed of aluminum (sidewalls), polycarbonate (front wall and ceiling), white opaque acrylic (back) pieces and a grid floor of stainless steel rods (4.8 mm thick) spaced 1.6 cm apart. A sound-attenuating chamber with fans (60 dB) provided background noise and white house lights enclosed the chamber. After each animal, the chamber was cleaned with 10% ethanol.
Before every experiment, all animals were gently handled for 3 consecutive days.
In Experiment 1, during the training session, the rats were individually placed into the conditioning chamber for 2 min, received a 1 s, 0.8 mA footshock, and were returned to their homecage after 1 min. One additional control group of SHAM animals (Imm) was placed in the conditioning chamber, received an immediate footshock and was immediately returned to the homecage. Half of the cohort was re-exposed to the context 48h later for 5 min to test contextual fear memory. Behavior was recorded in both sessions by a micro-camera in the chamber. An experimenter blind to the grouping measured the freezing behavior, defined as complete immobility except for breathing movements [94], which served as our measure of contextual fear memory.
In Experiments 2 and 3, rats were placed into the conditioning chamber for 2 min, but received three 1 s, 0.8 mA footshocks, with 30 s inter-trial interval, and were returned to their homecage after 1 min. The rest of the procedure is identical to Experiment 1, except that there was no Imm control group.
Phosphorylated CREB (pCREB) has a two-phase peak expression profile, which the latter (3–6 h) was shown to present a clearer associative learning-specific expression [95, 96]. Therefore, we used a 3h time window of pCREB expression in our study. Three hours following training in Experiment 1, half the cohort was deeply anesthetized and perfused transcardially with buffered saline and 4% paraphormaldehyde (PFA) in 0.1 M sodium buffer (pH 7.4). The brains were extracted, post-fixed in PFA, cryoprotected in 20% buffered sucrose, frozen and stored at -80°C. The brains were coronally sectioned in 30 μm thick slices in a cryostat (Leica, Wetzlar, Germany) and stored in 4 serial sets. One set was collected in glass slides and stained with cresyl violet for morphological and lesion analysis, another set was used for phospho-CREB immunolabelling and the two remaining were stored for future studies.
Immunolabelling was performed in free-floating sections using anti-phospho-CREB (1:1000, Santa Cruz, Dallas, TX) as primary rabbit polyclonal antibody. A Biotinylated goat anti-rabbit antibody (1:800, Vector Labs, Burlingame, CA) was used as secondary antibody. The reaction was revealed using the avidin-biotin peroxidase method conjugated to diaminobenzidine as the chromogen (ABC and DAB kits, Vector Labs, Burlingame, CA) as described previously [97].
The pCREB expression was measured in 30 brain regions including hippocampal, parahippocampal, amygdalar and prefrontal regions (see Fig 2) previously shown to have involvement in FC and/or context learning. The dHPC group had 27 regions measured, since dCA1, dCA3 and dDG were damaged. The regions were delimited manually using ImageJ free software. The anatomical delimitation was based on the Rat Brain Atlas Paxinos and Watson [98] as on other anatomical studies [see Fig 2; 99, 100, 101]. Images (32-bit RGB) were taken at 4X and 10X magnifications using a light microscope (Olympus, Waltham, MA), and pCREB-positive cells quantified using the automated, high-throughput, open-source CellProfiler software [102]. A pipeline was created to calculate the area of each region in mm2 and to identify stained nuclei based on their intensity, shape and size (20–150 μm2; Fig 2B and 2C). The quantification was performed bilaterally in 6 sections/region (3 in each hemisphere). The data was expressed in nuclei density (nuclei/mm2). In each region and animal, three sections quantified bilaterally were averaged and computed as the expression data.
The pCREB is known to possess both a higher baseline and a higher expression profile (around twofold) compared to c-fos, an IEG more commonly used as a proxy for neuronal activity [95, 103, 104]. Although a baseline signal close to zero is preferable in most studies, for correlation-based connectivity inference it blunts sensitivity to observe negative correlations, as a diminished expression is less observable. Detecting possible negative correlations was desired in our study, making pCREB a suitable proxy for neuronal activity. Further, pCREB has a well distinguishable expression in associative learning studies [95, 96, 103].
In all experiments, the histological examination of the lesions was performed in the cresyl violet stained slices (150 μm apart) using a light microscope (Olympus, Waltham, MA). Lesions were identified visually as presence of tissue necrosis, absence of tissue or marked tissue thinning. Animals with no bilateral lesions of the target region or with lesions present in less than half the slices analyzed were excluded. An expressive bilateral lesion (50%) of untargeted regions was also an exclusionary criterion.
We photographed the histological sections at 4X magnification from bregma -2.1 0mm to -7.10 mm (150 μm apart). Next, we quantified the percentage of damaged tissue in the dHPC, RSC and dHPC_RSC groups. Using the NIH/ImageJ open source program, we quantified the areaof spared tissue in each region and then calculated their overall volume. We quantified the intact RSC and RSGd in the dHPC group and the intact dHPC in the RSC group, then calculated the mean total volume of each of these regions and used these measures as the total intact volume for each region. For each region and subject, we calculated the percentage of spared tissue by dividing the spared volume by the total intact volume, then estimated the percentage of damaged tissue [1—spared volume].
Different from the typical neuroimaging studies in humans, which acquire multiple measurements across time (i.e. EEG, fMRI), task-dependent large-scale brain activity in experimental animals is more limited. As immunohistochemistry provides a single post-mortem measure per region per animal, inter-regional co-activation is assessed across subjects. We used the pCREB-positive nuclei density to compute the Pearson correlation coefficient between all possible pairs of regions in each group (total of 435 coefficients in SHAM and 351 in the dHPC group). As SHAM matrix has 3 regions (dCA1, dCA3 and dDG) more than dHPC matrix, a “SHAM with no dorsal hippocampal regions” (SHAM-nH) was also calculated. The network derived from this matrix served to directly compare the network of these groups. Three thresholds were applied to the correlation matrices, maintaining only coefficients with (uncorrected) two-tailed significance level of p ≤ 0.05, 0.025 or 0.01. This resulted in weighted undirected network graphs composed by the brain regions (nodes) and the remaining inter-regional correlations (edges), representing connections between the regions (Fig 3). The network analyses were performed in the networks of all thresholds.
In the cohort tested for fear memory, we compared the Total Freezing Time during memory test between the groups by three statistical tests: one-way ANOVA, Kolgomorov-Smirnov (KS) tests and Cohen’s d effect size. In the ANOVAs and KS tests, we used a bootstrap resampling. The bootstrap resampling was defined by 1) randomly resampling the sample, with replacement of subjects by others (from the sample), 2) calculating the statistics of interest (i.e. Fresampled) and 3) repeating it many times (10000). It generates an empirical sample-based artificial distribution of the statistics of interest under the null hypothesis, and allows to test if the empirical data statistics (Fempirical) differs from random null hypothesis distribution. The p-value was calculated as the frequency of Fempirical occurring in the resampled distribution [p = (Fresampled > Fempirical)/10000]. There is no normality (or any other) assumption to bootstrap resampling tests, allowing comparisons when the population distribution is not normal or unknown. Multiple comparisons were assessed by t tests with bootstrap resampling, as above, correcting the p-value by the number of concomitant comparisons.
In the cohort that had their brains immunolabelled for pCREB, the pCREB expression was quantified in 30 regions (27 in the dHPC group) as positive nuclei/mm2, and each region was compared between the groups using t tests with bootstrap resampling (as above), correcting the p-value with a false discovery rate (fdr) test [106].
In the hypothesis test for small-world network, each empirical network was ‘rewired’ as described previously [107] to generate 10000 random, null hypothesis networks with the same number of nodes, edges, weights and degree distribution. Each network was rewired a number of times equal to half the number of their edges to generate the randomized networks. We calculated the Geff and mean Leff empirical/random ratio for each randomized network. It was expected for the Geff ratios to be around 1 and the mean Leff ratios to be above 1. We also evaluated group differences in Geff and mean Leff ratios. Lack of overlap between any two 95% confidence interval was considered a significant difference.
After the hub identification, we directly compared the centrality level of each region (in each threshold) between the dHPC and SHAM-nH networks using a permutation test. In the permutation procedure, we 1) randomized the grouping labels without replacement, 2) calculated the centrality values differences [Diff = CSHAM—CdHPC] in each region and 3) repeated it 10000 times. The p-value was calculated as the frequency of the empirical difference (Diffempirical) occurring in the resampled (Diffresampled) distribution [p = (Diffresampled > Diffempirical)/10000]. No comparisons with the SHAM network were performed as both networks have to be the same size.
To test whether dHPC lesion influences interactions between other regions in the network, we compared the correlation coefficients between SHAM-nH and dHPC networks. We normalized the thresholded matrices using a Fisher’s Z transformation and compared the normalized correlation coefficient distributions in the dHPC and SHAM-nH networks with a two-sample KS test. Next, we calculated the z-score of the correlation coefficient difference between each cell of the matrices as in the formula bellow, defining an index of connectivity change, as done previously [74]. The Z-score values above |2| were considered significant (corresponding to a level of significance of α = 0.05). We verified which group possessed each significantly higher coefficient, and which nodes they connect.
dC=Rsham-Rdhpc1(dfsham-3)+1(dfdhpc-3)
where df is the degree of freedom in each group.
In all analyses, a corrected-p < 0.05 was considered significant. All statistical and graph theory analyses and figures were performed in R studio [108] using custom-written routines (available at https://github.com/coelhocao/Brain_Network_analysis) and the packages igraph [109], Matrix [110], lattice [111], ggplot2 [112], corrplot [113], car [114] and VennDiagram [115].
|
10.1371/journal.pbio.1001536 | A Positive Feedback Loop Involving Gcm1 and Fzd5 Directs Chorionic Branching Morphogenesis in the Placenta | Chorioallantoic branching morphogenesis is a key milestone during placental development, creating the large surface area for nutrient and gas exchange, and is therefore critical for the success of term pregnancy. Several Wnt pathway molecules have been shown to regulate placental development. However, it remains largely unknown how Wnt-Frizzled (Fzd) signaling spatiotemporally interacts with other essential regulators, ensuring chorionic branching morphogenesis and angiogenesis during placental development. Employing global and trophoblast-specific Fzd5-null and Gcm1-deficient mouse models, combining trophoblast stem cell lines and tetraploid aggregation assay, we demonstrate here that an amplifying signaling loop between Gcm1 and Fzd5 is essential for normal initiation of branching in the chorionic plate. While Gcm1 upregulates Fzd5 specifically at sites where branching initiates in the basal chorion, this elevated Fzd5 expression via nuclear β-catenin signaling in turn maintains expression of Gcm1. Moreover, we show that Fzd5-mediated signaling induces the disassociation of cell junctions for branching initiation via downregulating ZO-1, claudin 4, and claudin 7 expressions in trophoblast cells at the base of the chorion. In addition, Fzd5-mediated signaling is also important for upregulation of Vegf expression in chorion trophoblast cells. Finally, we demonstrate that Fzd5-Gcm1 signaling cascade is operative during human trophoblast differentiation. These data indicate that Gcm1 and Fzd5 function in an evolutionary conserved positive feedback loop that regulates trophoblast differentiation and sites of chorionic branching morphogenesis.
| Abnormal placental development during pregnancy is associated with conditions such as preeclampsia, intrauterine growth restriction, and even fetal death in humans. Here we focus on the earliest steps of placenta formation, which involves the development of the labyrinthine layer, a specialized epithelium that sits between the maternal blood and fetal blood vessels and facilitates the exchange of nutrients, gases, and wastes between the mother and fetus. Pivotal to the development of a functional labyrinth layer are the processes of folding and branching of a flat sheet of trophoblast cells (originally the outer layer of the blastocyst), and of trophoblast cell differentiation. Here, we show in mice that Frizzled5, a receptor component of the Wnt signaling pathway, and Gcm1, an important transcription factor for labyrinth development, form a positive feedback loop that directs normal placental development. We find that Gcm1 up-regulates Fzd5 specifically at branching sites and that elevated Fzd5 expression in turn maintains expression of Gcm1. Moreover, Fzd5-mediated signaling is required for the disassociation of cell junctions and for the up-regulation of Vegf expression in trophoblast cells. Finally, with implications for human disease, we demonstrate that the FZD5-GCM1 signaling cascade operates in primary cultures of human trophoblasts undergoing differentiation.
| The placenta is a temporary organ first formed during pregnancy that is essential for the survival and growth of the fetus in eutherian mammals. Abnormal placental development is often associated with intrauterine growth restriction, preeclampsia, and even fetal death in humans [1]–[3]. The development of placenta starts at embryonic day 4.5 (E4.5) in mice, when the formation of different trophoblast cell types is underway. By around E10.5, a placenta with complete structure has formed. The mature placenta is composed of three major layers: the outermost layer is comprised of trophoblast giant cells and is adjacent to maternal decidua; spongiotrophoblast cells form a layer between the labyrinth and outer giant cells, and the innermost layer is the labyrinth layer, a layer important for the exchange of nutrients, gases, and wastes between the mother and fetus. Development of the labyrinth is divided into three stages: chorioallantoic attachment at E8.5, initiation of branching in trophoblast cells at the base of the chorionic plate, and branching morphogenesis and vascularization in the chorionic plate. Disturbance to any one of these stages would lead to an impaired labyrinth development, resulting in failure of pregnancy. The Glial cells missing–1 (Gcm1) gene lies at a key step in labyrinth development, since its expression in clusters of trophoblast cells at the base of the chorion define the initiation of branchpoints [4]. While Gcm1 expression appears autonomously before chorioallantoic attachment, the maintenance of its expression during subsequent branching is dependent on contact with the allantois [5]. The signals that establish the initial Gcm1 pattern and which maintain it have not yet been defined.
Gene targeting experiments have shown that development of labyrinth is regulated by numerous signaling molecules [1]–[3] including the Wnt signaling pathway [6]. For example, mice with null mutation of Wnt7b, expressed in the chorion, die at mid-gestation stages due to a defect of chorioallantoic attachment [7]. Mutation of R-spondin3, a molecule that promotes the Wnt-β-catenin signaling pathway, leads to failure of branchpoint initiation in the chorionic plate [8]. Similarly, deletion of Bcl9l, a vertebrate ortholog of Drosophila legless and an essential intracellular member of Wnt pathway, also results in defective branchpoint initiation and impaired differentiation of trophoblast cells in the chorion into syncytiotrophoblast layer II (SynT-II) cells [9]. Moreover, targeted disruption of Wnt2 causes an impaired development of the labyrinth at a slightly later stage of gestation but still leading to perinatal embryo demise [10]. Defective labyrinth development has also been reported in Frizzled5 (Fzd5) mutant mice [11] though the details of the phenotype have not yet been reported. Therefore, it remains largely unknown how different Wnt ligands signal via Fzd5 receptors to regulate chorionic branching morphogenesis and/or vascularization of the labyrinth. Moreover, it is unknown how Wnt-Fzd5 signaling interacts with other essential regulators during the development of labyrinth.
In the present study, we have employed a variety of in vivo and in vitro models to address how Fzd5 regulates chorioallantoic development during placentation. We provide direct genetic evidence that an amplifying signaling hierarchy between Gcm1 and Fzd5 directs branching morphogenesis and trophoblast differentiation during placental development in both mice and humans.
To explore the pathophysiological significance of Fzd5-driven signaling during placental branching morphogenesis, we first performed in situ hybridization to examine the spatiotemporal expression profile of Fzd5 receptors in the developing placenta. Fzd5 mRNA expression was mainly detected in trophoblast cells of the chorion at E8.0, and was strikingly high at the branching points in the chorion at E9.0 (Figure 1A and Figure S1). Low levels of Fzd5 mRNA were also detected in the floating allantois at E8.0, from which the fetal vessels in the labyrinth are derived; its expression declined to undetectable levels in the allantois upon attachment with the chorion at E8.5 (Figure 1A and Figure S1). Fzd5 was also expressed in the yolk sac at later developmental stages (Figure S1), consistent with a previous report ascribing its necessity during yolk sac angiogenesis [11]. This spatiotemporal expression profile of Fzd5 suggests that Fzd5-coupled signaling may play a role during early placental labyrinth development.
To unveil the physiological significance of Fzd5 during chorionic villus development, we employed global Fzd5-null mutant mouse models achieved by crossing Fzd5loxp/loxp mice [12] with Zp3-Cre+/− mice. The yolk sacs of Fzd5-null mutant placentas at E10.5 were pale and devoid of blood vessels, with severely retarded fetal growth (Figure 1B). These defects are consistent with previous observations showing that Fzd5 is essential for yolk sac angiogenesis [11]. In addition to changes in the yolk sac, the labyrinth layer of the placenta was also significantly underdeveloped. Attachment of the chorion and allantois occurred normally in Fzd5 mutants with normal expression of vascular cell adhesion molecule–1 (VCAM-1) and α4 integrin at E8.5 (Figure S2), which are required for chorioallantoic attachment [13]–[15]. However, the initiation and progression of branching morphogenesis in the chorion failed to occur at E9.5 in Fzd5 mutants (Figure 1C). Immunostaining analysis of cytokeratin, which marks the placental trophoblast cells, and laminin, which stains the blood vessel endothelial cells, clearly revealed that the chorion remained flat and the primary villous branches did not initiate (Figure 1D).
The defective chorioallantoic branching was associated with altered trophoblast proliferation and differentiation. In control (Fzd5+/−) placentas, trophoblast cells lining the branchpoint site in the chorionic plate ceased proliferation and mitotic division, showing no staining for either Ki67 or phospho-histone H3 (pH3), consistent with previous findings [16]. By contrast, most trophoblast cells within the Fzd5−/− chorion still underwent proliferation showing positive staining for Ki67 and pH3 (Figure 1E), reinforcing the notion that Fzd5 deficiency derails the normal initiation of branching morphogenesis.
Since the development of chorionic villi involves both the chorionic trophoblasts and the blood vessels from the allantois, both of which express Fzd5, to ascertain the relative contribution of chorionic versus allantoic Fzd5 in branching morphogenesis, we established a Cyp19-Cre+/−/Fzd5loxp/loxp mouse line by intercrossing Fzd5loxp/loxp mice with hemizygous Cyp19-Cre mice [17],[18] to achieve conditional deletion of the Fzd5 gene in the trophoblast cells. Fzd5 gene can be selectively deleted only in the trophoblast cells, while no depletion of Fzd5 in embryonic tissues (fetal endothelial cells) and yolk sacs was observed (Figure S3A and B). Upon conditional deletion of Fzd5 in trophoblast cells, retarded fetal growth and blocked branching morphogenesis was observed, similar to that observed in Fzd5-null mutants (Figure 2A and Figure S3C and D). Laminin-stained and Fzd5-intact fetal blood vessels failed to penetrate into the trophoblast-specific Fzd5-null chorion, unlike the well-interdigitated maternal-fetal interface in the control (Fzd5loxp/loxp) chorioallantoic plate (Figure 2A). These data suggest that trophoblast-expressed Fzd5 is essential for the normal labyrinth development.
On the basis of this finding, we surmised that the placental defects in Fzd5-null conceptus would be corrected regardless of the genotype of embryo proper upon complementation with wild-type 4N trophoblasts via tetraploid aggregation assay. Indeed, when diploid Fzd5-null embryos were aggregated with wild-type tetraploid Egfp+/− embryos and analyzed at E12.5, the GFP-expressing cells contributed exclusively to the trophoblast cells of the placenta and endoderm of the yolk sac (Figure S4), rescuing the development of diploid Fzd5−/− embryos (Figure 2B). By histological and laminin immunostaining analysis, we further observed that the labyrinth, and its maternal-fetal vascular network developed normally in the tetraploid Egfp+/−/Fzd5−/− conceptuses (Figure 2C). These findings highlight the requirement for trophoblast-expressed Fzd5 during labyrinth development.
It is known that Gcm1 is expressed in small clusters of chorion trophoblast cells at the flat chorionic plate stage as early as E7.5 [19]–[21] and known to determine the sites where branching initiates [4]. To search for the underlying molecular basis intrinsic to chorionic defects in Fzd5 mutants, we speculated that there may be regulatory hierarchy between Gcm1 and Fzd5 during chorionic branching initiation. To test this idea, we examined expression patterns of Gcm1 and Fzd5 in wild-type, Gcm1−/−, and Fzd5−/− placentas. Double in situ hybridization analysis revealed a partially overlapping expression pattern of Gcm1 and Fzd5 in wild-type chorions (Figure 3A). Before the chorioallantoic attachment stage, patchy expression of Gcm1 but not Fzd5 was detected in chorion trophoblasts (Figure 3A). At E8.75 and E9.0, when branching is initiated at the basal layer of the chorionic plate, the Gcm1 expression was mainly located at the tips of branchpoint sites and lining the branching folds. Expression of Fzd5 was difficult to detect and scattered in the chorionic plate before E8.5. However, after chorioallantoic attachment Fzd5 expression became detectable in the chorion but in two distinct regions—in some scattered cells in the apical region as well as in the basal at the branchpoint sites where Gcm1 is expressed and branching is initiated between E8.75 and E9.0 (Figure 1A and Figure 3A). In Gcm1 mutant mice, we found a decrease or lack of Fzd5 expression in the basal chorion at E8.5, while Fzd5 expression at the apical region of the chorion appeared normal (Figure 3B). Interestingly, we noted that Gcm1 expression was almost diminished in Fzd5 mutants (Figure 3C). These interesting findings suggest that Gcm1 up-regulates Fzd5 specifically in chorion trophoblast cells at the sites where branching occurs, while this elevated Fzd5 expression in turn maintains Gcm1 expression at the branchpoints.
To better understand the sequence of events that are regulated by Gcm1-Fzd5 function, we analyzed the molecular regulatory machinery governing the syncytiotrophoblast development, trophoblast cell junction disassociation, and blood vessel development from the allantois in Fzd5 mutants.
In mice, there are two layers of syncytiotrophoblast cells, SynT-I and -II, that lie between the fetal vessels and maternal blood sinuses in the labyrinth, with Syn-I cells lying closest to the maternal blood sinuses and with SynT-II cells lying closest to fetal endothelial cells. In Fzd5 mutants, in addition to down-regulation of Gcm1 expression in the chorionic plate at E8.5–10.5 (Figure 4A), expression of Gcm1 target genes syncytin b (Synb) and CCAAT/enhancer binding protein α (Cebpa), which are all localized to SynT-II cells [21]–[23], was significantly down-regulated (Figure 4A). Moreover, basal chorionic trophoblast cells did not fuse to form syncytiotrophoblast layer II and a functional labyrinth layer failed to form in Fzd5 mutants (Figure S5A and B). By contrast, expression of Syna, a marker for Gcm1-negative Syn-I cells [21]–[23], was apparently normal in Fzd5 mutants (Figure 4A). These results suggest that Gcm1-directed trophoblast differentiation and syncytialization is greatly hampered in Fzd5 mutants
Observations of dynamic morphological changes in chorion trophoblast cells at branchpoint (Figure 1C) prompted us to further explore the cellular events that occur during the initiation of branching morphogenesis. Trophoblast cells at the base of chorionic plate are epithelial-like cells, which are linked to each other by cell adhesions, including tight junctions (Figure S6A and B). Dissociation of tight junctions is required for epithelial cells to be transformed to mesenchymal cells [24]. Therefore, we explored the status of the tight junction protein zonula occluden 1 (ZO-1) in the developing chorion by immunofluorescence staining. ZO-1 proteins were localized to the apical side of the trophoblast cells throughout the base of the flat control chorionic plate at E8.5 right before the initiation of branching morphogenesis (Figure 4B), whereas its expression was significantly down-regulated specifically at the branching sites at later times when branching began. By contrast, we noted with interest that ZO-1 expression was sustained in chorion trophoblast cells of Fzd5 mutants even at E9.5 (Figure 4B). In addition, claudin 4 and 7 underwent similar changes to that of ZO-1 upon Fzd5 deletion (Figure S6A and B). These observations suggested that Fzd5/Gcm1 is essential for cell junction disassociation, an early step during chorionic branching morphogenesis.
Coincident with chorionic branching in wild-type placentas, the villi are immediately filled with vessels from the allantois. It is conceivable that the invasion of fetal vessels from the allantois into the space created by chorionic branching may be attracted and facilitated by vascular endothelial growth factor (VEGF) [25]. Expression of Vegf164 mRNA was readily detectable in chorion trophoblast cells at E9.0 in control placentas. However, Vegf164 mRNA expression was extremely low in Fzd5 mutants (Figure 4C), suggesting that fetal vessel infiltration is impaired by Fzd5 mutation.
Since Fzd5 deficiency leads to aberrant expression of Gcm1, ZO-1, and VEGF in the chorionic plate and all these genes are known to be regulated by canonical Wnt pathway in other systems [9],[26],[27], we speculated that Fzd5 may mediate the canonical Wnt activation in trophoblast cells during branching morphogenesis. To test this hypothesis, we performed immunohistochemistry to localize β-catenin and active β-catenin in the E9.0 placentas. While the expression of β-catenin was detected in trophoblast cells in both the control and Fzd5-null chorionic plate (Figure 5A), nuclear accumulation of active β-catenin was only detected in chorionic trophoblast cells lining branching folds in control placentas, but not in trophoblast cells of Fzd5-null placentas. Moreover, active β-catenin was also detected in the allantois, and its expression was not affected by Fzd5 deletion (Figure 5B), reinforcing the notion that trophoblast-expressed Fzd5 is essential for normal branching morphogenesis. Nonetheless, this finding provides a new line of evidence suggesting that Fzd5-mediated canonical Wnt pathway plays a role during chorioallantoic development.
Since recent evidence shows that Gcm1 can be regulated by canonical Wnt pathway in human BeWo choriocarcinoma cells [9], we tested whether Fzd5-mediated canonical Wnt signaling would directly regulate Gcm1 expression during trophoblast differentiation and syncytialization in mice. The PGL3-Gcm1 constructs, which contain binding sites for LEF/TCF, were transfected into HEK293T cells and mouse trophoblast stem (TS) cells. One binding motif (CTTTGTA: −3,661 bp) in the promoter region of Gcm1 was found to be activated by LiCl and CHIR99021, activators of canonical Wnt pathway (Figure 6A and B and Figure S7). Quantitative RT-PCR analysis further revealed that Gcm1 expression was dramatically decreased in Fzd5−/− TS cells (Figure 6C), as well as the extent of trophoblast syncytialization (Figure 6D and E), whereas CHIR99021, which bypasses Frizzled receptors [28], could largely restore the Gcm1 expression and trophoblast syncytialization in Fzd5−/− TS cells (Figure 6C–E). These results indicate that Fzd5-mediated canonical Wnt signaling is essential for normal Gcm1 expression during trophoblast cell differentiation. However, a question remains as to which Wnt ligand(s) can signal through Fzd5 receptor during placental development.
Since a reciprocal interaction between allantoic mesoderm and chorionic trophoblast is critical for branching morphogenesis [13]–[15], we surmised that Wnt genes expressed in the allantois would contribute to Fzd5 activation during the initiation of branching morphogenesis. In assessing potential candidates, we performed in situ hybridization analysis of Wnt2, Wnt5a, and Wnt7b gene expression in the developing placentas. We observed that Wnt2 mRNA was expressed in the allantois before chorioallantoic attachment and to the endothelial cells of the fetal blood vessels at later stages after chorioallantoic attachment. While Wnt7b was localized to the base of the chorion plate, Wnt5a was detected in both the chorion and allantois (Figure S8). Since the trophoblast cells in the chorion begin to differentiate until the attachment of allantois, we surmised that Wnt2 may be a prospective ligand for Fzd5 to trigger trophoblast differentiation in the chorion. To test this hypothesis, we overexpressed Wnt2 in TS cells. Overexpression of Wnt2 significantly upregulated Gcm1 expression (Figure 6F) in control (Fzd5+/−) TS cells, but not in Fzd5−/− trophoblast cells, accompanied by intracellular β-catenin (active β-catenin) accumulation (Figure 6G and H). However, adding of IWP-2, a small-molecule inhibitor interfering with the ability of cells to produce active Wnt proteins [29]–[32], reduced the levels of Gcm1 expression and active-β-catenin significantly in both control and Fzd5-null TS cells (Figure 6G and H), suggesting that Wnt2 is at least one potentially important ligand directing the normal trophoblast differentiation in the chorion during placental development.
As in mice, GCM1 regulates trophoblast syncytialization in humans [33], and so we examined if WNT2-FZD5 signaling was also involved. FZD5 mRNA was detected in both cytotrophoblasts and syncytiotrophoblasts in human placental villi (Figure 7A and B), and its expression was up-regulated in cultured primary cytotrophoblast cells undergoing spontaneous fusion into syncytiotropblasts (Figure 7C). Moreover, WNT2 was primarily expressed in cytotrophoblast cells of the human villous (Figure 7A and B) and its expression was increased during spontaneous fusion of primary cytotrophoblast cells (Figure 7C). To determine whether FZD5 would regulate trophoblast syncytialization in humans, we employed siRNA against different sequences within the FZD5 mRNA in human BeWo cells. Upon down-regulation of FZD5 expression after introducing siRNA (Figure 8A), expression of GCM1 was down-regulated (Figure 8B). Moreover, FZD5 siRNA also reduced the expression of Syncytin 1 (Figure 8B), a downstream target gene of GCM1, which mediates the fusion of cytotrophoblast cells into syncytiotrophoblast cells [34],[35]. Since up-regulated GCM1 and Syncytin1 are required for forskolin (FK)-induced fusion of BeWo cells [33], we subsequently examined whether silencing of FZD5-mediated signaling would hamper FK-induced cell–cell fusion of BeWo cells. Indeed, we noted that FZD5 siRNA largely abolished the cell fusion events in FK-treated BeWo cells (Figure 8C–F). These findings suggest that the role of FZD5-GCM1 signaling in regulating trophoblast syncytialization is conserved from mouse to human.
The labyrinth layer of the placenta is the only site for exchange of nutrients, gases, and wastes between the maternal and fetal circulations from midgestation to term. Chorioallantoic attachment is the first step during labyrinth development, but soon thereafter, primary villi begin to develop at specific sites along the basal surface of the chorion that quickly become lined by fetoplacental blood vessels from the allantois. Defects in these processes are one of the most common causes of midgestation embryonic lethality. However, much remains unclear about the mechanisms. We provide here genetic, molecular, pharmacological, and physiological evidence that an amplifying feedback loop between Gcm1 and Fzd5 is essential for normal initiation of branching and trophoblast differentiation in the chorion of mice (Figure 9). Moreover, our studies reveal that this signaling axis is also functional in the human placenta.
Previous studies have proposed that the trophoblast cells at the branching sites within the chorion express Gcm1 and that changes in cell shape—thinning and elongation—are involved in driving the branching morphogenesis [16]. Deletion of Gcm1 in mice leads to a complete block to branching at the chorioallantoic interface. We observe a similar phenotype of impaired chorionic branching in Fzd5 mutant mice, even those lacking only trophoblast-expressed Fzd5. Fzd5 is expressed in clusters of cells in the apical as well as the basal chorion, though only the latter sites correlate with branching morphogenesis and overlap with Gcm1. Gcm1 expression precedes that of Fzd5 in the chorion and it was of great interest to note that Gcm1 deficiency remarkably attenuates Fzd5 expression at the site of branchpoint initiation in the basal chorion. This implies that Gcm1 regulates the onset of Fzd5 expression in the basal chorion, but we assume that this is not due to direct transcriptional activation by Gcm1 since the regulatory elements of Fzd5 gene have no Gcm1 binding motifs (unpublished observation) [35]. Therefore, it is possible that the regulation of Fzd5 by Gcm1 was mediated by other indirect ways.
While Gcm1 precedes Fzd5 expression, Fzd5 is in turn essential for the maintenance of Gcm1 expression at the selected branching site. Fzd5 mutant chorions had diminished Gcm1 expression and impaired nuclear localization of β-catenin at E9.0 and beyond. Employing TS cells, we found that Fzd5 through nuclear β-catenin signaling directly regulates Gcm1 expression during trophoblast differentiation. Our findings are consistent with recent studies about two important members of canonical Wnt pathway, R-spondin3 and Bcl9l. Mutations in genes encoding R-spondin3, a protein that promotes the Wnt-β-catenin signaling pathway, result in failure of chorionic branching and reduced Gcm1 expression [8]. Bcl9l is an essential intracellular member of Wnt pathway and functions as an adaptor linking β-catenin and Pygopus. Its deficiency leads to defective branching initiation and impaired differentiation of trophoblast cells in the chorion into syncytiotrophoblast layer II (SynT-II) cells [9]. In general, these observations further testify the important role of Fzd5-mediated canonical Wnt pathway on Gcm1 regulation during placental development.
Our studies add to the understanding of the upstream events that initiate where chorionic branching will occur. However, the cellular events downstream of Gcm1 and Fzd5 that drive chorion trophoblast differentiation and morphogenesis have not previously been well documented. In mice, trophoblast cells at the basal layer of the chorionic plate are aligned and tightly adherent to each other, similar to other polarized epithelia. This epithelium integrity is maintained by cell junctions, particularly the tight junctions at the apical side of epithelium. With the initiation of branching, the trophoblast cells that express Gcm1 at the branching sites become thin and elongated, and disassociated similar to an epithelial to mesenchymal transition. However, a reduction of E-cadherin and up-regulation of vimentin has not been observed in chorion trophoblasts at the branching sites (unpublished observation). By contrast, the expression of ZO-1 and claudin4 and 7, important components of tight junctions, is dramatically reduced or diminished at the branching sites in the control chorion, whereas this down-regulation does not occur in Fzd5−/− mutants. Moreover, ZO-1 can be down-regulated by Wnt-β-catenin signaling in human colorectal carcinomas [26]. This suggests that Fzd5 is essential for down-regulation of ZO-1 and claudins expression during branching initiation.
Any defect in branching morphogenesis of the chorion results in a small labyrinth layer, thus limiting the surface area for nutrient transfer as well as the extent to which fetal blood vessels can grow into the placenta and come into proximity to the maternal blood spaces. To the untrained observer and at a superficial level, mutants with small labyrinth layers due to chorion branching defects may appear to be undervascularized, but it is critical to distinguish the actual events and determine whether the volume of the fetal vascular network in the labyrinth is simply small compared to wild-type because it is proportionally limited by a reduced villous volume, as is true in many cases [2], or whether it is disproportionately reduced. It is worthy of further investigation to determine if there are vascular defects in Fzd5 mutants. Previous studies on Wnt2 mutants described an impaired fetal vascular network in the labyrinth [10], but previous descriptions of placental vascular defects in Fzd5 mutants [11] cannot be confirmed based on our findings of primary chorion branching defects. Whether Wnt2-Fzd5 signaling is essential for regulating the subsequent vascularization of villi after the primary villous branching occurs will require further studies. However, we provide evidence here that trophoblast-expressed Fzd5 is essential for Vegf expression in the chorionic plate, coincident with the sites of primary villous branching that are filled in by vessels from the allantois. VEGF can function as a chemoattractant as well as a growth factor to promote vessel growth [25] and can be strongly up-regulated by Wnt signaling during tumorigenesis [27]. Aberrant expression of Vegf has also been shown to be associated with severely impaired labyrinth morphogenesis, although the branching can be initiated in Lkb1 and Tfeb mutant mice [36],[37]. These findings suggest that Gcm1/Fzd5 signaling initiates not just differentiation and branching morphogenesis in the chorion trophoblast but that the trophoblast may in turn regulate vascularization of the labyrinth.
In summary, we provide direct genetic evidence of an amplifying feedback loop of Gcm1-Fzd5 signaling in the chorion and propose three main conclusions: (1) Gcm1, first expressed in chorion trophoblast cells and further upregulated by canonical Fzd5 signaling, determines the branching sites and differentiation into syncytiotrophoblasts; (2) the initial events in chorion trophoblast morphogenesis include trophoblast cell cycle exit and downregulation of ZO-1 expression, inducing the disassociation of tight junctions at the base of the chorionic plate for branching initiation; and (3) Wnt-Fzd5 signaling also up-regulates Vegf expression in the chorion and may in turn promote vascularization of the primary villi in the labyrinth. Besides shedding light on the fundamental mechanisms of branching morphogenesis during placental development, the finding has high clinical relevance, since Gcm1-Fzd5 signaling cascade is operative during human trophoblast differentiation and its aberrant regulation is often associated with trophoblast-related diseases, such as preeclampsia [38]–[40].
Fzd5loxp/loxp mice, Gcm1-null mice, and Cyp19-Cre transgenic mice were generated as previously described in Drs. Hans Clevers, Gustavo Leone, and James C. Cross's groups, respectively [4],[12],[17],[18]. Enhanced green fluorescent protein (Egfp), Rosa26loxp/loxp, and Zp3-Cre transgenic mice were obtained from Jackson Laboratory. Mice were housed in Institutional Animal Care Facility according to institutional guidelines for laboratory animals. Females were mated with fertile males of the same strain to induce pregnancy (E0.5, vaginal plug). Conceptuses for RNA extraction and histology were dissected from uteri from E8.0 to E12.5 as previously described [41],[42]. For double in situ hybridization, whole implantation sites or dissected placentas were fixed with 4% paraformaldehyde (PFA) in phosphate-buffered saline (PBS) at 4°C overnight. After PBS washes, tissues were immersed in 10% and 25% sucrose in PBS, and then embedded in the Tissue-Tek OCT compound and frozen with dry ice-cooled ethanol. All tissues used for other analysis were fresh-frozen or fixed in 10% neutral buffer formalin (NBF).
Tissues of human chorionic villi at gestational weeks 7 and 8 from pregnant women undergoing therapeutic termination of pregnancy, and human placental tissues at full term (36–38 wk) were obtained from Xuan-Wu Hospital and the Department of Obstetrics and Gynecology, Peking University Third Hospital in Beijing, China. Termination of pregnancies at weeks 7 and 8 and virginal deliveries at full term were conducted as per usual clinical practice, with no specific procedures relevant to this research study. The study was approved by the Research Ethic Committee in the Institute of Zoology and that in Xuan-Wu Hospital and Peking University Third Hospital. All the pregnant women provided the written informed consent of using the placenta tissues for research work regarding the expression of Wnt signaling before they denoted the placentas. A total of three chorionic villi at weeks 7 and 8 and three placenta specimens at full term (36–38 wk) were enrolled in this study. All placental tissues were washed with ice-cold PBS and fresh-frozen or fixed in 4% PFA for further analysis.
For hematoxylin and eosin staining, isolated implantation sites or whole dissected placentas were fixed in 10% NBF, dehydrated and embedded in paraffin wax, and cut into 5-µm sections. For semithin and ultrathin resin histology, implantation sites were fixed in 2% glutaraldehyde and embedded in JB-4 epoxy resin according to the manufacturer's instructions (Electron Microscopy Sciences). Sections (semithin, 2 µm; ultrathin, 100 nm) were then cut using glass knives on a Leica RM2265 microtome. Semithin sections were stained with Toluidine blue (Amresco) and ultrathin sections were contrasted with uranyl acetate and lead citrate. For immunohistochemistry analysis, antibodies specific to Laminin affinity-isolated antigen-specific antibody (Sigma), cow cytokeratin (Dako), phosphor-histone H3 (Cell Signaling), Ki67 (Epitomics), and active-β-catenin (Millipore) were used in 5-µm thick paraffin embedded sections. A Histostain-SP Kit (Zhongshan Golden Bridge Biotechnology) was used to visualize the antigen. For immunofluorescence, antibodies specific to ZO-1 (Abcam), WNT2 (R&D), FZD5 (Abcam), Claudin4 (Anbo), and Claudin7 (Anbo) and secondary antibodies conjugated with Cy3 dyes (Jackson ImmunoResearch Laboratories) were used. To analyze the cell fusion status, immunolocalization using β-catenin (Abcam) or Alexa Fluor 555 Phalloidin (Life Technology) was performed. Immunofluorescence images were captured in a Zeiss LSM 510 confocal scanning laser microscope.
In situ hybridization with isotopes, digoxygenin (DIG), or fluorescein isothiocyanate (FITC)–labeled antisense RNA probes was performed on cryosections as described previously [21],[43]. Sections hybridized with the sense probes served as negative controls.
Tetraploid aggregation chimeras were generated as described previously [44] with some modifications. Briefly, wild-type tetraploid embryos were generated by electrofusion of two-cell embryos derived from EGFP intercrosses. Fused embryos were cultured overnight in KSOM medium. Diploid eight-cell or morula-stage embryos generated from Fzd5+/− intercrosses were collected on E2.5. Zona pellucidae were removed with acidic Tyrode solution, and each diploid embryo was aggregated with two tetraploid embryos. Aggregated chimeric embryos were allowed to develop to the blastocyst stage and were then transferred into the pseudopregnant uteri of wild-type females. Chimeric embryos were dissected at E12.5. Both the embryo proper and placenta were visualized for GFP under a dissecting microscope and subsequently fixed and stained with hematoxylin-eosin and laminin. The visceral endoderm and embryo tissue was used as a DNA source for genotyping of diploid embryos.
Total RNA was isolated from chorionic plate or placenta by using TRIZOL (Invitrogen). One microgram of total RNA was used to synthesize cDNA. Expression levels of different genes were validated by real-time RT-PCR TaqMan analysis using the ABI 7500 sequence detector system according to the manufacturer's instructions (Applied Biosystems). All primers for real-time PCR were listed in Table S1. Assays were performed at least three times with each in duplicate.
The coding sequence region for Wnt2 was amplified from mouse placenta and cloned in to PWPI expression vector. The expression of Wnt2 was confirmed by transfecting into HEK293T cells followed by RT-PCR or Western blot. The upstream region of the Gcm1 gene relative to the transcription start site was generated by PCR (primers listed in Table S1) using placental genomic DNA as template. The amplified fragments were cloned into pGL3-Basic vectors (Promega). Mutations of the LEF/TCF binding sites were achieved by Fast Mutagenesis System (Stratagene).
HEK293T cells were kept in DMEM medium (HyClone) supplemented by 10% serum, 1 mM sodium pyruvate, 2 mM L-glutamine. Fzd5+/− and Fzd5−/− TS cells were derived from E3.5 mouse blastocysts as described previously [45]. Established TS cells were maintained in a proliferative state in media containing 70% embryonic fibroblast-conditioned medium, 30% TS cell medium, FGF4 (25 ng/ml), and heparin (1 µg/ml). For differentiation conditions, TS cell medium was used but without supplementation with bFGF, heparin, and embryonic fibroblast pre-conditioning in the presence of CHIR99021 (Biovision) or not. All constructs were transiently transfected into HEK193T cells and TS cells using Lipofectamine LTX and PLUS reagents (Invitrogen) according to the manufacturer's instructions. pRL-TK, internal control plasmid expressing Renilla (Promega), was co-transfected into the cells to normalize firefly luciferase activity of the reporter plasmids. LiCl (Sigma) and CHIR99021 were added 24 h after transfection and cells were collected after another 24 or 48 h, for HEK293T and TS cells, respectively. Luciferase assay was performed by Dual-Luciferase Reporter System (Promega) according to the manufacturer's instructions. Assays were performed at least three times with each in duplicate.
The human choriocarcinoma BeWo cell line was obtained from American Type Culture Collection and maintained as monolayers at 37°C, 5% CO2 with F-12K/DMEM, (1∶1) medium (Gibco) supplemented with 10% fetal bovine serum, and 2 mM glutamine. For RNA interference experiments, 20 nM of siRNAs were reverse transfected into 6.5×104 BeWo cells per 500 µl in 24-well plates (or adjusted proportionally to the plate size) by Lipofectamine RNAiMAX (Invitrogen) according to the manufacturer's instructions. At 24 h after transfection, medium was changed to that containing 50 µM FK or vehicle (dimethyl sulfoxide) and collected at 72 h posttransfection for assay. Sequences for siRNA oligonuclotides are listed in Table S1. Stealth RNAi siRNA Negative Control Hi GC or Med GC was used as control siRNA.
The analysis of cell fusion is according to the procedure described previously [9]. In brief, BeWo cells with siRNA transfected in a 24-well plate were immunostained by standard procedure and then observed at a final magnification of 400×. Six microscopic fields per sample were randomly selected for examination; three independent experiments were performed.
Statistical analysis was performed with SPSS11.5 program. Comparison of means was performed using the independent-samples t test. Data were showed as means ± SEM.
|
10.1371/journal.pntd.0000686 | Co-ordinated Gene Expression in the Liver and Spleen during Schistosoma japonicum Infection Regulates Cell Migration | Determining the molecular events induced in the spleen during schistosome infection is an essential step in better understanding the immunopathogenesis of schistosomiasis and the mechanisms by which schistosomes modulate the host immune response. The present study defines the transcriptional and cellular events occurring in the murine spleen during the progression of Schistosoma japonicum infection. Additionally, we compared and contrasted these results with those we have previously reported for the liver. Microarray analysis combined with flow cytometry and histochemistry demonstrated that transcriptional changes occurring in the spleen were closely related to changes in cellular composition. Additionally, the presence of alternatively activated macrophages, as indicated by up-regulation of Chi3l3 and Chi3l4 and expansion of F4/80+ macrophages, together with enhanced expression of the immunoregulatory genes ANXA1 and CAMP suggests the spleen may be an important site for the control of S. japonicum-induced immune responses. The most striking difference between the transcriptional profiles of the infected liver and spleen was the contrasting expression of chemokines and cell adhesion molecules. Lymphocyte chemokines, including the homeostatic chemokines CXCL13, CCL19 and CCL21, were significantly down-regulated in the spleen but up-regulated in the liver. Eosinophil (CCL11, CCL24), neutrophil (CXCL1) and monocyte (CXCL14, CCL12) chemokines and the cell adhesion molecules VCAM1, NCAM1, PECAM1 were up-regulated in the liver but unchanged in the spleen. Chemokines up-regulated in both organs were expressed at significantly higher levels in the liver. Co-ordinated expression of these genes probably contributes to the development of a chemotactic signalling gradient that promotes recruitment of effector cells to the liver, thereby facilitating the development of hepatic granulomas and fibrosis. Together these data provide, for the first time, a comprehensive overview of the molecular events occurring in the spleen during schistosomiasis and will substantially further our understanding of the local and systemic mechanisms driving the immunopathogenesis of this disease.
| Schistosomiasis is a significant cause of illness and death in the developing world. Inflammation and scarring in the liver and enlargement of the spleen (splenomegaly) are common features of the disease. Changes occurring in the spleen have the potential to influence the way in which the body deals with infection but the mechanisms driving these changes are not well characterised. In the present study we determined, for the first time, the gene expression profile of the mouse spleen during infection with Schistosoma japonicum and compared these results to those previously reported for the liver to determine if processes occurring in these organs co-operate to promote hepatic inflammation and granuloma formation. Our data indicated that gene expression in the spleen is related to the types of cells present and suggest that the spleen might be important in controlling schistosome-induced inflammation. Comparison of the liver and spleen showed that expression of cell signalling molecules (chemokines) was much higher in the liver, potentially promoting the recruitment of specific cell types to this organ, causing inflammation and scarring. The results from this study enhance our knowledge of the mechanisms that drive schistosome-induced splenomegaly and liver inflammation.
| Schistosomiasis, characterised by extensive hepatic fibrosis and splenomegaly, is a significant cause of parasitic morbidity and mortality [1]. Although extensive studies have been carried out to identify the processes driving hepatic granulofibrotic response, the immunopathogenesis of schistosome-induced splenomegaly has been largely neglected.
Splenomegaly is a common feature of many infectious diseases and can lead to alterations in the splenic architecture as well as the inherent immunological function of the organ. Changes in the splenic architecture following Leishmania and some viral infections have been shown to influence the nature of the immune response to subsequent infections [2], [3]. Schistosome infections induce significant splenomegaly characterised by loss of definition between the red and white pulp [4], [5], [6], [7]. Additionally, schistosome infections are known to modify the nature of the immune response to a number of other pathologies, including allergic responses and other parasitic infections, by as yet undetermined mechanisms [8]. Furthermore, undefined processes occurring in the spleen during active schistosome infections enhance the granulofibrotic response occurring in the liver [5]. The precise molecular mechanisms and transcriptional modulations corresponding to these cellular and immunological changes, however, have not been fully evaluated. Characterising the molecular processes occurring in the spleen during schistosomiasis is an important research priority if we are to fully comprehend the immunopathogenesis of this disease and the mechanisms by which schistosome infections modulate the immune response to other pathogens.
The study presented here describes the use of whole genome microarray analysis combined with flow cytometry and histology, to provide a comprehensive profile of the transcriptional and cellular response occurring in the murine spleen during Schistosoma japonicum infection. As well, we compare and contrast these results with those we have previously reported for the liver during the progression of S. japonicum egg-induced granuloma formation and hepatic fibrosis [9]. Our results reveal that there is co-ordinated expression of chemokines and cell adhesion molecules in the liver and spleen that may regulate the recruitment of effector cells to the liver during schistosome infection. Additionally, we demonstrate the up-regulation of several immunomodulatory elements in the spleen that may be involved in the control of the immune response to S. japonicum infection. The results of microarray analysis, flow cytometry and histology of the livers of the S. japonicum infected mice used in the present study are available [9] and the liver gene expression data are in the public domain (NCBI's Gene Expression Omnibus; Series Accession Number: GSE14367). Taken together these results provide insight into the integrated molecular mechanisms driving the development of schistosome induced pathology.
All work was conducted with the approval of the Queensland Institute of Medical Research Animal Ethics Committee.
Full details of the time-course experiments undertaken on S. japonicum-infected mice are described elsewhere [9]. Methods related to the analysis of the spleen are outlined below. Livers and spleens from the same animals were used for the purposes of parasitological, histological, microarray and flow cytometry analyses.
Four to six week old female C57BL/6 mice (Animal Resource Centre, Canningvale, Australia) were percutaneously infected with 20 S. japonicum cercariae (Chinese mainland strain, Anhui population). Mice were euthanized at 4 (n = 7), 6 (n = 7) and 7 (n = 8) weeks post-infection (p.i.) and spleen tissue collected. Three additional mice were used as uninfected controls. An identical experiment was performed for flow cytometry (n = 5 per group). Total adult worm pairs per mouse were recorded as a measure of parasite burden and eggs per gram of liver was calculated as a measure of hepatic egg burden as reported [9], [10].
Formalin fixed, paraffin embedded spleen sections were stained by haematoxylin and eosin to assess splenic structure. Within spleen tissue, eosinophils were identified by Giemsa staining; neutrophils by Leder stain (naphthol AS-D chloroacetate) [11]; and macrophages by immunoperoxidase staining for the macrophage specific cell surface marker F4/80 (Primary antibody: rat anti-mouse F4/80, Abcam, Cambridge, USA; Secondary antibody: biotin-conjugated anti-rat immunoglobulin, Jackson ImmunoResearch Laboratories, Inc, West Grove, USA. Detection: Streptavidin-HRP, Jackson ImmunoResearch Laboratories, Inc, West Grove, USA). Slides were scanned using an Aperio slide scanner (Aperio Technologies, Vista, USA). The number of eosinophils and neutrophils in the spleen of each mouse was quantified by calculating the average number of positively stained cells in 20 high power fields (×400). Positive staining for F4/80 was measured using Aperio's Spectrum Plus Software positive pixel count algorithm (Version 8.2; Aperio Technologies, Vista, USA).
Leukocytes were isolated from whole spleens as described [12]. Briefly, spleens were digested in collagenase D (1mg/ml; Roche Diagnostics, Mannheim, Germany) and DNAse I (0.5mg/mL; Roche Diagnostics, Mannheim, Germany) for 30 mins at 37°C. The tissue was then passed through a 70µm cell strainer (BD Falcon, Bedford, USA) and washed with FACS buffer (1% bovine serum albumin (w/v), 0.1% sodium azide (v/v) in phosphate buffered saline). Red blood cells were lysed with Gey's lysis solution. The solution was then underlayed with FACS buffer and centrifuged at 1300rpm for 5 mins. The resulting cell pellet was resuspended in FACS buffer and the cells counted.
Cells were stained for specific cell markers by first incubating with anti-Fc-receptorIII antibodies (Monoclonal antibody producing hybridoma; Clone: 24.G2) to block non-specific binding and then with commercially available fluorochrome-conjugated antibodies for 30 mins on ice (APC-anti-CD4, FITC-anti-CD8b and PE-anti-CD19: BD Pharminogen; FITC-anti-CD3, Miltenyi Biotec, Germany). Cells were defined as CD4+ T-cells (CD3+/CD4+), CD8+ T-cells (CD3+/CD8+) and B-cells (CD19+). Data were acquired on a FACS Calibur Flow Cytometer (BD Bioscience) and analysed using FlowJo Software (Treestar Inc) and GraphPad Prism, version 5.0 (GraphPad Software, San Diego, USA).
Total RNA was extracted from spleen tissue as described [13]. Briefly, spleen tissue was homogenised in Trizol (Invitrogen, Carlsbad, USA) using a Qiagen Tissuelyser (Qiagen Inc., Valencia, USA). A portion of the homogenate was then processed by phase extraction with Trizol and by column chromatography using an RNeasy Mini Kit (Qiagen Inc, Valencia USA). RNA quantity was measured using the Nanodrop-1000 (Nanodrop Technologies, Wilmington, USA) and quality was assessed using an Agilent Bioanalyzer (Agilent Technologies, Foster City, USA).
Each mouse group was normalised by log transformation for egg burden and outliers were excluded on the basis of 95% confidence intervals as described [9]. An equal amount of total RNA from the spleens of four mice with the highest quality total RNA were pooled for cRNA and cDNA synthesis.
cDNA was synthesised from pooled splenic total RNA using a Quantitect Reverse Transcription kit (Qiagen Inc., USA). cDNA concentration was measured using a Nanodrop-1000 (Nanodrop Technologies, Wilmington, USA.). Real-time PCR was used to validate a subset of the microarray data. Forward and reverse primers were sourced from the literature [16], [17], [18] or designed using Primer-Blast software (http://www.ncbi.nlm.nih.gov/tools/primer-blast) (Table S1). Hypoxanthine phosphoribosyltranferase (HPRT) was used as a housekeeping gene [18]. Real time PCR was performed using SYBR Green master mix (Applied Biosystems, Warrington, UK) on a Corbett Rotor Gene 6000 (Corbett Life Sciences, Concorde, Australia). Rotor-Gene 6000 Series software (version 1.7), Microsoft Office Excel 2003 and GraphPad Prism Version 5.00 for Windows (San Diego, California USA) were used in the analysis of results. Correlations between microarray and real-time PCR results were assessed using Spearman's Rho measure of correlation in GraphPad Prism Version 5.00 for Windows.
Changes in parasitological, histological, real time PCR and flow cytometry data were assessed by One Way ANOVA with post hoc Tukey testing (p≤0.05) using the GraphPad Prism Version 5.00 for Windows (San Diego, California USA). Statistical analysis of microarray data was performed using GeneSpring GX (version 7.3.1). Correlations between microarray and real-time PCR data were measured using Spearman's Rho correlation in GraphPad Prism Version 5.00 for windows as described [19].
Details of the parasitological burden and kinetics of granuloma formation and fibrosis in the livers of the S. japonicum infected mice used in these experiments are reported elsewhere [9]. Briefly, mice were moderately infected with an average of 5 worm pairs. Schistosome eggs were first observed in the liver at 4 weeks post infection (p.i.) and hepatic egg burden increased significantly thereafter (1-Way ANOVA, p≤0.05) [9]. The kinetics of granuloma formation and fibrosis were consistent with previous reports for S. japonicum [20], [21].
Total spleen weight increased significantly from 4 weeks p.i. onwards (Figure 1, upper panel). Changes to the splenic architecture were observed as early as 4 weeks p.i. and were characterised by increasing congestion of the red-pulp associated with loss of definition between the white and red pulp (Figure 1A–D, lower panel).
The number of neutrophils in the splenic red pulp increased significantly from 6 weeks p.i. and eosinophil numbers were significantly elevated compared with uninfected mice at 7 weeks p.i. F4/80 staining for macrophages was significantly increased from 6 weeks p.i., reaching a peak of 19% total section area at 7 weeks p.i. (Figure 2A–I).
Flow cytometry for CD8+ T-cells, CD4+ T-cells and B-cells revealed an early increase in the total number of these cells at 4 weeks p.i. followed by a return to baseline levels by 6 weeks p.i. (Figure 3A). In contrast, the ratio of T-cells and B-cells to total splenocytes decreased significantly after 4 weeks p.i. (Figure 3B).
Real-time PCR was performed on a subset of genes representative of transcripts that were highly up- or down-regulated in the spleen (NE, EPX, Chi3l3, CXCL13); exhibited contrasting expression in the liver and spleen (CXCL13, CXCL1, CXCL9) and are key genes from important biological categories identified by DAVID analysis (Cell cycle: Mki67; Chemotaxis: CXCL1, CXCL4, CXCL9, CXCL13); as well as genes encoding Th1/Th2 cytokines (IFN-γ, IL-4). The results of the real-time PCR analyses correlated closely with those observed by microarray analysis (Spearman's correlation r = 0.93, p≤0.0001, n = 36) (Figure S1).
To date, there have been no studies conducted to specifically define the molecular or transcriptional processes occurring in the spleen during infection with any schistosome species. We have reported here the first such study employing microarray analysis in combination with flow cytometry and histochemistry to characterise the transcriptional and cellular profile of the spleen following schistosome infection, comparing these changes with those occurring in the liver [9].
Overall, the transcriptional response of the spleen reflected cellular changes in this organ during progression of S. japonicum infection. Significant up-regulation of proliferation markers and genes associated with the cell cycle and lymphocyte proliferation paralleled the expansion of T- and B-cells within the spleen. Similarly, the significant down-regulation of genes associated with T- and B-cell receptor signalling reflected the decrease in the relative proportion of these cells in the splenic compartments over time. The down-regulation of the cytokines that promote Th1 development from 6 weeks p.i. could favour the development of a Th2 response and reflects the shift from a Th1 to a Th2 dominant response observed in the liver at this time point [1], [8], [9]. Further, down-regulation of a large cohort of genes predominantly associated with immune responses is likely to reflect attempts by the host to regulate the immune response to schistosome infection.
Loss of B-cells/T-cells following initial expansion could be due to cell death or migration from the spleen. We did not detect elevated expression of apoptosis-associated genes in the spleen but there was decreased expression of several chemokines, especially lymphocyte chemokines such as CXCL13, CCL21 and CCL19. Combined with the accumulation of these cells in the liver [9], these results suggest that cellular loss from the spleen is more likely due to migration than apoptosis. Further, it has been shown that S. mansoni infections are able to influence hematopoietic processes occurring in the bone marrow as well as the activity of bone marrow-derived cells (e.g [22], [23]). This raises the alternative possibility that infection-induced modifications in haematopoiesis occurring up-stream of the spleen, contribute to the loss of B- and T-cells from the spleen during S. japonicum infection. Further studies to define the contribution of specific chemokines, apoptosis and alterations to the haematopoietic process within the bone marrow are required if we are to fully comprehend the mechanisms involved.
Accumulation of eosinophils, neutrophils and macrophages in the spleen was paralleled by enhanced expression of neutrophil and eosinophil markers as well as genes known to be expressed at high levels in these cells such as annexin A1 (ANXA1) and Cathelicidin antimicrobial peptide (CAMP). ANXA1 regulates polymorphonuclear leukocyte trafficking and function during innate immune responses and T-cell dependent inflammation during adaptive immune responses [24]. CAMP is known to have direct antimicrobial activity, is chemotactic for a variety of cells, and regulates production of pro-inflammatory cytokines [25]. The role of these genes in schistosomiasis has not been investigated but their increased expression in the spleen suggests that they may be involved in shaping the immune response to schistosome infections.
Significant up-regulation of Chi3l3 and increased staining for the macrophage marker F4/80 suggests that there may be an increase in the population of alternatively activated macrophages in the spleen during schistosome infection. Similarly, there was significant induction of the alternatively activated macrophage markers Retnla and Mrc1 and increases in the number of F4/80+ macrophages in the liver. In light of recent studies demonstrating immunoregulatory roles for alternatively activated macrophages during Th2 responses [26], [27], [28], it is possible that their presence in the liver and spleen during schistosome infection represents a mechanism whereby the host regulates the immune response to infection at both the local and systemic level.
Comparison of the transcriptional profiles of the liver and spleen revealed common up-regulation of genes associated with progression through the cell cycle indicating that cellular proliferation is occurring in both organs during infection. Up-regulation of components of the heme and porphyrin metabolism pathways was specific to the spleen and likely reflects increased blood volume passing through this organ associated with portal hypertension.
Up-regulation of extracellular matrix components, as well as the profibrotic cytokines TGF-β, EDN1 and PDGF-β, was enhanced in the liver but remained unchanged in the spleen reflecting the development of fibrosis [9]. Similarly, there was liver specific down-regulation of many components of several metabolic pathways including, but not restricted to, metabolism of xenobiotics, bile acid biosynthesis, fatty acid metabolism and glutathione metabolism [9]. As discussed previously [9], these results are consistent with studies of the metabolic function of the liver following schistosome infection and are indicative of decreased liver function associated with hepatic injury [29], [30], [31].
The most striking difference between the liver and spleen was the contrasting expression of genes involved in cellular recruitment (Figure 5). Several chemokines, including lymphoid homing and T-cell chemokines (e.g. CXCL13, CCL19, CCL21a–c), were significantly up-regulated in the liver [9] but down-regulated in the spleen. Further, expression of eosinophil (CCL24), neutrophil (CXCL1) and macrophage/monocyte (CCL6, CCL7, CXCL14) chemokines and the cell adhesion molecules VCAM1, NCAM1, PECAM1 was enhanced in the liver [9] but was unchanged or undetectable in the spleen. The differential expression of these genes likely contributes to the generation of a chemotactic signalling gradient promoting the recruitment of effector cells, including eosinophils, neutrophils and lymphocytes, to the liver during infection leading to the development of granulomas and fibrosis [9]. The observed loss of T- and B-cells from the spleen after initial expansion at 4 weeks p.i. is, therefore, more likely due to migration of these cells from this organ to the peripheral tissues than cell death. Similar induction of CCL21, CXCL16, CXCL9 and CXCL13 in the liver in other models of hepatic disease suggests that common mechanisms regulate the recruitment of lymphocytes to the liver following inflammation [32], [33], [34]. Down-regulation of lymphoid homing chemokines in the spleen has been implicated in alterations in lymphoid structure during Leishmania infection [2], altered motility of dendritic cells and lymphocytes following several viral infections [3], and with decreased responsiveness to secondary infection following primary infection with lymphocytic choriomeningitis virus (LCMV) or Listeria monocytogenes [3]. The down-regulation of these chemokines during S. japonicum infection may contribute to the significant changes observed in the spleen during schistosome infection and may go some way to explaining how schistosomes skew the immune response to other infections.
Comparison of the expression profiles of the liver and spleen clearly indicate that there is co-ordinated expression of chemokines in these organs during S. japonicum infection. Up-regulation of lymphocyte, eosinophil and monocyte chemokines in the liver [9], and down-regulation of the same chemokines in the spleen may contribute to the development of a chemotactic signalling gradient that promotes recruitment of these cells to the liver, thereby facilitating the development of granulomas and fibrosis. Furthermore, the down-regulation of homeostatic lymphoid chemokines, such as CXCL13 and CCL21, in the spleen could lead to disruption of the splenic architecture and the altered immune responses associated with schistosome infections. Additionally, we observed up-regulation of the alternatively activated macrophage marker Chi3l3 and the immunoregulatory molecules ANXA1 and CAMP in the spleen. These results suggest that the spleen may be an important site for the regulation of S. japonicum-induced immune responses. Together these data highlight the importance of the spleen to the immunopathogenesis of schistosomiasis and significantly enhance our understanding of the chemokine signalling pathways regulating the development of schistosome-induced granulomas and fibrosis.
|
10.1371/journal.pmed.1002519 | Effect and cost-effectiveness of educating mothers about childhood DPT vaccination on immunisation uptake, knowledge, and perceptions in Uttar Pradesh, India: A randomised controlled trial | To assess the effect of health information on immunisation uptake in rural India, we conducted an individually randomised controlled trial of health information messages targeting the mothers of unvaccinated or incompletely vaccinated children through home visits in rural Uttar Pradesh, India.
The study tested a brief intervention that provided mothers face-to-face with information on the benefits of the tetanus vaccine. Participants were 722 mothers of children aged 0–36 months who had not received 3 doses of diphtheria–pertussis–tetanus (DPT) vaccine (DPT3). Mothers were randomly assigned in a ratio of 1:1:1 to 1 of 3 study arms: mothers in the first treatment group received information framed as a gain (e.g., the child is less likely to get tetanus and more likely to be healthy if vaccinated), mothers in the second treatment group received information framed in terms of a loss (e.g., the child is more likely to get tetanus and suffer ill health if not vaccinated), and the third arm acted as a control group, with no information given to the mother. Surveys were conducted at baseline (September 2015) and after the intervention (April 2016). The primary outcome was the proportion of children who had received DPT3 measured after 7 months of follow-up. The analysis was by intention to treat. A total of 16 (2.2%) participants were lost to follow-up. The coverage of DPT3 was 28% in the control group and 43% in the pooled information groups, giving a risk difference of 15 percentage points (95% CI: 7% to 22%, p < 0.001) and a relative risk of 1.52 (95% CI: 1.2 to 1.9, p < 0.001). The information intervention increased the rate of measles vaccination by 22 percentage points (risk difference: 22%, 95% CI: 14% to 30%, p < 0.001; relative risk: 1.53, 95% CI: 1.29 to 1.80) and the rate of full immunisation by 14 percentage points (risk difference: 14%, 95% CI: 8% to 21%, p < 0.001; relative risk: 1.72, 95% CI: 1.29 to 2.29). It had a large positive effect on knowledge of the causes, symptoms, and prevention of tetanus but no effect on perceptions of vaccine efficacy. There was no difference in the proportion of children with DPT3 between the group that received information framed as a loss and the group that received information framed as a gain (risk difference: 4%, 95% CI: −5% to 13%; p = 0.352; relative risk: 1.11, 95% CI: 0.90 to 1.36). The cost per disability-adjusted life year averted of providing information was US$186, making the intervention highly cost-effective with respect to the WHO-recommended threshold of once the gross domestic product per capita (US$793 in the case of Uttar Pradesh). Key study limitations include the modest sample size for this trial, limiting power to detect small differences in the framing of information, and the potential for contamination among households.
Providing mothers of unvaccinated/incompletely vaccinated children with information on tetanus and the benefits of DPT vaccination substantially increased immunisation coverage and was highly cost-effective. The framing of the health information message did not appear to matter.
The trial is registered with ISRCTN, number ISRCTN84560580.
| Only 51% of children aged 12–23 months are fully immunised in Uttar Pradesh, an Indian state of over 200 million people.
Few studies have evaluated the effect and cost-effectiveness of health information messages to improve the uptake of childhood vaccinations, and whether the framing of the information has implications for outcomes.
We conducted an individually randomised controlled trial to test a brief intervention that provided mothers through home visits with health information messages on tetanus and the benefits of the diphtheria–pertussis–tetanus (DPT) vaccine, involving 722 mothers of children aged 0 to 36 months in Uttar Pradesh.
We framed the health information messages in 2 alternative ways: as a gain—e.g., the child is more likely to be healthy if vaccinated—and as a loss—e.g., the child is more likely suffer ill health if not vaccinated.
Seven months later, children whose mothers received the information were 52% more likely to have received 3 doses of DPT vaccine than children in the control group, but there was no effect of framing.
The cost per disability-adjusted life year averted was US$186, making the intervention highly cost-effective with respect to the WHO-recommended threshold of once the gross domestic product per capita (US$793 in the case of Uttar Pradesh).
The study suggests that providing mothers of unvaccinated/incompletely vaccinated children with health information messages on the benefits of vaccination can be a highly cost-effective means of improving child health through increases in immunisation coverage.
| An estimated 5.9 million children die each year globally, of which 1.2 million are in India [1]. The majority of these deaths are preventable with existing low-cost health technologies, such as improved water and sanitation, zinc supplementation, oral rehydration solutions, and vaccines [2]. Indeed, such interventions have contributed to remarkable improvements in child mortality in many developing countries [1]. Despite well-documented evidence on the health and developmental benefits of immunisation [3], a huge number of children fail to get vaccinated. In Uttar Pradesh, a state of more than 200 million people and the setting for this study, only 51% of children aged 12 to 23 months are fully vaccinated [4].
There are many potential reasons for inadequate levels of immunisation coverage. They include problems in the supply of vaccines as well as demand-side factors such as time costs, high discount rates, distrust, fear, and limited knowledge [5,6]. If parents underestimate the true efficacy of vaccines or are simply unaware of their existence, it is plausible that providing information could increase uptake of vaccinations. There is reason to believe that parents may be poorly informed as to the benefits of immunisation. Female literacy is far from universal in many in low- and middle-income countries (LMICs), and information problems are likely to be pervasive. Accurately inferring the risk of disease and protective effect of vaccines through observations in daily life is unrealistic. In fact, provision of health information and education to parents and community members is commonly used to try to stimulate uptake for health interventions in LMICs. Information interventions have the benefit of being low cost to implement, and their effects are thought to be sustainable if they succeed in changing behaviours. For health interventions, such as vaccinations, even short-lived changes in behaviour could lead to large health impacts because once vaccinated, the child is immunised for years to come.
A range of interventions to increase uptake of childhood immunisation in LMICs have been studied, including monetary incentives, health information and education, training of providers, outreach sessions, and home visits [7–9]. However, the handful of studies that have tested the effect of information shed little light on how information increases coverage of immunisation, its cost-effectiveness, and whether the framing of information has implications for outcomes. With the introduction of relatively new vaccines, to prevent pneumococcal pneumonia and rotavirus diarrhoea, and lagging coverage in routine vaccinations, the need for rigorous research on cost-effective ways to increase uptake of childhood vaccinations is urgently needed.
This study reports findings from a randomised trial that examined the extent to which health information messages designed to educate mothers on the benefits of the combined diphtheria–pertussis–tetanus (DPT) vaccine increased immunisation coverage.
The study received ethical approval from the Indian Council of Medical Research (HMSC/2014/10/HSR), the Public Healthcare Society in India (10/Nov/2013), and the London School of Hygiene & Tropical Medicine in the UK (8610). Mothers gave written informed consent to participate in the study. The trial is registered with ISRCTN (ISRCTN84560580).
We undertook a 3-arm randomised controlled trial to evaluate the effect of a brief information intervention on the uptake of vaccination services in 180 villages (clusters) across 6 districts of Uttar Pradesh, India, between 12 September 2015 and 29 April 2016 (S1 Text). The districts and villages were selected as part of a broader research project in which this study was embedded; these sampling procedures are described elsewhere (S2 Text) [10]. Data from the Indian Census 2011 suggest that the study clusters were demographically similar to the study districts and the state (S1 Fig).
Mothers were the intended recipient of the information intervention. They were eligible for inclusion in the study if their child was alive, was aged 0–36 months, and had not received 3 doses of DPT vaccine (DPT3) and if the mother intended to remain in the study area for at least 6 months. Eligible mothers were randomly assigned in a ratio of 1:1:1 to 1 of 3 study arms: mothers in the first treatment group received information framed as a gain, mothers in the second treatment group received information framed in terms of a loss, and the third arm acted as a control group, with no information given to the mother. A computer random number generator in CSPro (version 6.1) assigned participants to 1 of the 3 study arms during the baseline visit. The sequencing of events during the household visit was as follows: mothers were invited to give consent to participate in the study, they were interviewed for the baseline survey, they were assigned to treatment or control, and, if assigned to the treatment group, they received the information intervention.
Potentially eligible participants were identified from lists of mothers generated using 2 sources of information: (i) a representative household maternal and child health survey of 3,600 mothers conducted by the research team in the same villages 9 months prior to the start of this study and (ii) a list of mothers who had given birth in the past year provided by the community health worker (accredited social health activist [ASHA]) in each village. The ASHA in each village was identified by field staff through a home visit during the baseline survey. Field staff visited the household of each potentially eligible mother and assessed eligibility based on either vaccination cards or self-reports. A total of 722 eligible mothers were identified—459 through the previous household survey and 263 using the lists provided by ASHAs.
The study tested an information intervention that provided mothers with information on the benefits of the tetanus vaccine. The intervention was implemented by Sambodhi Research and Communications, a research organisation in Uttar Pradesh. Field staff were mostly male, had completed secondary school, and were from the same state but were not known to the communities. The information was delivered to mothers face-to-face in the privacy of their home, and field staff followed a script that they were trained to deliver in a standardised manner. Specifically, it described the causes and symptoms of tetanus, possible health consequences, the individual benefit of the combination DPT vaccine in terms of mortality and morbidity gains, and the wider community benefits associated with herd immunity.
There were 2 versions of the script that differed in the way the information was framed. The first framed the information on tetanus vaccination as gains—e.g., the child is less likely to get tetanus and more likely to be healthy if vaccinated. The second framed information on tetanus vaccination as a loss—e.g., the child is more likely to get tetanus and suffer ill health if not vaccinated. Visual aids were used to help convey the information in an accessible manner to illiterate women, and a Hindi leaflet containing the information was left with the mother (S2 and S3 Figs). A short question and answer session followed the provision of the information to ensure comprehension of the information. The intervention took about 10 minutes to deliver. The intervention design was informed by the theory behind framing [11–14], previous research on framing in health [15], and extensive piloting.
Both variants of the intervention informed mothers where in the public sector they could get their child vaccinated. The Indian Academy of Pediatrics recommends that 3 doses of DPT should be given, at 6 weeks, 10 weeks, and 14 weeks. The minimum age for this vaccine is 6 weeks. If any of these doses are missed, then the recommended age for catch-up is any time up to 7 years [16]. All study participants were asked questions during the baseline survey on the vaccination status of their child.
Data were collected at baseline in September 2015 and 7 months later at endline in April 2016. At baseline, 722 eligible mothers were interviewed. The survey tools were designed to capture the immunisation status of the child and the mother’s knowledge of the causes of, symptoms of, and prevention methods against tetanus. Immunisation status was assessed in the standard way using the vaccination card and, if not available, self-reports from the mother. The interview also included ‘games’ with chickpeas designed to elicit women’s perceptions of the efficacy of tetanus and measles vaccination, as well as several verification questions to test comprehension of these games. Data were collected on tablets using computer-assisted personal interviewing. Field staff were blinded to group assignment in the endline survey.
The pre-specified primary outcome was the proportion of children who had received DPT3 measured after 7 months of follow-up. Pre-specified secondary outcomes were the proportion of children fully vaccinated against tuberculosis, diphtheria, pertussis, tetanus, and measles; the mother’s knowledge of any symptom of tetanus; and the mother’s perception of the efficacy of tetanus vaccination. To understand better the effect of the intervention and how it worked, the secondary outcomes also included measures that were not pre-specified in the study protocol but were planned for prior to data analysis: the proportion of children with measles vaccination, the proportion of children with Bacillus Calmette–Guérin (BCG) vaccination, the mother’s knowledge of any cause of tetanus, and her knowledge of any tetanus prevention method.
Immunisation status was measured using vaccination cards, where available, or self-reports by mothers. Perceptions of the efficacy of the tetanus vaccine were obtained using interactive games, in which women were asked hypothetical questions on the chances of children in 2 villages being infected under different immunisation coverage scenarios. Chickpeas were used to elicit responses between 0 and 10, with 0 corresponding to a perception of 0% efficacy of the vaccine and 10 corresponding to 100% efficacy (S3 Text). Knowledge outcomes on causes, symptoms, and prevention methods were defined as binary variables where success corresponded to knowledge of at least 1 of the possible causes, symptoms, and prevention methods, respectively. Given the nature of the intervention, we did not anticipate, nor did we measure, any adverse events.
To assess balance at baseline, we compared the characteristics of participants across the 3 study groups. We then analysed the effect of the interventions by intention to treat using 2 pre-specified approaches. The first was a pooled analysis in which participants in the 2 intervention arms were grouped together and compared with those in the control group. The second was a treatment group analysis in which each intervention group was compared with the control group and with each other. For binary outcomes, we report the proportion in each group, the difference in proportions between groups, and the risk ratio. For continuous outcomes, we report the mean in each group and the difference in means between groups. Adjusted absolute differences were estimated using ordinary least squares linear regression of the outcome on treatment group assignment indicator(s), controlling for age of the child in months and the outcome at baseline.
We conducted several further analyses. With the exception of the first, these analyses were unplanned and were conducted after the analysis of the pre-specified outcomes. First, we studied the presence of information spillovers by exploiting random variation in the geographical density of households assigned to the treatment groups [17]. Information on the geographical coordinates of households collected during baseline was used to generate our variable of interest, measuring for each household the share (proportion) of other study households within a given radius (250 m, 500 m, and 1 km) that received the information intervention.
Second, we hypothesised that information would increase perceptions of efficacy most for those with inaccurate perceptions at baseline. We compared the mean in our measure of perception of efficacy between treatment and control for 2 subgroups of women, with baseline perception of efficacy above and below the 50% threshold. This allowed us to assess the effect of information on mothers with an inaccurate perception of the efficacy of the tetanus vaccine at baseline versus the effect on mothers with a more accurate perception of efficacy.
Third, we carried out a cost-effectiveness analysis based on effect estimates from the pooled analysis. We considered costs from a provider perspective only [18]. Costs were collected through project accounting systems and categorised as start-up or implementation costs. Costs related to research activities were not included. Calculations were based on 2 scenarios: (i) actual costs within the study and (ii) the cost of scale-up to reach the entire target population in the 6 study districts. The scale-up scenario made no changes to average costs, except that we assumed almost double the number of households could be reached each day if the intervention were delivered outside the confines of a research study. To calculate child deaths averted, we used estimates of the effectiveness of DPT3 and measles vaccines on disease-specific mortality [19,20]. Various sources provided estimates of the proportion of under-5 mortality caused by tetanus, pertussis, and measles [21,22]. We did not consider morbidity benefits, positive externalities (herd immunity), or the fact that more than 1 child in a household may have benefited. We calculated disability-adjusted life years (DALYs) by using a period life expectancy of 67.8 years at 2 years of age from the Indian Census 2011 [23] and applying a discount rate of 3% to future years of life. More details on the assumptions and underlying data are provided in S4 Text. All costs are reported in US dollars.
All analyses were conducted using Stata version 14.2 and ArcGIS 10.3. The primary analysis was done blinded to treatment group assignment.
We conducted sample size calculations recognising that the achievable sample size would be constrained by the prior decision to work in the 180 study villages. In the pooled analysis, we estimated that a sample of 465 participants (155 control, 310 pooled treatment) would provide 80% power to detect an absolute difference of 10 percentage points in the proportion of children with DPT3, assuming a 5% level of significance and a DPT3 vaccination rate of 10% in the control group. In the treatment group analysis, we estimated that a sample of 882 participants (294 in each group) would provide 80% power to detect an absolute difference of 10 percentage points in the rate of DPT3 vaccination between any 2 of the treatment groups, assuming a 5% level of significance and a DPT3 vaccination rate of 20% in the comparison group.
Between 12 and 30 July 2015, 2,359 mothers were assessed for eligibility (Fig 1). Of these, 1,637 (69%) mothers were excluded, most commonly because the child had already received DPT3. Overall, 722 participants were enrolled and randomly assigned to 1 of the 3 treatment groups: information positively framed (n = 237), information negatively framed (n = 246), or no information (n = 239). A total of 16 (2.2%) participants were lost to follow-up, resulting in a final analytical sample of 706. There were no further missing data. Attrition was similar across treatment groups.
Table 1 presents baseline characteristics of the study sample by treatment group. The mean age of children was 10 months. Two-thirds of children had had their first dose of DPT, and measles vaccine coverage was 16%. Two out of 5 mothers could identify at least 1 cause of tetanus and 1 method of prevention; however, less than 1 in 10 mothers were able to identify any of the symptoms. Mothers underestimated the efficacy of the tetanus vaccine—they believed the tetanus vaccine to have an average efficacy of 70% while actual efficacy is around 95%. Participants in the 3 groups were similar in terms of child age, child vaccination coverage, knowledge, perceptions of efficacy, and access to health facilities. For a subset of the sample (n = 459), we have a richer set of socioeconomic characteristics from the household survey conducted 9 months prior to the start of the trial. These data also suggest the treatment arms were well balanced (S1 Table).
The results of the pooled analysis are shown in Table 2. The proportion of children with DPT3 was 28% in the control group and 43% in the 2 groups receiving information, giving a difference of 14.6 percentage points (95% CI: 7.3 to 21.9, p < 0.001) and a relative risk of 1.5 (95% CI: 1.2 to 1.9, p < 0.001). In other words, children whose mothers received the information were 52% more likely to receive DPT3 than children in the control group. There was a positive effect on DPT3 vaccination regardless of the source of data, suggesting that self-reports were not driven by social desirability bias.
Results for the secondary outcomes follow a similar pattern (Table 2). The effect of the information intervention on the proportion of children fully immunised was an increase of 14 percentage points (95% CI: 7.7 to 21.1, p < 0.001), equivalent to a relative risk of 1.7 (95% CI: 1.3 to 2.3, p < 0.001). Despite the information never mentioning measles, the effect on measles coverage was 22 percentage points (95% CI: 14.3 to 29.6, p < 0.001). There was no measurable effect on BCG vaccination coverage. In terms of possible pathways, the information intervention had a large positive effect on knowledge of causes, symptoms, and prevention of tetanus. We did not find an effect on perceptions of efficacy (difference: 0.17, 95% CI: −0.17 to 0.51, p = 0.318).
Table 3 reports results from the treatment group analysis to determine whether the framing of information mattered. The effect of positive framing on the proportion of children with DPT3 was an increase of 12.4 percentage points compared to control (95% CI: 3.9 to 21.0, p = 0.005), compared with an increase of 16.7 percentage points (95% CI: 8.2 to 25.2, p < 0.001) when the information was negatively framed. However, there was no difference when the 2 information groups were compared with each other (95% CI: −5 to 13, p = 0.352). Looking across all the secondary outcomes, the framing of information had no effect, with the exception of knowledge of prevention. Results remained qualitatively the same when we adjusted estimates with the inclusion of covariates (S2 Table) and when we used a Bonferroni correction to deal with the problem of multiple hypothesis testing (S3 Table).
Fig 2 shows the effect of the information intervention on each knowledge indicator, showing that the negative framing had a larger impact than the positive framing on most measures of knowledge (differences were significant at the 5% level for knowledge of rapid heartbeat as a symptom of tetanus, and knowledge of child vaccination and vaccination during pregnancy as methods of prevention).
We conducted several additional analyses to better understand the main findings. First, Fig 3 shows the effect of the information intervention on perceptions of vaccine efficacy for 2 subgroups of mothers with perceptions of efficacy at baseline above and below 50%. The intervention had a significant positive effect on mothers who, at baseline, believed the DPT vaccine to have an efficacy level below 50% (difference: 0.88, 95% CI: 0.06 to 1.69, p = 0.04). The intervention did not have an effect on mothers who were already convinced of the efficacy of the DPT vaccine—i.e., those who had perceptions of efficacy above 50% at baseline (difference: 0.01, 95% CI: −0.36 to 0.37, p = 0.97). Second, using variation in the geographical density of study households assigned to the information intervention, we found no evidence of spillovers. Proximity to study households that received the information intervention did not affect the probability of children receiving DPT3 (S4 Table).
The total cost of the information intervention was $11,353, of which $2,923 (26%) was start-up costs and $8,430 (74%) was implementation costs. Table 4 shows that the cost of the information intervention was $165 per additional child with DPT3, $109 per additional child with measles vaccine, $186 per DALY averted, and $5,572 per under-5 death averted. The cost per DALY averted was considerably lower than standard thresholds used to define highly cost-effective interventions. The approach of the World Health Organization is to use a threshold based on the country’s gross domestic product per capita ($793 in the case of Uttar Pradesh), while the World Bank’s World Development Report 1993 recommended at the time $150 per DALY ($246 in current dollars). Under our scale-up scenario, the cost of the information intervention falls to $95 per DALY averted and $2,840 per under-5 death averted. While these estimates are only for the purposes of illustration, they encompass a wide range of valuations for which the intervention would be considered highly cost-effective.
This paper presented evidence on the role of information in raising demand for immunisation in India. Our analysis yielded 3 key findings. First, providing mothers of unvaccinated or incompletely vaccinated children with information on tetanus and the benefits of vaccination substantially increased immunisation coverage of DPT3, full immunisation, and measles. The large effect on measles vaccination was not anticipated, given that the information intervention focused solely on tetanus. We speculate that the increase in measles vaccination was generated by increased engagement with the public health system and, in turn, health workers ensuring children were up to date on all their vaccines, not just DPT3. Second, the framing of the information did not appear to generate large differences in outcomes. Although the effects of negative framing were consistently larger than when information was framed as a gain, differences between the 2 groups were small and rarely significant. Third, information improved mothers’ knowledge of causes of, symptoms of, and methods of prevention against tetanus. There was no effect on perceptions of vaccine efficacy, but there was suggestive evidence of an increase in perceptions of efficacy for mothers who initially had inaccurate perceptions.
The findings leave open the possibility that the intervention worked through various channels, most notably by increasing basic awareness on the existence of the DPT vaccine as a method of prevention. An alternative explanation is that the information served simply to remind mothers to get their child vaccinated, in which case cheaper interventions—such as SMS text message reminders—could be envisaged. We caution against such an interpretation. Not only do our results on knowledge suggest otherwise, but also all study participants were exposed to questions about vaccines in the baseline survey, making the immunisation status of the child salient. We also note that studies of the effect of text message reminders on vaccination uptake in developing countries have produced mixed results, showing either small or zero effects [24–28]. The effectiveness of reminders relative to the information intervention we tested should be the focus of future research.
A key strength of the study is that it is one of the first to evaluate the effectiveness of information provision on immunisation coverage in India, where a substantial proportion of the world’s unvaccinated children live. Other strengths of the study include the randomised design, the wide range of outcomes measured, and the novel nature of the information intervention. The information intervention targeted the mothers of unvaccinated/incompletely vaccinated children rather than the wider population of mothers. To inform the optimal design of the intervention, we tested 2 variants of how to frame the information.
The study had a number of limitations. We conducted an individual rather than a cluster randomised controlled trial, making contamination a potential concern. However, the information was given in private, and our analysis suggests that the outcomes of neighbours were not influenced by their proximity to treated households. If contamination was present, our results would provide an underestimate of the effect of the information intervention. We did not achieve the target sample size for the treatment group analysis such that the study was underpowered to detect small differences between the 2 intervention groups. Further research on framing in the context of immunisation is warranted to optimise the intervention. Qualitative research may have shed more conclusive light on how the intervention worked. We were unable to locate more than 400 potentially eligible women who could not be found at home. This may have limited the extent to which our study sample was representative of the wider population. Finally, the follow-up period was only 7 months. Tracking the study participants for longer would have provided evidence on whether the effects on outcomes were sustained.
Our study contributes to the growing body of evidence on interventions to increase uptake of vaccinations. A recent systematic review and meta-analysis found that demand-side interventions lead to an increase in the uptake of vaccinations, with a relative risk of 1.30 (95% CI: 1.17 to 1.44) [8]. The relative risk in this study was 1.52 (95% CI: 1.21 to 1.91), making this information intervention among the more effective ways to increase demand. The effect of information has been studied in the context of other health behaviours—such as maternity care [29], water purification [30–33], prevention of worms [34], malaria bed net use [35], and circumcision for HIV prevention [36]—with mixed results. We also contribute to evidence on the framing of health information. Our findings support a review that concluded, contrary to expectations, that there is no consistent effect of framing on health behaviours [15].
For policymakers, the findings suggest that targeted information may be a highly cost-effective means of increasing uptake of childhood immunisation. While caution must be taken in generalising the findings beyond the study setting, we highlight a number of context-specific factors that may be relevant when considering the wider relevance of the findings. First, the 6 study districts were not the worst performing in the state. In areas where immunisation rates are lower, such as remote rural areas in Uttar Pradesh and other Indian states, the intervention might produce larger effects as long as the vaccine supply is in place. Second, a local organisation implemented the intervention; however, any future scale-up would likely rely on government delivery channels, which may reduce costs but at the same time could present risks in terms of implementation fidelity. Third, we note that mothers’ knowledge and perceptions of efficacy at baseline were quite low; the intervention is likely to be less effective in areas where awareness and knowledge levels are higher. Finally, the supply of vaccines must be in place, at least intermittently, if information is to have any effect in increasing immunisation rates.
In light of the results on framing, a prudent strategy would be to adopt the negative framing of the information script since there are no cost implications. There may of course be cheaper ways of delivering the same information to parents. In the context of India, one obvious option could be to engage ASHAs given that they are integral to the delivery of community health services and have a strong focus on maternal and child health. A potential concern would be the loss of fidelity in the implementation of the intervention. At the same time, the messages are simple and quick to deliver.
Policymakers may also want to consider whether information should be combined with other interventions. Recognising the multiple barriers to behaviour change that exist, a small number of studies have combined information with price subsidies or monetary incentives. Ashraf et al. [37] studied the effects of door-to-door marketing of water purification products with different combinations of price subsidies and information about the benefits of the target product compared to a traditionally used product. They found that additional information increased the effectiveness of price subsidies by 60%. Banerjee et al. [38] considered demand- and supply-side interventions in combination, showing that incentives alongside immunisation camps were far more effective in raising immunisation rates than increasing supply alone.
There are a range of unanswered questions about how best to deliver information to improve immunisation coverage. Alternative, less costly ways of delivering information could make use of social networks. There is a growing literature on social networks being exploited to spread information and behaviours, e.g., by targeting highly connected individuals, nominated friends of individuals, or community leaders [39,40]. Understanding how information spreads is an active area of research that could yield useful insights for public health [41].
Our results demonstrate that targeted and clear information delivered to mothers of unvaccinated/incompletely vaccinated children can be effective in improving immunisation coverage. These findings contribute to a growing body of evidence on what are the most effective strategies to improve vaccination rates in developing countries. Although the barriers to immunisation uptake are multiple, ranging from social norms to the reliability of supply systems, in contexts where knowledge and awareness are a key binding constraint, interventions that provide information to parents and carers of unvaccinated children have the potential to be a simple and cost-effective way of increasing demand for immunisation.
|
10.1371/journal.pntd.0007306 | Relationship between toxoplasmosis and obsessive compulsive disorder: A systematic review and meta-analysis | A few studies investigated the relationship between toxoplasmosis and mental disorders, such as obsessive compulsive disorder (OCD). However, the specific nature of the association between Toxoplasma gondii (T. gondii) infection and OCD is not yet clear. The aim of this study was to collect information on the relationship between OCD and toxoplasmosis and assess whether patients with toxoplasmosis are prone to OCD.
For the purpose of this study, 6 major electronic databases and the Internet search engine Google Scholar were searched for the published articles up to July 30th, 2018 with no restriction of language. The inverse variance method and the random effect model were used to combine the data. The values of odds ratio (OR) were estimated at 95% confidence interval (CI).
A total of 9 case-control and 3 cross-sectional studies were included in our systematic review. However, 11 of these 12 articles were entered into the meta-analysis containing 9873 participants, out of whom 389 were with OCD (25.96% positive for toxoplasmosis) and 9484 were without OCD (17.12% positive for toxoplasmosis). The estimation of the random effect model indicated a significant common OR of 1.96 [95% CI: 1.32–2.90].
This systematic review and meta-analysis revealed that toxoplasmosis could be as an associated factor for OCD (OR = 1.96). However, further prospective investigations are highly recommended to illuminate the underlying pathophysiological mechanisms of T. gondii infection in OCD and to better investigate the relationship between OCD and T. gondii infection.
| Toxoplasma gondii (T. gondii) is an obligate neurotropic parasite that infected about 25–30% of the total human population in the developed and developing countries. The obsessive compulsive disorder (OCD) is a psychiatric disease that affects the income and quality of life. Some studies confirmed an association between infectious agents as the associated or protective factors specifying the development of psychiatry diseases. Among various pathogens associated with psychological disorders, most of the attention is on T. gondii, which has a life-long asymptomatic latent phase after a short acute stage in healthy individuals. The detrimental effect of T. gondii on immunocompromised people and pregnant women is an important concern for public health. The correlation between toxoplasmosis and OCD is still relatively understudied with a paucity of documented findings. The previous meta-analysis reviewed only two studies and reported a 3.4-fold greater chance of OCD. The results of our study presented stronger evidence of a positive relationship between toxoplasmosis and OCD. Eventually, our research team hopes to present an overview of what is known and encourage more intensive research to determine the real impact of this parasite on the occurrence of OCD that may contribute to the prevention of OCD worldwide.
| The T. gondii is a neurotropic apicomplexan protozoan that infects one-third of the world’s human population by affecting some tissues, including brain, eyes, and testes in warm-blooded mammals [1]. Infection with this parasite is due to the consumption of raw or undercooked meat containing tissue cysts or consumption of food or drinking water contaminated with oocysts shed by cats. Moreover, organ transplantation, blood transfusion, and vertical transmission during pregnancy from mother to fetus are other causes of T. gondii transmission [2]. The T. gondii infection is generally asymptomatic in immunocompetent individuals. However, immunocompromised patients may experience severe clinical complications, such as chorioretinitis, encephalitis, and pneumonitis. Toxoplasmosis also leads to psychotic symptoms and changes in the personality of individuals [3]. The T. gondii has a specific tropism for brain tissue, where tachyzoites can invade to microglia, astrocytes, and neurons and create cysts in these cells. The considerable production of neurotransmitters, such as dopamine by T. gondii, induces the increased production of bradyzoites and destruction of cyst walls that may be responsible for behavioral changes [4,5].
Recently published systematic review and meta-analysis studies have examined the relationship between T. gondii infection and various psychiatric disorders; such as bipolar disorder [3,6], schizophrenia [6,7], epilepsy [8], and depression [6,9]. The results of these studies showed that toxoplasmosis is an associated factor for bipolar disorder, schizophrenia, epilepsy, but not for depression.
The OCD is a common, chronic, and debilitating psychiatric condition that affects about 3% of the general population [10,11]. This disorder is identified by unwanted and recurrent thoughts, which cause marked distress. Individuals with OCD are struggling to reduce their anxiety by mental acts and repetitive behaviors [12]. According to the World Health Organization, OCD is one of the top ten disorders which affect people’s income and quality of life although it has the least effect [13].
Some of the available data indicate the possibility of an association between toxoplasmosis and OCD [14,15] although there are some contradictory results [16]. Therefore, the main purpose of this systematic review and meta-analysis was to evaluate the relationship between T. gondii and OCD.
This study was designed according to the Preferred Reporting Items for Systematic Reviews and Meta-Analysis (PRISMA) guidelines [17]. The protocol was registered in the PROSPERO with the registration number of CRD42018106354 [18].
To identify the published studies on the association between toxoplasmosis and OCD, the researchers performed a systematic search in 6 databases, namely PubMed, Scopus, ScienceDirect, Web of Science, EMBASE, ProQuest, and the Internet search engine Google Scholar. This systematic review was conducted through gathering the articles published up to July 30th, 2018 with no restriction of language. The search process was accomplished using the following keywords “Toxoplasma” OR “toxoplasmosis” AND “Obsessive-Compulsive Disorder” OR “OCD”.
The inclusion criteria included: (1) studies published until July 30th, 2018, (2) case-control and cross-sectional studies about the relationship between toxoplasmosis and OCD, (3) original research papers, (4) studies with available full texts, and (5) studies with information on the exact total sample size and positive samples in the case and control groups. The exclusion criteria were: (1) studies with no exact information about the sample size in the case and control groups, (2) review articles, and (3) non-human studies.
All the retrieved articles from the search strategy were imported to EndNote (version X7). After the removal of duplicated papers, the titles and abstracts were independently reviewed by two researchers. In the next step, eligible articles were selected for full-text download (Fig 1). Data from relevant studies were extracted into a Microsoft Excel datasheet. The extracted variables included the name of first author, year of publication, location of the study, diagnostic method, OR, number of seropositive cases and control, as well as the age and gender of the participants in the case and control groups. The researchers of the current study were very careful about extracting the correct information. In this regard, the authors of the three selected articles were contacted for more detailed information [19–21].
Two researchers independently assessed the quality of the included papers using standard strengthening the Reporting of Observational Studies in Epidemiology checklist (STROBE). This scale includes 22 items that are related to the title, abstract, introduction, methods, results, and discussion sections of the articles. This checklist included items assessing objectives, different components of the methodology (e.g., study design, study size, study population, bias, statistical methods), key results, limitations, generalizability, and funding of the studies. The assigned scores were within the range of 0–44. Based on the STROBE checklist assessment, articles were categorized into 3 groups (low quality: less than 15.5, moderate quality: 15.5–29.5, and high quality: 30.0–44.0). The S1 Checklist indicates the quality of the included studies [22].
The data entered into Microsoft Excel were exported to Stata version 14 (Stata Corp, College Station, TX, USA) for the analysis [23]. The common OR were estimated using inverse variance and random-effects model for each included study. Furthermore, the heterogeneity index was determined using Cochran’s Q and I squared statistics. I squared values less than 25%, 25–50%, and greater than 50% were defined as low, moderate, and high heterogeneity, respectively [23]. The publication bias was examined by the Egger test. A sensitivity analysis was performed using Stata version 14 (Stata Corp, College Station, TX, USA) to identify the possible effect of each study on the overall results by removing each study.
Out of 2500 identified articles, 392 articles were excluded due to the duplication, and 2056 articles were also eliminated on the basis of their titles and abstracts. After reading the full text of the articles, 12 papers were included in our systematic review [14–16,19–21,24–29]. Eventually, 11 of these 12 articles [14–16,19–21,24,25,27–29] were entered into this meta-analysis with respect to the inclusion/exclusion criteria (Fig 1). One of the papers was excluded due to the lack of detailed information about the number of patients with OCD [26]. Information and characteristics about the investigated publications are presented in Table 1 and Table 2.
Studies were published from 2006 to 2018. Accordingly, 9 out of the 12 studies had a case-control design, and 3 of them were cross-sectional studies (Table 1). One of the articles was not analyzed due to the unclear data about the exact number of patients with OCD [26]. The total number of participants involved in the 11 included studies in the meta-analysis was 9873, including 389 OCD patients and 9484 controls. Studies were conducted in Turkey [14,16,25], Czech Republic [15,19,21], China [27,29], USA [20], Mexico [28], Saudi Arabia [24], and Iran [26]. Anti-Toxoplasma antibodies (IgG and IgM) were determined using enzyme-linked immunosorbent assay [14–16,19,20,24–29], indirect immunofluorescence assay [14], complement fixation test [15,19], and enzyme immunoassays [26]. One of the studies did not address the method through which Toxoplasma is diagnosed [21].
Meta-analysis results showed that the OR of the chance of toxoplasmosis in OCD patients compared to control groups was 1.96 (95% CI: 1.32–2.90) (Fig 2). The test of heterogeneity showed a moderate heterogeneity among the studies included in the meta-analysis (chi2 = 15.37, P = 0.119, I2 = 34.9%).
Publication bias was assessed by Egger’s test and the results showed no publication bias (P = 0.540). Sensitivity analysis using the “one study removed at a time” technique demonstrated that the impact of each study on meta-analysis was not significant on the overall estimates (Fig 3).
Toxoplasmosis in the individuals leads to psychotic symptoms and changes in personality [3]. The T. gondii has a relationship with schizophrenia [30,31] and bipolar disorder [3] ]; however, its relationship with the OCD is understudied and there are few documented findings. The inconsistent results among the included studies in our meta-analysis demonstrate a discrepancy in the relationship between T. gondii and the chance of OCD. Therefore, we designed this systematic review and meta-analysis to assess the overall prevalence and ORs of this infection in the individuals with OCD compared to those in the control group.
A total of 12 articles on the prevalence of toxoplasmosis in OCD patients were included in the current paper. Although few studies were included in this meta-analysis, our findings indicated higher T. gondii seropositivity in the OCD patients compared to those in the control group with the OR of 1.96 (95% CI: 1.32–2.90). This agrees with the results of the ecological study by Flegr [32] showing a very strong correlation between incidence of toxoplasmosis and OCD-related burden in European (p = 0.02) and especially in non-European countries (p<0.0001). These results showed that there is a strong correlation between the prevalence of toxoplasmosis and OCD. The results of the current study 1.96 (95% CI: 1.32–2.90) differed from those of previous meta-analysis 3.4 (95% CI: 1.73–6.68) [6]. The previous meta-analysis was performed only on the basis of two studies in 2015 [6]. Since the current study investigates the updated evidence of the association between toxoplasmosis and OCD, it includes nine studies, which were not examined in the previous meta-analysis [15,16,19–21,24,25,28,29]. Moreover, a published study in 2006 was not included in the previous study [28] and this leads to discrepancies in the results of our study with the previous ones.
The included studies in our meta-analysis study were from three continents of Asia (Turkey: 3 studies, China: 2 studies, Iran: 1 study, Saudi Arabia: 1 study), Europe (Czech and Slovak Republics: 3 studies), and America (USA: 1 study, Mexico: 1 study). However, data gaps were identified for Africa, Australia, and many European countries where no data were available.
The status of the disease mainly depends on two quantities, the sensitivity and the specificity of the serological tests. However, all of the relevant studies have presented the prevalence of disease without mentioning tests sensitivity and specificity. Nevertheless, false positive and negative results can be significant because they do not show the prevalence of the infected people [33]. Variation in the sensitivity and specificity of enzyme-linked immunosorbent assay kits and the different cutoff values are effective factors on the prevalence of infection [34]. Different results of studies evaluated the relationship between various variables (including age, sex, education level, and history of blood transfusion) and the prevalence of toxoplasmosis reduced the ability to meta-analysis for these variables. In addition, the lack of evaluation of various associated factors in the eligible studies can be considered as basic gaps.
Identification of time evaluation is considered as an important variable for the temporal relationship between T. gondii exposure and disease onset. The evaluation of this variable helps to improve the precision of future studies describing the association between infectious agents and psychiatric disorders. However, none of the studies included in the current article considered this variable. Yolken et al. in 2017 [35] conducted a study for measured serological evidence of exposure to T. gondii in people. The results of the study indicated an increased odds of T. gondii exposure in people with a recent onset of psychosis (OR = 2.44).
Since the genetic characteristics of an individual can influence the forms of OCD family; therefore, there is a need to consider this issue in evaluating the relationship between OCD and T. gondii. However, only one study has addressed this important variable [26]. Rh phenotype is also an important variable that should be considered in various studies. Recent studies on women showed that Rh-positive women had lower levels of depression, obsession, and other psychiatric disorders. Although Rh-positive is an important variable, it has not been sufficiently addressed in previous studies [36]. The prevalence of T. gondii in patients with OCD was different among various studies. This difference in the prevalence of the included studies might be explained by the difference in the prevalence reported in the general population of each studied place. One of the reasons for the difference in the prevalence among the psychiatric and control populations might be due to the differences in the sanitary conditions among the groups. Indeed, most psychiatric inpatients belonged to a lower socio-economic level and had lower housing conditions than the control populations [28].
Some of the psychiatric disorders in humans are due to the ability of T. gondii to alter immune responses and neurotransmitters [37]. One of the important neurotransmitters is dopamine, which plays an essential role in the etiology of different neuropsychological diseases, such as major depression, schizophrenia, Parkinson’s disease, and Alzheimer’s disease [38]. Latent toxoplasmosis significantly affects dopaminergic and glutamatergic systems [39]. The higher chance of schizophrenia and OCD in the T. gondii infected individuals can be due to the increased dopaminergic activity [40]. Additionally, current studies have reported that brain cells infected with Toxoplasma contain high concentrations of dopamine [25]. The migration of Toxoplasma to the brain, formation of cysts, and changes in the production of neurotransmitters, such as dopamine can lead to the high rate of OCD prevalence in people with serum positive for T. gondii [25]. Treatment of two children with toxoplasmosis and OCD using anti-protozoan medications decreased Toxoplasma antibodies and completely cured OCD [41]. Furthermore, treatment in a 34-year-old woman with AIDS and neurotoxoplasmosis consuming antiprotozoal decreased OCD symptoms [42]. These findings supported a possible relationship between toxoplasmosis and OCD.
It has been suggested that changes in the hypothalamic-pituitary-adrenal gland axis, immune reactions [43], hormonal disorders caused by Toxoplasma infection [44], neuroimmune function and serotonin function disorder could lead to OCD [25]. Moreover, OCD could be due to a dysfunctionality of the front striatal loops, involved in frontal differentiation, as well as the lack of inhibition of automatic behavior [45,46]. Furthermore, some immune-mediated basal ganglia processes may be operating in OCD [41]. Denys et al. reported the observation of reduced TNF-alpha production and NK cell activity in patients with OCD [47]. Regardless these facts, it is possible that the OCD could be the cause rather than the effect of the Toxoplasma infection. It should be reminded, however, that OCD-induced behavioral changes such as fear of contamination, repeated washing of hands and social avoidance reduce rather than increase the chance of toxoplasmosis [15]. It is still possible that some unknown factor influences both the chance of toxoplasmosis and OCD. Therefore, further studies will be necessary to clarify the nature of the association between T. gondii and OCD.
One of the limitations of the included studies in the present research was that the individuals were invited to participate in some of these studies through snowball sampling technique using Facebook, fliers, and electronic media [15, 19, 21]. In this regard, the researcher(s) posted a Facebook announcement to invite people to take part in diverse psychological, ethological, and psychopathological experiments. However, the samples recruited in the mentioned studies cannot be representative of the general population since all people do not have access to Facebook. Moreover, the provided information were not based on the medical records; therefore, there were possibilities of wrong or at least obsolete data. To clarify, some patients may be infected with Toxoplasma after being tested for the presence of anti-Toxoplasma antibodies using serological methods. This could result in positively biased incidence rates of particular disorders. Accordingly, the obtained results cannot be generalized to the whole population. In one of these studies, the questionnaire contained many questions related to sexual behaviors and sexual preferences [21]. As a result, the participants were composed of those who were interested in these topics. Another limitation was that some studies were conducted only on children and adolescents, which made it difficult to generalize the findings to the society as a whole [25, 29].
There were also, some limitations in our research, including (1) few numbers of studies that investigated the relationship between T. gondii infection and OCD, (2) small sample size in the included studies, (3) reports with various quality, (4) available studies with no sufficient information on disease status/severity, (5) lack of the published articles in many parts of the world regarding the seroprevalence of toxoplasmosis among patients with OCD, (6) lack of the evaluation of various associated factors, such as familial history and Rh phenotype.
Based on the currently available data, T. gondii infection was more frequent in OCD patients than the control group. The results of this study were indicative of a probability of positive association between the prevalence rate of toxoplasmosis and OCD. However, many questions remained to be answered in future studies. Therefore, further research should be performed to evaluate the reduction rate regarding the prevalence of OCD following the treatment of toxoplasmosis and the recognition of the physiopathological mechanisms involved in T. gondii infection in OCD. Also, it is highly desirable to obtain empirical data from other parts of the world.
|
10.1371/journal.ppat.1006428 | Vibrio cholerae ensures function of host proteins required for virulence through consumption of luminal methionine sulfoxide | Vibrio cholerae is a diarrheal pathogen that induces accumulation of lipid droplets in enterocytes, leading to lethal infection of the model host Drosophila melanogaster. Through untargeted lipidomics, we provide evidence that this process is the product of a host phospholipid degradation cascade that induces lipid droplet coalescence in enterocytes. This infection-induced cascade is inhibited by mutation of the V. cholerae glycine cleavage system due to intestinal accumulation of methionine sulfoxide (MetO), and both dietary supplementation with MetO and enterocyte knock-down of host methionine sulfoxide reductase A (MsrA) yield increased resistance to infection. MsrA converts both free and protein-associated MetO to methionine. These findings support a model in which dietary MetO competitively inhibits repair of host proteins by MsrA. Bacterial virulence strategies depend on functional host proteins. We propose a novel virulence paradigm in which an intestinal pathogen ensures the repair of host proteins essential for pathogenesis through consumption of dietary MetO.
| The virulence program of intestinal pathogens such as Vibrio cholerae depends on the continued function of target host proteins. If these proteins are inactivated by methionine oxidation, virulence may also depend on repair of these proteins by host methionine sulfoxide reductases such as MsrA. Dietary methionine sulfoxide competitively inhibits repair of host proteins by MsrA. Here, using the model host Drosophila melanogaster, we demonstrate a novel virulence paradigm in which V. cholerae uptake of dietary methionine sulfoxide frees host MsrA to repair host proteins essential for virulence.
| Childhood diarrheal disease is a leading cause of morbidity and mortality, particularly in the developing world, and bacterial pathogens figure prominently in this entity [1]. Traditional virulence factors are pathogen-specific and synthesized expressly for the purpose of colonizing, entering, and manipulating host intestinal epithelial cells. However, well-conserved metabolic pathways of diarrheal pathogens may also contribute to disease by altering the metabolite profile of the intestinal contents, the composition and physiology of the resident intestinal microbiota, and local and systemic host metabolism. This, in turn, may have immediate and long-lasting impacts on host intestinal function and nutrition.
Vibrio cholerae, an important pathogen in many regions of the developing world, causes a life-threatening diarrheal disease when ingested in contaminated water or food [2]. Two intensively studied virulence factors of V. cholerae are the toxin co-regulated pilus and cholera toxin, which is responsible for the severe secretory diarrhea of cholera [2]. To identify and study additional virulence factors in a genetically tractable host, we developed the arthropod Drosophila melanogaster as a model for V. cholerae infection [3]. In our experimental design, Drosophila are fed V. cholerae in LB-broth, and a lethal infection ensues. This infection is independent of the toxin co-regulated pilus and is only partially mitigated by deletion of the genes encoding cholera toxin. V. cholerae colonizes the fly midgut and rectum, disrupts adherens junctions, suppresses intestinal stem cell division, causes accumulation of large lipid droplets in enterocytes, and suppresses insulin signaling [3–7]. While cholera toxin plays only a small role in pathogenesis in this model, the accumulation of large lipid droplets in Drosophila enterocytes is correlated with mortality [3, 6, 7].
We recently reported a Drosophila-based screen for V. cholerae transposon insertion mutants with decreased virulence [7]. This screen identified the virulence factor CrbRS, a two-component system that regulates acetate uptake in a process known as the acetate switch [8]. In response to an unknown signal, CrbRS activates transcription of acs1, a gene encoding the enzyme acetyl-CoA synthase, which uses acetate as a substrate. We reported that elimination of intestinal acetate uptake by V. cholerae prevented lipid droplet formation in enterocytes and interruption of host insulin signaling. In mammals, short chain fatty acids (SCFA) such as acetate, propionate and butyrate, are principally produced by the intestinal microbiota and serve as both signals and nutrients that maintain the intestinal epithelium [9, 10]. Intestinal uptake of SCFA by pathogens suggests a difference in metabolism between pathogenic and commensal bacteria in the intestinal environment and provides one mechanism by which pathogen metabolism may contribute to virulence. Here we report an additional mechanism by which pathogen metabolism may manipulate intestinal function.
We previously identified several V. cholerae glycine cleavage system mutants that were attenuated for virulence [7]. While exploring the virulence attenuation of V. cholerae glycine cleavage system mutants, we uncovered an intestinal phospholipid degradation cascade that is responsible for lipid droplet coalescence in enterocytes during infection. Here we report that the host proteins required for this cascade are inactivated by dietary methionine sulfoxide (MetO), disruption of the V. cholerae glycine cleavage system, and inhibition of host methionine sulfoxide reductase A (MsrA), an enzyme that reduces both free and protein-associated MetO to methionine [11, 12]. In the absence of repair by MsrA, oxidation of exposed methionines can lead to protein inactivation. We propose that free MetO competitively inhibits MsrA-dependent repair of a protein required for V. cholerae-activated phospholipid degradation. Consumption of dietary methionine sulfoxide by wild-type V. cholerae relieves this inhibition. This represents a novel virulence paradigm in which an intestinal pathogen, through its metabolism, promotes the repair of host proteins essential for virulence.
A genetic screen for V. cholerae virulence determinants in a Drosophila model of infection identified several transposon insertions in genes encoding components of the bacterial glycine cleavage system [7]. The glycine cleavage system, which is highly conserved among bacteria, plants, and animals, is involved in the catabolism of serine and glycine. It consists of the proteins GlyA and GcvH, P, T, and L (Fig 1A) [13]. GlyA is a serine hydroxymethyltransferase that generates glycine from serine along with donation of a methyl group to tetrahydrofolate (THF). GcvH, which is modified with a disulfide bond-containing lipoic acid, activates the glycine decarboxylase GcvP. After release of CO2, the glycine-derived reaction intermediate is transferred to a GcvH-associated lipoic acid sulfhydryl group generated by disulfide bond reduction and shuttled to GcvT, an aminomethyltransferase. From here, another methyl group is transferred to THF, and ammonia is released. To re-initiate the cycle, the GcvH-associated lipoic acid disulfide bond must be regenerated by GcvL, a dihydrolipoamide dehydrogenase (DHLD). The methyl groups donated to THF enter the folate cycle, giving rise to key biological compounds such as purines and methionine.
In V. cholerae, glyA2, gcvH, gcvP, and gcvT are found in one chromosomal locus, while VC2638, a putative DHLD and a second glyA homolog are found in other regions of the chromosome (Fig 1B).
To explore the role of the glycine cleavage system in virulence, we constructed strains carrying in-frame deletions in glyA1, glyA2, gcvH, gcvP, gcvT, and VC2638. As shown in Fig 1C, mutations in gcvT, P, and H, but not VC2638 or glyA1/2 greatly decreased virulence in the fly model. The virulence defect of a ΔgcvT mutant could be rescued by an expression plasmid encoding the gcvT gene (S1 Fig). This strongly suggests that interference with V. cholerae glycine but not serine catabolism alters virulence. Furthermore, because the ΔVC2638 mutant had no virulence defect, we hypothesize that it is not essential for glycine catabolism, possibly due to the presence of another dihydrolipoamide dehydrogenase with redundant function such as that at locus VC2412.
Because our results suggested that the glycine cleavage system was important for virulence, we first tested the possibility that glycine catabolism was required for survival and growth of V. cholerae within the fly intestine. However, we found that the bacterial burden of strains defective for glycine catabolism was similar to that of the wild-type parental strain (Fig 1D).
We previously showed that V. cholerae infection greatly suppresses intestinal stem cell (ISC) division in the Drosophila intestine and that activation of this process prolongs host survival [6]. Therefore, we questioned whether mutation of the glycine cleavage system had an additional effect on ISC division. Histone 3 is phosphorylated during cell division. To assess ISC divisions, we used immunofluorescence to enumerate cells in which histone 3 is phosphorylated in the intestines of uninfected Drosophila as well as those infected with wild-type V. cholerae or a glycine cleavage system mutant. As shown in Fig 1E, V. cholerae glycine cleavage system mutants with reduced virulence did not suppress ISC division. V. cholerae ΔgcvT, P, and H mutants had similar infection phenotypes, suggesting a common mechanism of virulence attenuation. Therefore, for simplicity, further investigation of mechanism focused on only one of these, the ΔgcvT mutant.
Infection with wild-type V. cholerae results in accumulation of lipid droplets in the fly intestine in tandem with depletion of lipid droplets from the Drosophila adipose tissue or fat body, and interventions that decrease accumulation of lipid droplets in the intestine prolong host survival [7]. To determine the effect of Drosophila infection with a V. cholerae ΔgcvT mutant on lipid droplet distribution, we stained the fat bodies and intestines of flies with the lipophilic dye Nile Red. Similar to what was observed previously, we found that lipid droplets accumulated in the intestines of flies infected with wild-type V. cholerae and were depleted from the fat body. However, in the absence of V. cholerae gcvT, none of these derangements were observed (Fig 1F and 1G). We previously observed that depletion of lipid droplets from the Drosophila fat body results in diminished signaling through the insulin/insulin-like signaling (IIS) pathway. Activation of the llS pathway results in protein kinase B phosphorylation (p-AKT), which can be detected using a p-AKT-specific antibody. We used Western blot analysis to assess signaling through the llS pathway. As shown in Fig 1H, infection with a V. cholerae ΔgcvT mutant decreased p-AKT to a lesser extent than that with the wild-type strain.
In a process known as the acetate switch, V. cholerae generates acetate through fermentation of available sugars and then consumes acetate when other carbon sources are scarce [7, 8]. We previously showed that V. cholerae ΔcrbS mutants, which cannot consume acetate, also have a significant defect in virulence in a Drosophila model [7]. To determine if the virulence defect of the glycine cleavage system mutants was due to an inability to consume acetate, we compared acetate concentrations in the spent supernatants of wild-type V. cholerae, ΔgcvT mutant, and ΔcrbS mutant cultures. As shown in Fig 1I, acetate concentrations in the spent supernatants of ΔgcvT mutant cultures were not significantly different from those in wild-type V. cholerae supernatants. V. cholerae acetate uptake mutants increase food accumulation in the Drosophila intestine suggesting an effect on appetite [7]. Ingestion of the ΔgcvT mutant did not have this effect (Fig 1J). These data suggested to us that V. cholerae glycine cleavage system and acetate uptake pathways are independent mediators of virulence in the fly model.
We reasoned that if mutations in crbS and gcvT rescued fly survival through the inability to consume distinct nutrients, co-infection with ΔcrbS and ΔgcvT mutants would result in net consumption of all differentially secreted metabolites, leading to a restoration of virulence. In fact, we found that, while each mutant alone had a significant virulence defect, co-infected flies died at the same rate as flies infected with wild-type V. cholerae (Fig 1K). This is consistent with the hypothesis that ΔcrbS and ΔgcvT mutants rescue virulence through secretion or an inability to utilize distinct small molecules that are normally consumed by V. cholerae.
Glycine cleavage system mutants have previously been shown to secrete glycine into the culture medium [14]. As an indication that glycine might be secreted into the fly intestine by the V. cholerae ΔgcvT mutant, we measured transcription of the Drosophila gcvH (ppl), gcvL (CG7430), gcvT (CG6415), and gcvP (CG3999) genes in the intestines of LB-fed Drosophila and those infected with wild-type V. cholerae or a ΔgcvT mutant (S2A–S2D Fig). We found that infection with wild-type V. cholerae decreased intestinal transcription of Drosophila gcvH and gcvL, while infection with the V. cholerae ΔgcvT mutant increased transcription of these genes. We hypothesize that this represents the transcriptional response of the Drosophila intestine to consumption and secretion of glycine by wild-type V. cholerae and the ΔgcvT mutant, respectively. We reasoned that if secretion of glycine by the V. cholerae ΔgcvT mutant was responsible for prolonged host survival, diet supplementation with glycine during wild-type V. cholerae infection should also have this effect. However, we found that glycine ingestion did not alter host survival of infection (S2E Fig). We also administered glycine to flies in phosphate buffered saline (PBS) and measured ISC division. This treatment decreased ISC division (S2F Fig). These results demonstrate that the host intestine senses and responds to glycine secretion by a V. cholerae ΔgcvT mutant, but this response neither increases host resistance to infection nor restores ISC division.
We hypothesized that the metabolic interaction of a glycine cleavage system mutant with its environment might extend beyond secretion of glycine. To test this, we analyzed polar metabolites in sterile LB broth as well as the spent culture supernatants of wild-type V. cholerae and a ΔgcvT mutant by targeted LC-MS/MS (Fig 2A and 2B, and S1 Table) [15]. Alanine and methionine sulfoxide (MetO) were several fold lower in the supernatants of wild-type V. cholerae as compared with sterile LB broth (Fig 2B). However, levels of these two metabolites in the V. cholerae ΔgcvT mutant supernatant were more like those of sterile LB, suggesting very little consumption of these amino acids by the ΔgcvT mutant. As expected, levels of glycine were similar in LB broth and the spent supernatant of wild-type V. cholerae but higher in that of the ΔgcvT mutant (Fig 2B).
To assess the relevance of our findings in culture to the infected Drosophila intestine, we performed metabolomics on the intestines of flies fed LB broth alone or inoculated with wild-type V. cholerae or a ΔgcvT mutant (Fig 2C and S2 Table), using a sample preparation protocol that eliminated protein-associated amino acids and metabolites. In general, the differences observed were smaller than those observed in culture (Fig 2A–2C, S3 Fig and S1 and S2 Tables). In an infection model, we reason that nutrients left behind by the pathogen are rapidly consumed by the host. However, similar to our culture results, free alanine and MetO were significantly elevated in the intestines of ΔgcvT-infected flies as compared with those of flies infected with wild-type V. cholerae. Interestingly, in contrast to LB broth, to which both the pathogen and host had access during infection, the level of MetO in the intestines of flies fed LB broth alone or inoculated with V. cholerae were much higher than that of methionine (Fig 2E). This may reflect a highly oxidizing intestinal environment.
We extended our studies to the mammalian intestine. In the recently resuscitated infant rabbit model of cholera [16], cecal fluid accumulates during infection and is easily harvested. We infected infant rabbits with either wild-type V. cholerae or a ΔgcvT mutant, assessed colonization, and then harvested cecal fluid for metabolomic analysis. Similar to our findings in the fly, no significant difference in colonization or fluid accumulation was observed in the terminal ileum or cecum of rabbits infected with these two strains (S4A–S4C Fig). We collected cecal fluid from infected rabbits, removed cells and other particulates by centrifugation, and then performed metabolomic analysis on the resulting fluid (Fig 2D and S3 Table). While there were very low concentrations of free methionine and MetO in the intestines of infant rabbits as compared with the concentrations of these metabolites in LB broth and the fly intestine (Fig 2E), methionine, methylcysteine, and adenine were more abundant in the cecal fluid of rabbits infected with a V. cholerae ΔgcvT mutant. These metabolites are linked to methionine sulfoxide via methionine sulfoxide reductase and the methionine cycle (Fig 2F), and their presence in the cecal fluid suggests that there is more capacity for MetO reduction and metabolism in the infant rabbit intestine as compared with the adult Drosophila intestine. We conclude that mutation of V. cholerae gcvT alters pathogen methionine metabolism in LB broth, the arthropod intestine, and the mammalian intestine and, therefore, that the V. cholerae glycine cleavage system has the potential to modulate host intestinal physiology.
Alanine and MetO were more abundant in the intestines of flies infected with the ΔgcvT mutant as compared with wild-type V. cholerae. We then explored whether diet supplementation with these amino acids might increase host resistance to infection (Figs 3A and S5A). While MetO significantly prolonged fly survival, alanine and a number of other metabolites identified had no effect. Importantly, MetO did not significantly affect V. cholerae growth either in LB or in the fly and did not impact fly mortality in the absence of infection (Figs 3B, S5B and S5C). These data suggest that a decrease in MetO uptake as a result of mutation of V. cholerae gcvT leads to an interaction that is more favorable for the host.
Free methionine as well as protein-associated methionine is easily oxidized to MetO [17]. This occurs in vivo through exposure to reactive oxygen species. The reverse process, conversion of MetO to methionine, is carried out by methionine sulfoxide reductases that are widely conserved and present in both flies and bacteria [18, 19]. To determine whether the impact of MetO supplementation on infection was direct or attributable to increased methionine availability, we performed an infection in LB supplemented with methionine. We found that methionine supplementation had no effect on bacterial growth or survival of V. cholerae-infected flies (Figs 3A, S5C and S5D). These data suggest that MetO but not methionine promotes host survival of V. cholerae infection.
Wild-type V. cholerae but not ΔgcvT mutant infection suppresses ISC division in the Drosophila gut. We hypothesized that this might be the result of MetO fueling the methionine cycle through conversion to methionine. We tested this by measuring dividing cells in the intestines of flies fed PBS supplemented with either MetO or methionine. Methionine but not MetO increased ISC division (Fig 3C). Therefore, while dietary methionine stimulates intestinal stem cell division, intestinal MetO is not likely to be the cause of increased ISC division in flies infected with a V. cholerae ΔgcvT mutant.
Infection with wild-type V. cholerae but not a V. cholerae ΔgcvT mutant leads to accumulation of lipid droplets in the intestine, depletes lipid droplets in the fat body, and suppresses signaling through the insulin pathway (IIS). We reasoned that if MetO supplementation prolonged fly survival by the same mechanism as mutation of V. cholerae gcvT, it would also reverse the metabolic phenotype of flies infected with wild-type V. cholerae. To assess lipid redistribution, we used Nile Red to stain lipid droplets in the fat bodies and intestines of flies infected with wild-type V. cholerae alone or supplemented with methionine (Met) or MetO. MetO prevented loss of lipid droplets from the fat body and lipid accumulation in the intestine (Fig 3D and 3E). To assess the impact of MetO on signaling through the IIS pathway, we used Western blot analysis to estimate the abundance of p-AKT in wild-type V. cholerae-infected flies supplemented with this amino acid. As shown in Fig 3F, supplementation with MetO preserved signaling through the IIS pathway. In contrast, both in the presence and absence of infection, supplementation with Methionine increased lipid droplets in the gut without depleting lipids in the fat body or decreasing insulin signaling (Figs 3D–3F and S5E–S5G). This suggests that depletion of lipids from the fat body deactivates insulin signaling during V. cholerae infection.
Lipid droplets are also present in Human Embryonic Kidney 293 cells (HEK293 cells). To determine whether the effect of MetO was specific to fly enterocytes, we incubated these human cells with MetO and the hydrophobic fluorescent dye boron-dipyrromethene (BODIPY) and enumerated lipid droplets. As shown in S5H and S5I Fig, fewer lipid droplets were also observed in human cells incubated with MetO.
Taken together, these data suggest that, similar to infection with a V. cholerae ΔgcvT mutant, MetO but not methionine supplementation promotes host survival by limiting intestinal lipid droplet accumulation.
Lipid droplets consist of an inner triglyceride core and an outer monolayer of polar phospholipids that forms the interface between the hydrophilic cell cytoplasm and the hydrophobic triglyceride core [20, 21]. The radius of curvature accommodated by these phospholipids depends on the size of the polar headgroup relative to the hydrophobic fatty acid chains. Phospholipids with larger head groups and shorter fatty acid chains or just one fatty acid chain such as a lysophospholipid form smaller lipid droplets (Fig 4A). We reasoned that large lipid droplets in the intestines of Drosophila infected with V. cholerae could result either from an increase in triglycerides within enterocytes leading to lipid droplet enlargement or from a decreased supply of short chain phospholipids leading to lipid droplet coalescence (Fig 4B). To distinguish between these two possibilities, we performed untargeted LC-MS/MS lipidomic analysis on the intestines of flies fed LB broth alone or inoculated with wild-type V. cholerae to quantify a vast array of lipid classes and fatty acids (S6 Fig and S4 Table). While triglyceride levels did not change with infection (Fig 4C), we noted a large decrease in short chain phosphatidylcholine (PC) and phosphatidylethanolamine (PE) species with total fatty acid carbons totaling less than or equal to 30 (Fig 4D–4G). This suggests that coalescence of lipid droplets in the gut is precipitated by catabolism of short chain phospholipids.
To explore the increased survival of Drosophila infected with a V. cholerae ΔgcvT mutant, we then applied untargeted lipidomics to the intestines of these flies (S7 Fig and S5 Table) [22]. Interestingly, we again observed no difference in triglyceride levels (Fig 4C). However, a significantly larger amount of short chain PE and PC was detected in flies infected with the ΔgcvT mutant as compared with those infected with wild-type V. cholerae (Fig 4D–4G). Interestingly, we also observed an accumulation of lysophosphatidylcholine (LPC) and lysophosphatidylethanolamine (LPE) (Fig 4H–4K), which are phospholipids from which one fatty acid chain has been enzymatically removed (Fig 4A). We hypothesize that V. cholerae infection induces an intestinal phospholipid degradation cascade and that this cascade is partially blocked by infection with a V. cholerae ΔgcvT mutant resulting in accumulation of lysophospholipid intermediates (Fig 4L). To demonstrate that this intestinal phospholipid profile was unique to V. cholerae infection attenuated by mutation of gcvT, we additionally explored the intestinal phospholipid profile of flies infected with the attenuated V. cholerae Δacs1 mutant (S8 Fig and S4 Table). In this case, we observed that, similar to a V. cholerae ΔgcvT infection, the level of short chain PC’s was increased as compared with wild-type V. cholerae (Fig 4M). However, in contrast to the ΔgcvT infection, there was no difference in PE, LPC, or LPE (Fig 4N–4P). This demonstrates that resistance to infection is correlated with inhibition of short chain phospholipid degradation in the intestine and also that V. cholerae gcvT and acs1 mutations alter the host intestinal phospholipid profile in unique ways.
Our data suggested that MetO rather than methionine was critical for resistance to infection. Both free and protein-associated MetO is abundant in oxidizing environments such as the fly intestine. For this reason, animals and bacteria alike possess intracellular methionine sulfoxide reductases that reduce MetO to methionine [23]. The fly genome encodes two stereospecific methionine sulfoxide reductases, MsrA and B. MsrA (CG7266), originally termed Eip71CD in the fly due to its regulation by ecdysone [24], reduces methionine-S-sulfoxide, while SelR or MsrB (CG6584) reduces methionine-R-sulfoxide [25]. While both proteins can reduce free MetO, studies in yeast and mammals show that MsrA reduces free MetO much more efficiently than MsrB [26]. We hypothesized that higher levels of MetO in the intestines of ΔgcvT mutant-infected flies might alter Drosophila msrA and/or msrB transcription. In fact, we found that Drosophila msrA transcription was increased 5-fold in a ΔgcvT mutant infection (Fig 5A). In contrast, transcription of Drosophila msrB was unchanged (Fig 5B). This suggested to us that the interaction of Drosophila with the V. cholerae ΔgcvT mutant might hinge on host MsrA.
We first characterized two Drosophila transposon insertion mutants. The Mi{MIC}Eip71CDMI14018 transposon is inserted in a C-terminal region of MsrA that is predicted to be non-coding [27]. The P{EPgy2}Eip71CDEY05753 transposon is inserted near the start of the second MsrA intron [28]. We tested the susceptibility of the msrAMI14018 and msrAEY05753 mutant fly lines to infection with wild-type V. cholerae after confirming decreased transcription of msrA in these lines (Fig 5C). As shown in Fig 5D, both mutants were highly resistant to infection as compared with controls in spite of similar bacterial burdens (Fig 5E). Infected msrA mutant flies showed normal lipid droplet accumulation in the fat body and enterocytes and active insulin signaling as compared with control flies (Fig 5F–5H). ISC division was equally suppressed in control and msrA mutant flies (Fig 5I). As an additional test, we obtained a fly line carrying RNAi targeting msrA and confirmed knockdown of msrA by a ubiquitous driver (Fig 5J). We then tested the effect of msrA knockdown on fly survival when the RNAi was driven ubiquitously or specifically to enterocytes (Fig 5K and 5L). In each case, the flies were resistant to infection as compared with driver-only controls. Taken together, these results show that MetO supplementation or msrA inactivation in enterocytes limits intestinal lipid droplet size leading to survival of infection.
We hypothesized that lipidomic analysis of the intestines of msrA mutant flies fed LB broth alone or inoculated with wild-type V. cholerae might shed light on the mechanism by which mutation of host MsrA prevents lipid droplet coalescence in the intestine. In an MsrA mutant fly, levels of short chain PC, PE, LPC and LPE did not decrease in response to V. cholerae infection (Fig 5M–5P and S6 Table).
We hypothesized that, during V. cholerae infection, proteins essential for the phospholipid degradation cascade defined here depend on repair by host MsrA. However, infection with a ΔgcvT mutant results in competitive inhibition of MsrA and increased protein oxidation. To assess the extent of protein oxidation in the Drosophila intestine, we harvested intestines and performed Western blot analysis using an antibody reported to recognize MetO [29]. LB broth has been shown to activate dual oxidase in Drosophila enterocytes, resulting in a highly oxidizing environment [30]. Therefore, to test whether this antibody could detect protein-associated MetO in the Drosophila intestine, we first performed Western blot analysis on the intestines of flies fed fly food or LB broth. As shown in Fig 6A, many more bands were observed in the samples prepared from flies fed LB broth, consistent with a higher amount of protein-associated MetO. Furthermore, densitometry quantification both of specific bands and total staining was increased for samples prepared from flies fed LB broth as compared with those fed fly food (Fig 6C–6E). We then performed the same experiments with flies fed wild-type V. cholerae or a ΔgcvT mutant. Although the difference was subtler in this case, samples prepared from the intestines of flies infected with a ΔgcvT mutant produced darker bands on Western blot analysis (Fig 6B). Densitometry analysis again supported our subjective observations (Fig 6C–6E). These results are consistent with our hypothesis that infection with a V. cholerae ΔgcvT mutant leads to decreased repair of protein-associated MetO in the fly intestine.
Here we present evidence that V. cholerae ingestion by the model host Drosophila melanogaster activates a phospholipid degradation cascade in enterocytes that results in lipid droplet coalescence, depletion of lipids from adipose tissue, and host death. We show that the activity of this cascade is ensured by V. cholerae consumption of dietary MetO and is inhibited by knockdown of methionine sulfoxide reductase MsrA within host enterocytes. MsrA is an intracellular protein that reduces dietary and protein-associated MetO to methionine [31]. Protein function can be activated or inactivated by methionine oxidation, and a large body of evidence suggests that reversible oxidation of protein-associated methionine is a mechanism by which cells adjust their physiology in response to reactive oxygen species [32]. Pathogens that co-opt host proteins either through delivery of toxins or type lll secretion system effectors to host cells depend on the continued function of their protein targets for pathogenesis [33, 34]. Based on the findings reported here, we propose a novel mechanism by which a pathogen ensures the continued functioning of host proteins required for virulence during intestinal infection (Fig 7). A large proportion of dietary methionine is consumed in the form of MetO and must be reduced to methionine by MsrA prior to utilization [35]. Thus, dietary MetO competes with protein-associated MetO for reduction by MsrA. During infection, V. cholerae consumes MetO in the host intestinal lumen. Because very little MetO reaches enterocytes, MsrA is free to reduce protein-associated MetO (Fig 7A). This ensures the continued function of host proteins required for phospholipid degradation and promotes host death. When MsrA expression in enterocytes is decreased by RNAi or the host is infected with a bacterium unable to consume dietary MetO, such as a V. cholerae ΔgcvT mutant, less MsrA is available for repair of host proteins, the phospholipid cascade is blocked, and the host survives (Fig 7B). Therefore, consumption of dietary MetO by V. cholerae promotes the function of host proteins essential for its virulence.
Our experiments demonstrate that the V. cholerae glycine cleavage system alters the metabolic profile of the intestinal lumen of flies and rabbits during infection. While methionine and its metabolites were increased in the cecal fluid of rabbits infected with the V. cholerae ΔgcvT mutant, only small amounts of methionine and MetO were present in the cecal fluid of infant rabbits regardless of the infecting V. cholerae strain. While there are no studies of the composition of rabbit or murine breast milk, methionine is one of the least abundant amino acids in human breast milk [36], and the finding that bacterial methionine synthesis is essential for V. cholerae colonization of the neonatal mouse intestine suggests that this is also the case in the mouse [37]. While only neonatal mammalian models of cholera are available, the defined diet of neonatal animals underscores a limitation in their use to explore the role of dietary manipulation in protection of young children and adults against cholera.
Our findings may be relevant to human disease. The mechanism we describe for lipid droplet coalescence is also believed to underlie some types of non-alcoholic fatty liver disease (NAFLD) [38]. Furthermore, NAFLD is associated with small intestinal bacterial overgrowth and inflammatory bowel disease, both of which could affect phospholipid genesis, catabolism, and supply in enterocytes and hepatocytes [39–42]. While intestinal or hepatic lipid accumulation has not been explored in cholera, host MsrA and its interaction with luminal MetO may play a role in the osmotic diarrhea of this disease. Calmodulin is a well-established facilitator of secretory diarrhea, and calmodulin antagonists have been developed as anti-diarrheal treatments [43, 44]. This is true, in particular, for cholera, whose secretory diarrhea depends on the CFTR chloride and SK potassium channels, both of which are calmodulin-dependent [45, 46]. Calmodulin is a Ca2+-binding regulatory molecule, which becomes unresponsive to Ca2+ activation upon oxidation of methionines 144 and 145 [47, 48]. Function is restored when these methionines are reduced by MsrA [49, 50]. Our findings suggest a rationale for the investigation of dietary MetO as an inhibitor of cholera toxin-induced diarrhea.
Diarrheal disease is responsible for 1.7 billion childhood infections per year worldwide, which may cause death or lead to life-altering sequelae such as undernutrition, growth faltering, cognitive impairment, poor response to childhood vaccines, and increased risk of death from other causes [1]. Here we describe a novel virulence mechanism by which intestinal microbes may disrupt enterocyte lipid metabolism and diarrheal pathogens may prolong secretory diarrhea. This mechanism points to MsrA as a new target for metabolic and anti-diarrheal treatments and also suggests inexpensive dietary interventions such as MetO supplementation to mitigate disease.
Drosophila were maintained on Bloomington formulation medium at 25°C. Fly strains used are listed in S7 Table. Where not otherwise noted, L-amino acids and metabolites (Sigma) were used at a concentration of 50 mM. Vibrio cholerae strains were cultured in Luria-Bertani (LB) broth or on LB agar supplemented with streptomycin (100 μg/ml) at 27°C. E. coli strains were grown in LB broth supplemented with ampicillin (100 μg/ml) when necessary at 37°C. Bacterial strains used are listed in S8 Table. Human Embryonic Kidney 293 cells (HEK293 cells, Thermo Fisher Scientific) were cultured in Dulbecco's Modified Eagle Medium (DMEM, Corning) supplemented with 10% fetal bovine serum (Cyclone). For experiments, cells were seeded into 96-well plates at a density of 50,000 cells/well. Where indicated, cells were supplemented with L-methionine sulfoxide (Sigma) (100 mM).
Mutagenesis was performed by double homologous recombination as previously described [51]. Plasmids and strains used are listed in S8 Table.
Infections were performed at 25°C, as previously described [3]. Groups of thirty 5–10 day old female flies were infected with the indicated strains of V. cholerae. For each condition, flies were divided into three groups of ten and placed in vials containing a cellulose plug infiltrated with 3 mls of LB broth inoculated with a 10-fold dilution of an overnight culture of V. cholerae and chemicals as noted. For rescue experiments using pBAD expression vectors, LB broth was also supplemented with 100 μg/ml ampicillin and 0.2% L-arabinose. Mortality was enumerated at least once each day. A non-parametric Kaplan–Meier test was used to estimate log-rank values.
1 to 2 day old New Zealand white rabbit kits (Charles river Research Models & Services) were given two doses of oral cimetidine (50mg/kg) 24 and 3 hours prior to administration of the noted V. cholerae strain. Bacteria were prepared from an overnight culture grown at 30°C. 200μl of a 1010 cfu/ml suspension of V. cholerae in sodium bicarbonate buffer (2.5 g in 100ml; pH 9) was administered to rabbit kits by gavage using 3.5 Fr red rubber catheter. The kits were inspected individually for signs of trauma or aspiration immediately following gavage and then observed periodically for 18 hours post inoculation to monitor dehydration, diarrhea and progressing symptoms. At the first sign of dehydration, kits were sacrificed. Cecal fluid was processed as described below, and intestinal tissues were collected, homogenized and plated to quantify colonization.
The acetate concentration in spent supernatants was measured using an Acetic Acid Assay Kit (Megazyme International Ireland) according to the manufacturer’s instructions. Bacteria were cultured overnight in LB broth, diluted into fresh LB broth to yield a starting OD600 of 0.04, and then incubated at 37°C with shaking overnight. 100 μl of this culture was centrifuged to remove bacteria, and the resulting supernatant was diluted in a 1:5 ratio with water and used in the assay. Sodium acetate was used to generate a standard curve.
For each infection, sixty 7 day old flies were divided equally into six separate vials containing a cellulose plug infiltrated with 3 mls of LB broth into which 300 μl of an overnight culture of the indicated strain had been added. After two days, flies were collected, rinsed in 70% ethanol to remove or lyse bacteria attached to the fly exterior, and homogenized in PBS. Serial dilutions of this suspension were plated on LB agar supplemented with streptomycin (100 μg/ml). After overnight incubation at 27°C, the resulting colonies were enumerated.
This was carried out as previously described [7]. Briefly, thirty flies divided into three vials per conditions were given access to LB broth alone or inoculated with the indicated strain of V. cholerae and supplemented with 1% fluorescein (Sigma) for the length of time noted. Ten flies were then washed, homogenized, and centrifuged. The fluorescence intensity of the resulting supernatants was recorded using a microplate spectrophotometer with fluorescence capability (Infinite 200, Tecan). Measurements were normalized to the fluorescence levels of flies fed LB alone and reported as a fluorescence ratio.
Thirty to forty-five female flies divided equally into three vials considered experimental replicates, treated as indicated, and harvested for mRNA quantification. For validation of RNAi constructs, whole flies were used. For intestine-specific transcription, intestines were dissected and removed 42 hours after exposure to V. cholerae. RNA was extracted using a High Pure RNA isolation kit (Roche Life Science) and treated with TURBO DNase treatment (Ambion). Quantification of total RNA was done with a NanoDrop 1000 spectrophotometer (Thermo Fisher Scientific), and quality was monitored by agarose gel electrophoresis. 500 ng of the resulting RNA was used for cDNA synthesis using a Quantitech Reverse transcription kit (Qiagen). Real time q-PCR was performed on the StepOnePlus real-time PCR system (Applied Biosystems) using iTaq Universal SYBR Green supermix (Bio-Rad). Relative expression was calculated using the 2-ΔΔCq method. RP49 (CG7939) gene transcription levels were used for normalization. Primers used are listed in S9 Table.
Detection of p-AKT: After three days of exposure to V. cholerae, 10 flies were homogenized in PBS (100 μl) and heated at 95°C for 15 min. Proteins in the resulting lysates were separated on a 12% SDS-PAGE gel (Biorad) and transferred to a PVDF membrane (Biorad) for hybridization. Primary antibodies were used in the following dilutions: Rabbit anti-Akt: 1:1,000 and Rabbit anti-phospho-Drosophila-Akt (Ser505): 1:1,000 (Cell Signaling Technology). HRP-conjugated anti-rabbit IgG was used in a 1:5,000 dilution as a secondary antibody (Cell Signaling Technology). Detection of MetO: After three days of exposure to fly food, LB, wild-type V. cholerae, or a ΔgcvT mutant, the intestines of 30 flies per condition were isolated, homogenized in PBS (100 μl), and heated at 95°C for 15 min. Proteins in the resulting lysates were separated on a 12% SDS-PAGE gel (Biorad) and transferred on to a PVDF membrane (Biorad) for hybridization. A methionine sulfoxide polyclonal antibody (Cayman) was used in a dilution of 1:200 as a primary antibody and an HRP-conjugated anti-rabbit IgG antibody (Cell Signaling Technology) was used in a 1:5,000 dilution as a secondary antibody. The tubulin loading control was visualized using mouse 12G10 anti-alpha-tubulin (DSHB) as a primary antibody in a dilution of 1:5000 and horse radish peroxidase-conjugated anti-mouse IgG in a 1:5,000 dilution (Cell Signaling Technology) as a secondary antibody. BSA-MetO protein was used as a positive control. Bands were quantified using ImageJ densitometry analysis.
For lipid droplet analysis of Drosophila intestines or fat bodies, infected flies were dissected, fixed in 4% formaldehyde, washed 3 times in PBS supplemented with 0.1% tween 20 (PBT), and stained with 1 μg/ml DAPI (Sigma) and 2 μg/ml Nile Red (Sigma). Human cells were incubated with the indicated supplements for 3 days, fixed with 4% paraformaldehyde, incubated with BODIPY 493/503 (Thermo Fisher Scientific) and DAPI for 30 mins, washed with PBS, and mounted on slides for confocal microscopy. Lipid droplet number and size were counted using the ImageJ particle analyzer. For phospho-Histone H3 (PH3+) staining, guts were incubated first with a polyclonal rabbit anti-phospho-Histone H3 (Ser10) antibody (EMD Millipore) diluted in a ratio of 1:500 in PBT supplemented with 2% BSA. Alexa Fluor 594-conjugated goat anti-rabbit IgG (H+L) antibodies (Thermo Fisher Scientific) were used in a 1:200 ratio to visualize PH3+ cells. Samples were then mounted in Vectashield mounting media (Vector Lab Inc) and imaged using an LSM700 confocal microscope (Zeiss).
Extraction of metabolites followed a previously published protocol [15]. Briefly, to prepare samples for metabolomics, bacteria were cultured overnight in LB broth. For supernatants, 1 ml of a bacterial culture was collected by centrifugation, filtered through 0.22 μm filter (Thermo Fisher Scientific) to remove remaining bacteria, and then combined with methanol to yield a methanol:water solution (80:20). For Drosophila studies, the intestines of twenty flies treated as indicated and derived from three independent vials were dissected and homogenized in 500 μl of a cold methanol:water solution (80:20). After incubation for 2h at -80°C, samples were centrifuged for 10 min at 14,000 X g. The supernatants were transferred to a new vial, and the pellets were again extracted with a methanol:water solution (400 μl) as described above. Supernatants from the two extractions were combined. Cecal fluid was prepared by centrifugation for 10 min at 5,000 X g to remove particulates, cells, and bacteria. Methanol was added to yield a methanol:water solution (80:20). This solution was incubated for 6h at -80°C and then centrifuged for 10 min at 14,000 X g. All methanol:water suspensions were lyophilized or dessicated at ambient temperature in a SpeedVac concentrator (Savant). The resulting suspensions were stored at -80°C until use. Metabolite pellets were resuspended in 20 μL LC/MS grade water, and 5 μL were injected over a 15 min gradient using a 5500 QTRAP triple quadrupole mass spectrometer (AB/SCIEX) coupled to a Prominence UFLC HPLC system (Shimadzu) via SRM of a total of 287 SRM transitions using positive and negative polarity switching corresponding to 258 unique endogenous water soluble metabolites. The dwell time was 3 ms per SRM resulting in ∼10–14 data points acquired per detected metabolite. Samples were separated using a Amide XBridge HPLC hydrophilic interaction liquid chromatographic (HILIC) column (3.5 μm; 4.6 mm inner diameter (i.d.) × 100 mm length; Waters) at 300 μl/min. Gradients were run starting from 85% buffer B (HPLC grade acetonitrile) to 40% B from 0–5 min; 40% B to 0% B from 5–16 min; 0% B was held from 16–24 min; 0% B to 85% B from 24–25 min; 85% B was held for 7 min to re-equilibrate the column. Buffer A was comprised of 20 mM ammonium hydroxide/20 mM ammonium acetate (pH = 9.0) in 95:5 water/acetonitrile. Peak areas from the total ion current for each metabolite SRM transition were integrated using MultiQuant version 2.1 software (AB/SCIEX) via the MQ4 peak integration algorithm using a minimum of 8 data points with a 20 sec retention time window.
The Folch method was used for extraction of lipids [52]. Briefly, the intestines of twenty female flies treated as indicated and harvested from three independent vials were removed and homogenized in a chloroform:methanol solution (2:1, 500 μl). After shaking for 30 minutes, 100 μl of a 0.9% NaCl solution were added, and the mixture was centrifuged for 5 min at 2, 000 rpm to separate the aqueous and organic phases. The lower aqueous phase was dessicated at ambient temperature using a SpeedVac concentrator (Savant). All samples were stored at -80°C prior to LC-MS/MS analysis.
Lipid samples were analyzed as previously described [53]. Briefly, samples were re-suspended in 35 μL of 50% isopropanol (IPA)/50% MeOH. 10 μL of sample were injected onto liquid chromatography tandem mass spectrometry (LC-MS/MS) system. A Cadenza 150 mm x 2 mm 3μm C18 column (Imtakt) heated to 40°C at 260 μL/min was used with a 1100 quaternary pump HPLC with room temperature autosampler (Agilent). Lipids were eluted over a 20 min gradient from 32% B buffer (90% IPA/10% ACN/10 mM ammonium formate/0.1% formic acid) to 97% B. A buffer consisted of 59.9% ACN/40% water/10 mM ammonium formate/0.1% formic acid. Lipids were analyzed using a hybrid QExactive Plus Orbitrap mass spectrometer (Thermo Fisher Scientific) in DDA mode using positive/negative ion polarity switching with 1 MS1 scan followed by 8 MS2 HCD scans per cycle (Top 8). DDA data were acquired from m/z 225–1450 in MS1 mode and the resolution was set to 70,000 for MS1 and 35,000 for MS2. MS1 and MS2 target values were set to 5e5 and 1e6, respectively. Lipidomics data were analyzed using LipidSearch 4.1.9 software (Thermo Fisher Scientific). The software identifies intact lipid molecules based on their molecular weight and fragmentation pattern using an internal library of predicted fragment ions per lipid class and the spectra are then aligned based on retention time and MS1 peak areas are quantified across sample conditions.
All bar graphs represent the mean of at least three biological replicates. Metabolomics experiments were performed in triplicate. For survival curves, thirty female flies were used. Graphpad prism 2.0 software was used to calculate means of pooled data and statistical significance for small data sets and survival curves. A student’s t-test was used to determine the significance of differences between two measurements. The significance of differences in survival curves was calculated using log-rank analysis. The p values for survival curves are shown in the respective graphs. For other data, statistical significance is indicated by stars placed above the compared values, which are defined in the legend. For analysis of metabolomics data, MetaboAnalyst software was used [54]. Data sets were normalized to sum. Means were calculated for each metabolite based on three values, and statistically significant means were assessed using a student’s t-test. In all cases, a p value of less than 0.05 was considered statistically significant.
Animal experiments were performed in accordance with standards outlined in the National Research Council’s Guide for the Care and Use of Laboratory Animals and Boston Childrens Hospital’s public health service Assurance. The protocol was approved by the Boston Children’s Hospital Institutional Animal Care and Use Committee (IACUC) appointed to review proposals for research involving vertebrate animals (Protocol number 14-06-2706). For euthanasia, rabbits were anesthetized with ketamine/xylazine followed by administration of fatal plus. All efforts were made to minimize distress, pain, and suffering.
|
10.1371/journal.ppat.1002005 | Invasive Extravillous Trophoblasts Restrict Intracellular Growth and Spread of Listeria monocytogenes | Listeria monocytogenes is a facultative intracellular bacterial pathogen that can infect the placenta, a chimeric organ made of maternal and fetal cells. Extravillous trophoblasts (EVT) are specialized fetal cells that invade the uterine implantation site, where they come into direct contact with maternal cells. We have shown previously that EVT are the preferred site of initial placental infection. In this report, we infected primary human EVT with L. monocytogenes. EVT eliminated ∼80% of intracellular bacteria over 24-hours. Bacteria were unable to escape into the cytoplasm and remained confined to vacuolar compartments that became acidified and co-localized with LAMP1, consistent with bacterial degradation in lysosomes. In human placental organ cultures bacterial vacuolar escape rates differed between specific trophoblast subpopulations. The most invasive EVT—those that would be in direct contact with maternal cells in vivo—had lower escape rates than trophoblasts that were surrounded by fetal cells and tissues. Our results suggest that EVT present a bottleneck in the spread of L. monocytogenes from mother to fetus by inhibiting vacuolar escape, and thus intracellular bacterial growth. However, if L. monocytogenes is able to spread beyond EVT it can find a more hospitable environment. Our results elucidate a novel aspect of the maternal-fetal barrier.
| Infection of the placenta and fetus is an important cause of pregnancy complications and fetal and neonatal morbidity and mortality. Listeria monocytogenes is an intracellular bacterial pathogen that causes pregnancy-related infections in humans. The pathogenesis of listeriosis during pregnancy is poorly understood. We have previously shown that transmission of L. monocytogenes from maternal cells and tissues to fetal cells occurs in the uterine implantation site, and that a small subpopulation of specialized fetal cells called extravillous trophoblasts are the preferred initial site of infection. Here we use primary human placental organ and cell culture systems to characterize the intracellular fate of L. monocytogenes in extravillous trophoblasts. We found that these cells entrap bacteria in vacuolar compartments where they are degraded and therefore reduce bacterial dissemination into deeper structures of the placenta. Our study provides new insights into the nature of the maternal-fetal barrier. Extravillous trophoblasts that are accessible to infection with intracellular pathogens from infected maternal cells have host defense mechanisms that constitute a bottleneck in maternal-fetal transmission.
| L. monocytogenes is a ubiquitous, facultative intracellular, Gram-positive bacterium that causes food-borne disease in humans and other mammals [1], [2]. Humans are exposed relatively frequently to L. monocytogenes: healthy adults in the United States are estimated to ingest 105 bacteria at least four times per year [3]. Ingestion of L. monocytogenes by an immunocompetent host is relatively innocuous, but in immunocompromised individuals and pregnant women listeriosis is a severe disease [1], [4]. In the US there are ∼530 cases per year of listeriosis during pregnancy (FDA, 2009). The clinical manifestations depend on the gestational age. During the second trimester L. monocytogenes is the cause of ∼3% of spontaneous abortions [5], [6]. Infection around term results in neonatal disease with mortality of up to 50% [7]. The mechanisms by which L. monocytogenes infects the placenta and crosses the maternal-fetal barrier are controversial and still poorly understood.
The intracellular life cycle of L. monocytogenes has been characterized in a variety of different cell lines as well as primary murine bone marrow-derived macrophages [8], [9]. L. monocytogenes is taken up either by phagocytosis or internalized via interaction of bacterial surface proteins, such as internalin A (InlA), with host cell receptors, such as E-cadherin [10], [11]. After internalization, the bacterium finds itself in an endocytic vacuole that develops into a late endosome and acidifies slightly [12]. Acidification activates the pore-forming toxin listeriolysin O (LLO) that is important for escape of the bacterium into the host cytosol, where L. monocytogenes replicates rapidly [13], [14]. The listerial protein ActA nucleates actin and allows L. monocytogenes to spread from cell-to-cell without exposure to the extracellular milieu [15].
The placenta has to protect the fetus from vertical transmission of pathogens while also providing an environment of immunological tolerance for the fetal allograft [16]. How the placenta accomplishes these contradictory tasks is unknown. It has long been postulated that the placenta is an immune-privileged organ that has diminished adaptive immune defenses in order to establish tolerance. However, low placental infection rates in the face of frequent pathogen exposure suggest that the organ itself must have defense mechanisms against infection.
What are the barriers of the placenta to infection and how is L. monocytogenes able to breach them? Understanding the structure of the human placenta is critical for addressing these questions. The placenta is comprised of maternal and fetal cells (Fig. 1) [17]. Prior to implantation, the maternal uterine lining transforms into the receptive decidua. Shortly after, specialized fetally derived cells called trophoblasts differentiate into several subpopulations that perform critical placental functions. Specifically, invasive extravillous trophoblasts (EVT) anchor the placenta in the uterus and invade the decidua and maternal spiral arteries. Consequently, maternal blood flows into the intervillous space, bathing the fetally derived villous trees. These villi are covered by a continuous layer of multinucleate syncytium (SYN), a specialized trophoblast layer that mediates gas, nutrient and waste exchange between mother and fetus. The syncytium is underlaid by progenitor cells called subsyncytial cytotrophoblasts (sCTB), which are separated by a basement membrane from the stroma of the villi where fetal capillaries are found.
We and others have previously shown that primary human placental organ cultures are relatively resistant to infection with L. monocytogenes [18], [19]. We utilized organ cultures from first trimester placentas to define where and how L. monocytogenes breaches the maternal-fetal barrier [19]. The syncytium that is in direct contact with maternal blood is highly resistant to infection. The other maternal-fetal interface is in the decidua where invasive EVT are in direct contact with maternal cells and tissues. Even though this interface has a much smaller surface area, it is the preferred site of initial placental infection. Indirect evidence suggested that EVT are capable of restricting intracellular growth and cell-to-cell spread of L. monocytogenes [19]. In this study, we further characterize the intracellular fate of L. monocytogenes in EVT.
We used a cell culture model system of primary human EVT [20] to understand how these specialized cells are able to delay or inhibit the intracellular life cycle of L. monocytogenes. We found that isolated EVT were able to restrict intracellular bacterial growth and spread. Furthermore, EVT prevented vacuolar escape and steered vacuolated bacteria towards degradation in lysosomes. This phenotype was strongest in invasive EVT; cells that in vivo are in contact with maternal tissues. Our results suggest that EVT have effective defense mechanisms against intracellular pathogens and form a significant bottleneck in the transplacental transmission of pathogens.
In order to examine the role of EVT in placental infection, we characterized the intracellular fate of L. monocytogenes in isolated primary human EVT. We used a well-established model system that has been previously used to study differentiation of progenitor cytotrophoblasts along the invasive pathway [21]. In this cell-culture system, cytotrophoblasts are isolated from second trimester placentas and induced towards differentiation along the invasive phenotype by culture on extracellular matrix such as Matrigel [22]. Tissue from the second trimester is used because the yield of cytotrophoblasts is higher than from first and term placentas [20]. Cytotrophoblasts are isolated by a series of enzymatic digestions and ficoll gradients. In order to assure a pure population of cytotrophoblasts, we performed a CD45-based depletion to remove any remaining immune cells prior to culture and differentiation. Cytokeratin 7, a cytotrophoblast marker [23], was used to determine the purity of the cell population. Generally our cell preparations contained over 95% cytotrophoblasts, with the remaining <5% being predominantly placental fibroblasts from the villous stroma (data not shown).
EVT were infected with wild type L. monocytogenes at a multiplicity of infection (MOI) of 5. Gentamicin was added to the culture medium at 1 hour post-inoculation (p.i.) to eliminate extracellular bacteria. We observed a ∼100-fold variation in susceptibility to infection across EVT from different placentas: 0.1–12% of EVT were infected, on average with one bacterium, at 2 hours p.i. This may be due to genetic differences between donors or heterogeneity in the condition of the placentas. We therefore normalized the growth of bacteria over time to the number of bacteria at the 2-hour time point for each placenta. Normalized data from 10 individual donor placentas was averaged to minimize the effects of individual differences. In contrast to almost all other previously studied cell types—which generally support L. monocytogenes growth [9]—intracellular bacteria in EVT decreased 2-fold between 2 and 5 hours p.i. and continued to decrease by 5-fold over the next 19 hours (Fig. 2).
For comparison, the three commercially available human trophoblast-derived choriocarcinoma cell lines (BeWo, Jeg3, and Jar) were also infected at an MOI of 5, which resulted in significantly greater infection: ∼14% of cells at 2 hours p.i. in BeWo (p = 0.004 by Student's T-test). Invasion of EVT and BeWo cells is InlA-dependent [18], [19], [24]. We observed lower levels of E-cadherin expression by immunofluorescence microscopy, the host cell receptor for InlA, in isolated EVT in comparison to BeWo cells (data not shown), which most likely accounts for the difference in invasion between these two cell types.
In all of the choriocarcinoma cell lines L. monocytogenes grew with a similar doubling time of about 77 minutes between 2 and 5 hours p.i. (Fig. 2 and data not shown). This is 2-fold slower than intracellular growth rates in murine macrophage cell lines [25], and in sharp contrast to the “halving time” of L. monocytogenes in EVT of about 310 min (p<10−4 by Student's T-test). In addition, we tested the fate of L. monocytogenes in primary human placental fibroblasts. These fibroblasts were isolated from a first trimester placenta [26] and propagated in culture for at least 10 generations before infection to ensure purity. To compare with EVT, placental fibroblasts were infected at an MOI of 60, which resulted in infection of ∼1% of cells at 2 hours p.i. Subsequently, L. monocytogenes grew with a doubling time of 40 minutes between 2 and 5 hours p.i. (Fig. 2).
L. monocytogenes grows rapidly in the host cell cytosol [9], while mutants that are unable to access the cytosol generally do not replicate [25]. Therefore, the lack of intracellular growth of L. monocytogenes in EVT could be due to an inability to escape from the primary vacuole. We thus determined whether L. monocytogenes can escape from the primary vacuole in EVT and whether bacteria can be found in the host cell cytosol.
To determine vacuolar escape, we utilized a L. monocytogenes strain that expresses red fluorescent protein (RFP) under the actA promoter (pactA-RFP). ActA nucleates actin, and its transcription is up regulated 200-fold in the host cell cytosol [27], [28]. Therefore, the expression of RFP in this strain correlates with entry of bacteria into the cytoplasm [29]. In addition, polymerization of host actin filaments around bacteria indicates cytosolic localization of L. monocytogenes and can be visualized by staining fixed cells with fluorescently labeled phalloidin, a compound that binds F-actin [30].
First we analyzed vacuolar escape rates of L. monocytogenes in BeWo cells. We infected BeWo cells with pactA-RFP L. monocytogenes and fixed the cells for immunofluorescence microscopy at 2, 5, 8 and 24 hours. The preparation was counterstained with polyclonal anti-Listeria antibody to visualize the total number of bacteria per cell. Microscopic inspection of BeWo cells at 8 hours p.i. showed that the vast majority of bacteria expressed RFP (Fig. 3A). The percentage of RFP-expressing bacteria increased from 17% to 95% between 2 and 8 hours p.i (Fig. 3C). Consistent with high vacuolar escape rates in this cell line, the number of bacteria that co-localized with phalloidin steadily increased from 13% to 76% over the same time period (Fig. 3C). Between 8 and 24 hours p.i. RFP expression remained at 95%, consistent with the long half-live of RFP [31], which led to RFP persistence in all bacteria that had escaped the primary vacuole. In contrast, phalloidin co-localization decreased to 53% at 24 hours p.i. The significant difference between RFP expression and phalloidin co-localization at 8 hours p.i (p = 0.04 by Student's T-test) and the observed decrease in phalloidin co-localization at 24 hours p.i. is most likely due to the fact that phalloidin staining provides a snapshot of intracellular bacteria that are in the actin-nucleating stage of their life cycle. In contrast to RFP expression, phalloidin does not co-localize with bacteria that have spread to neighboring cells and are still in the secondary vacuole. It is unlikely that host cell death at 24 hours p.i. contributes substantially to the decrease in phalloidin co-localization, because host cell death would lead to a significant decrease in intracellular bacteria as well [32], which we do not observe (Fig. 2).
In contrast, infection of EVT with pactA-RFP L. monocytogenes counterstained with polyclonal anti-Listeria antibody revealed that the vast majority of bacteria did not express RFP at 8 hours p.i. (Fig. 3B). Quantitation showed that less than 10% of L. monocytogenes expressed RFP over the 24-hour course of infection (Fig. 3C), compared to nearly 100% in BeWo cells (p<10−5 by Student's T-test). In addition, no co-localization of L. monocytogenes with phalloidin was observed in EVT (data not shown). Our results suggest that bacteria are unable to grow in EVT because they are trapped in the primary vacuole.
In the murine model of infection the virulence factor LLO, a cholesterol-dependent pore-forming cytolysin, is essential for vacuolar escape [33]. We evaluated the role of LLO in the intracellular fate of L. monocytogenes in EVT. We tested two bacterial strains: DP-L2161 which is deficient in LLO [34] and unable to grow in BeWo cells (data not shown) and DP-L4057 which has a mutation in LLO (S44A) that increases phagosomal escape in murine bone marrow derived macrophages [35]. The outcome of infection did not differ from wild type infection - both strains were eliminated in EVT over 24 hours - although with slightly different kinetics (see Supplementary Fig. S1). These results suggest the possibility that LLO function is impaired in EVT.
The maturation of L. monocytogenes-containing vacuoles has been studied in detail in murine macrophage cell lines (RAW 264.7 and J774A.1) [12]. Wild type L. monocytogenes escapes from a vacuolar compartment that includes the late endosomal marker Rab7. The early endosomal marker Rab5 does not associate with L. monocytogenes even at very early time points after phagocytosis. If the vacuole matures further and acquires the lysosomal marker Lamp1, the rate of vacuolar escape is minimal.
To characterize the vacuolar compartment that L. monocytogenes occupies in EVT, we examined these same markers. The early endosomal marker Rab5 was associated with less than 10% of bacteria at 2 and 5 hours p.i. (Fig. 4A). The late endosomal marker Rab7 was found to co-localize with 55% of bacteria at 2 hours p.i. and with over 40% of bacteria at all other time points through 24 hours of infection (Fig. 4B,C). Lamp1 co-localized with only 12% of bacteria at 2 hours p.i. and increased steadily to 40% at 24 hours p.i. (Fig. 4B,D).
To test whether the L. monocytogenes-containing vacuole in EVT becomes acidified, we used the acidotropic dye Lysotracker. Lysotracker staining followed a similar trend to Lamp1 staining: 17% of bacteria were found in an acidified compartment at 2 hours p.i., increasing to 51% at 24 hours p.i. (Fig. 4B,E).
While it is generally believed that L. monocytogenes replicates in the cytoplasm and not in vacuoles, there have been a few reports suggesting the possibility of slow replication in vacuolar compartments. Bhardwaj et al. described the presence of multiple bacteria in membrane-bound vacuoles in mononuclear cells in the liver of SCID mice with chronic listeriosis [36]. Furthermore, Birmingham et al. found that 13% of bacteria in a murine macrophage cell line were replicating slowly in autophagosome-like vacuolar compartments (LC3-positive, LAMP1-positive, non-acidified) and named these structures SLAPS (spacious Listeria-containing autophagosomes) [37]. We therefore evaluated whether L. monocytogenes co-localizes with the autophagy marker LC3, but found little to no co-localization in our system (Fig. 4A). We concluded that bacteria in EVT are trapped in vacuoles that mature into acidified lysosomes, suggesting that L. monocytogenes is degraded in this compartment.
To look more closely at the subcellular localization of L. monocytogenes in EVT, transmission electron microscopy was performed. EVT were infected with wild type L. monocytogenes at an MOI of 60. This high inoculum was used to increase the number of infected cells and the number of bacteria/cell for better visualization. Because the most significant decrease in intracellular bacterial numbers occurred between 2 and 5 hours p.i. (Fig. 2), infected EVT at those time points were examined (Fig. 5A,B). The number of vacuolar L. monocytogenes was enumerated: at both time points, 81–86% of bacteria were confined to vacuoles (Fig. 5C). These escape rates (14–19%) are slightly higher than those measured using the pactA-RFP strain above. This difference is significant (p = 0.004 by Student's T-test) and is likely due to differences in the infection (MOI of 5 versus 60) and/or due to a more limited detection threshold of RFP fluorescence as compared to electron microscopy. Furthermore, we enumerated the number of bacteria that appeared intact versus degraded. Intact appearing bacteria decreased from 67% to 50% between 2 and 5 hours p.i., and degraded bacteria increased from 33% to 50% during the same time interval (Fig. 5D,E,F).
A vacuolar compartment derived from the primary vacuole consists of a single lipid bilayer, whereas secondary vacuoles (a result of infection via cell-to-cell spread) and autophagosomes typically consist of two lipid bilayers [15], [38]. With a membrane contrast-enhancing stain and at higher magnification the membranes of the L. monocytogenes-containing vacuoles were visualized and appeared to consist of a single lipid bilayer (Fig. 5E). These ultrastructural results are consistent with bacterial entrapment in the primary vacuole and degradation in lysosomes.
We previously found that EVT are the preferred initial site of infection for L. monocytogenes in first trimester placental organ cultures [19]. Furthermore, we observed that L. monocytogenes is able to spread beyond the EVT along sCTB in some placentas. Under the conditions Robbins et al. used, such spread occurs in 50% of placentas over a time period of 72 hours. The inability of L. monocytogenes to escape from the primary vacuole in EVT could explain the delay or lack of listerial dissemination in placental organ cultures. Thus, we analyzed the rates of vacuolar escape in first trimester placental organ cultures infected with pactA-RFP L. monocytogenes and counterstained with polyclonal anti-Listeria antibody as described above (Fig. 6A,B). At 8 hours p.i., only 14% of bacteria had escaped the vacuole, while 39% and 37% were in late endosomes and lysosomes respectively (Fig. 6C). By 24 hours p.i. the percentage of RFP-expressing bacteria increased to 23% and the proportion of L. monocytogenes co-localizing with Rab7 and Lamp1 remained around 40% (Fig. 6C).
Vacuolar escape rates in placental organ cultures were somewhat higher than those observed in isolated second trimester EVT (Fig. 3C; p = 0.19 by Student's T-test). We therefore decided to test whether vacuolar escape rates differ between distinct trophoblast subpopulations. When isolated cytotrophoblasts are grown on Matrigel they differentiate along the invasive pathway and therefore consist of a more homogeneous EVT population [20], [21]. In contrast, there are several distinct subpopulations of trophoblasts in vivo and ex vivo that are in different stages of differentiation ranging from progenitor cytotrophoblasts near the stroma to invasive EVT at the outer villus margin. Therefore, infection of placental organ cultures leads to infection of a mixed population of trophoblasts.
To test whether listerial escape rates differ in different trophoblast subpopulations, we compared escape rates in three distinct populations of trophoblasts: (1) trophoblasts that were in contact with Matrigel (invasive border EVT), (2) trophoblasts that were surrounded by other trophoblasts on all sides (middle EVT), and (3) those that were in contact with the basement membrane and its underlying stroma (parastromal trophoblasts) (Fig. 7A–C). We increased the dose of L. monocytogenes to 2x107 bacteria/ml for 5 hrs before addition of gentamicin in order to achieve infection of all three subpopulations within one placenta at 24 hours p.i., and compared escape rates between invasive border EVT, middle EVT and parastromal trophoblasts at 24 and 48 hours p.i. (Fig. 7D,E). The average escape rate in invasive border EVT at 24 hours p.i. was 40% (range 11% to 55%). Because of this large variability between placentas from different donors, we normalized the escape rates in middle EVT and parastromal trophoblasts to the escape rate in invasive border EVT from the same placenta. At 24 hours p.i. we determined the fold-difference in escape rates in middle EVT and parastromal trophoblasts in comparison to the escape rate in invasive border EVT from the same placenta. Vacuolar escape rates increased the closer the trophoblasts were to the core of the placental villus. The average increase in escape rates compared to invasive border EVT was 1.21-fold in middle EVT and 1.51-fold in parastromal trophoblasts (Fig. 7E). At 48 hours p.i. we determined the fold difference in escape rates in all three subpopulations in comparison to the escape rate in invasive border EVT at 24 hours p.i., and found similar results. The average increase in escape rates was 1.14-fold (invasive EVT), 1.54-fold (middle EVT), and 1.76-fold (parastromal trophoblasts) (p = 0.02 by Student's T-test for combined 24- and 48-hour time points). We concluded that EVT at the invasive border—a cell type that is in direct contact with maternal cells in vivo—are especially prohibitive for listerial vacuolar escape. However, if L. monocytogenes is able to spread beyond the invasive EVT it can find a more hospitable environment.
Much of the pioneering work on the L. monocytogenes life cycle and intracellular growth kinetics has been performed in murine bone marrow derived macrophages as well as various murine and human cell lines [9], [15], [39]. In these cells, L. monocytogenes vacuolar escape rates are 80% or higher [29], [40], and bacteria grow rapidly (generation time of ∼40 min) in the nutrient-rich cytosol [25]. In contrast, the vacuolar escape rates in isolated primary EVT were less than 10%.
It is possible that vacuolar escape and growth rates vary depending on the specific cell type, especially in cells that play a role in host defense against infection. For example, primary murine dendritic cells are less hospitable to L. monocytogenes than primary bone marrow-derived mouse macrophages [40], [41]. Westcott et al. showed that bacterial doubling time is about 2-fold slower in primary murine dendritic cells (∼70 min), and only ∼40% of the bacteria escape into the cytosol. Specific endosomal maturation features in dendritic cells that are important for efficient processing and presentation of bacterial antigens to T cells are thought to be the underlying reason for these decreased vacuolar escape rates. Primary murine peritoneal macrophages are even more hostile to L. monocytogenes: Portnoy et al. has demonstrated that these cells kill ∼80% of L. monocytogenes during the first 2 hours p.i. and that surviving bacteria grow at a generation time of ∼120 min or longer [42]. Furthermore, if resident peritoneal macrophages are stimulated with IFNγ, bacterial growth is eliminated, and 95% of bacteria are found in vacuolar compartments [42]. While these activated professional immune cells are known to be critical in scavenging and containing infectious particles, it is more surprising that epithelial cells in the placenta, the EVT, would possess a similar bacteriocidal phenotype. In this context, it is interesting that IFNγ is crucial for a successful pregnancy and present at high levels at the maternal-fetal interface [43]. IFNγ is produced by uterine natural killer cells, which comprise approximately 20–40% of the leukocytes in the decidua [44], [45], and IFNγ receptors are expressed on human trophoblast cells throughout pregnancy [46]. It is possible that residual effects of in utero IFNγ exposure contribute to decreased vacuolar escape and increased bacterial degradation in EVT.
Why are bacteria not able to escape the vacuole in EVT? In the murine model of infection the virulence factor LLO is essential for vacuolar escape [33]. We found that lack of LLO or increased hemolytic activity of LLO did not alter the outcome of infection in EVT, suggesting that LLO function is impaired in this cell type. LLO-mediated pore formation is a pH dependent process, with a pH optimum of 5.5 [47], [48]. Although the Listeria-containing vacuole in EVT acidifies, the kinetics or extent of acidification could present unfavorable conditions for LLO function. For example LLO loses its hemolytic activity at neutral pH in less than 10 min [49]. Another possibility is that the Listeria-containing vacuole has a different lipid composition that renders LLO non-functional. LLO is dependent on the presence of cholesterol, which is utilized by EVT for the synthesis of progesterone [50]. Specialized hormone synthesis in EVT could lead to differences in cholesterol metabolism and/or distribution in these cells, rendering it inaccessible to vacuolar LLO. Moreover, the active form of a host-derived thiol reductase (GILT) involved in antigen processing has been shown to be required for the activation of LLO [51], and may not be present or accessible in the Listeria-containing vacuole in EVT. However, all of the above mentioned studies have been performed in the murine model of infection. In contrast, in many human cell types L. monocytogenes deficient in LLO is capable of vacuolar escape [25], [52], [53]. The mechanisms of LLO-independent vacuolar escape are poorly understood [54], but the existence of these examples opens up a myriad of other pathways that may be different in EVT, that ultimately could lead to vacuolar entrapment of L. monocytogenes. Further studies will be needed to assess these possibilities.
Work in several pregnant animal models of listeriosis supports our findings that L. monocytogenes has to pass several bottlenecks to infect the placenta and spread to the fetus. We have shown previously that the placenta in the pregnant guinea pig model is relatively protected from colonization, characterizing the kinetics of bacterial spread from maternal organs to the placenta and to the fetus [55]. The guinea pig placenta is colonized with 104-fold fewer bacteria than maternal liver and spleen after intravenous inoculation, and the bottleneck between placenta and fetus is again 1∶104 bacteria. Studies in the pregnant mouse and gerbil models also require high intravenous inoculums, >106 bacteria, to induce placental infection [56], [57].
Interestingly, there are several lines of evidence that suggest EVT are a suboptimal niche for the growth of intracellular pathogens in general. Human CMV infection, for example, is inefficient in trophoblasts, progresses slowly, and releases only small amounts of progeny virus [58]. Likewise, placentas infected with CMV in utero show rare viral replication in EVT, with membrane-clustered virions [59]. Recent studies with HIV-1 indicate that EVT are also non-permissive to HIV-1 replication due to active degradation and/or passive inactivation of critical viral replication mechanisms [60]. Others have observed that the majority of HIV-1 virions are trapped within endosomal compartments [61]. The common thread in these studies is that vacuolar or endosomal trafficking is hindering the normal life cycles of pathogens and preventing growth and spread of the virus or bacterium. While little is known about EVT in general, ultrastructural studies of uninfected human placentas report many unidentified vesicles and vacuoles in EVT [62], [63]. It is possible that the invasive role of EVT and their active degradation of extracellular matrix may require unique degradative and/or endosomal pathways that interfere with the life cycle of intracellular pathogens. As a result, EVT create a significant barrier to infection, and pathogens must get past the bacteriocidal EVT into more permissive cells in the placenta for the infection to progress.
If the primary site of infection is an inhospitable cell-type, then how does placental infection progress to cause pregnancy complications and fetal infection? One possibility is that even though EVT are the preferred site of initial infection with L. monocytogenes [19] and Toxoplasma gondii (our unpublished observations), and can harbor CMV in utero [64], they are a dead end for pathogens. This seems unlikely because we have not observed placental infection without infection of EVT, and L. monocytogenes can spread beyond EVT in some placentas [19]. In addition, other routes of crossing the trophoblast barrier appear even more difficult, since the syncytium is highly resistant to infection with L. monocytogenes [19] and T. gondii (our unpublished observations). It is possible that EVT could differ in their resistance to infection due to host genotypic differences. This would mean that some people are simply more predisposed to placental infection and pregnancy complications than others. However, to our knowledge no genetic basis for differences in susceptibility to vertical transmission has ever been identified.
We hypothesize that EVT can either contain or eliminate an infection until a certain threshold of cellular damage or placental inflammation is surpassed. For instance, non-infectious pregnancy complications that influence oxygen tension or pH in the placenta could alter the biochemical and/or physiological condition of EVT and decrease their resistance to infection. Co-infection with other pathogens could similarly escalate an immune imbalance at the maternal-fetal interface. If these imbalances threaten the healthy progression of pregnancy, spontaneous abortion or preterm labor are initiated to avoid continuation of pregnancy with a compromised placenta.
The placenta has developed a marvelous defense against infection, most likely consisting of multiple layers of physical and biochemical barriers. Both subpopulations of trophoblasts—syncytium and EVT—that are in direct contact with maternal cells and tissues are effective barriers against infection. The syncytium is in direct contact with maternal blood and is highly resistant to infection. The EVT are in close contact to maternal cells and tissues in the implantation site, and are the preferred initial sites for infection, but are inhospitable to a variety of intracellular pathogens. Both barriers can probably be breached by additional damage, resulting in infection of subsyncytial cytotrophoblasts, which appear to be more hospitable to intracellular replication of pathogens. Nevertheless, L. monocytogenes still has to pass another physical barrier: the basement membrane [19], to reach the villous stroma where the fetal capillaries are. L. monocytogenes will serve as an excellent model to characterize the precise molecular basis of the maternal-fetal barrier.
This study was conducted according to the principles expressed in the Declaration of Helsinki. The study was approved by the Institutional Review Board at the University of California, San Francisco, where all experiments were performed (H497-00836-29). All patients provided written informed consent for the collection of samples and subsequent analysis.
All chemicals were purchased from Sigma-Aldrich unless otherwise stated. For human placental organ cultures, placentas from elective terminations of pregnancy (gestational age 4 to 8 weeks) were collected and prepared as previously described [65]. Briefly, fragments from the surface of the placenta were dissected into 1–3 mm tree-like villi, placed on Matrigel (BD Biosciences, San Jose, CA)-coated Transwell filters (Millipore, Bedirica, MA, 30-mm diameter, 0.4 um pore size) and cultured in Dulbecco's modified Eagle-F12 medium (DMEM-F12; 1∶1, vol/vol) supplemented with 20% fetal bovine serum (FBS, Fisher Scientific), 1% L-glutamine and 1% penicillin/streptomycin (Invitrogen, Carlsbad, CA).
For EVT isolation, placentas from elective terminations of pregnancy (gestational age 14 to 24 weeks) were collected and prepared as previously described [21], [66]. Briefly, placentas from normal uncomplicated pregnancies were obtained immediately after aspiration and subjected to a series of enzymatic digestions followed by purification over a Percoll gradient. Remaining leukocytes were removed using a magnetic-bead-based EasySep CD-45 Depletion Kit with RoboSep device (Stem Cell Technologies, Vancouver, Canada). For growth curves, purified cells were plated on Matrigel-coated Transwell filters (Millipore, Bedirica, MA, 12-mm diameter, 0.4 um pore size) in serum-free DMEM-high glucose, with 2% Nutridoma (Roche Diagnostics, Indianapolis, IN), 1% L-glutamine, 1% sodium pyruvate, 1% 25 mM HEPES, 1% penicillin/streptomycin at a concentration of 1.25×105 cells/transwell. For immunofluorescence microscopy, purified cells were plated on Matrigel-coated 6-well plates at a concentration of 2×106 cells/well.
Placental fibroblasts were isolated as described [26] from a placenta at gestational age 8 weeks, and were cultured in DMEM-high glucose with 10% FBS, 18% M-199, 1% penicillin/streptomycin. For growth curves and immunofluorescence microscopy, cells were plated on glass coverslips in 24-well plates at 2.5×105 cells/well.
The choriocarcinoma cell line BeWo (ATCC CCL-98) was cultured in Ham's F12 medium with 10% FBS, 1% L-glutamine, 0.15% sodium bicarbonate, 1% penicillin/streptomycin. For growth curves and immunofluorescence microscopy, cells were plated on glass coverslips in 24-well plates at 2.5×105 cells/well.
L. monocytogenes 10403S expressing green fluorescent protein (GFP) (strain DH-L1252) was a gift from Darren Higgins [67]. The pactA-RFP strain (PL512) was constructed as follows: The ORF encoding TagRFP from Entacmaea quadricolor [31] was codon optimized for expression in L. monocytogenes using Gene Designer software [68] and the gene was synthesized de novo (DNA2.0, Menlo Park, CA). The synthetic gene was cloned downstream of the actA promoter in the vector pPL2 and stably integrated at the tRNAArg locus of the bacterial chromosome in the wild type L. monocytogenes strain DP-L4056 as described previously [69]. Molecular constructs were confirmed by DNA sequencing. For infections, bacteria were grown overnight to stationary phase in BHI (Brain Heart Infusion broth) at 30°C and washed once with PBS before dilution and infection.
Cells were incubated in antibiotic-free medium for 1 hr before infection. Bacteria were added for 30 minutes, followed by three washes with PBS and addition of antibiotic-free medium. For CFU (colony forming units) determination gentamicin (50 µg/ml) was added at 60 minutes p.i. EVT were inoculated with 3×106 bacteria/ml (MOI 5), and placental fibroblasts with 4×107 bacteria/ml (MOI 60). At indicated times, cells were lysed with distilled water, aliquots were plated on BHI agar plates, and CFU were enumerated. Infection for immunofluorescence microscopy was performed as outlined above with following modification: at 60 minutes p.i. Matrigel was dissolved by incubation with BD Cell Recovery Solution (BD Biosciences, San Jose, CA) for 40 minutes, and cells were re-plated on fresh Matrigel on Transwell filters in media containing gentamicin (50 µg/ml). Therefore, gentamicin was added at 1 hour 45 minutes p.i. to infected cells that were analyzed by immunofluorescence microscopy. CFU after exposure to the enzymatic solution and gentamicin addition at 1 hour 45 min were not significantly different from those under standard CFU (gentamicin at 1 hour p.i.) conditions (data not shown). For electron microscopy, EVT were infected as above with the following alteration: the infectious dose was 4×107 bacteria/ml (MOI 60). Infection of placental explants was performed as previously described [19] with the following alteration: the infectious dose was lowered to 3×106 bacteria/ml for 30 minutes.
Explants were fixed in 3% paraformaldehyde, passed through a sucrose gradient and snap-frozen in OCT (Ted Pella, Redding, CA). Histological slicing was performed on a Hacker-Slee cryostat. Glass slides with sections were incubated in acetone, soaked in blocking solution (1% bovine serum albumin (BSA) in PBS), then incubated with primary antibodies, rinsed in PBS, incubated with secondary antibodies, and affixed over Vectashield mounting medium with DAPI (Vector Laboratories, Burlingame, CA).
Cultured cell lines and EVT were fixed in 3% paraformaldehyde. For Lysotracker visualization, the dye was added to cells for 30 minutes at 5 µM and washed in PBS before fixation. For Rab7 staining, cells were rinsed in glutamate lysis buffer (25 mM HEPES, 25 mM potassium chloride, 2.5 mM magnesium acetate, 5 mM EGTA, 150 mM K-glutamate), dipped into liquid nitrogen, rinsed in lysis buffer, and fixed in paraformaldehyde. Transwell filters were cut out of wells, blocked and permeabilized in 1% BSA and 0.1% Triton-X100, then stained as described above in BSA/TritonX-100/PBS solution.
Primary antibodies: polyclonal rabbit Listeria O antiserum (1∶1000 BD Biosciences, San Jose, CA), mouse polyclonal Lamp1 antiserum (1∶100 DSHB at University of Iowa), mouse monoclonal Rab5 antibody (1∶100, BD Biosciences, San Jose, CA), mouse monoclonal LC3 antibody (1∶100, gift from Dr. Jay Debnath), rabbit monoclonal Rab7 antibody (1∶1000, gift from Dr. Suzanne Pfeffer). Secondary antibodies: Alexa Fluor 594 goat anti-mouse IgG (1∶500, Invitrogen), Alexa Fluor 488 and 594 goat anti-rabbit IgG (1∶1000 & 1∶500, Invitrogen).
Slides were viewed using an inverted TE2000-E microscope (Nikon, Tokyo, Japan) equipped with a 12-bit cooled CCD camera (Q imaging, Surrey, Canada). Images were collected using Simple PCI software (Hamamats, Sewickley, PA).
For the 2-hour time point, cells were fixed overnight at 4°C in 3% glutaraldehyde, 1% paraformaldehyde in 0.1 M cacodylate buffer. Fixed cells were post-fixed with 2% osmium tetroxide, dehydrated in ethanol and embedded in Epon. Thin sections (70 nm) were cut using a Leica Ultracut-UCT Microtome (Leica Microsystems, USA). Observations were made under a Philips Tecnai 10 transmission electron microscope (Department of Pathology, UCSF), and digital acquisition was performed with a CCD camera (Maxim DL Software, Cyanogen, Canada). For the 5-hour time point, cells were fixed as above, and post-fixed with 1% osmium tetroxide and 1.6% potassium ferrocyanide, stained with 5% uranyl acetate solution, dehydrated with ethanol and embedded. Sections were cut using a microtome (RMC MTX, Reichert Ultracut E, RMC MT6000) and observations made under a Philips Tectani 12 transmission electron microscope (EM lab, UC Berkeley). For quantification, 100 bacteria at each time point were counted and categorized by cytoplasmic versus vacuolar localization and intact versus degraded bacteria.
Images were prepared using ImageJ (RSB, Bethesda, MD), Photoshop and Illustrator (Adobe, San Jose, CA). RGB hues were linearly adjusted but no non-linear alterations were performed.
|
10.1371/journal.ppat.1003948 | Dimerization of VirD2 Binding Protein Is Essential for Agrobacterium Induced Tumor Formation in Plants | The Type IV Secretion System (T4SS) is the only bacterial secretion system known to translocate both DNA and protein substrates. The VirB/D4 system from Agrobacterium tumefaciens is a typical T4SS. It facilitates the bacteria to translocate the VirD2-T-DNA complex to the host cell cytoplasm. In addition to protein-DNA complexes, the VirB/D4 system is also involved in the translocation of several effector proteins, including VirE2, VirE3 and VirF into the host cell cytoplasm. These effector proteins aid in the proper integration of the translocated DNA into the host genome. The VirD2-binding protein (VBP) is a key cytoplasmic protein that recruits the VirD2–T-DNA complex to the VirD4-coupling protein (VirD4 CP) of the VirB/D4 T4SS apparatus. Here, we report the crystal structure and associated functional studies of the C-terminal domain of VBP. This domain mainly consists of α-helices, and the two monomers of the asymmetric unit form a tight dimer. The structural analysis of this domain confirms the presence of a HEPN (higher eukaryotes and prokaryotes nucleotide-binding) fold. Biophysical studies show that VBP is a dimer in solution and that the HEPN domain is the dimerization domain. Based on structural and mutagenesis analyses, we show that substitution of key residues at the interface disrupts the dimerization of both the HEPN domain and full-length VBP. In addition, pull-down analyses show that only dimeric VBP can interact with VirD2 and VirD4 CP. Finally, we show that only Agrobacterium harboring dimeric full-length VBP can induce tumors in plants. This study sheds light on the structural basis of the substrate recruiting function of VBP in the T4SS pathway of A. tumefaciens and in other pathogenic bacteria employing similar systems.
| Agrobacterium tumefaciens causes crown gall disease (tumors) in agriculturally important plant species. It initiates infection through its Ti plasmid, which integrates a portion of its own DNA (T-DNA) into that of the host genome. The T-DNA is bound to VirD2 relaxase, and this complex is required for the efficient translocation and integration of the T-DNA into the plant genome for tumor formation. Two additional proteins, among others, are also required for Agrobacterium tumorigenesis: VirD4-coupling protein (CP) and VirD2-binding protein (VBP). VBP is responsible for recruiting VirD2–T-DNA to VirD4 CP to help localize T-DNA to the Type IV Secretion System apparatus for transfer. However, it is still unclear how VBP recruits the complex to VirD4 CP. Here, we report the crystal structure and associated functional studies of the C-terminal domain of VBP. We show that the C-terminal domain is the dimerization domain of VBP and only dimeric VBP is functional and essential for the induction of tumor in plants. This study enhances the understanding of the role of VBP in recruiting VirD2–T-DNA in A. tumefaciens prior to its transfer into the host plant. This mode of action can be extended to other pathogenic bacteria employing similar secretion systems.
| The Type IV Secretion System (T4SS) has an unmatched versatility among the seven different secretion systems known in bacteria. T4SS can translocate not only proteins but also DNA to phylogenetically diverse taxa, including many bacterial species and various eukaryotic cells, as well as import and export DNA from the extracellular milieu. The T4SS shares a common ancestry with bacterial conjugation systems [1], with three types of T4SS described to date: (1) conjugation systems, defined as machines that translocate DNA substrates to recipient cells by a contact-dependent process; (2) effector translocation systems, functioning to deliver proteins or other effector molecules to eukaryotic target cells; and (3) DNA release or uptake systems, which translocate DNA to or from the extracellular milieu [2].
The VirB/D4 secretion system of Agrobacterium tumefaciens is a typical example of a T4SS, which serves as both a conjugation and an effector translocation system. Conjugation systems within the T4SS facilitate the translocation of protein-DNA complexes from the bacterium into the cytoplasm of the host cell. For instance, A. tumefaciens translocates to the host cell a segment of the Ti (tumor inducing) plasmid between the right and left borders (T-DNA) in complex with a cytoplasmic relaxase protein (VirD2) [3]. Proteins involved in T-DNA processing and translocation are classified into three functionally distinct classes [4]. Class I constitutes proteins involved in the processing of the DNA intermediates, such as VirC1, VirC2, VirD1 and VirD2 relaxase [5]. VirC1 and VirC2 proteins assemble the relaxosomal complex at the border sequences of the Ti plasmid and initiate T-DNA processing [6], whereas VirD2 relaxase, when present in the relaxosomal complex, cleaves the bottom strand of the T-DNA, where it remains covalently attached to the 5′ end of the single-stranded T-DNA [7]. Class II comprises 11 VirB proteins that form the T4SS apparatus which is responsible for the translocation of the VirD2–T-DNA complex and the effector proteins into the host cell cytoplasm [4], [8]. Class III constitutes the coupling proteins (CP), which mediate the interaction between the substrate (VirD2–T-DNA complex) and the transport apparatus [9]. VirD4 CP, the coupling protein in the A. tumefaciens VirB/D4 system, is an inner membrane protein with a large cytoplasmic domain essential for the transfer of both the T-DNA strand and VirE2 to host cells [10]–[12]. However, the VirD2–T-DNA complex and VirE2 are translocated separately [7], [13], [14]. The translocated effector proteins— VirE2, VirE3 and VirF which are involved in nuclear targeting, import and integration of the T-DNA into the host genome—mediate the interactions between the VirD2–T-DNA complex and host cellular factors [1]. In particular, VirE2 coats single-stranded T-DNA and protects it from host cell nucleases [15], whereas both VirE2 and VirD2 carry the nuclear localization signals that aid in the nuclear import of the T-DNA. VirE2 also interacts with host cell proteins, such as VIP1 and VIP2 [16]. VIP1 and VIP2 localize to plant nuclei and probably facilitate delivery of the T-DNA complex to its site of integration [16] VirF, on the other hand, localizes in the host nucleus and targets VIP1 and VirE2 for proteolysis and thus uncoats the T-DNA from its cognate proteins. This uncoating mechanism is a crucial step that must occur prior to integration of the T-DNA in the host genome [17].
The VirD2–T-DNA complex forms in the bacterial cytoplasm, but there is no evidence to suggest that VirD4 CP can recruit the bulky substrate complex to the T4SS apparatus [18]. In 2007, Guo et al. reported the existence of a subset of proteins defined as ‘recruiting proteins’ that are involved in bringing the nucleoprotein substrate complex formed in the bacterial cytoplasm to the VirD4 CP. The VirD2-binding protein (VBP) was subsequently identified as a key protein belonging to this subset and was shown to recruit the VirD2–T-DNA complex to the VirD4 CP [8]. Site-directed mutagenesis experiments have indeed shown that the interaction between VBP and VirD2 is important for T-DNA transfer [8], with VBP interacting with both the VirD2–T-DNA complex and with several of the T4SS components independently, including VirD4 CP, VirB4 and VirB11. However, the molecular mechanism(s) by which VBP recruits the complex to VirD4 CP remains unknown.
A. tumefaciens is a gram-negative bacteria that causes crown gall disease (tumor formation) in over 140 species of dicots [19].Despite this negative role, researchers have started to exploit the ability of Agrobacterium to infiltrate and infect other organisms to create transgenic plants. Thus, given the importance of this bacterium in both crop infection and as a research tool, we sought to better understand how VBP recruits the protein-DNA complex to VirD4 CP by exploring the structure of VBP.
Here we report the crystal structure of the C-terminal domain of VBP along with its functional studies. We show that the two monomers of the asymmetric unit form a tight anti-parallel dimer, with the dimerization confirmed by solution studies. Moreover, we demonstrate that this C-terminal domain is the dimerization domain of full-length VBP. Pull-down analyses and plant virulence assays showed that only Agrobacterium harboring dimeric VBP can interact with the key proteins to induce tumor formation in plants. These studies broaden our understanding about the role of VBP in the VirD2–T-DNA transfer from A. tumefaciens to the plant cell.
We initially attempted to crystallize full-length VBP with intact N- and C-terminal domains (Figure 1A). Peptide-mass finger printing analysis on the crystals showed that only the C-terminal domain of VBP had crystallized. The boundaries of the crystallized protein were determined using N-terminal sequencing and mass spectrometric analysis. Subsequently, we generated a new construct that consisted of the C-terminal domain, and obtained crystals that diffracted up to 2.7 Å resolution. The structure was determined using the single wavelength anomalous diffraction (SAD) method (Table 1).The asymmetric unit consists of two monomers forming a tight dimer. Each monomer contains a six-helix bundle with three long anti-parallel α helices (α1, α2, and α6) that forms the major part of this domain, and three shorter helices (α3, α4 and α5) that stack at an angle to the long helices (Figure 1B, Figure S1A, S1B). The monomers are arranged as anti-parallel dimers with a large buried surface area of 1203 Å2 (or 14% of the total surface area of each monomer). The dimer is held together by tight interactions between α1 and α2 helices from both monomers (Figure 1C).
PSI–BLAST searches of the non-redundant protein database using the full length VBP (gi|159141484) resulted in several hits of nucleotidyltransferase proteins from Rhizobium, Sinorhizobium and other species of Agrobacterium. Sequence-based predictions revealed that VBP has two domains: an NT_KNTase -like domain at the N-terminus and a HEPN (higher eukaryotes and prokaryotes nucleotide-binding) domain at the C-terminus. The NT (nucleotidyltransferase) domain is implicated in nucleotidyl transfer function, whereas the HEPN domain is implicated in nucleotide binding[20]. A DALI [21] search for the structural homologs of VBP C-terminal domain identified several proteins with a HEPN domain (Table S1); this confirmed that the C-terminal domain of VBP adopts a HEPN fold (hereafter referred to as HEPN domain). Notably, the HEPN domain of VBP aligns with the C-terminal domain of kanamycin nucleotidyltransferase from Staphylococcus aureus (PDB code: 1KNY) with an rmsd of 2.9 Å for the 104 Cα atoms. Although VBP has very low sequence identity (14 to 16%) with its structural homologs, it might have similar nucleotide binding and transfer function.
The crystal structure of the HEPN domain shows the presence of a tight dimer in the asymmetric unit. The molecular mass based on the sequence of VBP is 37.5 kDa (including the 6His tag). However, size-exclusion chromatography showed that VBP elutes as a single peak at an elution volume corresponding to an apparent molecular mass of 75 kDa (Figure 2A). Further, the analytical ultracentrifugation (AUC) analysis showed that VBP sediments as a single species corresponding to an apparent molecular mass of dimeric VBP (75 kDa) (Figure 2B). Taken together, these results show that VBP forms a homodimer in solution and likely functions as a dimer in the cell.
We examined the role of the HEPN domain in VBP oligomerization by generating individual constructs for the N-terminal (NT) and C-terminal (HEPN) domains of VBP. The size-exclusion chromatography of the purified proteins showed that the NT domain elutes as a single peak at an elution volume corresponding to an apparent molecular mass of 16.8 kDa (NT as monomer), whereas the HEPN domain elutes as a single peak at an elution volume corresponding to an apparent molecular mass of 37 kDa (HEPN as a dimer) (Figure S2). These results were further confirmed using analytical ultracentrifugation experiments (Figure 3A, 3B), and are consistent with the crystal structure findings that show the presence of tight dimeric HEPN domains in the asymmetric unit. Taken together, these results suggest that HEPN domain is responsible for the dimerization of the full-length VBP.
The structural analysis indicated that the HEPN dimer is held together by contacts throughout helices α1 and α2; in particular, residues Asp173, Lys184 and Asn186 of the HEPN domain play important roles in maintaining the dimer. Asn186 is located at the edge of a loop connecting the α1 and α2 helices of the dimer interface (Figure 1C). This residue, along with Lys184 and Asp173 of one monomer, is involved in hydrogen bonding contacts with Asp173 and Asn186, Lys184 of the second monomer. We found that a single substitution of Asp173Asn, Lys184Asp or Asn186Asp in the HEPN domain is sufficient to disrupt dimer formation, as verified by size-exclusion chromatography and analytical ultracentrifugation experiments (Figure 4A, Figure S3A, S3B and Figure S4).
Next, we verified the role of these key residues using full-length VBP. Single point substitutions of Asp173Asn, Lys184Asp or Asn186Asp in full-length VBP caused the dimer to break and the protein to elute as a single peak at an elution volume corresponding to an apparent molecular mass of 37.5 kDa (monomeric VBP). Further AUC analyses of full-length VBP bearing one of these point substitutions showed that the protein sediments as a single species at an apparent molecular mass corresponding to monomeric VBP (37.5 kDa) (Figure 4B, Figure S5A, S5B and Figure S6). The circular dichroism (CD) spectrum of the wild-type VBP, mutated VBP and HEPN domain suggested that these mutants have the same structural fold as the wild-type proteins (Figures S7A, S7B). These results indicate that the hydrogen bonds at the dimeric interface play a key role in maintaining an intact dimeric HEPN domain and reiterate the involvement of the HEPN domain in VBP dimerization.
VBP is the key recruiting protein that binds to the VirD2–T-DNA complex and brings it into contact with VirD4 CP for subsequent translocation to the host cell [8]. Thus, we sought to verify the binding property of VBP with VirD2 and VirD4 CP using pull-down assays. Using an in vitro pull-down assay with recombinant proteins (6His-VBP and MBP-VirD2), we showed that wild-type VBP binds to VirD2, whereas VBP with one of these aforementioned single point substitutions—Asp173Asn, Lys184Asp or Asn186Asp (which results in monomeric VBP)—does not bind VirD2 (Figure 5A). To confirm these results, we used His-VBP and substituted His-VBP (with (Asp173Asn/Lys184Asp/Asn186Asp mutations) as bait and pulled down the VBP-interacting proteins from a vir gene-induced Agrobacterium null mutant that lacks VBP: GMV123 (A triple vbp null mutant strain GMI9017Δvbp2Δvbp3 (for which all three existing vbp genes were knocked out)). WT His-VBP was able to pull down VirD2 and VirD4 CP, whereas substituted VBP could not pull down either of these proteins. These results indicate that the dimeric nature of VBP is important for its interaction with VirD2 and VirD4 CP (Figure 5B).
The observed dimeric nature of VBP in solution and in the crystal structure prompted us to verify the functional state of VBP inside the cell using a plant virulence assay. GMV123 (vbp KO strain of A. tumefaciens) was complemented with pQH300 plasmid harboring substituted constructs for a functional vbp gene, a gene expressing the HEPN domain, or a gene expressing the NT domain. We found that null mutants transformed with a plasmid expressing VBP were able to cause a tumor-like phenotype in the wild-type plants (Figure 6A and 6B and Figure S8). Strains expressing a substituted VBP (Asp173Asn/Lys184Asp/Asn186Asp) or one of the other deletion mutants (pQH-NTD or pQH-HEPN) did not cause tumors (Figure 6A and 6B and Figure S8). These results indicate that VBP functions as a dimer in the cells and that full-length VBP is required for tumor formation in plants.
Proteins homologous to VBP are predicted to bind ATP [22]. A previously reported structure of kanamycin nucleotidyltransferase (PDB code: 1KNY), a structural homolog of VBP, in complex with an ATP analog and kanamycin, shows that the nucleotide binding pocket involves residues from the N-terminal domain of one monomer and the C-terminal domain of the second monomer. We verified the ATP binding property of VBP using isothermal titration calorimetry (ITC) (Figure 7A and Table S2). The ITC analysis showed that VBP binds to the ATP analog (AMPPNP) (Kd = 2.0 µM), whereas VBP with the three point substitutions described above does not bind to AMPPNP (Figure 7B, Figure S9A and S9B), indicating that only dimeric VBP binds to ATP. Further, using ITC experiments, we sought to verify whether the HEPN domain alone can bind nucleotides (Figure 7C). Our results indicated that the HEPN domain alone cannot bind nucleotides. We infer that, similar to the kanamycin nucleotidyltransferase, the ATP binding in VBP might involve both N-terminal and C-terminal domains. Notably, the structure of the kanamycin nucleotidyltransferase (PDB code: 1KNY) complexed with a nucleotide analog and kanamycin shows that the two monomers of the dimer interact in an anti-parallel fashion to form the ATP binding pocket. Similarly, the structure of the HEPN domain from VBP shows that the two HEPN monomers form a tight dimer in which the monomers run in anti-parallel. Although the relative orientation of monomers in the dimers of both proteins is not the same, both proteins might have a similar ATP-binding mechanism.
A. tumefaciens affects more than 140 species of dicots [19], instigating infection through the efficient translocation of the VirD2–T-DNA complex [2], a prerequisite for the integration of T-DNA into the plant genome and eventual tumor formation in plants [23]. The T-DNA is a segment of the Ti plasmid that encodes most of the proteins that are involved in VirD2–T-DNA complex translocation and T-DNA integration into the host genome, with each protein catering to a particular stage of the translocation process. The VirD4 CP is known to couple the VirD2–T-DNA complex mediated by VBP as a recruiting complex (VBP: VirD2–T-DNA) to the T4SS secretion apparatus. VBP is thus a key component of the recruiting complex [8].
Based on the sequence analysis, it has been predicted that the NT domain of VBP belongs to the DNA polymerase β superfamily of proteins [21]. This superfamily includes nucleotidyltransferases that catalyze nucleotidylation of proteins in yet unidentified pathways [21]. Similarly the C-terminal of VBP is predicted to have the HEPN domain [24]. VBP interaction with VirD2, and several other energizing components of the T4SS (VirD4, VirB4 and VirB11) [13], makes it difficult to predict the exact nucleotidylation site of the protein.
In this study, we sought to analyze the structural and functional aspects of the C-terminal domain of VBP. We show that this domain adopts a HEPN fold and facilitates the dimerization of VBP, forming tight anti-parallel dimers. The structural similarity observed between the HEPN domain of VBP and kanamycin nucleotidyltransferase, as well as the results from our ITC experiments, suggest that a nucleotide binding pocket is formed by the dimeric interface in this protein. Furthermore, our pull-down assays show that only dimeric VBP can bind VirD2 and VirD4 CP and that dimerization of VBP is essential for A. tumefaciens-induced tumor formation in plants (Figures 5, 6 and 8).
Previously, we showed that VBP has independent interactions with VirD4 CP, VirB4 and VirB11 ATPases [8]. VirD4 CP has a prominent cytoplasmic domain, besides its membrane embedded domain [1], whereas VirB4 and VirB11 have both membrane and the cytoplasmic regions [25]. Our previous studies have shown that VBP is a cytoplasmic protein that localizes at the poles only in the presence of T4SS [8]. This is probably because of the interaction between VBP and the cytoplasmic regions of the T4SS proteins, particularly VirD4 CP, VirB4 and VirB11. Furthermore, the polar localization of VBP does not depend on the presence of VirD2 [8].
Although VirD2 is a cytoplasmic protein, it localizes to the poles in the presence of VBP and T4SS apparatus, but remains in the cytoplasm in the absence of VBP, irrespective of the presence of T4SS apparatus [8]. This indicates that when T4SS is present, VBP binds to the cytoplasmic region of VirD4 CP either independently or as a complex with VirD2, when it is available. In light of this finding, we propose that either VBP binds to the VirD2–T-DNA complex and recruits it to the VirD4 CP, or VBP initially binds to VirD4 CP proteins at the cytoplasmic region and serves as a docking station for the VirD2–T-DNA complex (Figure 8). Using a transfer DNA immunoprecipitation (TrIP) assay, Cascales et al. elegantly showed that T-DNA recruited to VirD4 CP is transferred to VirB6 through VirB11 [7] and that VirB4 and VirB11 ATPases interact to drive the export of T-DNA [7].
The present study sheds light on the role of VBP in the VirD2–T-DNA complex translocation in VirB/D4 T4SS of A. tumefaciens and other similar bacterial system that use the Type IV secretion conjugation systems.
The strains and plasmids used are given in Table S3. Intact vbp and vird2 genes were amplified from A. tumefaciens C58 plasmid and Ti plasmid, respectively. These genes were then cloned to pET32a (Novagen; Madison, WI, USA) and pRSET (Invitrogen, Carlsbad, CA, USA) vectors, respectively. N-terminal nucleotidyltransferase (NT) domain and C-terminal HEPN domain constructs were created using specific primers that amplify these regions and were cloned into pGEX-6p1 (GE Healthcare; Buckinghamshire, UK). Site-specific mutations in vbp were introduced by overlapping PCR, as described previously [26]. Each construct was verified by DNA sequencing. A fragment of virF cassette cloned from pTiBo542 (GenBank: DQ058764.1) was inserted into the SphI–ApaI site on pCB301 [27]. The virF coding sequence was substituted with a multiple cloning site, resulting in pQH300.
The plasmid pET32a-vbp was transformed into E. coli BL21 (DE3) cells and was grown in LB broth at 37°C overnight. The overnight culture was transferred into 1 L of LB broth and the protein expression was induced at an absorbance of 0.6 with 350 µM IPTG for 20 h at 20°C.Cells were harvested and lysed in lysis buffer (20 mM Tris-HCl, pH 8.0, 200 mM NaCl and 1 mM PMSF). Cell lysates were centrifuged and the supernatants transferred to affinity columns containing Ni-NTA agarose (Qiagen; Valencia, CA, USA), pre-equilibrated with the lysis buffer. The 6His-VBP bound to Ni-NTA was eluted with 400 mM imidazole following three wash steps to remove non-specific proteins. The eluted protein was purified through size-exclusion chromatography using a HiLoad 16/12075 Superdex75 gel filtration column (AKTA FPLC UPC-900 system, GE Healthcare) containing buffer (20 mMTris-HCl, pH 8.0, 200 mM NaCl, and 5% glycerol). The GST fusion proteins (GST-HEPN and GST-NTD) were expressed as described above using M9 media [28]. The fusion proteins were purified by affinity chromatography on GST-Sepharose resin, and the tags were removed by cleavage with PreScission proteases (GE Healthcare; Buckinghamshire, UK). The HEPN domain was additionally purified by size-exclusion chromatography in gel-filtration buffer (30 mM CHES pH 9.0, NaCl 200 mM, 5% glycerol). The NT domain was purified in the same way but using a buffer containing 30 mM Tris-HCl, pH 7.5 and 200 mM NaCl buffer.
Initial crystallization conditions were identified by hanging drop vapor diffusion method using an index screen (Hampton Research, Aliso Viejo, CA, USA). Diffraction-quality crystals were obtained by equilibrating l.0 µl drop of protein (4 mg/ml) in 30 mM CHES, pH 9.0, 200 mM NaCl and 5% glycerol mixed with 1.0 µl of reservoir solution (8% (w/v) PEG 3350, 2% v/v tacsimate, 5% v/v 2-proponal, and 0.1 M imidazole) suspended over 1 ml of reservoir solution. Crystals grew in 1–3 days at 16°C. For data collection, 15% glycerol was added as a cryo-protectant and the crystals were flash-cooled in an N2 cold stream.
A complete single wavelength anomalous diffraction (SAD) [29] dataset was collected to 2.7 Å resolution at the synchrotron beamline X6A (National Synchrotron Light Source, Brookhaven National Laboratory, Upton, NY) using a Quantum4-CCD detector (Area Detector Systems Corp., Poway, CA). The datasets were processed and scaled using HKL2000 [30]. The crystals belonged to a P212121 space group. There were two monomers in the asymmetric unit corresponding to Vm = 2.49 Å3 Da−1 with a solvent content of 50.6%. The position of the selenium atoms were determined using the program Phenix-Autosol [31]. The obtained phases were further improved by density modification using RESOLVE [31]. Over 50% of the backbone atoms of the model were built by RESOLVE. The remaining residues were manually built using Coot [32] and subsequently refined using Refmac [33]. Refinement was continued until the R-value converged to 0.22 (Rfree = 0.28) for reflections I>σ (I) to 2.7 Å resolution (Table 1). The model had good stereochemistry, with 99.3% residues within the allowed regions of the Ramachandran plot. Subsequently, the importance of the key residues at the dimeric interface was validated by structure-based in vitro studies, such as analytical ultracentrifugation and pull down assays, and in vivo plant virulence studies. Coordinates of HEPN domain of VBP have been deposited in the Protein Data Bank (http://www.pdb.org) under accession code 4NQF.
The oligomeric state of full-length VBP, HEPN and NTD domain of VBP and their mutants was investigated by monitoring the sedimentation properties of each protein in sedimentation velocity experiments. For these experiments, 400 µl of samples at 1 mg/ml in appropriate buffer were used, with the experiments carried out in duplicate. The sedimentation velocity profiles were collected by monitoring the absorbance at 280 nm. The samples were centrifuged at 40,000 rpm at 24°C in a Beckman Optima XL-I centrifuge fitted with a four-hole AN-60 rotor and double-sector aluminum centerpieces and equipped with absorbance optics. Eighty scans were collected and analyzed using the Sedfit program [34].
MBP-VirD2 bound to amylose resin (New England Biolabs, Ipswich, MA, USA) was incubated with purified 6His-VBP, with or without substitution at Asp173Asn/Lys184Asp/Asn186Asp. The beads were washed several times before resolving on 12.5% SDS-PAGE. For western blot analysis, the proteins were transferred to a PVDF membrane. 6His-VBP was detected by the addition of diluted anti-His antibody (Santa Cruz Biotechnology; Santa Cruz, CA, USA). The signal was detected using the Super Signal WestPico Chemiluminescent substrate (Pierce Biotechnology; Rockford, IL, USA) under the conditions recommended by the manufacturer. For pull down assay from A. tumefaciens crude extracts, E. coli strain BL21 was used to produce 6His-VBP/substituted 6His-VBP as described above.6His-VBP/substituted 6His-VBP bound to Ni-NTA metal affinity resin was incubated with freshly prepared A. tumefaciens (vbp KO strain) crude extracts (2.5 mM MgCl2, 50 mM NaCl, 2 mM PMSF, 50 mM Tris-HCl, pH 7.4, 0.5% Triton X-100). After incubation at 4°C for 1 h, the resin was washed four times. The bound proteins were eluted with 250 mM imidazole. The eluted proteins were resolved on 12.5% SDS-PAGE. For western blot analysis, the proteins were transferred to a PVDF membrane. VirD2 and VirD4 CP in the eluent were detected by western blot using anti-VirD2 (1∶5000) and anti-VirD4 (1∶4000) antibodies. The signal was detected using the Super Signal WestPico Chemiluminescent substrate (Pierce Biotechnology; Rockford, IL, USA) under the conditions recommended by the manufacturer.
6His-VBP, with or without substitution of key residues, were purified in gel filtration buffer containing 30 mM Tris-HCl, pH 8.0 and 200 mM NaCl. ITC experiments were carried out using a VP-ITC calorimeter (MicroCal, LLC, Northampton, MA, USA) at 25°C using 0.01 mM VBP protein in the sample cell and 0.25 mM AMP-PNP in the injecting syringe. All samples were thoroughly degassed and then centrifuged to remove precipitates. With the exception of the first injection, 10 µl volumes per injection were used for different experiments. Consecutive injections were separated by 5 min to allow the peak to return to baseline levels. ITC data were analyzed with a single-site binding model using Origin 7.0 (Origin Lab Corp., Northampton, MA, USA) software.
Far UV spectra (260–190 nm) of VBP, HEPN, NTD domains and their mutants were measured using a Jasco J-810 spectropolarimeter (Jasco Europe, MI, Italy) in phosphate buffer (pH 7.5) at room temperature using a 0.1-cm path length-stoppered cuvette. Six scans were recorded, averaged and then baseline-subtracted.
A. tumefaciens strains were grown in MG/L liquid medium overnight at 28°C supplemented with appropriate antibiotics. The bacterial cells were collected by centrifugation and re-suspended in a buffer solution consisting of 10 mM MgCl2 and 10 mM MES, pH 5.5. Cell concentrations were adjusted to OD600 = 0.1. The leaves of Kalanchoe plants were wounded with a hypodermic needle and 5 µl of bacterial cell suspension was inoculated onto each wound area. The tumors were photographed at different time points after inoculation.
|
10.1371/journal.pntd.0001493 | Cognitive Changes and Quality of Life in Neurocysticercosis: A Longitudinal Study | Few studies have focused on the cognitive morbidity of neurocysticercosis (NCC), one of the most common parasitic infections of the central nervous system. We longitudinally assessed the cognitive status and quality of life (QoL) of patients with incident symptomatic NCC cases and matched controls.
The setting of the study was the Sabogal Hospital and Cysticercosis Unit, Department of Transmissible Diseases, National Institute of Neurological Sciences, Lima, Peru. The design was a longitudinal study of new onset NCC cases and controls. Participants included a total of 14 patients with recently diagnosed NCC along with 14 healthy neighborhood controls and 7 recently diagnosed epilepsy controls. A standardized neuropsychological battery was performed at baseline and at 6 months on NCC cases and controls. A brain MRI was performed in patients with NCC at baseline and 6 months. Neuropsychological results were compared between NCC cases and controls at both time points. At baseline, patients with NCC had lower scores on attention tasks (p<0.04) compared with epilepsy controls but no significant differences compared to healthy controls. Six months after receiving anti-parasitic treatment, the NCC group significantly improved on tasks involving psychomotor speed (p<0.02). QoL at baseline suggested impaired mental function and social function in both the NCC and epilepsy group compared with healthy controls. QoL gains in social function (p = 0.006) were noted at 6 months in patients with NCC.
Newly diagnosed patients with NCC in this sample had mild cognitive deficits and more marked decreases in quality of life at baseline compared with controls. Improvements were found in both cognitive status and quality of life in patients with NCC after treatment.
| Neurocysticercosis (NCC) is one of the most common parasitic infections of the central nervous system. Cognitive changes have been frequently reported with this disease but have not been well studied. Our study team recruited a group of new onset NCC cases and a matched set of healthy neighborhood controls and new onset epilepsy controls in Lima, Peru for this study. A neuropsychological battery was administered at baseline and at 6 months to all groups. Brain MRI studies were also obtained on NCC cases at baseline and at 6 months. Newly diagnosed patients with NCC had mild cognitive deficits and more marked decreases in quality of life at baseline compared with controls. Improvements were found in both cognitive status and quality of life in patients with NCC after treatment. This study is the first to assess cognitive status and quality of life longitudinally in patients with NCC and provides new data on an important clinical morbidity outcome.
| Neurocysticercosis (NCC) is caused by an infection of the human central nervous system (CNS) by the larval stage of the pork tapeworm Taenia solium. NCC is one of the most common parasitic infections of the human CNS and has become an increasingly important emerging infection in the United States [1]. It is the most common cause of symptomatic epilepsy worldwide [2], [3].
NCC is a dynamic, pleomorphic disease due to the variety of locations, numbers, and stages of lesions within the individual host [4]. The disease may mimic many neurologic syndromes, and a uniform presentation has not been described. Seizures are the most commonly reported symptom at presentation, occurring in up to 50–80% of patients [5], [6]. Headaches and focal neurologic deficits are also common.
Patients with NCC often display cognitive impairment. Mild to moderate cognitive dysfunction has been reported in up to 88% of NCC patients [7]. Dementia, defined through bedside testing, has been found in 6 to 15% of NCC patients in large clinical series [5], [8]. Levav, et al was the first group to use a standardized neuropsychological battery in assessing the cognitive status of patients with NCC [9]. They found significant deficits in motor control and impulsivity in NCC cases compared to controls. More recently, Ciampi De Andrade, et al used a comprehensive neuropsychological battery to evaluate a group of patients with NCC and matched healthy and epilepsy controls [10]. The group found impairment in multiple cognitive domains in patients with NCC prior to treatment as compared to both groups of controls. The use of antiepileptic drugs and seizures did not account for the cognitive abnormalities.
The studies that have utilized standardized cognitive batteries in NCC have been cross-sectional in nature. Because NCC is a heterogeneous disorder with great differences in disease duration, clinical features, and disability, it is important to control for as many of these factors as possible. Several unanswered questions remain about the temporal nature of cognitive dysfunction in the disease. For example, how do cognitive changes evolve over time? Are there risk factors that influence the course? Finally, what is the impact of NCC on the quality of life of patients with this infection?
We used a comprehensive standardized battery to assess the cognitive status of patients with newly diagnosed NCC at baseline and at six months. A group of demographically matched healthy neighborhood controls and epilepsy controls were recruited for comparison. Similar to NCC cases, the epilepsy controls had new onset seizures to adjust for the effects of antiepileptics on cognitive performance.
This study was approved by the Institutional Review Board (IRB) at the of the Universidad Peruana Cayetano Heredia, the Instituto Nacional de Ciencias Neurologicas, and the Hospital Alberto Sabogal all in Lima, Peru, as well as the IRBs of Georgetown University and the VA Medical Center, both in Washington, DC. All subjects enrolled in the study provided written informed consent.
A total of 14 patients with NCC were recruited at the Sabogal Hospital and Cysticercosis Unit, Department of Transmissible Diseases, National Institute of Neurological Sciences in Lima, Peru. To make the baseline time point consistent across cases, patients with NCC were recruited within six months of a new-onset seizure associated with parenchymal NCC. These incident symptomatic cases had a positive serologic enzyme-linked immunotransfer blot (EITB) assay and met standard diagnostic criteria for NCC [11]. Seizures must have been controlled with standard antiepileptic drugs (AED) for inclusion. Exclusion criteria for cases included central nervous system disease not related to NCC, active alcohol or drug abuse, primary psychiatric disease prior to developing NCC, disorders of vision that would preclude cognitive testing and pregnancy.
For comparison, 14 healthy neighborhood controls were recruited along with the NCC cases. Healthy controls were recruited from the neighborhood of the patients with NCC and were matched to cases on age (+/−10 years) and education (+/−5 years). Controls were EITB negative, had no chronic health conditions and had similar exclusion criteria as cases.
Epilepsy controls were also recruited at the Sabogal Hospital after new onset seizures (N = 7). Epilepsy was defined according to revised criteria from the International League against Epilepsy [12]. All seven epilepsy controls had a brain CT prior to enrollment which was normal. Seizures were both generalized and focal and their epilepsy was classified as unknown etiology. Seizures were controlled in all subjects with standard AEDs. All epilepsy controls were EITB negative, and met exclusion criteria as per NCC cases.
All patients participating in the study had a screening history and physical at baseline. Patients with NCC were treated with a standard 10-day course of albendazole [13] (15 mg/kg/day divided bid) and started on an AED regime that was consistent throughout the 6 month follow-up period. Similarly, patients with new onset seizures were started on a standard AED regime during the study period. Neuropsychological testing was performed at a mean interval of 8 weeks from starting AED medications for new onset epilepsy cases and a mean interval of 11 weeks after completing antihelminthic medication and initiating AEDs for new onset NCC cases.
A brain MRI was performed at baseline and month 6 for all NCC patients at a private radiological center on a 1.5-Tesla unit, as in previous trials [13], [14]. The protocol included T-1 weighted axial with and without intravenous gadolinium (0.1 mmol/kg), and T2-weighted and FLAIR axial studies. A radiologist experienced with NCC cases read each imaging study to determine the number, stage and location of CNS parenchymal cysticerci. Cure rate was determined by the number of patients who had resolution of cysts by brain MRI at 6 month follow-up.
The neuropsychology testing battery for the current study, administered at baseline and 6 month follow-up to all patients, was chosen to tap into the various domains of cognitive functioning thought to be affected by NCC. It is sensitive to lapses in attention, decreased visual and motor processing, and problems with working memory. It was developed in consultation with Dr. Mirsky based on his experience of testing NCC patients in Ecuador [9]. In addition, a standard quality of life (QoL) instrument, the SF-36, was included in the battery. The tests in the battery are listed and referenced under their cognitive domain in Table 1. All tests were administered by a single native Spanish speaking psychology technician using standard Spanish translations of each test. The instructions for tests were given to patients in Spanish using a standard script.
Group means for individual neuropsychological tests at each visit (baseline, and 6 months) were calculated for NCC cases and controls. SF-36 transformed scale scores were calculated from the raw scale scores. A p-value≤0.05 was considered significant. Differences between groups were assessed by the χ2 test for categorical variables and the t-test for continuous variables. Because there are no national normative data for Peru, results for the NCC cases were assessed in the context of healthy controls, and epilepsy controls.
Due to the number of individual neuropsychological tests administered, a composite score was computed for the assessed cognitive domains. The tests comprising the composite scores for the major cognitive domains of attention, processing speed, learning and memory are defined in Table 1. Additional tests covering the following neuropsychological areas are also included in the battery: a) vigilance/sustained attention; b) general cognitive function; c) affect; and d) quality of life. These tests were not part of the composite scores.
The composite score is made up of the individual tests within the specified domain and is highly correlated to the individual test scores. Using composite scores focuses the analysis on specific cognitive domains and reduces the likelihood of making Type 1 errors by minimizing the number of variables in the analysis. Composite scores also attenuate ceiling and floor effects of individual test scores and are helpful in finding longitudinal changes when there is heterogeneity in cognitive function [15], [16]. In each case, raw scores were converted into z-scores based on US population-based normative data for each standardized test. The z-scores were averaged within each domain yielding a standardized measure of a participant's (both cases and controls) average performance on key measures within each cognitive domain. For example, the memory composite score was calculated by adding the individual z-scores of the Wechsler Memory Scale -III Logical Memory II and the Rey Auditory Verbal Learning Test delayed recall and dividing by 2, the total number of scores in the index.
We used both one-way and multiple ANOVA to examine how NCC cases performed on various measures of neuropsychological function and QoL as compared with controls at each testing time point. We also compared changes in neuropsychological testing scores longitudinally within the same group using t-tests. For multiple regression, demographic and variables were entered as covariates in the standard linear models. Adjustments were not made for multiple comparisons. Within the four composite cognitive domains, we considered a z-score below 1.5 standard deviations to be impaired. We compared the proportion of subjects that were impaired between groups at the baseline assessment. SPSS version 14 (Chicago, IL) was used for the statistical analysis.
There were no statistically significant differences in the major demographic variables (Table 2) among the NCC cases, neighborhood controls and epilepsy cases in this study. There were no clinical seizures during the study follow-up period in either the NCC cases or epilepsy controls. Patients with NCC and epilepsy remained on the initial antiepileptic therapy they were prescribed during the 6-month study follow-up period. There were no cases of hydrocephalus or increased intracranial pressure in the NCC group per neuroimaging at any time point in the study.
Comparisons were made among the groups (NCC, neighborhood healthy controls, and epilepsy controls) at baseline using multiple ANOVA (Table 3) with scores adjusted for age and education. At baseline, patients with NCC had lower Attention Composite Scores than the epilepsy controls (p<0.04). For other composite cognitive domains, patients with NCC generally scored lower than controls. However, there were no significant differences in baseline composite scores for learning, memory, processing speed or other neuropsychological tests when comparing NCC cases with epilepsy controls or NCC cases with neighborhood controls (all p≥0.13).
The proportion of cognitively impaired patients with NCC at baseline within the composite domains of attention, processing speed, learning and memory was 57%, 57%, 50% and 36% respectively. Similarly, neighborhood controls had 43%, 43%, 29% and 29% of subjects impaired in the same respective cognitive domains. Epilepsy controls had 14% impairment in each of the composite domains. Overall, there was a greater trend for patients with NCC to be impaired in all four cognitive composites when compared with neighborhood controls and epilepsy controls.
QoL summary scores at baseline revealed a significant decrement in social function (p = 0.002) and borderline decreases in physical function (p = 0.067) and mental function (p = 0.072) in the NCC group compared with neighborhood controls (Table 4). Epilepsy controls had a significant QoL decrease in mental health function (p = 0.028) and borderline low social function (p = 0.065) compared with neighborhood controls. The number of seizures at onset were not significantly correlated with any baseline QoL summary score.
QoL scores at the six month follow-up visit showed an overall trend for improvement across physical health, mental health and social function domains in both NCC and epilepsy groups (Table 4). There were no significant or borderline differences cross-sectionally between any group with all p values>0.1.
All groups were examined for cognitive changes over time (i.e., between baseline and 6-month follow-up testing). Within the cognitive battery of tests, NCC cases had significant improvement in psychomotor speed and working memory as assessed by the WAIS-III Digit Symbol Coding Subtest (p = 0.012) and by the Trail Making Test Part B (p = 0.016) between the baseline and 6-month follow-up sessions but there were no significant changes in any composite score (Table 5).
Impressive improvement in QoL for NCC cases was noted in the SF-36 domain assessing social function (p = 0.0062) over the 6-month follow-up period (Table 4). Despite improved scores for NCC and epilepsy groups between baseline and follow-up visits, there were no other significant or borderline longitudinal changes in the remaining QoL domains for any of the 3 patient groups (p values>0.4).
Among the demographic and clinical variables collected in the study, there were no significant longitudinal predictors of cognitive dysfunction or QoL in any group. Variables utilized in bivariate and multivariate statistical models included age, sex, education, occupation, number of CNS cysts, type of cysts, and cure rate of cysts.
This study demonstrated that patients with parenchymal NCC had relatively mild cognitive dysfunction shortly after presentation of seizures compared with two sets of control groups. Moreover, quality of life was significantly impaired in the NCC and epilepsy groups compared with healthy controls. Many reports of altered cognition in NCC have been related to hydrocephalus and intracranial hypertension which is not the case in this series.
After six months of follow-up, cognitive tasks involving attention/psychomotor speed and working memory improved in patients with NCC as did QoL. There were no significant demographic or clinical predictors of cognitive dysfunction or QoL in NCC cases or controls. This is the first study to longitudinally assess cognitive function and QoL in patients with NCC.
There are few publications that have specifically addressed cognitive changes in NCC. Most of the information on this topic comes from case series and anecdotal reports of NCC patients who experience psychiatric disturbances or mental status changes. How the cysticercotic CNS lesions exert their effects on cognition is unclear, but at least some of the mental status changes described in clinical reports could be explained by partial seizures, mass effect of cysts and increased intracranial pressure. The inflammatory immune response is critical in NCC pathogenesis but is poorly understood. CNS inflammation has been correlated with the cognitive dysfunction in a number of neurological conditions including multiple sclerosis and Alzheimer's dementia [17], [18]. Finally, the effects of AEDs on cognition in patients with epilepsy has been well studied [19]. Controlling for these effects with an epilepsy control group is prudent.
Longitudinal improvement trends in QoL were seen in patients with both NCC and epilepsy in this study. The mechanisms for decrements in QoL may be different than those producing cognitive dysfunction. Reduction in seizures which was seen in both the epilepsy and NCC groups over the follow-up period may be playing a role in improved QoL outcomes. We did not see a significant correlation between the number of seizure and QoL at baseline, however.
The earliest reports of mental status changes with NCC come from case series in the early 20th century, where both clinical and pathologic aspects are described [1], [2]. Other early reports emphasize the importance of altered mental status as a presenting symptom of NCC [5]. McArthur made some relevant insights into British soldiers with neurocysticercosis in the 1930s:
“As well as gross disturbance which suggests diagnosis such as these above, there may be mental dullness, impairment of memory, temporary periods of disorientation, or a change in disposition so that a previously efficient soldier may become careless and untrustworthy. Indeed, Colonel Benson, commanding the military hospital, Millbank, has told me with some feeling that if any breach of ward discipline is reported, usually a cysticercosis patient proves to be the delinquent [20].”
Roselli et al. described a severe case of dementia associated with parenchymal NCC in a 15-year-old girl [21]. Treatment with praziquantal decreased intellectual deterioration but may have precipitated hydrocephalus. Latovski et al. reported five patients with NCC who presented to a New York hospital with signs of dementia, seizures and increased intracranial pressure [22]. All improved with surgical and/or medical therapy. Larger case series have reported the prevalence of dementia between 6% [3] and 16% [4]. Psychotic behavior has been the presenting symptom in individual NCC cases [5] and in 2% of 112 NCC cases from Houston, TX [23].
Forlenza et al. reported on the psychiatric and cognitive manifestations of NCC from a case series from Brazil [7]. All 38 NCC cases were examined cross-sectionally. The authors used the mini-mental state exam, the present state exam, and Strub and Black's mental status examination to assess cognitive function. Psychiatric disease was found in 66% of cases and cognitive decline in 88%. Mild to moderate cognitive deficits were more common than frank dementia. Interestingly, the number, type of brain lesions, use of corticosteroids, or epilepsy did not significantly correlate with the severity of psychiatric symptoms. The high rate of cognitive impairment was striking considering the relative insensitivity of the tests used.
The first study to use standardized cognitive testing in NCC patients and controls was published in 1995 by Levav et al [8]. The authors recruited a representative community sample of patients from a larger epidemiologic study on epilepsy and NCC in Ecuador. A total of 123 subjects agreed to participate in the study. A series of 11 standardized NP tests were administered to 20 patients with NCC and controls to measure five dimensions of neurocognitive function. The testing revealed that patients with NCC had significant deficits in attention and motor control compared with controls.
The most recent report on cognitive changes and NCC came from Ciampi De Andrade et al [10]. This Brazilian group recruited 40 treatment-naïve NCC cases, 49 healthy controls and 28 patients with cryptogenic epilepsy. The assessment of neuropsychological function was cross-sectional with NCC cases exhibiting significant impairment in executive function, memory, constructive praxis and verbal fluency compared with healthy controls and epilepsy controls. Similar to our study, there were no significant predictors of cognitive dysfunction based on clinical variables. In contrast to Ciampi De Andrade, et al, we did not find the degree of cognitive dysfunction in our NCC cases. However, the healthy controls in the study by Ciampi De Andrade, et al were made up in part of staff at the hospital and the epilepsy controls were established cases (vs. new cases) at the authors' institution. The controls used could have produced some bias in the cognitive testing results. For example, patients with established epilepsy may have accommodated to any untoward effects of anti-epileptic medications through modification of the dose or agent as compared to new onset epilepsy cases. We recruited healthy controls exclusively from the same neighborhood as cases and all epilepsy controls had new onset seizures of unknown etiology. These control groups mirrored the experience of NCC cases in our population.
To our knowledge, there have been no longitudinal QoL studies in patients with NCC. A recent cross-sectional study of QoL in patients with NCC from Mexico reported deficits in both mental and physical domains compared with controls [24]. Other neurological conditions that have demonstrated impairments in quality of life in Latin America include head trauma [25] and spinal cord injury [26].
There are limitations to our study. Our sample size was relatively small. This was due in part to the type of patients we recruited, new onset patients with NCC with seizures, and the longitudinal nature of the project. Second, we had relatively mild disease burden of parenchymal NCC with low number of cysts and few neurology symptoms outside of seizures. This is fairly typical of patients with NCC in endemic regions and the US. A more broad morbidity spectrum of NCC cases would likely provide a wider range of cognitive dysfunction. Finally, it would have been useful to have Peruvian neuropsychological test norms for comparison in this study. Unfortunately, these data are lacking for Peru and many Latin American countries. US norms were available for the majority of the cognitive tests and served as a reference for composite scores.
In summary, our study showed patients with parenchymal NCC have mild cognitive dysfunction along with more significant deficits in QoL compared with healthy neighborhood controls and epilepsy controls. Altered mental status and psychiatric presentations in NCC are the tip of the iceberg while more subtle cognitive dysfunction is quite common. Importantly, in our series, the cognitive and QoL deficits improve with time. Future epidemiologic and clinical studies of NCC should include cognitive function and QoL as outcome variables.
|
10.1371/journal.pcbi.1000296 | Effects of Ploidy and Recombination on Evolution of Robustness in a
Model of the Segment Polarity Network | Many genetic networks are astonishingly robust to quantitative variation,
allowing these networks to continue functioning in the face of mutation and
environmental perturbation. However, the evolution of such robustness remains
poorly understood for real genetic networks. Here we explore whether and how
ploidy and recombination affect the evolution of robustness in a detailed
computational model of the segment polarity network. We introduce a novel
computational method that predicts the quantitative values of biochemical
parameters from bit sequences representing genotype, allowing our model to
bridge genotype to phenotype. Using this, we simulate 2,000 generations of
evolution in a population of individuals under stabilizing and truncation
selection, selecting for individuals that could sharpen the initial pattern of
engrailed and wingless expression. Robustness was measured by simulating a
mutation in the network and measuring the effect on the engrailed and wingless
patterns; higher robustness corresponded to insensitivity of this pattern to
perturbation. We compared robustness in diploid and haploid populations, with
either asexual or sexual reproduction. In all cases, robustness increased, and
the greatest increase was in diploid sexual populations; diploidy and sex
synergized to evolve greater robustness than either acting alone. Diploidy
conferred increased robustness by allowing most deleterious mutations to be
rescued by a working allele. Sex (recombination) conferred a robustness
advantage through “survival of the compatible”: those
alleles that can work with a wide variety of genetically diverse partners
persist, and this selects for robust alleles.
| Most so-called “higher organisms” are diploid (have two
copies of each gene) and reproduce sexually. Diploidy may be advantageous if one
functional copy can mask the effects of a mutation in the other copy; however,
it is a liability if most mutations are dominant. Sex can increase genetic
diversity and the rate of evolution by creating new combinations of alleles that
might function better together but can also disrupt working combinations. Given
these trade-offs, why are sex and diploidy so common, and why do they occur so
often together? We hypothesize that sex and diploidy allow gene networks to
evolve to function more robustly in the face of genetic and environmental
variation. This robustness would be advantageous because organisms are exposed
to constantly changing environments and all genes undergo mutation. To test this
hypothesis, we simulated evolution in a model of the segment polarity network, a
well-studied group of genes essential for proper development in many organisms.
We compared the robustness of haploid and diploid populations that reproduced
either sexually or asexually. Sexually reproducing diploid populations evolved
the greatest robustness, suggesting an explanation for the selective advantage
of diploid sexual reproduction.
| Phenotypic robustness, also called canalization [1], is the ability of a
phenotype to persist when challenged by a perturbation to the system producing it.
Many phenotypes are not the product of an individual gene, but rather arise from
interactions within larger gene networks. The functions of several well-studied
networks have been shown or predicted to be robust to quantitative variation in the
biochemical kinetics [2]–[8]. This variation can
come from both intrinsic (genetic) and extrinsic (environmental) sources: Genetic
diversity (polymorphism) within populations can produce variation in gene expression
levels and in the activity of gene products [9]–[13]. In a
genetically diverse, sexually reproducing population, recombination is continuously
producing new combinations of alleles, and robustness to genetic variation would
confer a fitness advantage. This intuition is supported by experiments showing much
genetic variation is hidden—i.e. quantitative variation between
individuals has no detectable effect on phenotype [10]. Another source of
perturbation is environmental: Individuals can transiently experience a broad range
of potentially noxious environments (due to pH, oxygen level, starvation conditions,
or temperature) that alter protein activity and potentially disrupt gene networks.
While only genetic effects are heritable, genetic and environmental variation both
perturb network dynamics, and robustness to one may confer robustness to the other
[14]–[16].
A possible mechanism to increase robustness is diploidy, as mutations can be masked
by a functional copy (a recessive mutation), allowing greater tolerance to mutation.
However, it is unclear whether diploidy is an advantage in genetic networks, because
it is also potentially harmful: a diploid network will have mutations twice as often
as a haploid, and a single bad allele could break the network (a dominant mutation).
Most deleterious mutations in enzyme coding genes are recessive to the wild type
alleles [17]–[19]. For metabolic
networks, Kacser and Burns [20] showed theoretically that most mutations are
recessive because in long metabolic pathways each individual enzyme contributes
weakly to the total flux. This theory was formulated for metabolic networks where
all gene products were enzymes, and it may not hold for gene regulatory networks
[21],[22]. Importantly, a majority of disease-causing
mutations in transcription factors are dominant [23]. Experimental
evolution on yeast, which can exist either as haploids or diploids, has shown that
different ploidies are advantageous under different conditions [24]–[26]. The
advantage of diploidy depends on the frequency of deleterious dominant mutations,
mutation rates, and other factors [27]–[30]. However,
this is an oversimplification because if most deleterious mutations are recessive,
the evolutionary advantage of diploidy remains questionable as the effects of rare
beneficial recessive mutations could likewise be masked. Such masking of beneficial
mutations in a diploid population has been observed in antibiotic resistance
evolution in yeast [25]. Thus, models investigating the effects of ploidy
on robustness need to incorporate both the spectrum of possible mutations, and the
functional context in which they occur (e.g. participation in a network).
Theory predicts that genetic variation combined with gene interaction favors the
evolution of phenotypic robustness [14],[31],[32]. The evolution of increased robustness to
mutation (mutational robustness) has been predicted by models of RNA folding [33],[34] and
randomly wired transcriptional networks [5],[35],[36]. Theory and modeling
predict that sexually reproducing populations, with recombination shuffling alleles,
should experience stronger selection for robustness than asexual populations [37], and
has been shown to hold for randomly-wired interaction networks [35]. However, it is
unknown whether these results hold for real networks because interactions between
mutations may be more complicated than theoretical studies assume. Additionally,
real networks may have subtle topological or regulatory architecture that differ
from randomly-wired model networks in important ways. Sex and diploidy are commonly
found together, and both may produce greater robustness, but this has not been
tested for gene regulatory networks.
In this study, we investigate how ploidy and sex (recombination) affect the evolution
of robustness in a detailed model of the segment polarity network. Previous modeling
studies focused on highly simplified and abstract networks [5], [33]–[36], and it
is essential to test whether these findings hold in a realistic network with a known
function. The segment polarity network is a canonical example of a pattern forming
network that is robust to variation in its underlying biochemical kinetics [2],[38]. It
is essential for development in many insects, and the function of its genes and
their interactions within the network are well-understood. In this network, gene
expression is regulated at both pre- and post-transcriptional levels, with some
regulations requiring cell–cell communication. During development prior to
the operation of the segment polarity network, gap and pair-rule genes activate
expression of wg and en in a noisy prepattern of stripes. The segment polarity
network in Drosophila development then sharpens and maintains these
stripes through the lifetime of the organism. Correct location of these stripes of
expression is essential for development, as they provide positional information to
activate downstream genes and processes in the proper locations. Previous work
showed that a haploid model reconstituting the known interactions within this
network can robustly reproduce the observed pattern of gene expression (i.e. the
phenotype) despite large changes in the model parameters representing the
biochemical kinetics [2],[38],[39].
To investigate the evolution of phenotypic robustness of the segment polarity
network, we developed a novel approach where model parameters were calculated from a
digital genotype, allowing our model to bridge genotype to phenotype (the pattern of
gene expression) in a way that can capture the quantitative and qualitative effects
of mutation and recombination. Mutations can alter the strength of interactions, and
all connections/processes in the network can vary and evolve in a simulated
population of organisms. Additionally, we built a diploid model of this network,
which allows 2 versions of each gene and all resulting gene products to have
potentially different kinetics. Using these, we explore how and whether a diploid
model is more robust compared to the haploid. We simulate a population of
individuals (organisms endowed with the network), with selection only to stabilize
the correct spatiotemporal pattern of expression (phenotype). Using this more
biologically detailed representation of the segment polarity gene network we
compared evolution of robustness in 4 different populations: sexual haploid, asexual
haploid, sexual diploid, and asexual diploid. We find that diploid sexual networks
evolve the greatest robustness increase and the combination of the two produces
greater robustness than either alone.
We took as a starting point a previous haploid model of the segment polarity network
[2],[38],[40]. This
model, shown in Figure 1A,
reconstitutes the core biological interactions as a set of ordinary differential
equations that govern the time evolution of mRNA and protein concentrations in a row
of 4 cells, starting from the prepattern of wg and en mRNA expression shown in Figure 1C. The spatiotemporal
pattern of expression depends on the biochemical parameters in the model. Thus, the
model is a bridge between a kinetic description of the network and the spatial
pattern of gene expression, the phenotype.
In the following paragraphs, we describe extensions to this model that allow us to
simulate evolution of the segment polarity network in response to selection on the
pattern of en and wg expression (the phenotype). We present a diploid version of the
model that allows us to directly compare evolution and robustness in haploid and
diploid models. We also use a novel framework of deriving model parameter values
from a digital genotype, which allows mutations to alter many gene properties (i.e.
changes in expression level, stability and activity). Using these, we start with
initially viable identical founders and follow them through 2,000 generations of
evolution as shown in Figure 2A.
We use the model to calculate phenotype (the en and wg pattern of expression) from
genotype, apply truncation or stabilizing selection on the phenotype, using a
multinomial sampling scheme to simulate random mating with a fixed population size
(N = 200) and a per-gene mutation rate (μ)
of 0.03.
Mathematically, our model of the haploid segment polarity network is the same as
described previously [2],[38] with 2
modifications: (1) The equations incorporate parameters for transcriptional and
translational synthesis rates (which were previously removed by
nondimensionalization). Including these parameters does not alter the dynamic
repertoire of the system, and allows mutations to alter the expression levels of
the mRNA & proteins. (2) Cells were 4-sided (changed from 6-sided) to
allow faster computation. This change did not alter the hit rate of successful
solutions in a random parameter search, nor did we notice a change in the
dynamical behavior of the system.
The segment polarity network was reconstituted into a system of ordinary
differential equations. The dependent variables in this system represent the
concentrations of the biomolecules in each cell (for cytoplasmic/nuclear
molecules) or membrane compartment (for membrane-bound molecules). The system
simulates a row (4 cells wide) of square cells with repeating (toroidal)
boundary conditions to represent a 2-D sheet of interacting cells. The
concentration of a membrane-bound protein can be different on each of the 4
sides of a cell (each side is treated as a separate compartment), and we
simulate diffusion by allowing molecules to transfer between cells and membrane
compartments where appropriate. The time rate of change for a given
concentration is simply the sum of the processes/mechanisms influencing it:(1)where is the concentration of molecule X in side
j of cell i.
Decay, binding, and translation follow standard mass-action kinetics
(1st or 2nd order). The detailed kinetics of enzymatic
activity and translational activation have not been measured in the segment
polarity network, so these processes are constructed from Hill functions as
described previously [2],[40]. Briefly, if protein A activates production
of molecule X, then:(2)where is the Hill function:(3)and where A is the concentration of the
activator, X is the concentration of target, K
is the concentration of A where activation is half maximal, and
ν is the cooperativity (Hill coefficient). This
parameterization is attractive because it can be tuned to capture a wide range
of activation curves with parameters that are commonly used in standard enzyme
kinetics and these parameters are, in principle, measurable. Additionally, this
function enforces expected qualitative behavior of biological processes:
saturation (biological processes tend to saturate above some level of
activation, after which further addition of activator ceases to have an effect)
and monotonicity.
The complete list of equations and parameters are listed in Protocol S1
and Tables
S1 and S2. All software was written in Mathematica version 5.2 (Wolfram
Research). The system of equations was integrated using Mathematica's
built-in NDSolve numerical differential equation solver. To guard against errors
in numerical integration, we tested a subset of the solutions generated by
Mathematica to that returned by Ingeneue [40],[41]. Ingeneue uses a
different numerical integration scheme than Mathematica, and shares no code, and
we found no difference between the solutions returned by the two programs.
The model shown in Figure 1A
is a haploid network, with a single form of each gene. We constructed a diploid
model of the network with 2 versions of each gene and gene product. Figure 1B shows the diploid
network for only the ptc and cid genes; both
the number of distinct biomolecules (boxed items) and the number of interactions
(lines) can increase by a factor of 2 or more. In the diploid model, there are 2
distinct versions of each mRNA and protein. However, for complexes, such as the
Patched-Hedgehog dimer, there are 4 possible distinct dimers (4 ways to combine
the 2 HH and 2 PTC proteins).
In the diploid network, all molecules maintain the same activities as in the
haploid, but the presence of two alleles must be correctly implemented to follow
the established biology of diploidy. Fluxes/conversions (solid lines in the
network diagram) are doublings of the haploid version: translation, decay, exo
& endocyctosis, and diffusion. For example, each protein is translated
only from the corresponding mRNA; i.e. CID1 protein is translated only from cid1
mRNA, while CID2 protein is translated from cid2 mRNA (and we assume it is
independent of cid1 translation). Similarly, the two versions of each
biomolecule decay independently with 1st order kinetics (we assume
the decay of one allele does not affect the rate of decay of its homologue).
Regulatory interactions (dotted lines in the network) become more complex in the
diploid network, as we must account for the combined regulatory activity of both
alleles. In the example in Figure
1B, each of the two CID proteins can have a potentially different
effect on the activity of each of the ptc target genes, so the number of arrows
(regulatory interactions) has quadrupled compared to the haploid case.
For diploid networks, we construct an extension of the Hill function to allow for
two activators controlling expression of a target. Here, we extend the example
of Equation 2 for two activators (A1 and A2) that can have different efficacies
in activating two targets (X1 and X2):(4)where X1 and X2 are the
concentrations of the two alleles of target gene X, and A1 and
A2 are the concentrations of the two alleles of activator
A, and is an extension of (described below).
KA1X1 describes
how efficiently A1 influences X1 synthesis
(i.e. how well transcription factor A1 activates the production synthesis of X1
by binding productively to the X1 enhancer sequence),
KA2X1 describes how efficiently
A2 influences X1 synthesis(i.e. how well
transcription factor A2 activates the production synthesis of X1 by binding
productively to the X1 enhancer sequence), etc. We assume that A1 and A2
proteins do not interact with each other in activating X (i.e. we do not
consider that A1 might block activity of A2 by nonproductively binding to the
enhancer sites on X1) and the net activity of A1 and A2 is simply their average
activity in binding to the affector for gene X. Furthermore, we assume the
cooperativity reflects the number of occupied binding sites on the target gene.
Substituting into the Hill equation yields:(5)The K parameters
(KA1X1,
KA2X1, etc.) in
the diploid model cannot strictly be interpreted as half maximal activities like
their haploid counterparts because activation depends on both
A1 and A2. Note that a completely homozygous
diploid is identical to the haploid; when the concentrations of both diploid
activators and activities are the same (when
A = A1 = A2
and
K = K1 = K2)
then . Figure 3
shows the behavior of Equations 3 and 5.
There are many ways to extend the Hill function (or implement alternative
formulations) to approximate the effects of diploidy, and increased realism
comes at additional computational complexity. A highly realistic model would
ideally track the bound state of each enhancer site for a gene (perhaps
including current availability of the site based on histone acetylation, etc.),
the affinity of each activator allele for each site, and the contribution of
each bound transcription factor to the initiation of transcription. We settled
upon the formulation in Equation 5 because it is simple (both to use and
understand) but captures attractive features of diploidy. Specifically, our
scheme: (1) Captures the same qualitative biological behavior as the Hill
function did in the haploid case: saturation and monotonicity. (2) Does not
dramatically increase the parameter count or complexity of the model. (3) Allows
for direct comparison between the haploid and diploid models. The homozygous
diploid model reduces to the haploid equivalent when
A1 = A2
and
K1 = K2.
This allows us to compare directly the evolution of diploid and haploid
networks.
Our formulation of Equation 5 has consequences for the behavior of heterozygous
diploid networks. Activation in the diploid model depends on the average
activity (concentration divided by K parameter) of the two
activators. Thus, the loss of either A1 or A2 can be compensated by a
sufficiently large increase in the concentration of the other (shown in Figure 3). In the case of a
homozygous diploid, if
A1 = A2
and both have the same activity (both have identical
K's), then the total activity is the same as the
haploid. Depending on the activities of the two activator proteins A1 and A2,
the loss of either could result in anything from an insignificant change (if
A1 and A2 were both far above their
respective K's) to a dramatic change (if
A1 and A2 were near their respective
K's).
We emphasize that the segment polarity network has highly nonlinear behavior
[2],[38], and the loss of
one allele in an otherwise homozygous individual will usually not result in a
simple halving of expression in the affected gene. There is substantial feedback
between different genes and different cells, and some perturbations can result
in a complete change in the pattern of expression, while others will produce
almost no change. Because there are multiple cells in the network that are
co-regulating each other, many genes must be expressed within a correct window
of expression (above one threshold but simultaneously below another) in each of
the cells. Additionally, when there is high cooperativity in Equations 3 and 5,
the resulting gene activity may be unchanged (if far from the threshold for
activity) or completely lost (if near threshold).
Our implementation of diploidy does not allow for the possibility of interactions
between the two activator alleles: for example that A1 and A2 compete for
binding sites in such a way that A1 fails to activate production of X and also
(dominantly) blocks the activity of A2 (by binding nonproductively to enhancer
sites). Similarly, we do not allow for overdominance effects such as A1 and A2
somehow synergizing so that their combination has greater activity than an
equivalent amount of either alone.
The models of the segment polarity network described above are insufficient to
predict the effects of mutations on phenotype because many parameters in the
model are not properties of individual gene products, but instead reflect
interactions between biomolecules. For example, many
parameters in our model determine how well a transcription factor activates or
inhibits its target's gene expression. In reality, the strength of such
regulatory interactions could be altered by mutating either the transcription
factor or enhancer sequence, resulting in different patterns of inheritance
depending which gene combination is passed on to the offspring. Additionally, a
single mutation can perturb multiple parameters in the model: a mutation in a
transcription factor will affect its ability to recognize both enhancer
sequences.
Biophysically, interactions in genetic networks rely on physical binding of
biomolecules in regions with complementary surface chemistry and topology. To
capture the qualitative behavior of such binding, we abstract genotype as a bit
sequence (digital genotype) comprised of 1's & 0's
that can be imagined as a surrogate for the physical surface of molecules that
participate in a binding interaction (i.e. an enzyme's active site or
the binding surface offered by an enhancer consensus sequence) as shown in Figure 2B. The
strength/kinetics of an interaction (represented by biochemical parameters in
our model) are determined by the degree of complementarity between two bit
sequences, weighted by bit position. Each bit in the sequence is weighted twice
that of its neighbor on the right to allow mutations that alter bit-sequences to
have graded effects from very small to large (the motivation for this choice is
further discussed in the section “simulating mutation”
below). The parameter is derived from the interacting bit sequences by simply
scaling the normalized bitwise XOR value of the bit sequences according to
either a linear(6)or logarithmic scaling:(7)bitSequenceA and bitSequenceB
are the numeric representations of the binary interacting sequences and
N is the length of the bit sequence, set to 20 for our
simulations. We used linear scaling (Equation 6) for K
parameters, and logarithmic scaling (Equation 7) for all others. Linear vs. log
scaling was used so that mutations usually resulted in a weak/nonexistent
interaction as described in the “simulating mutation”
section below. Cooperativities were restricted to integer values by rounding the
results of Equation 6 to the nearest integer in order to speed numerical
integration.
Several parameters reflect the interaction of the segment polarity genes with
genes products outside of the network. Table S1 lists the general categories of
parameters in the model, what they represent, and indicates whether the
parameter is derived from two different bit sequences (i.e. is an interaction
between 2 genes with the segment polarity network) or is derived from the
comparison of a bit sequence from a single gene with 0 (indicating interaction
with general cellular machinery that we assume is constant). For example,
maximal transcription and translation rates (C and
L parameters) are determined by how well the SPN genes interact
with the initiation machinery for these processes. Evolution of global cellular
behavior is slow, while transcription factors evolve quickly [42], therefore we did not allow global machinery to
change, and held the corresponding bit sequences fixed at 0 (i.e. all
0's in the bit sequence, this was chosen for convenience since again,
this sequence did not evolve). This allows the maximum translation rates of
genes to be changed and inherited as any other property, but does not allow, for
example, heritable ribosomal mutations that would globally alter all translation
rates. Thus, our model explicitly represents the genotype of 5 genes in the
haploid network (10 in the diploid).
For the special case of the lifetime of the PTC-HH protein
dimer(HPH), we reasoned that the stability of
the complex reflects a tripartite interaction involving both proteins with the
degradation machinery in the cell: (8)Where min and max are the range of allowed values (Table S2),
HPHptc is the bit sequence representing the
ability of the PTC part of the PTC HH dimer to interact with the degradation
machinery, and HPHhh represents the same for the HH
part of the dimer. The lifetime of the dimer is the average of the contribution
of the ability of HH to be recognized by the degradation machinery (bit sequence
fixed at 0) and the PTC part.
In the model, different parameters for each gene describe distinct
sub-activities/properties such as mRNA stability, protein stability, protein
activity, expression level, etc., as shown in Figure 2C. In reality, the DNA sequences
determining these different activities are usually spatially separated on the
gene: enhancer sites (affecting transcription rate) are on the non-coding region
usually away from the ribosomal recognition sequence (which affects translation
rate) and likewise distant from the coding region of the active site (which
affects protein activity). Thus, most point mutations alter only one or a few
properties of the gene products: for example a mutation in the coding sequence
for a real protein might alter the protein's activity and stability
[43], but not its transcription rate. Additionally,
mutations in a transcription factor can alter its interactions with a subset of
targets while leaving other interactions unaffected [44]. To capture this in
our model, we use a separate bit sequence for each of a gene's
parameters (sub-activities). For example, we use separate bit
sequences for the maximum transcription rate of a gene, the stability (mean
lifetime) of the mRNA, maximum translation rate into protein, stability of the
protein, and each of the protein's activities. Thus, though there are 5
genes (10 in the diploid model), there are far more bit sequences (∼71
in the haploid model, 142 in the diploid) than genes. From these bit sequences,
all model parameters (57 haploid, 140 diploid) are determined using Equations
6–8. The number of model parameters is more than half the number of
bit sequences because several parameters are derived by comparing bit sequences
describing properties of segment polarity genes with fixed cellular machinery
(fixed at a value of 0, and not included in the bit sequence count). Thus,
parameters that reflect interaction with cellular machinery are simply inherited
(though they can still be mutated).
Equations 6 & 7 capture important relationships between different
parameters in the model. In a diploid organism, consider a mutation in a
transcription factor that affects the surface of the transcription factor that
binds the enhancer. Such a mutation will alter the ability of the transcription
factor to recognize the enhancer sequences of both target alleles: in Equation
4, a mutation that alters the ability of transcription factor A1 to bind to
enhancer sites will alter both KA1X1 and
KA1X2. Conversely, a mutation in an enhancer
sequence will alter the ability of both transcription factor alleles to regulate
the mutated gene. In Equation 4, a mutation that alters the enhancer sequence of
X1 will alter the ability of both transcription factors, A1 and A2, to recognize
it and will alter both KA1X1 and
KA2X1. If bit sequence BA1
transcription represents the surface of A1 that binds to
enhancers, BA2 transcription represents the surface
of A2 that binds to enhancers, BX1 enhancer is the
surface presented by gene X1 recognizable by transcription factors, and
BX2 enhancer is the surface presented by
gene X2 to transcription factors, then we can calculate the relative strengths
of the two transcription factors to activate each target gene using Equation 6:(9a)(9b)(9c)(9d)All 4 KAX parameters share a common
range from to . A single mutated bit sequence can affect multiple parameters,
as expected from the underlying biology, and our model properly captures the
qualitative effects of cis and trans
mutations.
To reiterate, our scheme of calculating parameters from Equations 6–8
is attractive because: (1) It is conceptually consistent with the underlying
biophysical mechanism of binding. The binding surfaces/active sites are
specified either directly by the genotype (i.e. a regulatory consensus sequence)
or indirectly (the genotype specifies the 3-D shape of a protein), but the
ultimate origin of both is a mutable sequence (the DNA sequence of the gene).
(2) It allows us to compute how well a gene product can interact with any
partner, allowing us to easily simulate the effects of recombination (which will
produce new combinations of alleles that may not have worked together before)
and inheritance, as parameter values are interactions (not heritable) that
depend on the interacting genes. Our scheme allows us to calculate the strength
of an interaction when, for example, both a transcription factor and the
enhancer sequence it binds are mutated. (3) It allows us to simulate both
cis and trans mutations. Transcriptional
regulation can be altered by a mutation in either the transcription factor or
the enhancer, with different consequences depending on which is mutated. Our bit
sequence representation allows this aspect of biological reality to be captured.
(4) It allows us to capture the general qualitative features of mutations (see
next section). (5) It is computationally trivial.
To simulate evolution of the network, it was necessary to generate founder
genotypes that produced a viable phenotype (Figure 1C). To do this, we performed a random
search for viable haploid parameter sets, then converted them to genotypes. To
reduce the number of free parameters in the random parameter search, we
restricted the transcriptional and translational rates (C and
L parameters) to the inverse of the mRNA and protein
lifetimes (H parameters):(10)This is equivalent to the nondimensionalization scheme used
previously [2],[40]. This strategy was used for the en, wg, ptc,
and cid mRNA and proteins. However, because the HH protein will heterodimerize
with PTC protein on adjacent cells, we allowed for a stoichiometric
excess/scarcity of PTC and HH. In the random parameter search the
LHH parameter varied from to . This allows the maximal HH protein concentration to vary from
0.2 to 5 times that of PTC. The restriction in Equation 10 was not applied
during evolutionary simulation (i.e. synthesis and stability were independent).
Table S2
shows the range explored for each parameter in the random search for founders.
The constraints we impose in Equation 10 enforce that en and wg have a maximal
value of 1, and so all founders have similar patterns of wg and en: in cells
that should express them highly, wg and en are expressed between 0.8 and 1
during the relevant simulation time from 200–500 minutes. As shown in
Table
S2, during evolution we allow model parameters to explore a much larger
range for most parameters, so wg and en expression can rise above the founder
levels to a maximum of 20.
To generate the founder genotypes, we converted working parameter sets into the
corresponding genotypes by inverting Equations 6 & 7. The parameter
value uniquely defines only the XOR difference between pairs of bit sequences.
Thus, we chose a random value for one bit sequence (each bit position was
randomly set to 0 or 1 with equal probability), then assign a unique value to
the other using the inverse of Equations 6 and 7 above:(11)for linearly scaled parameters or(12)for logarithmically scaled parameters. In Equations 11 &
12, bitSequenceB has each position randomly set to 1 or 0,
allowing us to find a unique value for bitSequenceA. Here, min
and max represent the extremes of the allowed values for the parameter during
evolution according to Table S2. During evolution, we allow a much
wider range of parameter values than during the search for founders, as
mutations should often weaken (but rarely strengthen) an interaction.
We emphasize that the bit sequences described in the previous section are
abstract surrogates for the 3-D physical surfaces of molecules that participate
in an interaction. They do not attempt to represent base pairing between
complementary nucleotide sequences as all interactions in the segment polarity
network are protein–protein or DNA/RNA-protein interactions. There is
no general theory that allows us to calculate the strength of binding between an
arbitrary gene product and its partners, nor to predict the effect of a general
mutation on the strength of this binding. Quantitatively, a point mutation can
have varied effects on an interaction: a complete quenching of an interaction
(mutation of the nucleotide/amino acid that is essential for
binding/interaction), an almost imperceptible change (mutation at a site
peripheral to the key interaction that slightly perturbs the interaction
strength), a (less likely) strengthening of the interaction (a mutation that
slightly increases the affinity of binding). In general, most mutations lower
the expression and activity of gene products, though rare mutations may
strengthen them.
After each mating/division in our evolutionary simulation, we allow a
3% chance per gene of a mutation. In a mutated gene, we mutate a
randomly-chosen bit sequence, with a recursive 10% chance that an
additional bit randomly-chosen bit sequence in the same gene is mutated. Thus,
mutations typically change one bit sequence (90% probability per
mutation) or more than one (10% probability per mutation) in the
gene, allowing mutations to, for example, change both the activity and stability
of a gene product. When a bit sequence is mutated, we randomize each position of
the bit sequence to a 0 or 1 (mutations result in an independent random draw).
The effect of this is the corresponding parameter(s) is/are set to a value
between the min and max shown in Table S2. For example, the mean lifetime for
a gene product (H parameters) will have a log-distributed
random value between 1×10−6 and 100 after a
mutation
(mean = 1×10−2;
a factor of 500 lower than the most unstable founder). Thus, 75% of
mutations will result in near to complete elimination of a gene product (with a
mean lifetime less than 1; in the founders, mean lifetimes vary from
5–100). Only a fraction (∼12%) of mutations will
produce protein stabilities comparable to those of the founders. Similarly,
values for transcriptional regulation (half maximal concentrations or
K parameters) will have a random value uniformly distributed
between 0.001 and 100 (mean = 50, a factor of
100 higher than most founders and no gene product in any simulation evolved
expression high enough to activate a process with such a weak interaction). This
biases the parameter towards extremely high values (i.e. weak activity), with
>99% of mutations producing ineffective (or dramatically
lowered) transcriptional regulation. For the special case of cooperativities
(ν parameters), we restricted these values to a
narrow range (mutations produce integer cooperativities between 1 and 10), as
high cooperativities are computationally expensive. Thus mutations usually
produce a limited change in cooperativity, biased towards low cooperativity (log
scaled). For all other parameters (>80% of parameters),
mutations, on average, produce interactions 2+ orders of magnitude
weaker than the founders.
In our model, mutations usually result in very weak or absent interactions (i.e.
the corresponding parameter has a value so the interaction is silent). We have
not attempted to reproduce the real distribution of mutational effects. Our
model parameters abstract a wide variety of processes (RNA stabilities, protein
stabilities, transfer/diffusion rates, etc.). For many processes, the mutational
effects are not well known, and capturing the remaining known mutational spectra
would require a separate mutational scheme (or genotype→parameter
mapping function or both) for each class of parameter. Our goal was to allow
mutations to have graded effects that usually disrupt interactions but
occasionally strengthen them, and also allow us to calculate the strength of
interaction between arbitrary pairs of partners (that may not have co-existed
within the same individual before). Additional limitations of our mutation
scheme: (1) We do not allow the possibility of whole gene duplications or genes
to evolve novel interactions that are absent from Figure 1A (i.e. dimerization between en and
wg or PTC degrading en). (2) We do not attempt to capture the relative rates or
magnitudes of mutational effects: one could imagine that mutations may more
frequently alter a protein's mean lifetime than the per-molecule
maximal catalytic rate due to the differences in mutational target size.
Similarly, the magnitude of mutational effects may differ: individual amino
acids may contribute weakly to the overall protein stability while mutations in
the active site may dramatically alter catalytic rate.
In sexual populations, mating was random, with randomly chosen (with replacement)
pairs of parents producing a single offspring. Recombination proceeded as
follows: In diploid sexual populations, each parent would randomly pass on one
of its two alleles for each gene to the offspring (we did not include the
effects of genetic linkage in this study). In haploid sexual populations, the
haploid offspring produced by mating two haploid parents would randomly inherit
(with 50% chance) one of the two parents' alleles for each
gene. In both cases, all bit sequences corresponding to an inherited gene were
passed on together, and we did not allow recombination within genes. Division in
asexual populations was implemented by allowing a randomly chosen individual to
produce a clonal offspring that had the same genotype as the parent. In all
simulations (sexual and asexual), the genotype was subject to mutation as
described above, and individuals reproduced until the specified number of viable
offspring reached the population limit. Drift is present in our simulations, as
an unlucky individual may stochastically not mate/produce any offspring, and
individuals could mate with more than one partner in each generation.
We began by screening many randomly generated haploid genotypes to find 40
“founder” genotypes that sharpened the pattern of wg and en
mRNA expression as shown in Figure
1C. We simulated evolution for 2,000 generations starting each
simulation with a single founder. For each founder, we simulated 4 independent
parallel runs: sexual haploids, asexual haploids, sexual diploids, and asexual
diploids. Forty diploid founder genotypes were constructed from the haploid
founders by making them homozygous for the haploid alleles (again, diploids
homozygous for all genes produce the identical phenotype as the haploid). Each
generation in our model of evolution comprised the 5 phases shown in Figure 2A: Prediction of model
parameters from genotype, determining phenotype (spatiotemporal pattern of wg
and en expression), selection on phenotype, reproduction (either sexual or
asexual cloning), and mutation. Population size was fixed at
N = 200, giving 100 (diploids) or 200
(haploids) in each generation.
We used one of two selection criteria in our simulations: stabilizing or
truncation. Genomic data suggests that gene expression in
Drosophila is under stabilizing selection [45]–[47], or
selection for an unchanging pattern of expression. In our stabilizing selection
simulations, the founder phenotype is optimal (fitness ), with fitness falling as the en and wg patterns diverge from
the founder phenotype. Quantitatively, fitness under stabilizing selection is: (13)where d is the phenotypic distance between the
(optimal) founder and the evolved individual. For haploid individuals:(14)and for diploid individuals where there are 2 potentially
distinct en and wg alleles:(15)where eni,e and
wgi,e are the en and wg mRNA concentrations in
the ith cell position in an individual whose fitness
is being determined, eni,f and
wgi,f are the levels of en and wg expression of
the (optimal) founder, and horizontal lines indicate time averages of the
concentration from 200 to 500 minutes of development. When diploids are
homozygous for all alleles (producing identical expression of both en and wg
alleles), d reduces to the haploid case.
The developmental function of the segment polarity network is to stabilize
stripes of gene expression to pattern subsequent development. From the
perspective of this function, mutations that produce insufficiently sharpened wg
and en stripes are disastrous while those that result in an over sharpened
pattern are viable. To explore the consequences of this, we simulated truncation
selection where individuals are dead () if wg and en have expression levels outside of the expression
thresholds shown in Figure
1C, or take too long to stabilize their correct patterns. Otherwise,
individuals are viable with . In other words, as long as en and wg are sufficiently high in
the correct cells (and sufficiently low in the rest), the developmental
processes that depend on wg and en expression are unperturbed and the individual
will be viable. These two criteria approximate two biologically plausible
extremes, truncation selection penalizing insufficient sharpening of the pattern
but allowing the pattern to change, while stabilizing selection penalizes any
deviation from the founder pattern.
We tested robustness to 3 types of perturbations that the real segment polarity
network might be exposed to: (1) Perturbation of a single bit sequence in a
single randomly chosen gene. This usually caused a dramatic change in one or two
parameters, and is conceptually similar to a point mutation that produces a
specific effect. (2) Perturbation of all parameters. We multiplied each
parameter (after calculating it from genotype) by a randomly-chosen
(log-sampled) value from 0.66 to 1.5, independently (i.e. all parameters were
perturbed by a factor up to 1.5). Extreme environmental stress (pH change,
temperature change, starvation, etc.) could alter the cellular environment so
many parameters are substantially altered. (3) Perturbation of initial
conditions. We multiplied the initial amount of wg and en mRNA by a
randomly-chosen (log-sampled) value from 0.5 to 2, independently in each cell
(i.e. noise was added to the en and wg prepattern, but this never changed the
positions of the cells with the highest initial en and wg). A variety of sources
(developmental noise, mutations in genes responsible for the en and wg
prepattern, etc) could result in a perturbed prepattern.
We quantified robustness to these sources of variability and, for clarity, we
will use the term ‘survivorship’ when describing results
from truncation selection and ‘fitness’ for stabilizing
selection. Under truncation selection, we measured the fraction of trials where
the ability to sharpen the pre-pattern (according to the criteria in Figure 1C) continued in the
face of perturbation. Under stabilizing selection we measured the fitness
decrease (Equation 13) using the distance between the unperturbed and perturbed
wg and en expression levels analogously to Equations 14 and 15.
As described in Models, we generated 40 viable random haploid genotypes that
stabilized and sharpened the pre-pattern to produce the phenotype shown in Figure 1C. These genotypes
were not the product of evolution, but of randomly searching for genotypes
satisfying the above criteria. We then measured how robustly the phenotype
persisted in the face of perturbation (see Models), comparing the randomly
generated haploid genotypes with homozygous diploid genotypes (homozygous for
the haploid genotype for all genes). Figure 4 shows the robustness of the diploid
and haploid networks. Homozygous diploid networks were substantially more robust
to perturbations than their haploid equivalents: diploids had a higher chance to
maintain the wg and en sharpening and showed a smaller change in their en and wg
patterns compared to their haploid equivalents. The diploid robustness advantage
varied with the specific genotype we tested, but diploids had greater robustness
than haploids in >90% of the genotypes.
We next tested whether the greater robustness of diploid networks persisted when
we simulated evolution for 2,000 generations using the same 40 randomly
generated, viable genotypes as founders. In these simulations, we used a high
mutation rate (μ = 0.03) with small
population sizes (N = 200). We used each
genotype to generate a genetically identical founder population and simulate
evolution under either truncation or stabilizing selection with the following
conditions: haploid sexual, haploid asexual, diploid sexual and diploid asexual.
Thus, each founder was used in 8 parallel simulations. Our simulations allow us
to incorporate key features of diploidy: (1) Genotype is the product of
evolution, not from a random search of genotypes that happen to produce the
right pattern. (2) There is usually genetic diversity in a population [9],
[11]–[13], [48]–[50] and diploid
individuals can be heterozygous at loci. (3) Diploid individuals experience
twice as many mutations as haploids during evolution (assuming a constant
per-gene mutation rate).
Regardless of ploidy or reproduction mode (sexual or asexual), our evolutionary
simulations quickly produced a genetically diverse population, with several
quantitatively different alleles co-existing for most genes in any given
generation (expected since the expected number of mutations per gene per
generation μN = 6). Initially,
populations were genetically identical at all loci, but the founder allele
became extinct within a few hundred generations, after which there was a
diversity of several alleles present in the population, and diploid individuals
were heterozygous for most genes.
After simulations were complete, we measured the robustness at each generation to
3 types of perturbation; results are shown in Figure 5A–C for the average of all
40 simulations in each condition. Simulations under truncation and stabilizing
selection showed the same qualitative behavior. All populations evolved
increased robustness to the perturbations. Diploid populations continued to
exhibit increased robustness compared to haploid populations, especially when
combined with sexual reproduction. Comparing the terminal generations that share
a common founder, diploid sexual populations evolved the greatest robustness at
generation 2,000 in almost all (38/40 truncation; 39/40 stabilizing) tests of
robustness to point mutations, most (32/40 truncation; 31/40 stabilizing) tests
of robustness to all parameter perturbations, and a substantial fraction (19/40
truncation; 18/40 stabilizing) of tests of robustness to initial conditions.
While we cannot determine whether the robustness advantages of diploid sexual
populations persist forever (i.e. the asymptotic behavior), extrapolating from
data in Figure 5 suggests
that diploid sexual populations should maintain higher robustness than other
conditions far into the future.
The data shown in Figures 4A
and 5A suggest that most
mutations in the diploid network model are recessive: simulated point mutations
had a smaller effect in diploids than haploids. This is not built in; our
network allows for the possibility of dominant deleterious mutations. Examples
of possible dominant (and lethal) mutations that we observed during simulated
evolution: (1) A sufficiently hyperactive WG protein (which is initially
expressed at non-zero levels in all cells in our simulation) could disrupt the
normal gene expression pattern through excessive global wg autoactivation or
global en activation. (2) A mutation in the enhancer of cid that causes loss of
inhibition by en would result in overexpression of CID that disrupts the wg and
en patterns. In our simulations, mutations usually result in nonproductive
interactions, so mutation (1) is far less likely than mutation (2); the former
requires the (unlikely) mutation that produces strong autoactivation while the
latter requires a (more frequent) loss-of-function.
There are two mechanisms that may contribute to the increased robustness in
diploid populations: First, diploidy allows masking of a perturbed allele by its
homologue (i.e. most mutations are recessive). Second, diploid populations may
evolve increased robustness faster than their haploid counterparts through a
mechanism independent of dominance. To separate these, we measured robustness in
diploid populations by simulating symmetric mutations that perturb both versions
of an allele by the same amount, so there is no unperturbed homologue to mask
the perturbed allele. Symmetric point mutations altered the same bit sequence in
both alleles of the perturbed gene by the same amount (both homologous bit
sequences were altered by an XOR operation with the same random value). Figure 5D shows the results of
symmetrically perturbed diploid populations compared to their singly-perturbed
haploid counterparts. The robustness of the symmetrically perturbed diploid
populations was very close to the haploids, and changed only slightly over time,
indicating that the majority of mutations are recessive in our diploid model of
the segment polarity network.
The ability of the network to mask the effects of mutation may itself be evolving
(i.e. over time, the network evolves so that more mutations are recessive). Such
evolution would manifest in diploid populations as an increase in robustness to
(single) point mutation without an increase in robustness to symmetric point
mutations. Our data indicates this is the case for diploid sexual populations,
as the dramatic increase in robustness to point mutations over time is almost
eliminated under symmetric mutation. Diploid asexuals show a far smaller
increase, indicating that sex accelerates evolution of greater masking (i.e.
greater dominance of functional alleles).
Why does sex produce more robust populations? In our simulations, individuals
have reduced fitness/survivorship if they fail to sharpen the correct en and wg
patterns sufficiently. Fitness/survivorship can be reduced by two sources: a new
mutation or, in sexual populations, recombination of alleles that do not
function properly together. Figure
6 shows the relative effect of recombination and mutation on
survival. During the simulation, we recorded the number of dead individuals and
their genotypes, and whether they had a new mutation. Figure 6A shows the fraction of individuals
with a new mutation that were viable. This data is qualitatively consistent with
Figure 5A, but includes
mutations that could alter multiple genes and bit-sequences during evolution. To
determine how often recombination produced incompatible allele combinations, we
measured the fraction of deaths where individuals did not have a new mutation
(i.e. the fraction of the dead due to recombination). Figure 6B shows diploid sexual populations
showed a near doubling of this fraction compared to the haploid sexual
populations. Thus, diploid sexual populations experience a greater pressure to
maintain alleles that both produce the correct phenotype and that are also
highly compatible with the other alleles in the population. Recombination
constantly produces new allele combinations that cause quantitative variation;
thus sexual populations (especially diploid sexual populations) more strongly
select for genotypes (and alleles) that are robust to quantitative variation.
In Figure 6C, we plot the
fitness load for each of our simulations defined as:(16)where is the fitness of the most fit individual in the generation,
and is the mean fitness of all individuals in that generation.
Consistent with Figure 6B,
we see that recombination produces a higher fitness load in diploid populations
(the fitness load is noticeably higher in diploid sexual populations compared to
diploid asexuals), but not haploid populations (the fitness load of the two
haploid populations are nearly equal). Proulx and Phillips [14] showed the upper
bound for selection for mutational robustness is the fitness load minus the
mutation rate. All 4 populations have fitness loads higher than μ, with
diploid sexual populations having the greatest expected pressure to evolve (and
maintain) mutational robustness. Taken together, our data shows that the
combination of sex with diploidy synergize to produce the strongest selection
for mutational robustness.
Under truncation selection, individuals were dead if they failed to sharpen en
and wg sufficiently or if they did so too slowly. Populations under stabilizing
selection were penalized if the en or wg pattern was altered, but fitness was
independent of the time the prepattern was sharpened. To explore how aspects of
the phenotype and network function evolved, Figure 7 plots the time at which the pattern
was sharpened sufficiently and the average wg and en levels at the time
selection acted (200–500 min). Under both selection types, populations
evolved to sharpen wg more rapidly, with all populations showing similar
speeding. In contrast, sexual populations maintained the time to sharpening of
the en pattern, but asexual populations (particularly diploid asexual) showed
slowed sharpening. Thus, evolution did not exclusively favor faster sharpening.
Under truncation selection, the expression levels at which both wg and en
stabilized (in the different cells that should express those genes highly)
evolved to higher and higher values. Expression of wg showed more change
compared to en, and wg expression often decreased in cells that had to express
it at low levels. The highest possible en and wg level is 20 in our simulations,
and requires both maximal transcription and highly stable products (long mean
lifetimes). In general, diploid sexual populations show the greatest tendency to
move away from thresholds of failure (high expression in the appropriate cell,
and low elsewhere), while diploid asexual populations sometimes move towards
expression thresholds (higher expression in cells that should express low
levels, and slightly later en sharpening).
Moving away from thresholds of failure could confer increased robustness by
buffering the system to tolerate to small changes in expression. However, we
emphasize the segment polarity network has been shown to exhibit highly
nonlinear behavior, with successfully larger perturbations first producing
almost no change in the pattern of expression followed by an abrupt collapse of
the normal pattern [2]. Each of the 40 founders evolved slightly
different phenotypes and robustness, and Figure 8 shows the correlation between the
phenotype and robustness. Figure
8A plots mutational robustness against time to stabilization of the en
and wg patterns for both truncation and stabilizing selection after 2,000
generations. Faster stabilization of the pattern was associated, on average,
with only a modest increase in robustness. The average robustness of the diploid
and haploid founders is also plotted (large circles). The best-fit lines
indicate the correlation between evolved robustness and sharpening time;
intersection of this line with the mean founder behavior indicates the
robustness increase was due solely to changes in expression time. However, the
best-fit lines lie above the founders, indicating that the robustness evolved
through a mechanism independent of a faster time to sharpening. Similarly, Figure 8B correlates
mutational robustness with expression level in the highest-expressing cell for
truncation selection; there was little expression change under stabilizing
selection. We did not fit lines to the data, as such a fit would be dominated by
the outliers; most of the simulations showed little change in expression.
However, there is only weak correlation between expression level and robustness,
and the robustness that evolves is clearly not due solely to superthreshold
buffering. Thus, both stabilization and truncation selection evolves greater
robustness, particularly diploid sexual populations through mechanisms that do
not have profound changes in phenotype.
In the previous simulations, mutational robustness was expected to evolve due to
the high mutation rate. Theory predicts such robustness should evolve when there
is substantial genetic diversity, specifically when μN>1. Sex may
allow selection for robustness at lower mutation rates, and this has been shown
in randomly-wired transcriptional networks [35]. To test whether
this holds in our network, we ran simulations with
μ = 1/N = 0.005.
Figure 9 shows the
results of this simulation for 38 founders. We observed little robustness
evolution in haploid populations, with no significant increase in robustness by
generation 5,000. In contrast, diploid sexual populations evolved higher
mutational robustness, while asexual diploid populations showed a transient
decrease in robustness that stabilized by generation 1,000. As before, symmetric
double mutations eliminated the diploid robustness advantage, indicating the
diploid advantage was due to dominance of functional alleles. Again,
recombination resulted in a greater fitness penalty in diploid populations
compared to haploid (Figure
9C), and diploid sexual populations had the highest fitness load (Figure 9D). Thus, diploid
sexual populations still experience the strongest selection for robustness when
μN = 1.
We explored how ploidy and sex shape the evolution in a model of an actual,
well-characterized, developmental genetic network. The segment polarity network is
one of the best characterized networks, and comprises a functional module [2] that
is conserved across insects and beyond. Previous theoretical and modeling studies
have predicted mutational robustness can evolve, but we believe it is essential to
test these findings in as detailed a model as possible. Our model allows us to
bridge genotype to phenotype and to capture fundamentally important aspects of
allelic fitness which no previous model has represented. We found that diploidy and
sex combine to allow populations to evolve the greatest robustness to mutation,
global perturbations affecting all interactions, and initial conditions. Diploidy
confers an immediate robustness advantage as most deleterious mutations are
recessive in our network, and over time the network evolves so that functional
alleles become more dominant. Recombination, especially in diploid populations,
produced a greater fitness load that selected for greater robustness evolution even
at lower mutation rates of μ = 1/N.
Recombination in our network constantly shuffled alleles and prevented the
stabilization of matched allele combinations that could be maintained in asexual
populations. In sexual populations, the constant shuffling of alleles by
recombination in a genetically diverse population selects for those alleles that are
highly compatible with others—i.e. alleles that are robust to genetic
variation and mutation.
It is useful to compare our evolutionary model to that of previous computational
studies on robustness evolution. Wagner [5] simulates evolution in
randomly-wired haploid regulatory networks, which have been used in numerous studies
[35],[36],[51],[52]. The Wagner model
assumes a fixed time step, steep nonlinearities that result in effectively discrete
expression levels and additive regulatory effects. All parameters reflect the
strength of transcriptional activation/repression and mutation allows single
mutations to change an inhibitor to an activator with 50% probability. In
contrast to this, our model allows continuously variable expression levels with more
graded nonlinearities (the maximum Hill coefficient in our simulations was 10).
Previous models with fixed time steps [5],[36] reported a dramatic
increase in speed in generating the target pattern of gene expression, while we
observed only a slight speeding of wg (but not en) sharpening. In our model,
molecular half lives can be mutationally altered as they would be in real life,
which is difficult to translate to a fixed-time step model. It also permits
non-additive interactions between multiple transcriptional regulators. Our model
does not allow the sign of a regulatory interaction to change (inhibitors never
switch to activators), and allows mutations to be cis or
trans (the Wagner model parameters represent
cis effects only [5]), thus allowing a meaningful exploration of
diploidy. In our model, mutations are qualitatively similar to Wagner [5], as most
result in nonproductive/weak interactions.
Other simulations have attempted to capture more accurate quantitative effects of
mutation and other biological parameters. Robustness evolution has also been
explored in models of mRNA secondary structure prediction [33],[34]. These models allow the
detailed quantitative prediction of effects of mutation which we cannot do for our
model; additionally it is difficult to explore the effects of recombination and
diploidy in these models in a way that would meaningfully translate to genetic
regulatory networks. It would be possible to alter our model so that the bit
sequences represented mutable DNA/RNA sequences from which the interaction strength
is calculated. We did not explore this because we wanted a general scheme to capture
interactions within the network, and most model parameters reflect protein-protein
or protein-DNA interactions. If we replaced our binary bit sequences with sequences
of DNA bases (ATGC) or the 20 amino acids, there is no tractable function to
describe how well the two would interact, as such a calculation would require
prediction of the tertiary or quaternary structure. For the case of protein-DNA
interactions with known binding motifs, the effects of mutations can be approximated
[53],[54], and it would be an interesting extension to this
work to incorporate a similar approximation. However, there are many interactions in
addition to transcriptional regulation in our model, and such a scheme would not
allow us to model all parameters. One final limitation of computer simulation is
that we are limited by available computing power to relatively small populations and
high mutation rates. Real Drosophila effective population sizes and
mutation rates differ from our simulations by more than an order of magnitude.
Drosophila populations are monomorphic for most genes
(Nμ<1), so robustness is unlikely to evolve through the mechanism in
our model. The small population size we use strongly increases the effect of drift,
and may lead to increased genetic load and heterozygosity compared to
larger/infinite populations [55]. Additionally, the increased drift due to low
population sizes can hide the effect of weak selective pressures [56],[57].
Despite these limitations, our simulation incorporates a more realistic network and
mutational effects than those in previous studies, and further advances in computing
power will allow larger simulations.
Theory has predicted that sex and diploidy can evolve increased robustness in the
presence of genetic variation [14], [27]–[29],[31],[32],[37].
Mutational robustness can evolve without recombination when there is sufficient
genetic variation (Nμ>1). In randomly-wired haploid transcriptional
networks, recombination leads to evolution of robustness when
Nμ = 1 [35], a result that we did
not observe in our haploid segment polarity network, though this may be due to the
short duration of our simulations (5,000 generations) or small population sizes.
More generally, Proulx and Phillips[14] predict that selection for robustness depends on
the fitness load (effect of variation from all sources), and we clearly see diploid
sexual populations have the greatest load (from mutation and recombination), while
sex has little effect on haploid populations (Figures 6 and 9). Our results are generally consistent with
this theory except for the substantial decrease in mutational robustness under
conditions of lower mutation rate in asexual populations (Figure 9). The most likely explanation for this
decrease is because the diploid founders are homozygous for all alleles, and thus
both ‘halves’ of the network are identical. Theory predicts
networks would rapidly accumulate deleterious recessive mutations[14],[58] that
were masked by the working counterpart, and such mutations would persist in asexual
populations without recombination to remove them. Because there is only a single
working allele, there is nothing to rescue this network when that allele is mutated,
resulting in a decrease in robustness compared to the founder. The decrease in
robustness does not continue forever, reaching a minimum by approximately generation
1,000. The initial decrease in robustness reflects the loss of functionally
redundant alleles possessed by the founders, consistent with theory that suggests
selection to maintain both alleles is weak [58].
We found diploidy confers a robustness advantage primarily because most deleterious
mutations are recessive to their working counterparts. Our model allows the
possibility of dominant mutations, but predicts that most deleterious mutations are
recessive in the segment polarity network. This is consistent with metabolic
networks, however, we do not allow for the possibility of interference between two
alleles (i.e. that wg1 might bind nonproductively to its targets, blocking wg2
activity as shown in Figure 3
and discussed in the Models section). Because of this, our model may underestimate
the rate of dominant deleterious mutations, which are important for dominance
evolution [59]. Future studies could explore the effect of more
detailed allelic interaction, and incorporate more realistic rates of the different
types of mutation and their quantitative effect, once such data is available.
Additionally, our scheme allows us to simulate the effects of both
cis and trans mutations, and future studies could
also explore differences in mutational rates and whether they are consistent with
genomic data [60].
The selection pressure that acts upon the real segment polarity network is not known.
Since the segment polarity network stabilizes stripes of gene expression that
activate downstream processes at the proper location, fitness must depend on the
pattern produced. Our truncation selection explores the simple assumption that the
expression of a segment polarity gene must be above a threshold for activation of
those processes in the correct location and below this threshold everywhere else,
for development to proceed normally. Alternatively, genomic data [45]–[47] suggest that many
genetic networks are under stabilizing selection—maintaining specific,
optimal levels of gene expression through time. Our simulations show truncation
selection leads to evolution of higher gene expression (far above threshold) in
those cells that should express the gene. Intuitively, very high gene expression
levels should buffer the system to tolerate perturbations that cause slight changes
in expression level [61], and our simulations are consistent with this
intuition and previous modeling [62]. We do not impose a cost associated with higher
expression, though presumably greater synthesis comes with a metabolic cost that
would eventually limit the expression. The ultimate level of expression depends upon
on the balance of synthesis and degradation, and mutations that solely increase the
stability of a gene product likely have little metabolic cost, but it is difficult
to determine the upper limit for gene product stability. Thus, in truncation
selection, our founders had non-optimal patterns of gene expression that satisfied
the developmental task; and evolved towards a more optimal phenotype with high
expression levels of essential genes. However, the increase in expression alone
shows only a weak correlation with increased robustness (Figure 8), and robustness in both truncation and
stabilizing selection shows similar increases despite an unchanging pattern under
stabilizing selection. Many parameters in our model reflect the activity of a gene
product (K parameters) and so gene activity can change without
changes in expression. It is likely that the absolute expression level is less
important than the amount by which the expression exceeds the minimum/maximum
threshold for activity. Thus, populations rapidly produce increased robustness
regardless of whether the initial phenotype is optimal, and can evolve increased
robustness without dramatic changes in phenotype.
Several extensions of this work warrant future study. We do not allow for the
possibility of new regulatory interactions or gene duplication events (but we do
allow for interaction loss) that alter the topology of the network. The topology of
the segment polarity network to robustly stabilize stripes of wg and en expression
may be nearly optimal, as indicated by a search of nearby network topologies in a
simplified network [63]. It would be interesting to extend our
simulations to allow the topology to change (i.e. the rise of new regulatory
interactions) and gene duplication events, to see whether this topology is
evolutionarily preserved or, if evolution settles on an alternate network. Gene
duplication events would be particularly interesting because a duplication of all
genes in the network would effectively increase the ploidy. Many organisms exist as
tetraploid, octaploid or beyond, and others can amplify their genomic content
through endoreplication [64] to attain very high ploidy(>1000C).
Additionally, some viruses can have high effective ploidy when multiple viruses
infect the same cell [65]. Our study suggests that having 2 copies of each
gene can confer a robustness advantage over just one because most mutations are
recessive and this more than compensates for the doubling of mutation rate. It would
be interesting to explore under what conditions an increase in ploidy ceases to be
advantageous in real networks, and why diploidy, as opposed to tetraploidy or beyond
is so common. Finally, our scheme for translating genotype to model parameters would
easily extend to randomly-wired networks used in previous studies [5], and
allow diploid networks to be explored.
It is an open question as to how general our results are for other real networks.
Theory and modeling studies indicated that increased phenotypic robustness readily
evolves under conditions of interacting genes and variation in haploid networks
[5],[31],[35],[36]. Based on these studies and ours, we speculate
diploid sexual populations will evolve greater mutational robustness in networks
when most deleterious mutations are recessive, and there is sufficient interaction
between gene products so that recombination will select for alleles that can combine
robustly with other alleles. By allowing both masking (by diploidy) and allele
shuffling (recombination), the two can combine to achieve greater robustness than
either alone. Thus, a sexual population for which robustness is important would
likely favor a dominant diploid, not haploid, life cycle.
|
10.1371/journal.pbio.1002364 | An Interferon Regulated MicroRNA Provides Broad Cell-Intrinsic Antiviral Immunity through Multihit Host-Directed Targeting of the Sterol Pathway | In invertebrates, small interfering RNAs are at the vanguard of cell-autonomous antiviral immunity. In contrast, antiviral mechanisms initiated by interferon (IFN) signaling predominate in mammals. Whilst mammalian IFN-induced miRNA are known to inhibit specific viruses, it is not known whether host-directed microRNAs, downstream of IFN-signaling, have a role in mediating broad antiviral resistance. By performing an integrative, systematic, global analysis of RNA turnover utilizing 4-thiouridine labeling of newly transcribed RNA and pri/pre-miRNA in IFN-activated macrophages, we identify a new post-transcriptional viral defense mechanism mediated by miR-342-5p. On the basis of ChIP and site-directed promoter mutagenesis experiments, we find the synthesis of miR-342-5p is coupled to the antiviral IFN response via the IFN-induced transcription factor, IRF1. Strikingly, we find miR-342-5p targets mevalonate-sterol biosynthesis using a multihit mechanism suppressing the pathway at different functional levels: transcriptionally via SREBF2, post-transcriptionally via miR-33, and enzymatically via IDI1 and SC4MOL. Mass spectrometry-based lipidomics and enzymatic assays demonstrate the targeting mechanisms reduce intermediate sterol pathway metabolites and total cholesterol in macrophages. These results reveal a previously unrecognized mechanism by which IFN regulates the sterol pathway. The sterol pathway is known to be an integral part of the macrophage IFN antiviral response, and we show that miR-342-5p exerts broad antiviral effects against multiple, unrelated pathogenic viruses such Cytomegalovirus and Influenza A (H1N1). Metabolic rescue experiments confirm the specificity of these effects and demonstrate that unrelated viruses have differential mevalonate and sterol pathway requirements for their replication. This study, therefore, advances the general concept of broad antiviral defense through multihit targeting of a single host pathway.
| How infected cells respond to a virus during the first minutes to hours after infection can determine whether a disease develops and influences the host’s long-term survival. In mammals, unlike plants and flies that use small RNAs to fight viral infections, virus-induced interferon responses are a critical early event resulting in broad protection against infection. Interferon is a secreted host protein that binds to receptors on the surface of infected and uninfected cells and activates biochemical pathways that profoundly change the expression of hundreds of cellular genes, including those encoding microRNAs. The antiviral functions of only a handful of these genes are understood, and it is not known how the majority contribute to broadly protect against many different viruses. In this study, we uncover an interferon-regulated microRNA (miR-342-5p) that contributes to broad host cell immunity against infection through the cholesterol biosynthesis pathway. We show that miR-342-5p does this through a multihit strategy, turning down the master regulator of sterol biosynthesis as well as several specifically targeted enzymes within the pathway. A wide range of viruses depend on a number of the metabolite side-branches of the sterol biosynthesis pathway for their replication. Notably, our study reveals that by utilising multihit targeting of key branch-points in a single pathway, miR-342-5p is able to inhibit the replication of unrelated, clinically significant pathogens ranging from Herpes to Flu viruses.
| The innate immune response plays a critical role in cellular resistance to infection. In plants and invertebrates, small interfering RNAs (siRNAs) play a vital role in cell-autonomous immunity to viruses. siRNAs are generated and amplified from viral RNA molecules by the cellular RNAi machinery and instil a profound protection against the respective pathogen [1,2]. In mammals, interferon (IFN)-mediated JAK-STAT signalling orchestrates the cellular response to infection and, for several decades, research has focused on the identification and characterisation of antiviral proteins [3]. In contrast, roles for small RNAs in infection of mammalian cells have yet to be fully established (Fig 1) [4].
Several cellular microRNAs have been shown to contribute positively or negatively to infection by targeting host or viral gene expression [5–8]. In this context, IFN stimulation alters the expression of hundreds of cellular genes including some microRNAs [6,9]. Importantly, however, antiviral functions have been attributed to very few of these, although evidence for IFN-induced miRNAs targeting specific pathogens has been documented [6]. A key unresolved issue with important therapeutic implications is: how can IFN-regulated miRNA enhance antiviral functionality? In 2007, Pedersen et al. showed that an IFN-elicited down-regulation of a virus-targeting host miRNA (miR-122) leads to a reduction in Hepatitis C virus replication. This was notable as, historically, studies of IFN antiviral effects have focused on induced genes and this work, identifying a reduction in the expression of a virus-targeting host miRNA, has led to significant clinical advances [6,10]. Attention has also focused on interferon-regulated miRNA enhancement of IFN-mediated antiviral effects through the suppression of negative regulators such as the SOCS proteins [11]. Importantly, whilst antiviral miRNAs targeting host transcripts are emerging, few, if any, studies have characterised the precise mechanisms through which these function. miRNA targeting of multiple transcripts opens up potential for the simultaneous regulation of divergent and convergent cellular pathways and functions. It still remains an open question whether mammalian IFN-induced host-directed miRNAs act as effectors of the cell-intrinsic antiviral immune response. Such a strategy would arguably confer an advantage, as it has the potential to endow a broad-spectrum antiviral activity and greater resistance to the development of escape mutants.
Viruses rely on metabolic and biosynthetic cellular pathways for their replication, and a common feature is a dependency on cellular lipid metabolism [12]. As such, pharmacological inhibition of lipid biosynthesis can curtail virus replication [13–19]. Moreover, there is increasing evidence that the immune system and lipid pathways are tightly coupled [20–22] and defining how innate immunity and lipid metabolism are integrated, share resources, and crossregulate one-another during infection may result in new therapeutic strategies [20,23–26]. In this regard, IFN-induced inhibition of sterol biosynthesis serves as an integral component of the very early cellular response to virus infection and we, and others, have shown that cholesterol 25-hydroxylase (CH25H) and its cognate metabolite 25-hydroxycholesterol (25-HC) are important effectors in this response [18,24,27].
In this study, we sought to further unravel a previously unreported cellular mechanism underpinning immune regulation of lipid pathways and the inhibition of a broad range of viruses. We present evidence for a new miRNA-mediated cell-intrinsic antiviral effector arm of the IFN response, activated during the very early stages of infection. The miRNA under investigation has the capacity to instil a broad antiviral state in vitro and in vivo. Mechanistically, we present data to show the broad antiviral functions of the miRNA occur via a specific multihit suppression of the sterol-biosynthetic pathway.
In experiments, seeking to address whether the activity of CH25H (UniProt: Q9Z0F5) was sufficient to account for the IFN-induced down-regulation of cholesterol biosynthesis in macrophages, we measured the abundance of several key sterol pathway transcripts including HMGCS1 (Entrez Gene: 208715), HMGCR (Entrez Gene: 15357), MVD (Entrez Gene: 192156), SQLE (Entrez Gene: 20775), and SREBF2 (Entrez Gene: 20788) in CH25H-/- cells treated with interferon gamma (IFN-γ) (UniProt: P01580). Importantly, in the absence of CH25H, the effects of IFN-γ on these transcripts were reduced but not abrogated (Fig 2A and 2B) at 6 h (HMGCR, SQLE and SREBF2) and 24 h (HMGCS1, HMGCR, MVD and SQLE) after addition of the cytokine. These data suggested that 25-HC-dependent and independent mechanisms are involved early in the suppressive effect of IFN-γ on the sterol pathway.
To gain further kinetic insights into IFN-γ-elicited alterations in mRNA expression, we undertook a serial (every 30 mins for 8 h) high-resolution, systematic microarray analysis of de novo RNA synthesis and overall RNA abundance in bone marrow-derived macrophages (BMDM) treated with IFN-γ. A flow diagram summarising our protocol for the isolation and parallel analysis of newly transcribed and total RNA is presented (S1 Fig). These experiments revealed a significant down-regulation in the rate of transcript synthesis for 13/19 members of the cholesterol pathway (Data: Fig 3A and 3B and S2, S3 Figs Pathway: Fig 4) [28].
Decreases in the rate of pathway transcript synthesis commenced 2 h to 3 h after treatment (Fig 3A, S2 Fig) and were maintained for the duration of the time course (8 h). In addition, the rate of synthesis and abundance of SREBF2 (but not SREBF1 (Entrez Gene: 20787) also decreased in response to IFN-γ treatment (Fig 3C and 3D, S2 and S3 Figs). We note a more pronounced repression of SREBF2 synthesis over abundance, which likely reflects an IFN-γ action on the well-characterised transcriptional autoregulation of SREBP2 [29]. Taken together, this data is consistent with previous studies [18,24,30,31], demonstrating that transcriptional mechanisms play a role in IFN-γ-mediated down-regulation of cholesterol. However, unexpected reductions in the abundance, but not synthesis of transcripts such as MVD, NSDHL, and DHCR7 (Fig 3A and 3B, S2 and S3 Figs) suggested that cholesterol pathway transcripts may also be subject to 25-HC-independent post-transcriptional regulation. As IFN-γ-mediated suppression of the sterol pathway is strictly dependent on JAK-STAT signalling, we hypothesised that a likely post-transcriptional mechanism might involve IFN-stimulated miRNAs specifically targeting transcripts within the sterol metabolic network.
In support of the above hypothesis, studies have documented miRNA which can regulate lipoprotein uptake (e.g., miR-125a and -455), lipid biosynthetic enzyme expression (e.g., miR-155, miR-21, and miR-185) and, in particular, cholesterol efflux (e.g., miR-33, miR-144) [32–37]. IFN-γ treatment of a melanoma cell line suggested that some of these miRNA (e.g., miR-125a, -455 and -185) may be part of an IFN response; however, it is not known if they are directly coupled to IFN [9]. Thus, we assessed changes in the expression of miRNA precursors (pri/pre-miRNAs) in BMDM stimulated with IFN-γ. Using conservative criteria for detection, we identified 66 pri/pre-miRNAs in our macrophage data (Fig 5A and 5B). Temporal analysis of our time course microarray data, however, revealed two of these, namely pri/pre-miR-155 (Entrez Gene: 387173) and pri/pre-miR-342 (Entrez Gene: 723909), to be significantly up-regulated during the first 8h of IFN-γ treatment (Fig 5A and 5B). MiR-155 is an evolutionarily highly conserved, NF-κB-responsive miRNA encoded by the MIR155 host gene (MIR155HG). It is highly expressed in activated macrophages and lymphocytes [38]. Much less is known about the conserved miR-342 located in an intron of the Ena-vasodilator stimulated phosphoprotein gene (EVL–Entrez Gene: 14026) in the mouse or Ena-Vasp-Like (EVL–Entrez Gene: 51466) in the human (Fig 6A) [39]. Gene expression data from 89 mouse cells and tissues analysed in the BioGPS GeneAtlas indicates the EVL transcript is primarily expressed in cells of the immune and nervous system [40]. In macrophages, miR-342 has previously been identified as a PU.1-regulated miRNA contributing to myeloid differentiation and a proinflammatory mediator capable of enhancing miR-155 expression [41,42]. To confirm the presence of mature miR-155, miR-342-3p and -5p derived from the precursors detected in our array analysis, we stimulated BMDM with IFN-γ (10 U/ml) or interferon beta (IFN-β) (Uniprot: P01575) (25 U/ml) and analysed miRNA expression using quantitative reverse transcription polymerase chain reaction (Q-RT-PCR). In these analyses, significant increases in the expression for all 3 miRNAs were observed (Fig 6B and 6C, S4A and S4B Fig).
Fig 6D and S4C Fig show the coordinate regulation of both the synthesis and abundance of the primary EVL transcript and pri/pre-miR-342 in IFN-γ stimulated BMDM. Both the primary transcript and precursor decreased in synthesis rate (relative to the mock) in the first 60 min after treatment. This was followed by an elevated synthesis rate and increased abundance from around 120 min onwards. Although the transcription factors IRF1 (UniProt: P10914) and IRF9 (UniProt: Q00978) have been implicated in the regulation of EVL in the context of long-term retinoic acid treatment of human promyelocytic leukaemia cells [43], it is not known whether the EVL promoter can be directly regulated by IFN and whether IFN signalling alone is sufficient for its induction [42]. To test the cell-autonomous requirement for IFN signalling in the up-regulation of EVL and pri/pre-miR-342 during infection, we analysed by microarray the endogenous regulation of these transcripts in BMDM with a genetic ablation in IFN beta (IFN-β), the type 1 IFN receptor (IFNAR1 –UniProt: P17181) (Fig 6E) or the down-stream signalling molecule tyrosine kinase 2 (TYK2 UniProt: Q9R117) (S4D and S4E Fig). Ablation of IFN signalling abolished murine cytomegalovirus (MCMV)-induced induction of EVL and pri/pre-miR-342 in all cases demonstrating that virus-induced up-regulation of EVL and pri/pre-miR-342 is dependent on an intact type 1 IFN response. By computational promoter analysis, we identified three putative binding motifs (one IFN-stimulated response element/IRF1 binding site [ISRE/IRF1] and two potential IFN regulatory factor [IRF] binding sites) within a 1 kb region of the EVL promoter (S5A Fig). Luciferase reporter plasmids containing a wild-type promoter or promoters with one or multiple mutations in the three predicted motifs validated the ISRE/IRF1 motif (Fig 6F). Mutation of either the proximal or distal interferon regulatory factor (IRF) motifs resulted in an intermediate reduction in promoter activity. We next performed chromatin immunoprecipitation assays to test for STAT1 (UniProt: P42225) and/or IRF1 (UniProt: P15314) binding to the murine EVL/miR-342 promoter. In IFN-γ-treated murine BMDM, the positive control CXCL10 promoter, but not the EVL/mir-342 promoter, was occupied by STAT1 or IRF1 2 h after treatment (S5B Fig upper graph). By 6.5 h, however, we observed an IRF1-specific enrichment (~3-fold) of EVL/miR-342 promoter sequences (Fig 6G) and by 24 h, IRF1 recruitment was still detectable (S5B Fig lower graph). We conclude that canonical IFN signalling is required for increased expression of miR-342 and that this involves the recruitment of IRF1 to the EVL/miR-342 promoter.
MiR-342 has recently been implicated in the regulation of SREBP2 in a cancer cell line; however, biological roles and precise mechanisms for the miRNA in relation to sterol biosynthesis and the immune response were not addressed [44]. To systematically test whether miR-155 or miR-342 could regulate the cholesterol pathway, miR-155, -342-3p or -342-5p mimics were transfected into a fibroblast cell line and RNA levels for selected cholesterol biosynthesis enzymes quantitated. While miR-155 and miR-342-3p had no significant effect on gene expression in the cholesterol pathway, miR-342-5p reduced the overall abundance of HMGCS1 (−32%), HMGCR (−51%), MVD (−48%), SQLE (−51%) and NSDHL (Entrez Gene: 18194) (−62%) relative to cells treated with the nontargeting control (Fig 7A). In a comprehensive sterol pathway analysis of RNA extracted from primary murine embryo fibroblasts (pMEF) transfected with miR-342-3p or -5p, we proceeded to show that miR-342-5p but not miR-342-3p induced a significant, coordinate reduction in the abundance of 12 out of 19 sterol biosynthesis pathway transcripts analysed (Fig 7B and 7C). S6 Fig shows a comparison of the reduction in abundance of the sterol pathway transcripts by miR-342-5p with that elicited by a SREBF2 targeting siRNA. Notably, the abundance of the fatty acid-associated transcript FASN (Entrez Gene: 14104) was not altered by miR-342-5p in these experiments (Fig 7B and 7C).
The concordant reduction of so many genes was consistent with the repression of a master regulator of the sterol biosynthesis pathway. SREBP2, the master transcriptional regulator for the majority of genes encoding sterol pathway enzymes, contains a predicted miR-342-5p binding site that is conserved in both human and mouse (Fig 8A). To directly test the function of this predicted miR-342-5p target, we generated dual-luciferase reporter constructs for the human and mouse SREBF2 and mouse SREBF1 3’UTRs. Cotransfection of these reporters with miR-342-5p significantly reduced luciferase activity for the murine SREBF2 but not the SREBF1 reporter construct (Fig 8B, 8C and 8D). In addition, mutation of the seed region in both the human and mouse SREBF2 3’UTR increased luciferase expression (Fig 8B and 8C). Transfection of several mouse or human cell types with the miR-342-5p mimic decreased SREBF2 transcript abundance (Fig 9A: pMEF, Fig 9B: BMDM, Fig 9D: NIH/3T3, S7A and S7B Fig). To avoid any complications arising from miRNA overexpression, we also sought to test the effect of inhibiting endogenous miR-342-5p on IFNG-mediated SREBF2 regulation. In these experiments, we found miR-342-5p inhibitor significantly decreased but did not eliminate the repressive effects of a nonsaturating dose of IFNG [45] on SREBF2 copy number in BMDM (Fig 9C). The results of these investigations show a contributory role of approximately 30% to 50% for miR-342-5p in the macrophage IFN suppression of SREBF2.
In further experiments, protein levels for both the uncleaved (P) and cleaved (N) form of the SREBP2 (UniProt: Q3U1N2) protein were also decreased in cells transfected with the miR-342-5p mimic (Fig 9E). The latter result contrasts with the action of 25-HC that reduced the cleaved protein (N) abundance alone (Fig 9E). Human SREBF1 (Entrez Gene: 6720) has recently been identified as a target of miR-342-5p (confirmed in S7B Fig) [44]. Our bioinformatics analysis did not identify a target for this miRNA anywhere in the murine transcript, and the transfection of mouse fibroblasts with a miR-342-5p mimic did not elicit a significant decrease in overall SREBF1 transcript abundance (Fig 9A and 9D).
The transcript SREBF2 contains an intronic miRNA (miR-33) that regulates fatty acid degradation and cholesterol homeostasis [32]. In this connection, it is worth noting that type 1 and 2 IFN treatment and transfection of miR-342-5p reduced miR-33 abundance in BMDM (Fig 10A and S7C Fig). Since miR-342-5p targets the mature, spliced SREBF2 transcript in the cytoplasm, the reduction of miR-33 must be occurring in the nucleus prior to splicing. Whilst the underlying molecular mechanism for this remains to be fully elucidated, we anticipate a decrease in miR-33 occurs as a result of miR-342-5p effects on the transcriptional autoregulation of the SREBF2 promoter by SREBP2 (Fig 10D). Since miR-33-5p targets ABCA1 (Entrez Gene: 11303) and ABCG1 (Entrez Gene: 11307) in mice [32], the question arose: can miR-342-5p regulate the abundance of these cholesterol efflux-related transcripts? The nuclear hormone Liver X receptors LXRα (NR1H3 UniProt: Q9Z0Y9) and LXRβ (NR1H3 UniProt: Q60644) are activated by oxysterol binding and are well-known regulators of cholesterol homeostasis acting to regulate the transcription of ABCA1 and ABCG1. As a consequence, they mediate cholesterol efflux from the cell [46]. To analyse miR-342-5p modulation of ABCA1 and ABCG1 expression, we measured the abundance of these transcripts in transfected BMDM treated with the LXR agonist T0901312. A schematic illustrating the logic of this approach is presented (S7D Fig). MiR-342-5p suppression of miR-33 was confirmed by Q-RT-PCR in these experiments (S7E Fig) and, as previously reported, miR-33 significantly reduced the effects of LXR activation on ABCA1 and ABCG1 transcript abundance in our experiments (Fig 10B and 10C) [32]. In contrast, miR-342-5p enhanced the effect of T0901312 treatment on ABCA1 and, significantly, on ABCG1 expression (Fig 10B and 10C).
We conclude that miR-342-5p targets and acts to suppress SREBF2 via a single, conserved miRNA binding site in its 3’UTR. In addition, our results point to a coupling of miR-342-5p to miR-33 regulation—likely through the SREBP2 transcriptional autoregulation pathway [29]. MiR-342-5p, therefore, also has the potential to counter-regulate LXR-mediated control of cholesterol homeostasis.
We next investigated whether miR-342-5p could regulate intracellular sterol levels. Mass spectrometry analyses of transfected BMDM showed miR-342-5p-induced reductions in metabolites from the mevalonate-shunt (24S,25-epoxycholesterol, Fig 11A), Bloch (desmosterol, Fig 11B) and Kandutsch-Russell pathways (7-dehydrocholesterol + 8(9)-dehydrocholesterol, Fig 11C) demonstrating miR-342-5p inhibits sterol biosynthesis. Accordingly, free (Fig 11D) and total cholesterol (S8A Fig) were also reduced by miR-342-5p transfection in BMDM and fibroblasts, respectively. In agreement with our gene expression data, we further showed that an inhibitor of endogenous miR-342-5p could moderately increase total cholesterol in transfected fibroblasts (S8B Fig). In these experiments, miR-342-5p consistently reduced intracellular cholesterol concentration in cells cultured in medium containing 10% serum. This agrees with experiments showing that miR-342-5p targets biosynthesis through SREBF2 and probably also influences other components of the cholesterol regulatory network responsible for influx and/ or efflux, e.g., ABCG1.
Our previous studies have indicated that although IFN suppression of sterol biosynthesis and the antiviral activities of 25-HC are partly SREBP-dependent, SREBP independent sterol pathway-related mechanisms predominate in the repression of virus infection [18]. In this study, candidate targets for miR-342-5p in the sterol pathway members IDI1 (Entrez Gene: 319554), SC4MOL (MSMO1 Entrez Gene: 66234), and DHCR7 (Entrez Gene: 13360) suggested a possible SREBP-independent mechanism for miR-342-5p (Fig 12A and S1 Table). IDI1 (UniProt: P58044) in particular plays a key role in the synthesis of isoprenoids from sterol pathway metabolic intermediates—a process of great significance to a range of viruses [18,47]. Dual luciferase assays were, therefore, used to validate miRNA target sites for miR-342-5p in the 3’UTR of IDI1 and SC4MOL (Fig 12B and 12C). MiR-342-5p also reduced the expression of a WT DHCR7 3’ UTR reporter (S8C Fig). Taken together, our data show that the IFN-induced miR-342-5p targets sterol metabolism at multiple levels in both a SREBP2-dependent and -independent manner and can regulate both cholesterol biosynthesis and homeostasis.
We, and others, have previously demonstrated that sterol pathway regulation is an integral part of the IFN antiviral response [18,24,46]. We speculated; therefore, that the regulation of sterol biosynthesis and homeostasis by miR-342-5p might mediate antiviral functions and protect cells from virus infection. In support of this hypothesis, we found miR-342-5p but not miR-342-3p significantly inhibited growth of MCMV (S9A Fig), and that this occurred in a dose-dependent manner (Fig 13A). We further showed that an inhibitor of endogenous miR-342-5p could partially rescue MCMV replication in NIH/3T3 fibroblasts treated with IFN-γ (Fig 13B, IFN-γ function confirmed in S9B Fig). We next tested whether miR-342-5p is capable of inhibiting MCMV in vivo. BALB/c mice were injected intraperitoneally with 10 μg (total) of miR-342-5p mimic, inhibitor, or a nonspecific control RNA. Four d postinfection, MCMV titres in miR-342-5p-treated mice (relative to the nonspecific control miRNA) were significantly reduced (~1 log10) in the kidney and lung (Fig 13C). In further, independent in vivo experiments, antiviral miR-342-5p effects were consistently observed; however, organs with a reduced virus titre varied between runs. This is likely attributable to variability in the delivery of relatively small doses of the miRNA (S9C and S9D Fig). In our in vivo experiments, the miR-342-5p inhibitor did not increase viral replication relative to the negative control miRNA treatment (Fig 13C). While further genetic loss-of-function studies are required to unequivocally investigate the in vivo function of miR-342-5p, these studies support the view that this miRNA imparts antiviral activity in vivo and in vitro.
We have previously reported that IFN-induced effects arising via sterol pathway regulation play a key role in limiting MCMV replication and are mediated via the isoprenoid-prenylation branch [18]. This results in severely restricted viral spread at a postentry stage of infection. In agreement with our 25-HC-related work, miR-342-5p transfection of cells did not alter viral entry (Fig 14A) but did significantly reduce infectious virus production (Fig 14B), immediate early gene expression (Fig 14C and 14D), and MCMV plaque diameter. Fig 14E shows a miR-342-5p elicited reduction in the diameter of plaques at 3 d post infection, whilst Fig 14F shows data from a quantitative analysis of plaque diameter in both WT and STAT1-/- cells at 4dpi. The latter cell type was used to minimise potential side effects of transfection on immune activation and subsequent antiviral responses. Of note, miR-342-5p did not notably alter the viability of cells used in any of the described analyses (S10A and S10B Fig). Since miR-342-5p targets the sterol metabolic network, we tested whether it exerted broader antiviral activity by analysing its effects on Human Cytomegalovirus (HCMV), Herpes Simplex virus 1 (HSV1), and Influenza A virus (H1N1) (Fig 15A, 15B and 15C). Replication of all three viruses was inhibited by miR-342-5p (80% for HSV-1, 50% for HCMV and 60% for Influenza virus A).
As previously discussed, whilst IFN-induced suppression of sterol biosynthesis involves a SREBP-dependent mode of action, alternative sterol pathway-related mechanisms predominate in the repression of virus replication [18]. With this in mind, we sought to further investigate SREBP-independent antiviral mechanisms for miR-342-5p. For these experiments, we evaluated whether the specific knockdown of predicted non-SREBP targets of miR-342-5p (IDI1, DHCR7 and SC4MOL) alone were sufficient to impart antiviral effects observed for the miRNA (Fig 16). Knockdown of specific targets by these siRNA was confirmed in pMEF by Q-RT-PCR (S9F Fig). Notably, siRNA-mediated knockdown of IDI1 consistently inhibited (−70%) MCMV replication in pMEF with a similar magnitude as inhibition mediated by a virus-specific (M54, MCMV DNA polymerase) or HMGCR-targeting siRNA (Fig 16). In contrast, siRNA targeting of SC4MOL and DHCR7—enzymes that function downstream of the isoprenoid pathway branch—did not limit MCMV replication. Fig 12C shows functional targeting of SC4MOL. Since SC4MOL catalyses six distinct enzymatic steps in the Kandutsch-Russell/Bloch arms of the sterol biosynthesis pathway (Fig 4), we anticipate miR-342-5p targeting of this enzyme could have significant effects on cholesterol biosynthesis in the cell and antiviral effects on pathogens dependent on this arm of the pathway. This data is in agreement with previous studies and demonstrates that SREBP-independent targeting of the sterol biosynthesis pathway can elicit a profound antiviral effect [18,27].
It is known that unrelated viruses have a common dependency on the mevalonate- sterol pathway. Evidence suggests, however, that the relative significance of individual pathway members differs according to virus type and the respective demands of its replication strategy [48,49]. In the case of MCMV, we have previously demonstrated that the addition of exogenous metabolic intermediates (mevalonate or geranylgeranyl diphosphate but not squalene [SQL]) to cells can partially rescue MCMV growth in vitro by compensating for IFN or 25-HC-induced suppression of the sterol pathway [18,24]. In this study, we sought to test whether the antiviral effects of miR-342-5p require the suppression of metabolic intermediates in the mevalonate and sterol biosynthesis arms of the pathway. In a series of MCMV metabolite rescue experiments, and in agreement with our 25-HC-related work, we found that geranylgeraniol (GGOH) (Fig 17C and 17D) and, to a lesser degree, mevalanolactone (MEV) (Fig 17A and 17B) (but not farnesol [FOH] or SQL, Fig 17E and 17F), could significantly increase and partially restore MCMV growth in miR-342-5p transfected pMEF. These results highlight the importance of host-targeting of the mevalonate arm of the sterol pathway by miR-342-5p and are in good agreement with our previous studies [18,24].
In contrast to MCMV, a series of Influenza virus metabolite rescue experiments demonstrated that MEV, FOH, SQL and, to a lesser extent, GGOH all partially rescued infectious virus production by A549 cells in the context of miR-342-5p but not RAB11A siRNA transfection (Fig 18A, 18B, 18C and 18D). RAB11 (UniProt: P62491) is known to be required for Influenza A virus budding and filament formation [50]. Whilst caution should be taken to avoid overinterpreting the relative effects of specific metabolites in unrelated cell types, when taken together, our metabolite rescue data emphasise two key mechanistic issues. Firstly, unrelated viruses depend on distinct metabolic aspects of the sterol biosynthesis pathway for their replication. Secondly, despite these distinctions, “multihit” targeting of the sterol pathway by miR-342-5p enables this miRNA to elicit a broad antiviral effect.
MiR-342-5p has recently been reported to specifically target Cocksackie B virus [51]. While we cannot exclude the possibility that miR-342-5p also directly targets MCMV, HCMV, and/or HSV-1 (S2 Table), we failed to find any direct viral-RNA targets for the miRNA in Influenza virus A. Given the relatively low number of cellular miRNAs shown to have antiviral activity [5], we consider it unlikely that miR-342-5p exerts broad antiviral effects by directly targeting viral transcripts. On the basis of these observations, combined with evidence from the metabolic rescue experiments, we propose that miR-342-5p exerts its broad antiviral effects by targeting host metabolic activities associated with the mevalonate-sterol biosynthesis pathway.
In this study, we identify miR-342-5p as a cellular miRNA with broad antiviral properties whose transcriptional regulation is coupled to IFN signalling by IRF1. MiR-342-5p is, therefore, an integral component of the cell-autonomous IFN response and exerts its antiviral effects by inhibiting the sterol metabolic network through SREBP-dependent and -independent mechanisms. The latter are, at least in part, mediated via the targeting of key genes in the sterol metabolic network. Metabolic rescue experiments show that unrelated viruses have a common requirement for the mevalonate biosynthesis arm yet depend on subtly different aspects of the distal sterol pathway for their life cycle.
In its role, miR-342-5p complements and reinforces the antiviral functions of the rapidly induced oxysterol 25-HC on sterol biosynthesis—in particular affecting the mevalonate-isoprenoid branch of the pathway. S11 Fig summarises our current understanding of the mechanisms by which IFN induced 25-HC and miR-342-5p function to coordinately regulate sterol biosynthesis. In murine BMDM, synthesis of CH25H mRNA is controlled by STAT1 and is up-regulated in the first 30 min after cells are activated by IFN-γ [18]. In contrast, EVL and pri/pre-miR-342 RNA expression increases 2 h to 3 h after IFN-γ activation of BMDM and is regulated by IRF1. Our data demonstrate, therefore, an IFN-elicited sequential regulation of the sterol metabolic network in which 25-HC provides an immediate, rapid mechanism for decreasing sterol biosynthesis and mediating antiviral effects. This involves a 25-HC blockade of SREBP2 translocation to the nucleus and the proteolytic degradation of HMGCR [18,52]. The effects of 25-HC are followed by miR-342-5p further promoting a more sustained fine-tuning of sterol metabolism and antiviral effects in the cell by targeting SREBF2 RNA and transcripts encoding select enzymes of the sterol biosynthesis pathway (e.g., IDI1 and SC4MOL).
25-HC has recently been shown to mediate anti-inflammatory activities via SREBP2 [30]. Here, we find miR-342-5p functions in a SREBP-dependent and -independent manner to regulate sterol biosynthesis and viral infection. Whether miR-342-5p has SREBP-related inflammatory functions remains to be fully investigated; however, the current study strongly supports the possibility that miR-342-5p might contribute to the regulation of inflammatory responses by targeting SREBP2 or the prenylation arm of the sterol pathway. In this context, a proinflammatory role for miR-342-5p in the enhancement of miR-155 expression has been described and attributed to the former miRNA targeting BMPR1 (Entrez Gene: 12166) and AKT1 (Entrez Gene: 11651) transcripts in macrophages [42]. In this study, we did not observe any notable alterations in BMPR1 transcript synthesis or abundance during 8 h of IFN-γ BMDM treatment. We did, however, observe a small but significant reduction in AKT1 abundance over time in IFN-stimulated macrophages. A knockdown of this transcript with a siRNA, however, had no effect on MCMV replication (S9E Fig).
Intriguingly, miR-342-5p targets multiple members of the sterol biosynthesis pathway including its master regulator SREBP2. The suppression of SREBP2 function results in a generalised decrease in sterol biosynthesis as well as a reduction in the abundance of key contributors to cholesterol homeostasis such as INSIG1 (Entrez Gene: 231070; Fig 7C, S2 and S3 Figs) and LDLR (Entrez Gene: 16835; S8D Fig). Both DNA and RNA viruses require the sterol biosynthesis pathway for optimal growth capacity, and we have recently shown that an IFN-mediated suppression of sterol metabolic network activity is an integral part of the antiviral response [24]. Importantly, however, whilst the IFN regulation of SREBP2 is undoubtedly important, alternative mechanisms play a dominant role in the sterol-related antiviral responses we have observed to-date.
Notably, in this study we found miR-342-5p directly targets enzymes of both the proximal and distal arms of the sterol biosynthesis pathway—in particular IDI1. IDI1 is an enzyme critical to protein prenylation and catalyses the isomerisation of the inactive carbon-carbon double bond of isopentyl diphosphate (IPP) to generate an isomer dimethylallyl diphosphate (DMAPP). Prenylation is a post-translational modification enabling the membrane association of modified proteins and involves the covalent addition of prenyl lipids (e.g., farnesyl or geranylgeranyl) derived from mevalonic acid to conserved residues at the C-terminus of proteins. This process is integral to host intracellular protein trafficking, leukocyte chemotaxis, and phagocytosis and has also been implicated in inflammatory cytokine production [53–56]. In this connection, we have recently highlighted the importance of the mevalonate-isoprenoid branch point to the antiviral effects of 25-HC [24].
The replication of several viruses requires prenylation of host and/or virus proteins. For example, Hepatitis D virus requires prenylation of its large delta antigen for optimal virion morphogenesis, and prenylation inhibitors have shown promise in the treatment of this pathogen [48]. Further, Hepatitis C virus requires the geranylgeranylated host protein FBL2 for replication and respiratory syncytial virus (RSV) F glycoprotein binds to the prenylated host protein RHOA enabling membrane fusion [57,58]. Here, we found that siRNA knockdown of IDI1 inhibits MCMV replication in pMEF [18] and, most importantly, the inhibitory effects of miR-342-5p on MCMV could be partially rescued via exogenous administration of GGOH and, to a lesser degree, MVA to cells. This points to an involvement of the sterol pathway prenylation branch point in the anti-MCMV functionality of miR-342-5p. However, the mode of action of miR-342-5p on the prenylation branch point remains to be determined. For Influenza virus, the inhibitory effects of miR-342-5p were partially rescued via the administration of exogenous MEV, FOH, SQL and, to a lesser extent, GGOH to cells. Thus, different targeting mechanisms of miR-342-5p facilitate broad antiviral effector functions.
The metabolic rescue data described above is crucial to our findings. It confirms sterol pathway targeting is essential to the antiviral effects of miR-342-5p we observed and highlights that whilst many viruses depend on the sterol metabolic network for their replication, unrelated virus types may exploit subtly different aspects of this pathway for their replication cycle. As a result, “multihit” targeting of a single pathway by miR-342-5p has beneficial outcomes for its broad antiviral effects.
Studies supporting a role for antiviral RNA-mediated interference in mammalian systems are beginning to emerge; however, the IFN response is still considered the pre-eminent mechanism by which mammalian cells resist viral infection [59–61]. Specific targeting of viral transcripts by host miRNAs has been reported, e.g., for hepatitis C virus (HCV) [6]. However, this is unlikely to represent a major mechanism for restricting viruses, as they are predisposed to the emergence of escape mutants [62]. In contrast, IFN-elicited miRNAs targeting cellular pathways required for virus replication arguably provide more robust and durable effects against a broad spectrum of viruses. Here, to our knowledge, we identify for the first time an IFN-induced host-targeting miRNA that elicits broad antiviral effects. Whilst we cannot preclude the possibility that miR-342-5p has a direct effect on viral RNA expression, the breadth of viruses repressed by this miRNA (both DNA and RNA) argues against a direct targeting of viral transcripts. Although miR-342-5p has been shown to specifically inhibit coxsackie B virus [51], our work emphasises the multifunctional, dual potential of this IFN-regulated antiviral miRNA.
In conclusion, the multihit targeting of sterol synthesis and antiviral effects mediated by miR-342-5p represent a new arm of the IFN-induced cell-autonomous immune response to viral infection and a new mechanistic link between lipid metabolism and the very early innate immune response. In this regard, miR-342-5p and its sterol pathway targets provide the foundation for future therapeutic exploitation, and this study highlights a general principle for blockade of infection.
C57BL/6 mice were housed in the specific pathogen-free animal facility at the University of Edinburgh. BALB/c mice were housed in the specific pathogen-free animal facility at the Institut d’Investigacions Biomèdiques August Pi i Sunyer, Barcelona, Spain. CH25H-/- (B6.129S6-Ch25htm1Rus/J) mice were purchased from Charles River (Margate, United Kingdom) and housed in the specific pathogen-free animal facility at the University of Edinburgh. All procedures were carried out under project and personal licences approved by the Secretary of State for the Home Office, under the United Kingdom's 1986 Animals (Scientific Procedures) Act and the Local Ethical Review Committee at Edinburgh University. All procedures involving animals and their care in Spain were approved by the Ethics Committee (protocol number CEEA 308/12) of the University of Barcelona and were conducted in compliance with institutional guidelines as well as with national (Generalitat de Catalunya decree 214/1997, DOGC 2450) and international (Guide for the Care and Use of Laboratory Animals, National Institutes of Health, 85–23, 1985) laws and policies.
BMDMs were isolated and differentiated with CSF-1 derived from L929 cells. Details of all cell culture conditions are provided in S1 Methods.
Unless otherwise stated, murine recombinant IFN gamma (IFN-γ) (Perbio Science) and IFN-β (Stratech, UK) were added to cells at a final concentration of 10 U/ml or 25 U/ml, respectively. For experiments investigating the effect of the miR-342-5p inhibitor on SREBF2 transcript abundance in BMDM, subconfluent macrophages were transfected for 24 h and then treated with 2.5 ng/ml murine recombinant IFNG (Life Technologies) for 24 h. Lipopolysaccharides from Escherichia coli 026:B6 (Sigma-Aldrich, UK) were reconstituted in SPBS (1 mg/ml) and added to BMDM at a final concentration of 100 ng/ml.
The GFP-encoding MCMV (MCMV-GFP) has been previously described [63]. HCMV-GFP (AD169-GFP) has been previously described [64]. HSV-1-eGFP (C12) was propagated and titred by plaque assay in Vero cells. A/WSN/33 (H1N1) influenza virus was propagated and titred in MDCK cells. Growth and titration conditions are provided in S1 Methods.
MiRNA mimics or siRNA were transfected into cells using DharmaFECT 1 (Thermo Fisher Scientific). After 48 h or 72 h, cells were infected with virus and growth was measured using a POLARstar OPTIMA plate reader (BMG Labtech) according to manufacturers’ recommendations. See S1 Methods.
Incorporation of 4-thiouridine (Sigma) into newly-transcribed RNA was undertaken as described by Dölken et al. [65]. See S1 Methods for further information.
Processing of ntRNA samples (100 ng) for hybridisation to Affymetrix Mouse Gene 1.0 ST arrays was undertaken according to manufacturer’s instructions (Affymetrix). Hybridisation, washing, staining, and scanning of the arrays were also undertaken following standard Affymetrix protocols. After scanning and data capture, open-source R-based software “Bioconductor” was used to implement all quality control and statistical analyses. See S1 Methods for further information.
Total RNA was extracted from cells with RNeasy Mini kit (QIAGEN). Quantitative gene-expression analyses were then performed using Roche UPL reagents, IDT PrimeTime (IDT, United States) assays or Taqman primer probe sets (Applied Biosystems). Expression of target genes was normalized to Actb unless otherwise stated. See S1 Methods for further detailed information.
Total RNA from in vitro tissue culture experiments was isolated using a Qiagen miRNeasy kit according to manufacturer’s recommendations (Qiagen, US). MiRNA expression analyses were then performed using reagents and assays from Quanta Biosciences, US as per manufacturers recommendations. See S1 Methods for further detailed information.
Stratagene MXPro software was used to analyse the data. Threshold determinations and differences in transcript abundance relative to controls were automatically performed by software for each reaction.
MicroRNA mimics, control miRNAs and siRNA were purchased from Dharmacon RNAi Technologies, Thermo Fisher Scientific (Lafayette, USA). Mimics and siRNA were transfected into cells using DharmaFECT 1 (Thermo Fisher Scientific). ZEN-AMO and 2’OMe/LNA-PS miRNA inhibitors were obtained from IDT and for NIH/3T3 cells were transfected using DharmaFECT 1 [66]. For BMDM or RAW cells, miRNA mimics or inhibitors were transfected (final concentration 25 nM unless otherwise stated) into cells using Viromer Blue (Lipocalyx, Germany) as per manufacturers recommendations. See S1 Methods for further information.
For in vivo experiments, miRNA were administered by an intraperitoneal injection route as previously described [67]. For infection, mice were injected with 1 x 106 PFU MCMV in SPBS on day 3 of the experiment. Mock-infected animals were injected in an identical manner with SPBS only. Tissues were collected and snap frozen for subsequent analysis of MCMV titre. See S1 Methods for further information.
Regions from the 3’UTR of the gene were synthesized and subcloned into the 3’UTR MCS of the psiCheck2 renilla luciferase by Eurofins MWG Operon (Ebersberg, Germany). MiRNA mimics or controls were reverse transfected into cells with either wild-type or mutant luciferase reporter in DharmaFECT DUO. After 24 h, luciferase expression in the transfected cells was measured using a dual-luciferase reporter assay kit (Promega, UK). See S1 Methods for further information.
25-HC (Sigma, H1015) was dissolved in 100% Ethanol (1000x stock, 20 mM) and stored at −20°C under argon in 2 ml opaque tubes with gasket screw-top lids. The powder (2-Hydroxypropyl)-β-cyclodextrin (HβCD) (Sigma, H107) was dissolved in medium at 37°C just before use.
Cholesterol was extracted from cells using Chloroform/Methanol protocol detailed in S1 Methods. Total Cholesterol quantitation was carried out using an Amplex Red Cholesterol Assay Kit as per manufacturers recommendations (Invitrogen, UK).
Sterols and oxysterols were analysed using liquid chromatography—mass spectrometry (LC-MS) on an Orbitrap Elite (ThermoFisher) operated as described previously[68]. Sterols and oxysterols were identified by comparison of m/z, retention time and MSn fragmentation with reference standards. Quantification was by stable isotope-dilution. See S1 Methods for further information.
For the prediction of potential microRNA targets in genomic 3’ UTR regions the database TargetScan was used [69]. For the prediction of microRNA targets in 5’UTR, coding regions and 3’UTRs the database miRWalk was used [70].
MiRanda version 3.0 was used to scan viral coding sequences for predicted miRNA binding sites. The viral coding sequences were obtained from GenBank via NCBI for the following accessions: MCMV (NC_004065.1), HCMV (FJ527563.1), HSV1 (X14112.1), and influenza A (A/WSN/33 (H1N1): X14112.1, CY010795.1, CY010794.1, CY010793.1, CY010788.1, CY010791.1, CY010790.1, CY010789.1 and CY010792.1). MiRanda results were extracted as key-value pairs and sorted according to total score and free energy. See S1 Methods for further information.
To analyse and predict potential Stat1, Irf1 and Irf9 binding sites in the promoter of Human and Mouse EVL/Evl, the open source software Toucan was used [71]. See S1 Methods for further information.
BMDM were treated with Ifn-γ (10 U/ml) for 2, 6, and 24 h, fixed and then Chromatin immunoprecipitation (ChIP) was performed as described previously [72]. Primers for the amplification of promoter regions from EVL and the positive control gene CXCL10 were designed using PrimerBLAST and are provided in S5 Fig. Quantitative-PCR using SYBR-green incorporation (Quanta PerfeCTa SYBR Green FastMix, Low ROX) was used to analyse enrichment of sequences relative to input DNA. See S1 Methods for detailed information.
A 421-bp region from the Human EVL promoter containing three predicted IFN-activated transcription factor binding sites was synthesized and subcloned into the MCS of the pGL4.1 luciferase plasmid by Eurofins MWG Operon (Ebersberg, Germany). In parallel, corresponding mutants of each individual site (designated: ISRE, proximal Irf7 and distal Irf7) and a mutant in which all predicted sites were mutated (designated: All) were produced. Promoter activation by type 1 IFN was then tested as described in S1 Methods.
After washing, normal medium containing vehicle (Ethanol) or GGOH (Sigma G3278), Mevalonolactone (MEV) (Sigma M4667), FOH (Sigma F203), or SQL (Sigma S3626) was added to the infected wells. T0901317 (Tocris Bioscience, Bristol, UK) was resuspended in 100% ethanol (5 mM stock), diluted in normal medium (50 nM final concentration) and cells were treated for 18 h.
Unless otherwise stated, a two-sample Welch t test was used to test statistical significance of results in Microsoft Excel. Prior to parametric testing, a Shapiro-Wilk test was undertaken in R to confirm normal distribution of data. Statistical testing of Q-RT-PCR data from independent experiments (normalised to housekeeping gene ACTB) was undertaken using a one-sample t test in Microsoft Excel. Experimental group sizes (where n = number of independent biological samples per group) are stated in figure legends. Statistical analyses of growth curves was undertaken using a permutation based approach using a browser-based implementation of the “compareGrowthCurves” function originally developed for the statmod software package for R (http://bioinf.wehi.edu.au/software/compareCurves).
The cholesterol biosynthesis model used in this study was derived from a comprehensive consensus Systems Biology Graphical Notation diagram of the regulation and feedback of cholesterol metabolism [28].
|
10.1371/journal.pbio.2001655 | Reduced insulin signaling maintains electrical transmission in a neural circuit in aging flies | Lowered insulin/insulin-like growth factor (IGF) signaling (IIS) can extend healthy lifespan in worms, flies, and mice, but it can also have adverse effects (the “insulin paradox”). Chronic, moderately lowered IIS rescues age-related decline in neurotransmission through the Drosophila giant fiber system (GFS), a simple escape response neuronal circuit, by increasing targeting of the gap junctional protein innexin shaking-B to gap junctions (GJs). Endosomal recycling of GJs was also stimulated in cultured human cells when IIS was reduced. Furthermore, increasing the activity of the recycling small guanosine triphosphatases (GTPases) Rab4 or Rab11 was sufficient to maintain GJs upon elevated IIS in cultured human cells and in flies, and to rescue age-related loss of GJs and of GFS function. Lowered IIS thus elevates endosomal recycling of GJs in neurons and other cell types, pointing to a cellular mechanism for therapeutic intervention into aging-related neuronal disorders.
| Insulin and insulin-like growth factors play an important role in the nervous system development and function. Reduced insulin signaling, however, can improve symptoms of neurodegenerative diseases in different model organisms and protect against age-associated decline in neuronal function extending lifespan. Here, we analyze the effects of genetically attenuated insulin signaling on the escape response pathway in the fruit fly Drosophila melanogaster. This simple neuronal circuit is dominated by electrical synapses composed of the gap junctional shaking-B protein, which allows for the transfer of electrical impulses between cells. Transmission through the circuit is known to slow down with age. We show that this functional decline is prevented by systemic or circuit-specific suppression of insulin signaling due to the preservation of the number of gap junctional proteins in aging animals. Our experiments in a human cell culture system reveal increased membrane targeting of gap junctional proteins via small proteins Rab4 and Rab11 under reduced insulin conditions. We also find that increasing the level of these recycling-mediating proteins in flies preserves the escape response circuit output in old flies and suggests ways of improving the function of neuronal circuits dominated by electrical synapses during aging.
| Synapses undergo age-associated morphological and functional changes in a number of model organisms [1–3] and in humans [4–6]. In the fruit fly Drosophila melanogaster and the worm Caenorhabditis elegans, synaptic changes were seen during normal aging in both central and peripheral parts of the nervous system and linked to cognition, memory, learning, locomotor, and homeostatic deficits [7–14].
In neurons, gap junctions (GJs) constitute the morphological substrate of electric (i.e., electrotonic) synapses characterized by electrical coupling, and permeability for small molecules generally up to approximately 1 kDa (i.e., metabolic or biochemical coupling). GJs play vital roles in the distribution of metabolic substrates [15], tissue development, homeostasis [16], and cell-to-cell communication via calcium waves [17]. Electrical synapses serve important functions in the sensory and motor neurons [18], and in learning and memory [19]. Gap junctional communication is also critical for (nonexcitable) glial cells, providing a pathway that contributes to the uptake of ions and the release of neuroactive substances, so called “gliotransmitters” [20]. A previous work demonstrated a significant decline with age in the density of connexins Cx43 and Cx30, the most abundant astrocytic gap junctional proteins, in mouse brains [21]. The decline in function of the GJs with age is likely to have a pervasive role in both invertebrate and vertebrate nervous systems because electrical synapses play a major role in both [18]. Considering that 25% of the fruit fly brain is composed of glial cells, while in the human brain this percentage reaches about 90% [22], the importance of GJs for nervous system function is profound.
Reduced insulin/insulin-like growth factor (IGF) signaling (IIS) can ameliorate the effects of aging in multiple model organisms and, probably, humans [23,24], pointing the way to a broad-spectrum, preventative medicine for the diseases of human aging. The nervous system is an important case in point, as both insulin and IGF are expressed across physiologically distinct brain regions [25,26] and have well known roles in the development, growth, and survival of the central nervous system [27]. In addition, lowered IIS can lead to insulin resistance and diabetes, neuronal injury [28], and compromised neural control of metabolism [29]. Despite this, chronically lowered insulin/IGF signaling can improve metabolic, synaptic, and cognitive defects in rodent and Drosophila models of several neurodegenerative diseases [30–36], leading to what has become known as the “insulin paradox” [37]. The molecular mechanisms through which lowered IIS mediates improved health, particularly in the nervous system [37,38] are poorly understood.
Here, we chose to study electrical transmission in the giant fiber system (GFS), the fly’s escape response neuronal circuit. Electrical synapses in the GFS are formed by the products of the shaking-B gene, SHAK-B [39] and represent the dominant synapse type in the circuit. Conduction through the GFS in SHAK-B mutants is very weak [40,41], with individual animals producing either no response or a significantly delayed response to a stimulus. In addition, when the chemical (cholinergic) synapses are disabled with tetanus toxin, the circuit function is largely unaffected [42], further demonstrating the predominantly “electrical” nature of this neuronal system. We showed that IIS silencing improves function of the electrical component of the GFS in aging flies by preventing the loss of the principal gap junctional shaking-B protein (SHAK-B). This preservation of the GJ density and circuit function is likely mediated by the recycling-promoting Rab4 and Rab11 proteins. Our experiments in human cells showed increased lysosomal targeting with elevated IIS, with IIS reduction and Rab4/11 over-expression resulting in increased density of GJs.
The bilaterally symmetrical GFS (Fig 1A), a well-characterized, multicomponent neuronal circuit [43,44], mediates fast escape behavior by extension of the legs [45] followed by flight. It consists of 2 giant fiber (GF) interneurons that descend from the brain and synapse in the thoracic neuromere both with peripherally synapsing interneurons (PSI), which in turn synapse with dorsal longitudinal flight muscle motor neurons (DLMns) innervating the dorsal longitudinal flight muscles (DLMs), and with motor neurons (tergotrochanteral muscle motor neuron [TTMn]) innervating the tergotrochanteral (jump) muscles (TTMs). Electrical brain stimulation activates the GF interneurons, and the 2 output pathways can be monitored by recording from the 2 muscles (Fig 1A). Rapid conduction of nerve impulses through the escape response pathway has survival value [46]. Response latency, the time between the brain stimulus and muscle response, is a reliable measure of GF conduction velocity and circuit functionality [39,42,47].
To assess aging of the GFS, we measured response latencies in wild-type (WT; wDah) flies at different adult ages (old flies are defined as ≥45 days old). At day 5, the latencies were similar to those previously reported for young WT flies [39,48], and then increased, in both TTMs and DLMs (Fig 1B), indicating that aging slowed transmission in the circuit.
We then assessed whether reduced IIS affected transmission, by using da-GAL4 driver [49] to ubiquitously and constitutively express a dominant-negative (DN) form of insulin receptor (InRdn) [50], previously shown to extend fly lifespan [51]. Reduced IIS has also been shown to have a positive effect on an olfactory neuronal circuit and odor-related behavior in young flies [52]. The extension of response latency with age, seen in both control groups, was abolished by lowered IIS both in the TTM (Fig 1C) and DLM branch of the pathway (S1A Fig). We also assessed transmission through the TTMn-TTM and DLMn-DLM neuromuscular junctions by directly stimulating the GFS motoneurons of young and old flies. Response latencies of all genotypes were similar to each other and to those previously measured in WT flies [42,44] and were unaffected by fly age (S1B Fig); lowered IIS thus probably acted upstream of the motoneurons and neuromuscular junctions (NMJs) to improve electrical transmission in the aging GFS.
Prompted by these findings, we next asked whether the Drosophila insulin receptor (IR) is expressed in the GFS. Indeed, the antiserum previously shown to immunolabel IRs in abdominal neurons [53] detected strong IR presence in the GF interneurons (Fig 2A), allowing for the possibility that GFS is modulated by Drosophila insulin-like proteins. To identify mechanisms by which constitutively and ubiquitously reduced IIS maintained GFS function during aging, we assessed the density of electrical synapses, which are assembled from the GJ proteins encoded by SHAK-B, and which belong to the innexin family of transmembrane proteins [40,54,55]. In SHAK-B mutants, transmission in both the GF-TTM and GF-PSI-DLM pathways is disrupted [39,42]. SHAK-B proteins concentrate in 2 regions of the mesothoracic neuromere: the midline and 2 more posterior bilateral tracts [54] (Fig 2B). We quantified anti-SHAK-B staining intensity and area in isolated central nervous system preparations (that comprise the brain and ventral nerve cord), and found marked reductions in the bilateral tracts (Fig 2C) that label GJs at the GF-TTMn synapse (S2A Fig), potentially explaining the preservation of GFS functionality in da-GAL4/UAS-InRdn flies.
We also quantified choline acetyltransferase (a marker for cholinergic neurons) and a subunit (α7) of Drosophila nicotinic acetylcholine receptors (nAChRs) [56] in immunoblots from isolated central nervous system preparations and from heads but found no difference in immunoreactivity between the young and old IIS mutants or controls (S2B Fig), consistent with absence of structural changes at the level of cholinergic synapses with age. Together with previously mentioned physiological experiments [40–42], these results indicate that longer latencies in aging flies are a consequence of SHAK-B deficiency.
Constitutive and ubiquitous down-regulation of IIS could affect development of the nervous system and could also act from other tissues, systemically, to maintain GFS function during aging. To address these possibilities, we first confined lowered IIS to adult neurons using the inducible GeneSwitch (GS ELAV-GAL4) line [57] to drive expression of the DN IR [58]. Response latency did not differ between induced flies and controls at day 7 and increased with age in control but not induced flies (Fig 3A, S3A Fig). To assess the response to a physiologically relevant stimulus, we activated the GFS using a light air-puff directed to the fly’s head, transmitted to the GFs via mechanosensory afferents [59]. The mean frequency of the DLM responses within 5 seconds was significantly higher in mid-aged RU(+) flies (>70%), demonstrating an enhanced physiological outcome in response to lowered IIS in the nervous system (S3B Fig). Lowered IIS in neurons only during adulthood thus maintained GFS function during aging. If neuronal IIS impairs GFS function during aging, then increased IIS would be predicted to impair it further. Indeed, over-expression of WT IR in adult neurons significantly increased response latencies in mid-aged animals (Fig 3B).
To investigate a possible systemic role of lowered IIS in other parts of the nervous system in maintaining GFS function during aging, we drove expression of InRdn within a small subset of neurons including the GFS itself using the A307-GAL4 driver, which drives strongly in the GF interneurons and, to a lesser extent, the TTM and DLM motoneurons and PSI interneurons. This resulted in complete rescue of age-associated increase in response latency (Fig 4A, S4A Fig). We demonstrated the GF neuron-specific effect of lowered insulin signaling by using the recently described split-GAL4 line [60] that drives expression exclusively in the GF interneurons (S4B Fig).
IIS down-regulation in the GFS also caused a marked increase in the density of thoracic SHAK-B in old flies (Fig 4B and 4C). SHAK-B mutants display longer response latencies [54], while forced expression of SHAK-B(n+16), the isoform crucial for functional hemichannel formation in the GFS [61], prevented the age-related functional decline (Fig 4D, S4C Fig) in both branches of the circuit. These results further strengthen the hypothesis that loss of SHAK-B caused the age-associated loss of functionality within the GF circuit. The diameter of GF interneurons in old flies did not vary significantly between the genotypes (S5A Fig), suggesting that the size of principal circuit interneurons cannot account for the measured differences in conduction speed in these animals.
Because SHAK-B transcript levels were unaffected by lowered IIS (S5B Fig), we hypothesized that SHAK-B protein level was regulated by degradation, but saw no impact of lowered IIS on age-related loss of proteasomal proteolytic activity (S5C Fig). We therefore investigated the effect of reduced IIS on trafficking into lysosomes. To test this, we initially used confluent human retinal pigment epithelial (RPE1) cells, because the formation and trafficking of GJs can be studied in high temporal and spatial resolution and can be easily manipulated in this system. With persistently elevated IIS (the cells were maintained in insulin-supplemented complete medium throughout the experiments; S6A Fig), the cell surface levels of the main GJ protein connexin 43 (Cx43), but not that of the glucose transporter 1 (GLUT1) or integrin α3 (ITGA3), was low (Fig 5A, S6A–S6E Fig), consistent with the short half-life and rapid turnover of GJs in tissue culture models [62]. Lowering IIS by incubating cells in serum-free media containing glucose but no insulin for 13 hours (S6A Fig) or by using a dual inhibitor for the IR and IGF-1 receptor (IGF-1R) for 1 hour, increased the levels of GJs to 165 +/− 4.20 and to 171 +/− 8.13% (P < 0.05, P < 0.01), respectively (Fig 5A–5C). Upon reduced IIS, an acute (1 hour) stimulation with insulin (S6A Fig) was sufficient to decrease the levels of Cx43 down to that of cells subjected to constant elevated IIS or to acute activation of protein kinase C (PKC), a known mediator of Cx43 internalization [62] (Fig 5A–5C). Cx43 accumulated at the cell surface (labeled by ITGA3) under decreased IIS, but not upon acute insulin stimulation in which it was significantly shifted into lysosomal-associated membrane protein 1 (LAMP1)-positive lysosomes (Fig 5E). These phenotypes could have been achieved by a change in either endocytosis or endosomal recycling. Acute insulin stimulation did not change endocytosis and endosomal recycling of the transferrin receptor—which cycles constitutively between endosomes and the plasma membrane (Fig 5F, S6F Fig). However, elevating IIS induced targeting of Cx43 to lysosomes and degradation, which could be blocked upon inhibition of lysosomal activity by NH4Cl or bafilomycin A (BafA) treatments (Fig 5D and 5E) or reverted by over-expressing the small GTPases Rab4 or Rab11 (Fig 5G and 5H), which regulate recycling to the plasma membrane of synaptic receptors, gap junctional proteins, and ion channels in Drosophila [63–65] and mammals [66,67].
Over-expressing Rab5 or Rab7 (which regulate early endosomes or late endosomes-lysosomes formation, respectively) could not revert the phenotypes (Fig 5G and 5H). Consistent with these findings, over-expression of the constitutively active (CA) forms of Rab4 or Rab11, but not that of their DN forms, derouted Cx43 from the degradative to the recycling pathway and blocked degradation (Fig 5G and 5H), thus mimicking lowered IIS. Inhibiting lysosomal functions was sufficient to revert Cx43 levels but not plasma membrane localization upon elevated IIS (Fig 5D and 5E), further confirming that elevated IIS targets GJ proteins to lysosomes for their degradation. Cumulatively, these results show that reducing IIS enhanced the formation of GJs in human cells by stimulating their recycling to the plasma membrane, and decreasing their trafficking into lysosomes and subsequent degradation, consistent with the higher levels of SHAK-B protein in the IIS mutant flies.
Consistent with our in vitro results, over-expression of Rab4 or Rab11 in the Drosophila GFS led to increased SHAK-B density in the thorax of 40- to 45-day-old animals (Fig 6A and 6B). Furthermore, overexpression of WT or CA Rab4 or Rab11 completely blocked the age-related increase in response latency (Fig 6C, S7A Fig). Down-regulation of Rab4 and Rab11 by RNAi, on the other hand, increased response latencies even in young flies (Fig 6C, S7A Fig), consistent with the effect of Rab4 and Rab11 suppression on thoracic SHAK-B density in young adults (Fig 6D, S7B Fig). Thus, experimentally increasing the expression of these Rabs could rescue loss of GJs and of GFS function during aging, and their suppression could impair neurotransmission in young flies by reducing synaptic SHAK-B levels. Furthermore, Rab11 was required for reduced IIS to maintain response latency in old flies (Fig 6E, S7C Fig), supporting the idea that reduced IIS exerted its synaptic effects through the recycling-mediating Rabs.
A number of experimental results demonstrated the impact of long-term IIS manipulations on the nervous system. For example, systemic injections of IGF-1 mimicked some of the effects of exercise in the brain [68], and genetically reduced IGF-1 signaling in the whole organism reduced inflammation and neuronal loss in a mouse Alzheimer disease model [32]. Likewise, chronic IIS manipulations only in the nervous system can have consequences on the whole organism: attenuated IR substrate/IR substrate 2 signaling in aging brains promoted healthy metabolism and extended the lifespan in mice [69], and neuron-specific reduction of IIS increased longevity in Drosophila [70]. At the synaptic level, basal IGF-1 activity has recently been shown to regulate ongoing neuronal activity in hippocampal circuits [26]. While infusion of IGF-1 does not appear to have short-term influence on Cx43 levels in various regions of the rat brain [71], no study so far has examined the effect of chronic IIS manipulations in the aging nervous system on GJs.
In this work, we demonstrated a role for IIS in regulating the trafficking of gap junctional proteins that is conserved over the large evolutionary distance between Drosophila and humans, and between different cell types. Elevated IIS induces the targeting of GJ proteins to lysosomes and degradation, thereby decreasing their cell surface assembly (Fig 7).
Specifically, reduced insulin signaling throughout adulthood leads to Rab4/11-mediated increase in the synaptic targeting of SHAK-B-encoded gap junctional components in the Drosophila escape response circuit, resulting in the maintenance of the “youthful” functional output even in old flies. Previous studies demonstrated a positive effect of reduced insulin signaling on neuronal circuit function. For example, visual acuity in is improved in mice with reduced insulin signaling in the visual cortex [72]. In the nematode C. elegans, mutations of the IR gene resulted in improved chemical transmission at the neuromuscular synapse, and delayed decline in the synaptic function with age [73]. Our findings have revealed a novel restorative and adaptive cellular mechanism by which lowered IIS can maintain electrical transmission in a neuronal circuit during aging, and that could potentially be harnessed to prevent decline in neuronal function. A recent report [70] demonstrated a negative effect of neuron-specific IIS reduction on age-specific walking behavior in Drosophila, suggesting that the effect of insulin signaling depends on the type of neuron(s) mediating a specific behavior. For example, physiological roles of different (chemical) neuronal circuits can be preferentially mediated by either evoked or spontaneous transmission [74]. Interestingly, blockade of insulin signaling has opposing effects on these 2 types of transmission [26], possibly explaining some of the seemingly contradictory experimental data about the role of IIS in the nervous system. Together, these findings indicate that studies of insulin signaling in the nervous system should be circuit- and synapse type-specific, taking into consideration the physiological properties of the neuronal system under study, and precluding simplified generalizations about the effectiveness of specific IIS manipulations across the nervous system.
Ubiquitous and neuron-specific expression was achieved with the GAL4-dependent upstream activator sequence (GAL4-UAS) system [50]. Daughterless(da)-GAL4 flies (w1118; P13 [#8641]) were obtained from the Bloomington Drosophila Stock Center (BDSC); GS ELAV-GAL4 was derived from the original GS ELAV 301.2 line [57] and obtained as a generous gift from Dr. H. Tricoire (CNRS, Paris, France). The UAS-InRdn (BDSC #8252) transgene encodes an amino acid substitution in the kinase domain (K1409A) of the Drosophila IR (dInR), resulting in its DN activity [75]. The A307-GAL4 and UAS-SHAK-B(N+16) lines were a kind gift from Dr. P. Phelan (University of Kent, Canterbury, UK); the UAS-Rpn11 line was a gift from the lab of Dr. M. Miura (University of Tokyo, Tokyo, Japan), the split-GAL4 was a gift from G. Card (Janelia Farm, Ashburn, VA); these stock are a combination of 2 split-GAL4 halves; 1 has the activation domain of the GAL4 and the other has the DNA binding domain. Only the cells that express both halves will reconstitute a complete GAL4 [76]. Other BDSC stocks include the following: UAS-Rab4(WT) (#23269), UAS-Rab4(CA) (#23268), UAS-Rab4(RNAi) (#33757, TRiP), UAS-Rab11(WT) (~8506), UAS-Rab11(CA) (#9791), UAS-Rab11(RNAi) (#42709) and UAS-InR (#8262). To standardize genetic background, parental GAL4 and UAS strains used to generate experimental and control genotypes were back-crossed to laboratory control strain wDah (Wolbachia-infected) for at least 6 generations, beginning with an initial cross between wDah females and transgenic males, followed by 5 subsequent back-crosses between transgenic females and wDah males. The white Dahomey (wDah) stock was derived by incorporation of the w1118 mutation into the outbred Dahomey background by back-crossing. All stocks were maintained and all experiments were conducted at 25°C on a 12 hour to 12 hour light:dark cycle at constant humidity using standard sugar/yeast/agar (SYA) media (15gl−1 agar, 50 gl−1 sugar, 100 gl−1 autolyzed yeast, 100gl−1 nipagin, and 3ml l−1 propionic acid) [77]. In RU experiments, adult-onset neuronal expression was induced by adding mifepristone (RU486; Sigma -Aldrich, St. Louis, MO) to the standard SYA medium at 200 mM starting at day 1 post-eclosion. For all experiments, flies were reared at standard larval density and eclosing adults were collected over a 12-hour period. Flies were mated for 48 hours before separating females from males. Female flies were used in all experiments.
Recordings from the GFS of adult flies were performed as described by Allen et al. [78]; a method based on those described previously [44,79]. Flies were anaesthetized by cooling on ice and secured in wax placed inside a small Petri dish, ventral side down, with the wings held outwards in the wax to expose lateral and dorsal surfaces of the thorax. A tungsten earth wire served as a ground electrode and was placed in the abdominal cavity. Extracellular stimulation of the GF neurons was achieved by placing 2 electrolytically (NaOH) sharpened tungsten electrodes through the eyes and into the brain (the supra-oesophageal ganglion) to deliver a 40V pulse for 0.03 ms using a Grass S48 stimulator. Threshold for the short-latency, direct excitation for GF stimulation was previously demonstrated to be a 10 to 20 V pulse that lasts 0.03 ms [44,48]. We therefore applied pulses 2 to 3 times threshold to ensure that threshold was always exceeded.
Intracellular recordings were made following GF stimulation from the TTM and contralateral DLM muscle using glass micropipettes (resistance: 40–60 MΩ). The possibility that descending neurons other than the GFs might be simultaneously activated, leading to a possible TTM or DLM response, was previously excluded [78]. The electrodes were filled with 3M KCl and placed into the muscle fibers through the cuticle. Responses were amplified using Getting 5A amplifiers (Getting Instruments, San Diego, CA) and the data digitized using analogue-digital Digidata 1320 and Axoscope 9.0 software (Molecular Devices, Sunnyvale, CA). For response latency recordings, at least 5 single stimuli were given with a 5-second rest period between each stimulus; measurements were taken from the beginning of the stimulation artifact to the beginning of the EPSP (i.e., muscle depolarization). For direct activation of motoneurons (“thoracic stimulation”) [42], stimulating electrodes were removed from the brain and placed at the anterior end of the thorax through the cuticle and into the fused ganglia in the ventral region of the thorax. The signals were amplified and stored on a PC with pCLAMP software and a DMA interface board (Molecular Devices). Analysis was performed on the PC using pCLAMP and Microsoft Excel 2010 software (Microsoft, Seattle, WA).
Nervous systems (for the SHAK-B staining) were dissected in Drosophila saline [S3], fixed in 4% para-formaldehyde in phosphate buffered saline (PBS) for 30 minutes at room temperature and washed in PBS. After pre-incubation in blocking solution containing 4% normal goat serum in PBS + 0.5% Triton X-100 (PBT-X), preparations were incubated overnight at 4°C in primary antibodies diluted in blocking solution. Primary rabbit anti-SHAK-B antibody, raised against a C-terminal peptide common to all members of SHAK-B group of proteins (gift from P. Phelan), was used at 1:100. Preparations were washed 5 times in PBT-X and incubated with Alexa Fluor488-conjugated goat anti-rabbit secondary antibodies (Molecular Probes, 1:500) for 2 hours. Images were taken on Leica TCS SP2 inverted confocal microscope (Leica Microsystems GmbH, Wetzlar, Germany) or Zeiss 700 (Carl Zeiss MicroImaging GmbH, Jena, Germany). Three- to 5-day-old flies were used for immunostaining with anti-InR antibody (Cell signalling, #3024) at a concentration of 1:1000 followed by goat anti-rabbit dylight 649 secondary antibody at a 1:1000 concentration. In most cases, CNS preparations from animals with different genotypes were mounted on the same slide to control for possible variability in mounting procedure or properties of cover slips. For all preparations on the slide the same confocal settings were used (zoom, laser strength, PMT, gain, digital offset, averaging), and images represent projections of confocal z series composed of approximately 20 focal planes (slices) taken at approximately 1 μm steps at 400x magnification. Images were generated using the sum slices option in ImageJ (NIH, Bethesda, MD) that creates a real image that is the sum of pixel intensities in all focal planes. The abundance of GJs (labelled by anti-SHAK-B antibody) in the SHAK-B-positive bilateral tracts in the mesothoracic neuromere (T2) was quantified by drawing a line around the SHAK-B signal around each SHAK-B-labelled tract, and measuring the area and mean grey value (MGV). The area and MGV are first calculated for each tract separately; weighted mean MGV was then calculated for both tracts together and integrated density obtained as mentioned above.
The GF axons were injected in the connective with a dye solution of 10% w/v Neurobiotin (Vector Laboratories, Burlingame, CA) and tetramethyl rhodamine-labeled dextran (Invitrogen, Carlsbad, CA) in 2 M potassium acetate by passing depolarizing current, respectively. For GJ labelling, rabbit anti-SHAK-B (1:100) and goat anti-rabbit Dylight 649 (1:1000; Jackson ImmunoResearch Laboratories, West Grove, PA) was used. Confocal images were obtained using a Nikon A1 plus confocal with an Apo 60X oil lambdaS objective. Axon diameters of rhodamine-dextran and neurobiotin labelled GFs were measured in the first thoracic ganglion. Nikon Elements Advanced Research 4.4 Binary Editor was used to trace and 3D reconstruct the GF and ND Images Arithmetic function was used to extract anti-SHAK-B labelling that localizes to the 3D reconstructed GFs.
All western blots were run on 10% nongradient sodium dodecyl sulphate poly-acrylamide gel and proteins were transferred onto nitrocellulose membrane using a semi-dry blotter (Bio-rad). For ChAT (choline acetyltransferase) western blots, 10 CNS were dissected in PBS, boiled twice in Laemmli sample buffer and run on a 10% nongradient sodium dodecyl sulphate (SDS) polyacrylamide gel. Proteins were transferred onto nitrocellulose using a semidry blotter (Bio-rad) and probed with mouse monoclonal anti-ChAT 4B1 antibody (Developmental Studies Hybridoma Bank, Iowa City, Iowa) at [1:100] in 5% milk + TBST. Detection was performed with anti-mouse horseradish peroxidase-conjugated secondary antibody and Amersham ECL detection reagent (GE Healthcare, Little Chalfont, UK). Bands were normalized to actin.
Western blots for the recycling of membrane components and for Dα7 (nAChR) used fly heads, homogenized and boiled in Laemmli sample buffer (10μL/lane). Western blots were run as above. Primary Drosophila anti-Dα7 antibody concentration (incubation was performed overnight at 4°C) was 1:1000 [56]; the incubations were done in 3% BSA/TBS-T. Bands were normalized to actin, using (Abcam, Cambridge, UK) mouse anti-Actin [1:10000] in 5% milk/TBS-T. Secondary antibodies were diluted [1:10000] in 3% BSA/TBS-T, using either (Abcam) goat anti-mouse HRP (Abcam), or goat anti-rabbit HRP or anti-rat HRP (Sigma-Aldrich). Detection was performed with LuminataTM Crescendo or Forte (for Dα7) western blot HRP substrate (#WBLUR0500; Millipore, Billerica, MA) and imaged using Image Quant LAS4000.
RNA was extracted from 8 CNS, dissected in PBS, with 2 x 500 ul of TRIzol (Invitrogen) using a RiboLyser homogenizer and precipitated overnight at −20°C using 1 volume of isopropanol. Glycogen (Invitrogen) was included at 50 μg/ml to act as a carrier. The pellet was washed in 2 x 1 ml of 75% ethanol and re-suspended in DEPC water. RNA was treated with TURBO DNase (Ambion; Thermo Fisher Scientific, Waltham, MA) and reverse transcribed using Superscript II (Invitrogen) and Oligo dT. SHAK-B RNA was quantified with real-time PCR using primers (forward: (CAACGCACAACCAAAAAGG, reverse: GCGAAAAACAGGTGAATCG). Total RNA levels were compared after normalization to tubulin. The gene of choice for normalization was selected following a comparison of its expression level with actin and Tat using NormFinder.
Fly heads were homogenized in 25mM Tris, pH 7.5, and protein content determined by Bradford assay. Chymotrypsin-like peptidase activity of the proteasome was assayed using the fluorogenic peptide substrate Succinyl-Leu-Leu-Val-Tyr-amidomethylcoumarin (LLVY-AMC), based on a previously published protocol [80]. Twenty micrograms of crude fly head homogenate total protein was incubated at 37°C with 25 μM LLVY-AMC in a final volume of 200 μLs. Enzymatic kinetics were conducted in a temperature-controlled microplate fluorimeter (Infinite M200; Tecan, Männedorf, Switzerland), at excitation/emission wavelengths of 360/460 nm, measuring fluorescence every 2 minutes for 30 minutes. Proteasome activity was determined as the slope of AMC accumulation over time.
Human Rab4a, Rab4a N121I (DN mutant, called “Rab4 DN” in this study), Rab4a Q70L (CA mutant, called “Rab4 CA” in this study), Rab5a, Rab7a, Rab11a, Rab11a S25N (DN mutant, called “Rab11 DN” in this study), and Rab11a Q70L (CA mutant, called “Rab11 CA” in this study) genes were cloned into pEGFP-expressing vectors. Antibodies used for immunofluorescence analysis were rabbit polyclonal anti-Cx43 (called “Cx43” in this study; 3512; Cell Signaling Technologies, Danvers, MA); mouse monoclonal anti-integrin α-3 (called “ITGA” in this study; GTX11767; Genetex, Irvine, CA); rabbit polyclonal anti-GLUT1 (15309; Abcam); rabbit anti-TfR antibody (CBL47; Millipore); mouse anti-EEA1 (MAB8047; Bio-techne, Minneapolis, MN), mouse anti-LAMP1 (DSHB H4A3), and mouse and rabbit IgG isotype controls (DA1E and G3A1; Cell Signaling). Alexa Fluor 488-, 555-, and 633-conjugated donkey, anti-mouse, and anti-rabbit antibodies were from Life Technologies (Durham, NC).
Insulin (used at 1 μM for 1 hour, unless indicated [MP Biomedicals, Santa Ana, CA]), BMS 536924 (called IR inhib in this study and used at 5 μM [4774; Tocris, Bristol, UK]), phorbol 12-myristate 13-acetate (PMA; called PKC active in this study and used at 100 nM for 1 hour [10008014; Cayman Chemical, Ann Arbor, MI]), Gö6983 (called PKC inhib in this study and used at 5 μM for 1 hour [2285; Tocris]), NH4Cl (Sigma-Aldrich), BafA 1 (called BafA in this study and used at 1 μM [J61835; Alfa Aesar, Haverhill, MA), DAPI (Sigma-Aldrich), Alexa Fluor 488-conjugated transferrin (referred to as Tf,-A488 and used at 200 μg/ml [Life Technologies]), desferroxamine (Sigma-Aldrich), and holo-transferrin (Sigma-Aldrich).
RPE1 cells (ATCC CRL-4000) were maintained at 37°C, 5% CO2, in complete medium (DMEM:F12 HAM (1:1 v/v [D6421; Sigma-Aldrich]) supplemented with 10% fetal bovine serum (FBS [Life Technologies), 0.5% (w/v) sodium bicarbonate [Sigma-Aldrich], 2 mM GlutaMAX [Life Technologies], antibiotic-antimycotic [Sigma-Aldrich], and 20 μg/ml hygromycin). Cells were regularly tested for mycoplasma contamination. For all assays RPE1 cells were seeded in appropriate culture dishes (approximately 8 x 104, 2 x 105; cells were seeded in each well in a 96-well glass bottomed or 12-well plastic tissue culture dish, respectively) and grown as monolayers for 4 days in complete medium (containing insulin from FBS). Persistent elevated IIS (elevated IIS) was achieved by maintaining the cells thorough the experiment in complete medium. Persistent reduced IIS (reduced IIS) was achieved by incubating the cells in serum-free medium (containing glucose but no insulin) for 13 hours. Alternatively, cells were incubated in insulin-free medium (DMEM:F12 HAM (1:1 v/v) supplemented with 7.5 mM GlutaMAX (Life Technologies), antibiotic-antimycotic (Sigma-Aldrich), 1 μg/mL hydrocortisone (Sigma-Aldrich), 50 μg/mL ascorbic acid (Sigma-Aldrich), and 5 ng/mL basic FGF (Life Technologies) for 13 hours. IIS was acutely reduced by treating cells maintained in complete medium with IR and IGF-1R dual inhibitor for 1h (elevated IIS + IR inhib). Acute IIS (reduced IIS + insulin) was achieved by incubating the cells in serum-free medium (containing glucose but no insulin) for 12 hours, followed by addition of insulin (10 nM to 1 μM for 1 hour). In some experiments, cells grown under elevated IIS or reduced IIS conditions were treated with PKC activator (PMA, 100nM), PKC inhibitor (Gö6983, 5 μM), NH4Cl (50 mM) or BafA 1 (1 μM) with or without insulin for 1 hour.
All transfections were undertaken using GeneJuice (Millipore) transfection reagent according to the manufactures instructions in 96-well glass bottomed dishes at 100 ng of DNA per 96 well. Following 36 hours of transfection, cells were treated for experimentation and fixed, then prepared for analysis.
Treated cell monolayers were fixed in 4% paraformaldehyde (PFA) in PBS for 20 minutes at room temperature, followed by 2 washes with PBS and then incubated for 1 hour in 50 mM NH4Cl in PBS to quench the residual PFA. Cells were then surface-labeled with anti-ITGA3 followed by permeabilizaton with 0.1% Saponin-PBS and immuno-labeled in the presence of horse serum (Life Technologies) at an antibody dilution of 1:200 for all other primary antibodies, 1:1000 for all secondary antibodies, and 1:10,000 for DAPI staining of DNA. Fluorescently-labeled samples were imaged using either a high-throughput automated epifluorescence microscope (ImageXpress Micro XLS; Molecular Devices) or a confocal laser-scanning microscope (LSM700; Carl Zeiss MicroImaging).
Image analysis was performed using a protocol established in CellProfiler image analysis software [81]. A set of image analysis algorithms, or pipeline, was constructed to measure the properties of interest within the RPE1 cell culture labelled with DAPI, anti-Cx43 and/or anti-ITGA3, anti-EEA1, and anti-Lamp1, as required. Each image-set, corresponding to 1 field of view or site and comprising 3, or 4 where required, fluorescent channels, were analyzed independently using this pipeline. Twelve sites per well were analyzed and repeated in triplicate experiments.
In brief, an illumination correction function was calculated for each channel using a median filter (200 x 200 pixels) to correct for illumination variations across each 96-well plate. Each image set was then processed in an imaging pipeline as follows. The 4 channels' raw images were divided by their respective plate/channel illumination function. Firstly, segmentation of the nuclei of each cell in the field of view was identified corresponding to an arbitrary fluorescence intensity, median size (20 to 60 pixels) and shape (circular) according their DAPI DNA labelling. Next, cells were identified extending from the nuclei using an arbitrary fluorescence intensity corresponding to the red fluorescent channel. Identified cells were further partitioned, where appropriate, into transfected or nontransfected populations by their fluorescence associated with their GFP expression. Next, punctae corresponding to Cx43 and either LAMP1 or EEA1 were identified within the masked cellular region, with a typical diameter range of range of 8 to 20 pixels and having an arbitrary threshold of fluorescence intensity associated with their corresponding antibody labelling were identified as primary punctae objects for analysis. Finally, where appropriate, the surface of the cells was identified by labelling associated with anti-ITGA3 in nonpermeabilized samples, which were subsequently permeabilized and labelled with DAPI and Cx43. This ITGA3 labelling was designated as an object surface. The intensity, frequency, and size of all punctae were measured, and the ITGA3 labelled surface area was also measured. Cx43 punctae intersecting with early-endosomes and late-endosomes synapses were subsequently identified if 2 primary punctae objects in the case of early (Cx43/EEA1) or late (Cx43/LAMP1) were determined to colocalize in their respective fluorescent channels. The intensity and frequency for the colocalized object were measure for all channels. Cx43 objects localized to the surface were identified if Cx43 designated primary punctae objects intersected with ITGA3 labelled surface objects. All subsequent object data were partitioned into either transfected or nontransfected groups where appropriate. Levels of cellular Cx43 data were calculated using numbers of punctae per field of view normalized to cell number, and the intensity of the identified punctae. Levels of Cx43 localized to surface were normalized total number of Cx43 and their intensity, to surface area and the cell number in each of view. Levels of Cx43 localized to early-endosomes and late-endosomes were normalized to the total number of Cx43, and their intensity and the cell number in each field of view and expressed as a percentage of this value. Fractional results of the 3 subcellular compartments measured are expressed as a percentage of their values normalized to a total of 100%.
Data per well were determined by first aggregating the data of images taken within the same well for all sites and then over replicate wells and experiments. The average number of RPE1 cells per condition per experiment were >1,000. The median numbers of Cx43 punctae, EEA1 punctae, or LAMP1 punctae measured per condition per experiment were >12,000, >8,000, and >30,000, respectively.
Treated samples were then fixed and immunolabeled as outlined above. Epi-fluorescent, three-dimensionalprojections were created for each sample by acquiring a z-stack of images using a slice increment of 0.4 μm with 63 x oil objective. A three-dimensional reconstruction of the field of view was created using the image slices and Volocity image analysis software. Individual Cx43 GJ plaques from cells were selected automatically using a protocol established in Volocity image analysis software (Perkin Elmer, Waltham, MA). GJ plaques were identified as objects within a volume range of 5 to 100 μm3, according to an arbitrary threshold of fluorescence intensity associated with anti-Cx43 antibody labeling. Once individual objects (GJ plaques) were identified, their mean fluorescent signal and frequency of occurrence were quantified using Volocity software. These values were further partitioned, where appropriate, into those arising from transfected and non-transfected cells, determined by GFP. At least 10 fields of view from each condition were quantified for each sample with approximately 400 plaques being detected per field of view. Where possible, anti-ITGA3 or anti-Glut1 labeling determined the cell–cell interface and were utilized to assess Cx43, ITGA3, or Glut1 cell–cell interface localization through pixel–pixel colabeling.
Treated cell monolayers were detached using enzyme free cell-dissociation solution and resuspended in ice-cold PBS. Cells were further washed twice in ice cold PBS, and once in acid wash solution (150 mM NaCl, 100 mM Glycine, 5 mM Kcl, 1 mM CaCl2, pH4.5), as required. Cells were then fixed in 4% PFA on ice for 5 minutes followed by 20 minutes at room temperature. The fixation reaction was then quenched for 1 hour in the presence of 50 mM NH4Cl in PBS. Cells were washed twice in PBS and then immunolabeled as required with appropriate primary antibodies at a dilution of 1:200 in the presence of horse serum and 0.1% saponin in PBS (total staining) or just PBS (surface staining), as appropriate. Control cells were left unlabeled or labeled with an isotype control. This was followed by secondary antibody labeling with anti-rabbit Alexa Fluor 488 at a dilution of 1:1000. A total of over 10,000 events were analyzed per condition on a FACS LSRII flow cytometer (Becton Dickinson) for fluorescence intensities associated with Alexa fluor 488 labeling. Unlabeled and isotype control labeled cells were used to calibrate the instrument settings and to determine background fluorescence labeling respectively. Data were analyzed with FlowJo 8.6.3 (Tree Star Inc., Ashland, OR) software. Results were determined using fluorescence associated with the 488 nm laser excitation and resultant emission per cell and represent the geometric mean of each population.
Treated cell monolayers were incubated with 200 μg/ml Alexa Fluor 488-conjugated Tf at 37°C for up to 45 minutes (to fully saturate the endocytic network) in appropriate media supplemented with indicated reagents. Internalization was halted by chilling on ice, and cells were washed 3 times with ice-cold PBS to remove unbound Tf and once with mild acid wash buffer (150 mM NaCl, 100 mM Glycine, 5 mM KCl, 1 mM CaCl2, pH4.5) to remove surface-bound Tf. Cells were then incubated with prewarmed media (associated with the relevant conditions) containing 0.1 mM desferroxamine and 0.5 mg/ml unlabeled holo-Tf (Sigma-Aldrich) at 37°C for up to 30 minutes. At the indicated timepoints, cells were then incubated on ice, washed with ice-cold PBS, collected, fixed and processed for flow cytometry analysis. Fluorescent Tf remaining in cells at each time point were measured and the efflux rate was expressed as percentage of initial intracellular Tf measurement of resting cells after 7.5 minutes of efflux. Results were further normalised against the saturated level of intracellular Trf488 at time 0 after 45 minutes pre-incubation uptake. Results displayed comprise the fluorescence associated with Trf488 labeling per cell normalized as outlined and represent the geometric mean of each population.
Treated cell monolayers were collected using enzyme free cell dissociation solution, fixed and either immunolabelled in the presence of 0.1% Saponin-PBS (for total transferrin receptor [TfR]) or with PBS (for surface TfR) with anti-TfR antibody at a dilution of 1:200 at RT, or with isotype control under the same conditions. All samples were then secondary antibody labeled with Alexa fluor 488 and processed for surface and total TfR quantitation by flow cytometry. Results were determined from the ratio of surface TfR to total TfR using the geometric mean of each population. This value was used to normalize the results from associated samples in the Tf uptake assays.
Cells were prepared for the assay in the same way as above (surface and total TfR) and then incubated with 200 μg/ml Alexa Fluor 488-conjugated Tf at 37°C for up to 10 minutes in appropriate media supplemented with the indicated reagents. Internalization of Tf was halted by putting the cells on ice and washing immediately with ice-cold PBS. Surface bound Tf was further removed by mild acid washing followed by 3 ice-cold PBS washes. Cells were then detached with enzyme-free cell dissociation solution, fixed and processed for flow cytometry as outlined above. Results were normalized against the ratio of surface TfR to total TfR and control cells and represent the geometric mean of each population.
Statistical analyses were performed using GraphPad Prism 5 software (GraphPad Software Inc., La Jolla, CA). A 2-way ANOVA test was used to perform (age x genotype) interaction calculations. For other comparisons between 2 or more groups, a 1-way ANOVA followed by a Tukey-Kramer post hoc test was used. In all instances, P < 0.05 is considered to be statistically significant (*P < 0.05; **P < 0.01; ***P < 0.001). Log-rank tests were performed for survival. All data were tested for Gaussian distribution with Kolmogorov-Smirnov test with the Dallal-Wilkinson-Lillie for corrected P value. In case of Gaussian distribution, the following parametric tests were used: Student t test (2 groups) or 1-way ANOVA and Dunnett test (2+ groups), as appropriate. All values are reported as the mean ± SEM.
|
10.1371/journal.ppat.1003355 | Hepatitis C Virus p7 is Critical for Capsid Assembly and Envelopment | Hepatitis C virus (HCV) p7 is a membrane-associated ion channel protein crucial for virus production. To analyze how p7 contributes to this process, we dissected HCV morphogenesis into sub-steps including recruitment of HCV core to lipid droplets (LD), virus capsid assembly, unloading of core protein from LDs and subsequent membrane envelopment of capsids. Interestingly, we observed accumulation of slowly sedimenting capsid-like structures lacking the viral envelope in cells transfected with HCV p7 mutant genomes which possess a defect in virion production. Concomitantly, core protein was enriched at the surface of LDs. This indicates a defect in core/capsid unloading from LDs and subsequent membrane envelopment rather than defective trafficking of core to this cellular organelle. Protease and ribonuclease digestion protection assays, rate zonal centrifugation and native, two dimensional gel electrophoresis revealed increased amounts of high-order, non-enveloped core protein complexes unable to protect viral RNA in cells transfected with p7 mutant genomes. These results suggest accumulation of capsid assembly intermediates that had not yet completely incorporated viral RNA in the absence of functional p7. Thus, functional p7 is necessary for the final steps of capsid assembly as well as for capsid envelopment. These results support a model where capsid assembly is linked with membrane envelopment of nascent RNA-containing core protein multimers, a process coordinated by p7. In summary, we provide novel insights into the sequence of HCV assembly events and essential functions of p7.
| Viroporins are small hydrophobic viral membrane proteins which oligomerize and modulate membrane properties to facilitate virus propagation. Within their membrane environment these proteins can form membrane pores or channels which change the permeability of membranes for ions. These properties are known to contribute to release of infectious enveloped virus particles from infected cells and/or to facilitate viral cell entry by catalyzing virus uncoating. In case of HCV, p7 function is essential for production of infectious progeny and its ion channel activity is well documented in vitro and in cell-based systems. Recent evidence indicated that p7 channel activity dissipates the low pH of the cellular secretory compartment thus protecting the viral glycoproteins from low pH induced misfolding and inactivation. In this investigation we highlight the involvement of the p7 ion channel in the assembly and envelopment of viral RNA-containing capsids. Our results indicate that p7, likely in concert with the viral envelope proteins, comprises a membrane-bound recipient complex that provides a scaffold to initiate unloading of core protein from lipid droplets for capsid assembly and membrane envelopment. Collectively, these findings highlight novel facets of p7 function in the course of HCV morphogenesis.
| Approximately 160 million people are chronically infected with HCV [1] and viral persistence is associated with severe liver diseases such as steatosis, cirrhosis and hepatocellular carcinoma [2]. HCV is an enveloped virus with a positive-strand RNA genome that encodes a polyprotein of about 3,000 amino acids. The polyprotein is cleaved co- and post-translationally by cellular and viral proteases into 10 mature proteins. Briefly, the structural proteins are resident in the N-terminal portion of the polyprotein and include the core protein and the envelope glycoproteins E1 and E2.The nonstructural proteins NS3, 4A, 4B, 5A and 5B are encoded on the C-terminal part of the polyprotein, and are components of the viral replicase complex [3]. Finally, the p7 and NS2 proteins reside in between structural proteins and NS3-5B. Both are dispensable for replication but necessary for the production of progeny viruses [4], [5], [6].
The pathways of HCV morphogenesis including capsid assembly, particle envelopment and virus egress are incompletely understood. Moreover, there is limited information about the structure and composition of HCV particles as well as the precise association of the viral structural proteins and accessory factors during virus production. However, recent work mostly based on the JFH1-based HCV infection model [7], [8], [9] has highlighted some unique features of the late steps of the HCV life cycle. Firstly, lipid droplets, cellular lipid storage organelles, have been recognized to be essential for production of infectious HCV progeny [10]. Core protein resides on the surface of these lipid storage organelles [11], [12] and during virus assembly recruits viral proteins and RNA, which is an essential prerequisite for virus production [10]. Core protein itself is loaded onto these lipid bodies through an interaction with diacylglycerol acyltransferase-1 (DGAT-1) [13], a host enzyme which catalyzes the final step in the biosynthesis of triglycerides that is essential for lipid droplet biogenesis [14]. Moreover, host factors crucial for the biosynthesis and secretion of human lipoproteins have emerged as cofactors for HCV morphogenesis. These include apolipoprotein B (ApoB), apolipoprotein E (ApoE) and microsomal triglyceride transfer protein (MTTP) [15], [16], [17]. Presumably as a consequence of the tight coupling of HCV assembly with lipoprotein biogenesis, infectious HCV is thought to circulate as so called “lipo-viro particle”, rich in cholesteryl esters and comprising viral proteins, ApoB and ApoE [18], [19], [20]. Finally, it is recognized that besides the viral structural proteins (core, E1, E2) and p7, essentially all NS proteins contribute to production of infectious viral progeny [4], [5], [21], [22], [23], [24], [25], [26]. Yet, how each of these proteins contributes to infectious HCV particle assembly and release is currently poorly defined.
It is thought that HCV core proteins form the viral capsid which encases the viral RNA. Monomeric, mature core protein has a molecular weight of 21 kDa and exerts dual affinity for RNA and for lipids. Based on a hydropathicity profile, the N-terminal domain 1 (D1) is hydrophobic and extends to residue 117 [27]. It mediates core protein homo-oligomerization [28], [29], [30], [31] as well as its interaction with the viral RNA [32], [33], [34]. The latter seems to be poorly sequence-specific and to depend primarily on the presence of structured RNA [29], [30], [35]. Domain 2 (D2, ca. residues 118–174 [27]) is on the contrary hydrophobic and composed of two amphipathic α-helices connected by a hydrophobic loop [36]. This domain is responsible for the localization and anchorage of the mature core protein at the surface of lipid droplets [37]. Finally, domain 3 (D3) is composed of the C-terminal portion of core and acts as a signal sequence for insertion of E1 into the ER lumen. Notably, an interaction between core and the E1 envelope protein has been proposed [38], [39]. The envelope shell harbors the E1 and E2 glycoproteins, whose heterodimers appear to be crosslinked by disulfide bridges [40]. The NS5A protein, more specifically the C-terminal domain III, is also essential for production of infectious HCV [25], [41], [42]. Interestingly, phosphorylation of specific serine residues within this domain of the protein has been implicated in virus production, suggesting a model where NS5A phosphorylation governs replication and assembly [42]. Moreover, accumulating evidence from several independent studies supports a key role of NS2 in coordinating assembly by interacting with the glycoprotein complex, but also with p7, NS3 and NS5A thus bringing together the different viral components for envelopment [43], [44], [45], [46]. While NS2 function has been studied extensively the precise role of p7 in this process remains unclear.
P7 is a small membrane protein of 63 aa with both termini oriented towards the ER lumen. It contains an N-terminal α-helix followed by two transmembrane (TM) domains connected by a cytosolic loop (aa 33–39) [47]. Importantly, p7 monomers can assemble into hexamers or heptamers [48], [49], [50] and thereby form cation-selective ion channels in planar lipid bilayers, thus classifying p7 as a viroporin [48], [51], [52], [53]. A fully conserved di-basic motif within the cytoplasmic loop (residues K33 and R35 for genotype 2a J6) is important for the channel function in artificial membranes as well as for virus assembly and release in cell culture [4], [48], [54]. Finally, the essential role of p7 was also confirmed in vivo [55]. Notably, recent evidence suggested that p7 acts as a proton channel in the secretory compartment of the cells and dissipates the low pH in these compartments to protect low pH-sensitive nascent particles and envelope proteins [54]. However, it was unclear if this was the sole function of p7 during HCV assembly. Therefore, in this study we developed a set of complementary biochemical and cell biological methods to dissect the individual steps involved in HCV morphogenesis in order to shed further light on the specific function(s) of p7 in the course of infectious virion morphogenesis.
To investigate the precise function of p7 during HCV morphogenesis, several p7 mutants with published defects in virus production were utilized, among them a point mutant of the dibasic motif in the loop region (p7-KR33/35QQ, or p7-QQ [4]) abrogating ion channel activity, and a more severely impaired deletion mutant lacking the predicted p7 N-teminal transmembrane helix (p7 residues 1 to 32) (Δp7half, [4]). For comparison, various genomes with mutations of one or more structural proteins were characterized in parallel. These included a mutant carrying four consecutive alanine residues in place of amino acids 69 to 72 of J6CF-core (core-C69-72A [56]), and a construct with a mutation in the transmembrane region of E1, known to disturb E1–E2 heterodimerization and virus production (E1-K179Q [4], [57]). Finally, we also created a mutant which lacks the entire coding region of E1 and E2 (ΔE1E2) (Figure 1A). Note that all mutants were generated in the context of the highly infectious Jc1 chimeric HCV genome [58]. To verify and compare the influence of these mutations on HCV assembly and release of infectious virions, extra- and intracellular virus titers in HCV-transfected cells were determined with a limiting dilution assay (Figure 1B). While the point mutations similarly reduced secreted virus titers ca. 100-fold, virus production was completely abrogated for the two deletion constructs. Furthermore, the parallel decrease in intracellular infectivity strongly argues for a specific impairment of HCV assembly rather than release. Only for the p7-QQ mutant, extracellular infectivity was slightly more reduced compared to intracellular infectivity, suggesting the possibility of an additional block in virus secretion. To rule out that the changes observed were indirectly caused by an effect of the mutations on HCV RNA replication, translation or specific infectivity of virus particles, extra- and intracellular levels of core protein were determined by ELISA (Figure 1C). As expected, we observed comparable intracellular core amounts for all virus variants indicating that the assembly defects were not due to disturbed RNA replication or translation (Figure 1C). For most mutants the decrease in extracellular infectivity correlated with a reduction in extracellular core levels. Therefore, the decreased extracellular infectivity is attributable to reduced assembly of infectious virions rather than secretion of poorly infectious, aberrant particles. Notably, this was not the case for the E1-K179Q mutant virus, for which core release was comparable to WT Jc1 despite a ∼100-fold reduction in extracellular infectivity titers (Figure 1B and C). This data suggests the secretion of non- or poorly infectious particles in case of this mutant. In conclusion, with the exception of E1-K179Q, all mutants clearly showed defects in HCV assembly with downstream effects on the number of particles released and the total intracellular and secreted infectivity. Moreover, the residual titers were comparable among point and deletion constructs, facilitating the comparison between these groups of mutants.
To pinpoint the specific defect(s) in HCV morphogenesis caused by these mutations, we dissected the assembly process into distinct steps such as core trafficking to the lipid droplets (LDs), core oligomerization, capsid formation, and, finally, envelopment. First of all, core protein resistance to proteinase K digestion was assessed as a measure of the degree of envelopment of the protein into membranes, which is expected to protect the protein from proteolytic digestion. Briefly, cells transfected with mutant or wild-type (WT) Jc1 were lysed mechanically and treated with proteinase K. Residual core protein was quantified by immunoblotting or ELISA and normalized to the total amount of core protein in the untreated sample. As a control, cell lysates were pretreated with Triton X-100 to solubilize all membranes before proteinase K digestion. Under these latter circumstances, all core proteins are expected to be sensitive to proteolytic digestion, ensuring that the proteinase K concentration was not limiting in the assay. The core amount detected under these conditions was used for background subtraction. A representative immunoblot stained for HCV core is shown in Figure 2A. In the context of WT-Jc1, quantification of signal intensities revealed that about 60% of intracellular core protein was resistant to protease digestion and thus, had already acquired a membrane envelope and proceeded to a post-budding step (Figure 2B). In contrast, all mutants with the exception of E1-K179Q, which was not significant (p = 0.084, Welch's t-test), showed a highly significant (p<0.01, Welch's t-test) reduction in amount of protected core, indicating that most core species present were digested. Together, these data suggest that the mutations/deletions under investigation confer a defect in infectious particle production occurring at, or prior to, core envelopment.
With the reduced amounts of membrane-protected core observed, we next analyzed if this was due to a defect in capsid envelopment itself or due to blockade of events prior to this. To address this question, HCV-transfected-cells were lysed by multiple cycles of freeze and thaw. Subsequently, post-nuclear lysates were separated by rate zonal centrifugation in the absence of detergent. The core protein content of each fraction was determined by ELISA and normalized to the total core amount in the lysate. While there was a single prominent peak of rapidly sedimenting core in fraction 7 with a refraction index of 1.360 for WT-Jc1 and the E1-K179Q mutant (Figure 3), the other mutants clearly showed a second peak, sedimenting more slowly at fraction 4 with a refraction index of around 1.345. In fact, in the case of the deletion constructs and p7-QQ, core protein species with reduced mobility were the predominant species. Notably, expressing core protein from a replicon RNA lacking E1, E2, p7 and NS2 caused a similar accumulation of core protein in fraction 4 (Supplementary Figure S1). In order to further characterize different core protein species, including those forms with slower-sedimentation, HCV infectivity and RNA content were determined in each gradient fraction. Interestingly, the core peak in fraction 7 which is predominant in the case of WT-Jc1 was highly infectious (Figure 4A). In contrast, fraction 4, which displays the accumulation of core protein that is characteristic for the p7 mutants and the deletion constructs (Figure 3 and Figure 4B) contained relatively little infectivity: 100-fold reduction compared to fraction 7. (Figure 4A). Since fraction 4 of the WT-Jc1 preparation contained only 3-fold less core protein than fraction 7 (Figure 4B) and as most of the viral RNA sediments in fraction 4 (Figure 4C), it is unlikely that the dramatic reduction of infectivity is attributable to lack of viral RNA or core protein. Of note, the sedimentation of viral RNA in the gradient was independent of core protein expression, since distribution of viral RNA after transfection of a Jc1 mutant lacking most of the core coding region (Jc1- Δcore) was comparable to Jc1-WT transfected cells (Supplementary Figure S2). Given these findings we suspected that core protein species sedimenting in fraction 4 may represent incomplete, non-enveloped, particles. In contrast, fraction 7 carries a substantial amount of infectious particles. Notably, the fraction with highest infectivity in the sedimentation was fraction 6. This suggests that either fraction 7 besides infectious virions carries an excess of non-infectious viral RNA and core compared to fraction 6 or that some HCV particles may “mature” causing differential sedimentation (fraction 6) and higher infectiousness. To address the observed differences between fractions 4 and 7, we subjected these fractions to proteolytic digestion by protease K treatment. Interestingly, only approximately 10–20% of core protein was protected from proteolysis in fraction 4 whereas about 50–60% of core protein was protease-resistant in fraction 7, giving a highly significant difference in the two fractions (p = 3.909e-5 over all mutants, Figure 4D). Notably, the degree of core protein protection specific to fraction 4 and 7 was similar between WT-Jc1, the E1 and the p7 mutant, indicating that all three genomes yield comparable core protein structures – albeit with different efficiency. Moreover, specifically the E1 mutant displayed very low infectivity throughout all fractions (Figure 4A), which may be due to a defect of particles carrying this mutation in cell entry. Altogether, these results indicate that fully-enveloped infectious intracellular HCV particles preferentially sedimented in fraction 7 and that the p7 mutant is heavily impaired in producing core protein species with these sedimentation properties. Strikingly, supplementation of the rate zonal gradient with detergent to strip core protein from lipid envelope led to a shift in core sedimentation of WT-Jc1 from fraction 7 to fractions 3 in the low density range of the gradient (Figure 4E). These results indicate that non-enveloped capsids and/or oligomeric core protein liberated after detergent treatment preferentially sediment in fractions 3of the detergent-carrying gradient. Therefore, mutations/deletions in p7 can increase accumulation of non-enveloped capsids and/or oligomeric core proteins which sediment in fraction 4 of our rate zonal gradient (in the absence of detergent). It is possible that these structures may be similar to true capsids released from intracellular WT-Jc1 particles which sediment in fraction 3 of the detergent gradient.
To further investigate the natural course of capsid envelopment, HCV transfected cells were collected at 12, 24 and 48 h after transfection and subjected to rate zonal centrifugation as described above (Figure 5A). Interestingly, at an early time point (12 h post-transfection), the sedimentation profile of core protein was identical for WT-Jc1 and the p7 mutants, with a peak at fraction 4, similar to the one observed 48 h post-transfection for the p7 mutants (Figure 3 and Figure 5A). While this peak remained unchanged over time for the Δp7half mutant, a fraction of core protein displayed increased mobility as early as 24 h post-transfection and sedimented in fraction 7 in the case of WT-Jc1. Finally, at 48 h almost all core protein in the WT-Jc1 transfected cells sedimented in fraction 7. An intermediate phenotype was observed for the p7-QQ mutant, which exhibited a delayed and incomplete shift in core sedimentation profile. This change in mass and/or size of core complexes over time, evidenced here by different rate zonal centrifugation profiles, likely reflected capsid envelopment, a process apparently impaired or abrogated respectively for the p7-QQ and Δp7half mutants. Therefore, early during assembly, capsid envelopment and possibly budding represent a rate-limiting step that involves p7. We next investigated whether this p7 function could be complemented in trans and thus, transfected Jc1 Δp7half RNA into a packaging cell line expressing the core to NS2 region [59]. Strikingly, transfection of the p7 mutant into the packaging cell partially restored infectivity (data not shown). Concomitantly, the core sedimentation profile was partially shifted back to enveloped core species with rapid sedimentation (Figure 5B) indicating that the defect in membrane envelopment could be at least partially rescued in trans.
All assembly mutants tested, with the exception of the E1-K179Q mutant, showed an abnormal accumulation of non-enveloped capsids or core oligomers (Figure 3). To further explore the molecular mechanisms underlying this defect, we next compared the subcellular localization of core protein between Jc1-WT, E1-K179Q, p7-QQ and the core-C69-72A mutants by immunofluorescence staining using core-specific antibodies and co-staining of lipid droplets by the lipid dye BODIPY (Figure 6A). As published previously in the context of WT-Jc1, core protein demonstrated a reticular staining pattern with relatively little LD association [60], [61]. Moreover, LDs were well dispersed throughout the cell. The same phenotype was observed for E1-K179Q (Figure 6A). In contrast, accumulated core protein was located at the surface of clustered lipid droplets in the case of p7-QQ and core-C69-72A mutants (Figure 6A). To quantify these differential phenotypes, we analyzed between 100 and 200 cells for each mutant and assigned HCV expressing cells into two categories characterized by elevated LD clustering (i.e. aggregation of at least 10 LDs in close proximity to each other) or increased LD association of core protein (circular core staining around the surface of LDs; Figure 6B). In the case of WT-Jc1 and the E1-K179Q mutant, only about 10% of cells showed LD clustering and even less cells displayed core staining around the entire surface of LDs. Strikingly, the frequency of both phenotypes was strongly increased for the p7-QQ and the core-C69-72A mutants with more than 90% of p7-QQ-transfected cells being positive for both (Figure 6B). To further confirm these observations, lipid droplets were isolated from cell lysates by flotation through a discontinuous density gradient. Enrichment of ADRP (a LD marker) and exclusion of calreticulin (a typical ER-resident protein) from the LD fraction confirmed the efficiency of the cell fractionation (data not shown). The amount of core protein associated with the LD fraction was determined by ELISA and normalized to the total core content, with the value obtained for WT-Jc1 set as 1 (Figure 6C). An over 30-fold increase in the amount of LD-associated core could be observed for the p7 and the core mutants whereas for E1-K179Q only ca. 10-fold more core protein was detected at LDs. Therefore, the enrichment of non-enveloped capsids for the p7 mutants (and the core-C69-72A mutant) correlates with an accumulation of core at the surface of lipid droplets. These findings strongly argue that these mutations cause the retention of core protein and possibly core oligomers and higher order capsid-like structures at those organelles.
To further investigate if the observed defects in capsid envelopment were caused by prior alterations in capsid assembly, core complexes from HCV-transfected cell lysates were separated by rate zonal centrifugation in the presence of detergent. As observed before, the dominant core species from WT-Jc1 transfected, detergent treated cells sedimented in fraction 3 (Figure 7A). These core proteins likely represent mostly assembled capsids since the highly infectious enveloped core species from fraction 7 were converted to this sedimentation behavior by detergent treatment (Figure 4E). An indistinguishable profile was observed for the core-C69-72A mutant indicating that its inefficient envelopment was not caused by a defect in capsid assembly, as far as it can be resolved with this method (Welch's t-test against WT, Fraction 5: p = 0.3046, no significant difference). In contrast, a moderate, but significant (see right panel of Figure 7A) enrichment of slightly faster sedimenting core complexes was detected for both p7 mutants (fraction 5: p7QQ: p = 0.01556, Δp7half: p = 0.0417).
To further characterize these putative capsid assembly intermediates (or aberrant capsid species), we tested their ability to protect the viral RNA from RNase digestion (Figure 7B). To this end, gradient fractions 3 (corresponding to the core peak obtained with the WT-Jc1 construct), 4 and 5 (specifically enriched in the p7 mutants) were subjected to RNase digestion and residual viral RNA was quantified by qRT-PCR. To control the efficiency of the RNase digestion, RNA-protecting capsids or core oligomers (or other proteins) were removed by proteinase K pretreatment, before RNase addition. Strikingly, the viral RNA in fraction 3 was mostly protected from RNase digestion, unless the samples were treated with proteinase K beforehand. In contrast, RNase digestion strongly reduced HCV RNA levels in fraction 5, and pretreatment with proteinase K only marginally increased RNA degradation. Comparable results were also obtained for p7-QQ, Δp7half and core-C69-72A mutants (Supplementary Figure S3). Notably, HCV RNA contained in crude replication complexes prepared from Jc1-WT transfected cells according to the protocol by Quinkert et al. [62] was almost completely protected from RNase digestion even when protease K was added (Figure 7B). However, when 1% DDM was added to these CRCs to mimic conditions of our rate zonal gradient, addition of RNase A alone was sufficient to degrade more than 99.9% of viral RNA. Collectively, these results argued that a large proportion of RNA complexes sedimenting in fraction 3 in the presence of 1% DDM are protected from ribonuclease digestion unless protease K was added. Since HCV RNA in CRCs does not share this property and is readily degraded even in the absence of protease, we believe it is unlikely that the RNA structures in fraction 3 predominantly represent CRCs. In contrast, due to the high abundance of core protein in fraction 3 (Figure 7A) and the observation that enveloped capsids are shifted to this fraction 3 (Figure 4E) we believe that the viral RNA in fraction 3 may be encased in a capsid-like structure thus protecting it RNase digestion. On the contrary, core complexes sedimenting in fraction 5 and enriched in mutant-transfected cells, might be loosely associated with the viral RNA and form incomplete, loose or unstable capsid structures that do not fully protect the RNA. Notably, we cannot dismiss the alternative interpretation that other proteins differentially sedimenting in fractions 3, 4 and 5 may cause a differential protection of the viral RNA to ribonuclease digestion.
Finally, in order to gain further insights about the size and oligomerization status of the various core complexes observed, lysates of HCV-transfected cells normalized for equal levels of HCV core were separated by native PAGE using 4-16% gradient gels with a resolution from 15 to 1,000 kDa. Strikingly, a large amount of core protein was detected as part of high-order complexes between 480 and 1,048 kDa for the p7-QQ mutant (Figure 7C) whereas for Jc1-WT and the core-C69-72A mutant, little core protein signal was visible. Possibly, most of the core protein could not enter the gel for WT-Jc1 and the core-C69-72A mutant, indicating that it belonged to large structures exceeding 1,000 kDa, presumably the fully-assembled capsids. Alternatively, the antibody used for detection (C7-50), may not recognize the predominant core species for these constructs. To circumvent this problem, two-dimensional gel electrophoresis was performed with native polyacrylamide gel electrophoresis (PAGE) in the first dimension and denaturing, reducing SDS-PAGE in the second. With this setup, all core complexes are dissociated into monomers during the second dimension of the electrophoresis thus avoiding possible problems caused by masking of the antibody epitope in native core protein complexes. Notably, the mobility of core in the gel reflects oligomerization status since the first dimension of the gel is resolved under native conditions. A 3–12% gradient gel was used in the first dimension to further broaden the resolution of the assay. Subsequent immunoblotting of core essentially confirmed the data obtained from the native PAGE (Figure 7D). Due to the inherent variability of the assay, three independent experiments are shown side by side for comparison. A variety of core complexes, ranging from monomeric to high-order structures, could be detected for WT-Jc1, however, the most prominent species exceeded the resolution of the gel and, thus, probably corresponded to fully-assembled capsids. In contrast, an enrichment of diverse intermediate high-order core complexes could be seen for the p7-QQ mutant.
In summary, the p7 mutants showed an accumulation of putative capsid assembly intermediates which may not have fully encased the viral RNA, while contrastingly the core C69-72A mutation interfered specifically with a downstream envelopment and/or budding event. Taken together, our results suggest that the presence of functional p7 is required for a late step of capsid assembly as well as capsid envelopment suggesting a tight linkage between these two processes.
Accumulating evidence supports the crucial role of non-structural proteins in HCV assembly and release [63] and in particular the essential part played by the p7 viroporin [4], [5]. Here, we unraveled the precise role of p7 in production of HCV progeny virions by characterizing the defective viral particle production of two p7 mutants, a double mutant impaired in its ion channel activity (p7-KR33,35QQ) [4], [48], and a more extreme deletion mutant lacking the first transmembrane domain (Δp7half) [4]. These mutants were studied in the context of the genotype 2a chimera Jc1, chosen for its high assembly efficacy [58], [60], allowing us to biochemically dissect the sequence of events involved in this process and to better resolve potential points of interference.
In a Jc1 background, we observed a ca. 60-fold reduction, or a complete block in infectious particle production, respectively for the p7-QQ and Δp7half mutants (Figure 1B). Intracellular core levels of these mutants were unchanged (Figure 1C), ruling out gross RNA replication defects for both mutants. In case of p7-QQ the specific infectivity of released particles was comparable to WT-Jc1 thus ruling out impaired cell entry. Production of intracellular infectivity of the p7-QQ mutant was also impaired, although not quite to the same extent as for the generation of extracellular infectious particles which may point to a potential role of the ion channel function during virus secretion [4]. Importantly, Wozniak et al. reported that p7 could prevent the acidification of intracellular vesicles and, thus, might protect pH-sensitive immature intracellular virions from premature fusion during release [54]. While in this study an ion channel-defective mutant could be rescued by treatment with pharmacological inhibitors of intravesicular acidification, this was not the case for p7 deletion constructs [54] highlighting additional functions of p7 independent of its ion channel activity. In accordance with this, we observed a significant reduction of protease-resistant core protein in cells transfected with p7 mutants as compared to WT-Jc1 (Figure 2), clearly showing that virus production is arrested or impaired at a step prior to envelopment and capsid budding into the ER. Rate zonal centrifugation of detergent-free cell lysates revealed an accumulation of slow sedimenting, non-infectious, non-enveloped core complexes in the case of the p7 mutants (Figures 3 and 4). Interestingly, a similar profile was also obtained in the case of WT-Jc1 but only early (12 h) after transfection (Figure 5), suggesting that these structures were naturally occurring assembly intermediates that, in case of a non-functional p7, could not complete their maturation. These species disappeared 24 h post-transfection for the WT-Jc1 in favor of faster sedimenting enveloped infectious particles, indicating that capsid envelopment might be a rate limiting step early during assembly. On the contrary, time dependent emergence of these latter, enveloped, highly infectious core structures was abrogated, or delayed and incomplete for the Δp7half and p7-QQ mutants, respectively. Finally, rate zonal centrifugation of Jc1-transfected cell lysates in presence of detergent also showed slow sedimenting core species (Figure 4E, fraction 3), confirming that p7 mutants accumulate core structures resembling naked capsids. Interestingly, the production of enveloped virus could be partially rescued for the Δp7half mutant when providing functional p7 in trans leading to the two-peak phenotype observed for the p7-QQ mutant (Figure 5B).
Secondly, the accumulation of non-enveloped core protein structures for p7 mutants correlated with an enrichment of core protein at the surface of lipid droplets as shown by immunofluorescence analysis and cellular fractionation (Figure 6). These results suggested that, in absence of functional p7, capsid assembly takes place at lipid droplets but recruitment back to the ER and subsequent budding are disturbed. These observations confirm and complement recent results obtained by Boson et al. who reported that p7 expression alone induces a redistribution of core from lipid droplets to the ER [61]. This was modulated by NS2 in a genotype-dependent manner and unrelated to p7 ion channel function. However, the authors only expressed individual HCV proteins and did not investigate p7 mutants in the context of virus infection where a more complex interaction pattern between viral proteins likely occurs. Notably, several independent groups have recently shown that NS2 establishes a number of protein-protein interactions required for HCV assembly [23], [43], [44], [45], [46]. A complex consisting of NS2, NS3, NS5A and the E1–E2 envelope glycoproteins was shown to form at the ER in proximity to the LDs [45]. Although data for the localization of p7 in infected cells are missing, the polypeptide is able to interact with NS2 [43], [45] and could therefore belong to the same complex. Since NS5A interacts with core at the surface of lipid droplets [10], [25], [41], it might recruit the whole complex to assembling capsids, thereby facilitating envelopment. While p7 was only assigned a subsidiary role of regulating NS2 protein interactions [45], [46] possibly by modulating its topology [44] or stability [46], the results of Boson et al. clearly show that it affects core localization independently of NS2 indicating a putative direct or indirect core-p7 interaction between the two proteins and additional functions of p7 during assembly.
Strikingly, cells transfected with the p7-QQ mutant showed an enrichment of high-order core complexes mostly ranging in size from 480 kDa to over 1,000 kDa (Figure 7C and 7D) compared to WT-Jc1. Moreover, particularly in the separation of core structures by native, gradient electrophoresis, we observed clearly more core protein signal for p7-QQ (Figure 7C). It is possible that the dominant core species present in cells transfected with WT-Jc1 was too large to enter the polyacrylamide gel as would be expected for fully-assembled capsids. Alternatively, antibody recognition of core present in native protein complexes may be impaired which could result in the observation of less core protein signal for WT-Jc1 compared to the p7-QQ mutant (Figure 7C). Since we normalized the samples prior to loading for equal protein content by a core-specific ELISA, we can exclude that differential abundance of core protein is responsible for this observation. In contrast, these findings suggest that in the mutant either more core protein species were able to enter the gel or they were better detectable by the antibody, thus arguing that the mutant p7 protein altered capsid assembly. This conclusion is also corroborated by our rate zonal centrifugation assay performed in presence of detergent. Indeed, compared with WT-Jc1 and core-C69-72A, the p7 mutants showed an enrichment in fast sedimenting core complexes (Figure 7A, fraction 5) which poorly protected RNA from enzymatic digestion (Figure 7B). In contrast, the dominant core species of WT-Jc1 present in fraction 3 efficiently protected RNA from RNase digestion unless the shielding capsid was previously removed by protease treatment. Hence, mutation in p7 caused an accumulation of high-order core complexes which might be loosely associated with but have not yet completely incorporated the viral RNA. Whether those core complexes represent capsid assembly intermediates or aberrant core aggregates remains to be investigated but the results clearly point to a role of p7 during RNA encapsidation.
To get a better understanding of the phenotypes observed for the p7 mutants, we compared them to structural protein mutants, namely the core-C69-72A, E1-K179Q and ΔE1E2 mutants. These point and deletion mutations reduced virus titers to levels comparable to the p7-QQ and Δp7half mutations, respectively, indicating that differences in phenotypes were not due to varying potencies of the mutations but rather pointed to distinct mechanisms of action. The core mutant core-C69-72A was first described by Murray et al. who used an alanine scanning approach to identify residues important for virus production [56]. This mutation inhibited assembly of infectious virus particles without altering the protein stability or its association with LDs [56]. We confirmed those observations in the Jc1 context and further characterized the mutant assembly defect. Similarly to p7-QQ mutant, this core mutant showed an accumulation of naked capsid-like core complexes at the surface of LDs. However, no impairment of capsid assembly, as far as it could be resolved in our assays, was observed. Thus, this mutation seemed to interfere with budding via a mechanism of action different from the p7 mutations. Interestingly, cells transfected with the core mutant often exhibited large lipid droplet aggregations at the cell periphery (data not shown) contrasting with the perinuclear concentration typically seen for the p7 mutants and for most HCV genotypes, or with the diffuse localization pattern observed in Jc1-transfected cells. A correlation between inefficient virus assembly and core accumulation at LDs has been previously reported [60]. However, core accumulation at the LD surface typically displaces ADRP leading to a redistribution of LDs towards the MTOC (microtubule organization center) in the perinuclear region [64]. In case of the C69-72A mutant, LDs seemed to be relocalized in the opposite direction, thereby increasing the distance between sites of capsid assembly and the ER budding site, which could explain the impairment in capsid envelopment. Interestingly, Murray et al. reported a rescue of infectious virus production by two independent second site mutations, one in the first TM segment of p7, the other one in the third TM segment of NS2, supporting the notion that p7 and NS2 are important for budding of the capsid into the ER [56].
Residue K179 in E1 has been shown to be important for heterodimerization of the viral glycoproteins [57] and its substitution with a glutamine residue leads to a reduced E1/E2 interaction [46] and a strong decrease in infectious virus production [4]. Surprisingly, we found that extracellular core levels remained nearly unchanged (Figure 1C) suggesting that only the specific infectivity of virus particles was affected and not particle assembly and release per se. Accordingly, E1-K179Q behaved as WT-Jc1 in all subsequent characterization assays. Hence, efficient heterodimerization of the viral glycoproteins is no prerequisite for particle assembly and envelopment but essential for subsequent entry steps into new cells [65], [66].
To further analyze possible contributions of the viral glycoproteins in HCV assembly, we created a deletion mutant lacking both E1 and E2. Not surprisingly, envelopment of capsids was completely abrogated as seen by rate zonal centrifugation (Figure 3) pointing towards a role of the glycoproteins during the budding process. This supports the model of Popescu et al. where E2 is part of the protein interaction complex ultimately leading to virion assembly [45]. This argues that the glycoproteins, just like p7, are necessary not only for envelopment but also for the final stages of capsid assembly. One possible explanation for this link between envelope glycoproteins and capsid assembly could be that, due to the high hydrophobicity of core domain 2, the membrane scaffold provided by the ER during the budding process is needed for the formation of fully-closed capsids. Thus, we propose the following model of HCV assembly (Figure 8): HCV RNA is translocated from the replicase complex to lipid droplets by NS5A through interaction with core. In the presence of viral RNA, assembly of high-order core complexes is initiated. The bridging function of NS2 between the glycoproteins and replicase components provides a close proximity of replication and assembly sites while p7 might directly or indirectly recruit the core assembly intermediate to the ER membrane, thereby providing the scaffold necessary for completion of capsid assembly and initiating the budding process. However, several open questions remain. Is there a direct interaction between core and p7? Which other proteins does p7 interact with? Is the ion channel activity involved? Is p7 part of the virus particle and does it have a role during cell entry? These are interesting questions to be resolved in the future.
Huh-7.5 cells [67], the Huh-7.5-derived stable cell line expressing core, E1, E2, p7 and NS2 of the J6CF isolate [59], and Huh7-Lunet-hCD81-Gluc cells [68] were grown at 37°C and with 5% CO2 in Dulbecco's modified Eagle's medium (DMEM; Invitrogen, Karlsruhe, Germany) supplemented with 2 mM L-glutamine, non-essential amino acids, 100 U/ml of penicillin, 100 µg/ml of streptomycin, 10% fetal calf serum, and for the latter two 5 µg/ml of blasticidin. Note that Huh7-Lunet-hCD81-Gluc cells are derived from the highly permissive Huh7-Lunet-hCD81 cells [69] via lentiviral gene transfer to express the secreted gaussia luciferase which is a simple biomarker to assess cell density and viability.
The plasmids pFK-Jc1, pFK-Jc1-Δp7half, pFK-Jc1-p7-KR33,35QQ (p7-QQ mutant) and pFK-Jc1-E1-K179Q have been previously described [4], [58]. The pFK-Jc1-core-C69-72A plasmid which carries 4 consecutive alanine substitutions of core residues 69-72 [56] was created via PCR mutagenesis. Plasmid pFK-Jc1-ΔE1E2 which carries a complete deletion of the E1–E2 coding region encompassing residues 192 to 750 of the J6CF [70] isolate was created by a PCR-based strategy. Likewise, construct pFK-Jc1Δcore was created by PCR-based mutagenesis. It encodes a Jc1 derivative that has an in frame deletion encompassing residues 17 through 131 of the core coding region. Finally, to express core protein in the absence of E1, E2, p7 and NS2, we created a derivative of the bicistronic JFH1 replicon pFK PI-EI-NS3-5B/JFH1 [71] expressing J6CF-derived core from the first cistron and designated pFK PI-Core/J6-EI-NS3-5B/JFH1. Each construct was verified by sequencing. Additional information including the sequences of these constructs is available upon request.
Methods for in vitro transcription of HCV RNA and its electroporation into Huh-7.5 cells have been previously reported [72]. HCV core protein and RNA were quantified respectively by ELISA [68] and qRT-PCR [72]. Intra- and extracellular infectious titers were determined as previously described [68].
For experiments under denaturating conditions, protein samples were obtained from scraped cells lysed directly in SDS sample buffer (final concentrations (1×): 75 mM Tris-HCl pH 6.8; 0.6% (v/v) SDS, 15% (v/v) glycerol, 0.001% (w/v) bromophenol blue; 7.5% (v/v) ß-mercaptoethanol) for 10 min at 95°C and separated by SDS-PAGE. Immunoblotting was performed essentially as described before [72].
For blue-native PAGE, cells were harvested and lysed using the NativePAGE Sample Prep kit (Invitrogen, Karlsruhe, Germany) with 1% n-dodecyl beta-D-maltoside (DDM, AppliChem, Darmstadt, Germany) under non-reducing conditions without heating according to the manufacturer's recommendations. Typically, 1×106 HCV-transfected Huh7-Lunet-hCD81-Gluc cells were lysed in 100 µl volume, of which 25 µl were loaded onto 4–16% or 3–12% polyacrylamide gradient gels (Invitrogen) and electrophoresis was performed at a constant voltage of 150 V. Proteins were transferred on a PVDF membrane using a dry blotting system (IBlot, Invitrogen).
For two-dimensional resolution, individual lanes from blue-native PAGE were cut out of the gel, equilibrated in 2× SDS sample buffer for 30 min, stacked peripendicular above a 11% polyacrylamide SDS gel in one large stacking gel pocket and overlayed with 1× SDS sample buffer. Gel electrophoresis and immunoblotting were performed as above [72].
For western blotting, the following antibodies were used: anti-core (C7-50; kindly provided by D. Moradpour; [73]; diluted 1∶1,000), anti-ADRP (adipose-differentiation related protein, AP125 antibody; Progen, Heidelberg, Germany; diluted 1∶200), anti-Calreticulin (Stressgen, Victoria, BC, Canada; diluted 1∶2,000) and anti-mouse-HRP (Sigma, Steinheim, Germany; diluted 1∶20,000).
HCV-transfected cells seeded in 6-well dishes were scraped in 170 µl proteinase K buffer (50 mM Tris-HCl pH 8.0; 10 mM CaCl2; 1 mM DTT) 48 h post-transfection and subjected to five cycles of freeze and thaw. Subsequently, 50 µl of the crude lysate was left untreated, while 50 µl was treated with 50 µg/ml proteinase K (Roche, Mannheim, Germany) for 1 h on ice and another 50 µl was lysed with 5% (v/v) Triton X-100 prior to proteinase K treatment. Proteinase K digestion was terminated by addition of PMSF (phenylmethylsulfonyl fluoride; AppliChem) at a final concentration of 5 mM and incubation of the sample on ice for 10 min. Subsequently, 13 µl of 5× SDS sample buffer were added and the sample was heated to 95°C for 10 min. The amount of residual core protein was determined by SDS-PAGE and immunoblotting. Alternatively, to quantify core amounts by ELISA, proteinase K was inactivated with 10 mM PMSF, heating to 95°C and addition of 50 µl 2× protease inhibitor cocktail (Roche, Mannheim, Germany).
Three 50-µl samples from individual 1-ml gradient fractions were either (i) left untreated, (ii) treated with RNase A (Sigma) (50 µg/ml for 4 h at RT) or (iii) pretreated with proteinase K (5 µg/ml, for 1 h on ice; reaction was stopped by the addition of 10 mM PMSF and 5 µl of 10× protease inhibitor cocktail) prior to RNase A digestion (as above). Subsequently, total RNA was extracted using the NucleoSpin RNAII kit (Macherey-Nagel, Düren, Germany) according to the manufacturer's instructions and HCV RNA quantified by qRT-PCR as previously described [72].
HCV-transfected cells seeded in 10-cm dishes were scraped in 250 µl TNE buffer (10 mM Tris-HCl pH 8.0; 150 mM NaCl; 2 mM EDTA) 48 h post-transfection and subjected to five cycles of freeze and thaw. Postnuclear supernatants collected after centrifugation of the crude lysate for 5 min at 4°C and 1,700× g were layered on top of preformed continuous 0–30% sucrose/TNE gradients and spun at 270,000× g for 1 h at 4°C in a Sorvall TH-641 rotor. Subsequently, 10 fractions (1 mL each) were harvested from the top and the respective refraction index was measured by refractometry. For rate zonal centrifugation in the presence of detergent, 1% DDM was added to the sucrose solutions and to the postnuclear supernatant.
Lipid droplets were isolated from HCV-transfected cells following a previously published protocol [74]. Briefly, 48 hours post-transfection, pellets of HCV-transfected cells were harvested from 15-cm dishes by scraping and resuspended in 1 ml ice-cold HLM (hypotonic lysis medium; 20 mM Tris-HCl pH 7.4; 1 mM EDTA) followed by incubation on ice for 10 min and lysed by 6–8 strokes in a Potter-Elvehjem tissue homogenizer. The postnuclear supernatant was mixed with one third volume of HLM containing 60% sucrose and layered under 5 ml HLM containing 5% sucrose and 5 ml sucrose-free HLM. Rate zonal centrifugation was performed at 28,000× g for 30 min at 4°C in a Sorvall TH-641 rotor. The uppermost fractions enriched in lipid droplets were collected and the purification process was controlled by SDS-PAGE and immunoblotting of ADRP and calreticulin in the isolated fractions.
Indirect immunofluorescence staining was performed as described before [72]. HCV core protein was detected using the mouse monoclonal C7-50 antibody diluted 1∶1,000 in PBS supplemented with 5% goat serum (Sigma) followed by a secondary antibody specific for murine IgG conjugated with Alexa-Fluor 568 (Invitrogen) at a dilution of 1∶1,000. For LD detection, BODIPY (493/503) (Invitrogen) was added to the secondary antibody at a dilution of 1∶1,000. Cell nuclei were counter-stained for 1 min at RT with DAPI (Invitrogen) diluted 1∶3,000.
Data were analyzed statistically in R (http://www.r-project.org). Normality of data was assessed using histograms and qq-plots. Differences between sample populations were assessed using a two-sample Welch's t-test. P-values were calculated and statistical significance reported as highly significant with (***) if p≤0.001, (**) if p≤0.01, and significant (*) if p≤0.05. Differences were considered not significant (n.s.) for p>0.05.
|
10.1371/journal.ppat.1005149 | The Role of the Antiviral APOBEC3 Gene Family in Protecting Chimpanzees against Lentiviruses from Monkeys | Cross-species transmissions of viruses from animals to humans are at the origin of major human pathogenic viruses. While the role of ecological and epidemiological factors in the emergence of new pathogens is well documented, the importance of host factors is often unknown. Chimpanzees are the closest relatives of humans and the animal reservoir at the origin of the human AIDS pandemic. However, despite being regularly exposed to monkey lentiviruses through hunting, chimpanzees are naturally infected by only a single simian immunodeficiency virus, SIVcpz. Here, we asked why chimpanzees appear to be protected against the successful emergence of other SIVs. In particular, we investigated the role of the chimpanzee APOBEC3 genes in providing a barrier to infection by most monkey lentiviruses. We found that most SIV Vifs, including Vif from SIVwrc infecting western-red colobus, the chimpanzee’s main monkey prey in West Africa, could not antagonize chimpanzee APOBEC3G. Moreover, chimpanzee APOBEC3D, as well as APOBEC3F and APOBEC3H, provided additional protection against SIV Vif antagonism. Consequently, lentiviral replication in primary chimpanzee CD4+ T cells was dependent on the presence of a lentiviral vif gene that could antagonize chimpanzee APOBEC3s. Finally, by identifying and functionally characterizing several APOBEC3 gene polymorphisms in both common chimpanzees and bonobos, we found that these ape populations encode APOBEC3 proteins that are uniformly resistant to antagonism by monkey lentiviruses.
| Many human pathogens are of zoonotic origin, meaning they originated in animals. This includes HIV-1, the cause of the human AIDS pandemic, which is the result of cross-species transmissions of lentiviruses from chimpanzees and gorillas. However, little is known about the host factors that provide natural protection against viral emergence in a new species. Chimpanzees, which are humans’ closest relatives, harbor only a single lentiviral lineage, despite their frequent exposure to lentiviruses that infect monkeys on which they prey. Here, we investigate the capacity of the accessory protein Vif from different primate lentiviruses to antagonize the APOBEC3 antiviral gene family found in chimpanzees. We found that the Vif protein from most monkey lentiviruses was not able to antagonize chimpanzee APOBEC3G. Furthermore, other APOBEC3 proteins from chimpanzees were also resistant to Vif antagonism. Finally, we showed that, despite polymorphism in the APOBEC3 genes, common chimpanzee and bonobo populations are uniformly resistant to monkey lentiviral Vif antagonism. Our results are consistent with the hypothesis that the host APOBEC3 antiviral proteins protect chimpanzees against many HIV-related viruses commonly found in monkeys.
| Although lentiviruses are widespread in African monkeys, there have only been a few documented cases of cross-species transmission and lentiviral emergence into hominoids [1]. Chimpanzees are of particular interest because their lentivirus, SIVcpz, lies at the root of all HIV-1 infections [2]. SIVcpz has a complex evolutionary history as it resulted from the cross-species transmission and recombination of SIVrcm from red-capped mangabeys and SIVmus/mon/gsn from guenons [3,4]. However, only central and eastern chimpanzees are infected by SIVcpz, while western and Nigerian-Cameroonian chimpanzees as well as bonobos seem currently free of any lentiviral infection [1,5,6]. The fact that chimpanzees are infected by only a single lentiviral lineage is surprising given that they are exposed to SIVs that are present at high prevalence in their monkey prey [6,7]. Moreover, there have been multiple viral cross-species transmissions of simian foamy virus (SFV) and simian T-lymphotropic virus (STLV) to chimpanzees from their main prey, the western-red colobus [8–10], yet, no infection with this monkey species’ lentivirus, SIVwrc, has been documented in chimpanzees [6,7]. Overall, this suggests that there are host factors, rather than solely epidemiological or ecological barriers, that protect chimpanzees against the emergence of new lentiviral infections.
There have been four independent transmissions of HIV-1 into humans that originated from SIVcpz; two of these transmissions had their immediate source in chimpanzees (HIV-1 groups M and N), while two others passed through gorillas before infecting humans (HIV-1 groups O and P) [2,11]. HIV-2, on the other hand, is the result of cross-species transmissions of SIVsmm from sooty mangabeys to humans [2]. While SIVsmm has jumped to humans on over nine independent occasions, neither the equivalent SIVsmm infection of chimpanzees nor any other SIV other than the recombinant virus that gave rise to SIVcpz has been reported in apes. As chimpanzees are the closest relatives of humans, the mechanisms governing their susceptibility or resistance to lentiviruses have direct relevance for the potential of additional primate lentiviruses to adapt to hominoids and subsequently spread in humans.
Host restriction factors are intrinsic blocks to viral replication [12,13]. Therefore, to complete their lifecycle, viruses encode antagonists that target these innate immune factors. These antagonistic relationships have led to genetic conflicts driving the evolution and specificities of virus-host interactions, which may impose potent species barriers to cross-species transmission [12]. APOBEC3G is one of the antiviral proteins that have been implicated in the species-specificity of lentiviruses, although the role of APOBEC3G as a species barrier has been mainly investigated in experimental cross-species transmissions [14–17]. Moreover, there are at least four genes from the APOBEC3 gene family (APOBEC3D, F, G, and H) that potently block the lentiviral life cycle in the absence of the specific viral antagonist Vif (reviewed in [18,19]). Vif primarily counteracts the APOBEC3s by binding the host protein and targeting it for proteasomal degradation by recruiting an E3 ubiquitin ligase complex. Although vif is highly diverse within and between SIV and HIV lineages, it is present in all primate lentiviruses and has an ancient and conserved role in antagonizing the host APOBEC3G protein [20].
Here, we examined why chimpanzees harbor only a single SIV lineage despite being frequently exposed to various SIVs that infect their prey species. We show that Vif from diverse lentiviruses is incapable of antagonizing chimpanzee APOBEC3G. Moreover, additional chimpanzee APOBEC3 family members, especially APOBEC3D, also provide blocks to lentiviral replication. Consequently, we find that the potential of a lentivirus to replicate in primary chimpanzee CD4+ T cells is governed by its accessory protein Vif. Our data suggest that retention and evolution of the APOBEC3 family, where several host proteins are antagonized by a single viral protein at different motifs, set up a diverse battleground against viruses, which may overall enhance the protection of the host against viral emergence. Finally, we show that the APOBEC3 genes are polymorphic in common chimpanzees and bonobos, but that the populations are similarly resistant to lentiviruses with various vif. Overall, we propose that the restriction imposed by the APOBEC3 family of host restriction factors is a crucial mechanism by which common chimpanzees and bonobos may be naturally protected against most lentiviral cross-species transmissions.
Chimpanzees have overlapping ranges with many monkeys [21] that are widely and commonly infected by lentiviruses [1]. Each SIV bears a lineage-specific vif whose sequence varies greatly between lentiviral lineages (Fig 1A). It was previously shown that the inclusion of SIVmac vif in HIV-1 is necessary to experimentally generate a simian-tropic HIV-1 in rhesus macaques [14,15]. Moreover, in populations of African green monkeys (AGMs) naturally infected by SIVs, vif co-evolved with APOBEC3G polymorphisms in the different AGM species to maintain antagonism [17]. In addition, we previously found that adaptation of SIVcpz to chimpanzees involved the evolution of the vif gene to adapt and counteract chimpanzee APOBEC3G [4]. Therefore, to determine whether APOBEC3G could be responsible for the lack of transmission of diverse SIVs to chimpanzees, we tested an extended panel of Vifs from ten lentiviral lineages that spans the diversity of primate lentiviruses for its ability to antagonize chimpanzee APOBEC3G (Fig 1A). This panel included vif genes from SIVs that infect known preys of chimpanzees (e.g. SIVwrc from western-red colobus [6,7]). Each SIV vif gene was cloned into an HIV-1 backbone in place of HIV-1 vif as previously described [20]. The capacity of Vif to antagonize chimpanzee APOBEC3G was measured in single-round infectivity assays by co-transfecting the HIV-1 provirus containing an SIV vif gene with a plasmid encoding chimpanzee APOBEC3G [22]. The supernatant was normalized for p24gag expression and used to infect a T cell line, SupT1. The HIV-1 provirus encoded a defective envelope gene and was pseudotyped with VSV-G so that only one round of infection was assayed and the provirus expressed a luciferase gene used as the readout.
In the absence of Vif, chimpanzee APOBEC3G was able to block lentiviral infection (Fig 1B, white bar). As positive control, this was rescued by SIVcpz Vif, which fully antagonized chimpanzee APOBEC3G (Fig 1B, black bars, SIVcpzPts and SIVcpzPtt Vifs) [4]. By testing the Vif protein from eight different SIVcpz isolates, we found that all of them were also able to antagonize chimpanzee APOBEC3G (S1B Fig). However, Vif from other SIV lineages had differential capacities to antagonize chimpanzee APOBEC3G (Fig 1B, grey bars). As previously shown [4], Vif from both SIVrcm and SIVmus were able to partially restore infectivity in the presence of chimpanzee APOBEC3G (17–25% rescue of infectivity), which may have facilitated their evolution and adaptation to chimpanzees. Amongst the other monkey lentiviral vif genes, all but one lacked the capacity to rescue viral infection in the presence of chimpanzee APOBEC3G (less than 4% infectivity relative to SIVcpz Vif) (Fig 1B). In particular, the Vif from SIVwrc was unable to counteract chimpanzee APOBEC3G restriction (Fig 1B). This result alone may explain why chimpanzees are not infected by SIVs from one of their most common prey, the western-red colobus.
As controls, we also examined the Vif-APOBEC3G antagonism in two cases of known cross-species transmissions, that of SIVagm.ver (or SIVver) into baboons [23] and SIVagm.sab (or SIVsab) into Patas monkeys [24]. In both cases, the Vif from the donor species (i.e. SIVver Vif and SIVsab Vif) was capable to overcome the APOBEC3G of the recipient species (i.e. baboons and Patas monkeys) as well as it overcame the APOBEC3G of its natural host (S2 Fig). A tabulation of other cross-species transmissions of primate lentiviruses (Table 1) indicates that all known natural host switches that occurred were from primate lentiviruses that had some or full capacity to antagonize their new host APOBEC3G (Table 1 and S2 Fig). This suggests that at least partial antagonism of APOBEC3G may be a pre-requisite to natural cross-species infection. On the other hand, APOBEC3G antagonism is not sufficient to allow cross-species transmission, as other barriers may be involved. For example, SIVsmm Vif was the only monkey lentiviral protein able to completely antagonize chimpanzee APOBEC3G (Fig 1B), showing that APOBEC3G cannot explain the lack of infection of chimpanzee populations with this virus (see below). In summary, the lack of antagonism of APOBEC3G by lentiviral Vifs could explain the lack of SIV emergence into chimpanzees of most, but not all SIVs from monkeys.
Because other APOBEC3 genes may also be implicated in the Vif-dependent restriction of lentiviruses, we tested if vif genes from the various SIV lineages are capable of antagonizing chimpanzee APOBEC3D, APOBEC3F, and APOBEC3H. APOBEC3D is of particular interest, because the chimpanzee version of this protein is highly active against lentiviruses, while the human version has less such activity [25]. Using the single-round assay where exogenous APOBEC3 genes are transfected into 293T cells as described in Fig 1B, we confirmed that in the absence of Vif, chimpanzee APOBEC3D is more potent than the human APOBEC3D in restricting a lentivirus (S3A Fig, delta-Vif condition; similar expression level of chimpanzee and human APOBEC3D in a dose-dependent manner, S3B Fig) [25]. Despite the increased activity of chimpanzee APOBEC3D relative to the human protein, the Vif protein from two divergent SIVcpz isolates was able to antagonize chimpanzee APOBEC3D and to rescue viral infection (Fig 2A, black bars). On the other hand, the Vif protein from six monkey SIV lineages, including SIVwrc Vif, was not able to antagonize chimpanzee APOBEC3D (Fig 2A). Moreover, it was also notable that, although it could readily antagonize chimpanzee APOBEC3G, SIVsmm Vif was only poorly active against chimpanzee APOBEC3D (Fig 2A, 15% capacity versus SIVcpz Vif).
Chimpanzee APOBEC3F and APOBEC3H also reduced lentiviral infectivity in the absence of Vif, although APOBEC3F restriction was not as strong as the other APOBEC3s (Fig 2B and 2C, white bars). Most monkey SIV lineages encode a Vif that had a moderate activity against chimpanzee APOBEC3F (Fig 2B), while most Vifs were fully equipped to antagonize chimpanzee APOBEC3H (Fig 2C, levels of infectivity similar to SIVcpz Vif). This suggests that chimpanzee APOBEC3H and APOBEC3F on their own may not be major species barriers to SIVs in general, although some SIVs were more efficient than others at antagonizing the chimpanzee APOBEC3F in particular.
Overall, only the Vif protein from SIVcpz was able to antagonize all tested members of the chimpanzee APOBEC3 family. The antagonism of SIVcpz Vif corresponds to decreases in the levels of chimpanzee APOBEC3D, F, G, and H (S4 Fig), consistent with the known mechanisms of Vif-mediated degradation of APOBEC3 proteins [18]. The protein Vif from different lentiviral lineages had different specificities for a given APOBEC3 substrate (Fig 2D, read the heat map vertically). For example, chimpanzee APOBEC3G could be antagonized by SIVsmm, but not by SIVwrc or other SIVs (Fig 2D). Importantly, we also found that a given vif had different specificities amongst the APOBEC3 genes (Fig 2D, read the heat map horizontally). Indeed, while most of the Vif proteins retained the capacity to antagonize chimpanzee APOBEC3F and APOBEC3H, only a few of them were capable to counteract chimpanzee APOBEC3D and APOBEC3G (Fig 2D). This suggests that the evolution and retention of multiple antiviral APOBEC3 proteins in chimpanzees provide a potent restriction to a broad diversity of lentiviruses, and therefore likely confer an advantage to the species against cross-species infections. Our data also show that the various SIV lineages have different susceptibility to chimpanzee restriction factors. For example, SIVsmm strains would be expected to adapt and antagonize chimpanzee APOBEC3 proteins more readily than strains of SIVwrc, which would need to adapt to antagonize three APOBEC3 members. Indeed, SIVwrc was unable to cause degradation of chimpanzee APOBEC3G, F, and D proteins (S4 Fig). Hence, in addition to the strong barrier conferred by APOBEC3G, other APOBEC3 members, especially APOBEC3D, also pose an obstacle towards transmission and adaptation of diverse SIVs harbored by monkeys to chimpanzees.
We wished to determine if the Vif-dependent restrictions observed in the APOBEC3 over-expression assays (Figs 1 and 2) could be recapitulated in infections of primary chimpanzee CD4+ T cells. We modified the proviruses used in Figs 1 and 2 by replacing the HIV-1 env gene, so that the HIV-1 backbone containing different SIV vif genes would be replication-competent with an X4 envelope from HIV-1. The advantage of this system, as opposed to infection of chimpanzee cells with different entire SIVs, is that we could control for the other host factors that may interact in a species-specific manner with the virus, as only the vif gene was different between the replication-competent viruses. Moreover, certain HIV-1 strains have been shown to replicate in primary chimpanzee cells [26].
Because of the limiting amounts of primary chimpanzee (Pan troglodytes verus) CD4+ T cells, we used four HIV-1 clones, each containing a different SIV Vif; two SIV Vifs that fully antagonized chimpanzee APOBEC3G in the over-expression system (SIVcpz Vif, which served as a positive control, and SIVsmm Vif), one Vif that partially antagonized chimpanzee APOBEC3G (SIVrcm Vif), and one that failed to antagonize chimpanzee APOBEC3G (SIVsab Vif). An HIV-1 deleted in vif served as a negative control. This assay differs in two important ways from the single-round assay used in Figs 1 and 2. First, since the viruses are replication-competent, we measured p24gag production rather than a reporter gene. Second, and more importantly, viral inocula were initially produced in transfected 293T cells in the absence of any added APOBEC3 gene. Because APOBEC3G is active in the target cell, rather than the producer cell, this means that the first round of infection of the primary cells will proceed uninhibited by APOBEC3G and the effects of endogenous chimpanzee APOBEC3G would be observed only in the second (and subsequent) rounds of infection.
All viral constructs replicated with comparable kinetics in a “permissive” T cell line, SupT1, which does not express endogenous APOBEC3G (S5A Fig) [27,28]. This shows that none of the viral constructs had an inherent replication defect. In primary CD4+ T cells from three chimpanzee donors, we found that the delta-Vif construct replicated after the initial infection (as expected since it was produced in cells without APOBEC3G; Fig 3A, grey lines). However, it did not replicate beyond the first time-point measured (Fig 3A, grey lines). On the other hand, the construct that encoded an SIVcpz Vif replicated between one to two orders of magnitude better than that of the delta-Vif construct (Fig 3A, black versus grey lines). These data show that the vif gene is essential for efficient viral replication in primary chimpanzee CD4+ T cells.
We found similar replication patterns in all three chimpanzee CD4+ T cell cultures among viruses containing different vif genes (Fig 3A). The viral construct containing SIVsab vif replicated to similar levels as the negative control (absence of vif), showing that SIVsab Vif was not active in chimpanzee cells (Fig 3A, orange line). This result is consistent with the data that SIVsab Vif is unable to overcome chimpanzee APOBEC3G (Fig 1B). As control, the infectious proviral construct with SIVsab Vif was fully capable of overcoming AGM APOBEC3G in an infection assay (S6 Fig). Moreover, these results indicate that of the APOBEC3 proteins, APOBEC3G alone is capable of blocking virus replication since SIVsab Vif was able to antagonize chimpanzee APOBEC3D, F, or H (Fig 2D).
In contrast to viruses encoding SIVsab Vif, viruses that encoded SIVrcm Vif and SIVsmm Vif had intermediate capacities to replicate in the chimpanzee donor cells (Fig 3A, blue and green lines). The intermediate replication of the virus encoding SIVrcm Vif (Fig 3A) is consistent with the intermediate ability of SIVrcm Vif to overcome chimpanzee APOBEC3G (Fig 1B). However, the intermediate replication of the virus encoding SIVsmm Vif (Fig 3A) is not consistent with the full activity of SIVsmm Vif against chimpanzee APOBEC3G (Fig 1B), but could be explained by the poor activity of SIVsmm Vif against chimpanzee APOBEC3D (Fig 2A and 2D). Therefore, while the Vif-dependent restriction of SIVrcm and SIVsab in primary chimpanzee CD4+ T cells could be explained by APOBEC3G, the Vif-dependent restriction of SIVsmm must be due to another cellular factor, potentially APOBEC3D.
The APOBEC3 proteins are not induced by interferon (IFN) in activated human CD4+ T cells [27]. However, other Vif-dependent potential restriction factor might be induced by IFN. Thus, we also performed these experiments in the presence of IFN to determine if we would see different patterns of virus growth. We found that treatment of primary chimpanzee CD4+ T cells with IFNα or IFNβ lowered the overall amount of lentiviral replication, but did not change the relative Vif-dependency (Figs 3B and S5B). These data suggest that the non-IFN induced APOBEC3 proteins are the major Vif targets in activated primary chimpanzee CD4+ T cells.
The APOBEC3 proteins mediate their antiviral effects through the hypermutation of the newly synthesized viral genome by their cytidine deaminase activity, as well as other proposed mechanisms [19,29]. As one of the hallmarks of the APOBEC3 restriction is the induction of G-to-A hypermutation in viruses (reviewed in [18]), we looked for evidence of such hypermutation in integrated viral genomes nine days after infection of primary chimpanzee CD4+ T cells. Genomic DNA was extracted from infected cells and two fragments (of ~1,200 bp and of ~600 bp) encompassing the vif region were amplified, cloned, and sequenced (see Methods). Two methods were used to determine the significance of hypermutation signatures, Hypermut [30] and Hyperfreq [31], which both determine that a sequence is hypermutated when G-to-A mutations in a given hypermutation-associated context are more likely than mutations in a control context. The first method uses the Fisher exact test, while the second one uses a Bayesian approach and can evaluate the strength of various hypermutation contexts [31] (see Methods).
We found that the viral construct that lacked a vif gene accumulated many G-to-A mutations in its genome (G-to-A mutation rate of 0.65%, versus 0.05% for other mutations; Table 2, ΔVif column) and 50% of the sequences were found to be significantly hypermutated in an APOBEC3-context (p<0.05) (Table 2). These mutations occurred primarily in the GG context, which is characteristic of APOBEC3G activity (85% of the mutations were in the GG context and, using Hyperfreq, hypermutation of sequences was most frequently associated with the GG context, Table 2). However, G-to-A mutations also occurred in the GA context, which may be a signature of APOBEC3F, APOBEC3D, and/or APOBEC3H activity (Table 2) [25,32]. In contrast, the viral construct that expressed the SIVcpz Vif had no evidence of hypermutation (Table 2, cpzVif column). This suggests that the expression of the APOBEC3 proteins in activated primary chimpanzee CD4+ T cells was able to hypermutate the viral genome in the absence of Vif antagonism, with APOBEC3G being the main driver, and that SIVcpz Vif could counteract this APOBEC3-mediated hypermutation.
In the virus construct that encoded SIVsab Vif, we found an elevated G-to-A mutation rate in the GG context (0.81%, versus 0.07% for other mutations, Table 2) and 67% of the sequences were found to be significantly hypermutated in the APOBEC3G-context (p<0.05 with both Hypermut and Hyperfreq, Table 2). Importantly, seven recovered viral clones out of nine harbored premature stop codons in the sequenced open reading frames. Therefore, the accumulation of G-to-A mutations mediated by APOBEC3G may be directly responsible for the poor viral growth of viruses encoding SIVsab Vif in primary chimpanzee CD4+ T cells (Fig 3).
By analyzing the viruses encoding SIVrcm Vif and SIVsmm Vif, we did not find any evidence of APOBEC3G deaminase activity (Table 2). However, for the virus encoding SIVrcm Vif, two viral clones out of 21 were significantly hypermutated in the GA context (p<0.05 Hyperfreq, Table 2), suggesting that APOBEC3D, APOBEC3F, and/or APOBEC3H were active against this virus. This is consistent with our single-round infectivity results where SIVrcm Vif could not antagonize chimpanzee APOBEC3D (Fig 2A and 2D). The little or lack of hypermutation observed in the viral constructs containing SIVsmm vif or SIVrcm vif, for which replication was only partly inhibited, may be due to the fact that the mutational activity of the APOBEC3 family members was below our limit of detection at nine days post infection (e.g. due to outgrowth of a subpopulation of virus that escaped hypermutation or other causes). On the other hand, the low hypermutation rate may also suggest that a Vif-dependent host factor that does not rely on its deaminase activity was in part responsible for this phenotype.
Polymorphisms in APOBEC3 genes from various primates may impact protein function or the capacity to escape from viral antagonists [16,17,33]. Therefore, we sought to identify and functionally characterize genetic variants of the APOBEC3 genes present in bonobos (Pan paniscus) as well as different subspecies of common chimpanzees (Pan troglodytes). To date, the sequence of the APOBEC3 genes has only been reported from bonobos and western chimpanzees (P. t. verus)—two populations that appear currently free of lentiviral infection [1,5] (Fig 4A). Here, we analyzed the deep-sequencing reads from the Great Ape Genome Project [34] to determine whether additional polymorphisms existed in the non-human hominoid APOBEC3 sequences, in particular between chimpanzee subspecies that harbor SIVcpz and those that do not. Specifically, we mapped reads to the chimpanzee reference genome (panTro3) using BWA-MEM aligner, which allowed us to appropriately retrieve the polymorphisms for the APOBEC3 gene family. We also performed Sanger sequencing on a subset of 16 variants and validated the polymorphisms and the heterozygous/homozygous sites for all the variants tested (S7 Fig). Overall, we recovered genetic variants for APOBEC3D, G, F, and H genes, from a total of 36 individuals including five western chimpanzees, ten Nigerian-Cameroonian chimpanzees, four central chimpanzees, six eastern chimpanzees, and eleven bonobos (Fig 4A).
In the coding regions of the APOBEC3 genes, we found multiple sites that were polymorphic within and/or between chimpanzee populations (Table 3). The majority (56%) of the individuals were homozygous for the reference nucleotide at polymorphic sites, 31% were homozygous for the alternate nucleotide, while 12% were heterozygous. None of the variants were found in the cytidine deaminase domains. The APOBEC3G and APOBEC3F genes had the greatest number of polymorphic sites, and five out of eight SNPs in each gene coded for amino acid changes (Table 3). In addition, none of the SNPs identified in chimpanzees were overlapping with the common polymorphisms (MAF>1%) that have been described in human APOBEC3 genes [35].
To determine whether or not the identified APOBEC3 genetic variants impacted their anti-lentiviral function and/or their escape from Vif antagonists, we tested the effect of a subset of SNPs that were in, or proximal to, putative Vif binding regions (Table 3, underlined positions). In particular, we tested one SNP in APOBEC3D (N324D) and two SNPs in APOBEC3F (E282D and S322R), all of which were found in regions known to affect HIV Vif binding [36], as well as all five non-synonymous SNPs in APOBEC3G (Table 3). We found that all variants had similar levels of expression in transient expression assays (Fig 4B), and all retained their antiviral capacity in the absence of Vif (Fig 4C,—Vif). We also tested the APOBEC3 variants against a panel of Vif from three SIVs that have different antagonist specificities and are from different primate species: SIVcpz from chimpanzees, SIVwrc from western-red colobus, which is chimpanzee’s main monkey prey, and SIVsmm from sooty mangabeys (Fig 2D). Although SIVcpz infects only two chimpanzee subspecies, we found that SIVcpz Vif could antagonize all chimpanzee APOBEC3 variants tested (Fig 4C). Furthermore, the APOBEC3D, APOBEC3F, and APOBEC3G genetic variants, despite harboring variations in the potential Vif binding pocket, did not affect the antagonist capacity of SIV Vifs (Fig 4C). Although some variation in the ability of SIVwrc Vif to counteract APOBEC3G genetic variants was observed, SIVwrc Vif was not able to the rescue viral infection in any condition. Thus, our conclusions on chimpanzee APOBEC3 resistance against lentiviruses from monkeys that were made from the sequence of APOBEC3 genes from a single P. t. verus individual (Figs 1 and 2) and on primary cells from three P. t. verus donors (Fig 3) are representative of all the chimpanzee and bonobo populations. Therefore, the APOBEC3 genes may uniformly protect chimpanzees and bonobos against the emergence of most primate lentiviral lineages.
Although chimpanzees are frequently exposed to various lentiviruses that infect their monkey preys, they harbor only one SIV, SIVcpz. The fact that chimpanzees have remained resistant to other SIVs is especially intriguing because they have acquired other retroviruses from monkeys (reviewed in [1,6]). Here, we show that the APOBEC3 family of antiviral proteins may represent a powerful barrier against cross-species infection of most SIVs to chimpanzees. We found that chimpanzee APOBEC3G blocks viral replication of most SIV strains to which these apes are exposed [6–9]. Furthermore, our data are consistent with the hypothesis that encoding multiple APOBEC3s with different virus-host interfaces contributes to the protection of a host against cross-species transmission by being differentially resistant to the vif antagonist encoded by monkey lentiviruses. Moreover, the Vif-dependent lentiviral replication observed in primary chimpanzee cells, which varied according to the SIV of origin, highlights the importance of Vif in the spread of lentiviruses between species. Finally, although polymorphisms in the APOBEC3 genes are found across chimpanzee populations, we found that the different (sub)species (regardless of whether or not they currently harbor SIVcpz in their population) encode APOBEC3 proteins with similar specificity. We would therefore expect the common chimpanzee and bonobo populations to be similarly resistant to the Vif antagonism from most SIVs.
We found that chimpanzee APOBEC3G cannot be fully antagonized by the Vif protein of any circulating monkey SIV, except by Vif from the SIV infecting sooty mangabeys. Therefore, with the caveat that our experiments were done in vitro, and therefore are much more simplified than what occurs in actual transmissions in the wild, these results are consistent with the hypothesis that APOBEC3G in chimpanzees is a host factor that represents a potent species barrier to very diverse primate lentiviruses. This factor alone could explain why SIVwrc from western-red colobus has not crossed into chimpanzees despite ample evidence of exposure of chimpanzees to this virus in their prey [6,7]. This result is consistent with other reports on APOBEC3G as a selective barrier to heterologous viruses [14–17]. Moreover, our previous studies found that a dramatic change in the vif gene at the “birth” of SIVcpz was needed to allow vif adaptation to chimpanzee APOBEC3G [4]. Finally, Vif-APOBEC3G antagonism in natural host switches (Table 1 and S2 Fig) suggests that at least partial antagonism of APOBEC3G may be a pre-requisite to natural cross-species infection.
In primary chimpanzee CD4+ T cells, we observed a strong replication defect of lentiviruses with heterologous Vifs that cannot antagonize chimpanzee APOBEC3G, as shown with a viral construct bearing SIVsab Vif. Indeed, this construct accumulated numerous mutations in the GG context including stop codons that would be deleterious for the virus. Therefore, it is likely that APOBEC3G would not allow for sufficient rounds of replication of viruses that harbor a non-active Vif (such as SIVsab, SIVwrc and many others) for adaptation to chimpanzee.
While several other restriction factors may play a role as species barrier in different settings, their role in protection of chimpanzees against cross-species infection of SIVs is far less compelling than that of the APOBEC3 family. For example, TRIM5 is also known to be a species barrier in experimental transmissions of HIV-1 to rhesus macaques [14,15]. However, none of a diverse panel of SIVs tested were sensitive to the restriction by chimpanzee TRIM5 [37]. In addition, the restriction imposed by the antiviral gene Tetherin/BST-2 is less likely to block cross-species transmissions since many SIVs may have a Nef protein capable of antagonizing chimpanzee Tetherin/BST-2, which, in contrast to human has a cytoplasmic terminal domain similar to other monkeys [38]. However, our results do not rule out that additional barriers to SIV infections of chimpanzees (for example, receptor or co-receptor mismatches, and other possible antiviral genes) might also exist.
The duplication and subsequent evolution of the APOBEC3 genes in primates may be an efficient evolutionary strategy for the host to keep up in the virus-host arms race. This evolutionary strategy may generally be beneficial to target several viruses [39,40], as well as to target a single rapidly evolving virus that will need to develop multiple defense strategies. Here, the fact that a single antagonist, Vif, targets the various host APOBEC3 proteins may be an efficient strategy for the host to provide a broad spectrum of defense against a potential emerging virus. Moreover, although APOBEC3G exert the main lentiviral restriction amongst the APOBEC3 members, it was shown that human APOBEC3D, APOBEC3F, and APOBEC3H also restrict lentiviruses in primary cells and in humanized mice [27,28,41]. The caveat to this conclusion to in vivo transmission is that we do not yet know the expression of the different chimpanzee APOBEC3 proteins in the SIV target cells. Nonetheless, as the mRNAs corresponding to the human versions of the APOBEC3 proteins are widely expressed [27], it is likely that the single viral protein Vif would need to simultaneously adapt to multiple host restriction factors with different interfaces to allow viral replication in chimpanzees. Furthermore, Vif has distinct regions for antagonizing the APOBEC3 proteins [19,42,43]. Therefore, overcoming one APOBEC3 protein after adaptation will not necessarily allow the virus to overcome the other APOBEC3 members. In heterologous SIV infections of AGMs, viral adaptation was impaired when Vif had to adapt to two alleles of APOBEC3G, highlighting the role of “heterozygous advantage” [17]. The host advantage of multiple alleles can be extended to the harboring of multiple antiviral genes from the same family. Therefore, when a Vif protein needs to adapt to multiple APOBEC3 members, as SIVwrc Vif in chimpanzees, it would increase the host advantage and drastically constrain viral evolution and adaptation to the new species.
SIVsmm from sooty mangabeys has been able to cross multiple times to humans giving rise to HIV-2 viruses. Two independent SIVsmm transmissions were successful and are at the origin of the HIV-2 epidemic in West Africa, while at least seven others have been found in only few individuals with no or very limited secondary spread [44,45]. In contrast, no equivalent SIVsmm emergence in chimpanzees has been recorded. The present-day lack of transmission of SIVsmm into chimpanzees is most likely due to ecological factors since, amongst chimpanzees and bonobos, only western chimpanzees have overlapping ranges with sooty mangabeys and they do not frequently hunt sooty mangabeys [6]. Nonetheless, as SIV infection of monkeys has been ongoing for millions of year [20,46,47], we believe that it is likely that at some time in the past chimpanzees would have been exposed to an SIV with a vif gene with a similar specificity as modern-day SIVsmm. Our results on infections in primary chimpanzee CD4+ T cells show that there is a clear Vif-dependence of infection, where the SIVsmm Vif poorly supports viral replication in these cells. We found that this impairment cannot be explained by restriction by APOBEC3G, APOBEC3F, or APOBEC3H (Fig 2D, SIVsmm Vif). However, because chimpanzee APOBEC3D has stronger antiviral activity compared to human APOBEC3D ([25] and S3 Fig) and SIVsmm Vif is not able to fully antagonize chimpanzee APOBEC3D (Fig 2A), it is possible that this host protein may be responsible for the reduced infection of the virus with SIVsmm vif in primary chimpanzee cells (Fig 3). Notably, at least in human cells, APOBEC3D, is expressed in primary CD4+ T cells and can have antiviral effects [27,28,41]. Moreover, major evolutionary changes in Vif were necessary for adaptation of SIVcpz to chimpanzees, which included adaptation to chimpanzee APOBEC3D [4]. On the other hand, human APOBEC3D was unlikely a barrier to transmission of SIVsmm to humans [25].
In a nine-day infection of primary chimpanzee CD4+ T cells, we did not observe any signature of APOBEC3D deaminase activity in the presence of SIVsmm Vif. It is possible that this activity is not observed in our system, but can have some impact in vivo, or that it acts primarily by a deaminase-independent mechanism [29]. It is also possible that another Vif-dependent factor is responsible for the viral growth defect in chimpanzee cells.
Screening for variants in the APOBEC3 genes among the chimpanzee population, we found that these host proteins are polymorphic, but that functionally they are all similarly resistant or susceptible to circulating lentiviruses. This is similar to what has been described for the common polymorphisms found in APOBEC3 genes in different human populations [35]. Although only two subspecies of chimpanzees are infected by SIVs, these do not harbor functionally different APOBEC3 genes from the ones found in bonobos or western chimpanzees that appear free of lentiviral infection [5,7]. This emphasizes that even though this gene family is polymorphic in common chimpanzees and bonobos, these modern-day apes are similarly resistant to Vif from diverse SIVs across Africa. These findings are however in contrast to studies on APOBEC3G from captive macaques, APOBEC3G from wild African green monkeys, or human APOBEC3H, where polymorphisms influence the host protein restriction capacity or susceptibility to Vif [16,17,33,48]. This suggests that the main antiviral APOBEC3 genes in chimpanzees may have been under strong selection either now or in the recent past to prevent invasion by other primate lentiviruses. On the other hand, the recent introduction and low overall prevalence of SIVcpz relative to lentiviruses of other primates [5,46,49] may also explain why we do not observe an ongoing “arms race” between chimpanzee APOBEC3 genes and the SIV vif gene as observed in other primate species [17].
Both common chimpanzees and bonobos harbor multiple APOBEC3 gene members that are strongly active against lentiviruses, including APOBEC3D and APOBEC3H that otherwise have a poor anti-lentiviral capacity in humans [25,33]. Together with our population genetic data, the strong positive selection observed in hominoids’ APOBEC3 genes [25,33,50,51] suggests that ancient lentiviruses had infected primates before the split between common chimpanzees and bonobos, and thereafter shaped their host genomes by selection. Indeed, selection in the APOBEC3D gene for its increased activity against lentiviruses occurred at the chimpanzee/bonobo common ancestor [25]. It is tempting to speculate infection by a virus similar to SIVsmm might have driven this selection. An ancient selective event is further supported by the evidence of a selective sweep of the MHC class I repertoire in chimpanzees [52]. Therefore, the antiviral defenses of the APOBEC3 locus of modern-day chimpanzees and bonobos may have been shaped by ancient lentiviruses and have recently been under continued selection, such that now they are particularly resistant to most SIV cross-species transmissions.
Expression plasmids for chimpanzee APOBEC3 genes were previously described; briefly, chimpanzee APOBEC3D was cloned into pCS2+ with a 3’end HA epitope tag [25] and chimpanzee APOBEC3G, APOBEC3H, and APOBEC3F were cloned into pcDNA3.1 vector with a 5’end HA epitope tag [4,33]. All these APOBEC3 genes are from the Pan troglodytes verus chimpanzee subspecies. APOBEC3 mutants bearing the newly identified SNPs (Fig 4) were made by site-directed mutagenesis of the APOBEC3 plasmids (Quick Change II Site Directed Mutagenesis Kit, Agilent Technologies).
Recombinant HIV-1ΔVifΔEnvLuc2 proviral plasmids encoding vif from SIVsmm E041, SIVwrc 98CI04, SIVolc 97CI12, SIVmus-1 CM1085, SIVdeb CM5 were previously described [20], the ones with vif from SIVagm.sab-1, SIVagm.ver-90, SIVagm.gri-667, SIVagm.tan-1 are from Compton et al. 2012 [17], the ones with vif from SIVrcm CM8081, SIVcpzPts TAN3, SIVcpzPts UG38, SIVcpzPts TAN13, SIVcpzPtt Gab1, SIVcpzPtt MB66 are from Etienne et al. 2013 [4], and the ones with vif from SIVcpzPts BF1167, SIVcpzPtt EK505, and SIVcpzPtt DP943 are from this study. All were cloned as previously described [4]. The various Vifs are expressed at similar expression levels and have the capacity to fully antagonize at least one APOBEC3 protein [4,17,20].
For the spreading infections, replication-competent recombinant HIV-1 proviral plasmids encoding vif from SIVsmm E041, SIVagm.sab-1 (named SIVsab), SIVcpzPts TAN3, and SIVrcm CM8081 were cloned into an HIV-1 backbone in place of the HIV-1 vif and all harbor an intact env gene.
293T cells (obtained from the American Type Culture Collection (ATCC)) were co-transfected with 400 ng of APOBEC3 plasmid or an empty expression vector, 600 ng of proviral HIV-1 plasmid (HIV-1ΔVifΔEnvLuc2) with an SIV vif, and 200 ng of L-VSV-G for pseudotyping, using TransIT-LT1 (Mirus Bio). The cells and the virus supernatant were collected 48–72h post transfection. The harvested cells were used for the western-blot analyses and the following antibodies were used: mouse HA-specific antibody (Balco), mouse anti-tubulin and anti-actin antibodies (Sigma-Aldrich), and secondary goat anti-mouse horseradish peroxidase-conjugated antibody (Sigma-Aldrich). The total amount of virus in the supernatant was quantified by p24 Gag ELISA assay (Advanced Bioscience Laboratories). Each transfection condition was performed in 2–3 independent experiments. For the infection, SupT1 cells (obtained from the NIH AIDS Repository) were plated at 0.4 M cells/ml with 20 μg/ml of diethylminoethyl-dextran and infected with 2 ng of virus. Infections were performed in triplicate and luciferase activity was measured after 72h with the Bright-Glo Luciferase Assay Reagent (Promega).
293T cells were transfected with 1,200 ng of replication-competent proviral HIV-1 plasmids with an SIV vif using TransIT-LT1 (Mirus Bio). The virus supernatants were collected 48–72h post transfection. The total amount of virus in the supernatant was quantified by p24 Gag ELISA assay (Advanced Bioscience Laboratories) or p24 Gag AlphaLISA assay (Perkin Elmer).
Leftover blood samples from health examinations of uninfected chimpanzees housed at the Yerkes Regional Primate Center were shipped at room temperature and peripheral blood mononuclear cells (PBMCs) were isolated by gradient centrifugation using Ficoll-Paque Plus (GE Healthcase Life Sciences). These chimpanzee PBMCs were enriched for CD4+ T cells using non-human primate CD4 MicroBeads (MACS Miltenyi Biotec) and magnetic cell sorting (Militenyi Biotec), stimulated with staphylococcal enterotoxin B (Sigma-Aldridge) for 12 to 15 hours (3 μg/ml), and subsequently co-cultivated with autologous monocyte-derived macrophages for optimal activation as described [26]. 250 ng of each viral construct were used to infect 106 activated primary chimpanzee CD4+ T cells [26]. After 12h, cells were washed four times with PBS and were resuspended in new media. 50 μl of supernatant were collected every 48h up to 9–10 days post-infection. The total amount of virus at each time point was measured using the p24 Gag PE AlphaLISA assay. The experiment was performed using primary CD4+ T cells from three different chimpanzee donors also obtained from the Yerkes Primate Center. For the replication experiments in the presence of IFN, activated chimpanzee CD4+ T cells were pretreated for 24h with 500 U/ml of human IFN-alpha2 or 100 U/ml of human IFN-beta prior to infection. Each IFN was replenished every 48h throughout the replication kinetic. The experiment in the presence of IFN was performed using primary CD4+ T cells from two chimpanzee donors.
Human SupT1 cells were infected with 2 ng of each viral construct, using 20 μg/ml of diethylminoethyl-dextran, and spinoculated for 2h at 1,600 rpm. After 12h, cells were washed four times with PBS and were resuspended in new media. 50 μl of supernatant were collected every 48h up to nine days post infection. The total amount of virus at each time point was measured using ABL p24 Gag ELISA assay.
Primary chimpanzee CD4+ T cells from the donor 1 were harvested nine days post infection. Total DNA was extracted using the QIAamp DNA mini kit (QIAGEN). Two fragments were amplified using the AccuPrime Taq DNA polymerase (Invitrogen) with 40 cycles of amplifications. One small fragment encompassed the vif gene and was of approximately 600 bp (size depends on amplified SIV vif; primers are as followed Primer-vif-F, 5’-CAG CAA AGC TCC TCT GGA AAG GT-3’, and Primer-vif-R, 5’-CTA TGT CGA CAC CCA ATT CTG AAA TG-3’). A second fragment that starts in the pol gene and finishes in the vpr gene was of approximately 1,200 bp, and primers Primer-pol-F, 5’-GAA TTT GGA ATT CCC TAC AAT CCC C-3’, and Primer-vpr-R, 5’-CTA CTG GCT CCA TTT CTT GCT CTC C-3’, were used for amplification. Amplicons were gel purified and cloned using the TOPO TA cloning system (Invitrogen). Between nine and 25 clones were sequenced per sample and the total number of nucleotides analyzed per sample were 7,688–27,577 bp, as shown in Table 2. The hypermutation significance was calculated using two methods. First, we used Hypermut 2.0, which determines if the mutations in a given G-to-A context exceed the mutations out of context using a Fisher test, with a threshold of p<0.05 [30]. Second, we used Hyperfreq, which estimates the relative probability of G-to-A mutation in a given context versus a control context using a Bayesian approach [31]. The advantage of this second method is that it also evaluates the “strength” of various hypermutation contexts [31]. Here, we ran the Hyperfreq analyses using a level of significance of 0.05 in three different contexts: a GG context (consistent with primarily APOBEC3G activity), a GA context (consistent with primarily APOBEC3F and/or APOBEC3D activities), and a GR context (consistent with a combined activity of APOBEC3G and APOBEC3F and/or APOBEC3D).
In order to assess the diversity of the APOBEC3 gene family amongst wild chimpanzee and bonobo populations, we leveraged full-genome shotgun sequencing data of 25 common chimpanzees and 11 bonobos sequenced as part of the Great Ape Genome Sequencing Project [34]. Representatives from each of the four recognized common chimpanzee subspecies were included, five western chimpanzees, ten Nigerian-Cameroonian chimpanzees, four central chimpanzees, and six eastern chimpanzees. The raw reads for all chimpanzee and bonobo individuals were mapped to the complete chimpanzee reference genome PanTro3 using BWA-MEM (arXiv:1303.3997v2 [q-bio.GN]) with default parameters. SNPs were called using Samtools and default parameters with all individuals combined during the calling step [53].
We confirmed the deep-sequencing results of a subset of polymorphic sites (n = 16) in the chimpanzee APOBEC3 genes by Sanger sequencing. PCRs were performed from total genomic DNA (a kind gift from Evan Eichler [34]), using APOBEC3 primers (for APOBEC3G: F, 5’- TTT GGA GGC TCT AGC AAG TGA GTG-3’ and R, 5’- AGC TAC AGG AAG CAC AGG TGA-3’; for APOBEC3D: F, 5’- CCT GCC CTC TTC TCC CAT CG-3’ and R, 5’- CGT AGC ATT GTT TTC AGA AGT CG-3’; for APOBEC3F: F, 5’- TCC GCC CTC TGC TCT CAT C-3’ and R, 5’- CTG CAG CTT GCT GTC CAG GAA TAG-3’) and Accuprime Pfx DNA polymerase (Invitrogen). After gel purification, all the amplicons were sequenced and the SNPs were confirmed in all cases, including heterozygous and homozygous sites (S7 Fig).
The sequences of Vif from the SIV lineages were obtained from the Los Alamos HIV sequence database (http://www.hiv.lanl.gov/). The amino acid sequences were aligned using FSA [54] and the phylogenetic analyses were performed using PhyML with a JTT model [55].
|
10.1371/journal.pntd.0006025 | L-arginine availability and arginase activity: Characterization of amino acid permease 3 in Leishmania amazonensis | Leishmania uses the amino acid L-arginine as a substrate for arginase, enzyme that produces urea and ornithine, last precursor of polyamine pathway. This pathway is used by the parasite to replicate and it is essential to establish the infection in the mammalian host. L-arginine is not synthesized by the parasite, so its uptake occurs through the amino acid permease 3 (AAP3). AAP3 is codified by two copies genes (5.1 and 4.7 copies), organized in tandem in the parasite genome. One copy presents the expression regulated by L-arginine availability.
RNA-seq data revealed 14 amino acid transporters differentially expressed in the comparison of La-WT vs. La-arg- promastigotes and axenic amastigotes. The 5.1 and 4.7 aap3 transcripts were down-regulated in La-WT promastigotes vs. axenic amastigotes, and in La-WT vs. La-arg- promastigotes. In contrast, transcripts of other transporters were up-regulated in the same comparisons. The amount of 5.1 and 4.7 aap3 mRNA of intracellular amastigotes was also determined in sample preparations from macrophages, obtained from BALB/c and C57BL/6 mice and the human THP-1 lineage infected with La-WT or La-arg-, revealing that the genetic host background is also important. We also determined the aap3 mRNA and AAP3 protein amounts of promastigotes and axenic amastigotes in different environmental growth conditions, varying pH, temperature and L-arginine availability. Interestingly, the increase of temperature increased the AAP3 level in plasma membrane and consequently the L-arginine uptake, independently of pH and L-arginine availability. In addition, we demonstrated that besides the plasma membrane localization, AAP3 was also localized in the glycosome of L. amazonensis promastigotes and axenic amastigotes.
In this report, we described the differential transcriptional profiling of amino acids transporters from La-WT and La-arg- promastigotes and axenic amastigotes. We also showed the increased AAP3 levels under amino acid starvation or its decrease in L-arginine supplementation. The differential AAP3 expression was determined in the differentiation of promastigotes to amastigotes conditions, as well as the detection of AAP3 in the plasma membrane reflecting in the L-arginine uptake. Our data suggest that depending on the amino acid pool and arginase activity, Leishmania senses and could use an alternative route for the amino acid transport in response to stress signaling.
| Leishmania alternates its life cycle between the invertebrate host, in which the promastigote forms reside at pH 7.0 and approximately 25°C; and the mammalian host, in which the amastigote forms reside at pH 5.0 and approximately 37°C. These environmental changes submit the parasite to dynamic undergo modifications in morphology, metabolism, cellular signaling and gene expression regulation to allow for a rapid adaptation to the new environmental conditions. Leishmania is auxotrophic for many amino acids, such as L-arginine. The L-arginine availability is important for the uptake control by the parasite as well as the intracellular amino acids pool maintenance. Leishmania arginase uses L-arginine to produce urea and ornithine, being the last one a precursor of polyamine biosynthetic pathway, which is used by the parasite to replicate and to establish the infection. L-arginine uptake in L. amazonensis is performed through AAP3, which is encoded by two gene copies arranged in tandem on the genome (5.1 and 4.7 aap3). In this report, we characterized the AAP3 function and expression regulation in parasites maintained in different pH and temperature conditions simulating both insect and mammalian micro-environment. We submitted the parasites to amino acid starvation, simulating mid-gut starvation, a signal for promastigote metacyclogenesis. Our results demonstrated that the changes in temperature and pH, in addition with amino acid starvation or L-arginine supplementation might represent important signals for aap3 expression regulation, mainly for the 5.1 aap3 copy. We also linked the regulation of 5.1 aap3 transcript to the genetic background from the host macrophage. In addition, we localized the AAP3 in the plasma membrane and in the glycosome of L. amazonensis promastigotes and axenic amastigotes, indicating that arginine uptake is directed to this organelle.
| Leishmania is the causative agent of leishmaniasis, a complex disease characterized by cutaneous, mucocutaneous or visceral lesions [1–3]. It is currently endemic in 98 countries and territories around the world, with an annual incidence estimated at 1 million cases of cutaneous leishmaniasis and 300,000 cases of visceral leishmaniasis [4]. In its life cycle, the parasite alternates between the intestinal tract of the sand fly (promastigote form) and the phagolysosome compartment of mammalian host macrophages (amastigote form). These environmental changes submit the parasite to undergo extensive modifications and can be considered a trigger for the gene expression regulations that lead to adaptation to the new milieu.
Leishmania infection results in the specific activation of mammalian immune responses. Macrophages have a fundamental role in infection as the first line of defense. The production of nitric oxide (NO), a potent molecule effective against pathogens, is one of the key defense mechanisms of mammalian phagocytes [5]. NO is produced by nitric oxide synthase 2 (NOS2) using the amino acid L-arginine as substrate. Several studies have demonstrated that immune responses to infectious pathogens are strictly dependent on the expression of NOS2 [1,5,6]. Arginase is an immune-regulatory enzyme that can reduce the NO production by macrophages, limiting L-arginine availability to NOS2 and inducing resistance of some pathogens to host defense mechanisms [7–10]. In contrast, Leishmania also has arginase, which uses L-arginine to produce urea and ornithine as part of polyamine biosynthetic pathway, essential for the parasite replication and establishment of the infection [8,10–15].
The L-arginine uptake in macrophages is mediated by the cationic amino acid transporter family (CAT), such as cationic amino acid transporter 2B (CAT2B) in Leishmania infection [16]. By contrast, Leishmania has a selective uptake of this amino acid by AAP3 [17,18]. L. amazonensis AAP3 is encoded by two gene copies (5.1 and 4.7 aap3), arranged in tandem on the genome. The open reading frame (ORF) sequence shows 98% identity between the two copies and 93% identity to the AAP3 of L. donovani (LdAAP3) [18].
Previous studies have demonstrated that L. donovani responded to amino acid starvation by increasing both mRNA and protein levels of the L-arginine transporter LdAAP3 [17,19]. Castilho-Martins et al. (2011) described an increase in the 5.1 aap3 mRNA half-life in L. amazonensis [18]. Leishmania also has mechanisms that sense both external and internal concentrations of L-arginine and can respond with an increase in amino acid uptake [17,18]. Therefore, L-arginine uptake control might be an important factor in the resistance of some pathogens to host defense mechanisms [14,20].
RNA-seq data analyses revealed 14 amino acid transporters differentially expressed in La-WT and La-arg- promastigotes and axenic amastigotes in the stationary growth phase. The differentiation from promastigotes to axenic amastigotes, independent of arginase activity, showed up-regulation of some these transporters, which may be involved in the Leishmania polyamine biosynthetic pathway. In fact, the 5.1 and 4.7 aap3 transcripts were down-regulated in axenic amastigotes when compared to the promastigote expresson. Up-regulation of other transporters was also identified, as the amino acid transporter aATP11, suggesting that Leishmania senses the amino acid pool and regulates gene expression to use alternative route for parasite survival.
In this work, we showed that changes in pH, temperature as well as L-arginine availability and the background from the host macrophage canmodulated the aap3 mRNA expression and the AAP3 protein amount. We also showed the AAP3 plasma membrane localization correlated with the arginine uptake in La-WT mid-logarithmic growth phase promastigotes. The change conditions aimed to simulate the environmental changes of the parasite in its life cycle from the sand fly to the mammalian host. Furthermore, we demonstrated that in addition to its plasma membrane localization, AAP3 was also localized partially in the glycosome of promastigotes and axenic amastigotes forms of the parasite, indicating that arginine uptake can be directed to this compartmentalized organelle, supplementing the polyamine production.
L. amazonensis (MHOM/BR/1973/2269) wild type (La-WT) and L. amazonensis arginase knockout (La-arg-) [8] promastigotes were grown at 25°C in M199 medium supplemented with L-glutamine, 10% heat-inactivated fetal calf serum, 0.25% hemin, 12 mM NaHCO3, 100 μM adenine, 40 mM HEPES, 50 U/mL penicillin and 50 μg/mL streptomycin, at pH 7.0. La-WT and La-arg- axenic amastigotes were grown in M199 medium supplemented, as described above, at 34°C and pH 5.5 [21,22]. La-arg- cultures were grown in M199 supplemented as described above with hygromycin (30 μg/mL), puromycin (30 μg/mL) and putrescine (50 μM) addition.
For in vitro macrophages infection, bone marrow-derived macrophages (BMDMs) from BALB/c or C57BL/6 mice were derived from the femurs and tibiae of females (6-8-weeks) from the Animal Center of the Institute of Bioscience of the University of São Paulo. The femurs and tibiae were washed with cold PBS and the cells were collected at 500 x g for 10 min at 4°C. After lysis of erythrocytes, the cells were maintained in RPMI 1640 medium (LGC Biotecnologia, São Paulo, SP, Brazil), supplemented with penicillin (100 U/ml) (Life Technologies, Carlsbad, CA, USA), streptomycin (100 μg/ml) (Life Technologies, Carlsbad, CA, USA), 5% heat-inactivated FBS (Life Technologies, Carlsbad, CA, USA) and 20% L9-29 supernatant. The cells were cultivated for 7 days at 34°C and 5% CO2. After differentiation, cellular viability was evaluated with Trypan blue staining 1:1, and cells were counted in Neubauer chamber. Approximately 2x105 BMDMs were incubated on sterile 8 wells glass chamber slide (Lab-Teck Chamber Slide; Nunc, Naperville, IL, USA), overnight at 34°C and 5% CO2 to adhere to the coverslips. Non-adherent cells were removed by PBS washing.
THP-1 human monocytic cell line was maintained in culture at the same conditions for BMDMs. Differentiation was performed plating 5×105 cells in 8 wells chamber slide with 30 ng/mL of phorbol 12-myristate 13-acetate (PMA) (Sigma-Aldrich, St Louis, MO, USA) diluted in RPMI 1640 medium for 72 h, followed by a 72-h resting phase with fresh RPMI 1640 medium before infection.
The infection was performed with La-WT or La-arg- stationary growth phase promastigotes (MOI 5:1). After 4 h of infection, non-phagocytized parasites were washed with PBS and the cells were collected after 4, 24 and 48 h. Non-infected macrophages maintained in culture in the same conditions were used as control. The infections were evaluated by determining the percentage of infected cells after counting 200 Panoptic-stained (Laborclin, Parana, Brazil) macrophages. The infection index was determined by multiplying the percentage of infected macrophages by the mean number of parasites per infected cell [23,24]. Statistical analyses were performed using non-parametric two-tailed Student t tests.
Total RNA of BMDMs from BALB/c or C57BL/6 mice, and THP-1 derived macrophage infected with La-WT and La-arg- promastigotes; and La-WT promastigotes in different conditions of temperature, pH and amino acid starvation or L-arginine supplementation were isolated using TRIzol reagent (Life Technologies, Carlsbad, CA, USA), according to the manufacturer’s instructions. RNA samples were treated with DNase I (Thermo Scientific, Lithuania, EU) and RNA concentration was determined using a spectrophotometer at A260/A280 (Nanodrop ND1000, Thermo Scientific, USA).
For RNA-seq, total RNA from La-WT and La-arg- promastigotes and axenic amastigotes in the stationary growth phase were isolated using TRIzol reagent (Life Technologies, Carlsbad, CA, USA), according to the manufacturer’s instructions. RNA samples were treated with DNase I (Thermo Scientific, Lithuania, EU). Then, RNA concentration was determined using a spectrophotometer at A260/A280 (Nanodrop ND1000, Thermo Scientific, USA). In addition, RNA integrity was assessed using Agilent 2100 Bioanalyzer and Pico Agilent RNA 6000 kit (Agilent Technologies, Santa Clara, CA, USA), according to the manufacturer’s instructions.
Reverse transcription was performed using 2 μg of total RNA as a template, reverse transcriptase and random primers (Revertaid H minus Reverse Transcriptase kit, Thermo-Scientific, Canada), according to the manufacturer’s instructions. Equal amounts of cDNA were assessed in triplicate in a total volume of 25 μL containing Maxima SYBR Green qPCR Master Mix (Thermo Scientific, Lithuania, EU) and the following primers (200 nM): AAP3_F (5.1 UTR) 5´-GGTCCCCGATACACACATTC-3´, AAP3_R (5.1 UTR) 5´-GTCTCCCGTTTTGCAAGAGA-3´, AAP3_F (4.7 UTR) 5´-ACCATTGTGGGTTAGTTATACATCC-3´, AAP3_R (4.7 UTR) 5´-CAAGATCGC TAGCAGTGGAG-3´, GAPDH_Leishmania_F 5´-TCAAGGTCGGTATCAACGGC-3´ and GAPDH_Leishmania_R 5´-TGCACCGTGTCGTACTTCAT-3´. The mixture was incubated at 94°C for 5 min, followed by 40 cycles at 94°C for 30 s and 60°C for 30 s. A negative control in the absence of reverse transcriptase was included in RT-qPCR assays to detect DNA contamination in RNA samples. Reactions were carried out using an Exicycler 96 (Bioneer, Daejeon, Korea). The copy number of the target genes (aap3 5.1 and aap3 4.7) and reference gene (gapdh) were quantified in three biological replicate samples, considering the molar mass concentration, according to a standard curve generated from a ten-fold serial dilution of a quantified and linearized plasmid containing the target fragment for each quantification test. The normalized aap3/gapdh ratio of the absolute number of molecules of each target was used as the parameter to calculate the relative expression. Analyses were performed using Analysis Exicycler3 Software (Bioneer, Daejeon, Korea).
cDNA library preparations were performed using Stranded-specific TrueSeq RNA-seq Library Prep (Illumina), according to the manufacturer´s instructions.
Paired-end reads (125 bp) were obtained using the Illumina HiSeq 2000 platform at the Norwegian Sequencing Centre at the University of Oslo. Trimmomatic was used to remove the Illumina adapter sequences [25]. The quality of the produced data was analyzed using FastQC by Phred quality score [26]. Reads with Phred quality scores lower than 20 were discarded. Reads were aligned to the L. mexicana (MHOMGT2001U1103) genomic data obtained from TriTrypDB (tritrypdb.org/tritrypdb/) using TopHat [27,28]. Thereafter, read mapping was performed for transcript assembly using Cufflinks [29]. After assembly, the abundance of transcripts was calculated as the Fragments Per Kilobase of transcript per Million mapped reads (FPKM), which reflects the abundance of a transcript in the sample by normalization of the RNA length and the total read number [30]. Differentially expressed gene analysis was performed on four comparison pairs (La-WT promastigotes vs. La-arg- promastigotes; La-WT axenic amastigotes vs. La-arg- axenic amastigotes, La-WT promastigotes vs. La-WT axenic amastigotes; La-arg- promastigotes vs. La-arg- axenic amastigotes) [22].
Promastigotes in mid-logarithmic growth phase (day 4 of culture) were washed with Earl´s Salt Solution (EBSS) (LGC Biotecnologia, São Paulo, SP, Brazil) at pH 5.0 or pH 7.0. Then, cells were starved of amino acids or supplemented with 400 μM L-arginine for 4 h at 25 or 34°C. The control parasites were those collected before starvation and/or L-arginine supplementation [18].
Arginine uptake assays were performed after amino acids starvation or L-arginine supplementation, as previously described [31,32]. Briefly, 5x107 promastigotes in the mid-logarithmic growth phase were washed twice with EBSS medium, resuspended in PBS and incubated at 25°C or 34°C for 3 min. Then, 3H-arginine (1mCi/43Ci/mmol) (GE Healthcare, UK) was added. The uptake was stopped at different times by adding ice cold arginine. The parasites were washed twice with EBSS and the radioactivity was measured by liquid scintillation spectrometry Perkin-Elmer TRI-CARB 2910TR.
The epitope (ILYNFDPVNQP) designed for a specific region of AAP3 through a high affinity MHC was synthesized and used to produce a rabbit anti-AAP3 polyclonal antibody by Proteimax Biotechnology (São Paulo, SP, Brazil).
Approximately 107 parasites in the different conditions were washed with PBS and then lysed with lysis buffer (100 mM Tris-HCl pH 7.5, 2% Nonidet P40, 1 mM PMSF and protease inhibitor cocktail (Sigma-Aldrich, St Louis, MO, USA)). Cells were disrupted by five freeze/thaw cycles in liquid nitrogen and 42°C, and then were cleared of cellular debris by centrifugation at 12,000 x g for 15 minutes at 4°C. Equal amounts of total protein (50 μg) were solved using SDS-PAGE and then transferred to a nitrocellulose membrane (LI-COR Bioscience, Lincoln, NE, USA) using a Trans-Blot Semi-Dry apparatus (Bio-Rad, USA). The membrane was incubated with Blocking Buffer (LI-COR Bioscience, Lincoln, NE, USA) and then with anti-AAP3 serum (1:500 dilution), overnight, at 4°C. After incubation with primary antibody, the membrane was incubated with goat anti-rabbit DyLight 680 conjugated antibody (Thermo Scientific, IL, USA) (1:10000 dilution) for 1 h at room temperature. Anti-α-tubulin (Sigma-Aldrich, St. Louis, MO, USA) (1:5000 dilution) was used to normalize the amount of protein in the blot. All steps were followed by washing 3 times with PBS. The membranes were scanned using an Odyssey CLx apparatus (Li-COR, Lincoln, NE, USA) in 700 channel using an Odyssey System. Odyssey Imaging CLx instrument was used at an intensity setting of 5 (700 nm).
Approximately 106 promastigotes on days 3, 5, 7 and 9 of a growth curve, or in the mid-logarithmic growth phase after amino acid starvation or L-arginine supplementation in pH 7.0 or 5.0 maintained at 25°C or 34°C were washed with PBS and then fixed in 1% of paraformaldehyde (4°C, overnight). For the analysis of AAP3 on the external face of the plasma membrane, the cells were incubated with anti-AAP3 serum (1:500 dilution) at 4°C with overnight shaking. Then, the cells were incubated with goat anti-rabbit FITC conjugated antibody (Sigma-Aldrich, St. Louis, MO, USA) (1:500 dilution) at room temperature with shaking for 1 h. For the analysis of total AAP3, the cells were permeabilized with 0.05% Tween-20 for 20 min at room temperature. Then they were incubated with anti-AAP3 and anti-rabbit FITC, as previously described. The cells were analyzed using FlowSight image flow cytometer (Amnis-MerckMillipore, Darmstadt, Germany). 10,000 cells were acquired, sorted and analyzed using the gate based in gradient root mean square (RMS). Single cells were analyzed using the gate based in the bright field channel and fluorescence intensity of AAP3 in channel 2. Data were acquired using Inspire and analyzed using Ideas Software (Amnis Corporation, Seattle, WA, USA). All analysis was performed at the Core Facility of the Centro de Aquisição de Imagens e Microscopia from Instituto de Biociências (Caimi-IB) at the University of São Paulo.
Approximately 106 promastigotes La-WT, La-EGFP/SKL [33] or La-WT axenic amastigotes in the stationary growth phase were washed with PBS and adhered to coverslips treated with poly-L-lysine (Sigma-Aldrich, St. Louis, MO, USA) for 15 min. The cells were then fixed with 2% paraformaldehyde for 10 min at room temperature. The fixed cells were permeabilized and blocked with 0.1% Triton X-100 and 0.1% BSA in PBS for 1 h at room temperature. To analyze sub-cellular AAP3 localization, anti-AAP3 polyclonal antibody (1:500 dilution) was visualized using an anti-rabbit secondary antibody conjugated to Alexa546 or Alexa 488 (Life Technologies, Carlsbad, CA, USA) (1:500 dilution). Anti-α-tubulin (Life Technologies, Carlsbad, CA, USA) (1:1000 dilution) was visualized using an anti-mouse secondary antibody conjugated to Alexa594 (Life Technologies, Carlsbad, CA, USA) (1:500 dilution). Nuclear and kinetoplast DNA were labeled using DAPI. Each step was followed by washing 10 times with PBS. The coverslips were mounted in ProLong media (Life Technologies, Carlsbad, CA, USA). All imaging was performed using confocal microscope (Zeiss LSM 780 NLO) at the Core Facility of the Centro de Facilidades para Pesquisa (CEFAP) at the University of São Paulo.
The experimental protocols for the animals were approved by the Animal Care and Use Committee from the Institute of Bioscience of the University of São Paulo (CEUA 233/2015). This study was carried out in strict accordance with the recommendations in the guide and policies for the care and use of laboratory animals of the São Paulo State (State Law 11.977, de 25/08/2005) and Brazil government (State Law 11.794, de 08/10/2008).
Transcriptomic profiling by RNA-seq was used to identify differential gene expression in La-WT and La-arg- promastigotes and axenic amastigotes. Sequencing data obtained are available on the NCBI BioProject under accession number PRJNA380128 and Sequence Read Archive (SRA) under accession number SRX2661998 and SRX2661999 [22].
More than one billion sequence reads were obtained by Illumina HiSeq2000. Data were aligned to the L. mexicana genome (MHOMGT2001U1103), and 8253 transcripts, 180 hypothetical proteins and 443 novel transcripts were identified.
Based on the DE genes analyzed, we identified 14 amino acid transporters differentially expressed in the comparisons La-WT and La-arg- promastigotes and axenic amastigotes. As shown in Tables 1 and 2, we observed a down-regulation of both 5.1 and 4.7 aap3 in La-WT and La-arg- promastigotes and axenic amastigotes. In contrast, we observed up-regulation of other amino acid transporters.
Then, to investigate this modulation, we analyzed the changes in environmental signals that could regulate this gene expression, such as pH, temperature and L-arginine availability, as intrinsic factors that can influence the differentiation of the parasite life cycle from the sand fly to the mammalian macrophage host.
BMDMs from BALB/c or C57BL/6 mice or human lineage THP-1 derived macrophages were infected with La-WT or La-arg- (MOI 5:1) promastigotes and the infection index was determined at 4, 24 and 48 h post-infection. We did not observe differences in the infection index of BMDMs from BALB/c infected with La-WT or La-arg- after 4 and 24 h. A lower infection index was observed in BMDMs from BALB/c infected with La-arg- after 48 h compared to La-WT (Fig 1A), corroborating with previous data and indicating the importance of arginase activity to stablish the infection [8]. The infection index of BMDM from C57BL/6 infected with La-WT or La-arg- only presented significant difference after 48 h of infection (Fig 1B). In contrast, the infection index from THP-1 macrophages with La-WT was increased after 24 and 48 h of infection. And the infection index with La-arg- was lower in all time infections when compared to La-WT (Fig 1C).
In addition, we determined the 5.1 and 4.7 aap3 amount in the preparations from macrophages from BALB/c, C57BL/6 and THP-1 infected with La-WT or La-arg-. The 5.1 aap3 mRNA amount presented an increase during the time course of macrophages from BALB/c infected with La-WT (Fig 1D). The absence of arginase activity did not change the 5.1 aap3 mRNA amount during the time course of infection with La-arg-, but it was lower when compared to La-WT infection (Fig 1D). The 5.1 aap3 amount in macrophages from C57BL/6 mice infected with La-WT also did not change during the time course of infection. (Fig 1E). And in these C57BL/6 macrophages, the absence of arginase activity showed lower expression when compared to La-WT infection after 24 and 48 h (Fig 1E). The 5.1 aap3 amount in human THP-1 macrophage increased during the time course of infection with La-WT. Interestingly, in these THP-1 macrophages, the 5.1 aap3 amount was higher in La-arg- compared to La-WT after 4h of infection, and decreased after 24 and 48 h of infection (Fig 1F). Furthermore, the 4.7 aap3 amount appeared up-regulated during the time course of infection in macrophages from BALB/c infected with La-WT (Fig 1G). However, did not appear altered in macrophages from BALB/c or C57BL/6 or THP-1 with La-arg- infection (Fig 1G, 1H and 1I, respectively). Interestingly, as observed for 5.1 aap3, the 4.7 aap3 amount in THP-1 macrophages was higher in La-arg- compared to La-WT after 4h of infection (Fig 1I).
La-WT promastigotes in mid-logarithmic growth phase were starved of amino acids or supplemented with 400 μM L-arginine and incubated at 25°C or 34°C for 4 h. Then, total RNA was extracted and the 5.1 and 4.7 aap3 mRNA were quantified by RT-qPCR. Data were normalized to the gapdh transcripts in each condition. As shown in Fig 2A, in parasites maintained at 25°C and pH 7.0, the starvation of L-arginine caused a significant increase in 5.1 aap3 transcript level compared with parasites that were supplemented with L-arginine. Interestingly, at 34°C, an increase in 5.1 aap3 transcripts was observed during starvation of L-arginine at pH 5.0 (Fig 2A). By contrast, a significant decrease of the same transcript was observed at pH 7.0 during starvation as well asin parasites submitted to L-arginine supplementation. The decrease was also detected at pH 5.0 during L-arginine supplementation (Fig 2A).
Although no significant difference was observed in the amount of 4.7 aap3 transcripts in all conditions, a slight decrease in the mRNA was observed when the parasites were submitted to amino acid starvation at 25°C (Fig 2B). The profiles at 25°C and 34°C were very similar, except at 34°C in pH 5.0, when starvation slightly increased the mRNA level (Fig 2B). Our data suggest that L-arginine availability and increased temperature regulates only 5.1 aap3 mRNA expression.
To assess the amount of AAP3 present on the external face of the plasma membrane and the total AAP3 present in the whole cell, we performed a flow cytometry analysis of fluorescence labeled AAP3 antibody against non-permeabilized and permeabilized parasites. The fluorescence intensity values were normalized to the control (parasite collected before amino acid starvation or L-arginine supplementation). Initially, the total amount of AAP3 protein was measured during the time course of the growth curve. We observed an increase of AAP3 protein on days 3, 5, 7 and 9 compared to day 2 (S1 Fig). This data indicated that the increase of AAP3 in the plasma membrane was related from mid-logarithmic to late-stationary growth phase. Then, the total amount of AAP3 was measured during pH and temperature changes, and L-arginine availability.
The total amount of AAP3 protein was increased in parasites kept at 25°C and pH 7.0 under amino acid starvation, compared to the control parasites, but not in the parasites at pH 5.0 (Fig 3A). Interestingly, a decreased of total amount was observed in parasites at 25°C, pH 7.0 and supplemented with L-arginine when compared to parasites under amino acids starvation. No difference in the total AAP3 amount was observed in parasites at 34°C and pH 7.0, compared to the control parasites. In contrast, an increase in the total amount was observed in parasites at pH 5.0 independent of amino acids starvation or L-arginine supplementation (Fig 3A). Furthermore, the plasma membrane amount of AAP3 was measured and we observed that the parasites under amino acid starvation at 25°C in pH 7.0 presented increased AAP3 amount in the membrane compared to the control parasites (Fig 3B). A decreased in the membrane amount was observed in parasites at 25°C, pH 7.0, supplemented with L-arginine when compared to parasites under amino acids starvation. No significant difference was observed at pH 5.0 under amino acids starvation, compared to control parasites. However, an increased membrane AAP3 amount was observed at pH 5.0 under L-arginine supplementation when compared to the control parasites and to those at pH 7.0. The AAP3 membrane amount was increased at 34°C, in both pH 7.0 and pH 5.0, and in amino acid starvation or supplementation of L-arginine (Fig 3B). The differences in the total and plasma membrane AAP3 amount reflected the increase in the protein expression and in the traffic of the transporter carrier to the membrane. The increase in total and plasmatic membrane AAP3 amount occurred in amino acid starvation at 25°C and pH 7.0, and at 34°C and pH 5.0, contrasting with the amount of total AAP3 and plasmatic membrane in L-arginine supplementation at 34°C (Fig 3C).
In addition, we performed Western blot analysis of cell lysates of La-WT promastigotes and axenic amastigotes during the time course of growth curve, and no difference was observed in the AAP3 protein level (S2A Fig). Compared with the flow cytometry analysis results of total AAP3 protein (Fig 3C), the protein level detected by Western blotting showed a similar profile in La-WT promastigotes after amino acid starvation or L-arginine supplementation at both 25°C and 34°C (S2C and S2D Fig).
To evaluate and correlate the results obtained for mRNA and protein levels, we analyzed L-arginine uptake in parasites submitted to the same pH and temperature changes, and L-arginine availability.
The L-arginine uptake increased during the amino acid starvation at pH 7.0 (Fig 4), demonstrating a correlation with the increase exhibited by 5.1 aap3 mRNA in the same conditions (Fig 2A). Changing the parasites to 34°C caused a 3-times increase in the rate of L-arginine uptake independent of pH, as well as amino acid starvation or L-arginine supplementation (Fig 4).
Using the anti-rabbit antibody against the epitope AAP3, confocal microscopy analysis was performed to localized that transporter in the parasites. Confocal images from La-WT promastigotes and axenic amastigotes in the stationary growth phase, and La-GPF/SKL promastigotes in the stationary growth phase confirmed the AAP3 localization in the plasmatic membrane as well as partially in the glycosome (Fig 5).
The importance of the amino acid L-arginine in Leishmania has been related to parasite replication as well as a requirement to establish the infection in the mammalian host [8,17,32,34][22,35]. In the course of the Leishmania life cycle, environmental changes may act as signals to regulate gene expression, which enables the parasite to adapt to the new conditions [7].
Stress signaling starts with the consumption of all available nutrients at the end of blood meal digestion in the insect's digestive tract. The starvation signal can cause the release of procyclic promastigotes from the insect mid-gut epithelia and its migration to the proboscis, promoting the differentiation of procyclic promastigotes into metacyclic promastigotes [36]. The deprivation of amino acids is an important signal for metacyclogenesis, providing the parasites with the infective capacity to establish the infection.
The shift of temperature, from the sand fly (25°C) to the mammalian body temperature (37°C), is the next shock and is also known to be an important signal for parasite differentiation. The heat-shock proteins are good examples of gene activation that allow parasite survival in rapid temperature changes [37–39]. The pH change is the following step upon the fusion of the phagosome to the lysosome in the formation of the phagolysosome [37,40,41].
In this context, we described here how the parasite was able to regulate the AAP3 expression in response to the L-arginine availability, arginase activity, pH and temperature modifications. AAP3 is an amino acid transporter described for L-arginine uptake in L. amazonensis and L. donovani [17,18]. This transporter is encoded by two copies of aap3 gene (5.1 and 4.7). A possible explanation about the presence of these two copies can be related with the post-transcriptional gene regulation according to the environmental conditions, since the two open reading frames are similar [18].
The differentially gene expression based on the RNA-seq analysis in La-WT and La-arg- promastigotes and axenic amastigotes in the stationary growth phase revealed transcripts of amino acid transporters down- and up-regulated. The 5.1 and 4.7 aap3 appeared down-regulated in the comparisons of La-WT promastigotes vs. axenic amastigotes and La-WT vs. La-arg- promastigotes. Similar results were observed in RT-qPCR assays from intracellular La-WT or La-arg- amastigotes infecting macrophages with different genetic background (BALB/c and C57BL/6 mice, and human THP-1). Interestingly, in THP-1 macrophages, we observed an increased expression of both 5.1 and 4.7 aap3 after 4 h of infection with La-arg- when compared to La-WT. This differential behavior can be explained based on the distinct genetic background of the host that can influence in L-arginine accumulation. Then, in THP-1 and La-arg- infections, the absence of arginase activity can lead to a host transporter modulation. By its side, the parasite can respond to the L-arginine availability modulating the aap3 expression to establish the infection.
Previous studies demonstrated increased aap3 mRNA expression during the time course of macrophage infection. Muxel et al., 2017 showed that intracellular amastigotes increased the amount of 5.1 aap3 in the time course of macrophage infection with La-WT. On the other hand, the infection with La-arg- did not altered 5.1 aap3 levels. These data demonstrated the importance of L-arginine availability and arginase activity to regulate 5.1 aap3 mRNA expression and L-arginine transport to ensure the amastigote survival [7]. Additionally, when the L-arginine availability was lower in melatonin-treated infected macrophages, the levels of 5.1 aap3 and arginase mRNA of intracellular amastigotes were maintained on trial to keep the polyamine supply [42]. Goldman-Pikovich et al., 2016 showed that L. donovani intracellular amastigotes presented higher levels of LdAAP3.2 mRNA than in axenic amastigotes kept in L-arginine-starvation condition [19]
According to TriTryp database, we identified AAP3 orthologs in L. major (LMJLV39_310014600, LMJLV39_310014700, LMJSD75_310014300, LMJSD75_310014400, LmjF.31.0870 and LmjF.31.0880), L. gerbilli (LGELEM452_310014200), L. tropica (LTRL590_310015200 and LTRL590_310015300), L. turanica (LTULEM423_310014200), L. braziliensis (LbrM.31.1030; LBRM2903_000006400 and LBRM2903_310017700), L. donovani (LdBPK_310900.1 and LdBPK_310910.1) and L. infantum (LinJ.31.0900 and LinJ.31.0910). The AAP3 transcripts were annotated for L. donovani (LdBPK_310900.1.1) [43], L. infantum (LinJ.31.0910) [31], L. major (LmjF.31.0870) [44] and L. mexicana (LmxM.30.0870.1) [45] (TriTrypDB). These findings among the different Leishmania spp. show the importance of L-arginine metabolism and uptake indicating that it could contribute to the fine tuning of gene expression and consequently L-arginine uptake.
Still, according to the RNA-seq data, we also identified other amino acid transporters differentially expressed in the following comparisons: La-WT promastigotes vs. axenic amastigotes, and La-WT vs. La-arg- promastigotes. The transcript aATP11, an amino acid transporter, was previously described in association with amino acid starvation [19], confirmed the gene expression regulation in stress conditions. Notably, some aATP11 members were up-regulated (LmxM.30.0330, LmxM.30.0571 and LmxM.30.0570) and other was down-regulated (LmxM.30.0350). Other studies have shown that amastigotes activate signals when internalized into the phagolysosome, which has been linked to the down-regulation of many surface nutrient transporters. The remodeling of amastigote central carbon metabolism also represents a programmed response to stress that cells undergo in the host macrophage [46,47].
The response to starvation or L-arginine supplementation in different conditions of pH and temperature revealed that 5.1 aap3 mRNA was down-regulated at 34°C and pH 7.0 in both amino acid starvation and L-arginine supplementation, and at pH 5.0 in L-arginine supplementation. These observations can indicate a gene expression regulation at 34°C, during amastigote differentiation. On the other hand, 5.1 aap3 mRNA was up-regulated at pH 5.0 in amino acid starvation, as previously described for L. amazonensis [18] and L. donovani [19], suggesting up-regulation of the transporter in response to low L-arginine availability. The 4.7 aap3 mRNA did not show significant expression differences, indicating that probably only the 5.1 aap3 copy present the regulatory sequence for modulation.
The data obtained after L-arginine starvation showed increased amino acid uptake corroborating previous findings [18], indicating that Leishmania senses the concentration of this amino acid and regulates the expression of the transporter [17–19]. L-arginine starvation reduced the levels of arginine, ornithine and putrescine, but not spermidine, spermine and agmatine, in L. amazonensis promastigotes [48]. This condition allows the parasite to sense new signals, such as L-arginine availability in the phagolysosome environment, which is predominantly caused by the competition for L-arginine with the host cell [19]. The absence of L-arginine or polyamine could be surpassed by the polyamine transporter, described and characterized in L. major, with high affinity for putrescine and spermidine [49], L. donovani [50,51] and L. mexicana [50], and presenting an optimal transport function at pH 7.0–7.5 for promastigotes and pH 5.0 for amastigotes of L. mexicana [50]
The increase of AAP3 protein levels in the plasmatic membrane reflected the increase of L-arginine uptake at 34°C, highlighting the participation of AAP3 and the importance of the uptake of the amino acid during promastigote to amastigote differentiation, as a response to the change in temperature (25 to 34°C) [17–19]. And the increase amount of total AAP3 compared to the plasma membrane can suggest the directing of this transporter to other compartments. This hypothesis was confirmed with the cellular localization of AAP3 in promastigotes and axenic amastigotes. Previous studies already demonstrated the LdAAP3 localized in the flagella surface and in the glycosome [19]. As arginase enzyme was also localized in the glycosome [8] and as Szoor et al. (2010) reported that signaling in the nutrient-sensing pathway was targeted to this organelle in Trypanosoma brucei [52], we hypothesized that the AAP3 could also be localized in the glycosome to supply polyamines biosynthesis. In this study, we demonstrated AAP3 localized in the plasma membrane as well as in the glycosome of La-WT promastigote and axenic amastigotes in the stationary growth phase, indicating that L-arginine uptake is directed to this organelle. The results presented in this communication indicated that as a strategy for controlling Leishmania infection could be focused in the inhibition of L-arginine flux into the glycosome of the parasite.
|
10.1371/journal.pgen.1003821 | Ago1 Interacts with RNA Polymerase II and Binds to the Promoters of Actively Transcribed Genes in Human Cancer Cells | Argonaute proteins are often credited for their cytoplasmic activities in which they function as central mediators of the RNAi platform and microRNA (miRNA)-mediated processes. They also facilitate heterochromatin formation and establishment of repressive epigenetic marks in the nucleus of fission yeast and plants. However, the nuclear functions of Ago proteins in mammalian cells remain elusive. In the present study, we combine ChIP-seq (chromatin immunoprecipitation coupled with massively parallel sequencing) with biochemical assays to show that nuclear Ago1 directly interacts with RNA Polymerase II and is widely associated with chromosomal loci throughout the genome with preferential enrichment in promoters of transcriptionally active genes. Additional analyses show that nuclear Ago1 regulates the expression of Ago1-bound genes that are implicated in oncogenic pathways including cell cycle progression, growth, and survival. Our findings reveal the first landscape of human Ago1-chromosomal interactions, which may play a role in the oncogenic transcriptional program of cancer cells.
| Argonaute (Ago) proteins are an evolutionarily conserved family of proteins indispensable for a gene regulation mechanism known as RNA interference (RNAi) which is mediated by small RNA including microRNA (miRNA) and small interfering RNA (siRNA) and occurs mainly in the cytoplasm. In mammalian cells, however, the function of Agos in the nucleus is largely unknown despite a few examples in which Agos are shown to be involved in regulating gene transcription and alternative splicing. In this study, by taking a genome-wide approach, we found that human Ago1, but not Ago2, is pervasively associated with gene regulatory sequences known as promoter and interacts with the core component of the gene transcription machinery to exert positive impact on gene expression in cancer cells. Strikingly, the genes bound and regulated by Ago1 are mostly genes that stimulate cell growth and survival, and are known to be involved in the development of cancer. The findings from our study unveil an unexpected role of nuclear Ago1 in regulating gene expression which may be important both in normal cellular processes and in disease such as cancer.
| Argonautes (Ago) comprise a family of evolutionarily conserved proteins that are central to the RNA interference (RNAi) platform and miRNA function [1], [2]. Ago proteins are often recognized by their cytoplasmic function in which they regulate gene transcripts via post-transcriptional gene silencing (PTGS) mechanisms. However, nuclear functions have also been well-characterized in fission yeast and plants in which they assist in mechanisms of transcriptional gene silencing (TGS). In fission yeast, Ago partners with antisense transcripts to form the RITS (RNA-induced transcriptional silencing) complex at centromeric regions to induce heterochromatin formation [3]. Similarly, plant Argonautes interact with ribonucleoprotein complexes to induce histone and DNA methylation [4].
In mammals, the nuclear role of Ago proteins (Ago1–4) has remained largely unexplored. There have been scattered examples implicating mammalian Ago members in several nuclear processes including TGS [5]–[8], gene activation [9]–[11], and alternative splicing [12]. In the present study, we investigate the nuclear functions of Ago1 and Ago2 – the major facilitators of miRNA activity [13], [14] – from a global prospective using human cancer cells as a model system. Initial biochemical experiments indicate that nuclear Ago1 selectively interacts with RNA polymerase II (RNAP II). Chromatin immunoprecipitation coupled with massively parallel sequencing (ChIP-seq) reveals nuclear Ago1, but not Ago2, is pervasively associated with promoters of actively transcribed genes involved in growth, survival, and cell cycle progression. Ago1 knockdown experiments further indicate a positive correlation between Ago1 binding and gene expression. Additional evidence suggests that Ago1-chromosomal interactions may be dependent on miRNA. Our data represents the first landscape of Ago1-chromosomal interactions in human cells and reveals a novel function for Ago1 in modulating gene transcription within the nucleus.
We have previously shown that Ago1 and Ago2 exist in the nuclear fraction of mouse cells [11]. To determine if this feature is conserved in human cells, we examined Ago1 and Ago2 cellular distribution in the nuclear and cytosolic fractions of PC-3 (prostate adenocarcinoma) and RWPE-1 (normal prostatic epithelial) cells by immunoblot analysis. Nuclear distribution of endogenous Ago1 and Ago2 proteins was readily detectable in both cellular compartments (Figure 1A, 1B). Stable overexpression of exogenous HA-tagged Ago1 (HA-Ago1) or Ago2 (HA-Ago2) in PC-3 was also detected in both nuclear and cytosolic fractions (Figure 1C). Immunofluorescence (IF) analysis confirmed that the distribution of Ago1 and Ago2 was evident in both the cytoplasm and nucleus of PC-3 cells expressing HA or GFP-tagged Ago proteins, although signal appeared more prominent in the cytoplasm when observing whole cell distribution (Figure S1).
To determine if nuclear Ago proteins are associated with chromatin, we adopted a fractionation protocol [15] designed to selectively isolate chromatin-bound factors (Figure 1D). Immunoblot analysis revealed that Ago1 and Ago2 were detected in both chromatin fractions (P1 and S2), as well as present in the Triton X-100 soluble fraction (S1) comprising non-chromatin bound cellular proteins such as tubulin (Figure 1E); consistent with the canonical functions of Ago proteins in post-transcriptional gene silencing (PTGS) mechanisms. RNA polymerase II (RNAP II) was also detected and served as a marker for chromatin association (Figure 1E). Taken together, these results suggest that Ago1 and Ago2 are present in the nucleus of human cells in which a subfraction is bound to chromatin.
To analyze Ago protein distribution in only the nuclear compartment, we performed IF on isolated nuclei from the HA-Ago1 and HA-Ago2 stable cell lines. As shown in Figure 1F, 1G, Ago1 signals were generally scattered throughout the nuclear interior, whereas Ago2 was predominantly found on the inner nuclear periphery. Negative controls omitting the primary antibody or using cells without HA tag yielded no staining at all (Figure S2). This data indicates Ago1 and Ago2 have different nuclear localization patterns, which may reflect differences in their nuclear function.
Ago proteins have been implicated in regulating transcriptional mechanisms mediated by small RNA duplexes including gene activation and silencing [11], [16]. To determine if Ago proteins directly interact with transcriptional machinery, we performed immunoprecipitation (IP) assays on nuclear extracts from PC-3 cells using antibodies specific to endogenous Ago1 or Ago2 and immunoblotted for RNAP II. As shown in Figure 2A, RNAP II strongly co-precipitated with Ago1, but not Ago2. We further performed reciprocal RNAP II IP experiments followed by immunoblotting for Agos as well as TFIIB, a known RNAP II interacting protein, as a positive control (Figure S3). The result further confirmed RNAP II association with Ago1, but not Ago2 (Figure 2B). This interaction was also conserved in nuclear extracts from LNCaP (human prostate adenocarcinoma) cells (Figure 2C).
To address whether the Ago1-RNAP II interaction requires RNA species as intermediates, nuclear extracts were digested with a cocktail of RNase A and T1 (RNase A/T1) prior to IP (Figure 3A–C). RNase A/T1 treatment did not disrupt interactions between Ago1 and RNAP II (Figure 3A). Although it is possible that RNA molecules may have been protected from digestion by Ago1 or its associated protein complex [17], the data implies Ago1-RNAP II interactions are stable following depletion of nuclear single-stranded RNA species. To determine whether the interactions are DNA dependent, we treated the nuclear extracts with DNase and found that DNase treatment abolished Ago1-RNAP II association (Figure 3B, 3C), suggesting that DNA is required for their interaction.
To test if depletion of miRNA and/or components of the miRNA biogenesis pathway alter the Ago1-RNAP II interaction, we transfected PC-3 cells with siRNA designed to specifically knockdown Dicer (siDicer) or Drosha (siDrosha) (Figure S4A, S4B). Treatment with either siDicer or siDrosha resulted in ≥50% declines in several highly expressed miRNAs implying global downregulation of miRNA maturation (Figure S4C). It should be noted that siDicer and siDrosha treatments also upregulated endogenous protein levels of Ago1 including its nuclear abundance (Figure 3D–G), which may have resulted from a possible compensation mechanism in response to miRNA depletion [18]. Regardless, a moderate decrease in the amount of Ago1-associated RNAP II was observed following Dicer knockdown; the ratio of bound RNAP II to nuclear Ago1 decreased by ∼70% following siDicer treatment (Figure 3E),
Mutation to Dicer at exon 5 has been used to generate a stable cell line (Dicerexon5) derived from HCT116 (colorectal carcinoma) cells with impaired helicase function that interferes with miRNA maturation [19]. IP experiments revealed that co-immunoprecipitation of RNAP II with Ago1 antibody was reduced in Dicerexon5 cells compared to wild-type (WT) controls (Figure 3H), although the protein levels of neither Ago1 nor RNAP II changed in Dicer knockout line compared to its parental cells (Figure 3H). Taken together, these results indicate that Ago1 directly interacts with the core transcription machinery in human cells, which may require Dicer activity and/or the miRNA species it processes.
The physical association between Ago1 and RNAP II strongly suggests that Ago proteins may participate in transcriptional gene regulation by interacting with chromatin. Previous studies have demonstrated that Ago proteins programmed with small RNAs can bind to gene bodies or promoters by using chromatin IP (ChIP) assays [11], [12], [20]. To provide a more global view of nuclear Ago interactions, we mapped Ago1 and Ago2 binding in the genome by ChIP coupled with massively parallel sequencing (ChIP-seq). Antibody validation confirmed that ChIP antibodies for Ago1 and Ago2 had no detectable cross-reactivity ([21] and Figure S5A–C, Figure 2A) and are highly specific for RNA-protein IP and ChIP based applications ([20] and Figure S5D). ChIP-seq was also performed for H3K4me3; a histone mark associated with active gene transcription [22]. DNA quality and fragment size distribution for each library was roughly equivalent (Figure S6A). Approximately 80–100 million sequencing reads were obtained from each ChIP-seq library of which ∼80–90% could be uniquely mapped back to the human genome (Table S1, S2). To identify Ago1, Ago2, and H3K4me3-enriched regions, we applied the CCAT (control-based ChIP-seq analysis tool) peak calling algorithm [23] to the raw reads and obtained 110,533 Ago1, 144 Ago2, and 16,729 H3K4me3 peaks (Table S2). By conservatively setting the false discovery rate (FDR) cutoff to 0.054 based on independent ChIP validation results (Figure S6B–D), we obtained 44,684 Ago1 and 16,151 H3K4me3 bona fide peaks (Table S2, S3, S4). None of the Ago2 peaks passed the FDR cutoff (Table S2); therefore, we focused our subsequent analyses only on Ago1.
On average, Ago1 peaks were found once in every 70 kb of genomic sequence (Table S5) having a typical size of ∼1 kb, while the size of H3K4me3 peaks were generally broader (Figure S6E, S6F). Ago1 peaks were neither evenly distributed on chromosomes nor on genes; rather, their distribution on chromosomes correlated strongly with gene density (R2 = 0.75, P<0.0001) and GC% (R2 = 0.468, P = 0.0001), but not with % repetitive sequences (R2 = 0.034) (Figure 4A and Table S5). For example, the highest Ago1 binding density was seen on gene-dense chromosomes 19 and 17, while lowest Ago1 binding was on chromosomes Y and 13, which have the lowest gene density (Figure 4A, Figure S7, and Table S5). When multiple regression analysis was applied, gene density becomes the sole determinant of Ago1 binding density on chromosomes (P<0.001, Table S6). In addition, the majority of Ago1 peaks do not overlap chromosomal “HOT” (high occupancy transcription-related factors binding) regions [24], suggesting that Ago1 peaks we identified are not due to experimental or computational artifacts (Text S1).
Overall, Ago1-bound sequences were largely (64.9%) non-repetitive (Figure 4B). Statistical analysis indicated that Ago1 is associated with significantly less (35.1%) repetitive elements compared to overall abundance in the genome (49%, P = 4.9×10−324) (Figure 4B). Nonetheless, the major fraction of bound repetitive sequence consisted primarily of SINE, LINE, and LTR transposable elements (Figure 4B). SINE (56.8%), low complexity (4.1%) and simple repeat (3.6%) elements were overrepresented compared to their respective frequency in the genome, while LINE (18.2%) and LTR (10.2%) repeats were depleted in Ago1-bound sequences (Figure 4B). Nuclear RNAi has been implicated in transposon regulation in yeast and other eukaryotes by interacting with noncoding transcripts generated from repetitive sequence [25]. It is possible that transposable elements also mediated Ago1 interactions in the nucleus of human cells by a similar manner.
Ago1 peaks were also categorized based on gene proximity to include intragenic regions (i.e. introns, exons, and UTRs) and adjacent sequences (i.e. promoters and 3′ flanking region) within 5 kb of gene bodies. Overall, a majority of the reads corresponded to these genic locations. Compared to their respective composition in the genome, all genic regions were overrepresented in the Ago1 library including promoters, 5′UTRs, exons, introns, 3′UTRs, and 3′ flanking regions by 3.61-, 10.25-, 4.83-, 1.1-, 2.36-, and 2.27-fold, respectively (Figure 4C). In contrast, Ago1 peaks were significantly underrepresented in intergenic regions (0.41-fold, P = 5×10−324) (Figure 4C).
Given that Ago1 binding was primarily genic, we evaluated Ago1 peak distribution within ±5 kb of transcription start sites (TSS) of annotated genes. We found that the majority of Ago1 peaks mapped to a region within ±1 kb of TSSs in a distribution pattern similar to H3K4me3 peaks (Figure 4D, 4E). In fact, by further stratifying Ago1-bound genes (AbGs) for the presence or absence of H3K4me3 at TSSs, we found that within the ±1 kb region, 65.2% of Ago1 peaks overlapped with the H3K4me3 mark (Figure 4D). This data implies that Ago1 pervasively associates with chromatin at TSSs of transcriptionally active genes.
Select examples of AbGs include PIK3CA, PRKCH, CDC6, and RRM1, which have overlapping Ago1 and H3K4me3 peaks proximal to their TSSs (Figure 4F). To determine if RNAP II was also bound to AbGs, we performed ChIP analysis at the promoters of each gene. As shown in Figure 4G, we detected an enrichment of RNAP II as well as Ago1 at each TSS. Collectively, these results indicate Ago1, H3K4me3, and RNAP II are present at the promoters of the example AbGs.
To evaluate the impact of Ago1 perturbation on AbG expression, we performed microarray analysis in PC-3 cells following Ago1 knockdown with a pool of 3 Ago1-specific siRNAs (siAgo1) (Figure S8A). We identified a total of 3156 Ago1-responsive genes (ArGs) including 1592 up- and 1564 downregulated genes defined by >1.2-fold change in expression with a P value<0.05 (Table S7, S8 and Figure S8B). Twenty three genes were selected and independently assessed by qRT-PCR to confirm changes in gene expression (Figure S8C, 8D). AbGs identified by ChIP-seq analysis were subsequently correlated to the changes in global gene expression following Ago1 depletion (Figure 5A). The results indicated that 48.3% of up- and 55.4% of downregulated genes were also bound by Ago1 within 5 kb of their TSSs (Figure 5A) and the overlap between AbGs and ArGs are significantly higher than expected by chance (P = 1.4×10−6, Figure 5B, blue bars). However, when we stratified ArGs by up and downregulation, correlation was statistically significant only for downregulated ArGs (P = 2.0×10−7, Figure 5B, green bars) and not upregulated ArGs (red bars, P = 0.1, Figure 5B, red bars), suggesting that AbGs are more likely to be downregulated when Ago1 is perturbed.
Furthermore, we examined the positional effect of Ago1 binding (within ±5 kb distance) on changes in gene expression in response to Ago1 perturbation. To this end, we calculated the correlation between changes in gene expression and Ago1 binding events on the same gene for each location within −5 kb∼+5 kb region shifting one basepair each time. In consistent with the overall correlation analysis (Figure 5B), Ago1 binding events have a better correlation with down- (Figure 5C, green line) than upregulated (Figure 5C, red line) ArGs. The closer Ago1 binding was to the proximal promoter region, the greater the statistical significance was for enrichment of ArGs, especially for downregulated ArGs, with the enrichment for downregulated ArGs peaked at +111 location (P = 1.0×10−8) and upregulated ArGs at −135 location (P = 0.002) (Figure 5C). Taken together, the correlation between AbGs and downregulated ArGs suggests that Ago1 plays a positive role in maintaining transcription of a subset of genes. It is important to note that our data does not rule out the possibility Ago1 may also be functioning to suppress gene expression through promoter interactions for certain genes.
miRNAs have been shown to regulate gene transcription by binding to promoter sequences in an Ago-dependent manner [7], [11], [16], [26]. Since Ago proteins do not possess a known DNA binding domain based on protein sequence and structural analysis [27], [28], Ago1-chromosomal interactions might be mediated by miRNAs. As such, we performed miRNA target prediction analysis on the Ago1-bound DNA sequences identified by ChIP-seq. Compared to random selected matched control sequences, the frequency of putative target sites in Ago1-bound peaks were roughly equivalent for most miRNAs (Figure 6A, 6B). However, a total of 49 miRNAs were found to have a statistically higher number of target sites in the Ago1-bound peaks compared to the control sequences (>1.5 fold enrichment, P = 0∼6×10−41), while only 3 miRNAs, function of which is unknown, have higher number of targets in the control sequences (Figure 6A, 6B and Table S9). Interestingly, approximately one third of the 49 miRNAs are known oncomiRs including those from the miR-17-92 and miR-106b-25 clusters, as well as the miR-520/373 family (Figure 6B, Table S9).
We also preformed motif analysis on each miRNA with enriched target sites in Ago1-bound sequences. A common motif “AGUGCU/A” was found in 19 out the 49 miRNAs; 7 of which contained two incidences of this motif (Figure 6C, 6D). Interestingly, a similar motif (AGUGUU) was identified in the 3′terminus of miR-29b, which functions as a nucleic acid-based nuclear localization signal (NLS) [29]. Although the significance of our motif in context to Ago1-bound sequences is unknown, it shares ∼83% homology with the miR-29b NLS (Figure 6C). As certain miRNAs are known to preferably accumulate in the nucleus [30], the identification of putative target sites at Ago1-bound peaks supports the idea that such miRNAs may play a role in directing Ago1-chromomal interactions.
To test the regulatory effect of Ago1 binding on gene promoters, we depleted Ago1 in PC-3 cells using siAgo1 and evaluated its effect on 4 ArGs (i.e. SMC1A, CDC20, SMAD3 and BUB1) with overlapping Ago1 and H3K4me3 peaks at their TSSs (Figure 7A). ChIP analysis revealed reductions in bound Ago1 at the promoters for each gene (Figure 7B). Moreover, knockdown of Ago1 reduced RNAP II occupancy at TSSs (Figure 7C) with corresponding decreases in gene expression levels (Figure 7D). We also generated stable cell lines in RWPE-1 (non-malignant prostate epithelium) cells overexpressing Ago1 or a deletion mutant lacking the PAZ domain (Ago1 dPAZ) (Figure 8A), which is known to interfere with efficient miRNA loading into Ago proteins [31]. Ago1 overexpression resulted in a moderate induction of each gene (Figure 8B), while PAZ deletion attenuated this response (Figure 8B), further supporting a role for miRNAs in directing Ago1-chromosomal interactions. Furthermore, we performed Ago1 ChIP for the 4 example genes and were able to detect in Ago1 overexpressing RWPE-1 cells a concurrent increase in Ago1 binding at the same sites near TSSs detected in PC-3 cells (Figure 8C). Collectively, these results suggest Ago1 contributes to positive gene regulation of select ArGs by interacting with gene promoters and stimulating RNAP II enrichment.
Three overlapping AbG sets were defined to include AbGs-5 kb, -1 kb and -0.5 kb, which consist of genes with at least one Ago1 peak within ±5, ±1, and ±0.5 kb away from TSSs, respectively. AbGs-5 kb, -1 kb and -0.5 kb respectively contain 15503, 10074, and 8057 unique genes encompassing 27.5%, 17.9%, and 14.3% of all annotated genes in Ensembl human genome database (Table S10, S11, S12). Interestingly, clustering AbGs-5 kb, -1 kb or -0.5 kb genes by their chromosomal location reveal several cytobands implicated in different human cancers that are highly overrepresented (Figure S9, Table S13, S14). For example, the top-enriched cytobands 19p13.3 and 16p13.3 have been established by numerous studies to be susceptibility loci for several types of cancers including prostate, breast, thyroid, and lymphoma [32]–[35].
Gene pathway enrichment analysis further revealed a number of oncogenic pathways overrepresented by AbGs. The top 5 KEGG pathways highly enriched in AbGs-5 kb genes include “pathways in cancer” (P = 1.9×10−10), “MAPK signaling” (P = 1.7×10−8), “Wnt signing” (P = 1.1×10−7), “endocytosis” (P = 4.3×10−7), and “focal adhesion” (P = 6×10−7) (Figure 9A). Many proto-oncogenes and proliferation-promoting genes are exemplified in these pathways including growth factors, tyrosine/serine/threonine kinases, G-protein coupled receptors, membrane-associated G-proteins, and nuclear DNA-binding/transcription factors (Table S15). These enrichments hold when AbGs are narrowed down to AbG-1 kb and AbG-0.5 kb genes (Figure S10A, S10B). For instance, SMC1A, CDC20, SMAD3 and BUB1 are all example AbG-0.5 kb genes known to promote cell cycle progression and proliferation in various cancer cell types [36]–[39]. Gene Ontology (GO) classification of AbGs-5 kb genes also show enrichment for gene categories that regulate metabolic processes, transcription, cell cycle, chromatin modification, and cell death (Figure S10C).
KEGG pathway enrichment analysis also revealed that ArGs shared several cancer-related pathways with AbGs-5 kb genes including “MAPK signaling”, “p53 signaling”, “cell cycle”, “prostate cancer”, “colorectal cancer”, etc. (Figure 9B). Further GO analysis revealed that up- and downregulated ArGs were enriched in distinct biological processes with the latter significantly overrepresented by processes important for cancer growth/development including cell cycle, mitosis, DNA repair, chromosome organization, etc. (Figure S11A, S11B).
Analysis of Ago1 protein levels in non-tumorigenic (RWPE-1 and PWR-1E) and cancerous (PC-3, DU145, LNCaP, RV1, CWR22R, and C4-2) prostate cell lines indicated Ago1 is generally expressed at significantly higher levels in cancer cell lines (Figure S12A). Furthermore, knockdown of Ago1 in PC-3 cells caused G0/G1 arrest as indicated by the increase in G0/G1 cell number and corresponding reductions in S and G2/M populations (Figure S12B, S12C). Our data suggests Ago1 may be involved in oncogenic processes, in part, through its nuclear activity by affecting the expression of genes involved in cell growth/survival. In support, integrated analysis of ChIP-seq and gene expression profiling places Ago1 in various major cancer-related signaling pathways involved in regulating DNA damage response, mitogenic signaling, cell cycle, angiogenesis, and apoptosis (Figure 9C).
It has become clear that Ago proteins participate in gene regulation at multiple levels. In the present study, we reveal another layer to Ago1 in regulating gene expression within the nucleus of human cancer cells. We provide biochemical evidence that nuclear Ago1, but not Ago2, directly associates with RNAP II. ChIP-seq analysis indicates Ago1 is pervasively bound to multiple genomic loci including repetitive elements of transposons and euchromatic sites as defined by the histone mark H3K4me3. Interestingly, this observation is consistent with the chromosomal binding profiles of drosophila Ago2 (dAGO2); the primary Argonaute for mediating RNAi and miRNA function in the fly [15], [40]. Additionally, Ago1 binding at gene promoters functionally impacts active gene transcription as its loss of function results in reduced Ago1 and RNAP II occupancy at TSSs with corresponding reductions in gene expression, whereas gain of function causes the opposite changes. Our data represents the first landscape of Ago1-chromosomal interactions in human cancer cells, while revealing a novel non-canonical function for Ago1 in regulating gene expression.
It is currently unclear how Ago1 is targeted to selected chromosomal loci. Our analyses imply that miRNA may be involved in mediating interactions between nuclear Ago1, chromatin, and/or RNAP II. Ago1-bound sequences contained putative miRNA target sites and its binding activity to RNAP II was suppressed by perturbing Dicer function; an essential protein involved in miRNA maturation. Additionally, deletion of the RNA-binding domain (PAZ) in Ago1 interfered with gene activation further implicating a role for RNA (i.e. miRNA) in this process. In support, it has been reported that transfection of exogenous miRNA can promote enrichment of Ago proteins at highly-complementary sites in gene promoters to manipulate transcription [7], [11], [16]. Depletion of nuclear single-stranded RNAs by RNase A/T1 did not interfere with Ago1-RNAP II association; however, Ago1 may be loaded with miRNA forming a duplex with complementary target sequence and protecting bound RNA from RNase A/T1 digestion in manner similar to canonical target recognition [17], [41]. As we have not definitely confirmed the presence of miRNAs in these nuclear Ago1 complexes, it is also possible other classes of small RNA species mediate Ago1 interactions with chromatin. For instance, recent deep sequencing studies have shown that Ago1 can associate with small RNA species from non-miRNA sources [42], [43].
In contrast to Ago1, Ago2 apparently lacked pervasive association with chromatin. Additionally, it did not immunoprecipitate with basal transcription machinery (i.e. RNAP II). Although we cannot absolutely rule out technical reasons for the lack of Ago2 binding, the difference in binding may be reflective of their differential nuclear distribution as revealed by IF microscopy (Figure 1E). Ago1 and Ago2 have been reported to exhibit intrinsic preferences when selecting and/or loading RNA molecules. For instance, studies have shown that Ago2 binds perfect-complementary RNA duplexes (e.g. siRNAs) with higher affinity than Ago1; whereas, Ago1 preferably associates with duplexes containing bulges and mismatched bases (e.g. miRNA) [44], [45]. This intrinsic segregation in RNA binding may also be a key determinant in mediating Ago interactions in the nucleus. Alternatively, nuclear Ago2 may be sequestered to the nuclear envelope and only associate with chromatin in a signal-dependent manner. In support, cellular senescence has been shown to trigger nuclear accumulation of Ago2 and binding at gene promoters [8].
It is noteworthy that the magnitudes of gene expression changes for a vast majority of genes in response to Ago1 perturbation were less than two-fold. This observation is consistent with post-transcriptional gene regulation by miRNA [46] and suggests that the role of Ago1 is fine-tuning gene expression in a miRNA dependent manner both at the transcriptional and post-transcriptional levels. The short term (48 hrs) transfection of Ago1 siRNA may also be accounted for the subtle changes in gene expression. We chose this duration to minimize detecting potential secondary regulation but at the same time we might have missed the maximum responsiveness of gene expression to Ago1 perturbation.
In the cytoplasm, Ago proteins elicit pleiotropic effects on gene expression by utilizing miRNA to silence multiple transcripts and regulate various cellular processes [1], [2]. Similarly, nuclear Ago1 also possesses pleiotropy by affecting transcription of multiple genes. In PC-3 cells, Ago1 appeared to preferably drive the expression of genes involved in oncogenic pathways suggesting it may play a role in the cancer phenotype. In support, knockdown of Ago1 by siRNA inhibited cell cycle progression. However, its effects on cancer may be context dependent and vary between different cell types based on both its cytosolic and nuclear activities, as well as the gene profile it regulates. It would be of future interests to understand the crosstalk between Ago-mediated gene regulatory networks and oncogenic signaling pathways.
PC-3, LNCaP, DU145, LAPC4, RV1, CWR22R, C4-2, and HCT116 cell lines (ATCC) were maintained in RPMI-1640 media (UCSF Cell Culture Core) supplemented with 10% fetal bovine serum (Hyclone), penicillin G (100 U/mL), streptomycin (100 µg/mL) in a humidified atmosphere of 5% CO2 at 37°C. RWPE-1 and PWR-1E cells were cultured in serum-free keratinocyte medium supplemented with 5 ng/ml human recombinant epidermal growth factor and 0.05 mg/ml bovine pituitary extract.
Vectors pIRESneo-FLAG/HA-Ago1 (Addgene #10820) and pIRESneo-FLAG/HA-Ago2 (Addgene #10822) were used to establish stable cell lines overexpressing HA-tagged Ago1 (PC3-HA-Ago1) and Ago2 (PC3-HA-Ago2), respectively. Briefly, PC-3 cells were transfected with each corresponding vector and single colonies were subcultured following selection with G418. GFP-Ago1 (Addgene #21534) and GFP-Ago2 (Addgene #11590) plasmids were transiently transfected into PC-3 cells and imaged by fluorescence microscopy. Full-length human Ago1 and the PAZ deletion mutant (Ago1 dPAZ) were amplified from pIRESneo-FLAG/HA-Ago1 and pIRESneo-FLAG/HA-Ago1dPAZ, respectively. Each amplicon was cloned into the lentiviral cDNA expression vector pCDH-EF1-MCS-T2A-copGFP (System Biosciences) via EcoRI and BamHI restriction sites. For lentivirus mediated overexpression, lentivirus particles were generated by the ViraPower Lentiviral Expression System (Invitrogen) and used to infect RWPE-1 cells to generate stable cell lines. Expression of all constructs was confirmed by immunoblot analysis.
PC-3 cells were seeded on coverslips at 50% confluency. The following day, cells were washed with PBS and fixed in 4% paraformaldehyde at room temperature for 15 min. Cells were permeabilized in PBS containing 0.3% Triton-X-100 for 10 min, rinsed with PBS, and blocked with 10% goat serum at room temperature for 1 hr. Coverslips were incubated with primary antibodies anti-HA (Cell Signaling,cat # 2367, 1∶200) or anti-Ago2 (Wako, cat # 011-22033, 1∶200) diluted in 10% goat serum at room temperature for 1 hr. Cells were washed with PBS and subsequently treated with anti-mouse FITC antibody (Vector Lab; 1∶200) at room temperature for 1 hr. Coverslips were washed and mounted with mounting media containing DAPI. IF images were captured using a Zeiss AxioImager M1 fluorescence microscope. Purified nuclei for IF analysis were isolated as previously described [47]. Nuclei were fixed on slides with fixative reagent (methanol: acetic acid, v/v 3∶1) at room temperature for 5 min and washed with 4×SSC containing 0.1% Tween 20. The slides were subsequently incubated with anti-Ago1 (Santa Cruz Biotechnology, cat #sc-32657, 1∶200), anti-Ago2 (Wako, cat #011-22033, 1∶200), or anti-HA (Cell Signaling, cat #2367, 1∶200) diluted in dilution buffer (1% bovine serum albumin, 4×SSC, and 0.1% Tween 20) at 4°C overnight. Nuclei were washed and incubated with the appropriate Alexa Fluor® 488 secondary antibodies (Molecular Probes; 1∶200) for 30 minutes at 37°C. Following a series of washes, slides were mounted with DAPI II (Abbot Molecular) and IF signals were analyzed using the CytoVision imaging system (Applied Imaging).
Chromatin fractionation was performed as previously described [15]. Cell pellets were collected from two 150 mm plates and washed with PBS. Approximately 1/10th of the cell pellet was resuspended in RIPA buffer (50 mM Tris, pH 7.4, 150 mM NaCl, 1% Triton ×-100, 0.5% deoxycholate, 0.1% SDS, protease inhibitor cocktail, and phosphatase inhibitor) and incubated on ice for 30 min to generate whole cell lysate. The remaining pellet was lysed in cold CSKI buffer [10 mM PIPES, pH 6.8, 100 mM NaCl, 1 mM EDTA, 300 mM sucrose, 1 mM MgCl2, 1 mM DTT, 0.5% (v/v) Triton X-100, and protease inhibitor cocktail (Roche)]. The lysate was divided into two equal portions and centrifuged at 500×g for 3 min at 4°C. The resulting supernatant was collect and referred to as the S1 fraction. One pellet was washed twice in CSKI buffer and resuspended in RIPA buffer to generate the P1 fraction. The other pellet was resuspended in CSKII buffer [10 mM PIPES, pH 6.8, 50 mM NaCl, 300 mM sucrose, 6 mM MgCl2, 1 mM DTT, and protease inhibitor cocktail (Roche)] and treated with DNase (Qiagen) for 30 min. The resulting sample was extracted with 250 mM NH2SO4 for 10 min at room temperature and centrifuged at 1200×g for 6 min at 4°C to generate the S2 (supernatant) and P2 fractions (pellet). The P2 fraction was subsequently resuspended in RIPA buffer. Cytoplasmic and nuclear fractions were prepared by using the NE-PER Nuclear and Cytoplasmic Extraction Reagents (Thermo Scientific). Whole cell lysate was obtained by lysing cells in RIPA buffer for 15 minutes at 4°C. Lysates were clarified by centrifugation for 15 minutes at 14,000 rpm and supernatants were collected. 30 µg of protein from all fractions was analyzed by immunoblot analysis.
Immunoprecipitation was performed according to Cernilogar et al. [15]. Approximately 400–800 µg of protein from nuclear extracts was mixed with equal volumes immunoprecipitation buffer [10 mM Tris-HCl, pH 8.0, 150 mM NaCl, 1 mM EDTA, 1 mM DTT, 0.1% NP-40, and protease inhibitor cocktail (Roche)]. In Figure 3A–3C, nuclear extract was treated with 2.5 ul of RNase A/T (Ambion) cocktail for 30 min at 25°C or 100 ng/uL of DNAse I (Roche) for 20 min at 37°C. Each sample was subsequently treated with 5 µg of antibody and incubated overnight at 4°C. Antibody treatments included anti-Ago1 (Wako, clone 2A7, cat# 015-22411), anti-Ago2 (Wako, clone 4G8, cat# 011-22033), anti-RNAP II (Millipore, cat# 05-623), or mouse IgG (Millipore, cat# 12-371). The following day, 40 µl protein G dynabeads were added to each sample and rotated for 2 hrs at 4°C. The beads were subsequently washed five times with 500 µl immunoprecipitation buffer and resuspended in SDS-PAGE sample buffer. Immunoprecipitates were boiled for 5 min and the resulting supernatants were analyzed by immunoblot analysis.
Sample protein concentration was determined by BCA protein assay (Thermo Scientific). Equal amounts of protein were resolved by SDS-PAGE and transferred to 0.45 µm nitrocellulose membranes by voltage gradient. The resulting blots were blocked overnight in 5% nonfat dry milk and subsequently probed with primary antibody. The antibodies were used at the indicated dilutions: anti-Ago1 (Cell Signaling. cat #5053) at 1∶1000, anti-Ago2 (Wako, cat# 011-22033) at 1∶1000, anti-HA 6E2 (Cell Signaling, cat # 2367) at 1∶1000, anti-Tubulin (Sigma, cat # T6074) at 1∶1000, anti-Topoisomerase I (Santa Cruz, cat #sc-10783) at 1∶500, anti-RNAP II (Millipore, cat # 05-623) at 1∶5000, anti-Dicer (Santa Cruz, cat #sc-30226) at 1∶1000, anti-Drosha (Cell Signaling, cat #3364) at 1∶1000, and anti-TFIIB (Cell Signaling, cat #4169) at 1∶1000. Immunodetection occurred by incubating blots with appropriate secondary HRP-linked antibodies and utilizing the SuperSignal West Pico Chemiluminescent kit (Thermo Scientific) to visualize antigen-antibody complexes.
All siRNAs were designed using the BLOCK-iT RNAi Designer Program (Invitrogen). Ago1 knockdown was accomplished by using a pool of 3 siRNAs, while single duplexes were used to knockdown Dicer or Drosha. A pool of 3 non-specific siRNAs served as controls. Transfections were carried out using Lipofectamine RNAiMax (Invitrogen) according to the manufacturer's instructions. All siRNA sequences are listed in Table S16.
Total RNA was isolated by using the RNeasy Mini Kit (Qiagen). ∼1 µg of total RNA was reverse transcribed into cDNA with MMLV reverse transcriptase (Promega) in conjunction with oligo(dT) primers. The resulting cDNA samples were subjected to real-time PCR analysis using gene-specific primers. All primer sequences are listed in Table S16.
Chromatin immunoprecipitation (ChIP) was performed as previously described with slight modification [11]. Chromatin was prepared from PC-3 cells following crosslinking with formaldehyde. DNA was sheared to an average size of ∼500 bp using the Bioruptor sonicator (Diagenode) set to ‘high’ with 30 sec ON/OFF pulses for 8 min for a total of 8 cycles. Chromatin was immunoprecipitated overnight at 4°C using 5 µg of the following antibodies: anti-Ago1 (Wako, clone 2A7), anti-Ago2 (Wako, clone 2D4), anti-H3K4me3 (Millipore, cat# 07-473), and mouse IgG (Millipore, cat# 12-371). The following day, the samples were incubated with 25 µl Protein G Dynabeads (Invitrogen) for 2 hrs at 4°C. Immunoprecipitates were sequentially washed with low salt, high salt, and TE buffer. Eluates were collected and reverse crosslinked at 65°C overnight. ChIP DNA was treated with Proteinase K, purified with phenol/chloroform, treated with RNase A, and purified using the Qiaquick PCR purification kit (Qiagen). Target amplification and detection was performed by the 7500 Fast Real-Time System (Applied Biosystems). All reactions were prepared in 10 µl volumes containing 2 µl DNA, 2× Fast SYBR Green master mix (Applied Biosystems), and region-specific primer sets (Table S16). Each sample was analyzed in triplicate. Enrichment was determined by using the 2−ΔCT method relative to input DNA or IgG control. Primer specificity was confirmed by evaluating dissociation curves and independently analyzing amplified product on an agarose gel. For Ago ChIP-western analysis, IP was performed essentially the same way as above and the beads were resuspended in 2× SDS sample buffer and boiled for 5 min. Supernatant was collected and analyzed by western blotting analysis.
Each library was prepared by combining the eluates from two ChIP experiments and following the Illumina ChIP-seq library preparation protocol. Briefly, ∼10 ng DNA was end-repaired and subsequently labeled with an additional “A” base on the 3′ ends of the DNA fragments. The resulting DNA samples were ligated to oligonucleotide adaptors and amplified by PCR to construct the individual libraries. Each library was size-selected for DNA fragments ranging between ∼200–300 bp by gel electrophoresis purification. Sample quality was assessed on a Bioanalyzer (Agilent) using the Hypersensitive DNA kit (Agilent) prior to sequencing. Libraries were diluted to 10 nM and sent to the Vincent J. Coates Genomics Sequencing Laboratory at UC Berkeley (http://qb3.berkeley.edu/gsl) for sequencing analysis on a Hiseq2000 Sequencing System (Illumina). Additional detail on ChIP-seq is available in Text S2.
The ChIP-seq and microarray data from this study have been deposited into the GEO database under the accession numbers GSE40536 and GSE42600.
Other experimental procedures are available in Text S2.
|
10.1371/journal.pcbi.1003275 | Keeping up with the Joneses: Interpersonal Prediction Errors and the Correlation of Behavior in a Tandem Sequential Choice Task | In many settings, copying, learning from or assigning value to group behavior is rational because such behavior can often act as a proxy for valuable returns. However, such herd behavior can also be pathologically misleading by coaxing individuals into behaviors that are otherwise irrational and it may be one source of the irrational behaviors underlying market bubbles and crashes. Using a two-person tandem investment game, we sought to examine the neural and behavioral responses of herd instincts in situations stripped of the incentive to be influenced by the choices of one's partner. We show that the investments of the two subjects correlate over time if they are made aware of their partner's choices even though these choices have no impact on either player's earnings. We computed an “interpersonal prediction error”, the difference between the investment decisions of the two subjects after each choice. BOLD responses in the striatum, implicated in valuation and action selection, were highly correlated with this interpersonal prediction error. The revelation of the partner's investment occurred after all useful information about the market had already been revealed. This effect was confirmed in two separate experiments where the impact of the time of revelation of the partner's choice was tested at 2 seconds and 6 seconds after a subject's choice; however, the effect was absent in a control condition with a computer partner. These findings strongly support the existence of mechanisms that drive correlated behavior even in contexts where there is no explicit advantage to do so.
| In this study we examine the neural substrates of inter-personal error signals on behavior in an investment task using real historical markets. We show that behaviorally, subjects correlate their investments, despite the fact that another trader has no extra information about how the market may move. These behavioral results are supported by neural data showing large, parametric responses in brain areas related to reward and learning when information about another trader's behavior is revealed, even though this occurs after all useful information about the market has already been shown. These results promise to elucidate some of the subconscious processes that guide people to correlate their behavior in markets and other group environments.
| Humans learn a range of information from one another [1] and show a particular sensitivity to the influence of group behavior [2]. The ultimate evolutionary origins of these behaviors and their dependence on other relevant variables raise broad-ranging questions [3]–[6] however, they also invite important but narrower questions about the human propensity to assign value to the behavior of others even when there exists no external incentive to do so. Such assignments can reasonably be considered irrational because they explicitly violate external incentive structures. It has been suggested that this propensity to ‘follow-the-crowd’ – even in the face of information that suggests otherwise – is the basis of a range of herding behaviors displayed by humans interacting through markets including both bubbles and crashes [7]–[11]. One hypothesis for the origin of this class of ‘believe-the-group’ irrationalities is that while long ago group behavior tended to be a good proxy for value, the complexities of modern life, and especially modern markets, subvert this tendency, producing unpredictable behaviors in market settings.
We used a tandem (two-person) sequential choice experiment, framed as a market investment task, to test the degree to which neural and/or behavioral responses change depending solely on the behavior of one's partner, and whether they do so in the absence of incentives. The task asks a subject to invest some fraction (from 0 to 1) of their total holdings, shows the change in the market value which controls gains and losses, and later shows the fraction invested by their partner (Figure 1). The partner's investment has no bearing on the payoff of the subject or on the market's future movements. In addition to this tandem task we included a control condition in which subjects played in tandem with a computer that chose its investments randomly (uniformly over [0,1]). In this control condition, subjects were informed that the other “investor” was a computer and that its choices were random. This experiment asks two empirical questions. (1) Does a subject change their behavior based on the difference with their partner's choice (Jones)? (2) How does the brain respond to the difference between the subject's investment level and their partners? We repeated the experiment twice and varied the time at which the partner's choice was revealed (2 seconds and 6 seconds after the subject's choice).
In this task, there is no incentive for the answer to either question to be yes; however, a positive answer to either suggests that group behavior is deemed valuable by brain and behavior even in the absence of external economic incentives. The striatum is well known for encoding “prediction error” signals that aid humans and other animals learn the value of various stimuli and actions; therefore, we hypothesized that the “interpersonal prediction error”, i.e. the difference between the partner's and the subject's own bet (henceforth referred to as Jones), would (a) correlate with activation in the striatum and (b) correlate with the bet in the next round of play. Hypothesis (a) is based on the idea that the subject's brain assumes that this difference with the partner's bet is an informative error signal. Hypothesis (b) – the idea that this difference would correlate with a tendency to adjust ones behavior toward that of the other investor – suggests one bias that would encourage irrational herding behavior.
The setup for our tandem investment task and our framing of the behavior in terms that inform our notion of irrational herding behavior is also supported by economic ideas. Economists have laid out the theory of information cascades – situations where rational agents disregard their private signals and follow the choices of others [9], [12], [13] ‘as though’ the others have different or better information. This tendency to herd is also thought to play a role in more complicated situations, such as financial markets, where the phenomenon may lead to bubbles and crashes [14].
Recently neuroscientists have begun to explore the neural underpinnings of social learning [15]–[23]. We extend these results to consider the effects of others' past investment behavior on subsequent investment behavior when the risk parameters of the underlying market are fundamentally unknown. We hypothesized that modulation of the error signals in the ventral striatum would reflect the influence of social information on investment behavior.
In order to test the hypothesis that people's investment behavior is affected by social information, and to probe the neural substrates of this influence, we employed fMRI and two human versions and one control condition of a “tandem” implementation of an investment game previously used to probe intrapersonal fictive errors (the difference between the actual received reward, and the best possible outcome a subject might have achieved) [24], [25]. Figure 1 gives a schematic outline of the tasks. In the human conditions two subjects (who knew that there was another person playing but did not meet) played the investment game simultaneously while being scanned. In the investment game, both subjects were endowed with $20, and then each had to decide what percentage of their endowment to risk in the “market” (the markets were taken from actual historical markets. See Text S1 for details). After each person lodged their asset allocation (their “bet”), the next market outcome was revealed, the portfolio value and percentage gained or lost was updated, and after a short delay (2 sec in the first version, and 6 sec in the second) a pair of red arrows representing the other player's investment level percentage appeared on the slider bar. After another short delay the process was repeated. The choices of the players had no direct influence on future market fluctuations, and the choices of one player had no direct influence on the payoffs of the other player. In the computer control condition the subject was told they were playing a computer partner that chose randomly; the delay between the market revelation and the revelation of the computer choice was 6 sec. We examined the behavior from all three experiments, but focused on the imaging from the 6 sec. human and computer control conditions.
To examine differences among the three versions of the experiment we performed a mixed-effects linear regression separating the three groups (2 sec human, N = 68; 6 sec human, N = 24; 6 sec computer control, N = 24; see Tables S1, S2, S3 for demographic information) using indicator functions for the three groups (interacting with all of the variables of interest). The dependent variable was the normalized investment. The independent variables in the regression were a constant, the normalized previous bet, the previous market return (MKT), and a variable we call DJONES, equal to the difference between the other subject's investment and the subject's investment. Here we focus on the regression coefficient of DJONES (Figure 2). The coefficients from the 2 sec and 6 sec human experiments are both significantly greater than zero, and the coefficient in the computer control condition is not significantly different than zero. There is also a significant difference between the human 6 sec condition, and the computer control condition. See Text S1 for more regression details, and Table S4 and Table S5 for complete regression tables.
To investigate the neural underpinnings of these signals we constructed a regression model for the imaging data using regressors suggested by behavioral model (see Supporting Information for details). We limited our investigation of the neural data to the 6 sec human and computer control experiments. Specifically we included a parametric regressor for DJONES at the reveal of the other person's investment, and a parametric regressor for MKT at the time of the revelation of the market return to the subject. Figure 3A shows the activation corresponding to the DJONES regressor in the human condition while 3B shows the activation in the computer control (both N = 24; both displayed with p<.001 uncorrected, cluster size > = 5). Note that there were no regions of significant negative correlation. See Figure S1 and Figure S2 for regression tables and glass brains. In the human condition, this activation survived a small volume correction for multiple comparisons over an ROI consisting of 5 mm radius balls centered on bilateral caudate/putamen voxels taken from peak activations in [24]. (See Figure S3 for mask). Additionally, the comparison (two-sample t-test) of DJONES across the human and computer conditions survived a similar small volume correction yielding voxels in left caudate (Figure S4). Activation tables for both small volume corrections are in Figure S5.
While not our main focus, it is worth noting that the MKT regressor also produced, in both human and control conditions, robust activation in the striatum (Figure S6). Figure S7 shows a conjunction/disjunction analysis of the MKT and DJONES activation at the p<.001 and p<.05 levels in the human condition.
We were also interested in the possible differences between the neural and behavioral effects of the variables obtained by splitting DJONES into its positive and negative parts (e.g. POSDJONES = max(DJONES , 0); see Text S1 for details). We find a significant difference in the behavioral regression coefficients, with the coefficient of the negative part of DJONES being larger in absolute value (Table S3). Neurally, however, we find no difference between the two conditions (Figure S8).
Finally, we wanted to investigate the relationship between the neural correlates of DJONES and the individual behavioral regression coefficients of DJONES. Figure 4A shows the middle cingulate region for which the individual neural DJONES responses are significantly positively linearly related to the individual behavioral DJONES coefficients (p<.05, FWE whole-brain corrected; behavioral coefficients from individual subject regressions. See Text S1 for details.). Figure 4B shows (for illustrative purposes only) a plot of the neural coefficients against the (mean adjusted) behavioral coefficients.
Using a tandem sequential investment task we show that when subjects play a human partner the inter-personal fictive error guides behavior (subjects' next bet) and correlates with a robust neural signature in the striatum. These findings are significant because the partner's choice is revealed after the subject's monetary outcome is revealed and the partner's choice has no bearing on the payoff to the subject. Despite these facts, the inter-personal fictive error still influences the subject's behavior on their next bet, correlates with a robust and parametric neural signature in an important reward processing structure, and depends on whether the partner is a human. Specifically, if humans play a computer partner expressing random investments on each trial this same inter-agent fictive error term has no behavioral impact on the next bet and has no significant neural correlate in the striatum.
Our results are for the most part are consonant with the results of previous studies of social influence [15]–[23] that show neural responses to and behavioral influences of the choices of others. However, there are several key differences that allow us to expand on these results.
First, the timing of private and social outcome revelations was significantly different in this design. Here, information about the market is revealed first, giving the subject all the information relevant to their payoff, and then the social signal from the partner is revealed. Second, our design is parametric in the choices and outcomes. Our design thus allows us to show that the striatal response and immediate subsequent behavior is fully parametrically influenced by both the market return signal and the interpersonal error signal. Additionally, we see a behavioral asymmetry in the effect of the partner's investment between the outcome where the partner invested more than the subject versus the case where the partner invested less. Subjects adjusted their subsequent investment more when their partner invested less than they did on the previous trial as though they were fleeing their own over-exuberance on that trial. Finally, Burke et al. [17] show that ventral striatum activation to social information covaries with behavioral sensitivity to herd information. We do not see this in our experiment. Rather, we see that neural activation to DJONES in middle cingulate cortex covaries positively with behavioral sensitivity to DJONES. One possible explanation for this correlation is suggested by two studies. Kishida et al. [26] found that athletes showed increased middle cingulate activity when imagining themselves playing their own sport as opposed to a different sport. Further, they saw the same result in subjects when they took a first, as opposed to third person perspective when imagining a sports scene. On the other hand, Chiu et al. [27] found decreased activity during the “self” phase of the trust game in the middle cingulate in autistic subjects. The effect covaried with symptom severity. These results suggest that this area is key for identifying with conspecifics, pointing to a hypothesis that neural sensitivity in middle cingulate to the DJONES signal is dependent on the tendency of a subject to identify with the other investor. This hypothesis is also supported behaviorally by the findings of Burke et al. [17] showing that herding behavior is more pronounced when investing alongside human conspecifics as opposed to non-human primates, as well as by the absence of a DJONES effect in our control condition.
Our results suggest that the difference between the partner's investment and the subject's investment can be viewed as an error signal that guides behavior, rather than as simply an add-on affective response. The affective system has long been considered a necessary component effective decision-making [28] whose function can be seen as “ecologically rational” [29]. Neural signals correlated with affect may then be reinterpreted as error signals [30]. For example, much of the early work on anterior insula focused on emotions such as pain and disgust. [31], [32] Recently, however, the function of the anterior insula has been recast in the language of error signals [29], whereby activation in the insula is regarded as signaling a variance prediction error. Here our focus is on the striatum, but the idea is similar. Indeed multiple works [17], [23], [33]–[35] suggest that socially construed reward signals should appear in the striatum just as other control signals do. In this light, the results of this paper strongly suggest that we view the activation in the striatum not only as a hedonic signal, but also as a control signal.
Correlation is a property that is vitally important in asset management: in order to maximize return with a minimum of risk an investment manager must know the correlation of the assets under management [36]. Our ancestors living in small groups were not “asset managers”, but it is likely the members of the group correlated their activities in an optimal way, an activity that would require the brain to track and control individual correlations.
Finally these results provide biological evidence that standard theories of investment behavior that are variations on the Markowitz model [37] miss a fundamental driver of behavior by failing to account for the behavior of other investors. The response of the striatum to the Jones variable suggests that tendency to correlate actions is deeply rooted with potential evolutionary drivers. This lends weight to the “behavioral finance” approach espoused by Shiller and others [10], [38], [39].
In summary, previous work shows that the comparison of personal results to the results of another modulates neural activity. Our results further show that the comparison of the personal result to the outcome of the other person can be put in the context of an error signal, the interpersonal fictive error, which controls behavior and has a robust neural signature. Social comparison can thus be construed not merely as a possibly unseemly manifestation of envy, but rather as a potentially useful learning signal.
Informed consent was obtained for all research involving human participants, and all clinical investigation was conducted according to the principles expressed in the Declaration of Helsinki. All procedures were approved by the Institutional Review Board of the Baylor College of Medicine, or the Institutional Review Board of Virginia Tech.
Experiment 1: 76 participants were recruited and 74 scanned in accordance with a protocol approved by the Baylor College of medicine IRB. In the two behavioral only subjects the log files of the experiment were incomplete, leaving unusable data; in two scanned subjects the experiment terminated prematurely; in 4 other scanned subjects the functional images were unusable, leaving 68 subjects with both behavioral and imaging data. Table S1 summarizes the demographic information of these 68 subjects. All data mentioned in the text and supplementary information referring to the first experiment refers to the behavioral data only of these 68 subjects. Experiments 2 and 3: 49 participants (24 for the human condition and 25 for the computer control condition) were recruited and 49 scanned in accordance with a protocol approved by the Virginia Tech IRB. One subject's scanning session terminated prematurely in the control cohort leaving 24 subjects. All data mentioned in the text and SOM referring to the second experiment refers to these subjects. Table S2 summarizes the demographic information of these subjects.
Participants arrived at the lab, were consented, and then read task instructions. In the versions with human partners the partners did not meet. After they were loaded in scanner, the task began. Each subject participated in 10 markets in a random order. There were two groups of markets, A and B (originally described and used in Lohrenz et al., 2007). 30 subjects saw group A, and 41 subjects saw group B. After seeing initial market data, a participant selected an investment level (0% to 100% in increments of 10%) using one button box (shown on a slider bar on the screen) and submitted the decision using the other button box. In the human partner versions the next market result appeared 750 ms after the later of the two partners' choice was submitted. In the computer partner version the result was displayed 6 seconds later. 2 or 6 seconds later (depending on the experimental cohort, 1 or 2,3) the other partner's choice, was displayed by showing two red arrows on either side of the slider bar showing the level person's investment. This was repeated 20 times per market, for a grand total of 200 decisions.
Subject's behavioral data were analyzed in R (package nlme) [40], [41] (see Text S1 for full details).
|
10.1371/journal.pntd.0002501 | Effect of Maternal Schistosoma mansoni Infection and Praziquantel Treatment During Pregnancy on Schistosoma mansoni Infection and Immune Responsiveness among Offspring at Age Five Years | Offspring of Schistosoma mansoni-infected women in schistosomiasis-endemic areas may be sensitised in-utero. This may influence their immune responsiveness to schistosome infection and schistosomiasis-associated morbidity. Effects of praziquantel treatment of S. mansoni during pregnancy on risk of S. mansoni infection among offspring, and on their immune responsiveness when they become exposed to S. mansoni, are unknown. Here we examined effects of praziquantel treatment of S. mansoni during pregnancy on prevalence of S. mansoni and immune responsiveness among offspring at age five years.
In a trial in Uganda (ISRCTN32849447, http://www.controlled-trials.com/ISRCTN32849447/elliott), offspring of women treated with praziquantel or placebo during pregnancy were examined for S. mansoni infection and for cytokine and antibody responses to SWA and SEA, as well as for T cell expression of FoxP3, at age five years.
Of the 1343 children examined, 32 (2.4%) had S. mansoni infection at age five years based on a single stool sample. Infection prevalence did not differ between children of treated or untreated mothers. Cytokine (IFNγ, IL-5, IL-10 and IL-13) and antibody (IgG1, Ig4 and IgE) responses to SWA and SEA, and FoxP3 expression, were higher among infected than uninfected children. Praziquantel treatment of S. mansoni during pregnancy had no effect on immune responses, with the exception of IL-10 responses to SWA, which was higher in offspring of women that received praziquantel during pregnancy than those who did not.
We found no evidence that maternal S. mansoni infection and its treatment during pregnancy influence prevalence and intensity of S. mansoni infection or effector immune response to S. mansoni infection among offspring at age five years, but the observed effects on IL-10 responses to SWA suggest that maternal S. mansoni and its treatment during pregnancy may affect immunoregulatory responsiveness in childhood schistosomiasis. This might have implications for pathogenesis of the disease.
| Infections with the blood fluke Schistosoma mansoni that cause schistosomiasis (also called Bilharzia) were not usually treated during pregnancy until 2002, but in 2002 a World Health Organization (WHO) team of experts recommended that praziquantel treatment of S. mansoni during pregnancy should be done. However, there was limited information on the effects of maternal S. mansoni infection and treatment during pregnancy on the outcomes in the offspring. We conducted a study in the Entebbe peninsula within Lake Victoria in Uganda to examine whether maternal S. mansoni infection or its treatment during pregnancy may have effects on the children's susceptibility to the infection. The children were examined at age five years old for the level of S. mansoni infection and for immune responses to schistosomes. At five years old few of the children in our study cohort were infected with S. mansoni. Our findings suggest that maternal infection with, or praziquantel treatment of S. mansoni during pregnancy did not influence the level of S. mansoni infection among the offspring. However our findings suggest an influence on regulation of the body's immune responses to schistosomes, which may have some effect on the progress of disease manifestations. This is an issue that needs further investigation.
| In-utero sensitisation to schistosome antigens occurs in up to 50% of offspring born to women with schistosomiasis during pregnancy [1], [2]. Those who do not show an immune response may not have been exposed to relevant antigen (due to the placental barrier) or may have been exposed and developed tolerance. For those who are tolerant, this may be due to induction of regulatory mechanisms, or to deletion of schistosome antigen specific cells. When the baby is sensitised in-utero, profiles of infant responses resemble the mothers' responses: mixed type one, type two cytokine responses and a prominent IL-10 response, as shown by our recent study on cord blood responses to schistosome worm (SWA) and egg (SEA) [3] and one earlier study [2]. Our study suggested a direct correlation between maternal S. mansoni infection intensity and cord blood T cell sensitisation of the offspring [3]. Thus immune responses to schistosome antigens typical of those observed in adults can be established in-utero in offspring born to schistosome-infected women.
In communities where schistosomiasis is endemic, an age-dependent build up of immune response and resistance to S. mansoni infection has been described [4], [5]. There is a possibility that, in addition to this, immune responses established in-utero might influence early childhood responses and resistance to infection. On the other hand, children who are exposed in-utero, but do not become sensitised, may have acquired tolerance to schistosome antigens, which might render them more susceptible to infection, as has been observed among offspring of women with filariasis [6]. In-utero sensitisation to schistosome antigens might also influence schistosomiasis morbidity. For example, the low incidence of the acute schistosomiasis syndrome, Katayama fever [7], in endemic populations is attributed to early or in-utero sensitisation such that they are able to modulate the immune response and subsequent pathology associated with S. mansoni infection.
Balanced type one/type two immune responses are needed to limit morbidity [8], [9], [10], [11], [12], and responses established in-utero could influence this balance. Studies in mice support the hypothesis that in-utero anti-idiotypic exposure induces B and T cell responsiveness to schistosome antigens recognised by the idiotype [13], and that this exposure induces immmunoregulatory effects in subsequent schistosomiasis infection [14]. Recent immunoepidemiological studies in Kenya involving school age children (4–17 years) in S. mansoni endemic areas showed that regulatory cytokines are important in schistosomiasis morbidity [15], and that dysregulation of pro-inflammatory cytokines could be a mechanism involved in childhood hepatosplenomegaly observed in schistosomiasis.
It is estimated that in Africa up to 10 million women per year have schistosomiasis during pregnancy [16], and 40 million women of child-bearing age have schistosomiasis [17]. Here we report results on follow-up of offspring at five years old, exploring the hypothesis that schistosomiasis and its treatment during pregnancy could influence the immune responses of the children when they themselves are exposed to S. mansoni infection. Based on the immunoepidemiological studies that explored association of the various cytokines with S mansoni morbidity [15], [18], [19], we measured IFN, IL-5, IL-13 and IL-10 responses to schistosome worm (SWA) and egg (SEA) antigens and regulatory T-cells.
Written informed consent was obtained from each participant [20]. For all the children's participation, written informed consent was obtained from their parent or guardian. Ethical approval was obtained from the Science and Ethics Committees of the Uganda Virus Research Institute – Ministry of Health, the Uganda National Council for Science and Technology and the London School of Hygiene and Tropical Medicine.
This study was a follow-up of children within the ‘Entebbe Mother and Baby Study’ (EMaBS; ISRCTN 32849447) [20]. EMaBS was a randomised, double-blind placebo-controlled trial of praziquantel versus matching placebo and albendazole versus matching placebo during pregnancy using a 2×2 factorial design. It was conducted within Entebbe municipality and the adjacent Katabi sub-county in Uganda. Previous policy had excluded pregnant and lactating women from the control of schistosomiasis using praziquantel treatment [21], and although this policy was rescinded [22] there had been limited information on the effects of praziquantel treatment of schistosomiasis during pregnancy regarding anticipated benefits (on outcomes such as anaemia and birth weight) or possible adverse effects on the developing fetus. Hence there was considered to be the required equipoise to justify placebo-controlled trials [23]. Pregnant women were recruited between April 2003 and November 2005. At recruitment, the women provided stool samples that were examined for intestinal helminths including S. mansoni infection using the Kato Katz method [24]. Children were followed up annually and stool and blood samples were collected at each visit. For the current study, we focussed on offspring of the study women at age five years.
The stool samples were examined for helminth infections including S. mansoni by the Kato Katz method for all children. Blood samples (approximately 6 mL) were processed for immune responses, plasma was separated and stored at −80°C, and peripheral blood mononuclear cells (PBMCs) were isolated and stored in liquid nitrogen. For schistosome immunology studies, whole blood culture for responses to schistosome antigens, and anti-schistosome antibody assays, were performed in a subgroup of 436 children, comprising 190 whose mothers had S. mansoni infection during pregnancy and, for comparison, a consecutive series of 246 children whose mothers did not.
Whole blood stimulation for cytokine responses to SWA and SEA was performed as previously described [3], [25], [26], [27]. SWA and SEA antigens were supplied by Professor David Dunne (Cambridge University, UK). The levels of endotoxin in the antigen preparations were 0.086 EU/mg for SWA and 0.175 EU/mg for SEA. On dilution of the antigen to working concentration of 10 µg/ml, the endotoxin levels in culture were negligible (<0.1 ng/ml) and unlikely to influence cytokine responses in whole blood culture. Briefly, heparinised blood was diluted 1 in 4 with serum-free medium (RPMI supplemented with glutamine, penicillin and streptomycin) and stimulated with SWA, SEA or phytohaemagglutinin (PHA- Sigma, UK) at final concentrations of 10 µg/ml in 96-well round-bottomed cell culture plates (TC Microwell, NUNC A/S, Roskelde, Denmark), or left unstimulated. Media (200 µl per well) was added to the plates and incubated at 37°C and 5% CO2. Supernatants were harvested after six days and left to stand at room temperature for one hour with viral inactivation buffer (0.03% tributyl phosphate and 1% Tween 80 (Sigma)) before storage at −80°C until needed for analysis.
The supernatants were assessed for interferon gamma (IFNγ), interleukin (IL)-5, IL-13 and IL-10 responses to SWA and SEA. IFNγ, IL-5 and IL-10 concentration in the supernatants was measured using OptEIA ELISA Kits (BD Pharmingen, USA) while IL-13 was measured using antibody pairs (BD Pharmingen, USA), with standards from the National Institute for Biological Standards and Controls (NIBSC, UK). The sensitivity of each assay, and cut-off for detectable responses, was the lowest concentration on the standard curve (9 pg/ml for IFNγ, 8 pg/ml for IL-5 and IL-10 and 16 pg/ml for IL-13). Cytokine levels in unstimulated cultures were generally low (Figure S1). To obtain antigen-specific responses, cytokine concentrations in unstimulated wells were subtracted from the measured concentrations in antigen-stimulated wells.
Plasma levels of IgG1, IgG4 and IgE to SWA and SEA were measured in duplicate by ELISA [24] with a high through-put semi-automated system. Briefly, flat bottomed 384-well microplates (Greiner Bio-One Ltd, Stonehouse, UK) were coated with SWA (8 µg/ml) or SEA (2.4 µg/ml) in 25 µl bicarbonate coating buffer and incubated overnight at 4°C. A standard positive pool and myeloma immunoglobulin isotype were added to allow a standard curve for quantification of antibodies to be generated. Samples were added at 25 µl per well at dilutions of 1/200 for IgG1 and IgG4 or 1/20 for IgE and incubated overnight at 4°C. Biotinylated mouse anti-human monoclonal antibodies (BD Pharmingen San Diego USA) were used for detection and Poly-HRP-streptavidin conjugate (Sanquin, Netherlands) was added at a 1/4000 dilution. Plates were developed with OPD substrate and stopped with 2 M sulphuric acid on observing the colour change. Optical densities (ODs) were recorded using Gen5 software. Concentrations were calculated from ODs by interpolation from standard curves. Normal European serum was included as a negative control to determine the cut-offs for positive levels. The cut-offs were set at three standard deviations above the mean of the normal European sera, which was 0.4 µg/ml for IgG1 to SWA, 0.1 µg/mL for IgG1 to SEA, 0.544 µg/ml for IgG4 to SWA, 0.042 µg/mL for IgG4 to SEA, 0.16 µg/mL for IgE to SWA and 0.04 µg/mL for IgE to SEA.
PBMCs were stained for surface CD4, CD25, CD127 and intracellular FoxP3 to determine regulatory T cell (Treg) frequencies. Antibodies for surface marker staining were fluorescine isothiocyanate (FITC) labelled mouse anti-human CD4 (Invitogen-Caltag, Camarillo, USA) , phycoerythrin (PE) labelled mouse antihuman CD25 (Invitogen-Caltag, Camarillo, USA) and mouse anti-human CD127-Alexa647 (BD Pharmingen San Diego USA). Intracellular FoxP3 was detected using rat anti-human FoxP3-PE-cy5 (eBioscience, San Diego USA). PBMCs were thawed, washed and cell numbers determined. Between 1 and 2 million cells were stained in V-bottom 96-well plates. The cells were stained for 20 minutes in the dark at 4°C in 100 µl surface mix comprising surface antibodies CD4-FITC, CD25-PE and CD127-Alexa647 at dilutions 1∶40, 1∶20 and 1∶10 respectively. The cells were washed, permeabilized and stained for FoxP3 according to eBioscience's recommended protocol. The cells were acquired using FACS Diva software on a 17-colour flow cytometer (LSR II BD Biosciences) within 12 hours of staining. The acquired data was analysed using FlowJo software. Regulatory T cells (Tregs) were defined as CD4+ CD127low, CD25+ and FoxP3+. An example of the gating strategy applied can be viewed in Figure S2. The frequency of cells expressing FoxP3 was expressed as a percentage of CD4+ cells and the expression level of FoxP3 expressed in terms of geometric mean fluorescent intensity (GMFI).
The analysis had four main objectives. (1) In the whole cohort, we examined whether S. mansoni infection during pregnancy was associated with childhood S. mansoni infection among the offspring at age five years. This was done using logistic regression to calculate odds ratios of infection among offspring at age five years, comparing prevalence among children of women who had S. mansoni infection during pregnancy with children of those who did not. For this observational analysis we also adjusted for factors that had a potential association with maternal S. mansoni infection in both mother and child, such as maternal residence, contact with lake water, maternal education, household socioeconomic status, maternal hookworm and Mansonella infections, using multivariable logistic regression. (2) In the whole cohort, we examined the effect of praziquantel treatment of S. mansoni infection during pregnancy on S. mansoni infection among the offspring at age five years. This was done by logistic regression comparing children of women that were treated, with children of women that were not treated. The possibility of a differential effect of maternal praziquantel treatment by maternal S. mansoni status was evaluated by conducting a subgroup analysis and calculating the p-value for the interaction. Since the prevalence of S. mansoni infection among children at age five years was very low, we further explored the consistency of the observed effect for objectives 1 and 2 using data collected at all annual visits, from age one year to age eight years, using repeated measures analysis combining available infection data from all time points. (3) In the subgroup selected for schistosome immunology studies, we examined associations between the schistosome-specific immune responses of the offspring at age five years and maternal S. mansoni infection during pregnancy, and associations between schistosome-specific immune responses in five year olds and praziquantel treatment during pregnancy. This was done by comparing cytokine and antibody responses to SWA and SEA between offspring of women who had S. mansoni infection during pregnancy and those that did not (restricting to the placebo group), and between children of women of the praziquantel and placebo groups (restricting to infected women). Cytokine and antibody responses had skewed distributions. The data were therefore log10 transformed and analysed by linear regression with bootstrapping to estimate bias-corrected accelerated regression coefficients and 95% confidence intervals that were back-transformed to give geometric mean ratios. Levels were also compared between infected and uninfected children using Wilcoxon rank sum test. (4) We examined the effect of maternal schistosomiasis and treatment during pregnancy on regulatory T cells in the offspring. This was done by comparing the proportion of CD4+ cells that expressed FoxP3, and the levels of FoxP3 expression, between children of women with schistosomiasis and those without, and also between children of infected women that were treated or not. Similarly, these comparisons were done by linear regression with bootstrapping to estimate bias-corrected accelerated regression coefficients and 95% confidence intervals that were transformed to give geometric mean ratios.
Of the 2507 women enrolled in the main study, 2345 live births were recorded and 1343 of these children were examined for S. mansoni at age five years, 703 in the placebo group and 640 in the praziquantel group. The baseline characteristics of the mothers as recorded at enrolment into the study during pregnancy did not significantly differ between the praziquantel and placebo arms (Table 1). Among the placebo group (n = 703), 571 (81.2%) were uninfected, 85 (12.1%) had light, 24 (3.4%) moderate and 23 (3.3%) heavy infection. Among the praziquantel group (n = 640), 519 (81.1%) were uninfected, 80 (12.5%) had light, 23 (3.6%) moderate and 18 (2.8%) heavy infection. Thus of the 1343 children, 253 (132 placebo, 121 praziquantel) were of women that had S. mansoni during pregnancy and 1090 (571 placebo, 519 praziquantel) were of uninfected women. Women that had S. mansoni at enrolment were less likely to live in Entebbe town and more likely to have regular contact with lake water (Table 1).
Overall, 32 of the 1343 children (2.4%) had S. mansoni infection at age five years based on a single stool sample. Of these 11 children (3 placebo group and 8 praziquantel group) were of women that had S mansoni infection during pregnancy while 21 (10 placebo group, 11 praziquantel group) were of uninfected women. Despite the few children infected, some had a high infection intensity with a geometric mean infection intensity of 106 eggs per gram (epg) of stool (minimum 12 epg, maximum 3384 epg). Fifteen (45.9%) had light (<100 epg), twelve (37.5%) had moderate (100–399 epg) and five (15.6%) had heavy (>400 epg) infections [21]. Among the children with S mansoni in the placebo group (n = 13), 5 had light, 6 moderate and 2 heavy infection while among those in the praziquantel group (n = 19), 10 had light, 6 moderate and 3 heavy S. mansoni infection.
Soil transmitted helminth seen among the children at age 5 years were hookworm 6 (0.4%), Ascaris 13 (1.0%), and Trichuris 73 (5.5%) and were mainly among the children with no S. mansoni infection. No children with S mansoni had hookworm or Ascaris and four children had co-infection of S mansoni and Trichuris.
We explored the association of factors including maternal helminth infections and treatment, location of residence, contact with lake water, maternal education, and household socioeconomic status (a score based on the house building materials, number of rooms and items owned), with the presence of S. mansoni infection in the children. As expected, contact with lake water and location of residence were associated with the children having S. mansoni infection (Table 2). Children whose mothers who resided in Kigungu or Katabi village (lakeside locations) were more likely to have S. mansoni infection than those from Entebbe town, which is not directly on the lakeshore. It was also noted that the proportion of children that reported having been ‘in contact with lake water frequently’ was greater for Kigungu (38.5%) and Katabi village (24.8%) than Entebbe town (14.5%), Katabi town (6.9%) or Manyago (11.5%) (P<0.001). Although crude analysis suggested that maternal S. mansoni infection was associated with S. mansoni infection among the children at five years old, on adjusting for the other factors (location of residence, contact with lake water, maternal education, and household socioeconomic status) this association was lost. Trial analysis showed that praziquantel treatment of the women during pregnancy had a tendency to increase the chances of their offspring being infected (Table 3), although this was not statistically significant. The effect of praziquantel treatment appeared somewhat stronger among children of mothers with S. mansoni, although again it was not significantly different to the effect of treatment among children of uninfected mothers (interaction p = 0.249). Because the prevalence of S. mansoni infection was low at five years, we also explored for effects of maternal praziquantel treatment during pregnancy on prevalence of S. mansoni infection among the study cohort children using a repeated measures analysis combining all data from annual time points to date (comprising age one to six for the youngest, and one to eight years for the oldest children, depending on the child's age at the time of the analysis). Again, no association was observed OR 0.80 (95% CI 0.57, 1.13 p = 0.21).
Four hundred and thirty six children were examined for IFNγ, IL-5, IL-10 and IL-13 responses to SWA and SEA. Of these, 207 were the offspring of women who received praziquantel treatment during pregnancy (praziquantel group) and 229 were of women who did not receive the treatment (placebo group). Restricting our initial analysis to the placebo group, we examined whether cytokine responses to SWA or SEA among the children differed between children of S. mansoni infected women and those of uninfected women (Table S1 in Text S1). Maternal S. mansoni infection did not show any significant associations with cytokine responses among the children, with the exception of IL-10 responses to SWA; children of women that had S. mansoni during pregnancy had lower IL-10 responses to SWA (geometric mean (cytokine concentration +1) = 6.1 pg/mL) than children of uninfected women (geometric mean (cytokine concentration+1) = 3.5 pg/mL, geometric mean ratio 0.58, 95%CI 0.39, 0.90)). We examined for associations between cytokine responses and maternal infection intensity during pregnancy and found no significant correlation (data not shown).
When we examined for effects of praziquantel treatment of S. mansoni during pregnancy on cytokine responses among the children, significant effects were only seen for IL-10 responses to SWA where the responses were higher among children of women who received praziquantel treatment during pregnancy (geometric mean (cytokine concentration +1) = 3.5 pg/mL) than those of women who were not treated (geometric mean (cytokine concentration +1) = 7.0 pg/mL, (geometric mean ratio 1.97, 95%CI 1.25, 3.23)). More information can be viewed in Table S2 in Text S1. Of the children examined for cytokine responses, 20 children had S. mansoni infection at age five years. As expected, cytokine responses were higher in children with S. mansoni than those without (Figure 1) and the cytokine profile among the children was a mixed type 1, type 2, and regulatory response. Among the uninfected children, praziquantel treatment during pregnancy did not influence cytokine responses to SWA or SEA except for IL-10, for which the responses to SWA were still higher among children of women who received praziquantel treatment compared with those of untreated women (geometric means 6.1 and 4.4 pg/mL, respectively geometric mean ratio 1.38 (95% CI 1.01, 1.90); Figure 2). The low number of infected children limited similar subgroup analysis among the infected children. Nevertheless, the responses among the few (4 placebo group, 16 praziquantel group) infected children at age five years presented a different picture from the responses among the uninfected children (Figure 3; compare A with B for SWA or C with D for SEA). Praziquantel treatment of the women during pregnancy showed a tendency to reduce IFNγ, IL-5, and IL-13 responses to SWA and SEA, but not IL-10 in infected children.
Levels of IgG1, IgG4 and IgE to SWA and SEA were measured in 226 children at age five years. Of these, 142 (63 placebo and 79 praziquantel group) were the off-spring of women infected with S. mansoni during pregnancy and 84 (54 placebo and 30 praziquantel group) of uninfected women. Among the 226 children, SWA-specific IgG1 antibodies were detectable in 79 (35.0%) children, with SWA-specific IgG4 and IgE antibodies detected in 12 (5.3%) children. SEA-specific IgG1, IgG4 and IgE antibodies were detectable in 56 (24.8%), 52 (23.0%) and 33 (14.6%) children respectively. Of the children examined for antibody responses, 28 had S. mansoni at age five years. As expected, antibody levels showed associations with the children's infection status. Of the 28 children that had S .mansoni infection, IgG1, IgG4 and IgE to SWA were detectable in 21 (75%), 7 (25%) and 9 (32.1%) of the children respectively, and IgG1, IgG4 and IgE to SEA were detectable in 21 (75%), 19 (67.9%) and 15 (53.6%) children, respectively. Levels of IgG1, IgG4 and IgE to SWA or SEA were minimal among uninfected children and were significantly lower than the levels among the S. mansoni infected children (Figure 3, p<0.001 for the respective comparisons). Maternal infection intensity was not associated with antibody levels (data not shown) and there were no significant effects of praziquantel treatment during pregnancy on antibody levels among the offspring (Figure 4).
Considering that S. mansoni infection has been associated with developing a strong immune regulatory response early in the initial encounter between the host and parasite [28], we explored whether maternal infection or treatment during pregnancy may influence long-term Treg expression in the offspring. Two hundred and nine children at age five years were examined for CD4+ T cell expression of FoxP3. Of these, 128 were the offspring of women infected with S. mansoni during pregnancy and 81 of uninfected women. Of the infected women, 68 received praziquantel treatment and 60 did not. We examined for effects of S. mansoni infection and praziquantel treatment during pregnancy on FoxP3 expression among offspring of women that were not treated during pregnancy. Maternal S. mansoni infection was not associated with FoxP3 expression in the children at age five years (data not shown) and praziquantel treatment during pregnancy did not influence expression of FoxP3 among the children (Table 4). Of the 209 children examined for FoxP3 expression, only 17 were infected with S. mansoni at age five years. FoxP3 expression was higher among the S. mansoni infected children than the uninfected children (Figure S3). We examined for associations between FoxP3 Treg expression and cytokine responses and found no significant associations (data not shown). This was done on pooled data as well as separately according to the children S. mansoni infection status.
This study explored the effects of S. mansoni infection and of praziquantel treatment during pregnancy on S. mansoni infection and immune responsiveness among the offspring at the age of five years. This is an age by which the children are expected to have had their first exposure to infection with S. mansoni through playing in lake water. We found no conclusive evidence of an effect of maternal schistosomiasis, or of its treatment during pregnancy, on susceptibility to S. mansoni infection in the children, or on their anti-schistosome antibody levels. Similarly, no effect on cytokine responses to schistosome antigens was observed in the children apart from IL-10 responses to SWA, for which children of infected mothers tended to show lower responses than children of uninfected mothers, and children of praziquantel-treated mothers showed higher responses than children of placebo-treated mothers.
In this study the prevalence of S. mansoni among the five years old children was 2.4%. This was based on a single stool sample, implying that a number of infected children could have been misclassified due to the low sensitivity of a single stool sample Kato Katz test [29], and that these numbers may represent an under estimate of the actual prevalence [29], [30]. The observed prevalence was less than what has been reported from epidemiological surveys involving preschool children in other areas on the shores of Lake Victoria [31], [32], [33]. These studies were different from our study in that they applied consecutive multiple stool sample Kato Katz analysis. Further, the Entebbe area, where our study was carried out, is relatively urbanised and developing rapidly, and has wide availability of treated piped water. It is thus expected to have relatively low S. mansoni transmission. However, we noted that even under such low transmission conditions, infected young children might have quite high infection intensities. Children who were resident in the peri-urban areas of Kigungu and off road areas of Katabi were more likely to be infected with S. mansoni than children in the urban centres of Entebbe town and Katabi, and this is a direct connection to the frequency of contact with lake water. This underscores the concern that in situations where younger children are exposed to potential sources of infection, they will acquire the infection at an early age. This has implications on schistosomiasis control programs that have tended to neglect preschool children [34].
This study found no evidence that maternal S. mansoni infection per se or maternal praziquantel treatment may predispose children to infection with S. mansoni. Others have reported that children born to women with filariasis [35], and those affected by placental malaria [36], are likely to be infected earlier with the respective parasites, an effect that has been attributed to the induction of tolerance in the offspring [6]. Considering that our study included very few children showing infection at age five years, there is a further need to explore this effect in an area of high S. mansoni transmission. However, repeated measures analysis showed consistent observation among the cohort children at ages one to eight years gave further evidence that maternal praziquantel of S. mansoni during pregnancy may not influence prevalence of infection among offspring.
Cytokine responses among the children had a mixed type 1/type 2/regulatory profile. The same cytokine profile was found among mothers of the children in this study cohort [27], and immunoepidemiological studies have also shown a mixed responses with strong regulatory cytokine phenotype [37], [38]. It was also noted that expression of FoxP3 was prominent among the S. mansoni infected children. Our findings show that, in S. mansoni endemic areas, the profile of T-cell responses typical to those in S. mansoni chronically infected adults may be established in early childhood. For the observational analyses, where we compare infected and uninfected children, we observed quite marked differences in the immune responses in the children. Considering that we applied a single stool sample Kato Katz for diagnosis, misclassification of infected children as uninfected might have weakened the differences observed. Therefore these results are probably a conservative estimate of the effects of infection on immune responses in the children.
Praziquantel treatment during pregnancy did not influence most of the assayed cytokine responses to SWA or SEA. This finding is consistent with our earlier report in which we did not find significant effects of praziquantel treatment during pregnancy on cytokine responses to schistosome antigens among the offspring, in either cord blood or at age one year [3]. Of some interest, however, was the IL-10 response to SWA. This was an exception in that it was low in the offspring of mothers with S. mansoni infection and was increased by praziquantel treatment during pregnancy. IL-10 is a regulatory cytokine, capable of switching off both type 1 and type 2 immune responses, and could possibly explain why we saw consistently low IFNγ, IL-5, and IL-13 responses among infected offspring of women who received praziquantel treatment during pregnancy compared to offspring of those who did not (Figure 3 B&C). This could have implications on S. mansoni associated morbidity among the offspring. For example, low regulatory and type 2 responses to schistosome antigens have been associated with hepatosplenomegaly in childhood schistosomiasis [15]. In addition, low levels of regulatory cytokines and correspondingly higher levels of type 2 cytokines have been associated with a high risk of peri-portal fibrosis in chronic S. mansoni [18], [39] and S. japonicum [40]. In S. japonicum infection pro-inflammatory cytokines including IL-1, IL-6 and TNF have been associated with morbidity outcomes [41]. On the other hand, immunosuppressive responses are presumed to allow establishment and prolong survival of the worms in the host, preventing elimination of the parasite while at the same time minimizing tissue damage [42], [43], [44]. In our study the observations on IL-10, a regulatory cytokine, were not reflective of the regulatory T-cell data in terms of FoxP3 expression. However, IL-10 responses were measured after in vitro antigen stimulation of whole blood cultures, and it is likely that sources other than T cells also contributed to IL10 production. The observed effect on IL-10 response may have implications on schistosomiasis-associated morbidity, and this need further immunological and morbidity studies
In conclusion, and considering that the low S. mansoni prevalence in this study was a major limitation, our findings show no evidence that maternal S. mansoni infection or its treatment during pregnancy has an influence on the prevalence of S. mansoni infection among the offspring at age five years. However, our findings do suggest that maternal S. mansoni and its treatment during pregnancy may affect some cytokine responses among S. mansoni infected offspring at age five years. The magnitude of such an effect and possible implications in childhood schistosomiasis need to be elucidated further by conducting a larger study involving greater numbers of infected children.
|
10.1371/journal.pcbi.1004860 | Learning Pitch with STDP: A Computational Model of Place and Temporal Pitch Perception Using Spiking Neural Networks | Pitch perception is important for understanding speech prosody, music perception, recognizing tones in tonal languages, and perceiving speech in noisy environments. The two principal pitch perception theories consider the place of maximum neural excitation along the auditory nerve and the temporal pattern of the auditory neurons’ action potentials (spikes) as pitch cues. This paper describes a biophysical mechanism by which fine-structure temporal information can be extracted from the spikes generated at the auditory periphery. Deriving meaningful pitch-related information from spike times requires neural structures specialized in capturing synchronous or correlated activity from amongst neural events. The emergence of such pitch-processing neural mechanisms is described through a computational model of auditory processing. Simulation results show that a correlation-based, unsupervised, spike-based form of Hebbian learning can explain the development of neural structures required for recognizing the pitch of simple and complex tones, with or without the fundamental frequency. The temporal code is robust to variations in the spectral shape of the signal and thus can explain the phenomenon of pitch constancy.
| Pitch is the perceptual correlate of sound frequency. Our auditory system has a sophisticated mechanism to process and perceive the neural information corresponding to pitch. This mechanism employs both the place and the temporal pattern of pitch-evoked neural events. Based on the known functions of the auditory system, we develop a computational model of pitch perception using a network of neurons with modifiable connections. We demonstrate that a well-known neural learning rule that is based on the timing of the neural events can identify and strengthen the neuronal connections that are most effective for the extraction of pitch. By providing an insight into how the auditory system interprets pitch information, the results of our study can be used to develop improved sound processing strategies for cochlear implants. In cochlear implant hearing, auditory percept is generated by stimulating the auditory neurons with controlled electrical impulses, enhancing which with the help of the model would lead to a better representation of pitch and would subsequently improve music perception and speech understanding in noisy environments in cochlear implant users.
| The existence of a pitch processing center or a group of specialized “pitch neurons” in the mammalian auditory system has been debated in recent years. For example, through single unit recordings, Bendor and Wang [1] found a potential pitch center in the anterolateral border of primary auditory cortex in marmoset monkeys. These pitch neurons were characterized by sustained spiking in response to their preferred pitch, evoked by a pure tone or a harmonic complex. Human brain analogues of monkey’s lateral primary auditory cortex, postulated by Bendor and Wang [1] to be the pitch center, has also been found to perform pitch-related processing. For example, through positron emission tomography (PET), Zatorre and Belin [2] found that areas in the lateral Heschl’s gyrus responded to the pitch of pure tones. Using functional magnetic resonance imaging (fMRI), Patterson et al. [3] found the same cortical area to be consistently activated by periodic stimuli with a defined pitch. Penagos et al. [4] also confirmed the sensitivity of the Heschl’s gyrus area to the pitch of harmonic complexes through fMRI investigations.
Possible locations for pitch sensitive neural units along the auditory pathway have been postulated in a number of modelling studies. For example, the coincidence detector neurons of the model of Shamma and Klein [5] required strong phase-locked inputs; therefore, Shamma and Klein proposed the inferior colliculus as a possible pitch processing site. Inferior colliculus neurons receive inputs from the cochlear nucleus neurons that, due to having onset type cells [6], generate spectrally and temporally sharp responses suitable for coincidence detector units.
The functional role of the cochlear nucleus in varying the timing of spikes has been observed in earlier studies [7]. Spike time variation in the auditory nerve is partially caused by the cochlear travelling wave and results in the spiking of neurons with high characteristic frequency (CF) several milliseconds prior to low-CF neurons in response to a stimulus. Through experimental studies, Oertel et al. [8] showed that it was particularly octopus cells in the cochlear nucleus that had the ability to detect spiking coincidences among a population of innervating auditory nerve fibers. The octopus cells were found to compensate for the different arrival times of the auditory nerve spikes. The ability of octopus cells to extract precise temporal information from the auditory nerve was related to their special anatomical structure and biophysical characteristics [9]. The compensating role of octopus cells was further investigated in a modelling study by Spencer et al. [10]. They showed that different arrival times of the auditory nerve spikes were compensated by proportional dendritic delays in the octopus cells, thus enabling the detection of the spike coincidences to be carried out more effectively in later stages.
Given the uncertainties that still exist about the physiology of pitch centers in the auditory system, the focus of this paper is on modelling the known functions of the possible pitch neurons rather than replicating the anatomical stages (and their interactions) involved in extracting the pitch information. According to existing literature, pitch sensitive neurons have a preferred pitch [11], exhibit sustained spiking activity [12,13], respond to pitch as a unified entity (regardless of the spectral shape of the stimuli) [1], and are located in the subcortical part of the auditory pathway [5,7].
In the model developed in this paper, it was assumed that the spectral (place) and temporal pitch information were processed by different populations of neurons. These neuronal populations, despite being connected to each other, used different mechanisms to extract their component of pitch information. One reason for considering such a neuronal architecture was the functioning differences between the two hemispheres of the brain in processing the pitch of stimulus. For example, Zatorre and Belin [2] found that the right hemisphere exhibited a stronger response to pitch-related spectral variations, while the left hemisphere showed higher degrees of activation in response to temporal variations of the stimulus. Based on these observations, Zatorre and Belin [2] suggested that the auditory system had two parallel processing sub-systems that provided different spectral and temporal resolutions required for perceiving a wide range of stimuli, such as speech and music. Poeppel [14] proposed that the hemispheric functional differences were a result of different timescale integration windows applied by each hemisphere (i.e., shorter for the left and longer for the right hemisphere) when processing auditory information. In a lesion study, Johnsrude et al. [15] identified the right Heschl’s gyrus as responsible for making judgments on the direction of pitch changes (i.e., pitch ranking) because patients whose right temporal lobes were partially resected showed higher pitch-difference thresholds compared to the control group. They also found that, unlike pitch ranking, a pitch discrimination task (detecting a pitch difference regardless of direction) could be performed by either hemisphere.
Another observation that inspired the use of a separate population of neurons for temporal pitch processing in the auditory system is the special organization of brain tissue [16] as illustrated in Fig 1A. Inputs to the auditory cortex can be presented in terms of spatio-temporal maps that describe the activity of spatially different neurons over time [17]. Fig 1B shows an example of a spatio-temporal pattern for a synthesized vowel /ɑ/, with F0 of 110 Hz and the first three formants located at 710 Hz, 1150 Hz, and 2700 Hz, using a model of auditory periphery developed by Zilany et al. [18]. Observations have shown that tonotopicity (viz., neurons responding to a frequency based on their location, leading to a place-frequency map) exists in the areas of higher auditory processing centers like the auditory cortex [19]. Tonotopicity (indicated by the color map in Fig 1A) thus accounts for the extraction of power-based or place features from the signal. Spectral features including the first two formants are strongly represented in the spatio-temporal patterns. The first two formants are indicated with grey arrows in Fig 1B. The role of the tonotopically-arranged areas could be interpreted as averaging the spatio-temporal patterns over time, resulting in a profile of activity rates across the auditory nerve. Fig 1C shows the corresponding rate profile (normalized to maximum) that is considered as the place code of pitch. The first two formants (indicated with grey arrows) have strong representation in the place code shown in Fig 1C. The cortex also has columnar divisions with connections to the tonotopically-arranged areas. According to this area-column synergy, measuring the activity across columns would provide a temporal code for pitch. Fig 1D represents a possible temporal code, extraction of which is the topic of this paper. Of note is the spacing between the peaks of the temporal code in Fig 1D that corresponds to the period of stimulus (i.e., ~9 ms), which is the F0 of the vowel.
Unlike the place code, simple averaging would not capture the temporal code because of the temporal variations (e.g., jitter) that naturally exists in the neural code generated by the auditory nerve. Therefore, capturing the fine-time structures, such as spike coincidences, from the neural code required an intermediate processing stage that adjusted the spike timings before any sort of averaging occurred. This intermediate processing stage would possibly replicate the functional role of the cochlear nucleus [7]. Enabling a model of cochlear nucleus to perform this function required specific neural connectivity that could arise through neural plasticity.
The computational analogue of the cortical structure shown in Fig 1A is depicted in Fig 1E and 1F. The two phases were defined by the set of modelling components that together extracted the place (Phase I) or temporal (Phase II) cues for pitch perception. Phase I performed temporal averaging for auditory neurons. Phase II provided a biologically-inspired computational substrate for producing precisely-timed spikes that would lead to an efficient temporal pitch code. A spiking neural network with plastic input/output synapses constituted the second phase. It is apparent that Fig 1C and 1D represent the outputs of Phase I and Phase II of the analogue shown in Fig 1E and 1F, respectively.
This section describes the data used in the simulations and the process of place and temporal pitch information extraction. Extraction of the place code is described jointly with the model of auditory periphery in the Auditory periphery (Phase I) section. Extracting the temporal code requires a neural setup that is properly adjusted to fit the pitch perception task. The neural components and associated learning equations are presented in the Neural setup and Synaptic adjustments sections, respectively.
Sound stimuli for this study were synthesized and real-world sounds that a typical listener might experience. Synthesized sounds are advantageous because they can be generated easily and precisely. However, for the sake of generality, real-world recordings from various musical instruments were also included in the simulations. Types of stimuli and their descriptions are given in Table 1. All stimuli were 0.5 s long, had a loudness of 60 dB SPL, and were sampled at 16,000 sample/s.
Pure tones were desirable stimuli because they have simple spectral shapes and evoke salient pitches. Speech is possibly the most common sound stimulus that one might experience; therefore, voiced speech tokens (sung vowels /ɑ/ and /i/) were well-suited to the purpose of this study. Synthetically generated variations of the vowel stimuli (/ɑ/T and /i/T) were also included in this study. This enabled the investigation of how the auditory system encoded pitch in real-life listening conditions such as telephone conversation, wherein the low-frequency contents of speech, including the F0 for most speakers, would be eliminated. The telephone line was simulated by high-pass filtering the original vowels using a high-pass FIR (finite impulse response) filter with a sharp cut-off frequency of 300 Hz. The filter was designed using MATLAB Filter Design & Analysis Tool with Fstop = 300 Hz, Fpass = 350 Hz, Astop = 80 dB, and Apass = 1dB. Filter order was 110 (minimum) and sampling rate was 16,000 sample/s.
Musical instruments provide a variety of spectral shapes and were included in the simulations to investigate the behavior of the model in response to spectrally-different sounds. The instruments were selected based on availability in the database, spectral shape variety, and the range of pitches that each instrument could generate.
The purpose of this study was to investigate how place and temporal pitch cues were extracted from the spatio-temporal maps generated by the auditory nerve. The former was assumed to be a profile of rates (temporal averages) associated with different auditory neurons. Extraction of the place cues is described jointly with the generation of the spatio-temporal maps in the Auditory periphery (Phase I) section. The Temporal code of pitch (Phase II) section explains the neural structure configuration and the associated learning process leading to pitch-related temporal information.
During the course of learning, synaptic weight changes directed by STDP gradually decreased the initial spiking rate for each pitch neuron to an asymptote rate of ~30 spike/s. Synaptic weights were recorded as the learning progressed. Fig 3 shows the input/output connectivity patterns (wij) that developed for different types of stimuli at an initial (top row) and a final (bottom row) stage of learning. Graphs (A-D) are associated with the weight patterns recorded after 500 s presentation of pure tones, /ɑ/ vowels, /i/ vowels, and piano sounds, respectively, examples of each were shown in Fig 2A–2D. In each graph, input neurons (j index) are shown along the abscissa, sorted by their CF (in kHz). Output or pitch neurons (i index) are presented along the ordinate, sorted based on the pitch that they represent. Graphs (E-H) show the corresponding emerged patterns when learning progressed for 5000 s. Fig 3I and 3J show the average synaptic weight patterns that emerged after 500 s of mixed-stimulus learning and after 5000 s of mixed-stimulus learning, respectively.
Because pitch categories, as opposed to spectral shapes, were consistent during the course of learning leading to the patterns in Fig 3E–3H, it could be inferred that the common behavior observed amongst the four patterns in Fig 3E–3H would be associated with pitch. Therefore, the “wrinkle” (peak-trough-peak sequence) that started from the bottom-left and moved towards the top-middle in each graph would correspond to pitch. In all the four patterns, for each pitch neuron, the trough of the wrinkle appeared at input neurons with CFs similar to the pitch categories. This “pitch curve” was the only apparent feature in pure tone patterns (Fig 3E) due to their simple spectra. Vowels (Fig 3F and 3G), on the other hand, had spectral power concentrated around the formants. Connections originating from formant locations were strongly affected by STDP due to high driving rates. Formants thus resulted in vertical stripes in the synaptic patterns in Fig 3F and 3G. As shown in Fig 2L, piano stimuli led to a distinct harmonic structure with evenly-distributed energy across the low-frequency half of the cochlear regions, which resulted in harmonically-related pitch patterns (Fig 3H). The timbre-independent pitch curve was replicated by the mixed-stimuli model (Fig 3J) as well; however, due to various spectral shapes presented to the model during learning, type-specific behavior observed in Fig 3F–3H is absent in the weight patterns of the mixed-stimuli model.
In order to measure the efficiency of STDP in adjusting the spike timings (e.g., in terms of producing phase-locked responses), vector strength was calculated for pitch neurons during early and late stages of learning. Vector strength is a well-known measure of phase locking or stimulus-response synchrony; it describes a phase relationship between the periodic input stimuli and the discharge of the output neuron [38]. Fig 4 presents vector strength matrices (stacked vector strengths from all the pitch neurons) computed from 5 s initial (A) and final (B) intervals, using the mixed-stimuli model. According to the noisy pattern of Fig 4A, during the early stages of learning − when the initial uniform synaptic weights had not been modified by the plasticity rule–the pitch neurons generated spikes at random times. However, after sufficient learning (Fig 4B), it was observed that for each pitch neuron, vector strength was strongest for the input stimuli that had the same pitch as that represented by the pitch neuron (the diagonal lines). This indicated that STDP had adjusted the connection strengths so that the spikes were more likely generated in-phase with a sinusoid of the same frequency, i.e., one spike per sinusoidal peak.
A question was then posed as to whether the temporal adjustment of the spikes would be affected by the absence of F0. To investigate this matter, in another simulation, the spiking neural network was exclusively presented with high-pass filtered vowels during the course of STDP learning. Fig 4C shows the resulting vector strength matrix for a final 5 s interval. It was observed that spike times became entrained to F0 by STDP, even when F0 was missing.
The inter-spike-interval histogram (ISIH), has proven to be an efficient measure of pitch, compatible with pitch-related psychophysical findings for a wide range of stimuli and levels [23]. Cariani and Delgutte [23] found that peak locations and relative amplitudes in a histogram of inter-spike-intervals provided a cue for pitch that was robust against sound level changes and spectral shape variations. The latter thus provided an explanation for pitch constancy at a neural level.
In this study, the most frequent or the dominant interval was considered as the temporal code of pitch. Although deriving the most common interval was possible by taking into account the spiking activity of a single pitch neuron [39], it was decided to use the ISIH of the population of pitch neurons (a.k.a., pooled ISIH) to account for the role of higher-order pitch processing centers in integrating information across pitch neurons. This was necessary to explain phenomena such as perception of the missing-F0 pitch that reportedly engages higher-order auditory processing centers [40].
To calculate the ISIH for each pitch category, the mixed-stimuli trained model shown in Fig 3J was presented with each of the 29 pure tones for 0.5 s. The resulting inter-spike intervals were pooled across the 29 pitch neurons and distributed in 1 ms bins. Fig 5A–5C shows examples of the first 50 ms of the pooled ISIHs for pitch categories of 370 Hz, 131 Hz, and 104 Hz, respectively. For better visualization, all the histograms were smoothed (using a moving average filter with a span of three) and normalized to maximum. Pitch values presented in Fig 5A–5C were selected as representatives of high-, medium-, and low-pitch stimuli in order to demonstrate how the temporal pitch information changed as a result of pitch increase.
A stacked pooled ISIH graph was generated by accumulating all the 29 pooled ISIHs (e.g., Fig 5A–5C), arranged by the stimulus pitch. The stacked pooled ISIH is shown in Fig 5D, with pitch categories shown along the ordinate in Hz and histogram amplitude represented by color. Similar stacked pooled ISIH are shown for vowels /ɑ/ and /i/ in Fig 5E and 5F, respectively. Fig 5G and 5H show the stacked pooled ISIHs for high-pass filtered vowels /ɑ/ and /i/, respectively.
It was observed that for all stimuli types, as the pitch of stimuli increased, the amplitude and the number of histogram peaks became stronger and fewer, respectively, indicating that the model used shorter inter-spike-intervals (viz., rapidly-occurring spikes) to encode higher pitches. Stacked histograms thus provide a representation of how the model temporally processes the pitch.
To demonstrate the effectiveness of the temporal cues in providing pitch information, a pitch ranking model using the pooled ISIHs as input was simulated. Pitch ranking is a typical psychophysical experiment wherein listeners are asked to decide which of the two presented sound stimuli has a higher pitch. Normal-hearing humans score about 70%-100% depending on the pitch difference in a sound pair and type of stimuli. For example, at one-semitone pitch difference using sustained vowels, subjects scored about 81% [41], which increased to about 100% when the pitch difference was increased to six semitones.
The pitch ranking model in this study consisted of an artificial neural network (a single layer perceptron with two neurons) that received two sets of inputs corresponding to a pair of stimuli and generated two outputs, the higher of which would indicate the higher-pitch stimulus. For 20 trials, the model was trained on 1500 pitch pairs (10% reserved for validation) and tested on 500 unseen pitch pairs. Performance at each trial was computed as the number of correct answers divided by the total number of presentations (i.e., 500). At each trial, pitch pairs were selected randomly from a pool of all eligible combinations of vowel stimuli. For a fair comparison between simulated results and available psychophysical data, only same-type vowel pairs (e.g., /ɑ/-/ɑ/ and /i/-/i/) with pitch differences between one and twelve semitones were allowed in the pool. The weights of the artificial neural network were adjusted using the error back-propagation method [42]. The overall performance of the model was calculated as the average performance over the 20 trials. The exact same simulations were performed using the high-pass filtered vowels. Fig 6 presents the overall performance of the model as a function of pitch difference for the original and high-pass filtered vowels.
Humans are born with some pitch perception abilities [43,44]. For example, through measuring event-related potentials, Leppänen et al. [43] reported that newborns were able to detect pitch changes in sequential tones. However, perceiving the missing-F0 pitch does not happen until 3–4 months of age [45]. He and Trainor [45] concluded that unlike pure tones and complete harmonic complexes that possibly relied only on a peripheral representation of the stimulus, processing the pitch of missing-F0 stimuli required cortical engagement to integrate the information from across the auditory periphery and elicit a single pitch percept. Auditory cortical development is an unsupervised process that happens naturally during early infancy.
In this study, it was observed that a correlation-based, unsupervised, spike-based form of Hebbian learning could explain the development of the neural structure required for recognizing the pitch of simple and complex tones, with or without F0. The emerged neural structure led to precisely-timed responses that were necessary for a reliable population code for pitch. More specifically, the synaptic wrinkles (Fig 3J) constituted a mechanism to compensate for the travelling wave delay that was the main cause of temporal misalignment between the spikes coming from different cochlear positions. Similar compensatory mechanisms (i.e., through developing proportional dendritic delays) were found by Greenwood and Maruyama [7] and Oertel et al. [8] in the cochlear nucleus.
Another interesting finding of this study was that although the emerged synaptic connection patterns followed the spectral power of the signal (i.e., the rate profiles), which varied amongst different stimulus types (Fig 3E–3H), the ISIH pattern extracted from the mixed-stimuli learning (Fig 5D–5H) emerged regardless. In other words, the temporal pattern shown in Fig 5D–5H would appear for any type of sound source, given that the model has experienced sufficient variations of the spectral shapes. It can thus be concluded that the temporal code for pitch could successfully extract invariances (F0) among inputs, although the inputs were spectrally different. The temporal code of pitch, therefore, can explain the pitch constancy phenomenon.
From a computational standpoint, the resemblance between the rate profiles (Fig 2I–2L) and the evolved synaptic weight patterns (Fig 3E–3H) indicated that STDP was mainly driven by the average activity or spiking rate of the pre-synaptic (auditory) neurons. However, applying the learning window enabled this correlation-based learning rule to incorporate temporal precision and generate responses that were indicative of pitch, regardless of the spectral shape of the stimulus. That is, the learning algorithm could successfully compensate for rate-place inconsistencies among different types of stimuli and provide a rate-independent temporal code. The ability of the model to replicate the above-mentioned phenomenon was of special importance because finding a spectrum-independent code for pitch has been considered a substantial step forward in the research field of pitch perception [46]. Neural correlates for “pitch constancy” have been detected in the auditory cortex of primates by Bendor and Wang [1]. They found that the pitch selective neurons would respond to both pure tones and harmonic complexes of the same pitch, even when F0 was eliminated from the latter’s spectrum. Simulated cortical columns in this study, therefore, could be considered as a computational substrate for what Bendor and Wang [1] labelled as pitch neurons.
As demonstrated in Fig 2J and 2K, eliminating the low-frequency content of the vowels led to flat lines in the place code corresponding to low-CF neurons. As Fig 4C suggested, STDP was still able to fine-tune the timing of the spikes, despite the missing fundamental. Compared to the original vowels, high-pass filtering the vowels did, however, lead to a noisier vector strength matrix (Fig 4B vs. 4C), indicating that entraining the spikes to a correct phase became a more challenging task for STDP when the fundamental frequency was not available in the spectrum. This was nevertheless an issue when the spiking neural network learnt mixed-stimuli due to exposure to many more representations and spectral variations. The absence of the fundamental frequency did not impair the performance of the pitch ranking model (Fig 6), confirming that the extracted temporal cues were independent of the fundamental frequency.
The place and the temporal codes of pitch are the roots of current pitch perception models, dividing the modern models into pattern matching- and autocorrelation-themed classes [47], respectively. The former estimates pitch based on a pattern or template, which is normally derived from an auditory model simulating the frequency analysis of the cochlea. Better-known examples of this category are the harmonic sieves model of Cohen [48] and the harmonic templates of Shamma and Klein [5]. The autocorrelation class, however, requires self-similarity measures, such as autocorrelation, to estimate periodicity. Examples of this category include Licklider’s [49] duplex theory-themed models such as the ones developed by Meddis et al. [50] and Patterson et al. [3].
The pitch perception model developed in this study employed elements of both modelling approaches in a more biologically-plausible platform. In fact, the present model followed closely the schematic pitch perception model suggested by Moore [51] that also combined the place and temporal code of pitch to explain how the pitch of complex sounds might be perceived by the auditory system. Similar to the model presented in this study, Moore’s schematic model also consisted of a bank of cochlear filters (similar to Fig 1E), a spike generation process (Poisson neurons in Fig 1F) and a spike analyzer that computed the pooled ISIH. A final decision making step would pick the most prominent interval as an estimate of the stimulus period. The learning phase employed in the current model, additionally provided a description of the development of the neural structure leading to the required ISIHs. The learning phase would also provide a biological analogue to Shamma and Kleins’ detector units [5], as well as eliminating the need for long neural lags in autocorrelational models.
It should be noted that the current model could reproduce the inter-spike interval statistics similar to the actual auditory nerve recorded by Cariani and Delgutte [23] and taken into account in Moore’s [51] model. However, the artificial neural network that made pitch judgments based on the received ISIHs was a simple model to demonstrate the effectiveness of the temporal cues in performing a simple pitch perception task and was not intended to be a biologically-plausible model of higher-order auditory system. In addition, in future work, the STDP learning step would present all pitches to all neurons, with a soft winner-take-all mechanism implemented to achieve competition between the neurons to create the pitch map across them.
The cochlear implant or “Bionic Ear” is one of the most successful neural prosthesis that restores partial hearing in profoundly deaf people by directly stimulating the auditory nerve with controlled electrical current pulses. Many implantees have obtained functional speech perception in favorable conditions similar to their normal hearing peers [52]. However, there are still unresolved issues like tone perception in tonal languages and speech perception in noisy environments [53,54]. Music melody appreciation is also very limited in cochlear implant users [55]. It has been shown that pitch perception in implant hearing is correlated with the users’ abilities in performing the abovementioned tasks [56]. Accordingly, if pitch perception is improved in cochlear implant patients, their auditory performance should also get better.
Similar to normal hearing, pitch information in multi-channel cochlear implant hearing is also carried through place and temporal cues [55,57]. In electrical hearing, place cues for pitch perception are associated with the tonotopically-arranged electrodes. For example, Nelson et al. [58] reported that the pitch elicited by stimulating basal electrodes was generally consistently higher than that of the apically-located electrodes. On the other hand, the rate of stimulation and the frequency of amplitude-modulation of the stimulation pulses have impacts on the perceived pitch that could only be explained by the temporal cues for pitch perception. For example, Tong et al. [59] found that, in a cochlear implant listener, high-rate stimulation (in an isolated electrode) resulted in a high-pitch sensation and vice versa. The modulation frequency has a similar effect on pitch as that of the rate of stimulation [60,61].
Although cochlear implants are able to induce cues for pitch perception similar to those used by normal hearing listeners, the quality of the cues is considerably limited in electrical hearing. For instance, a limited number of electrodes and depth of electrode insertion confine the place cues to a limited frequency range [62,63]. Moreover, the tonotopic order in electrical hearing may be distorted in cochlear implant subjects (e.g., Schatzer et al. [64]), resulting in a poor frequency-to-place mapping. Temporal cues are also restricted to a cap rate of about 300 Hz in cochlear implant hearing. This means that stimulation rates or modulation frequencies above this limit do not induce distinctive pitch percepts [59,65,66].
Due to limited depth of electrode array insertion and implant filters suppressing the low-frequency content of the signal (lowest band-pass filters in cochlear implants have a center frequency of ~125 Hz), cochlear implants are not normally able to convey F0 information through the place code. From this point of view, hearing through a cochlear implant is analogous to hearing through a telephone transmission line. The results of this study showed how normal hearing listeners could perceive the missing-F0 pitch by using the temporal cues. Therefore, it can be inferred that improving the temporal cues in cochlear implant users may compensate for the impaired place cues and eventually lead to a better pitch perception. Application of a pitch perception model using the place code in evaluating the effect of stimulation field spread on pitch perception in cochlear implant hearing can be found in a study by Erfanian Saeedi et al. [67]. Similarly, with a modified auditory periphery (Fig 1E), the model developed in this study can be used to estimate the efficiency of experimental sound processing strategies (e.g., [68,69]) in terms of providing better temporal pitch perception cues. Extending the application of the current model to cochlear implant research would require replacing the normal hearing cochlear filters with descriptors of auditory neuron responses to electrical stimulation, examples of which can be found in [70–74].
|
10.1371/journal.pgen.1006732 | Transcriptional repression of frequency by the IEC-1-INO80 complex is required for normal Neurospora circadian clock function | Rhythmic activation and repression of the frequency (frq) gene are essential for normal function of the Neurospora circadian clock. WHITE COLLAR (WC) complex, the positive element of the Neurospora circadian system, is responsible for stimulation of frq transcription. We report that a C2H2 finger domain-containing protein IEC-1 and its associated chromatin remodeling complex INO80 play important roles in normal Neurospora circadian clock function. In iec-1KO strains, circadian rhythms are abolished, and the frq transcript levels are increased compared to that of the wild-type strain. Similar results are observed in mutant strains of the INO80 subunits. Furthermore, ChIP data show that recruitment of the INO80 complex to the frq promoter is IEC-1-dependent. WC-mediated transcription of frq contributes to the rhythmic binding of the INO80 complex at the frq promoter. As demonstrated by ChIP analysis, the INO80 complex is required for the re-establishment of the dense chromatin environment at the frq promoter. In addition, WC-independent frq transcription is present in ino80 mutants. Altogether, our data indicate that the INO80 complex suppresses frq transcription by re-assembling the suppressive mechanisms at the frq promoter after transcription of frq.
| Circadian clocks organize inner physiology to anticipate changes in the external environment. These clocks are controlled by the oscillation of central clock proteins which form the central oscillator. Transcriptional regulation is a critical step in the regulation of the oscillation of these core proteins. In eukaryotes, chromatin remodeling is a common mechanism to regulate gene transcription by conquering or establishing nucleosomal barriers for the transcription machinery. Here, we showed that a C2H2 finger domain-containing protein IEC-1 and its associated chromatin remodeling complex INO80 are required for transcriptional repression of the core clock gene frq in the Neurospora circadian system. Moreover, the activator WHITE COLLAR (WC) complex is responsible for the transcriptional activation of frq; thus, considering the different timing of the transcriptional activation and suppression of frq, there must be a mechanism that coordinates the two opposite processes. We also demonstrated that the WC-mediated open state of the frq promoter facilitates the binding of INO80 to this region, which prepares for subsequent transcriptional suppression. Collectively, our data provide novel insights into the regulation of the frq gene and the circadian clock.
| From the filamentous fungus Neurospora crassa to animals, circadian oscillation is a conserved mechanism based on an auto-regulatory feedback loop composed of negative and positive elements [1–6]. In Neurospora, the heterodimeric WHITE COLLAR (WC) complex, consisting of WC-1 and WC-2, acts as the positive element that binds to the frq promoter and activates frq transcription [7–13]. The negative elements FRQ and FRQ-interacting RNA helicase (FRH) form the FRQ/FRH complex and mediate the phosphorylation of WCs, which inhibits their WC complex activity and promotes the cytoplasmic localization of the WC complex [12,14–16]. Immediately following synthesis, FRQ is progressively phosphorylated by casein kinases (CKI and CKII) and other kinases throughout the subjective day and evening [12,17–20]. CKI participates in the regulation of both FRQ and the WCC [12]. The phosphorylation of FRQ by CKI increases the degradation rate of FRQ. FRQ protein also acts as a scaffold by bringing CKI to phosphorylate the WCC which leads to its inactivation and repression. Similar to CKI, many of the sites on FRQ protein can also be phosphorylated by CKII, which promotes its degradation [12,21]. While casein kinases regulate the stability of FRQ, the casein kinases are countered by multiple protein phosphatases, including PP1, PP2A and PP4 [22,23], The PP1 dephosphorylates and stabilizes FRQ protein while PP2a and PP4 activities influence frq transcription by dephosphorylating WC-2. Hyperphosphorylated FRQ is degraded through the ubiquitin-proteasome pathway [24]. However, a recent study showed that the cycle ends when FRQ is sufficiently hyperphosphorylated and becomes invisible to the circadian machinery [25]. When the activity of FRQ is not sufficient to suppress the activity of the WCs, frq transcription is reactivated by the WC complex. Recently, epigenetic modifications were reported to regulate frq transcription in Neurospora. SET-1 is required for normal expression of frq [26], and the SET-2 pathway is involved in the suppression of WC-independent frq transcription [27]. Moreover, antisense transcription was shown to inhibit sense expression by mediating chromatin modifications and premature termination of transcription in the frq locus [28].
Recent studies have shown that CLOCK:BMAL1 promote the removal of nucleosomes at its binding sites in mammalian clock genes during transcription activation [29]. Based on these results, nucleosomal barriers at the activator-binding sites should be established during transcriptional repression of the circadian cycle. A still unknown factor(s) may be responsible for the rhythmic incorporation of nucleosomes into chromatin at the activator-binding sites of the clock genes to establish rhythmic transcriptional repression.
Chromatin remodelers are capable of removing, destabilizing, ejecting, and restructuring nucleosomes by using the energy provided by ATP hydrolysis. In Neurospora, two ATP-dependent chromatin-remodeling factors, CLOCKSWITCH (CSW-1) and chromodomain helicase DNA-binding-1 (CHD1), have been reported to regulate frq transcription by modulating nucleosome density at the promoter region [7,30]. Recent studies have shown that SWI/SNF is recruited to the frq promoter by WC-1 to initiate frq transcription [31]. The INO80 complex is a highly conserved chromatin remodeler from yeast to humans [32]. INO80 can mobilize a mononucleosome to the center of a DNA fragment in vitro and remove histone variant H2A.Z from nucleosomes [33–36], which is involved in gene activation in yeast [37,38]. In yeast or mammalian systems, the recruitment of the INO80 complex by yeast Iec1 or mammalian Yin Yang 1 is a key event in gene regulation [39,40]. Moreover, the INO80 complex contributes to chromatin silencing of the boundaries of genes and heterochromatins [41]. However, little is known about the role of the INO80 complex in the suppression of protein-coding genes. Here, we demonstrate that the IEC-1-recruited INO80 complex is required for the Neurospora circadian clock system. WC-mediated transcriptional activation accounts for the rhythmic recruitment of the INO80 complex to the frq promoter; binding of the INO80 complex to the promoter region creates a dense chromatin environment and suppresses frq transcription. Loss of the INO80 complex leads to WC-independent frq transcription due to the accessible chromatin environment at the frq promoter.
To characterize the transcription factors that are involved in the regulation of frq expression, we examined the conidiation rhythms of available knockout transcription factor mutants with race tube assays. As shown in Fig 1A, the strain with deletion of the iec-1 (ras-1+) (NCU03206) gene showed arrhythmic conidiation in a race tube assay compared to the wild-type strain, suggesting that IEC-1 plays a critical role in Neurospora circadian conidiation rhythm. The iec-1 gene codes a C2H2 finger domain-containing protein IEC-1, a homolog of mammalian Yin Yang 1 (YY1), Drosophila PcG PHO, and yeast Iec1 [39]. When the IEC-1 protein sequence was used in a BLAST search against protein databases, its homologs were found to be highly conserved in ascomycetes (Fig 1B). To confirm the phenotype of the iec-1 mutant, the iec-1 gene, which is driven by the quinic acid (QA)-inducible qa-2 promoter, was reintroduced to the iec-1KO strain. In addition, five copies of the c-Myc epitope with six histidine residues [42] were inserted at the N-terminus of the IEC-1 ORF to facilitate the detection of the expression of Myc-IEC-1 using a c-Myc monoclonal antibody (9E10). In QA-containing race tubes, the circadian conidiation rhythms of the iec-1KO, qa-Myc-IEC-1 strains were similar to those of the wild-type strains (Fig 1C). These data indicate that Myc-IEC-1 can partially complement the function of the endogenous IEC-1 protein in the iec-1KO strain. To test whether light exposure could entrain the conidiation rhythms, the wild-type strain and iec-1KO strains were assayed under light/dark cycles (LD) at 25°C. The race tube assays showed that the conidiation process of the iec-1KO strains was still entrained by LD cycles (S1A Fig). Meanwhile, a reduced light induction of FRQ expression by light was observed in the iec-1 mutant (S1B Fig). Notably, the basal level of FRQ is higher in the mutant. Nonetheless, this result suggests that IEC-1 also affect the light induction of FRQ expression. These data suggested that the clock phenotype of the mutant strain is not caused by a defect in the input pathway of clock. To further confirm the arrhythmic phenotype of the iec-1KO strain, we introduced bioluminescence reporter constructs (frq-luc) into the his-3 locus of the iec-1KO strains and the wild-type strains, in which the luciferase expression of both strains are driven by the frq promoter [43]. The sequence analysis of genomic DNA showed that the frq promoters in the luciferase-expressing strains were intact (S2 Fig). Consistent with the phenotype in the race tube assay, the robust bioluminescence rhythm of the wt, frq-luc strain was abolished in the iec-1KO, frq-luc strains (Fig 1D and S3A Fig), indicating that IEC-1 is critical for circadian clock function at the molecular level.
To further determine the role of IEC-1 in the circadian clock, we examined the FRQ expression profile at different time points in constant darkness (DD). Both FRQ rhythms and FRQ phosphorylation profiles were disrupted in the iec-1KO strain (Fig 2A). Northern blot analyses showed that the abolished FRQ rhythms were due to constant transcription of the frq gene in the iec-1KO strain (Fig 2B). To test whether IEC-1 directly regulates frq transcription, a chromatin immunoprecipitation (ChIP) assay was performed by using an IEC-1-specific polyclonal antibody (Fig 2C). Because the IEC-1 antibody recognizes nonspecific bands, the iec-1KO strain was used as the negative control in the ChIP assays. The ChIP assay revealed enrichment of IEC-1 at the frq promoter, ORF 3’, and the 3’UTR but not at ORF 5’ and the middle regions at DD18 (Fig 2D). Furthermore, the association of IEC-1 with C-box fluctuated from DD10 to DD42 (Fig 2E). Altogether, these results indicate that IEC-1 functions at the frq locus to regulate frq transcription.
Previous studies showed that yeast Iec1or mammalian Yin Yang 1 (YY1) co-purified with the INO80 complex in yeast or mammalian cells [39,44]. These transcription factors are required for the recruitment of the INO80 complex to target genes. Since the INO80 complex acts as a conserved co-factor of Iec1/YY1 in different organisms, we performed an IP assay to test the association of the INO80 complex with IEC-1 in Neurospora. The results showed that FLAG-IEC-1 was immunoprecipitated (IP) with INO80 (NCU08919) by the INO80 antiserum but not immunoprecipitated with the preimmune (PI) serum (S4A Fig). In the INO80 complex, INO80 is the catalytic subunit and IES-1 is a structural subunit [33,37,45]. To further identify whether the INO80 complex functions in the same pathway as IEC-1 in the regulation of the Neurospora circadian clock, we generated ino80 (NCU08919) and ies-1 (NCU01362) deletion mutants. As expected, both ino80KO and ies-1KO strains exhibited arrhythmic conidiation phenotypes in the race tube assays (S4B Fig). We also tried to generate mutants with a band background (ras-1bd), but we only obtained heterokaryotic strains. These strains also showed arrhythmic conidiation phenotypes in the race tube assays (S4 Fig). To further confirm that deletion of ino80 or ies-1 is the only cause of the clock defect in these mutants, rescue strains of the mutants were generated. The conidiation rhythms of the ino80KO, qa-Myc-INO80 and ies-1KO, qa-Myc-IES-1 transformants were not restored in the race tubes in the absence of QA (S4C and S4D Fig). Meanwhile, obvious conidiation rhythms were observed in the rescue strains but not in the knock-out strains in the race tube assays with 10−3 M QA (S4C and S4D Fig). However, we also observed a phase difference between the wild-type and the rescue strains. The results suggest that qa-2 promoter-driven expression of Myc-INO80 could not fully complement the function of endogenous INO80 due to an abnormal expression level or the presence of the Myc tag. The race tube assays showed that the conidiation processes of the ino80KO and ies-1KO strains are entrained by LD cycles, excluding the defect in the input pathway of clock in these mutants (S4E Fig). Notably, the growth rates of ino80KO strains were strongly dependent on glucose, indicating that ino80KO strain might be sensitive to glucose or carbon starvation. Bioluminescence rhythms in ino80KO, frq-luc and ies-1KO, frq-luc strains were also abolished (S4F, S3B and S3C Figs). Western blots showed disruption of both the FRQ rhythms and FRQ phosphorylation profiles in the ino80KO and ies-1KO strains (S4G Fig). Northern blot analyses showed that the abolished FRQ rhythms were due to constant transcription of the frq gene in the ino80KO and ies-1KO strains (S4H Fig). Altogether, these results indicate that the INO80 complex is critical for frq transcription repression.
Previous studies have shown that the INO80 complex can be recruited to target genes by Iec1 or YY1 in yeast or mammalian cells [39,40]. To investigate the occupancy of the INO80 complex at the C-box of the frq gene, we performed a ChIP assay using INO80-specific antiserum at DD18 (Fig 3A). The ChIP results revealed that binding of INO80 to the frq gene peaks at the C-box and 3’-UTR (Fig 3B), similar to the binding pattern of IEC-1 (Fig 2D). Next, we assessed whether IEC-1 is required for recruitment of the INO80 complex at the C-box of the frq promoter. The ChIP assays conducted with an INO80 antibody showed a dramatic decrease in enrichment of INO80 at the C-box in iec-1KO strains (Fig 3C), indicating that efficient binding of the INO80 complex at the frq C-box is dependent on the expression of IEC-1.
To determine whether the association of INO80 with the C-box is rhythmic during circadian cycles, the recruitment of INO80 at four different time points in constant darkness were examined using a ChIP assay. The results demonstrated that the association of INO80 with the C-box of the frq promoter was rhythmic, peaking at DD18 (Fig 3C). In yeast or mammalian systems, recruitment of the INO80 complex by Iec1 or YY1 is a key event in gene expression [39,40]. Thus, these results suggest that IEC-1-dependent INO80 recruitment at the frq promoter is required for suppression of frq transcription after WCC-inactivation by FRQ. In the Neurospora circadian system, the transcriptional activators WC-1 and WC-2 are responsible for the transcriptional activation of the frq gene during circadian cycles [7,9–11,13,46,47]. In the frq9 mutant, constant frq transcription [48] causes a low H3 density at the frq promoter, whereas, in the wc-2KO strain, a lack of frq transcription [10] is associated with a high H3 density at the frq promoter (Fig 3D). These findings indicate that WC-mediated transcriptional activation leads to open chromatin states by nucleosomal removal, which increases the accessibility of the frq promoter. It is also possible that WC-mediated transcriptional activation of frq is required for rhythmic recruitment of the INO80 complex during the period of high expression of frq to prepare for repression of frq transcription. To assess this possibility, we examined enrichment of the INO80 complex at the frq C-box in the wc-2KO and frq9 strains. The ChIP results showed that the levels of INO80 enrichment in the wc-2 mutant were lower than that in the wild-type strain at the DD18 time point (Fig 3E). These data suggest that the rhythmic recruitment of the INO80 complex is dependent on WC-mediated transcriptional activation of frq. In contrast, in frq9 strains, the constant elevation of the levels of INO80 enrichment are due to the constant high elevation of frq expression, which is caused by malfunction of the negative feedback loop (Fig 3E). To verify that the high levels of recruitment of the INO80 complex in the frq9 mutant are due to the activity of the WC complex, we generated the wc-2KO frq9 double-mutant strain, and examined the recruitment of INO80 at the C-box. Loss of WC-2 restored the low levels of recruitment of INO80 at the C-box in the frq9 mutant (Fig 3E). Meanwhile, the INO80 expression of these mutants was similar to that of the wild-type strain (Fig 3F). These data indicate that the highly activated transcription of frq by the WC complex also contributes to the rhythmic recruitment of the INO80 complex to the C-box of frq for subsequent suppression of frq transcription during circadian cycles.
In the Neurospora circadian system, the WC complex, the positive transcription factor that binds to the C-box of frq, is responsible for rhythmic frq expression [7–9,13]. To test whether frq transcription is driven by the WC complex in ino80 and iec-1 mutants, we first examined the transcriptional activity of WC-2 in the ino80KO and iec-1KO strains. The ChIP results revealed that the rhythmic enrichment of WC-2 at the frq C-box corresponded to the rhythmic expression of frq in the wild type strain (Fig 4A). In contrast, the recruitment of WC-2 was dramatically decreased in the ino80KO and iec-1KO strains (Fig 4A), which was inconsistent with the constant expression of frq in these mutants. This finding suggests that the transcriptional activities of WC-1 and WC-2 are decreased in the ino80 and iec-1 mutants. Previous studies showed that hypophosphorylated WC-1 and WC-2 efficiently bind to the C-box for frq transcriptional activation [12,21] while hyperphosphorylated WC-1 and WC-2 exhibit lower binding activity at the C-box of frq [14,49]. Western blot analysis showed that the phosphorylation levels of WC-1 and WC-2 were increased in the ino80KO, ies-1KO and iec-1KO strains compared to those in the wild-type strains at different time points in DD (Fig 4B and 4D). In addition, we found no significant changes in the protein levels of WC-1 and a slight decrease in the levels of WC-2 in the ino80KO, ies-1KO and iec-1KO strains compared to those in the wild-type strains (Fig 4C and 4E). The levels of WC-1 protein did not exhibit a robust circadian rhythm in our hands, which is consistent with previous studies showing that the amplitudes of WC rhythms are variable. Since WC activity is mostly known to be regulated by phosphorylation, the variability of the WC-1 rhythms might be due to the use of different WC-1 antibodies in different laboratories which may have different sensitivity to different isoforms of WC-1. These results indicate the increased frq transcripts in these mutants are not caused by the increase of transcriptional activity of WCC. These data also suggest that the high levels of frq transcription could be partially independent of WC expression in these mutants. To further confirm this possibility, we generated an ies-1KO wc-1RIP double mutant and compared its FRQ levels to that of a wc-1 single mutant. Consistent with previous results, the FRQ levels were extremely low in the wc-1 single mutant (Fig 4F) [10]. However, the levels of FRQ protein and frq mRNA in the ies-1KO wc-1RIP double mutant were well detected with Western blot or Northern blot analyses (Fig 4F and 4G), indicating that WC-independent frq transcription exists in the ies-1 mutants. Altogether, these results demonstrate that recruitment of the INO80 complex at the C-box through high transcriptional activation prepares for suppression of frq transcription after WCC inactivation.
In S. cerevisiae, the first four nucleosomes at the transcription start site of genes have critical roles in the process of transcription initiation [50]. In Drosophila, the +1 nucleosome strongly inhibits the normal function of RNA polymerase II [51]. Although the INO80 complex occupied and peaked at the boundary of the genes, genome-wide ultra-high-resolution ChIP-exo data showed that the Arp5 subunit of the INO80 complex was particularly enriched at the +1 position in yeast [36,52]. Antisense transcripts qrf in the Neurospora circadian system revealed that induction of qrf promotes frq gene expression by creating a more accessible local chromatin environment, even in the absence of the WC complex [28]. Given that qrf-induced frq transcripts share the same properties with the WC-independent frq transcripts, the chromatin states in these two pathways should be similar. The ChIP assay showed that the H3 levels were dramatically reduced at both the C-box and TSS in the ino80KO and ies-1KO strains compared to that of the wild-type strains (Fig 5A). However, the H3 levels at the frq ORF in mutants were similar to that of the wild-type strains (Fig 5A). These results indicate that nucleosome density is decreased at the C-box and TSS of the frq gene in mutants.
SET-2-mediated H3K36me3 is an important landmark on chromatin during transcription elongation [53]. To confirm the enhanced transcription of frq induced by the decreased nucleosomal barrier in the ino80KO strains, a ChIP assay was performed to examine the recruitment of SET-2 and enrichment of H3K36me3 at the 3’ORF of the frq gene. As expected, the recruitment of SET-2 and the enrichment of H3K36me3 were significantly increased at the 3’ ORF of the frq gene in the ino80KO strains compared to those of the wild-type strains (Fig 5B). Together, these results demonstrate that the INO80 complex contributes to establishment of the dense chromatin environment at the frq promoter, which is essential for suppressing WC-independent frq transcription.
In this study, we identified a C2H2 finger domain-containing protein IEC-1 and its co-factor, the INO80 complex, which as part of their normal cellular roles in chromatin assembly, facilitate generalized repression of the clock gene frq in Neurospora. To investigate the role of IEC-1 and the INO80 complex, we generated the iec-1KO, ino80KO and ies-1KO strains and discovered that IEC-1 and the INO80 complex are required for normal circadian clock function. Based on the data presented in this manuscript, the INO80 complex is rhythmically recruited by IEC-1 to the frq promoter to suppress frq expression in the Neurospora clock. The recruitment of the INO80 complex to the frq locus by the transcription factor, IEC-1, is important for frq transcriptional repression, since the disruption of INO80 binding to the frq promoter in the iec-1KO, ino80KO and ies-1KO strains leads to high FRQ levels and loss of frq rhythmicity (Fig 2 and S4 Fig). A similar situation was observed in deletion of the transcriptional co-repressor RCO-1-RCM-1 [54,55]. Rhythmic activation and repression of frq transcription are required for the function of the Neurospora circadian clock. Therefore, normal suppression of frq expression is essential for the circadian auto-regulatory feedback loop function in the Neurospora clock. Nucleosomes at the TSS were identified as the strongest barriers for RNA Polymerase II in S. cerevisiae and Drosophila [50,51]. The FRQ expression levels in the iec-1KO, ino80KO and ies-1KO strains were high and arrhythmic (Fig 2 and S4 Fig), suggesting that the recruitment of the INO80 complex is a key step for establishing repressed state of chromatin at the frq promoter. Our ChIP data revealed that deletion of INO80 or IES-1 results in decreased nucleosome density at the frq promoter. These results suggest that the INO80 complex is required for establishing compact chromatin environments at the frq promoter.
We tried several times but failed to detect tight interaction between the WCC and the IEC-1-INO80 complex. However, weak or transient interaction between them might still occur. Our results showed that the rhythmic binding of the INO80 complex is also dependent on frq transcription activation by the WC complex in the wild-type strain (Fig 3E). Due to the constant transcriptional activation of frq, a significant decrease in nucleosome density along with increased recruitment of the INO80 complex at the frq promoter was observed in the frq9 mutant strains (Fig 3D and 3E). In contrast, the nucleosome density at the frq promoter was increased dramatically in the wc-2KO strains because of the inactivation of frq transcription and reduced INO80 binding (Fig 3D and 3E). These results indicate that binding of the WC complex to the frq promoter results in an open chromatin state in this region and recruitment of the INO80 complex by IEC-1. The progressive inactivation of the WC complex by accumulated FRQ stimulates the remodeling activity of the INO80 complex associated with the frq promoter, which increases the nucleosome density at the frq promoter (Fig 5A). In contrast, although high levels of INO80 recruitment at the frq promoter were observed in the frq9 mutant, the INO80 complex could not re-assemble the dense chromatin state at the frq promoter in this mutant due to the absence of the FRQ protein, which is needed to shut down the high levels of frq transcription driven by the WC complex. In the wild type strain, the lowest level of INO80 recruitment was found at DD22. At this time point the frq transcripts decline and the frq promoter is condensed which is in a line with the data above. These data suggest that at DD22, the chromatin structure mediated by IEC-1 and INO80 complex to suppress frq transcription has already been established which is unsuitable for the binding of INO80 complex. Current results indicate that the FRQ-FRH complex functions as the negative element in the Neurospora circadian auto-regulatory feedback loop to inactivate the WC complex [12,14–16] and activate INO80 to achieve complete repression of frq expression. Thus, the open chromatin state of WCC-driven frq transcription promotes the recruitment of the INO80 complex, which closes the negative feedback loop upon WCC inactivation by FRQ. Similar to the Neurospora clock, the rhythmic activation of the clock genes in Drosophila and mammals by a heterodimeric PAS domain containing the transcription factors, CLK:CYC or the CLOCK:BMAL1 complex is essential for circadian clock function. In the mammalian clock, the binding of CLOCK:BMAL1 to the E-box of clock genes promotes nucleosome eviction and incorporation of the histone variant H2A.Z [29], which suggests that activation of the clock genes by CLOCK:BMAL1 leads to changes in the chromatin structures. Whether the open chromatin states of animal clock genes activated by CLK:CYC or CLOCK:BMAL1 can promote the binding of a remodeler to promoters of clock genes, similar to our results in Neurospora, is not clear. Considering the conserved roles of IEC1/PHO/YY1 and the INO80 complex in different organisms, it is worth determining whether there are similar mechanisms in Drosophila and mammals. Altogether, these results suggest that a conserved repression mechanism involving chromatin regulation exists in eukaryotic circadian systems.
The wild-type strain (4200) was used as a control. The iec-1 or ies-1 genes were deleted by the replacement of their ORFs with a hygromycin resistance gene (hph) on the ku70RIP (bd, a) background strain. The ku70RIP iec-1KO and ku70RIP ies-1KO strains were crossed with the 774–10 (A, his-3-) strain to obtain homokaryotic iec-1KO and ies-1KO strains, respectively. The ku70::bar ino80KO strain was generated by the replacement of its ORF with the hygromycin resistance gene (hph) on the ku70::bar background strain and microconidia purification. The ku70::bar ras-1bd ino80KO and ras-1bd ino80KO strains were generated in the same manner as the ku70::bar ino80KO strain. For rescue strains, the plasmid qa-5Myc-6his-IEC-1 was transformed into the iec-1KO strain. The ino80KO, qa-5Myc-6his-INO80 and ies-1KO, qa-5Myc-6his-IES-1 transformants were obtained in the same manner as the iec-1KO, qa-5Myc-6his-IEC-1 transformants. A plasmid containing the full-length frq promoter fused to luciferase was transformed into the iec-1KO, ino80KO and ies-1KO strains to generate the iec-1KO, frq-luc, ino80KO, frq-luc and ies-1KO, frq-luc strains. Liquid culture conditions were the same as a previously published method [56]. The wc-2KO, frq9, wc-2KO frq9, wc-1RIP and ies-1KO wc-1RIP all contain the band mutation.
The race tube medium contained 1× Vogel’s salts, 0.1% glucose, 0.17% arginine, 50 ng/mL biotin and 1.5% agar supplemented with or without 10 mM H2O2. Conidia of different strains were inoculated at one end of each race tube and were grown under constant light (LL) for 1 day to synchronize the clock. The race tubes were then transferred to constant darkness (DD), and the position of the advancing mycelia front was marked at 24 h intervals on the tube. When growth was completed, tubes were scanned, and the growth period of each strain was calculated.
GST-INO80 (containing INO80 amino acids 1–341) and GST-IEC-1 (containing IEC-1 amino acids 9–175) fusion proteins were expressed in BL21 cells by induction of IPTG. After purification, the recombinant proteins were used as antigens to immunize rabbits, which yielded rabbit polyclonal antiserums [57].
ChIP assay was performed as previously described [55]. Neurospora tissues were fixed by shaking in 1% formaldehyde for 15 min at 25°C, and cross-linking reactions were stopped by adding glycine at a final concentration of 125 mM. The cross-linked chromatin was sheared by sonication to approximately 200–500 bp fragments. A 1 mL aliquot of protein (2 mg/mL) was used per immunoprecipitation, and 10 μL was maintained as the input DNA. The chromatin immunoprecipitation reaction was carried out with 2 μL antibody to WC-2, 2.5 μL antibody to H3 (2650; Cell Signaling Technology), 2.5 μL antibody to IEC-1, 5 μL antibody to INO80, 5 μL antibody to SET-2, and 2 μL antibody to H3K36me3 (4909; Cell Signaling Technology). Immunoprecipitated DNA was quantified using real-time PCR. The primers for real-time PCR were designed according to a previously published protocol [27]. The ChIP-qPCR data were normalized by the input DNA, and the results were presented as the percentage of input DNA. Each experiment was independently performed at least three times.
Protein extraction, quantification and western blot analysis were performed as previously described [58,59]. Western blot analyses were performed by using antibodies against the proteins of interest. Equal amounts of total protein (40 μg) were loaded in each lane. After electrophoresis, proteins were transferred onto PVDF membranes, and western blot analysis was performed.
RNA was extracted by TRIzol [60] and analyzed by northern blotting as previously reported [56]. Shortly, equal amounts of total RNA (20 μg) were loaded onto agarose gels for electrophoresis, and the gels were blotted and probed with an RNA probe specific for frq mRNA.
The luciferase reporter assay was performed as previously reported [54]. The bioluminescence reporter construct (frq-luc), in which luciferase expression is driven by the frq promoter, was introduced into the his-3 locus of the iec-1KO, iec-1KO, ino80KO and wild-type strains. One drop of conidia suspensions in water was placed on AFV medium and grown in constant light (LL) overnight at 25°C. The cultures were then transferred to constant darkness, and luminescence was recorded in real time using LumiCycle after one day in DD at 25°C. The data were then normalized with LumiCycle analysis software by subtracting the baseline luciferase signal, which increases as the cells grow. Under our experimental condition, luciferase signals are highly variable during the first day in the LumiCycle but become stabilized afterwards, which is likely due to the light-dark transfer of the cultures. Thus, the results were recorded after one day in DD at 25°C.
|
10.1371/journal.pcbi.1006317 | Exploration and stabilization of Ras1 mating zone: A mechanism with positive and negative feedbacks | In mating fission yeast cells, sensing and response to extracellular pheromone concentrations occurs through an exploratory Cdc42 patch that stochastically samples the cell cortex before stabilizing towards a mating partner. Active Ras1 (Ras1-GTP), an upstream regulator of Cdc42, and Gap1, the GTPase-activating protein for Ras1, localize at the patch. We developed a reaction-diffusion model of Ras1 patch appearance and disappearance with a positive feedback by a Guanine nucleotide Exchange Factor (GEF) and Gap1 inhibition. The model is based on new estimates of Ras1-GDP, Ras1-GTP and Gap1 diffusion coefficients and rates of cytoplasmic exchange studied by FRAP. The model reproduces exploratory patch behavior and lack of Ras1 patch in cells lacking Gap1. Transition to a stable patch can occur by change of Gap1 rates constants or local increase of the positive feedback rate constants. The model predicts that the patch size and number of patches depend on the strength of positive and negative feedbacks. Measurements of Ras1 patch size and number in cells overexpressing the Ras1 GEF or Gap1 are consistent with the model.
| Unicellular fission yeasts mate by fusing with partners of the opposite mating type. Each pair member grows towards its selected partner that signals its presence through secreted pheromone. The process of partner selection occurs through an exploratory patch (containing activated signaling protein Cdc42 and upstream regulator Ras1) that assembles and disassembles on the cell cortex, stabilizing in regions of higher opposite pheromone concentration. We present a computational model of the molecular mechanisms driving the dynamical pattern of patch exploration and stabilization. The model is based on reaction and diffusion along the curved cell membrane, with diffusion coefficients measured experimentally. In the model, a positive Ras1 activation feedback loop generates a patch containing most of the activating protein (Ras1 GEF). The fast diffusing inhibitor Gap1 that is recruited locally from the cytoplasm spreads on the cell membrane, limiting patch size and causing its decay. Spontaneous reinitiation of Ras1 activation elsewhere on the cortex provides a mechanism for exploration. Transition of the system’s behavior to that of a single stable patch is possible upon simulated pheromone sensing. The computational model provides predictions for the number of patches and patch size dependence on parameters that we tested experimentally.
| How cells sense a chemical concentration gradient and polarize towards regions of high or low concentrations is a fundamental biological question. Encoding of spatial information in chemoattractant gradients is essential for many processes, including the directional migration of leukocytes and neutrophils or the directional growth of yeast and neurons [1–4]. Cellular response of eukaryotic cells in response to extracellular cues generally involves establishing a cellular axis of polarity through accumulation of small GTPases, such as Ras and Cdc42, in cortical domains [5]. The establishment of such a self-organizing polarity zone is a process of pattern formation on the cell cortex.
As model eukaryotic cells, budding and fission yeasts have been used to study basic properties of chemical sensing mechanisms [6–8]. Under mating conditions, these organisms respond to the pheromone gradient secreted by the opposite mating type cells, in order to identify the closest potential partner, grow towards it by formation of a shmoo extension, and fuse with it [7]. Recent studies have shown that this process involves an initial assembly of a “patch” or “zone” that is established independently of the direction of the pheromone gradient (Fig 1A and 1B) [9–11]. Over time, the patch that contains activated Cdc42 and co-factors reorients and stabilizes towards the chosen pheromone-secreting partner and starts growing toward it via Cdc42-dependent recruitment of the cell growth machinery. The partner search strategy is different between the two yeasts. The budding yeast patch executes a biased diffusion along the cell cortex towards high pheromone concentration regions [10, 12, 13]. By contrast, fission yeast implements an exploratory search involving complete patch disappearance and re-appearance at a different cortical location, followed by patch stabilization (Fig 1B) [9]. It has been proposed that the fission yeast strategy helps the whole cell population explore space of possible mating configurations, thus optimizing population mating efficiency [11]. Furthermore, the localization of pheromone transporter and components of the signal transduction cascade suggest that the spatial gradient information is produced and sensed locally from the polarity zone [11].
Models of the mechanism of yeast pheromone gradient sensing have to explain both the ability of cells to organize a patch and the change of its location. Many theoretical studies have considered the process of Cdc42 patch formation for bud growth during the mitotic cell cycle. In budding yeast cells lacking Ras-like Rsr1, the bud forms randomly along the mother cell cortex through a process of spherical symmetry breaking [14]. Theoretical studies have suggested that a positive feedback in Cdc42 activation gives rise to a Turing-type pattern formation mechanism which breaks symmetry and establishes a single patch through amplification of a stochastic fluctuation at a random location [15, 16]. However, the relative contributions of Cdc42 activity regulators and actin to the positive feedback have been debated [12, 17–20]. To explain observed oscillations between competing patches in initial stages of bud site selection required implementing an additional negative feedback mechanism [21]. During budding yeast mating, experiments and mathematical models have suggested two different mechanisms for patch diffusion toward the pheromone gradient. McClure et al. suggested that perturbations caused by the localized delivery of vesicles near the patch site dilute polarity factors, an effect counteracted by polarized G proteins [13]. Hegemann et al. proposed that a secondary positive feedback between the actin-based receptor trafficking and Cdc42 allows for patch diffusion toward the gradient that stabilizes in the high concentrations of Cdc42-GTP [10].
Our aim in this work is to propose and test a model implementing a mechanism for fission yeast mating patch formation, exploration, and stabilization, a process that has not been explored theoretically. This study is motivated by recent findings emphasizing the role of Ras1, the only Ras GTPase homolog in fission yeast. Ras1 is an upstream regulator for Cdc42 and its activity is critical for both mating and polarity establishment during interphase, since ras1Δ cells are sterile and round [22]. During early stages of mating, active Ras1 (Ras1-GTP) localizes at the same exploratory sites as active Cdc42 [23]. The appearance and disappearance of the patch is an evidence for regulation through positive and negative feedback loops. Indeed, in S. pombe, Ras1 activation in mating conditions is promoted by the pheromone-induced Guanine nucleotide Exchange Factor (GEF) Ste6 [24]. It has been shown that Ste6 pheromone-induced expression is itself regulated by Ras1 activity, which in turn creates an additional positive feedback loop [25, 26]. Ras1 inactivation is promoted through interaction with its GTPase activating protein (GAP), Gap1. The deletion of Gap1 abolishes the Ras1 polarity patch [23], increases the lifetime of the Cdc42 patch during exploration and results in reduction of mating efficiency [11]. It was found that Gap1 is recruited to the active Ras1 patch, occupying a larger area on the membrane where it remains for a longer time compared to Ras1-GTP during patch disappearance [11] (Fig 1C). These observations are consistent with Gap1-dependent hydrolysis of Ras1-GTP forming a local negative feedback to temporally control the polarity patch.
In this study we developed a 3D reaction-diffusion model, which accounts for the fission yeast cell geometry, to study the dynamic regulation of Ras1 during fission yeast mating. To develop this model, we performed experiments to measure the kinetics of diffusion and membrane dissociation of active Ras1 (Ras1-GTP), inactive Ras1 (Ras1-GDP), and Gap1. Since many of the biochemical interactions of the Ras1 positive and negative feedbacks are not yet known we adopt a phenomenological approach, inspired by prior studies in budding yeast. We illustrate that by tuning the rate constants for the positive and negative feedback loops, the model can reproduce the appearance and disappearance of Ras1 patch at random locations. We investigate the switch to patch stabilization upon sensing of a high level of pheromone and show that the patch in our model can be stabilized at positions with a higher positive feedback rate constant, which might represent regions of higher pheromone concentration. The model predicts that the patch size can be regulated by positive and negative feedback. In simulations, increase in negative feedback resulted in narrower patches and increase in positive feedback resulted in formation of multiple patches. These results were tested and confirmed experimentally.
We performed FRAP experiments of cells under different conditions to narrow down the possible transport mechanisms of Ras1 and regulators to the active patch. GFP-Ras1, which shows Ras1 localization independent of nucleotide binding, localizes along the whole cell cortex during the exploration phase [23]. Its intensity fluctuates around the cortex as its active form concentrates at the exploring patch, complicating FRAP analysis [23]. During interphase, GFP-Ras1 localizes all around the cell cortex but its active form is concentrated at the cell tips (Figs 1D and S1A) [23]. Thus, to monitor the dynamics of Ras1-GDP at the cortex, we bleached the sides of cells expressing GFP-Ras1 during vegetative growth, assuming that Ras1-GDP dynamics are similar during interphase and mating. The recovery of intensity was slower for wider bleached areas, consistent with diffusion-dominated recovery (Fig 2A). We fitted the FRAP data with a 3D model of membrane diffusion and cytoplasmic exchange that accounts for the geometrical features of the system, assuming that the concentration in the cytoplasm is uniform and constant (Fig 2A and 2B, Materials and Methods). A range of diffusion coefficients and exchange rates give curves that lie within the standard deviation of the experimental measurements (Fig 2E). To select from those values, we further fitted the recovery of the concentration profiles (Figs 2F and S1B and S1C), which favored the largest range of diffusion coefficients, with D = 0.15 μm2/s and exchange rates < 0.005 s-1.
To estimate the diffusion coefficient of Ras1-GTP, we performed FRAP of cells expressing GFP-Ras1Q66L carrying a GTP-locked Ras1 allele, which decorates the whole cortex of interphase cells, Figs 2C and S1A. We repeated the same analysis procedure as for Ras1-GDP. We found the range of values of diffusion coefficient and exchange rates that provide fits for the total recovery in the bleached region versus time, within the experimental standard deviation (Fig 2D and 2E). To further narrow down the range of values in Fig 2E, we also fitted the individual recovery profiles (S1D and S1E Fig), which indicated a slower Ras1-GTP diffusion, D = 0.04 μm2/s and faster membrane dissociation rate, 0.02 s-1 as compared to Ras1-GDP. Cooperative interactions amongst activated Ras1 may further slow diffusion or dissociation in regions of high Ras1-GTP concentration, however these effects are harder to measure and, for simplicity, we neglect these effects in the model below.
Gap1 is recruited to the sites of Ras1 activation [23] and localizes to the zone of cell-cell contact during mating and to the tips of interphase cells, where Ras1 is active [23]. To measure Gap1 diffusion, we performed two types of FRAP experiments. In the first series, we bleached all of Gap1-GFP at the contact zone of pre-fusion paired cells expressing Gap1-GFP and Myo52-tdTomato (Fig 3A and 3B). In half of a total of 6 cells, the recovery of the middle of the fusion focus was faster than the side regions (the other half cells showed either small or no detectable difference) (Fig 3C). This behavior is consistent with a mechanism of recruitment of Gap1 at the patch center, followed by lateral diffusion to the cell sides. In the second series, we bleached half of the cell tip of vegetatively-growing cells expressing Gap1-GFP to monitor the diffusion process at the boundary between bleached and unbleached regions (S2A and S2B Fig). Using D = 0.2 μm2/s and dissociation rate 0.02 s-1 provides good fits for Gap1-GFP FRAP at the fusion site in mating cells (Figs 3C and S2E) as well as the smoothening out of the sharp intensity gradient at the boundary between the bleached and unbleached regions at the cell tip of the vegetatively-growing cells (S2C and S2D Fig). Thus the diffusion rate of Gap1 is similar to that of Ras1-GDP, to which it may remain bound after hydrolysis. These results support a Ras1-GTP focalization mechanism based on slow local activation coupled to a fast diffusing inhibitor (Gap1) and motivate a reaction-diffusion model we describe below.
We developed a model based on a system of reaction-diffusion equations to study regulation of Ras1-GTP exploratory behavior and stabilization during fission yeast mating (Fig 4A). The model accounted for the diffusion and reaction of Ras1-GTP, Ras1-GDP and Gap1 along a 3D curved surface with surface densities CRT,CRD, and CGAP that vary along the cell surface (see Material and Methods). These three components have diffusion coefficients DRT,DRD and DGAP, and cytoplasmic exchange rates, rRT,rRD and rGAP, which we have estimated from FRAP experiments as discussed previously. We assume that the cytoplasmic concentrations of Ras1-GDP and Gap1 are uniform in space and constant over time, but that there is a finite pool of GEF that provides an upper Ras1 activation limit [16]. Prior modeling work has shown how the combination of positive and negative feedbacks can lead to patch formation and patch oscillations of Cdc42-GTP [16, 21, 27]. We thus examined if similar mechanisms can underlie Ras1-GTP dynamics.
In the model, activation of Ras1 by GEFs is assumed to occur through an autocatalytic mechanism that has a functional form similar to the positive feedback in a winner-take-all mechanism of S. cerevisiae polarization [16]. This is motivated by the experimental observations of Ste6 (Ras1 GEF) colocalization with Ras1-GTP at the fusion focus [23] and the previously shown positive feedback between Ras1 and its own activator (Ste6) [25]. In this positive feedback mechanism, a finite amount of GEF in the system is assumed to be distributed with higher proportions at sites with higher active Ras1 concentration, i.e. we assumed that the density of Ras1 GEFs, CGEF, is determined by CRT.
A negative feedback is required to enable patch disappearance. We assume that Gap1 is recruited to the membrane through Ras1-GTP [28], where it hydrolyzes Ras1-GTP and diffuses laterally before dissociation. The Gap1 recruitment rate depends nonlinearly on Ras1-GTP concentration to generate oscillations, as in the GAP mechanism implemented to reproduce Cdc42 oscillations in S. cerevisiae [21]. This nonlinearity is supported by the experimental evidence that Gap1 full localization to sites of Ras1-GTP needs more than its GAP domain [23]. All membrane-bound Gap1 is assumed to be able to hydrolyze Ras1-GTP but we note that this is an approximation since one possibility is that Gap1 remains bound to Ras1-GDP, and thus inactive, after GTP hydrolysis.
By varying the unknown rate constants of activation and inactivation and using other parameter values from Table 1, we found that the model can reproduce exploratory behavior through the formation and disappearance of an active Ras1 zone (Fig 4B and S1 Movie). A zone initially forms when random fluctuations (implemented as stochastic noise) trigger the positive feedback to take off at a random location along the cortex. The Gap1 that is simultaneously recruited spreads by diffusion faster than Ras1-GTP, restricting the lateral expansion of the zone, contributing both to its finite size and to its eventual disassembly. Some Gap1 remains at the former zone site after Ras1-GTP decays (Figs 4C and S3), similar to experiment (Fig 6D in [23]). The spatial profile of the zone (Fig 4D) compares well with experiments, with a local accumulation of Ras1-GTP (as in Fig 3G in [23]) and a wider Gap1 compared to Ras1-GTP (as in Fig 6I in [23]). The exploratory period in the model is 60 sec, which is close to the observed 90 sec in wt mating mixture experiments [9].
Our model provides an explanation for the uniform cortical activation and absence of active Ras1 patch in gap1Δ cells [23]. Upon removal of the Gap1-dependent hydrolysis of Ras1-GTP from the model (while keeping other parameters fixed as in Table 1), a zone is formed that keeps spreading until covering the whole cortex (Fig 4E). Thus, a minimum threshold of Gap1-dependent hydrolysis of Ras1-GTP is required to stop the spread of the patch into a homogenous state.
The behavior of the system in parameter space, varying the main coefficients of the positive feedback (the GEF-mediated activation rate constant of Ras1-GTP), and negative feedback (the Gap1-dependent hydrolysis rate constant of Ras1-GTP) is shown in Fig 5. The figure shows the region with single patch exploratory behavior surrounded by a region that lacks patches and other regions that may contain multiple stable or fluctuating patches. The region that lacks patches shows two different qualitative behaviors: first, when the positive feedback parameter is too small to initiate patch formation; second, when the negative feedback is too small to keep the patch from spreading all over the surface towards a homogeneously activated state. The behavior of the system as a function of saturation coefficient in recruitment of Gap1 and Gap1-dependent hydrolysis rate of Ras1-GTP is plotted in S4 Fig, which shows a similar structure with a region of single patch exploratory behavior surrounded by regions of different dynamics.
We studied the mechanisms by which our model can reproduce the transition from exploration to a stable zone, as occurs in the cell’s response to sensing of external pheromone. This change in the dynamical behavior of the system upon pheromone sensing could occur by three possible mechanisms: (1) a uniform change of the Ras1 rates constants of our model across the cell cortex, (2) a local change involving differential regulation of the Ras1 rate constants around the Ras1 patch site, or (3) a local change in the Ras1 rate constants around the region of high pheromone concentration. The last two mechanisms may be equivalent because pheromone receptors are locally active at the patch location, though prior data cannot distinguish whether, as in the second option, the patch brings receptors along to probe for local signal, or whether, as in the third option, receptors may be present and activated elsewhere at the cortex and promote patch stabilization if it forms at this location [11]. We note that this regulation likely involves additional factors not included explicitly in our model such as activated receptors, actin or MAPK signaling components [10, 30, 31].
The three mechanisms of patch stabilization described in the preceding paragraph were studied in simulations. A uniform change in the positive feedback rate constant k0p or negative feedback k2n across the cell cortex (around the reference values in Table 1) does not readily produce single stable patches but rather states with either none or multiple patches (Fig 5). However, a local increase of the positive feedback rate constant over a region on the cell membrane around the center of the exploring patch results in stabilization of the patch at that location (Fig 6A). This could result through locally stimulated receptors that regulate Ras1 positive feedbacks to stabilize the patch when it happens to form at a region of higher pheromone concentration [11]. A relatively small increase in the rate constants of the positive feedback over a fixed region on the cell membrane (corresponding to higher external pheromone) can also result in patch stabilization on that site after a few rounds of exploration (Fig 6B). A larger local increase causes the zone to form directly at that location. In conclusion, the model allows patch stabilization through mechanisms 2 and 3.
Another way to obtain a single stable patch is to decrease the detachment rate of Gap1 rGAP (Fig 6C) or decrease the saturation coefficient of the Gap1 recruitment rate Csat and Gap1-dependent hydrolysis rate constant k2n (S4 Fig). Whether these parameter changes occur locally around the patch or uniform across the surface is not as important because Gap1 is recruited at the patch following Ras1-GTP.
In prior experiments with external P factor sensed by cells lacking the P factor protease, the period of a single exploratory zone increased with pheromone concentration, which led to a model of partner selection through mutual stimulation [11]. We found that the patch lifetime in our model did not change significantly during the transition from exploration to stabilization through local positive feedback regulation (around the reference values of Table 1): as the positive feedback rate constants were increased locally, or decreased elsewhere on the cortex, the patch abruptly transitioned from exploratory dynamics with lifetime ~ 60 s to a single stable non-exploring patch. Similar abrupt transition (exploration with 60 s lifetime to stable patch) can be observed with a local decrease in the Gap1-dependent rate hydrolysis (k2n).
Interestingly, we found that decreasing rGAP globally lead to an increase of the lifetime of the exploring patch (Fig 6C). If such a mechanism underlies the control of Ras1 patch lifetime, then we would expect that the longer the patch lifetime, the higher the Gap1 concentration accumulated locally. The observed increase in Gap1 intensity at the fusion focus compared to an exploring patch [23] is consistent with the Gap1 membrane residence time playing a role in regulation of the patch lifetime in response to sensed pheromone concentration. We note that this is not the only possible mechanism leading to an increase in Gap1 intensity.
In conclusion, the model suggests that regulation of both positive and negative feedbacks is required for the observed dynamical response to pheromone in experiments (i.e. exploration to stabilization). We note that we cannot exclude the possibility that additional adaptation mechanisms [32] (in which the patch duration under high pheromone is related to the time required for the system to adapt back to the exploratory state) or other regulatory mechanisms, such as multi-step positive or negative feedbacks that couple to the sensing and Cdc42 system, may also regulate the patch period during the transition to a single stable zone.
The model further suggests how positive and negative feedbacks regulate the width of the Ras1-GTP zone, which may be part of the partner distance sensing mechanism. Indeed, it has been recently demonstrated that Ras1-GTP patch at the fusion focus is narrower in comparison to the exploratory patch (Fig 3G in [23]). In this section we study in more detail how feedback mechanisms regulate patch size in our model, and test them experimentally.
Increasing the Gap1 recruitment rate (parameter k3n) or the Gap1-dependent hydrolysis rate of Ras1-GTP (parameter k2n), with respect to the reference values of the Table 1, produces smaller patches (Figs 7A and S5A). The patch size gets smaller until reaching the minimum size limit of the simulations, which is equal to one polygonal cell on the surface mesh (one voronoi cell) (Fig 5). As discussed above, local recruitment of Gap1 that spreads in the vicinity of Ras1-GTP by diffusing faster than Ras1-GTP, restricts patch size. As a consequence, larger net hydrolysis rates lead to narrower patches, containing a higher concentration of Ras1-GTP, with approximately the same amount of GEF distributed over a narrower area.
An increase in the total GEF amount (parameter Ectot) or a uniform increase in the positive feedback rate constants (k0p or k1p and k2p) along the cell membrane results in the formation of two patches that compete within the appearance and disappearance time period (Figs 7B and S5B and S5D and S2 Movie). This is due to the fact that the stronger positive feedback rates, the result of a change in the above parameters, lead to higher nucleation rate during the process of a prior patch inactivation; this increases the likelihood that two patches competing for a limited pool of GEF start growing within a short time of one another, such that none of them has enough time to become dominant [33]. The system transitions to a very slow-competing phase with two or more stable zones upon further increase of the above parameters (Ectot,k0p,k1p and k2p) (Figs 5 and S4).
We also explored the dependence of a stable patch size to changes in the positive and negative feedback rate constants (Fig 7C where k0p was locally increased by 20% compared to the Table 1 to get a stable patch). Such a stable patch could be the result of stabilization after local pheromone sensing during the process of exploration or a patch at the tip of a shmoo growing closer towards that of a partner cell. We found that further increase in k0p lead to higher concentration of Ras1-GTP and Gap1 and marginally narrower patch. Additional increase of the Gap1-dependent hydrolysis rate (k2n) caused the patch to become narrower, as in Fig 7A, however in this case the concentration of Ras1-GTP did not increase significantly.
In summary, we find that increase in negative feedback rates makes the patch narrower with the intensity of Ras1-GTP staying the same or increasing depending on the reference state (i.e. exploring or stable patch). Increase in the positive feedback can cause formation of multiple patches or cause single patches to recruit more Ras1-GTP and Gap1, becoming marginally narrower.
To test our prediction about patch size regulation through positive and negative feedback we measured the size of exploratory Ras1-GTP patches in cells expressing RasActGFP and Myo52-tdTomato in wt and mutants overexpressing Gap1 or the GEF Ste6 (Fig 7D). By imaging every 10 min through a medial focal plane, we identified exploring cells in which a RasAct patch appeared in different position along the cortex. We identified patches as regions of the cell cortex with intensity above the cytoplasmic background (three times the standard deviation of the cytoplasmic region excluding the nucleus; the cytoplasmic intensity and its standard deviation was similar in all three cases, see Materials and Methods and S6 Fig). We did not use Myo52 localization as a criterion for patch detection since RasAct patches did not always colocalize with Myo52 (though they were frequently found together). Since RasAct has a spottier distribution compared to other patch markers such as Scd2, RasAct dots that were close to one another (of order 2 pixels = 260 nm) were counted as a single patch (Fig 7D). Cells occasionally contained more than 1 patch, each one of which could be at different stages of appearance and disappearance. We quantified the distributions of size and number of exploratory patches in Fig 7E and 7F (averaged over fluctuations in z position with respect to the focal plane; See Materials and Methods).
The experimental results in Fig 7 are in agreement with the trends predicted by our model. The average exploratory patch size in Gap1 overexpressing cells, 0.73±0.02μm (SEM) was smaller than 0.91±0.02μm in wt cells (Fig 7E, inset) and the number of patches per cell was less compared to wt cells (Fig 7F). This trend is in agreement with Fig 7A (increase in negative feedback parameter k3n), as well as Fig 5, which shows that increase in the negative feedback parameter k2n moves a system originally in the single patch oscillation region towards a region of narrower patches and away from a region with multiple patches. The patch size in Ste6 overexpressing cells 0.78±0.02μm (SEM) decreased compared to wt cells, in agreement with Fig 7C (increase in positive feedback rate constant k0p or total GEF amount Ectot (S5C Fig)). The model results in Figs 7B and 5 predict an increase in number of patches with an increase in the total amount of GEF or uniform increase of positive feedback rate constant k0p. A larger number of patches on average was indeed observed in Ste6 overexpressing cells (Fig 7F) though the relative magnitude of the change is small.
Though the experiments agree with model predictions, we note that patch size may be controlled by multiple factors. First, we should be cautious about interpreting the results of Ste6 overexpression in terms of changes in model parameters because the mechanism for Ste6 expression, activation and interaction with the pheromone signaling pathway is not fully resolved [7]. Second, we note that we did not consider the interaction between Cdc42 and Ras1 [34] that might form feedbacks that regulate the patch size. Finally, the actin cytoskeleton, endocytosis and exocytosis also likely contribute to the regulation of patch size especially in the final stages of fusion when the pheromone-induced formin Fus1 assembles the fusion focus [35]. We note that at these late stages, the curvature of the cell tip may also play a role. Bonnazi et al. [36] showed that the curvature-dependence of Cdc42 zone size required formin nucleated cortical actin cables and fusion of secretory vesicles transported along the actin cables. In our model, curvature had no drastic effect: variation of the tip radius of curvature from 1.8 to 2.3 μm while keeping the total cell surface area fixed gave Ras1-GTP and Gap1 patch sizes that varied by 1–4 Voronoi cells (for stable patches generated using Csat = 20 μm-2 and k2n = 0.01 μm2s-1, see S4 Fig., and other parameters as in Table 1; this small effect depended on the different Voronoi cell discretization at different curvatures).
In this study we showed that Ras1-GTP patch dynamics during mating in fission yeast cells can be modeled as reaction-diffusion system on the cell membrane with combined positive and negative feedbacks. We used FRAP experiments to estimate the diffusion coefficients of the main components in the system. The measured value of Ras1-GDP diffusion coefficient is similar to that of Cdc42-GDP membrane diffusion in fission yeast [37]. The slower Ras1-GTP diffusion is consistent with the slower diffusion coefficient of Cdc42-GTP compared to Cdc42-GDP [37]. Our estimation for Ras1-GDP membrane diffusion is twice the measured value for GFP-Ras2 membrane diffusion in budding yeast [38]. A very slow cytoplasmic exchange rate, comparable to our estimate, was measured for both single and double lipidated GFP-Ras2 [38]. Our values are also comparable to measurements of lateral diffusion of H-Ras and K-Ras proteins in mammalian cells [39, 40]. The measured slower Ras1-GTP diffusion is also in agreement with the observed reduction in membrane diffusion of active H-Ras and K-Ras in rat cells in comparison to inactive forms [41]. In mammalian cells, Ras proteins organize into nanodomains with distinct dynamics [42, 43]; here our measurements represent an average of diffusion through such possible nanodomains.
Using the measured values of diffusion coefficients, the model reproduces an exploratory patch with a lifetime comparable to experiments, and allows a switch from exploration to a stable patch by changes in the model’s rate constants. A patch forms in our model because Gap1 is recruited through locally self-amplifying Ras1-GTP. Due to its higher diffusion coefficient, it spreads further than the Ras1-GTP patch and limits the spread of the Ras1-GTP patch. Due to limited amount of GEF in the cell, autocatalytic activation of Ras1 slows down and saturates as the cytoplasmic pool of GEF gets depleted. The patch then starts to decay when the locally accumulated Gap1 inhibitor inactivates Ras1-GTP, releasing some of the GEF. This process allows a competing new patch to start elsewhere in the cortex, drawing the GEF away from the old patch. A local increase in the positive feedback rate constants above a certain threshold, which could occur by local pheromone sensing, stabilizes the patch against competition from other patches that have smaller positive feedback rate constants.
In the model, a balance of positive and negative feedback reactions, diffusive flux along the cortex and between cortex and cytoplasm is established, which regulates the patch size. We find that the patch size depends on the magnitude of Ras1-GTP inhibition. Increase in Gap1 recruitment rate to the cortex results in smaller patches. This prediction was confirmed by the experiments with overexpression of Gap1.
We also described the dynamical behavior of the model surrounding the state of single patch exploration (Figs 5 and S4), which may be observable with mutant alleles. Here we kept the values of diffusion coefficients equal to those we extracted from experimental data. We note another possible dynamical behavior: when the diffusion coefficient of Gap1 becomes sufficiently smaller than the reference value, the activation process is not confined to a patch and the system can transition to travelling wave behavior.
The mechanism of gradient sensing in mating fission yeast is an example of a cell system exhibiting a state of active exploratory fluctuations or oscillations that stabilize along a direction upon sensing of external signal [1, 44–46]. Since the exploratory state in this system consists of a single polarity patch, this process can also be described as a system that establishes polarity prior to gradient direction sensing [9, 10]. At the single cell level, exploration here works as part of the search mechanism for an optimal configuration (similar to finding a potential well in a wide and flat energy landscape), while Cdc42 fluctuations and oscillations during the transition from monopolar to bipolar growth in fission yeast have been described as facilitating the escape from an asymmetric state towards a symmetric polarization [27] (similar to crossing a barrier separating two potential wells). At the level of a cell population, exploratory partner search has been proposed to aid the population reach an optimal number of mating pairs [11] (in this case, over a landscape with many barriers and local minima).
Gradient sensing mechanisms in other cell systems have also been modeled with reaction-diffusion equations and related methods. In mating conditions, budding yeast cells form a Cdc42-GTP patch that wanders around the cell cortex, executing a biased random walk rather than appearance and disappearance. The patch moves upstream of the pheromone gradient and stabilizes close to opposite mating type [12]. Two recent studies [10, 13] proposed mechanisms for this type of patch motility, which differ to what we considered in our work. McLure et al. provided a mechanism based on local exocytosis contributing to patch displacement. In the mechanism of Hegemman et al. patch lateral motility depends on stochastic fluctuations: patch formation depends on autocatalytic activation of Cdc42 (based on a prior model [47]); a secondary positive feedback between actin-based receptor trafficking and Cdc42 [10, 47] results in biased motion up the pheromone gradient and stabilization once a high level of Cdc42-GTP is achieved. By contrast, patch formation and disappearance in our model relied on Turing-type mechanisms previously proposed for interphase budding yeast polarization and oscillations: from these studies we borrowed the mechanisms of competition of finite GEF [16] and Gap1 inhibition [21] used to model transient polarity patch oscillations.
Gradient sensing of chemoattractant has been widely studied in motile cells, such as amoebas, neutrophils, neurons and fibroblasts [1, 10, 45]. In many cases, the process of response to the gradient has similarities to the exploration and stabilization process of fission yeast mating patch. For example, upon starvation conditions, D.discoideum cells start the process of multi-cell aggregate formation by first breaking symmetry by forming filopodia and pseudopodia along many directions; those that happen to form in the direction of a diffusible external cue (cAMP secreted by other cells) win over the rest [48]. Thus, models developed for these systems share common features to our study.
Meinhardt (1999) proposed a model with a generic autocatalytic fast diffusive activator and a slow diffusive long-range inhibitor model, which is able to generate a stable Turing pattern. Addition of an extra generic inhibitor, which acts locally on the activator maximum peak to deactivate it leads to the generation and subsequent decay of a local activator maxima [15]. These local maxima are more likely to form along the direction in which the positive feedback rate constants have higher values (analogous to the process shown in Fig 6B for our model). A problem with this model is that it does not exhibit perfect adaptation, which inspired the development of Local Excitation, Global inhibition model [1]. The LEGI model has a fast acting local activator and a slow global inhibitor responding in direct proportion to an external signal. This model achieved perfect adaptation to signal gradient as well as high degree of sensitivity to changes in the signal gradient. However this model in turn needs additional components to generate significant amplification of an external stimuli and cannot account for persistence of polarity after the external signal is removed, as occurs in neutrophils and fibroblasts [45, 48]. More recent models have combined the LEGI model with an actin excitable system [32], to account for the process of adaptation as well as the existence of multiple layers of the process of cell polarization (accumulation of signaling components, actin polymerization, etc). Such models can further explain how D.discoideum cells in which actin polymerization was inhibited, were still capable of accumulating signaling proteins without pseudopodia or filopodia in response to the external cue gradient, as well as adapting to uniform increase of external signal [1].
Ras has been proposed to be involved in the signaling network providing perfect adaptation in D.discoideum (consistent with the LEGI model), with positive and negative feedback forming a parallel incoherent feedforward loop in response to external signal [49]. This is a different Ras signaling network connectivity compared to our model: in our case negative feedback follows a positive feedback that self-amplifies to form an exploratory patch, even without an increase of external input. While little is known about the adaptation properties of mating fission yeast, the fact that increase of external pheromone leads to measurable changes in patch lifetime [11] suggests that adaptation is either not perfect or else it occurs over times much longer than the patch lifetime. Despite this difference, we nevertheless anticipate a multilayer mechanism (analogous to D.discoideum) to exist in fission yeast. This requires further modeling work in future studies that include the contributions of Cdc42 and the actin cytoskeleton in addition to Ras1.
Strains used in this study are listed in Table 2. Standard genetic manipulation methods for transformation and tetrad dissection of S. pombe were used. For FRAP (Fluorescence Recovery After Photobleaching) experiments during exponential growth, cells were grown in Edinburgh minimal medium (EMM) supplemented with amino acids as required. For FRAP experiments and microscopy of cells during the mating process, liquid or agar minimal sporulation medium without nitrogen (MSL-N) was used [50, 51].
Gene tagging was performed at endogenous genomic locus at the 3’ end, yielding C-terminally tagged proteins, as described [52]. N-terminal tagging of Ras1 with GFP was performed as in [11]. Gene tagging was confirmed by diagnostic PCR for both sides of the gene.
Construction of fission yeast strains expressing overexpression of Ste6 and Gap1 (Pnmt1-ste6 and Pnmt1-gap1) was done by integration of ste6 and gap1 under nmt1 promoter at the ura4+ locus. First, kanMX-Pnmt1 fragment was excised from plasmid pSM647 (pFA6a-kanMX6-Pnmt1) through digestion with XmaI and EcoRI and ligated into similarly treated pAV133 (pJK211, a kind gift from Dr. Aleksandar Vjestica, UNIL) to generate plasmid pSM2106 (pJK211-kanMX-Pnmt1); second, ste6 was amplified from genomic DNA with primers osm4924 (5’- tccccccgggATGAGGTTTCAAACGACCGCAATAAG) and osm4925 (5’- tccccgcggTCAAAAAATGCCAGAATCAATTAGC), digested with XmaI and SacII and ligated to similarly treated pSM2106 to generate plasmid pSM2111 (pJK211-kanMX-Pnmt1-ste6); gap1 was amplified from genomic DNA with primers osm4918 (5’- tccCCCGGGATGACTAAGCGGCACTCTGGTACC) and osm4919 (5’- aaggaaaaaagcggccgcgTTACTTTCGTAAAAACAATTGTTC), digested with XmaI and NotI and ligated to similarly treated pSM2106 to generate plasmid pSM2110 (pJK211-kanMX-Pnmt1-gap1). Finally, pSM2110 and pSM2111 digested with AfeI were stably integrated as a single copy at the ura4+ locus in the yeast genome. In primer sequences, restriction sites are underlined.
Construction of strains to visualize the constitutively active ras1Q66L allele was done by integration of GFP-ras1Q66L at the endogenous ras1 locus. First, pSM1221 (pREP41-Pras1-GFP-ras1, [11]) was subjected to site directed mutagenesis with primers osm2167 (5’- GTATTGGACACGGCCGGTCTAGAGGAATATTCCGCTATG) and osm2168 (5’- CATAGCGGAATATTCCTCTAGACCGGCCGTGTCCAATAC) to generate plasmid pSM1392 (pREP41-Pras1-GFP-ras1Q66L). Second, pSM1392 digested with PstI and XmaI was stably integrated as single copy at the ras1 locus in the yeast genome, through transformation of a ras1::ura4+ strain and selection on agar plates containing 5-Fluoroorotic Acid (5-FOA). In primer sequences, inserted mutations are bold.
The DeltaVision platform (Applied Precision) described previously [53] was used for time-lapse imaging in Fig 7D that were performed as in [51]. Briefly, pre-cultures of cells were grown over night in MSL+N at 25°C to reach an OD600 of between 0.5 and 1. Cultures were then diluted to an OD600 of 0.025 in MSL+N and grown for 18 hours to an OD600 of between 0.5 and 1 at 30°C. Cells were washed three times with MSL-N, diluted to an OD600 of 1.5 in 1 ml MSL-N and incubated at 30°C for 4 h. Cells were mounted onto MSL-N agarose pads (2% agarose) before imaging in overnight movies.
FRAP data in Figs 2, 3 and S2 were obtained with a Photokinesis module on a spinning disk confocal system previously described [35, 53]. The FRAP experiments described in Fig 2 were performed by bleaching a cortical region at the cell side, in S2 Fig by bleaching a cortical region that included half of the cell tip, in Fig 3 by bleaching the entire Gap1-GFP signal at the fusion site (in this case mating pairs with a stable fusion focus, visualized as a single Myo52-tdTomato dot, where selected, [35]). The selected region was bleached following two pre-bleach acquisitions and recovery was followed at regular intervals of 1”.
FRAP kinetics and reaction-diffusion patterns depend on a system’s dimensionality and geometry. We simulated these processes over a 3D curved surface representing the cell membrane. The simulated geometry was that of a cylinder with hemispherical caps at its two ends with radius 2 μm. The tip-to-tip cell length for simulations of mating cells was set to 6 μm and 10–12 μm for interphase cells, representative of experimental images. We simulated diffusion on a curved surface by implementing the algorithm of Novak et al. [54], in which the Laplace-Beltrami operator is approximated locally with the Laplacian operator over a tangential plane. The curved geometry of the cell membrane was discretized into a set of Voronoi polygons. The area of each Voronoi cell in the simulations was between 0.017 to 0.046 μm2. We tested this algorithm by simulating diffusion with zero initial concentration except at a point placed on a cylinder or on spherical surface and comparing to the analytical solutions (for δ-function initial conditions) for these two cases.
We did not keep track of cytoplasmic concentrations, assuming that diffusion in the cytoplasm occurs with typical diffusion coefficients for single proteins. Such diffusion will smooth out cytoplasmic gradients over the Ras1-GTP patch size (~ 0.5 μm) over a time faster than the timescale of membrane binding and dissociation. Free diffusion across the cell through the cytoplasm would occur over seconds, which is much faster than the period of Ras1 patch appearance and disappearance (tens of seconds). Thus we also approximated uniform cytoplasmic concentration across the cell throughout the process of patch appearance and disappearance.
For estimation of diffusion and rates of dissociation of Ras1-GDP and Ras1-GTP from the plasma membrane, we carried out FRAP experiments and fitted the data using numerical simulations.
To measure Ras1-GDP, we bleached GFP-Ras1 over a small (1.5±0.2μm) or a large (4.0±0.2μm) area at sides of a WT cells (Fig 2A). The acquired images of a single confocal slice through the bleached zone were corrected for photo bleaching by fitting an exponential decay function with decay constant rPB to the cytosolic signal after subtracting the out-of-cell background. The pixel intensity at every time point was corrected by multiplying by 1-erPBt after subtracting the out-of-cell background. To quantify the recovery of GFP-Ras1 intensity over time, the average intensity of all the pixels within the bleached area was calculated and normalized with respect to the corresponding value at the pre-bleaching image. The normalized average intensity was calculated for 5 cells and their mean and standard deviation (Fig 2B). To measure Ras1-GTP, we performed the same analysis by bleaching GFP-Ras1 at the sides of constitutively active ras1Q66L cells (Fig 2C and 2D).
We fitted the FRAP data with our 3D model, assuming uniform membrane diffusion and constant rates of association and dissociation to the membrane:
∂C∂t=DΔSC+j+−rC
(1)
Here C is the concentration of Ras1-GDP or Ras1-GTP, D is the diffusion coefficient, j+ is the association rate from the cytoplasm to the membrane and r is the dissociation rate from the membrane to the cytoplasm. The second order differential operator ΔS is the Laplace–Beltrami operator. Starting from a system at steady state (j+=rC), bleaching is simulating by setting the concentration equal to zero for all Voronoi cells within a short (1.5μm) or long (4μm) region at the cell sides and along 6.28μm around the cylindrical circumference (inset of S1B and S1C Fig), using a time step 0.01 s.
By varying parameters D and r in Eq (1), we fitted the data in Fig 1B and 1D as well as the full recovery profiles (S1B–S1E Fig). To compare simulations to experimental data, we calculated the normalized concentrations on the Voronoi cells on a medial focal imaging plane and within 0.6μm in the vertical direction (approximate vertical width of the microscope’s point spread function). Diffusion results in different recovery rates between small and large regions, however small (large) D can be partly balanced by large (small) r (Fig 2E). The fits to the recovery curves of Fig 2B and 2D give DRD=0.03-0.15μm2s-1,rRD<0.03s-1,DRT=0.02-0.07μm2s-1,rRT<0.025s-1 (Fig 2E). Further considering the fits to the full profiles (S1B–S1E Fig) provides a narrower range: DRD=0.145-0.155μm2s-1,rRD<0.005s-1,DRT=0.035-0.045μm2s-1,rRT=0.018-0.022s-1.
For estimation of diffusion and dissociation rates of Gap1 a different approach compared to Ras1 had to be used because of its localized recruitment to the polarity patch. In this case we either bleached the full Gap1-GFP signal at the fusion focus in cells expressing Gap1-GFP and Myo52-tdTomato in mating conditions (Fig 3A) or we bleached half of Gap1-GFP at the cell tip during vegetative growth (S2A Fig). The acquired images were corrected for photobleaching prior to analysis as described above. For the Gap1-GFP full fusion focus recovery we observed a recovery up to 50%, which we interpret as a large portion of the cell’s Gap1 being bleached (Fig 3A and 3C). We identified the center and the side of the Gap1 distribution using the Myo52-tdTomato signal (Fig 3B). The average of intensity of all the pixels within each region shown in Fig 3B was calculated, normalized to the corresponding value at the pre-bleaching image. An example of the recovery at the middle region and side regions is shown in Fig 3C.
We used the 3D model to simulate the recovery of Gap1-GFP at the fusion focus using a model with localized recruitment of Gap1 to the cell tip (where the fusion focus is typically located), diffusion on the surface and uniform rate of membrane dissociation:
∂C(r,t)∂t=DΔSC(r,t)+A2σ2πe−d2(r)2σ2−rC(r,t),
(2)
where C is the concentration of Gap1, D is the diffusion coefficient, r is the dissociation rate from the membrane to the cytoplasm, and d(r) is arc-length distance to the cell tip. The second term on right is the Gaussian recruitment function, with amplitude A and standard deviation σ=0.4μm estimated by measuring the full width at half maximum of Myo52-tdTomato signals at the fusion focus. The value of A was reduced to the half of the initial value after bleaching in the simulations to allow for 50% recovery similar to the experimental results (Figs 3C and S2E). A series of simulations starting with no Gap1 on the cell surface explored the dependence on the values of D and r. Then the average surface concentration of the Gap1 protein on the all Voronoi cells that were within the same middle and side area size as the experiments was calculated. This was compared to the normalized intensity of Gap1-GFP recovery at the middle and at the sides separately (Fig 3C). Good fits were obtained for D=0.1-0.3μm2s-1 and r=0.015-0.025s-1.
The half-tip Gap1-GFP bleaching in the vegetative cells was done for the purpose of tracking diffusion from non-bleached region to the bleached region. In these experiments we observed a recovery up to 70% at the bleached region, which we interpret as a large portion of cellular Gap1 being bleached during these experiments (S2B and S2D Fig). We followed the changes in the intensity of Gap1-GFP in the bleached and non-bleached area separately after bleaching, as shown in S2B Fig. The average intensity of all the pixels within each region (~1.8 μm in width) was calculated and normalized to the corresponding value of the pre-bleaching image. The recovery of Gap1-GFP was averaged over 3 cells (S2D Fig) that had similar non-bleached area size, 2.8-3.1μm.
To model the half-tip Gap1-GFP recovery we used the same model discussed for full fusion focus with the difference in reducing the amplitude of Gaussian recruitment function after bleaching to 0.7 of initial value to account for the lost portion of Gap1-GFP by bleaching (S2C and S2D Fig). We let the system to reach steady state and then the Gap1 concentration was deleted over half of the tip as in the FRAP experiments. The standard deviation of the Gaussian function, σ, was estimated to be 0.8μm by comparing the full width at half maximum of Gap1-GFP in vegetative cells with the width at half maximum of Gap1-GFP at fusion focus in mating cells. The set of D, between 0.1μm2s-1 to 0.3μm2s-1, and r=0.015-0.025s-1 values that were determined to be good fits from the full fusion focus Gap1-GFP recovery then were used for half-tip FRAP simulations to determine the best fit values D=0.2μm2s-1,r=0.02s-1 (S2D Fig). The curve with D=0μm2s-1 in S2D Fig is shown to demonstrate that cytoplasmic exchange by itself up to 70% of the initial amplitude cannot explain the loss of Gap1-GFP from the non-bleached region in the experiments even after adjusting r to the optimal value of 0.08s-1.
The model is described by Eqs (3)–(7) presented in this section. The equations for Ras1-GTP and Ras1-GDP surface densities are as follows:
∂CRT∂t=DRTΔSCRT+(k0pCGEF+rnoise)CRD−(k1n+k2nCGAP)CRT−rRTCRT,
(3)
∂CRD∂t=DRDΔSCRD+jRDp+(k1n+k2nCGAP)CRT−(k0pCGEF+rnoise)CRD−rRDCRD
(4)
where ΔS is the Laplace–Beltrami diffusion operator, k0pis the GEF-mediated activation rate constant of Ras1-GDP, k1n and k2n are the rate constants of spontaneous and Gap1-mediated hydrolysis of Ras1-GTP, and jRDpis the uniform and constant rate of Ras1-GDP to the membrane from the cytoplasm. We use symbol r to label rate constants involving dissociation from membrane and k for rate constants involved in Ras1 activation or inactivation with superscript p or n, respectively. We also included a comparatively small random background activation of Ras1 by random variable rnoise, the magnitude of which was set to correspond to approximately 6 activated molecules per second per cell, the number used for Cdc42 in [16]. For more details see section “Implementation and effect of random Ras1 activation” below.
This model includes a positive feedback for activation of Ras1 by an autocatalytic interaction with GEFs. This is implemented similarly to the positive feedback in [16] by assuming GEFs are in quasi-static equilibrium with Ras1-GTP and assuming a quadratic non-linear dependence (the linear term was not sufficient for symmetry breaking in [16]):
CGEF=k1pEcVCRT+k2pEcVCRT2
(5)
Here Ec is the available number of GEF molecules in the cytoplasm and V is the cell volume. The available number of GEF molecules in the cytoplasm in each time step, Ec, is calculated by Ec=Ectot-∫CGEFda, where Ectot is the total number of GEF molecules in the cell and the integral is over the cell’s surface area, implying
Ec=Ectot/(1+∫[k1p1VCRT+k2p1VCRT2]da).
(6)
The finite pool of GEF limits the number of patches that can form in the system and amount of activated Ras1 in the patch. See section “GEF Quasi-static equilibrium approximation” below for a more detailed examination of our assumptions on GEF properties.
There is also a negative feedback loop in this model to account for the role of Gap1. We assumed that Gap1 is recruited to the membrane through Ras1-GTP as previously shown [28] and similar to the model of negative feedback for Cdc42 oscillations in budding yeast [21]. The equation for Gap1 is as follows:
∂CGAP∂t=DGAPΔSCGAP+k3nCRThCsath+CRTh−rGAPCGAP
(7)
The non-linear recruitment rate of the second term on the right hand side was chosen to represent cooperative Gap1 recruitment at small Ras1-GTP concentrations, reaching a plateau for Ras1-GTP concentration above Csat. We used a value h = 2 that was the smallest integer value sufficient to provide a Ras1 exploratory zone.
Eqs (3)–(7) together with the model parameter values in Table 1 provide the complete model. For initialization, we start the simulations by setting CRD=jRDp/rRDplus or minus small relative random fluctuations, initialize a relatively smaller random Ras1-GTP field and set CGAP=0 along the Voronoi cells representing the cell membrane.
The noise term in Eq (4), which represents fluctuations in Ras1 activation, is needed for the generation of a few Ras1-GTP to get the positive feedback started. We calculated the number dN of activated Ras1 per integration time interval dt over a Voronoi cell of surface area da to be dN = dt daprnoiseCRD, where p is a random number picked from a uniform probability distribution between 0 and 1. The corresponding change in the surface density was added to the corresponding Voronoi cell. This implementation of multiplicative noise (proportional to the surface density of Ras1-GDP) is specific to the chosen integration time interval, which we kept constant.
To study the effect of noise amplitude, we ran simulations with different values of rnoise, keeping the rest of the model parameters as in Table 1. The results of these simulations are summarized S7 Fig, demonstrating the period of patch appearance and disappearance as a function of rnoise (with 0.002 s-1 the default value). There is a minimum threshold rnoise = 0.0005 s-1 to get appearance and disappearance: below this threshold there is not enough spontaneous activation to trigger new patch formation. As rnoise is increased above the default value, patch appearance and disappearance becomes more irregular and sometimes more two or more patches form in the simulations with one patch growing while other one shrinks/disappears or two competing patches forming simultaneously. There is an upper limit rnoise = 0.008 s-1 above which random activation results in uniform activation along the cell surface.
Here we discuss in more detail the assumption of GEF quasi-steady state of Eq (5). The precise mechanism by which Set6 gets recruited to the membrane and by which it diffuses on the membrane is not yet known. One way of achieving the required non-linear positive feedback is for Ste6 to get recruited by single Ras1-GTP molecules and by aggregates with two Ras1-GTP (possibly involving a scaffold protein). To check our assumption for GEF to be in quasi-static equilibrium, we added GEF binding and dissociation reactions to obtain the following expanded model:
∂CRT∂t=DRTΔSCRT+(k0pCGEF+rnoise)CRD−(k1n+k2nCGAP)CRT−rRTCRT
(8)
∂CGEF∂t=DGEFΔSCGEF+ρ1ECVCRT+ρ2ECVCRT2−rGEFCGEF
(9)
∂CRD∂t=DRDΔSCRD+jRDp+(k1n+k2nCGAP)CRT−(k0pCGEF+rnoise)CRD−rRDCRD
(10)
∂CGAP∂t=DGAPΔSCGAP+k3nCRThCsath+CRTh−rGAPCGAP
(11)
Ec=Ectot−∫CGEFda
(12)
The new equation for CGEF describes the surface density of Ste6, ρ1 is the rate constant for recruitment of Ste6 to the membrane by single Ras1-GTP molecules and ρ2 the rate constant of Ste6 recruitment by aggregates with two Ras1-GTP. We use the same membrane diffusion coefficient for membrane bound GEF as for Ras1-GTP. To choose ρ1 and ρ2 parameters consistent with the model of Eqs (3)–(7), we relate them to parameters k1p and k2p via:
k1p=ρ1/rGEF
(13)
k2p=ρ2/rGEF
(14)
In the limit of fast GEF dissociation (large rGEF), the reaction terms on the right hand side of Eq (9) balance, resulting in Eq (5).
We found that using rGEF = 10 s-1 and other parameters as in Table 1 (except for the need to use a smaller integration time step of 10−5 s) produced patch appearance and disappearance similar to Fig 5, with a slightly shorter period of 45 sec. For rGEF=0.1,1s-1 the patch did not reform after an initial appearance and disappearance. Thus the model of Eqs (3)–(7) requires that the patch establishes biochemical equilibrium faster than a sec timescale.
The requirement for rGEF ~ 10 s-1is of the same order as the value of 10 s-1 for Cdc24 GEF dissociation from the budding yeast Cdc42 patch in the model of Goryachev and Pokhilko [16]. However we note that these authors used a different GEF mechanism compared to Eqs (8)–(12). In their study, positive Cdc42 feedback arises from the combination of slow uniform GEF recruitment to the membrane (similar to random Ras1 activation in our model, needed to get the Ras1 activation started) followed by GEF-mediated Cdc42 activation and then complex formation of GEF with active Cdc42.
Finally, we note that Ste6 transcription is a target of the MAPK cascade and so it may take a while after pheromone exposure until it reaches steady state. However, patch dynamics continue in much the same way for at least 14h in cells without partners, when Ste6 levels should be at steady state. Thus, modulations in Ste6 expression levels are not likely to play a role in patch dynamics.
To calculate the spontaneous Ras1-GTP hydrolysis rate, we fitted the in vitro Ras1-GTP hydrolysis data of (Fig 4A in [23]) with an exponential decay function. This leads to a spontaneous hydrolysis rate k1n=(1.23±0.05)×10-3s-1. To estimate the Gap1-dependent hydrolysis rate we fitted the GST-Ras1 + MPB Gap1-1 graph (Fig 4A in [23]) with an exponential, which gives a decay rate 0.00293s-1. Assuming that this rate is linear in Gap1 concentration and equal to k1n+k2n,3DCGAP3D, where 3D indicated concentrations per unit volume (as opposed to per unit area) we obtain k2n,3D=7.2×10-7μm3s-1. Assuming that Gap1 bound to the cell membrane is within w = 10 nm off the cell membrane gives an estimate for the Gap1 dependent hydrolysis rate for the model k2n=k2n,3D/w=7.2×10-5μm2s-1. In the model we had to use a larger value (0.1 μm2s-1) such that a realistic concentration of Gap1 at the membrane is sufficient to cause zone disassembly. The latter is not an unlikely possibility since Gap1 may be better positioned for hydrolysis when bound to the membrane in cells. Use of a higher Ras1 spontaneous hydrolysis rate in the model does not change the previous conclusions though it brings the system closer to the winner-take-all mechanism, with less role of Gap1 in determining the zone size.
Wt, Ste6 overexpression and Gap1 overexpression cells expressing RasActGFP and Myo52-tdTomato, imaged every 10 minutes on a DeltaVision platform, showed exploratory patch of RasActGFP that appeared at different position along the cell cortex in early stages of mating. We identified any bright region on the cell cortex which was equal or larger than two pixels and was brighter than the average cell background signal. By drawing a line manually along the bright regions their size and average intensity was measured. We also measured the cell background and standard deviation between the intensity of cell background pixels by drawing an irregular shape inside the cell excluding the nucleus. RasActGFP patch was defined as any bright region with an average intensity higher than 3 times of standard deviation between the intensity of cell background pixels. The cytoplasmic intensity and its standard deviation was similar in all three cases (S6 Fig). Since RasActGFP has a spottier distribution compared to other patch markers such as Scd2, RasActGFP regions that were close to one another (of order 2 pixels = 260 nm) were counted as a single patch. We also quantified the number of simultaneous patches along the cell cortex in each cell type. Then we averaged over the patch sizes in every cell type and performing two sample t-test between the patch size mean of wt and Ste6/Gap1 overexpression mutants we could show that the averages are significantly different, P-value < 0.001.
|
10.1371/journal.pbio.3000290 | Representation of abstract semantic knowledge in populations of human single neurons in the medial temporal lobe | Sensory experience elicits complex activity patterns throughout the neocortex. Projections from the neocortex converge onto the medial temporal lobe (MTL), in which distributed neocortical firing patterns are distilled into sparse representations. The precise nature of these neuronal representations is still unknown. Here, we show that population activity patterns in the MTL are governed by high levels of semantic abstraction. We recorded human single-unit activity in the MTL (4,917 units, 25 patients) while subjects viewed 100 images grouped into 10 semantic categories of 10 exemplars each. High levels of semantic abstraction were indicated by representational similarity analyses (RSAs) of patterns elicited by individual stimuli. Moreover, pattern classifiers trained to decode semantic categories generalised successfully to unseen exemplars, and classifiers trained to decode exemplar identity more often confused exemplars of the same versus different categories. Semantic abstraction and generalisation may thus be key to efficiently distill the essence of an experience into sparse representations in the human MTL. Although semantic abstraction is efficient and may facilitate generalisation of knowledge to novel situations, it comes at the cost of a loss of detail and may be central to the generation of false memories.
| What is the neuronal code for sensory experience in the human medial temporal lobe (MTL)? Single-cell electrophysiology in the awake human brain during chronic, invasive epilepsy monitoring has previously revealed the existence of so-called concept cells. These cells have been found to increase their firing rate in response to, for example, the famous tennis player ‘Roger Federer’, whether his name is spoken by a computer voice or a picture of him is presented on a computer screen. These neurons thus seem to encode the semantic content of a stimulus, regardless of the sensory modality through which it is delivered. Previous work has predominantly focused on individual neurons that were selected based on their strong response to a particular stimulus using rather conservative statistical criteria. Those studies stressed that concept cells encode a single, concrete concept in an all-or-nothing fashion. Here, we analysed the neuronal code on the level of the entire population of neurons without any preselection. We conducted representational similarity analyses (RSAs) and pattern classification analyses of firing patterns evoked by visual stimuli (for example, a picture of an apple) that could be grouped into semantic categories on multiple levels of abstraction (‘fruit’, ‘food’, ‘natural things’). We found that neuronal activation patterns contain information on higher levels of categorical abstraction rather than just the level of individual exemplars. On the one hand, the neuronal code in the human MTL thus seems well suited to generalise semantic knowledge to new situations; on the other hand, it could also be responsible for the generation of false memories.
| Cognitive faculties enabling flexible adaption of behaviour are at the heart of the human species’ evolutionary success. Cognition operates on abstract representations of knowledge derived from prior experience [1]. Abstraction can have two separate but related meanings [2]. First, formation of a concept in semantic memory requires abstraction in the sense of generalisation across episodes. For example, the concept ‘dog’, a furry animal that barks, is learned by extracting regularities among multiple encounters with various exemplars of dogs. Second, abstraction can also refer to the extraction of meaning from sensory input in a single instance of perception. Abstraction in the latter sense ranges from lower, more concrete levels (e.g., labelling a percept as ‘terrier’) to intermediate levels (‘dog’) and high, superordinate levels (‘animal’). Abstraction both as a cross-episode generalisation and as an extraction of supramodal semantic information from sensory input are in constant interplay and shape episodic and semantic memory representations [3,4].
Our knowledge about semantic representations in the human brain is for the most part restricted to the cortex. Putative functional roles of involved neocortical regions correspond to sensory and/or motor features of an encoded concept [1,5]. Here, abstract categories such as, for example, living and nonliving things differ with respect to which portions of the neocortex are recruited for their encoding. Due to such macroscopic, topographical organisation of semantic representations in the neocortex, these representations can be investigated with rather coarse imaging techniques such as functional magnetic resonance imaging [5]. Large strides have also been made in elucidating the neuronal code of object and face recognition along the ventral processing pathway of nonhuman primates leading up to highly abstract representation in monkey inferotemporal cortex and the amygdala [6,7]. Next to categorical codes, influential approaches also entail mapping semantic concepts onto a multidimensional, semantic space along dimensions such as living–nonliving or abstract–concrete [2,8,9].
Investigating object recognition and semantic representations at the final stages of the ventral processing pathway in the human medial temporal lobe (MTL), including the amygdala, has been notoriously difficult. Investigation of neuronal representations in the human MTL at the relevant level of detail seems impossible with noninvasive imaging techniques because—unlike the neocortex—most MTL areas lack semantic topographical organisation [10,11]. Studies conducted in the setting of invasive epilepsy monitoring using additional microelectrodes to record action potentials of single units have been instrumental for this purpose [10–15]. A seminal finding of these studies is that some MTL units responded in a selective and invariant manner to various images of a familiar person and even to their written and spoken name, suggesting that they encode the identity of that person and thus the contents of a concrete semantic concept in an all-or-none fashion [13,14]. However, further studies emphasised that MTL neurons can also respond to a wider range of stimuli in graded fashions in which sometimes more abstract semantic relations between stimuli can be identified such as, for example, membership to a broad category [9,14,15,16]. Thus, rather than all-or-nothing responses to specific concepts, it could be that neurons in the human MTL encode semantic features along continuous dimensions, resulting in ‘semantic tuning curves’. Or as Kornblith and Tsao [6] put it in the context of face-patches in primate IT, they are ‘[…] measuring faces, they are not yet explicitly classifying them’.
Previous human single unit studies often preselected units based on rather conservative response criteria, which may have resulted in a potential overestimation of all-or-none responses to individual semantic concepts. In the current study, in contrast, we analyse representations at the level of the entire population of units we record from. By doing so, we investigate how and at what level of abstraction semantic information conveyed by visual input is encoded in activity of single units in the human MTL. In contrast to previous studies, we consistently used the same set of images across sessions and patients, and the images could be grouped at multiple levels of abstraction. This procedure, in combination with a large sample of epileptic patients, allowed us to record neuronal responses for each image in a population of neurons unprecedented in size. Using this procedure, we could characterise and compare the nature of representations and their level of abstraction at a population level for different regions of the MTL.
Subjects (N = 25; 59 sessions) were bilaterally implanted with depth electrodes for seizure monitoring in the amygdala, hippocampus, entorhinal cortex, and parahippocampal cortex. Subjects were presented with visual stimuli depicting objects from 10 semantic categories consisting of 10 exemplars each (100 images, 10 trials each). The subjects’ task was to indicate by button press whether a man-made or natural object was depicted. As expected, this task was very easy as reflected by high accuracy (median = 97.62%, IQR = 2.25%) and short reaction times (median = 669 ms, IQR = 146 ms).
We first analysed our data by classifying units into responsive and nonresponsive, according to an established criterion (see Neuronal response test section in Materials and methods) as in previous studies [12,13] (Figs 1 and 2). Our analyses confirm that some units in the MTL respond to only a few stimuli in the set (Fig 1). We recorded from a total of 4,917 units, 2,009 of which were classified as single units (41%). In the amygdala, we found 1,392 units (656 single units [47%]), in the hippocampus 1,863 units (706 single units [38%]), in the entorhinal cortex 828 units (328 single units [40%]), and 831 units (319 single units [38%]) in the parahippocampal cortex (Fig 2B). A subset of 785 units responded with increased firing rates to at least one of the 100 stimuli (see Neuronal response test section in Materials in methods; Fig 2B). Selectivity as determined by the number of response-eliciting stimuli for a given neuron was similar in the entorhinal cortex, amygdala, and hippocampus but was markedly lower in the parahippocampal cortex [12] (Fig 2C). Some units responded very selectively, sometimes to only one of the stimuli in the set (Fig 1D–1F). In the amygdala, this was the case in 43% of the responsive units, in the hippocampus 57%, and in the entorhinal cortex 54%. This number was markedly lower in the parahippocampal cortex, namely, 35%. When units responded to multiple stimuli, the response-eliciting stimuli were often from the same semantic category (Fig 1A–1C and 1G–1I).
We also calculated the probabilities with which images from a given category elicited a neuronal response, separate for each anatomical region in the MTL. To this aim, we computed the number of significant responses across all units and divided this number by the total number of stimuli and the number of units. Observed response probabilities ranged between approximately 0.25% and 2% across anatomical regions and stimulus categories (Fig 1D). Neurons responded more frequently to food stimuli than to stimuli of other categories, which was especially prominent in the amygdala and, to a lesser degree, also in the hippocampus and entorhinal cortex (Fig 2A).
Going beyond analyses of responsive versus nonresponding units, we next looked at responses of the whole population of units we recorded from. With these analyses, we find that population activity is determined by abstract, semantic features of the stimuli. We investigated population activity by representational similarity analyses (RSAs) [9,17,18]. To this aim, we quantified each neuronal response to a stimulus using a single Z score that expressed average firing across all trials of a stimulus in the 1,000 ms after stimulus onset, normalised using the distribution of baseline firing rates (−500 to 0 ms relative to stimulus onset) across all trials. The population response to a stimulus thus corresponded to a population vector of Z scores from all units in a given region. Representational dissimilarity (i.e., distance) between two stimuli was then quantified as 1 − Pearson’s correlation coefficient of their population vectors. Representational dissimilarities are displayed as matrices of colour-coded distance between all pairs of stimuli (Fig 3A–3D). Representational dissimilarity analyses showed that population firing patterns evoked by stimuli of the same category were more similar than those evoked by stimuli from different categories in all anatomical regions (Fig 3A–3D; all p < 10−5; random permutation test, Inference statistics on representational dissimilarity and confusion matrices section in Materials and methods).
To elucidate potential principles on higher levels of abstraction, we applied multidimensional scaling (Fig 3E–3H) and automated hierarchical clustering (Fig 3I–3M, S3 Fig) to these dissimilarity matrices. Remarkably, inspection of dendrograms obtained from hierarchical clustering revealed that the preconceived assignment of stimuli to superordinate categories was almost perfectly reflected in representational dissimilarity of the recorded population activity in the amygdala and hippocampus (Fig 3I and 3K). That preconceived categories matched information present in neuronal representations is evidenced by the sorting on the x-axis of the dendrograms. Perfect correspondence between neuronal similarity and category membership is indicated in that all exemplars of a category line up next to one another on the x-axis after sorting according to similarity. This is the case for all but two categories in the amygdala, in which only one exemplar of the ‘computer’ category ends up closer to other exemplars from the ‘musical instruments’ category. A similar pattern of exemplar sorting is evident in the hippocampus, whereas this was not the case in the entorhinal and parahippocampal cortex (Fig 3L and 3M). RSAs for units that did not show a response according to any of the stimuli in our set (according to the statistical response criterion used in this and previous studies) showed similar patterns of similarity (S1 Fig). Consequently, representational similarities of nonresponding units alone are statistically significantly higher for within- versus between-category pairs (all p < 10−5; see ‘Inference statistics on representational dissimilarity and confusion matrices’ section in Materials and methods), suggesting that even small variations in firing rate of MTL units contain considerable amounts of information at an abstract, categorical level.
Representations clustered beyond our preconceived categories in a highly abstract but meaningful way. Abstract semantic clusters of representational similarity emerging from neuronal representations are visualised by the dendrograms resulting from hierarchical clustering (Fig 3I–3K) and by projections of multidimensional scaling onto a two-dimensional space (Fig 3E–3H). In the amygdala, we saw a food cluster that consisted of all exemplars of man-made food and fruit categories. This food cluster becomes evident in that exemplars from the preconceived categories of ‘man-made food’ and ‘fruit’ are close together in the 2-dimensional projection generated by multidimensional scaling (Fig 3E). An animal cluster entailed exemplars of wild animals, birds, and insects. The categories of all man-made objects together constituted a further cluster. In the hippocampus, we additionally observed a clear separation between man-made and natural objects. This separation becomes evident when one draws a diagonal from top left to bottom right in Fig 3F that almost perfectly separates manmade from natural exemplars. Such clearly semantic principles governing representational similarity at a high level of abstraction were less evident in the entorhinal and parahippocampal cortex.
To assess whether low-level physical image similarity could have been responsible for these findings, we calculated four widely used statistics to compare physical properties of two images, namely, the Euclidean distance, the mean squared error, the peak signal-to-noise value, and the structural similarity index. We then performed analyses analog to the ones shown in Fig 3 using these image similarity measures (S2 Fig). These analyses showed no emergence of higher-order grouping of images according to abstract semantics as was the case for the neural data (Fig 3). Therefore, we conclude that low-level physical similarity cannot account for the findings of representation similarity in our neuronal response patterns.
Abstraction comes at a trade-off between generalisation of knowledge to new situations and confusion between similar exemplars. We used the population responses described above to train pattern classifiers (multiclass support vector machine models; see Decoding of stimulus identity and category section in Methods and materials). A classifier was trained on the population responses of half the stimuli per category to predict the category label and was then tested out of sample on population responses of the other half of stimuli. This procedure was repeated 100 times with random divisions of the data into training and test sets. Successful generalisation to untrained stimuli was indicated by highly accurate out-of-sample classification of category labels from population responses (Fig 4A; for separate analyses for each subject, collapsing across anatomical regions, see S4 Fig). Generalisation was best using population responses from amygdala units, intermediate using hippocampal and entorhinal units, and lowest using parahippocampal units. Nevertheless, generalisation exceeded chance performance in all MTL regions by far.
To assess performance in classifying individual stimuli, we calculated Z scored population responses of unit firing for each trial in the same manner as described above. Pattern classification algorithms were then trained on population responses of half of the trials for each stimulus and tested out of sample on the other half. Again, out-of-sample performance was assessed in 100 random divisions of the data into training and test set. Classification performance exceeded chance level in all regions of the MTL (Fig 4F). Interestingly, we found a systematic pattern of misclassifications when inspecting confusion matrices (Fig 4G–4K). Confusion matrices cross-tabulate the number of classifier outcomes by predicted stimulus label in columns and true stimulus labels in rows. These analyses show that pattern classification algorithms trained to decode individual stimulus identity more often confused stimuli from the same versus different superordinate categories (Fig 4F–4K; all regions p < 10−5, permutation test; see ‘Inference statistics on representational dissimilarity and confusion matrices’ section in Materials and methods; for analogous analyses separately for each subject but collapsing across anatomical regions, see S5 Fig).
Taken together, our results provide a novel perspective on how information is encoded in the human MTL. We demonstrate that despite selective tuning of individual neurons to only a few stimuli in the set, activity at the population level is determined by information with a high degree of semantic abstraction. We find that population activity is similar in response to exemplars of the same category and that response pattern similarity extends to highly abstract semantic categories. Pattern classification results show high levels of semantic abstraction, which, on one hand, can be useful for successful generalisation of knowledge to novel situations. On the other hand, semantic abstraction comes at the cost of confusion between semantically similar stimuli.
With respect to neuronal representations in the MTL, we demonstrate a semantic code that spans multiple layers of abstraction emerging at the population level. This perspective may aid to reconcile disparate findings from previous studies investigating response properties of individual units [11,13,16]. Some have concluded that unit activity encodes concrete concepts such as, for example, a person’s identity [13,14]. Others postulate superordinate category membership as a decisive feature driving unit activity [16,19]. Our study may reconcile these views as population-level analyses show that encoded information spans across multiple levels of abstraction ranging from the concrete exemplar level to the level of preconceived semantic categories and beyond. Pattern classification analyses demonstrate that information on the exemplar and superordinate categorical level can both be decoded from population activity, whereas categorical information seems predominant. These aspects may not become apparent when looking at response profiles of individual units and underscore the importance of analyses at the population level.
Furthermore, our data refine the view on sparseness of coding in the human MTL. Hallmark human single unit studies suggest that very few concepts drive activity in one single neuron [13,14,20]. In fact, considerably more than 50% of responsive units were found previously to respond to only one out of approximately 100 stimuli [12]. This is true in the amygdala, hippocampus, and entorhinal cortex, whereas selectivity is lower in the parahippocampal cortex [12]. These findings led to the conclusion that the MTL uses a very sparse, almost ‘grandmother cell’-like code [21]. Although some units in our data set indeed only fired in response to one stimulus in the set, the overall selectivity in our study was lower (see Fig 1F) than reported earlier [12,20]. Previous studies used stimulus sets that were tailored to the patients’ interests, depicting relatives, preferred celebrities, and job- and hobby-related objects [12,13]. The aim in these studies was to screen for response-eliciting stimuli using a wide range of different concepts, likely resulting in rather low semantic feature overlap between stimuli. Our current stimulus material had a systematic semantic structure because images were grouped into categories of semantically related exemplars. Assuming that unit activity is determined by a rather narrow ‘semantic tuning curve’, we would indeed expect that neurons fire less selectively when ‘semantic distance’ between stimuli is sufficiently low. Thus, semantic relatedness between stimuli in a set seems likely to influence estimates of sparseness of unit responses in the MTL.
Two previous studies have applied RSAs to single units in the human MTL. First, in 2011, Mormann and colleagues [17] used RSA in combination with images that could be grouped into 3 categories, namely, persons, animals, and landmarks. This study found that the amygdala is preferentially activated by animal stimuli but did not investigate the semantic nature and level of abstraction in amygdala unit activity. Furthermore, a 2015 paper again by Mormann and colleagues [18] used RSA to show that units in the amygdala encode face identity rather than gaze direction. Again, analyses focused on the amygdala, and semantic abstraction could not be assessed because stimuli consisted of pictures of faces with gazes pointed in different directions.
Furthermore, the notion of an all-or-nothing response behaviour as implied in earlier studies (for example, [13,20]) should be critically reevaluated. Obviously, response behaviour strongly depends on the exact definition of the statistical response criterion employed. Previous studies have used a rather conservative response criterion and tended to regard any activity not meeting this criterion as background noise [12,13,20]. Our analyses demonstrate that even after excluding all neurons that showed statistical responses to any of the presented stimuli, semantic category information is still present in the population activity of the ‘nonresponsive’ neurons. Thus, such subthreshold responses according to this criterion are likely to carry relevant information about the presented stimulus. For example, looking at Fig 1 A and 1C, we see such subthreshold responses. Here, the units clearly prefer stimuli from one category (for example, clothing items in case of 1A). Within this category, however, some images drive spiking activity more strongly than others. The jean jacket in Fig 1A is the fifth-most response-eliciting stimulus for that unit but falls short of being classified as a response by the criterion we use, as indicated by the absence of a grey box around the respective raster plot. In view of the other response-eliciting stimuli, we would probably conclude that this might be a true but subthreshold response. Arguably, there are some units in the data set for which we find only such subthreshold responses because the near-optimal stimuli for these units were not in our set. It thus seems that these subthreshold units carry a significant amount of categorical information at the population level. Together, these results suggest that neurons do not encode the identity of a concept in an all-or-none fashion but rather that firing patterns may be best described as graded with the assumption of an underlying ‘semantic tuning curve’.
The high levels of abstraction in population activity observed in this study could also suggest a single-unit mechanism in the MTL for the generation of false memories. Classically, false memories are studied by presenting semantically related words for study, for example, ‘giraffe’, ‘lion’, ‘elephant’, or ‘tiger’, followed by a recognition memory test requiring old–new judgments of old words (for example, ‘lion’), as well as new words that were either semantically related (‘leopard’) or unrelated (‘keyboard’) to the studied words [22]. False memories manifest in more frequent old judgments to new words with high versus low semantic relatedness [22,23]. Overlap of recruited neocortical regions corresponds to semantic feature overlap between studied and new words, which, in turn, is correlated with false-memory likelihood [24]. However, it seems likely that overlap in recruitment of neocortical regions is in fact the consequence of ‘false’ reinstatement initiated by the hippocampus rather than the cause of false memories [24,25]. The hippocampus has been shown to be equally active during false and true memories in humans [26], and optogenetic activation of neurons in the rodent hippocampus has been shown to trigger reinstatement of ‘false’ contextual fear memories [25]. Our data suggest that confusion between semantically similar stimuli is facilitated by the abstract semantic code utilised by neurons in the hippocampus, and thereby provides a link between human behavioural and functional magnetic resonance imaging versus rodent optogenetic studies of false-memory generation [22,24–26].
The combination of RSA and pattern classification applied to our single neuron data reveals novel insights about the neuronal code for semantics in the MTL. Although we think that the decoding of semantic generalisation (top row of Fig 4) and the RSA analyses (Fig 3) convey similar aspects of the data, the decoding results are by no means a trivial consequence of the RSA analyses. First, the decoding analyses allow for a comparison of decoding accuracy for exemplar versus category decision. Second, the fact that confusions within category are more frequent than those across category offers a mechanistic explanation for the generation of false memories. Both of these points do not become apparent from the RSA results alone. These RSA results, in turn, show higher-order organising principles of semantic information in populations of single neurons in the MTL.
Our study also contributes to the understanding of neuronal representations in the amygdala. We found a preference of amygdala units for stimuli depicting food items, which dovetails with findings of a potential role of the amygdala in modulating food consumption recently reported in rodents [27] and with views of the role of the amygdala in processing positive and negative value as well as relevance of stimuli [28,29]. However, human amygdala units have also been shown to preferably respond to animals [17], to be involved in processing of faces and parts of faces [30,31], and to encode the intensity of emotion in facial expressions [32]. More generally, the amygdala has been hypothesised to be involved in social cognition [31]. It is noteworthy that we do not see a preference for stimuli depicting animals in the amygdala as reported by Mormann and colleagues (2011) [21]. Response probabilities of animal stimuli in our study are comparable to this study (approximately 1%). Mormann and colleagues (2011), however, compared animal stimuli to pictures of persons, landmarks, and objects, which all had significantly lower response probabilities (approximately 0.2%). Thus, we may not see a preference for animals because the categories to which we compare them (for example, food, plants, musical instruments, etc.) are different. It may help to reconcile this broad range of findings to consider that the amygdala is a complex and heterogeneous structure consisting of multiple nuclei involved in a wide range of different functions [33] and that the exact location of microwires with respect to these nuclei cannot be determined with sufficient accuracy in human subjects.
Finally, our data connect to notions of hierarchical processing within the MTL. Strong tuning to highly abstract semantics has been found in the hippocampus and the amygdala. Both regions receive highly processed, supramodal input [12,33,34]. The use of a highly abstract semantic code appears plausible to aid in attributing value and relevance of stimuli, a function hypothesised to occur in the amygdala [28]. In the hippocampus, high levels of abstraction may facilitate efficient and sparse representations of large amounts of information encoded in neocortical firing patterns for subsequent encoding of episodic memories [35–37]. In contrast, abstract semantic representations were less pronounced in parahippocampal and entorhinal neurons. This finding connects with views that these structures are situated at a lower stage of the processing hierarchy within the MTL [12,34,38]. Here, the parahippocampal cortex acts as an input region for higher MTL regions. Parahippocampal neurons fire earlier, less selectively than in other MTL regions [12], and display a preference for images with spatial layout of visual input [10]. Similarly, the entorhinal cortex relays reciprocal connections between hippocampus and neocortex [34] and has also been found to be involved in spatial processing in humans [39,40].
A total of 25 epileptic patients (9 female) aged 19 to 62 y (M = 38, SD = 13) were implanted with depth electrodes for chronic seizure monitoring. Their average stay on the monitoring ward was 7 to 10 d.
The study was approved by the Medical Institutional Review Board of the University of Bonn (accession number 095/10 for single-unit recordings in humans in general and 245/11 for the current paradigm in particular) and adhered to the guidelines of the Declaration of Helsinki. Each patient gave informed written consent.
One hundred images from 5 man-made and 5 natural categories of 10 exemplars each were selected as stimuli. The experiment was subdivided into 10 runs. One run entailed sequential presentation of all 100 images in the set in pseudorandom order. A trial entailed the presentation of a blank screen for a variable duration (200–400 ms) and a fixation dot for 300 ms, followed by the image that stayed on screen until the subject responded with a button press. Subjects were instructed to press the left or right arrow key if the image on the screen depicted a man-made or natural object, respectively.
Nine microwires (8 high-impedance recording electrodes, 1 low-impedance reference; AdTech, Racine, WI) protruding from the shaft of the depth electrodes were used to record signals from MTL neurons. Signals were amplified and recorded using a Neuralynx ATLAS system (Bozeman, MT). The sampling rate was 32 kHz, and signals were referenced against one of the low-impedance reference electrodes. Spike sorting was performed using wave_clus [41] in 33 sessions and using Combinato (https://github.com/jniediek/combinato) [42] in 26 sessions. Different spike-sorting routines were used as the reported paradigm also served as a procedure to screen for response-eliciting stimuli in the morning of a day of testing. Therefore, manual optimisation of spike sorting was performed immediately after recording. The lab as a whole switched to using Combinato for reasons unrelated to the reported research.
A total of 5,033 units resulted from spike sorting, 4,917 of which were recorded in one of the anatomical regions considered (amygdala, hippocampus, entorhinal cortex, and parahippocampal cortex). The number of microwires per patient was on average 71.60 (SD = 21.32) and ranged from 32 to 96. On average, we recorded 1.38 units per microwire (SD = 0.44). These values ranged from 0.41 to 2.24 across all 59 sessions.
To determine whether a unit responded with increased spiking activity to one of the stimuli in the set, we calculated a binwise rank-sum test described earlier [12]. We obtained spike counts in 19 overlapping 100 ms bins ([0:100:1,000] and [50:100:950] ms after stimulus onset) for each trial in which a given image was presented. We computed 19 rank-sum tests, each of which compared the distribution of spike counts of one of the 19 bins against the distribution of spike counts in a baseline interval (−500 to 0 ms) of all trials in a session. The resulting 19 p-values were corrected for multiple comparisons using the Simes procedure. A stimulus was classified as eliciting a neuronal response in a unit when one or more of these 19 p-values was lower than α = 0.001. Furthermore, we considered only increases in firing rates. Also, neuronal responses were only considered as such if at least one spike in the response period was recorded in more than 5 out of the 10 trials per image and if the average firing rate during the response window (0 to 1,000 ms) was above 2 Hz.
We counted the neuronal responses across all sessions, separate for superordinate category and anatomical location. To make these values comparable across anatomical regions and with previous work [17], we calculated response probabilities by normalising these counts to the number of units in an anatomical region and the total number of stimuli presented (100). Response probabilities were calculated for each of the four anatomical regions of interest. They thus represent the empirical probability that a unit in a given anatomical region will respond to a stimulus from a given semantic category.
We obtained measures of dispersion of these response probabilities by using a subsampling procedure. We drew 2,000 random subsamples of 700 units without replacement from each region and derived 95% confidence intervals from the resulting distributions of response probabilities for each category of stimuli.
A Fisher’s exact test on the response probabilities was conducted for each category and each anatomical region. To this aim, data were arranged in a 2 × 2 contingency table of the frequencies of significant and nonsignificant neuronal responses in a superordinate category of interest, and the frequency of significant and nonsignificant neuronal responses in all other superordinate categories.
To assess the dissimilarity between neuronal representations of stimulus categories, firing rates during the response period (0 to 1,000 ms after stimulus onset) of each stimulus were expressed as Z scores using the mean and standard deviation of firing rates in a base line interval ranging from −500 ms to stimulus onset (0) across all trials. These Z scores were arranged in a matrix of NS × NU, where NU is the number of units recorded and NS the number of stimuli in the set (100). Representational dissimilarity between a pair of stimuli was calculated using 1 –Pearson’s correlation coefficient (1 − R) of the vectors of Z scores corresponding to the population activity evoked by the two stimuli in a pair [9,17]. To assess representational dissimilarity on the level of individual trials, we computed Z scores for each trial in the experiment. These Z scores were arranged in a matrix of NT × NU, where NU is the number of units recorded and NT the number of trials during the paradigm (1,000).
Hierarchical clustering for dendrograms in Fig 3 was performed using unweighted average distance method on correlation distances.
We used the matrices of Z scores described above (NT × NU) to assess pattern classification performance. We used the function fitcecoc.m from MATLAB’s (MathWorks; www.mathworks.com) statistics and machine-learning toolbox. This function was used to train a multiclass, error-correcting output codes model of linear support vector machines for binary choices. Binary support vector machines were specified according to a ‘one versus all’ coding scheme in which for each binary classifier, one class is positive and the rest are negative. The classifier was trained to predict the label of stimulus identity from individual trials (NT × NU). Out-of-sample performance was assessed for 100 pseudorandom divisions of the data into training and test set (50% holdout for test). To test for semantic generalisation to ‘unseen’ members of category, further classifiers were trained on the mean responses (NS × NU) of half of the stimuli to learn category labels and tested on the other half of stimuli. Again, out-of-sample performance was assessed for 100 pseudorandom divisions of the data into training and test set. Classification performance was quantified by Cohen′sκ=PO−PC1−PC, where PO is the observed agreement and PC is chance agreement. S4 Fig and S5 Fig show these same analyses repeated separately for each subject but collapsing across regions.
To assess whether dissimilarity (1 − R) was significantly different within versus across exemplars of superordinate categories, we implemented a label-shuffling procedure. To this aim, we arranged dissimilarity between all pairs of stimuli in matrices of the format NS × NU. Next, we selected a set of indices to the elements in these matrices that correspond to within-category dissimilarity. Another set of indices was selected corresponding to between-category dissimilarity. We then computed a Mann-Whitney U test with the hypothesis that within-category dissimilarity is lower than between-category dissimilarity. From this test we obtained a test statistic (rank-sum) of the original assignments of the labels (within- versus between-category dissimilarity) to the data. We repeated this test 105 times with randomly shuffled assignments of labels to the data, that is, indices to the matrix corresponding to within- versus between-category pairs were randomised and hence mostly false. Of these 105 tests with random labels, we saved the distribution of resulting test statistics (rank-sums). The reported p-values reflect the percentile of the test statistic that got the correct assignments of labels to the data within the distribution of test statistics derived with randomly relabelled data. The same procedure was carried out for the confusion matrices derived from pattern classification. Note that dissimilarity matrices were symmetric, whereas confusion matrices were not. We therefore computed statistics for dissimilarity on the triangular matrices only.
We used MATLAB and its statistics and machine-learning toolbox in combination with custom code for analyses of the data. Spike sorting of 33 sessions was done using wave_clus (https://github.com/csn-le/wave_clus) [41]. The remaining 26 sessions were sorted using Combinato [42] requiring Python (www.python.org). We used the psychtoolbox3 (www.psythoolbox.org) and octave (www.gnu.org/octave) running on a Debian 8 operating system (www.debian.org) on a standard laptop computer for stimulus delivery. All relevant data and custom code are available on https://github.com/rebrowski/abstractRepresentationsInMTL.git.
|
10.1371/journal.pntd.0004960 | Antigenic Variation of East/Central/South African and Asian Chikungunya Virus Genotypes in Neutralization by Immune Sera | Chikungunya virus (CHIKV) is a re-emerging mosquito-borne virus which causes epidemics of fever, severe joint pain and rash. Between 2005 and 2010, the East/Central/South African (ECSA) genotype was responsible for global explosive outbreaks across India, the Indian Ocean and Southeast Asia. From late 2013, Asian genotype CHIKV has caused outbreaks in the Americas. The characteristics of cross-antibody efficacy and epitopes are poorly understood.
We characterized human immune sera collected during two independent outbreaks in Malaysia of the Asian genotype in 2006 and the ECSA genotype in 2008–2010. Neutralizing capacity was analyzed against representative clinical isolates as well as viruses rescued from infectious clones of ECSA and Asian CHIKV. Using whole virus antigen and recombinant E1 and E2 envelope glycoproteins, we further investigated antibody binding sites, epitopes, and antibody titers. Both ECSA and Asian sera demonstrated stronger neutralizing capacity against the ECSA genotype, which corresponded to strong epitope-antibody interaction. ECSA serum targeted conformational epitope sites in the E1-E2 glycoprotein, and E1-E211K, E2-I2T, E2-H5N, E2-G118S and E2-S194G are key amino acids that enhance cross-neutralizing efficacy. As for Asian serum, the antibodies targeting E2 glycoprotein correlated with neutralizing efficacy, and I2T, H5N, G118S and S194G altered and improved the neutralization profile. Rabbit polyclonal antibody against the N-terminal linear neutralizing epitope from the ECSA sequence has reduced binding capacity and neutralization efficacy against Asian CHIKV. These findings imply that the choice of vaccine strain may impact cross-protection against different genotypes.
Immune serum from humans infected with CHIKV of either ECSA or Asian genotypes showed differences in binding and neutralization characteristics. These findings have implications for the continued outbreaks of co-circulating CHIKV genotypes and effective design of vaccines and diagnostic serological assays.
| Chikungunya virus (CHIKV) has caused large epidemics of fever, rash, and joint pain around the world in recent years. Three different CHIKV genotypes exist. Infection with one genotype is likely to lead to immune protection (or cross-protection) against future infections with a different genotype. However, little is known about the nature of this cross-protection. In this study, we used serum from Malaysian patients infected with CHIKV of either Asian or East/Central/South African (ECSA) genotypes. We compared the ability of the serum antibodies to bind to and neutralize two different viruses, from either Asian or ECSA genotypes. We found that both Asian and ECSA serum were more effective in binding and neutralizing ECSA virus. We identified the key amino acids/epitopes within the E1-E2 surface glycoprotein, and showed that variation of these impacts the efficacy of antiserum in cross-neutralizing different genotypes of CHIKV. We showed how sequence variation of a known linear neutralizing epitope could alter the cross-neutralization efficacy. This study aids understanding of the importance of different circulating genotypes within a country and has implications for the design of vaccines and diagnostic antibody tests.
| Chikungunya virus (CHIKV) is a re-emerging, mosquito-borne arbovirus which has caused unprecedented worldwide epidemics in recent years [1]. There are three major CHIKV genotypes circulating: West African, East/ Central/ South African (ECSA) and Asian [2]. After the global outbreaks of ECSA between 2005 and 2010, the Asian genotype has re-emerged to cause large outbreaks in the Americas and the Pacific islands [3, 4]. Malaysia has experienced CHIKV outbreaks due to two different genotypes, Asian and ECSA. The endemic Asian CHIKV strain was responsible for small, geographically-restricted outbreaks in 1998 and 2006 [5–7]. An imported ECSA outbreak was reported in 2006 prior to an explosive nationwide outbreak which affected over 15,000 people across different states in 2008 [8, 9].
CHIKV is an alphavirus from the family Togaviridae. A CHIKV virion is 60-70nm in diameter, with a single-stranded positive RNA genome of approximately 11.8 kb in a capsid with a phospholipid envelope carrying glycoproteins E1 and E2. Its genome has 2 open reading frames encoding the non-structural (nsP1-nsP2-nsP3-nsP4) and structural polyproteins (C-E3-E2-6K-E1) [10]. The E1 and E2 glycoproteins form heterodimers which enable interaction with cellular receptors and fusion of the virion envelope with the cell membrane to initiate infection [11], while the capsid protein is required during virus assembly [12]. These proteins are highly immunogenic, and most CHIKV-infected patients develop antibodies targeting the structural proteins (particularly E2) and, to a lesser extent, nsP3 [13, 14]. After the initial induction of type I interferon [15], CHIKV-specific antibodies have been shown as the major effector in immunity to control infection [16]. Among other immune factors, T cells may play a secondary role in suppressing infection [17], although others have found that CD4+ T cells are more important in orchestrating joint inflammation [18].
Currently, treatment for CHIKV is supportive and no licensed vaccine or antiviral are available. Phase I clinical trials have demonstrated the safety and efficacy of vaccination with virus-like particles using structural proteins derived from the West African genotype [19], and a recombinant measles virus-based CHIKV vaccine derived from the ECSA genotype [20]. Cross-reactivity can be achieved against heterogenous genotypes, by which CHIKV seropositive individuals infected with either ECSA or Asian CHIKV have cross-protection against both CHIKV genotypes [9]. However, the cross-neutralizing efficacy of CHIKV-specific antibodies against Asian and ECSA genotypes, which are both currently circulating in Malaysia, Brazil [21] and the Asian region [22], is poorly understood. A distinct antigenic relationship has been established between West African and ECSA genotypes, in which mice and hamsters immunized with the ECSA genotype had 4- to 8-fold differences in neutralizing capacity when tested against a West African strain [2]. In a Singaporean cohort, CHIKV-immune sera exhibited differential antibody binding and neutralizing capacity against isolates with a naturally occurring K252Q amino acid change in the E2 glycoprotein [14]. Given the ability of CHIKV to rapidly spread across different parts of the world with displacement of one genotype with another, the understanding of cross-neutralizing antibody and antigenic variation of different genotypes will have implications for both continued outbreaks and vaccine development.
In this study, we analyzed the neutralizing capacity of CHIKV ECSA and Asian immune sera against representative clinical isolates and rescued viruses of ECSA and Asian CHIKV. We demonstrated that both sets of serum panels have stronger neutralizing capacity against the ECSA isolate, which corresponded to strong epitope-antibody interaction. E1-E211K enhances the neutralization activity of ECSA serum, while E2-I2T, H5N, G118S and S194G within linear epitopes improve the neutralization activity of both sets of sera panels. Rabbit polyclonal antibody targeting a known linear neutralizing epitope (LP1) from ECSA virus could only neutralize homotypic virus, but not heterotypic Asian virus due to sequence variation. These findings indicate the antigenic variation of ECSA or Asian CHIKV genotypes in naturally-acquired infection alters the spectrum of cross-genotype protective antibody immunity.
This study included 63 human samples from two independent outbreaks in Malaysia. The Asian serum panel comprised 40 samples collected from patients 11–14 months after an Asian CHIKV outbreak in Bagan Panchor in 2006 [7]. The ECSA serum panel consisted of 23 samples from patients infected by ECSA strains in 2008–2010, collected 1–6 months after onset of symptoms, who were seen at the University Malaya Medical Centre in Kuala Lumpur [9]. Healthy controls (n = 15) with no past infection of CHIKV served as negative controls. Serum neutralization assay was performed on all the sera. To determine the neutralizing activity due to IgG, heat-inactivated sera were treated for 1 hour with dithiothreitol (DTT) (Life Technologies) at a final concentration of 5mM at 37°C.
This study was approved by the Medical Ethics Committee of the University Malaya Medical Centre (reference no. 800.70). Our institution does not require informed consent for retrospective studies of archived and anonymized samples.
Baby hamster kidney (BHK-21) cells (ATCC no. CCL-10) were maintained in Glasgow minimum essential medium (GMEM) (Life Technologies) supplemented with 5% heat-inactivated fetal bovine serum (Flowlab), 10% tryptose phosphate broth, 20mM HEPES, 5mM L-glutamine, 100 U/ml penicillin and 100μg/ml streptomycin. Infected cells were maintained in GMEM containing 2% FBS. The clinical isolates used, which have been previously characterized [23], were MY/06/37348, an Asian genotype strain isolated from a patient in Bagan Panchor in 2006 (accession number FN295483), and MY/08/065, an ECSA virus isolated from a patient in Kuala Lumpur in 2008 (accession number FN295485). Both isolates had been passaged two times in Vero cells (ATCC no. CCL-81) before propagated in BHK-21 cells. Virus passage (P3) of clinical isolates was used for subsequent work.
To study the neutralizing epitopes, viruses rescued from two different infectious clones, derived from ECSA and Asian genotypes of CHIKV, were included. The plasmid vectors capable of producing infectious viruses were constructed under the control of the human cytomegalovirus immediate-early promoter. The CHIKV infectious clone derived from the ECSA genotype was based on LR2006-OPY1, isolated in Reunion Island in 2006,and has been described previously [24]. The full-length infectious cDNA (icDNA) clone from the Asian genotype was engineered by gene synthesis and assembled by the restriction enzymes approach based on the consensus sequence for strain 3462, isolated in Yap State in 2013 (accession no. KJ451623); however, the protein coding regions in the non-structural and structural proteins were changed to be identical to isolate CNR20235 from the Caribbean outbreak, which was isolated in Saint Martin Island in 2013 (http://www.european-virus-archive.com/article147.html). Both molecular clones have ZsGreen gene incorporated as reporter and duplication of the subgenomic promoter. The ECSA molecular clone was named “ICRES1”, while the Asian molecular clone was designated as “CAR”.
For construction of the chimeric viruses, the ectodomain regions of envelope glycoprotein genes E1 (amino acids 1–381) and E2 (amino acids 1–341) in the ICRES1 backbone were replaced with those of Semliki Forest virus (SFV) E1 (amino acids 1–381) and E2 (amino acids 1–340) from icDNA SFV6 [25] using NEBuilder HiFi DNA Assembly Master Mix (NEB). In order to study the effects of point mutations on the neutralizing epitopes, conventional PCR-based site-directed mutagenesis was performed on the CAR construct using Q5 High-Fidelity DNA polymerase (NEB) with designed primers (S1 Table). The sequences of all the constructs were verified by control restrictions and sequence analysis. Primers and sequences for infectious clone constructions are available upon request.
The viruses were rescued from icDNA by electroporation. Stocks of rescued viruses (P0) were harvested and titrated by plaque assay on BHK-21 cells. To obtain P1 stocks, confluent BHK-21 cells grown in T75-cm2 flasks were infected with P0 stocks at a multiplicity of infection (MOI) of 1 plaque forming unit/cell and maintained in 2% FBS GMEM. P1 stocks were harvested after 24 or 48 hours, titrated and used for the neutralization assay. Infectious center assay was performed on all the viruses rescued from icDNA. Details on virus rescue and related protocols are shown in S1 Text and S2 Table.
Seroneutralization was performed with a previously described immunofluorescence-based cell infection assay in BHK-21 cells [26, 27], with minor modifications. The DTT-treated sera underwent 2-fold serial dilutions (1:100 to 1:6400) in 1X Dulbecco’s PBS prior to mixing with CHIKV pre-diluted with 2% FBS GMEM. Cells were infected with clinical isolates at an MOI of 10. The virus-antibody mixture was incubated for 2 hours at 37°C before inoculation into 104 cells in 96-well CellCarrier-96 optic black plates (Perkin Elmer), and further incubated for 1.5 hours at 37°C. The inocula were decanted and 2% FBS GMEM was added to the plates. The plates were fixed with 4% paraformaldehyde after 6 hours of incubation at 37°C, permeabilized with 0.25% Triton X-100 for 10 minutes, and immunostained using anti-CHIKV E2 monoclonal antibody B-D2(C4) [28] at 1μg/ml followed by rabbit anti-mouse IgG-FITC (Thermo Scientific) at 1:100 dilution. Cell nuclei were counter-stained with DAPI. Fluorescence intensity was analyzed with a Cellomics High Content Screening (HCS) ArrayScan VTI (Thermo Fisher) over 9 different fields at 5X magnification. Percentage of infectivity was calculated according to the following equation: % infectivity = (mean average fluorescence intensity from serum sample/mean average fluorescence intensity from virus control) × 100. The neutralizing titer (NT50) was expressed as the serum dilution that reduced infectivity by 50% using non-linear regression fitting in GraphPad Prism 5 (GraphPad Software).
For seroneutralization using rescued viruses, diluted sera were mixed with viruses pre-diluted with 2% FBS GMEM (with infection performed at an MOI of 50), followed by the steps described above. The plates were fixed after 7 hours of incubation at 37°C. The plates were only counter-stained with DAPI prior to acquisition of ZsGreen fluorescence.
To investigate the cross-reactivity of CHIKV sera against another alphavirus, SFV was rescued from icDNA SFV6 as previously described [25]. Diluted sera (1:25 and 1:100 dilutions) were mixed with SFV pre-diluted with 2% FBS GMEM (with infection performed at an MOI of 10), followed by the steps described above. The plates were fixed after 6 hours of incubation at 37°C, and stained with mouse anti-alphavirus monoclonal antibody (Santa Cruz) at 1:100 dilution.
To investigate the effect of sequence variation of neutralizing epitopes in ECSA and Asian genotypes, polyclonal rabbit anti-LP1 (STKDNFNVYKATRPY), anti-LP1A (SIKDHFNVYKATRPY) and anti-LP47 (NHKKWQYNSPLVPRN) were produced commercially (GenScript). LP1 is similar to E2EP3, an immunogenic peptide (from an ECSA virus) previously reported to elicit neutralizing antibodies [26]. LP1A is the corresponding variant peptide with Asian genotype sequences. The LP47 peptide sequence is conserved in both genotypes. Seroneutralization was performed with purified antibody at 25μg/ml against the rescued viruses.
For indirect IgG ELISA (antibody end-point assay) and Western blot, the antigen was partially purified virus prepared by sucrose-cushion ultra-centrifugation, treated with 1% Triton X-100 in TE buffer, clarified by centrifugation, and stored in 50% glycerol at -20°C.
For production of native recombinant proteins of E1 (rE1, from amino acids 1–412) and E2 (rE2, from amino acids 1–362), viral RNA was extracted from clinical isolates (Asian MY/06/37348 and ECSA MY/08/065). cDNA was synthesized using reverse-transcription, and the genes were amplified using high fidelity Platinum Taq (Invitrogen) with designed primers (S3 Table). The transmembrane regions and cytoplasmic tails of the glycoproteins were not included in the expression cassette, to ensure solubility of the recombinant proteins. The amplicons were ligated into a pIEX-5 vector (Novagen) directionally at BamH1 and Not1 restriction sites. Each plasmid construct together with a pIE1-neo vector were co-transfected into Sf9 cells (Novagen) using Cellfectin II reagent (Invitrogen) [29]. Stable clones expressing rE1 and rE2 were generated under selection with G418 sulfate at 1000μg/ml. The proteins secreted from stable clones were purified under native conditions with activated Profinity IMAC resins (Bio-Rad) or HisTrap FF (GE). The eluates were concentrated with an Amicon centrifugal unit and the buffer was exchanged with sodium phosphate buffer (50mM NaH2PO4, 300mM NaCl, pH 8.0). The proteins were stored at -20°C in 50% glycerol, except for the proteins used in the competitive protein blocking assay, which were filter-sterilized and kept at 4°C. Fusion sequences expressing for rE2 and rE1 was generated by overlapping PCR; recombinant proteins encoded by obtained sequence were linked via linker with sequence GGGS-His (8X)-GGGG (S1 Text). The fusion glycoprotein constructs were transfected into TriExSf9 cells (Novagen) by TransIT-Insect transfection reagent (Mirus Bio).
The proteins were resolved with 12% SDS-PAGE under reducing and non-reducing conditions and electro-transferred onto a nitrocellulose membrane (GE). The membrane was blocked with 10% skimmed milk in 0.05% PBS-Tween 20 (PBST). The immunoreactivity of recombinant proteins was evaluated with pools of CHIKV immune sera applied at indicated dilutions in the blocking buffer. The bound antigen-antibody complex was detected by anti-human IgG-HRP (DakoCytomation) at 1:5000 dilution in 1% bovine serum albumin (BSA)-0.05% PBST. The membrane was visualized by chemiluminescence (Bio-Rad) and images were acquired with a BioSpectrum AC imaging system (UVP). Mouse anti-His tag antibody (Merck Millipore) was included as a loading control. Mouse anti-E2 monoclonal antibody (clone: B-D2(C4); EIEVHMPPDT) [28] was also included as a control.
All incubation steps were performed at 37°C for 1 hour, using 1% BSA-0.05% PBST as diluent for serum and antibodies. The plates were washed 4 times with 0.05% PBST after each incubation step. To determine the relative level of anti-E2 antibodies, the plates were coated with 250 ng of virus antigen or 100 ng of rE2 in 0.05M carbonate-bicarbonate buffer (pH 9.6). The antigens were normalized with monoclonal antibody B-D2(C4) to determine the relative level of anti-E2 antibodies. The plate was blocked with 3% BSA in 0.05% PBST. The sera were tested at 2-fold serial dilutions from 1:512 to 1:1,048,000 or 1:640 to 1:655,000. The IgG end-point titer was determined as the reciprocal of the highest dilution that produced an optical density (OD) reading of three times greater than that of the negative control. Anti-human IgG-HRP at 1:5000 dilution was added to detect the bound antibodies. TMB substrate (KPL) was added to each well and the plates were incubated at room temperature for 5 min. The reaction was terminated by adding 1M phosphoric acid. The absorbance was measured at 450nm with 630nm as the reference wavelength using an automated ELISA reader (Biotek Instruments). The cut-off value was established as the OD obtained from healthy controls sera plus three standard deviations (SD). The relative level of anti-rE2 antibodies was calculated with the following formula: (end point titer for rE2/end point titer for whole virus antigen) × 100.
Soluble recombinant CHIKV proteins (15μg) were mixed with heat-inactivated immune sera diluted at 1:200, and incubated for 1 hour at 37°C. CHIKV (MY/08/065) in amounts corresponding to an MOI of 10 was mixed with the samples, which were incubated for a further 2 hours at 37°C.
Synthetic peptides were obtained from GenScript (LP1, STKDNFNVYKATRPY; LP24, TDSRKISHSCTHPFH; LP38, GNVKITVNGQTVRYK); 60μg of each peptide was mixed with immune sera diluted with 1X DPBS at 1:100 and incubated for 1.5 hours at 37°C. All the synthetic peptides for the blocking assay have a purity grade greater than 95% and are soluble in high-grade water. ICRES1 (sucrose-cushion purified virus in TE buffer pre-diluted using 2% FBS GMEM) at an amount corresponding to an MOI of 1 was mixed with the samples, which were incubated for a further 2 hours at 37°C prior to infection of BHK-21 cells. The plate was replenished with plaque medium (2% FBS GMEM containing 0.8% of carboxymethylcellulose), fixed with 4% paraformaldehyde after 15 hours of incubation, and this was followed by ZsGreen fluorescence acquisition. Infectivity corresponded to the fluorescence intensity acquired with a Cellomics HCS reader. The effect on infectivity of antibodies in the presence and absence of blocking peptides was compared.
Biotinylated synthetic peptides covering the E2 glycoprotein sequence from amino acids 1–362 from a previous study [28] were used to screen CHIKV immune sera for binding to linear epitopes. The length of each peptide is 15-mer with a 10-mer overlap based on the CHIKV MY/08/065 sequence (accession no. FN295485; S4 Table). Similar steps were performed as described above except that the plates were washed 6 times after incubation with human sera and secondary antibody. The plates were coated with 20μg/ml streptavidin (NEB) and blocked with 5% BSA-PBST. The dissolved peptides in dimethyl sulphoxide were further diluted to a working concentration of approximately 150μg/ml in 1% BSA-PBST. CHIKV immune sera and healthy control sera were diluted at 1:1000, and screened against peptides in duplicate. The peptides with the highest OD reading from 2 adjacent overlapping synthetic peptides were considered as identified B-cell epitopes. Computational analysis and epitope localization were performed on structural data retrieved from Protein Data Bank (PDB, ID 3J2W) with UCSF CHIMERA software [30]. As the LP1 sequence is unresolved in structural data, the structure of the E2 glycoprotein was predicted using the online I-TASSER server [31, 32]. The electrostatic potential of the E2 structure (amino acid 1–362) was evaluated with PDB2PQR and APBS [33–35].
Data are presented as means ± SD or means ± standard error of the mean (SEM). Differences between groups and controls were analyzed using appropriate statistical tests. A P-value of <0.05 was considered significant. Statistical analyses were performed with GraphPad Prism 5.
We employed a sensitive seroneutralization assay to compare the neutralizing capacity of the different sera panels against MY/08/065 (ECSA) and MY/06/37348 (Asian) (S1 and S2 Figs). The heat-inactivated intact sera and DTT-treated sera had similar neutralizing capacity against MY/08/065 for both sera panels (S3 Fig). ECSA sera demonstrated strong neutralizing capacity against homotypic CHIKV compared to heterotypic CHIKV (Fig 1A), with a NT50 against MY/08/065 that was a median 2.67 (range, 1.40–4.61) times greater than the NT50 against MY/06/37348 (Fig 1B). Unexpectedly, Asian sera demonstrated better neutralizing capacity against heterotypic ECSA CHIKV compared to homotypic CHIKV (Fig 1A), with a NT50 against MY/08/065 of a median 1.44 (range, 0.70–3.19) times greater than the NT50 against MY/06/37348 (Fig 1B). The greater neutralizing capacity corresponded to stronger antibody binding to MY/08/065 compared to MY/06/37348 by quantitative ELISA (Fig 1C). Seroneutralization was performed against rescued virus from icDNA of ECSA and Asian genotypes. Both ECSA and Asian sera demonstrated better neutralizing capacity against ICRES1 (ECSA) compared to CAR (Asian) (Fig 1D). Immunoblotting showed stronger reactivity of serum with the whole viral antigen (with a band of about 50kDa, consistent with E1 or E2, a known immunodominant antigen in alphaviruses) and recombinant E2 glycoprotein of similar size derived from ECSA, compared to the Asian genotype. Under non-reducing conditions, ECSA sera had stronger antibody binding to its homotypic CHIKV isolate MY/08/065, ICRES1 and recombinant E2 glycoprotein (rE2) (Fig 1E and 1F). Asian sera bound similarly to both genotypes of viruses (clinical isolates and rescued viruses), and more strongly to rE2 glycoprotein of MY/08/065 (Fig 1E and 1F). Under reducing conditions, both sets of sera retained stronger binding to ECSA CHIKV and rE2 of ECSA CHIKV. Both sets of sera had a similar proportion of total antibodies binding to rE2 (median 50%, range 20–63% for ECSA serum; median 50%, range 16–63% for Asian serum) (Fig 1G), and these percentages suggest that antibodies also target sites other than E2. Taken together, CHIKV serum shows strong neutralizing capacity and binding to CHIKV, particularly of the ECSA genotype, and the epitopes may be presented as part of the conformational E1-E2 glycoprotein and/or as linear determinants in the E2 glycoprotein.
To determine if CHIKV immune serum targets E1, recombinant E1 glycoprotein (rE1) was probed in ELISA with serially diluted pooled sera, and signal was detected at low serum dilutions from 1: 100 to 1:800 (Fig 2A). A competitive protein blocking assay was performed, and blocking of ECSA and Asian sera with native rE1 alone did not significantly alter the neutralizing capacity (Fig 2B). However, when the sera was blocked by a mixture of rE1 and rE2, significant increases of infectivity were observed in both panels of sera compared to unblocked sera or sera blocked by either rE1 or rE2 alone. We then hypothesized that antibodies may target conformational epitopes on E1 and E2 glycoproteins together. To test this hypothesis, we constructed 2 chimeras which swapped the ecto-domain regions of the E2 and E1-E2 glycoproteins with those of Semliki Forest virus (SFV). Both sets of sera demonstrated a low degree of cross-neutralization against SFV, another alphavirus which is a member of the same antigenic complex as CHIKV (S4 Fig). At 1:100 serum dilution, loss of neutralizing effect for both sets of sera was observed when CHIKV E2 was replaced with SFV E2. Furthermore, in ECSA serum, loss of neutralization activity was much higher against the chimera with E1-E2 from SFV compared to the chimera with SFV E2 alone (Fig 2C). This provides further evidence that neutralizing antibodies are not solely targeting E2, but are also targeting epitopes spanning E1-E2 glycoproteins. Alternatively, E1 may affect the conformation of E2 and alter its epitopes.
To further determine the importance of conformational epitopes resulting from interactions between E1 and E2, four sets of fusion E1-E2 glycoproteins were constructed. Each hybrid fusion protein contained E1 and E2 sequences from either MY/06/37348 (ECSA) or MY/08/065 (Asian), transiently expressed as secreted native recombinant proteins in insect cells (Fig 3A). The antibody binding capacity of ECSA sera against fusion E1-E2 proteins significantly increased when either the E1 or E2 sequence was changed from that of MY/06/37348 to that of MY/08/065, as shown in immunoblotting (Fig 3B) and quantitative ELISA (Fig 3C). Asian sera had almost equal antibody binding capacity for the 4 fusion glycoproteins, suggesting that Asian serum was not sensitive to sequence changes in E1-E2 glycoproteins. This data shows that the greater binding and neutralization of the ECSA isolate MY/08/065 by ECSA sera (Fig 1A, 1C and 1D) is due to critical conformational epitopes on the E1-E2 heterodimer, which are sequence-dependent.
Between ECSA (MY/08/065 and ICRES1) and Asian (MY/06/37348 and CAR) genotypes of CHIKV in this study, there are 10 amino acids differences in E1 (Fig 3D). Using the fusion rE2-E1-Asian construct as a template, site-directed mutagenesis was performed independently to replace each amino acid of Asian origin with the corresponding ECSA residue, and the proteins were expressed in insect cells. The antibody binding significantly increased with the amino acid changes at A145T, E211K, A226V and M269V, in comparison to hybrid rE2Asian-E1ECSA recombinant proteins (Fig 3E). Recombinant virus carrying E1-211K demonstrated a large increase in neutralizing capacity compared to the parental virus clone (CAR), while the E1-145T change caused a slight decrease in neutralizing capacity (Fig 3F and 3G). The critical 211K amino acid was localized at the surface of E1-E2 heterodimers (Fig 3H).
To study the linear epitopes in the immunodominant E2 glycoprotein (based on strain MY/08/065, of the ECSA genotype), overlapping synthetic peptides covering amino acids 1–362 were mapped by peptide-ELISA using the ECSA and Asian sera (Fig 4A, 4B and 4C). Both ECSA and Asian sera mapped to the same 9 peptides, and the Asian sera mapped to an additional 3 peptides (Table 1). Between the strains of ECSA (MY/08/065, ICRES1) and Asian (MY/06/37348, CAR) genotypes of CHIKV used in this study, there are 15 amino acid differences in E2 (from amino acids 1–362), of which 4 amino acid differences fall within the identified linear epitopes (Fig 4D). Using the rE2-Asian construct as a backbone, site-directed mutagenesis was performed to replace each amino acid of Asian origin with an ECSA residue, and the proteins were expressed in insect cells. The antibody binding significantly increased with I2T, H5N, G118S, R149K and S194G substitutions in comparison to the original rE2-Asian recombinant protein (Fig 4E and S5 Fig). Recombinant viruses carrying either E2-2T, 5N, 118S or 194G demonstrated increases in neutralizing capacity compared to the parental virus clone (CAR), while the E2-R149K change caused a decrease in neutralizing capacity (Fig 4F). Competitive peptide blocking assay indicated that the anti-CHIKV antibodies interact with the LP1, LP24 and LP38 peptides that cover amino acid sites 2, 5, 118 and 194 on E2 (Fig 4G). These 4 neutralizing linear epitopes are localized on the surface of the E1-E2 heterodimer complex (Fig 4H).
As naturally-acquired infection of the Asian genotype of CHIKV leads to higher cross-neutralizing efficacy against ECSA CHIKV, we hypothesized that an epitope-based vaccine derived from the Asian genotype might provide a substantial level of cross-protection against ECSA CHIKV. The peptide LP1 (STKDNFNVYKATRPY) is similar to E2EP3, a peptide derived from ECSA virus which has been found to be highly immunogenic in eliciting neutralizing antibodies in an animal model [26]. We generated a variant, LP1A (SIKDHFNVYKATRPY), derived from the sequence of the Asian virus. Rabbit polyclonal antibodies were commercially prepared against LP1A and LP1. Peptide-ELISA was performed using human ECSA and Asian serum with LP1A and LP1 as antigens. Human ECSA serum bound to LP1 but not LP1A (Fig 5A). Rabbit anti-LP1 antibody showed the lowest binding capacity against CAR (Asian), and demonstrated poor neutralizing activity against the CAR virus harboring the LP1A sequence (infectivity 91±10%, Fig 5B). Anti-LP1 binding capacity and neutralization efficacy was partially restored with the mutations I2T and H5N. The anti-LP1 antibody had maximum binding capacity and neutralizing efficacy against CAR-E2-I2T-H5N (Fig 5B), which has the LP1 sequence; a finding in line with the antibody binding of ECSA immune sera against LP1 peptide (Fig 5A).
Asian serum could recognize LP1A, although binding was marginally higher to LP1 (Fig 5C), which supports the earlier finding that Asian serum has stronger binding against LP1 with I2T and H5N amino acid changes (Fig 4E). Unexpectedly, rabbit anti-LP1A did not demonstrate significant neutralizing activity against CHIKV with either the LP1A or LP1 sequences (Fig 5D). However, a competitive peptide blocking assay indicated that neutralizing antibodies from Asian sera could still recognize and interact with both LP1A and LP1 peptides (Fig 5E). The electrostatic potential of the E2 surface was computed based on the CAR ecto-domain region to study the charge distribution of these epitopes which affect binding affinity [36]. The I2T change leads to higher electrostatic potential, which is associated with improved binding capacity and neutralization efficacy (Fig 5F). LP47, another linear neutralizing epitope in humans, also failed to induce any functional neutralizing antibodies in rabbits.
CHIKV has become a major public health concern worldwide and causes considerable socio-economic burden. Protective adaptive immunity is mainly provided by specific antibodies, particularly those directed against epitopes on the E2 and E1 glycoproteins [37, 38]. Understanding cross-immunity resulting from infections with different genotypes is particularly important and timely. Many Asian countries now have both endemic Asian and epidemic ECSA strains circulating, and the recent widespread outbreaks in the Americas are due to the Asian genotype rather than the previously epidemic ECSA strains, indicating that viruses from both genotypes are capable of global spread.
In this study, we showed differences in cross-genotypic neutralization efficacy of immune sera against ECSA and Asian genotypes of CHIKV. Both ECSA and Asian serum had greater neutralizing capacity against ECSA genotype (MY/08/065 and ICRES1) than Asian genotype (MY/06/37348 and CAR), indicating that neutralizing antibodies regardless of initial infecting genotype preferentially recognized the epitopes presented by the ECSA genotype. The presence of cross-genotype neutralization was clearly shown lasting up to 14 months post-infection. The clinical significance of the differential cross-protective capacity of ECSA and Asian sera remains unclear, as all the immune sera had more than the minimum neutralizing titer (≥10) which appears to correlate with immune protection from symptomatic CHIKV infection in humans [39]. This high degree of cross-neutralization likely contributed to the geographic restriction of CHIKV of different genotypes seen historically, which limited, for example, the spread of ECSA viruses in Asia, at least until CHIKV underwent mutations that facilitated sequential adaptation to the Aedes albopictus vector [40, 41].
Apart from the stronger antigenicity of epitopes of the ECSA genotype, we also showed that neutralizing capacity was also affected by the target and the amount of neutralizing antibodies. Both ECSA and Asian sera contain high levels of neutralizing antibodies to numerous linear epitopes on the E2 glycoprotein as well as conformational epitopes on the E1-E2 heterodimer complex. This supports recent findings that most of the reported CHIKV neutralizing monoclonal antibodies target conformational epitopes on the exposed, topmost outer surfaces of the E2/E1 spike, particularly in domain A and domain B [42–45]. Our findings also suggest that subunit vaccine candidates derived from E1 or E2 glycoproteins alone [46–48] may be insufficient to provide full protection against all genotypes, and that virus-like particle vaccines which present epitopes on E2/E1 in their native configuration may preferentially induce the most highly protective immune response [19, 49, 50].
The loss of neutralization activity against chimeric CHIKV is in line with the finding that total IgG and anti-rE2 antibody titers correlate with the neutralizing titer of Asian serum (S6 Fig), suggesting that most of the neutralizing epitopes are on the E2 glycoprotein. The lack of correlation between anti-rE2 antibodies and neutralizing antibodies seen in ECSA serum could be due to the greater importance of conformational epitopes at E1-E2 sites, but we cannot exclude that it may reflect differences in potency/quality of the circulating antibodies due to the different timings of collection between the Asian and ECSA serum panels (S6 Fig). Correlation between serum neutralization titers and antibody binding titers has been reported in other viral infections such as dengue and influenza [51, 52], and is important for developing serological assays which are accurate correlates of protective immunity following infection or vaccination. Therefore, E2, while appropriate for serological assays to diagnose acute or past infection [53], may not be a suitable candidate for assays to measure protective immunity due to all CHIKV genotypes. Such assays are necessary for vaccine development.
Amino acid changes in key epitope regions, such as naturally occurring mutations or antigenic variation between different genotypes could affect surface charge distribution and electrostatic interactions between epitopes and antibodies, affect binding affinity and ultimately alter neutralizing capacity [14]. The E211K mutation in domain II of the E1 glycoprotein is a significant change of a negatively-charged to positively-charged amino acid, and this appears to enhance antibody binding and neutralization efficacy. During the recent Indian outbreak of ECSA CHIKV, the key amino acid change E1-K211E was shown to be under positive selection pressure [54], which may confer a selective advantage for virus dissemination and escape from the action of neutralization in humans. In addition, E211 is highly conserved in strains of the Asian genotype. Peptide-specific rabbit polyclonal antibody prepared against a short linear epitope (GDIQSRTPESKDVY, position 201–214) including 211K did not show neutralization activity (S7 Fig), suggesting that the neutralizing activity of immune sera targeting this amino acid is highly conformation-dependent. As for the E2 glycoprotein, I2T, H5N, G118S and S194G changes increased antibody binding and neutralization efficacy. All these amino acid changes are positioned within linear epitopes, which interacted with neutralizing antibodies. This was supported by a previous report of well-characterized human neutralizing monoclonal antibodies targeting epitopes that cluster around the LP24 and LP38 peptide regions in our study [43]. Notably, the linear epitope LP1 in our study is similar to E2EP3, a well-characterized key neutralizing linear epitope which has been suggested as a serology marker [26, 55], and LP1 demonstrated cross-reactivity with ECSA and Asian serum in our study. However, we found no effect of K252Q in antibody binding capacity in our cohort, although this was reported recently [14], and this could be due to differential immune responses in different populations. Other linear epitopes (LP19, LP47, LP56 and LP70) were identified in this study which had higher binding than LP1, and as all demonstrated binding to both ECSA and Asian sera, they may be potential candidates for diagnostic serological assays. Furthermore, antibodies against LP19 and LP47 demonstrated neutralizing characteristics which warrant further investigation as vaccine candidates (S8 Fig).
It was interesting that the Asian serum had greater neutralizing capacity against the heterologous ECSA isolates. The previously reported human CHIKV monoclonal antibodies 5F10 and 8B10 had a broad neutralization activity against isolates of the ECSA and West African genotypes, but were also less potent against an Asian isolate from Indonesia [56]. Monkeys inoculated with a virus-like particle vaccine derived from the West African strain 37997 also developed better neutralizing activity to a heterologous ECSA strain LR2006 OPY-1 than to 37997, possibly due to better presentation of conserved epitopes by LR2006 OPY-1 [49]. ECSA and Asian CHIKV genotypes could have induced different immune mediator profiles; as shown in mice, infection with a Caribbean (Asian) strain was associated with a weaker pro-inflammatory Th1 and natural killer cell response and higher IgG1:IgG2c ratio compared to an ECSA CHIKV strain, resulting in less severe joint pathology [57, 58]. Different CHIKV viruses may also trigger differential regulation of key innate immune responses such as TLR3 [59], which plays an important role in shaping subsequent neutralizing capacity [60]. Further studies are needed to understand how differentially-induced immune mediators modulate the properties of circulating serum antibodies.
Two amino acids in LP1 (2T, 5N) of the ECSA virus are critical for binding and neutralization activity, and this further highlights the fact that sequence variation could impact vaccine development. The rabbit polyclonal antibody targeting the linear neutralizing epitope LP1 from the ECSA virus showed reduced cross-neutralization against the Asian genotype, and unexpectedly, rabbit anti-LP1A poorly neutralized the homotypic CAR Asian virus, despite immunization of 4 rabbits. The linear neutralizing epitope LP1A from the Asian virus was not recognized by the ECSA sera. However, clearly there are preexisting antibodies against LP1 and LP1A in the Asian sera. LP47, another linear neutralizing epitope in humans (S8 Fig), which has a sequence that is conserved in both genotypes, did not induce any functional neutralizing antibodies in rabbits despite a similar immunization approach. Future studies will be required to address these apparent underlying differences of neutralizing antibody production from either natural infection or immunization. Nevertheless, our findings indicate that the choice of virus strain for vaccines could impact the spectrum and efficacy of protection across genotypes. For antibody therapy of CHIKV, monoclonal antibodies should retain high potency against a broad diversity of CHIKV isolates [43].
In conclusion, immune serum from humans infected with CHIKV of either ECSA or Asian genotypes showed differences in neutralization and binding capacities. Our findings are relevant to current outbreaks with co-circulating genotypes and provide insights into antibody-mediated immunity resulting from infections with CHIKV of different genotypes.
|
10.1371/journal.pgen.0030028 | Sequential Loading of Cohesin Subunits during the First Meiotic Prophase of Grasshoppers | The cohesin complexes play a key role in chromosome segregation during both mitosis and meiosis. They establish sister chromatid cohesion between duplicating DNA molecules during S-phase, but they also have an important role during postreplicative double-strand break repair in mitosis, as well as during recombination between homologous chromosomes in meiosis. An additional function in meiosis is related to the sister kinetochore cohesion, so they can be pulled by microtubules to the same pole at anaphase I. Data about the dynamics of cohesin subunits during meiosis are scarce; therefore, it is of great interest to characterize how the formation of the cohesin complexes is achieved in order to understand the roles of the different subunits within them. We have investigated the spatio-temporal distribution of three different cohesin subunits in prophase I grasshopper spermatocytes. We found that structural maintenance of chromosome protein 3 (SMC3) appears as early as preleptotene, and its localization resembles the location of the unsynapsed axial elements, whereas radiation-sensitive mutant 21 (RAD21) (sister chromatid cohesion protein 1, SCC1) and stromal antigen protein 1 (SA1) (sister chromatid cohesion protein 3, SCC3) are not visualized until zygotene, since they are located in the synapsed regions of the bivalents. During pachytene, the distribution of the three cohesin subunits is very similar and all appear along the trajectories of the lateral elements of the autosomal synaptonemal complexes. However, whereas SMC3 also appears over the single and unsynapsed X chromosome, RAD21 and SA1 do not. We conclude that the loading of SMC3 and the non-SMC subunits, RAD21 and SA1, occurs in different steps throughout prophase I grasshopper meiosis. These results strongly suggest the participation of SMC3 in the initial cohesin axis formation as early as preleptotene, thus contributing to sister chromatid cohesion, with a later association of both RAD21 and SA1 subunits at zygotene to reinforce and stabilize the bivalent structure. Therefore, we speculate that more than one cohesin complex participates in the sister chromatid cohesion at prophase I.
| Meiosis is a specialized cell division by which sexually reproducing organisms prompt the formation of specialized cells presenting a half of the species chromosomal number. These cells, the so-called gametes, are able to fertilize or be fertilized, depending on the sex in which they are produced and thus restore the species chromosomal number after fertilization. The reduction in the chromosome number is achieved by two successive rounds of chromosome segregations preceded by a single replication of the genetic material. Different proteins, mainly referred to as cohesins, are implied in the correct establishment and maintenance of an intimate association between homologous chromosomes by ensuring their close association until their separation in the first meiotic division. Grasshoppers have been considered as a gorgeous model for meiotic studies for decades due to their low chromosomal number, the large size of their chromosomes, and the well-defined meiotic stages at cytological level. On these grounds, we have combined classical grasshopper chromosome knowledge with protein immunolocalization tools in order to precisely analyze the presence of cohesins throughout the prophase of the first meiotic division. The results not only describe the dynamic loading pattern of several cohesin subunits in two grasshopper species, but they also surprisingly bring into light that different cohesins are sequentially loaded onto meiotic chromosomes throughout the first meiotic prophase. Finally, we discuss the possible roles for this sequential protein loading in relation to the processes that operate during meiosis, proposing a model for meiotic chromosome structure. Besides the novel scientific contributions for a better understanding of the meiotic process, this study clearly points out that classical cytogenetic models can be used to solve modern biological problems.
| Duplication of genetic material and its proper transmission to daughter cells must be scrupulously regulated in order to avoid errors that could modify the chromosomal complement of the species leading to aneuploidies. For this purpose, cells invariably achieve a round of DNA replication before each nuclear division when duplicated genomes separate into two identical cells. To assure a correct distribution, the previously duplicated DNA molecules must be joined from the time of their replication until their segregation at anaphase. The tight association between sister chromatids along their entire length is established by the mitotic cohesin complex. The general features of this complex are conserved from yeast to human [1,2]. This complex is mainly composed of four subunits: a heterodimer of two structural maintenance of chromosome proteins (SMC1 and SMC3), associated with two non-SMC components corresponding to sister chromatid cohesion proteins (SCC1 and SCC3) [3–6]; and for review see [7,8]. The SCC1 subunit is also called a mitotic chromosome determinant (MCD1) in Saccharomyces cerevisiae [9] or radiation-sensitive mutant (RAD21) in Schizosaccharomyces pombe [10]. In addition, in vertebrates there are two SSC3 subunits, which were first characterized as stromal antigen proteins (SA1 and SA2) [6,11].
Recent studies suggest that the association of the cohesin complex with chromatin and the following establishment and maintenance of cohesion are functionally separable, and that additional specific factors are required for each process to be achieved [10,12,13] (and for review see [14]). Proper sister chromatid cohesion may be established most efficiently during S phase since the two replicated sister DNA strands are closely apposed [12]. Cohesin-mediated connections are created at centromeres and at regular intervals along chromatid arms [15–18]. Several proteins are involved in the establishment of sister chromatid cohesion [19–22], despite that they may also be involved in other cellular functions [23–26]. The maintenance of cohesion throughout G2 is thought to facilitate the efficient repair of DNA double-strand breaks by homologous recombination between sister chromatids [27,28]. Furthermore, the cohesin complex plays an essential role ensuring bipolar attachment of sister chromatids to microtubules [21]. Therefore, the establishment of proper cohesion and its regulation during the cell cycle are of fundamental importance for genome stability.
In meiosis, a single DNA replication event precedes two consecutive rounds of chromosome segregation. In this process, cohesin complex not only maintains cohesion along the length of sister chromatids until the first meiotic anaphase, but also contributes to centromere cohesion up to the second meiotic anaphase. Additionally, this complex supports the interactions between homologous chromosomes and between sister chromatids during the initiation of recombination at prophase I [29] and plays a role in the maintenance of chiasmate bivalents until metaphase I [30–32]. Due to the variety of particular meiotic chromosome processes in which cohesion is implicated, it can be expected that meiocytes may contain distinct molecular complexes in order to ensure the specific behavior of chromosomes [33,34]. At present, it is well established in a variety of species that during meiosis some of the canonical mitotic subunits of the cohesin complex [33,35,36] coexist with several meiosis-specific variants such as meiotic recombination proteins (REC8 and REC11), Stromal antigen 3 (STAG3), and SMC1β [29,37–39] (for reviews see [40,41]).
The analysis of the temporal expression and loading of cohesin subunits onto meiotic chromosomes can allow us to elucidate the existence of different cohesin complexes and their role in the dynamics and structure of meiotic bivalents. For this purpose, we have performed these analyses in grasshopper males because, although these organisms have been classically used to analyze meiosis under a cytological point of view, there are no data regarding the participation of the cohesin complex and its loading dynamics in the meiotic chromosome organization. Taking into account the advantage of a certain degree of evolutionary conservation of the cohesin subunits among species, we have tested in two grasshoppers, Eyprepocnemis plorans and Locusta migratoria, several antibodies previously generated against human SMC3 and Drosophila cohesin subunits (DSA1 and DRAD21). The expression pattern of these three proteins has been analyzed on squashed spermatocytes, since this procedure preserves the structure and volume of the nucleus, allowing an accurate analysis of the 3-D relationships among them [42,43].
We have found that, whereas the cohesin axes defined by SMC3 allowed us to infer the position of both the axial and lateral elements (AEs/LEs) of the synaptonemal complex (SC) from preleptotene onward, the non-SMC cohesin subunits RAD21 and SA1 are not loaded onto chromosomes until their synapsis at zygotene. We propose that this second round of cohesin subunit loading reinforces the cohesion in the bivalent until its segregation at anaphase I. Finally, a possible model for the loading of cohesin subunits and the structure of meiotic chromosome during the first meiotic prophase is proposed and discussed.
To determine the immunoreactivity and specificity of anti-hSMC3, anti-DSA1, and anti-DRAD21 polyclonal antibodies in the species analyzed here, E. plorans and L. migratoria, we performed immunoblot analyses of grasshopper testis nuclear fractions. Mouse nuclear testis fraction and nuclear extract of Schneider cells from Drosophila were used as positive controls (Figure 1). In the grasshopper testis, each antibody specifically recognized a single band, all of them representing a similar molecular weight to that detected in the positive control extracts. The molecular weights of the immunoreactive bands in both grasshopper species and in the corresponding control were around 140 kDa for SMC3, 130 kDa for SA1, and 120 kDa for RAD21. Therefore, the antibodies used in the present study allowed us to identify, in grasshopper testis, the homologs of the cohesin subunits SMC3, SA1, and RAD21.
The immunolocalization of the cohesin subunit SMC3 in squashed grasshopper testis preparations of the species analyzed (Figure 2) revealed a uniform pattern of small puncta over the spermatogonial nuclei (Figure 2A). These cells were easily distinguished due to an evident nuclear protrusion which corresponds to the single sex chromosome (Figure 2A and 2B). It is important to note that the sex chromosomal determination system in these grasshopper species is of the type XX for females, and XO for males; therefore, in males the X chromosome remains as a univalent throughout all meiotic stages. The labeling pattern of SMC3 on the chromatin of the X chromosome in spermatogonia is similar to that present in the rest of the autosomes (Figure 2A and Video S1). Preleptotene spermatocytes display a pattern of larger, albeit homogeneously distributed foci of SMC3 immunostaining in the nuclei, except at the X chromosome (Figure 2C and 2D and Video S2). At this stage, the X chromosome, which usually appears in the nuclear periphery (Figure 2D), exhibits a weaker and more diffuse SMC3 staining than that displayed by the autosomes (Figure 2C and 2D and Video S2). Leptotene spermatocytes show irregular discrete threads of SMC3, which appear to be continuous throughout the nuclear volume, denoting cohesin axis maturation (Figure 2E and Video S3). The peripherally located X chromosome (Figure 2F) presents a single SMC3 axis, which resembled in conformation and localization the AE observed in this chromosome under electron microscopy [44]. At the leptotene-zygotene transition, and concomitant with the onset of synapsis, it becomes evident that cohesin axes start to associate in pairs, forming thick filaments in one or two discrete nuclear regions (Figure 2G and Video S4). In zygotene spermatocytes, both paired and unpaired SMC3 axes are discernible as synapsis proceeds (Figure 2I and 2J). Additionally, we observe that all the cohesin axis ends congregate in a discrete nuclear region where they polarize in a bouquet-like arrangement (Figure 2I and Video S5). The single unpaired SMC3 axis of the X chromosome was also polarized into the bouquet configuration (Figure 2I and 2J and Video S5). At pachytene, cohesin axes achieve pairing at their full length, except the X chromosome, which remains unsynapsed (Figure 2K and 2L and Video S6). Hence, at this stage, the number of thick, paired axes corresponds in number and length to the 11 autosomal bivalents of the analyzed species (Figure 2K). All of the cohesin axis ends are distributed at the nuclear periphery where they seem to be associated to the nuclear envelope (this situation is clearly detected in 3-D reconstructed cells, as in the pachytene spermatocyte shown in Video S6). Once again, the X chromosome appears located in the nuclear periphery and displays a single unsynapsed axis, around half of the width of the paired axes present in autosomal bivalents (Figure 2K and Video S6).
In diplotene, desynapsis between homologs becomes evident after analyzing the DAPI staining of the spermatocytes (Figure 2N). This stage is characterized by the irregular appearance of the SMC3 cohesin axes, which present a barbed wire–like organization, with multiple excrescences running from the axes to the surrounding chromatin (Figure 2M and Video S7). From late diplotene onward, the SMC3 signals appear located in the bivalents at the so-called interchromatid domain [45] and also between the sister chromatids of the X chromosome (Figure 2O and Video S8).
In interphase spermatogonial cells, the anti-DSA1 antibody renders a weak uniform labeling, whereas in pachytene spermatocytes, lines resembling the structure of SCs are detected. In spermatogonial prophases, as chromatin condensation progresses, an increase of the labeling in the nuclear regions far apart from the chromosomal territories is observed. In these cells, centrioles and pericentriolar material are more intensively labeled. Cell poles are detected in all spermatogonial mitotic stages. Metaphase cells show bright protoplasm, but the chromosomes appear negatively labeled. It is interesting to note that no signaling is located either in the centromeric regions or between sister chromatids. This absence of labeling in the chromosomes is maintained until telophase. However, we have recently obtained results in the species Chorthippus jucundus, which indicated the absence of labeling in between chromatids despite the fact that SA1 is present at the centromeric region of spermatogonial-condensed chromosomes (unpublished data)
The immunolocalization of SA1 in grasshopper squashed spermatocytes (Figure 3) revealed that in leptotene cells, no appreciable amounts of SA1 labeling are present inside the nucleus, with the exception of nucleoli (Figure 3A and 3B). However, dispersed foci are detected along the nuclear periphery (Figure 3A and 3B). These signals appear paired and polarized in one discrete nuclear region at later meiotic stages. We interpret these foci as the visualization of the attachment plates of the SC due to their staining characteristics with this antibody (see Figure 3E–3G) and their pairing and polarization as the development of a bouquet-like configuration. At this stage, no SA1 label is detected on chromosome axis.
In zygotene nuclei, and concomitantly with the progression of synapsis, SA1 forms discrete linear threads resembling the initial SC stretches between homologous chromosomes (Figure 3C and 3D). Accordingly, pachytene spermatocytes display multiple stained linear structures over the chromatin, with their ends close to the nuclear periphery (Figure 3A–3D). After 3-D reconstruction of the nuclei, using either confocal or optical sections, it is obvious that in each spermatocyte there are 11 fluorescent lines that correspond to the number of bivalents of the species analyzed. The ends of these linear SA1 structures, which locate in close proximity to the nuclear periphery, present two separate expansions (Figure 3E–3G). In a frontal view, these signals appear as two well-defined dots (Figure 3F). Therefore, these structures may represent the attachment plates of the LEs of the SC that associate with the nuclear envelope. It is worth mentioning that the single AE of the X chromosome is never visualized after SA1 detection and 3-D reconstructions (Figure 3H–3J). On these grounds, our results indicate that SA1 is only located on those chromosomal regions that seems to achieve SC development.
In diplotene spermatocytes, the well-defined SA1 lines become irregular and with lateral excrescences along their length, except at the association plates (Figure 3K and 3L). At diakinesis, SA1 labeling is confined to the interchromatid domain and begins to be detectable in the centromeric region of all bivalents (Figure 3M and 3N).
To determine the distribution of RAD21 in prophase I spermatocytes, we performed the immunolocalization of RAD21 in grasshopper squashed spermatocytes (Figure 4). Leptotene spermatocytes do not show any RAD21 signals inside the nuclei (Figure 4A and 4B), but, in contrast to SA1, RAD21 is not detected in the periphery of the nucleus and is also undetectable in nucleoli. Afterward, by zygotene, short discrete stretches of RAD21 are observed in certain nuclear regions (Figure 4C and 4D). At pachytene, linear RAD21 structures, with both of their tips ending in the nuclear periphery, are visualized (Figure 4A–4D). After analyzing the 3-D reconstruction of these spermatocytes, it can be concluded that the number and size of these lines correspond to the number and size of the bivalents of the species (Figure 4A–4D and Video S9). Like SA1, and in clear contrast to SMC3, no threads of RAD21 are detected inside the sex chromosome chromatin (Figure 4E–4G and Video S9). A barbed wire–like localization of RAD21 is observed at diplotene (Figure 4H and 4I), which afterward localizes at the interchromatid domain by diakinesis (Figure 4J and 4K). These results indicate a similar spatio-temporal expression pattern of RAD21 and SA1 cohesin subunits during the first meiotic prophase in the two grasshopper species analyzed.
A double immunolocalization of SMC3 and RAD21 was performed in order to analyze their relative chromosomal distribution during prophase I (Figure 5). Whereas SMC3 reveals the AEs in leptotene spermatocytes, no RAD21 signal was evident in these nuclei (Figure 5A–5D and Video S10). Zygotene spermatocytes are characterized by the colocalization of both cohesin subunits in the synapsed chromosomal regions (Figure 5A–5D and Video S10). In contrast, the unsynapsed regions are only stained by the anti-SMC3 antibody (Figure 5A–5D and Video S10). A complete colocalization of both cohesin subunits is clearly apparent in pachytene (Figure 5A–5D and Video S10). These results were obtained and validated by both optical sections, under fluorescence microscopy (Figure 5), as well as by sequential scanning capture of images under confocal microscopy (unpublished data). Surprisingly, the complete colocalization pattern of these proteins is not maintained at diplotene and diakinesis (Figure 5E–5L and Video S11) since, in addition to the yellow-labeled regions that indicate colocalization, we can also observe a few chromosomal regions where only SMC3 (red-labeled regions) or RAD21 (green-labeled regions) labeling is present (Figure 5G and 5K).
SMC1 and SMC3 are involved in sister chromatid cohesion during meiosis and also seem to be essential for the organization of the AE structure in which chromatin loops are attached in mammal meiosis [46]. A participation of SMC3 in the structure of the LEs of Drosophila has also been proposed [47]. In the two grasshopper species analyzed, we have detected the presence of SMC3 in all prophase I stages, whereas in preleptotene cells there is no specific nuclear distribution of this protein (similar to that found in spermatogonial interphases). At the onset of prophase I in grasshoppers, SMC3 re-localizes in well-defined and continuous lines; this situation is slightly different from that previously reported in mammals since it has been demonstrated that, at least in spreads, both proteins localize in a beaded structure along the AEs/LEs at pachytene [46]. Therefore, the morphology of these lines and their development throughout prophase I lead us to propose that SMC3 is closely associated to the AEs/LEs at the base of the chromatin loops (Figure 6). This assertion, in agreement with mammal observations, is reinforced by the correspondence between the number and size of the SMC3 signals at pachytene and the number of SCs observed in the two species studied.
Therefore, although in grasshoppers it is not possible to assay the SC formation directly; it could be inferred from the identification of thin and thick SMC3 filaments that correspond to synapsed and unsynapsed regions, respectively. The morphology of the SMC3 axis and its dynamics allow an accurate identification of the different prophase I stages, and, a subsequent analysis of the presence and localization of other proteins throughout prophase I [43,48,49].
The cohesin complex is responsible for the maintenance of sister chromatid cohesion in both mitotic and meiotic divisions. However, its composition and dynamics seem to be different in both processes. Thus, while in mitosis sister chromatid cohesion is released in each cellular division, cohesion in meiosis is lost in the two temporally separated divisions: arm cohesion at anaphase I and centromeric cohesion at the onset of anaphase II [45].
We have demonstrated here that a differential loading of cohesin subunits takes place during prophase I in grasshoppers. In this sense, whereas SMC3 labeling is already present in preleptotene cells, SA1 and RAD21 are not detectable until the initiation of synapsis at zygotene. Likewise, the double immunolocalization of SMC3 and RAD21 clearly defines a sequential association of these proteins to meiotic chromosomes. A possibility that cannot be excluded is that a small amount of the cohesin subunits, SA1and RAD21, is already associated with the chromosome at premeiotic S phase, but their amounts are not enough to be detected by the immunolocalization protocol used in this study. However, since we are able to clearly visualize the localization of SMC3 from early meiotic prophase I onward, and considering that SA1 and RAD21 participate within the same cohesin complex with SMC3 in a similar stoichiometry, we should have necessarily detected them. This timing of appearance of SMC3 versus SA1 and RAD21, and its localization pattern and dynamics during the first meiotic prophase, also concurs with previous biochemical results from Drosophila embryos, where the immunoprecipitation experiments of RAD21 intriguingly suggest that RAD21 and SA1 are more tightly associated to each other than they are with the SMC subunits of the cohesin complex [50]. However, after the coimmunoprecipitation assays using grasshopper testis nuclear protein extracts, we cannot discern whether RAD21 and SA1 are part of another cohesin complex, or, on the contrary, whether or not there is a sequential addition of the cohesin subunits.
Since SA1 and RAD21 cohesin subunits only became detectable at zygotene, we speculate that following the initial association of the canonical cohesin complex to the chromatin in early meiosis, a second round of cohesin loading, which included at least SA1 and RAD21, could be necessary to increase the bivalent stability. This additional round of cohesion establishment may be necessary to reinforce the arm cohesion of bivalents, in order to counteract the polar forces throughout congression, during prometaphase I and metaphase I, until the segregation of recombined homologous chromosomes at anaphase I. In this sense, the subsequent loading of both RAD21 and SA1 at zygotene is consistent with the necessity of arm cohesion reinforcement in order to prevent a premature separation of homologs prior to anaphase I.
The differential loading of cohesin subunits has indirectly been suggested in other organisms. For instance, in mammals it has been proposed that REC8 may provide a basis for AE formation in early prophase I prior to the appearance of SMC1β and SMC3 [51]. On the other hand, the depletion of TIM-1 in Caenorhbditis elegans (clock protein TIMELESS in Drosophila) prevents the assembly of non-SMC subunits onto meiotic chromosomes. However, a cohesin complex with SMC components is already loaded, suggesting that SMC1 and SMC3 are associated onto chromatin independently of non-SMC proteins [52]. Finally, in Arabidopsis, the protein SWI1, which appears to be required for early meiotic events that are at the crossroad of sister chromatid cohesion, is expressed exclusively in meiotic G1 and S phase [53]. Other cohesin proteins, such as REC8 and SCC3, will appear later [54].
In grasshopper spermatocytes, the single X chromosome shows an AE that never appears partially or fully synapsed at prophase I [44]. Consequently, the study of this chromosome is of particular interest as regards the composition of its cohesin axis. Our current results undoubtedly demonstrate that in prophase I spermatocytes, a single SMC3 axis is always observable inside the sex chromosome chromatin. By contrast, whereas both SA1 and RAD21 are present in each autosomal bivalent from zygotene onward, neither SA1 nor RAD21 are detected in the X chromosome throughout prophase I. In strong agreement with our finding that RAD21 and SA1 are only present in those autosomal regions that have achieved synapsis, the special features of the X chromosome emphasize our previous assertion that only when homologous synapsis progresses, the RAD21 and SA1 proteins are loaded onto chromosomes. Whether RAD21 and SA1 participate in a distinct cohesin complex other than SMC3, or whether our observations only indicate a sequential addition of cohesin subunits in meiotic bivalents, or both, still remains an open question. However, the bulk of these data taken together strongly suggests that, as in other organisms, different cohesin complexes may coexist during meiosis in grasshoppers, and that the presence of distinct cohesin complexes may contribute to the different dynamic meiotic chromosome requirements.
JSR and coworkers [55] studied the chromosome organization in grasshopper spermatocytes by light microscope analysis of silver-stained cores and concluded that the chromatid core represents the scaffold of each sister chromatid. Afterward, it was proposed as a model of meiotic chromosome organization (based on light and electron microscope observations), where the chromatid core locates at the base of the chromatin loops, would act as the framework for the further assembly of the AE/LE proteins [56]. Here, we incorporate the contribution of the subunits of the cohesin complex (Figure 6). Some of these subunits, like SMC3, could be considered as a structural component of an initial cohesin axis (dark green in Figure 6), which may in fact be the real framework for the assembly of the AE/LE proteins (pink), and that is closely associated to the chromatid cores (yellow). The subsequent association of the AE proteins to this primary cohesin axis would represent the cytologically detected AEs. Simultaneous with the assembly of the transverse filament/central element protein/s (purple) at zygotene, a second cohesin subunit loading containing at least SA1 and RAD21 (light green) would occur just before, during, or immediately after homologous synapsis. At pachytene, when the tripartite structure of SCs is fully formed, not only the SC proteins, but also cohesin complexes, would contribute essentially to the maintenance of the correct association of homologous chromosomes. This model of meiotic chromosome organization provides new essential insights to explain both the existence of synapsis in the absence of SYCP3 [57] and the existence of homologous chromosome alignment in the SYCP1-deficient mouse model [58]. Furthermore, observations of later meiotic stages in grasshopper spermatocytes, from diplotene up to telophase I, indicate that the different cohesin subunits do not fully colocalize in these stages (JSR, unpublished data).
Adult males of the grasshopper species E. plorans and L. migratoria, (Orthoptera: Acrididae) collected in natural populations, or bred in the laboratory, were used in the present study. Drosophila S2 cells were grown in Schneider's medium (Sigma, http://www.sigmaaldrich.com) supplemented with 10% fetal bovine serum at room temperature. Adult male C57BL/6 mice from our animal facilities were also used for this study.
To detect the cohesin subunit SMC3, we employed a polyclonal rabbit anti-SMC3 antibody (AB3914; Chemicon International, http://www.chemicon.com) raised against a synthetic peptide from human SMC3. It is worth noting that only the lot number 220701985 of the cited antibody rendered tiny immunolabeling signals in two grasshopper species contrary to the actually commercialized stock provided by Chemicon. SA1 was detected by a rabbit anti-DSA1 antibody generated against Drosophila SA1 recombinant protein [59]. RAD21 was detected using a rabbit anti-DRAD21 antibody raised against a bacterially expressed carboxy-terminal fragment of Drosophila RAD21 [60].
Testes from adult E. plorans and L. migratoria males were removed and placed in 1 ml of 0.5% Triton X-100 in PBS for 5 min and then washed with PBS, as previously described [48]. Schneider cells (5 × 106 cells) were harvested and washed with PBS. Testes from adult male C57BL/6 mice were removed and processed as previously described [61]. Nuclear extracts from grasshopper testes and Schneider cells were obtained using the NE-PER Nuclear and Cytoplasmic Extraction Reagent Kit (Pierce, http://www.piercenet.com) according to the manufacturer's instructions. For Western blotting, proteins were resolved by 8% SDS-PAGE [62] and blotted with the following antibodies diluted in 4% nonfat dry milk in PBS: anti-DSA1 at a 1:1,000 dilution, anti-DRAD21 at a 1:500 dilution, and anti-SMC3 at a 1:5000 dilution.
Testes were removed and fixed for immunofluorescence as previously described [42]. Briefly, testes were fixed in freshly prepared 2% formaldehyde in PBS containing 0.1% Triton X-100 (Sigma). After 5 min, several seminiferous tubules were placed on a slide previously coated with 1 mg/ml poly-L-lysine (Sigma) with a generous drop of fixative, and tubules were gently minced with tweezers. After exerting pressure on the cover slip, slides were frozen in liquid nitrogen and the cover slip removed with a razorblade. The slides were then rinsed three times for 5 min in PBS and incubated with the corresponding primary antibody for 45 min at room temperature, or overnight at 4 °C. Primary antibodies diluted in PBS were used at the following dilutions: anti-DSA1 antibody at 1:50 dilution, anti-DRAD21 antibody at 1:50 dilution, and anti-SMC3 at a 1:30 dilution. Following three washes in PBS, the slides were incubated for 30 min at room temperature with a fluorescein isothiocyanate-conjugated goat anti-rabbit IgG (Jackson ImmuoResearch, http://www.jacksonimmuno.com) secondary antibody at a 1:150 dilution in PBS. In the double immunolabelling experiment with SMC3 and RAD21, since the two primary antibodies were generated in the same host species, we proceeded as previously described [63]. Subsequently, slides were rinsed in PBS and counterstained for 3 min with 5 μg/ml DAPI (4′,6-diamidino-2-phenylindole). After a final rinse in distilled water, slides were mounted with Vectashield (Vector Laboratories, http://www.vectorlabs.com) and sealed with nail varnish.
Observations were performed using an Olympus BX61 (http://www.olympus.com) microscope equipped with a motorized Z-axis and epifluorescence optics. The images were captured with a DP70 Olympus digital camera using the associated analySIS software (Soft Imaging System, Olympus). Samples were also analyzed under different confocal laser scanning microscopes, a Leica TCSNT (http://www.leica.com), a Bio-Rad radiance 2000 (http://www.bio-rad.com), and an Olympus IX-70 inverted microscope equipped with a confocal laser scanning system (Fluoview 300). Images were captured by sequential scanning, noise-filtered, corrected for background, and processed using the appropriate software. Images were finally analyzed and processed using Adobe Photoshop 6.0 software (http://www.adobe.com), the public domain software ImageJ (National Institutes of Health, United States; http://rsb.info.nih.gov/ij), and VirtualDub (VirtualDub, http://www.virtualdub.com).
|
10.1371/journal.pgen.1005680 | LncRNA-HIT Functions as an Epigenetic Regulator of Chondrogenesis through Its Recruitment of p100/CBP Complexes | Gene expression profiling in E 11 mouse embryos identified high expression of the long noncoding RNA (lncRNA), LNCRNA-HIT in the undifferentiated limb mesenchyme, gut, and developing genital tubercle. In the limb mesenchyme, LncRNA-HIT was found to be retained in the nucleus, forming a complex with p100 and CBP. Analysis of the genome-wide distribution of LncRNA-HIT-p100/CBP complexes by ChIRP-seq revealed LncRNA-HIT associated peaks at multiple loci in the murine genome. Ontological analysis of the genes contacted by LncRNA-HIT-p100/CBP complexes indicate a primary role for these loci in chondrogenic differentiation. Functional analysis using siRNA-mediated reductions in LncRNA-HIT or p100 transcripts revealed a significant decrease in expression of many of the LncRNA-HIT-associated loci. LncRNA-HIT siRNA treatments also impacted the ability of the limb mesenchyme to form cartilage, reducing mesenchymal cell condensation and the formation of cartilage nodules. Mechanistically the LncRNA-HIT siRNA treatments impacted pro-chondrogenic gene expression by reducing H3K27ac or p100 activity, confirming that LncRNA-HIT is essential for chondrogenic differentiation in the limb mesenchyme. Taken together, these findings reveal a fundamental epigenetic mechanism functioning during early limb development, using LncRNA-HIT and its associated proteins to promote the expression of multiple genes whose products are necessary for the formation of cartilage.
| A fundamental problem studied by skeletal biologists is the development of regenerative therapies to replace cartilage tissues impacted by injury or disease, which for individuals affected by osteoarthritis represents nearly half of all of all adults over the age of sixty five. To date, no therapies exist to promote sustained cartilage regeneration, as we have not been able to recapitulate the programming events necessary to instruct cells to form articular cartilage without these cells continuing to differentiate into bone. Our analysis of the early programming events occurring during cartilage formation led to the identification of LncRNA-HIT a long noncoding RNA that is essential for the differentiation of the embryonic limb mesenchyme into cartilage. A genome wide analysis of LncRNA-HIT’s distribution in the mesenchyme revealed strong association between LncRNA-HIT and numerous genes whose products facilitate cartilage formation. In the absence of LncRNA-HIT, the expression of these chondrogenic genes is severely reduced, impacting the differentiation of these cells into cartilage. Mechanistically, LncRNA-HIT regulates these pro-chondrogenic genes by recruiting p100 and CBP to these loci, facilitating H3K27ac and transcriptional activation. LncRNA-HIT also appears to be present in most vertebrate species, suggesting that the epigenetic program regulated by this lncRNA may represent a fundamental mechanism used by many species to promote cartilage formation.
| In the animal kingdom, embryogenesis proceeds through the coordinated expression of genes whose products mediate the formation of complex tissues and structures. While proteins encoded by mRNAs contribute extensively to the regulation of these developmental processes, recent studies of the human and mouse genomes suggest that long noncoding RNAs (lncRNAs) play an essential role in coordinating the expression of genes required for tissue formation and organ development [1–5].
As gene regulatory molecules, lncRNAs modulate target gene expression using a variety of mechanisms, particularly in the nucleus where they can function as decoys, scaffolds, guides, or even enhancers [6]. As decoys, lncRNAs can titrate away transcription factors (GAS5-Glucocorticoid receptor), the transcriptional machinery (DHFR minor-TFIIB) or even splicing factors (MALAT1- SR splicing factors) to modulate the expression of target genes [7–9]. As a component of their secondary structures, lncRNAs such as HOTAIR, ANRIL, and Kcnq1ot1 provide protein-specific scaffolds, assembling enzyme complexes such as the polycomb repressive complexes 1 or 2 (PRC1, PRC2), LSD-1, and CoREST-HDAC to facilitate changes in histone lysine methylation or acetylation to enforce the transcriptional state of a specific locus [10–14]. Nuclear lncRNAs may also function as guides to localize protein complexes to specific chromosomal regions. Notably, the lncRNA, XIST, functions as a guide to recruit proteins such as YY1 or components of PRC2 to promote X chromosome inactivation [15–19]. Finally a growing body of evidence indicates that lncRNAs may also function as enhancer RNAs, using chromosomal looping to place proteins bound by the lncRNA proximal to genes to facilitate their regulation [5, 20–22].
A key group of genes regulated by lncRNAs are the Hox genes, a conserved family of developmental transcription factors that exhibit temporally- and spatially-restricted domains of expression and function [5, 23–29]. Evidence for the unique functions of the vertebrate Hox lncRNAs was first shown with the trans-acting functions of HOTAIR, which is expressed from the HoxC locus to recruit PRC2 to the HOXD cluster, where it mediates H3K27 methylation to repress the expression of several 5’ HoxD genes [4]. During development, the 5’ HoxD proteins regulate limb and axial skeleton development, suggesting that de-repression of the 5’ HoxD genes, by perturbations in HOTAIR expression, would most likely affect these same skeletal elements [30–32]. Recent studies confirm this hypothesis, as mice lacking HOTAIR exhibit malformations of carpal/metacarpal elements in the limb as well as homeotic transformations of the lumbar, sacral, and caudal vertebrae, phenotypes consistent with de-repression of Hoxd11 and Hoxd13 [29].
Proximal to the 5’ HoxA gene cluster is the lncRNA HOTTIP, which functions as an enhancer lncRNA to regulate the expression of 5’ HoxA genes to control the growth and elongation of zeugopod and autopod skeletal elements [5]. Mechanistically, HOTTIP modulates gene expression by chromosomal looping, placing its recruited Trithorax/WDR5/MLL protein complexes proximal to 5’ HoxA genes to facilitate H3K4me3 and gene expression [5,33]. HOTTIP function has also been associated with endochondral ossification [34]. This finding in conjunction with the functional studies of HOTAIR indicate a role for lncRNAs associated with the vertebrate Hox genes as necessary components for the development, patterning, and maturation of skeletal tissues [5,29,34]. Interestingly, a second lncRNA, LncRNA-HIT, has also been identified within the HOXA locus [35]. While initially characterized as a TGFβ-induced modulator of epithelial-mesenchymal transformations in tumor cells [35] our subsequent analysis of this lncRNA indicates that it is also expressed in the early limb where we hypothesized a role for this lncRNA in mediating chondrogenic differentiation. Characterization of subcellular localization of LncRNA-HIT by RNA FISH indicated the transcript is predominantly localized to the nucleus where it associates with p100/CBP complexes, suggesting that LncRNA-HIT may regulate gene expression by recruiting modulators of H3K27 acetylation. To identify the genes most likely to be regulated by LncRNA-HIT, ChIRP-seq was performed using pre-chondrogenic limb bud tissue. From this analysis LncRNA-HIT was found to be associated within 25 kb of numerous pro-chondrogenic genes including: Bmpr1b, Gli2, Col14a1, Adam17, Kdelr2, Pik3cb, Hoxa13, Hoxa11, Ncam1, and Gpc6. Loss of function analyses confirmed LncRNA-HIT ‘s role as a modulator of pro-chondrogenic expression as, H3K27ac, and near peak gene expression were significantly reduced by LncRNA-HIT-specific siRNA treatments in the limb mesenchyme. Moreover, chondrogenic differentiation was also significantly reduced in limb mesenchyme treated with siRNAs targeting either LncRNA-HIT or the p100 transcript, confirming that the LncRNA-HIT and its recruited protein complex are necessary to maintain H3K27ac and chondrogenic gene expression which facilitates differentiation. Combined, these findings identify an epigenetic mechanism functioning during limb skeletogenesis, using an lncRNA to coordinate the expression genes necessary to direct undifferentiated limb mesenchyme towards a chondrogenic state.
The LncRNA-HIT transcript was first identified as the full length cDNA, 9530018H14RIK, by the RIKEN Mouse Gene Encyclopedia Project and was mapped as a single exon to mouse chromosome 6 between Hoxa11 and Hoxa13 by genome sequencing (S1 Fig) [36]. Conservation analysis of the LncRNA-HIT cDNA sequence using BLAST (NCBI) revealed a single 253 bp region present in most vertebrate species (S1 Fig). Characterization of LncRNA-HIT ‘s protein coding potential using the Coding Potential Assessment Tool (CPAT) revealed multiple stop codons in all six reading frames and a protein coding probability of 0.084, well below the 0.44 threshold predicted for protein coding genes (S1 and S2 Figs) (CPAT version 1.2; http://rna-cpat.sourceforge.net, [37]). In comparison, CPAT analysis of neighboring genes, Hottip and Hoxa13, revealed scores of 0.049 and 0.99 respectively, confirming the poor coding potential of the lncRNA Hottip as well as the favorable protein coding potential of Hoxa13 (S2 Fig). Finally analysis of the potential initiation codons present in the LncRNA-HIT transcript also suggested poor protein coding potential, as these sites lack a Kozak consensus sequence and exhibited poor translational potential as determined by the NetStart software package (S2 Fig), http://www.cbs.dtu.dk/services/NetStart/, [38–39].
Localization of the LncRNA-HIT transcript by in situ hybridization revealed expression in the pre-chondrogenic limb mesenchyme as early as embryonic day (E) 10.5 (Fig 1A). By E 11.5 LncRNA-HIT expression is expanded to a greater portion of the limb bud, encompassing the majority of the pre-chondrogenic tissues that normally express members of the 5’ HoxA and D gene clusters including: Hoxa9-13 and Hoxd11-13 (Fig 1B). By E 13.5, LncRNA-HIT expression continued to follow the expression pattern of Hoxa13; particularly in the limb perichondrial tissues, gut, genital tubercle, and urogenital sinus (Fig 1L–1N) [27, 40–42]. In situ hybridization using the same LncRNA-HIT sequence transcribed in the opposite orientation revealed no expression in the limb or other embryonic tissues, confirming reports [43] of its unidirectional transcription in the same orientation as the HoxA genes (Fig 1C, S1 Fig). Finally, quantitation of LncRNA-HIT expression in the E 11.0 limb confirmed that it is highly expressed (Ct = 21.4 ± 0.8) at levels nearly nineteen-fold greater than Hottip, which resides approximately 5 kb from LncRNA-HIT on mouse chromosome 6 (Fig 1O, S1 Fig).
A recent analysis of the subcellular localization of 61 lncRNAs revealed a majority localized to the nucleus, recruiting proteins to mediate histone modifications, chromatin accessibility, and gene expression [44]. To gain insight into the potential function of the LncRNA-HIT, we first examined its sequence composition, which identified the same nuclear retention motif present in the BORG lncRNA [45] (Fig 2, S1 Fig). Analysis of LncRNA-HIT’s subcellular localization by single molecule RNA FISH detected the lncRNA in the nucleus where it was distributed diffusely and in larger foci in the limb bud mesenchyme (Fig 2A–2D). In contrast, Gapdh transcripts localized primarily to the cytoplasm in these cells, a finding consistent with the protein-coding function of the Gapdh transcript (Fig 2E and 2F). Finally, pre-treatment of the fixed limb mesenchyme with RNase A resulted in the complete loss of detection of the LncRNA-HIT and Gapdh signals, indicating that the signals detected by each FISH probe set corresponds to RNA hybridization rather than hybridization to the corresponding sequence present in the chromosomal DNA. (Fig 2G and 2H).
To determine which proteins interact with LncRNA-HIT in the nucleus, we first tested whether it functions similarly to its chromosomal neighbor, Hottip, which recruits WDR5 to facilitate H3K4me3 [5, 46]. Immunoprecipitation of WDR5 from Flag-WDR5 293T cells transfected with an LncRNA-HIT expression vector revealed no enrichment of the LncRNA-HIT transcript, suggesting a separate mechanism for LncRNA-HIT function in the nucleus (S3 Fig). To identify the proteins binding to LncRNA-HIT, an RNA affinity assay was used to isolate limb proteins preferentially binding to the LncRNA-HIT transcript. Mass spectroscopy analysis of the U1 control and LncRNA-HIT elution fractions identified multiple proteins in the LncRNA-HIT and U1 control elution fractions (S4 Fig). As an initial filter, proteins common to both the U1 and LncRNA-HIT elution fractions were excluded from subsequent analysis. Several proteins exclusive to the LncRNA-HIT elution fractions were also excluded from subsequent analysis as they could not be reproducibly detected in replicate elution fractions. After these initial exclusions, a single protein was identified to be consistently enriched (> 23-fold average enrichment) in only the LncRNA-HIT elution fractions. The single protein was identified as p100, a 100 Kd transcriptional co-factor that partners with creb binding protein (CBP) to recruit histone acetyltransferase activity to the STAT6 locus (Fig 3, S4 Fig, and Methods) [47–50]. Analysis of Snd1 expression which encodes the p100 protein confirmed that p100 is co-expressed with LncRNA-HIT in many of the same embryonic regions including the undifferentiated fore- and hindlimb mesenchyme as well as in the developing genital tubercle (Fig 3A–3D). We next evaluated whether p100 and CBP form a complex with endogenous LncRNA-HIT in the limb mesenchyme. Western blot analysis of the limb bud proteins co-precipitating with biotinylated-LncRNA-HIT revealed enrichment of p100 and CBP, suggesting that both proteins may form a complex with LncRNA-HIT (Fig 3E and 3F). Testing this hypothesis, we examined whether endogenous LncRNA-HIT present in the limb mesenchyme would co-precipitate with p100 and CBP. Immunoprecipitation of p100 and CBP from limb bud mesenchyme revealed a consistent enrichment of LncRNA-HIT (> 2-fold) compared to parallel precipitations using IgG, confirming that the endogenous lncRNA forms a complex with p100 and CBP in the pre-chondrogenic limb mesenchyme (Fig 3G).
The nuclear retention of LncRNA-HIT as well as its ability to bind p100/CBP complexes suggested the lncRNA may function at the chromosomal level to modulate gene expression in the developing limb. To identify the chromosomal regions contacted by LncRNA-HIT, ChIRP-seq was used to precipitate chromatin fractions associated with the lncRNA from E 11.0 limb bud mesenchyme.
The analysis pipeline previously described [51] was used to process the LncRNA-HIT-associated peaks as determined by ChIRP-seq. DNA isolated from the precipitated chromatin was submitted to Elim Biopharm (Hayward, CA) for library preparation and next generation sequencing. The sequenced fragments were assembled against murine genome (NCBI37/mm9) using Bowtie and peaks were ranked by MACS using their assigned p-value [52–53]. Visual inspection of the ranked peaks using the UCSC genome browser revealed a marked drop-off in peak signal strength beyond peak 775 (p-value = 5.9 x 10−57). Based on this result, the top 775 peaks were selected to identify candidate genes potentially regulated by LncRNA-HIT. From the initial cohort of 775 peaks, 173 peaks were excluded from subsequent studies as they were also present in parallel assays using the same LncRNA-HIT probe sets and a glial cell line control that does not express LncRNA-HIT. The remaining 602 peaks were evaluated for their association with cis-regulatory elements in the murine genome using the Genomic Regions of Enrichment of Annotations software (GREAT) (S5 Fig) [54]. Interestingly, while 588 peaks were identified as associating with one or more cis-regulatory elements, a correlative function for these associated cis-regulatory regions was not identified by the GREAT software package (S5 Fig). Next, because GREAT provides only a computational prediction of function, a second analysis was performed focusing on high confidence peak-to-gene relationships using peaks mapping within 25 kb of a known gene. After this analysis, 42 peak-associated candidate genes were identified as potential targets for regulation by LncRNA-HIT-p100/CBP complexes (Fig 4). To validate the regulation of these genes by LncRNA-HIT-p100/CBP complexes, E 11.0 limb mesenchymal cells were transfected with two siRNAs specific for the LncRNA-HIT transcript (-6.15-fold knockdown, Fig 4) and evaluated for changes in gene expression in three independent assays using qRTPCR. From this analysis, 28 near-peak genes exhibited decreased expression in response to the LncRNA-HIT siRNA treatments with fold-change differences ranging from -1.5- to -9.75 compared to transfections using scrambled siRNA controls (Fig 4). The remaining near-peak genes exhibited no change in expression in response to the LncRNA-HIT siRNA treatments (6 genes) or could not be detected in the E 11.0 limb mesenchyme by qRTPCR (8 genes) (Fig 4). The 28 genes exhibiting decreased expression in response to the LncRNA-HIT siRNA treatments were queried as a group for ontological function using the AmiGO 2 term enrichment software which identified proximal/distal pattern formation and regulation of chondrocyte differentiation (p≤ 0.05) as the most significant ontological categories, suggesting a role for LncRNA-HIT in the regulation of genes required for limb chondrogenesis (S6 Fig, and Methods).
To test whether LncRNA-HIT regulates chondrogenesis in the limb mesenchyme, an in vitro micromass chondrogenesis assay was used, as it recapitulates many of the cellular events occurring during chondrogenesis including mesenchymal cell condensation, cartilage extracellular matrix expression, and cartilage nodule formation and is amenable to siRNA-mediated knockdown of pro-chondrogenic genes including Runx1, Ctgf, Notch, and Angptl4 [27, 55–62]. Transfection of limb mesenchyme with siRNAs specific for LncRNA-HIT significantly reduced LncRNA-HIT RNA levels by nearly eighty percent, confirming the effectiveness of the siRNAs to target LncRNA-HIT for degradation in the micromass assay (n = 6 independent replicates, Fig 5). Analysis of the micromass assays transfected with the LncRNA-HIT siRNAs revealed a substantial reduction in chondrogenesis, resulting in the formation of fewer cartilage nodules that stained with alcian blue (n = 6 independent assays, Fig 4). In contrast, parallel micromass assays using the same populations of limb mesenchyme cells transfected with a scrambled control siRNA confirmed that the transfected limb mesenchyme was competent to undergo chondrogenic differentiation, exhibiting robust cell condensation and the formation of multiple alcian blue positive cartilage nodules (n = 6 independent assays, Fig 5).
Next, if chondrogenesis is facilitated by LncRNA-HIT’s recruitment of p100/CBP complexes, then siRNA-mediated reduction in Snd1, which encodes p100, should also affect the chondrogenic capacity of the undifferentiated limb mesenchyme. Testing this hypothesis, micromass assays using limb mesenchyme transfected with Snd1 siRNAs produced a similar loss in Snd1 transcript levels (75% knockdown) and reduced chondrogenesis, resulting in the formation of few cartilage nodules staining with alcian blue (n = 6 independent assays, Fig 6). Control transfections using a scrambled Snd1 siRNA consistently resulted in robust cartilage formation producing numerous cartilage nodules staining positively for alcian blue (n = 6 independent assays, Fig 5), supporting the hypothesis that LncRNA-HIT and its recruited proteins are essential for chondrogenic differentiation of the early limb mesenchyme.
The decrease in Hoxa13 and Hoxa11 expression in response to the LncRNA-HIT siRNA treatments (Fig 4) suggested that LncRNA-HIT may be functioning as an enhancer RNA, using its recruitment of p100/CBP to its site of transcription to promote neighboring 5’ HoxA gene expression through its maintenance of H3K27ac. Testing this hypothesis, we first examined whether siRNA-mediated reductions in LncRNA-HIT or Snd1 impact expression of additional genes proximal to the nascent site of LncRNA-HIT transcription. qRTPCR analysis of limb mesenchyme treated with LncRNA-HIT- or Snd1 (p100)-specific siRNAs revealed significant reductions in expression for all 5’HoxA gene members (Fig 7), suggesting that LncRNA-HIT is functioning as an enhancer lncRNA to mediate expression of the 5’ HoxA genes. Interestingly, no changes in expression were detected for Creb5 and Skap2, which flank the HoxA cluster, and for Hoxd13, suggesting a specific effect on the 5’ HoxA genes which are proximal to the site of LncRNA-HIT transcription (Fig 7). Changes in 3’ HoxA gene expression (Hoxa1-Hoxa7) could not be determined, as their expression in the limb mesenchyme is below the level of detection by qRTPCR. Finally Col2a1, a strong indicator of chondrogenic differentiation, also exhibited reduced expression in response to the LncRNA-HIT and Snd1 siRNA treatments confirming the disruption in chondrogenesis exhibited by the micromass assays (Figs 5–7).
Next we hypothesized that H3K27ac should also be affected by siRNA-mediated reductions in LncRNA-HIT as CBP recruitment would be concomitantly affected in the limb mesenchyme. Mapping of the H3K27ac-tagged chromosomal regions in the E 10.5 limb was previously reported (GEO Dataset: GSE30641, [63]). Using this dataset, the H3K27ac sites proximal to the LncRNA-HIT -associated loci were examined for changes using an acetylation-specific H3K27 antibody and quantitative chromatin immunoprecipitation (qChIP) (Fig 8). Starting with the site of nascent LncRNA-HIT transcription, qChIP analysis of fourteen H3K27ac-tagged chromosomal regions between Hoxa11 and Hoxa13 revealed consistent reductions in precipitated fragment enrichment in response to the LncRNA-HIT siRNA treatments (n = 3 independent assays) (Fig 7). Most notable was a greater than five-fold reduction in H3K27ac fragment enrichment for the region associated with exon 2 of Hoxa13 (Fig 8, fragment 7). A greater than 3-fold decrease in H3K27ac chromatin fragment enrichment was also detected in the Hoxa13 intronic region (Fig 8, fragment 11) and in chromatin fragments more proximal to Hoxa11 (Fig 8, fragment 1). Several H3K27ac regions were also identified within the LncRNA-HIT locus (Fig 8B). qChIP analysis of these regions revealed reduced enrichment for the entire region, with fold change decreases ranging from 1.5 to 2.8 (Fig 8, fragments 2–6).
Expanding this analysis, we examined additional chromosomal loci bound by LncRNA-HIT for changes in H3K27ac fragment enrichment in response to the siRNA treatments. For this analysis, we selected the H3K27ac sites most proximal to five loci exhibiting the greatest decrease in expression in response to the LncRNA-HIT siRNA treatments: Pik3cb, Col14a1, Bmpr1b, Pbx1, and D15ertd621e (Fig 4). qChIP analysis revealed consistent reductions in immunoprecipitated H3K27ac fragments for four of the five candidate loci (n = 3 independent assays) including Pik3cb (> 4-fold decrease), Col14a1 (> 3-fold decrease), Pbx1 (>1.4-fold decrease) and D15ertd621e (>2-fold decrease) (Fig 9). Attempts to amplify the immunoprecipitated H3K27ac fragments proximal to the LncRNA-HIT associated region at the Bmpr1b locus were unsuccessful using several primer pair combinations and sheared chromatin from the LncRNA-HIT—or control-siRNA treated samples, suggesting that H3K27ac in this region is insufficient to facilitate immunoprecipitation by the acetylation-specific H3K27 antibody (Fig 9).
Finally to test whether changes in H3K27ac peak enrichment was specifically affected by the LncRNA-HIT siRNA treatments we examined loci lacking associated LncRNA-HIT peaks for changes in H3K27ac fragment enrichment. Analysis of H3K27ac fragments associated with the promoter regions of Creb5 and Skap2 revealed no changes in fragment enrichment in response to LncRNA-HIT siRNA treatments, supporting the conclusion that LncRNA-HIT-p100/CBP is functioning to maintain H3K27ac status in regions bound by the lncRNA-protein complex (Fig 10).
We next examined whether the recruitment of p100 by LncRNA-HIT contributes to the regulation of near-peak gene expression. To address this question, an RNA tethering assay was used to dissect the gene-regulatory contributions of LncRNA-HIT and p100 towards the activation of a synthetic UAS-luciferase reporter (Fig 11). Recruitment of LncRNA-HIT to the reporter locus was facilitated by adding five copies of the BoxB transcript to the LncRNA-HIT RNA. The BoxB transcript is strongly bound by the RNA binding protein λN, which when fused to the GAL4 DNA binding domain facilitates the recruitment of the BoxB-LncRNA-HIT transcript to the UAS luciferase locus (Fig 11) [5,64]. Co-transfection of the UAS-luciferase reporter, λN-GAL4, and p100-GAL4 expression vectors resulted in no activation of the UAS luciferase reporter, indicating that the recruitment of these factors to the synthetic UAS locus is not sufficient to activate gene expression (Fig 11). Similarly, co-transfection of the UAS-luciferase reporter and expression vectors encoding λN-GAL4, p100-GAL4, and the BoxB transcript lacking LncRNA-HIT also resulted in no activation of the UAS luciferase reporter, indicating that the recruitment of BoxB, p100, and λN to the synthetic locus is not sufficient to activate gene expression (Fig 11A and 11C). Additionally, luciferase expression was not detected in cells transfected with UAS-luciferase reporter and expression vectors encoding BoxB-LncRNA-HIT and λN-GAL4, suggesting that the recruitment of LncRNA-HIT to the UAS locus is not sufficient to activate gene expression (Fig 11B and 11C). Finally, co-transfections using the UAS-luciferase reporter, BoxB-LncRNA-HIT, λN-GAL4, and p100-GAL4 expression vectors resulted in strong activation of the luciferase reporter in a dosage-dependent manner, indicating that co-recruitment of p100 and LncRNA-HIT to a locus is required to activate gene expression (Fig 11A and 11C).
The development of the vertebrate limb requires a coordinated series of cellular events to facilitate the formation of bone and articular cartilage. Initiating this process is the proliferation and migration of mesenchymal progenitor cells from the lateral plate mesoderm. Once in the limb bud, the mesenchymal cells condense forming cartilage templates for all skeletal tissues. Finally the chondrocytes within these templates mature, forming bone through endochondral ossification or remain as a specialized population of chondrocytes to form the articular cartilage [65–68]. Interestingly, while SOX9 appears to be essential for many of the cellular processes required for skeletal development, emerging evidence suggests that an epigenetic component, mediated by lncRNAs, may also play an important role in the formation, maintenance, or pathology of the appendicular skeletal tissues [5,34,69–73]. Given the large number of lncRNAs expressed by the human (>15,000) and mouse genomes (>7600), it is likely that additional roles for these molecules will be identified in the appendicular cartilage and bone (GENCODE ver. 22, [74]). In this report, we identify the lncRNA, LncRNA-HIT, as an essential component for chondrogenesis, regulating multiple genes to facilitate the formation of cartilage from undifferentiated limb mesenchyme.
Ontological analysis of loci contacted and regulated by LncRNA-HIT suggests that the lncRNA coordinates the expression of genes whose products mediate chondrogenic differentiation in the limb mesenchyme. This conclusion is consistent with the decrease in cartilage nodule formation in response to the LncRNA-HIT and Snd1 siRNA treatments. Early studies of cartilage formation identified disruptions in the initial condensation of the limb mesenchyme as a primary cause of reduced cartilage formation [75–76]. More importantly the regulation of mesenchymal condensation was recently shown to require BMP-SMAD4 signaling, independent of SOX9 function [77]. This finding provides a mechanism to explain the loss of cartilage formation in the micromass assays, as Bmpr1b, a major receptor for GDF5 and other BMPs during limb development, is significantly down-regulated in response to the LncRNA-HIT siRNA treatments (Fig 4) [78–80]. In the absence of BMPR1B, the primary chondrogenic phenotype exhibited by humans and mice is the loss of prechondrogenic condensations and skeletal element hypoplasia, particularly in the distal limb, a region that also strongly expresses LncRNA-HIT (Fig 1) [80–83]. Reductions in Bmpr1b expression in response to the LncRNA-HIT siRNA treatments could also reflect a loss in p100 enrichment. This conclusion is supported by the RNA tethering assay which determined that interactions between LncRNA-HIT and p100 were sufficient to activate gene expression from a synthetic locus. By this mechanism, LncRNA-HIT could regulate Bmpr1b as well as additional loci by stabilizing p100 at the Bmpr1b locus; providing a functional explanation for the 4.75-fold decrease in Bmpr1b expression in response to the LncRNA-HIT siRNA treatments. Alternatively, the down-regulation of Bmpr1b expression in response to the LncRNA-HIT siRNA treatments could also reflect indirect regulation of Bmpr1b by another LncRNA-HIT-regulated gene product.
A second component contributing to the control of chondrogenesis by LncRNA-HIT is its regulation of Hoxa13 and Hoxa11 which previous studies indicate can regulate additional components of the BMP-signaling cascade including Bmp2 and Bmp7 by HOXA13 and Runx2 by HOXA11 [84–85]. Thus in the absence of LncRNA-HIT function, the expression of receptor, ligand, and transcriptional components of the BMP signaling pathway are reduced, which in combination, explains the reductions in mesenchymal cell condensation and cartilage nodule formation exhibited by the micromass assays treated with the LncRNA-HIT siRNAs.
The down-regulation of 5’ HoxA genes in response to LncRNA-HIT siRNA treatments suggests that LncRNA-HIT may be functioning as an enhancer lncRNA. Recent studies of enhancer lncRNAs indicate that these molecules regulate the expression of nearby genes by recruiting chromatin modifying proteins to increase the accessibility of the chromosomal region to gene-regulatory factors [6,22,86]. In the limb mesenchyme, our detected decrease in H3K27ac and 5’ HoxA gene expression in response to LncRNA-HIT siRNA treatments support a role for LncRNA-HIT as an enhancer lncRNA using its associated proteins (p100 and CBP) to facilitate expression of the surrounding 5’ HoxA genes (Fig 12). It is interesting to speculate that the presence of LncRNA-HIT within the 5’ HoxA cluster in many vertebrate genomes (S1 Fig) may reflect its conservation as an essential enhancer of 5’ HoxA gene expression, providing a fundamental epigenetic mechanism to promote chondrogenic differentiation; presumably through its recruitment of the p100/CBP complexes. P100/CBP complexes are important components of the STAT6-dependent enhanceosome which functions as a transcriptional enhancer of genes involved in the interleukin 4 gene regulatory cascade [47–48,87]. In limb mesenchyme, LncRNA-HIT may be working in a similar capacity, forming a complex with p100 and CBP to regulate pro-chondrogenic gene expression.
Interestingly, the results from the RNA tethering assay also suggest that interactions between LncRNA-HIT and p100 may be required to confer functional competence to the RNA protein complex, as the activation of the luciferase reporter occurred only when p100 and LncRNA-HIT were co-expressed. Recent studies support this conclusion, as mutations in WDR5 that prevent lncRNA binding impacts WDR5-MLL chromatin occupancy, maintenance of H3K4 trimethylation, and the regulation of genes necessary for embryonic stem cell self-renewal [46]. Taken together these findings suggest that LncRNA-HIT promotes gene expression by two mechanisms, first: by binding to p100 to stabilize the protein to promote its co-activator function and second: through its recruitment of CBP to promote chromatin accessibility by maintaining H3K27ac which in the 5’ HoxA region facilitates the expression of multiple factors whose products instruct the limb mesenchyme to form skeletal tissues [27, 30–31,42, 88–89].
Transcriptional profiling of breast tumor cell lines initially identified LncRNA-HIT as a TGFβ-induced transcript [35]. Interestingly, TGFβ1/SMAD signaling can also promote Snd1 (p100) expression which can facilitate epithelial to mesenchymal transitions (EMT) in mammary tumor cell lines, a causal event in tumor cell migration and invasion [90–94]. In these tissues, Tgfβ should induce both LncRNA-HIT and Snd1 (p100) which would activate Hoxa13, which has been previously shown to mediate the formation of vascular tissues, providing a mechanism to promote tumor vascularization and growth [95].
Interestingly several studies investigating predicative biomarkers of cancer progression and patient outcomes in hepatocellular carcinoma have implicated the expression HOXA13 and HOTTIP as indicators of poor prognosis with high disease progression [96–97]. Given that the physical distance between LNCRNA-HIT, HOXA13, and HOTTIP is only 6.5 kb in both humans and mice, it is possible that LNCRNA-HIT’s function may also be co-opted in these cancers. If human LNCRNA-HIT is determined to be expressed in hepatocellular carcinoma, a ChIRP-seq approach could be used to determine the loci contacted by the lncRNA-protein complex, providing additional therapeutic targets for this disease. Moreover, several non-chondrogenic regions also express LncRNA-HIT during murine development including the gut epithelium and the genital tubercle, suggesting that the lncRNA may have additional roles in the formation of these tissues. Using the approaches outlined in this study, including ChIRP-seq and RNA affinity chromatography the proteins interacting with LncRNA-HIT in these tissues as well as the loci contacted by the protein-lncRNA complexes can be identified providing new insights into the developmental regulation of these tissues. To discern the in vivo functions of LncRNA-HIT during development will be more challenging, as targeted disruptions of LncRNA-HIT should affect 5’ HoxA gene expression either by disrupting activation mediated by the enhancer RNA functions of LncRNA-HIT -p100/CBP complexes or by disrupting cis-regulatory elements required for 5’ HoxA gene expression. By either mechanism 5’ HoxA gene expression would be reduced, creating phenotypes that cannot be solely attributed to the loss of LncRNA-HIT function.
LncRNA-HIT is also capable of functioning as a guide RNA as it can recruit the same proteins to loci independent of its site of transcription. By this process LncRNA-HIT-p100/CBP complexes can regulate a larger cohort of genes. As an epigenetic modulator of chondrogenesis, LncRNA-HIT, may be one of several non-coding RNA species required for the formation and/or maintenance of cartilage tissues including micro RNAs MiR-145 and 337 as well as the small nucleolar RNAs U38 and U48 which are elevated in the serum of individuals affected by injury-induced osteoarthritis [98–99]. These studies in conjunction with the present work indicates a higher level of gene regulation is necessary for the initial formation of cartilage tissues as well as its loss during osteoarthritis, providing additional targets to exploit as therapies to minimize joint tissue loss as a consequence of injury or disease.
All investigations using mice were certified as compliant with AVMA guidelines by the OHSU IACUC board prior to implementation following an approved mouse use protocol IS00001648 to HSS. Euthanasia was conducted using CO2 gas until all signs of movement and respiration ceased, followed by prompt thoracotomy. This method is consistent with the recommendations of the Panel on Euthanasia of the American Veterinary Medical Association. PHS 398/2590 (Rev. 06/09).
Embryonic tissues were derived from timed matings of wild type Swiss Webster mice (Charles River Labs). Gestational development was measured in embryonic days (E) where E = 0.5 reflects the detected day of a vaginal plug.
Limb bud mesenchyme was dissected from E 11.0 embryos, dissociated, and grown in culture in two chamber cover-glass slides (Nunc Lab- TEKII, Thermo Scientific) as described [55]. After 24H the cells were fixed for 10 min using 4% formaldehyde/PBS and prepared for RNA FISH following the protocol described by the Lanctôt laboratory (http://lanctotlab.org/en/protfish_rnafisholigo20mer.html). The LncRNA-HIT RNA FISH probe sets were synthesized and labeled with CalFluor Red 610 by Stellaris Biosearch Technologies (S1 Table). Cells were imaged using a Zeiss Axioplan epifluorecence microscope fitted with a 100x 1.4 NA oil objective and a cooled monochrome CCD camera using a 2 second exposure. Detected signals were pseudo-colored red (LncRNA-HIT), green (Gapdh), and blue (DAPI) using the Zeiss AxioVision digital image processing software (release 4.7.2).
Approximately 1mg of E 11.5 limb bud lysates were used to immunoprecipitate p100 and CBP with endogenous LncRNA-HIT transcripts following the previously described RIP procedure [5, 68] using antibodies specific for SND1 (p100) (Cat. Ab65078 Abcam, Cambridge, MA) and CBP (Cat. 7389 Cell Signaling, Danvers, MA). After immunoprecipitation of p100, CBP, or IgG, the co-immunoprecipitated RNA was extracted using TRIzol following the manufacturer’s protocol (ThermoFisher). After extraction, the isolated RNA was treated with RQ1 Rnase-Free Dnase (Cat. M6101 Promega, Madison, WI) and evaluated for enrichment of LncRNA-HIT using qRTPCR as described in the RNA isolation and qRTPCR section of the Methods. Three independent immunoprecipitation assays were performed for p100, CBP, and the IgG control analysis of LncRNA-HIT co-immunoprecipitation.
Embryos were collected at E 10.5 as described [55]. Cells were transfected with two small interfering RNAs (siRNAs) specific for LncRNA-HIT or with a scrambled siRNA control at a concentration of 10 nM using Trifectin as recommended (IDT, Coralville, Iowa). 24H after transfection, the cells were trypsinized and resuspended to a final concentration of 2x107 cells/ml and seeded as 10 ul drops into 60 mm Falcon tissue cultures as described [56]. The LncRNA-HIT siRNA sequences used were:
For Snd1 repression, limb mesenchyme was transfected using a commercial Snd1 siRNA cocktail at 30 nM concentrations as recommended by the manufacturer (Life Technologies: Cat: 4390771, Grand Island, NY).
25 oligonucleotide probes corresponding to the murine LncRNA-HIT transcript were selected using the Stellaris Probe Designer Software set to masking level 5 (biosearchtech.com) (S2 Table). The oligonucleotide sequences were synthesized with an 18-atom spacer followed by a biotin tag located at the 3’ end and purified by HPLC as described by the manufacturer (IDT, Coralville, Iowa). The autopod regions of the E 11 limbs were dissected and digested for 10 minutes in sterile PBS containing 0.1% trypsin and 0.1% collagenase as described [56]. Approximately 2000 embryonic limb buds were required to produce the 200mg cells needed for each ChIRP assay. Cell lysis and chromatin shearing and streptavidin bead precipitation were accomplished using a Bioruptor instrument (Diagenode) as previously described [50]. As a control, chromatin from murine glial cells (a kind gift from Dr. Peter Hurlin), that do not express LncRNA-HIT, were used in parallel assays and hybridized with the same LncRNA-HIT probes. The precipitated DNA was quantified using a Qubit 2.0 DNA fluorometer (Life Technologies) and submitted to Elim Biopharm (Hayward, CA) for library preparation and next generation sequencing using an Ilumina HiSeq 2500 sequencer.
The LncRNA-HIT ChIRP-seq libraries were mapped, normalized, and analyzed using the ChIRP-seq analysis pipeline previously described [51]. Raw reads from each ChIRP-seq library were uniquely mapped to the reference mouse genome (mm9) using bowtie-0.12.9 and normalized to total read count as described [52–53]. Even and odd LncRNA-HIT ChIRP samples were merged by taking the minimum value of the even and odd tracks at each genomic position. The mouse glial cell line that does not express LncRNA-HIT was used as a control to identify false hybridization peaks detected by the LncRNA-HIT-specific even and odd probe pools. Peaks were called by MACS using the LncRNA-HIT ChIRP merge sample and the DNA input, with an initial threshold p-value ≤ 5.9 x 10−57 corresponding to 775 peaks based on visual inspection of the peaks using the UCSC Genome Browser. Peaks were additionally filtered by visual inspection with UCSC Genome Browser excluding any peak with a corresponding peak in the glial controls. The limb-specific peaks were then submitted for analysis using the Genomic Regions of Enrichment of Annotations Tool (GREAT) [54]. Following GREAT the peaks were additionally filtered to identify high confidence peaks mapping within 25 kb of a known gene resulting in the identification of 42 LncRNA-HIT-associated genes by visualization using the UCSC Genome Browser. The LncRNA-HIT ChIRP-seq data sets have been submitted to the NCBI GEO repository under the accession number, GSE70986.
42 near-peak genes were validated for regulation by LncRNA-HIT -p100/CBP using siRNA-mediated reduction of the native LncRNA-HIT transcript followed by gene-specific qRTPCR in limb mesenchyme primary cultures. For each gene, a minimum of three independent LncRNA-HIT siRNA treatments followed by qRTPCR was performed. A scrambled siRNA control was used in parallel assays for each validation experiment. Genes exhibiting decreased expression ≤ -1.5-fold, were assessed for their collective ontological function using the GO consortium ontology tool kit, AmiGo version 2.1.4, using the criteria: experimental biological processes in M. musculus (http://amigo2.berkeleybop.org/amigo) [100].
RNA was isolated using TRIzol (Life Technologies, Grand Island, NY) as instructed by the manufacturer. Gene expression levels were quantitated using a real-time quantitative reverse transcriptase polymerase chain reaction method (qRTPCR). First-strand cDNA was synthesized using an ImProm-II Reverse Transcription System (Promega, Madison, WI). A minimum of three independent samples were used for qRTPCR using a SYBR Green PCR Super Mix and an IQ5 thermal cycler according to the manufacturer’s instructions (BioRad Hercules, CA). Fold change expression levels were determined after normalization of the amplification products to Gapdh expression using the BioRad IQ5 software suite. Student’s t-tests were used to determine statistical significance. Data was plotted using Sigmaplot 10.0 (Systat, San Jose, CA). Primer sequences used for the qRTPCR assays are presented in S3 Table.
Cartilage nodule formation in the siRNA-treated micromass cultures was visualized by staining the micromass cultures with Alcian blue 8GX (Canemeo Inc, QC, Canada), as described [101]. Alcian blue stained nodules were photographed using a Leica MZFLIII stereoscope fitted with a Canon EOS 40D digital camera.
RNA tethering assays were performed using expression vectors encoding GAL4-p100 andGAL4-λN as described [5, 64]. Expression plasmids containing five copies of the BoxB transcript or containing the same five copies of the BoxB transcript fused to the LncRNA-HIT transcript (BoxB-LncRNA-HIT) were subsequently transfected into NG108-15 cells with the GAL4-λN and/or the GAL4-p100 expression vectors using increasing dosages of the BoxB or BoxB-LncRNA-HIT vectors at 0, 50, 100, or 250 ng per assay and assessed for luciferase expression 48H after transfection using the Dual-Luciferase Reporter Assay System (Promega, Madison, WI). All analyses were performed in triplicate and the average and standard deviation were plotted using Sigmaplot 10.1.
Flag-WDR5 293T cells were transfected with a LncRNA-HIT expression vector and evaluated for WDR5-LncRNA-HIT interactions by RNA protein co-immunoprecipitation as described [102].
Mouse LncRNA-HIT and U1 genes were cloned into the pCDNA3 vector (Life Technologies, Grand Island, NY) and transcribed using T7 RNA polymerase and biotin-16-UTP as described by the manufacturer (Roche, Indianapolis, IN). Cell lysates from E11.0 wild type embryos limbs which were homogenized in mRIPA buffer and centrifuged as described [103]. 1 mg of total protein was incubated with either the LncRNA-HIT or U1 transcripts tagged with biotin-16-UTP 12 H at 4°C. Following incubation, the RNA-protein mixtures were incubated with 100 μl prewashed streptavidin beads for 2.5 H at 4°C. After incubation, the streptavidin beads were washed five times in ice cold PBS and combined with 50 μl of an SDS elution buffer containing 125 mM Tris/HCl pH 6.8, 20% glycerol, 4% SDS, 10% 2-mercaptoethanol and 0.02% bromphenol blue. After two rounds of elution, the recovered protein mixture was heated to 95°C for 5 minutes and fractionated by SDS-PAGE electrophoresis using a 4–12% gradient gel (Nu Page, Invitrogen). The gels were stained with Coomassie blue and the individual bands digested with trypsin and submitted for identification by mass spectroscopy by the OHSU Proteomics Core Facility as described [104].
The antisense riboprobe specific for mouse LncRNA-HIT was generated using PCR amplification of the gene-specific sequence. The amplifying primers were: 5’-AGAGGAGGTTCCCAGACTCC-3’, and 5’-GCACACAAACACTGATATGCAA-3’. Riboprobe synthesis and in situ hybridization was performed on mouse embryos as described [105].
Limb buds from E 11.0 embryos were dissected, and the mesenchymal cells were transfected with the previously described LncRNA-HIT siRNAs. 24 H after transfection, qChIP assays were performed as described [106–108], using an anti-acetyl histone H3 (lys27) antibody (Cat. 07–360, Millipore, Billerica, MA) or mouse IgG (Diagenode: C15400001) as a negative control. DNA sequences specific to H3K27ac peaks proximal to LncRNA-HIT -associated peaks were identified using an overlay of the E 10.5 limb bud H3K27ac ChIP-seq dataset [63] and the UCSC Genome Browser. Primers were selected to amplify the DNA regions containing the H3K27ac peaks using Primer3 (simgene.com) (S4 Table). ChIP-enriched H3K27ac fragments were quantified by qPCR using an IQ5 quantitative PCR instrument (BioRad, Hercules, CA) as described [107–108].
|
10.1371/journal.pntd.0006510 | Mother-to-child transmission of Chikungunya virus: A systematic review and meta-analysis | Chikungunya virus (CHIKV) is an emerging arboviral infection with a global distribution and may cause fetal and neonatal infections after maternal CHIKV-infections during gestation.
We performed a systematic review to evaluate the risk for: a) mother-to-child transmission (MTCT), b) antepartum fetal deaths (APFD), c) symptomatic neonatal disease, and d) neonatal deaths from maternal CHIKV-infections during gestation. We also recorded the neonatal clinical manifestations after such maternal infections (qualitative data synthesis). We searched PubMed (last search 3/2017) for articles, of any study design, with any of the above outcomes. We calculated the overall risk of MTCT, APFDs and risk of symptomatic neonatal disease by simple pooling. For endpoints with ≥5 events in more than one study, we also synthesized the data by random-effect-model (REM) meta-analysis.
Among 563 identified articles, 13 articles from 8 cohorts were included in the quantitative data synthesis and 33 articles in the qualitative data synthesis. Most cohorts reported data only on symptomatic rather than on all neonatal infections. By extrapolation also of these data, the overall pooled-MTCT-risk across cohorts was at least 15.5% (206/1331), (12.6% by REMs). The pooled APFD-risk was 1.7% (20/1203); while the risk of CHIKV-confirmed-APFDs was 0.3% (3/1203). Overall, the pooled-risk of symptomatic neonatal disease was 15.3% (203/1331), (11.9% by REMs). The pooled risk of symptomatic disease was 50.0% (23/46) among intrapartum vs 0% (0/712) among antepartum/peripartum maternal infections. Infected newborns, from maternal infections during gestation were either asymptomatic or presented within their first week of life, but not at birth, with fever, irritability, hyperalgesia, diffuse limb edema, rashes and occasionally sepsis-like illness and meningoencephalitis. The pooled-risk of neonatal death was 0.6% (5/832) among maternal infections and 2.8% (5/182) among neonatal infections; long-term neurodevelopmental delays occurred in 50% of symptomatic neonatal infections.
Published cohorts with data on the risk to the fetus and/or newborn from maternal CHIKV-infections during gestation were sparse compared to the number of recently reported CHIKV-infection outbreaks worldwide; however perinatal infections do occur, at high rates during intrapartum period, and can be related to neonatal death and long-term disabilities.
| Chikungunya virus (CHIKV) is an emerging arboviral infection with a global distribution and can cause infections of the fetus and newborn after maternal CHIKV-infections during gestation. In this systematic review, we evaluated the risk for mother-to-child transmission (MTCT), antepartum fetal deaths (APFD) and symptomatic neonatal disease from maternal CHIKV-infections during gestation. Whenever meaningful, we also synthesized the data by random-effect-model (REM) meta-analysis. We also recorded the list of clinical manifestations of neonatal infections after maternal infections during gestation. Overall, published cohorts with pertinent data to estimate the impact to the fetuses and newborns of maternal CHIKV-infections were sparse compared to the number of recently reported CHIKV-infection outbreaks worldwide. Most cohorts reported data only on symptomatic neonatal infections rather than on all (symptomatic and asymptomatic) neonatal infections. By extrapolation also of these data, the pooled MTCT-risk was at least 15.5% (206/1331), (12.6% by REMs). Symptomatic disease occurred almost exclusively with maternal infections around the time of delivery. Overall, the pooled risk of symptomatic disease was 15.3% (203/1331), (11.9% by REMs); however, the risk of symptomatic disease from intrapartum maternal infections was 50.0% (23/46) vs 0% (0/712) from antepartum/peripartum maternal infections. The pooled APFDs-risk was low (1.7%); however, APFDs occurred with maternal infections in all trimesters. Infected newborns were either asymptomatic or presented during their first week of life, but not at the time of birth, with manifestations such as fever, irritability, rashes, hyperalgesia syndrome, diffuse limb edema, bullous dermatitis and occasionally also meningoencephalitis. Long-term neurodevelopmental delays occurred in 50% of symptomatic neonatal infections.
| Chikungunya virus (CHIKV) is a remerging arbovirus[1–5] in the family of Togaviridae, genus Alphavirus that is transmitted by the Aedes spp. mosquitos A. aegyptii and A. albopictus[6] causing a crippling musculoskeletal inflammatory disease in humans characterized by fever, polyarthralgia, myalgia, rash, and headache.[7] It was first identified in Tanzania in 1953[8], and the name comes from a Makonde word that means “that which bends up” due to the position taken by patients suffering from the severe joint pain.[9, 10] Since then, it has caused outbreaks in Africa[5, 11, 12], Indian Ocean islands, South East Asia [13–15], Central and South America[16–18], US territories[19] and Europe.[20–23] CHIKV has now been identified in 94 countries worldwide.[6, 24] CHIKV infections reemerged in India after a gap of 32 years with an estimated 1.38 million people been infected by the end of 2006; the outbreak subsequently declined and by 2009 there were only a few thousand cases reported yearly.[25, 26]La Reunion Island, a French territory in the Indian Ocean, had the best-studied epidemic, and over one third of the inhabitants of the island were affected in the 2005–2006 outbreak.[27] In other outbreaks, such as in India and Malaysia much higher post-outbreak seropositivity rates were reported (62%-68%[28, 29] and 56%[30] respectively).
The first case of CHIKV infection in the Western Hemisphere was reported in 2013[18], and it has now rapidly spread to 44 countries in the Americas,[19, 31] including also US territories and the Caribbeans.[32, 33] In the US, in 2014 there was the first report of local- autochthonous CHIKV transmission in Florida.[34] Among 2,799 CHIKV cases reported to ArboNET in 2014 from US states, 12 cases were locally-transmitted (from Florida); while all the remaining cases were from returning travelers from endemic areas. In contrast, 99% of the 4,710 CHIKV cases reported from US territories were locally-transmitted. CHIKV infection became a nationally notifiable condition in 2015.[4] The number of CHIKV-infections reported in the US, declined after 2015 and in 2017 there were only 36 reported cases from the US, with no locally transmitted cases, and 30 cases from US territories where local transmission continues.[35]
The almost global distribution of CHIKV as well as the possibility for autochthonous transmission in the US make CHIKV infections a threat to global health and also to domestic health in the US. According to the CDC predictive model estimates for the US based on climate data, the potential range where the Aedes aegypti and Aedes albopictus mosquitos could potentially live, survive and reproduce in the US is quite extensive.[36–38]These mosquitos are capable of transmitting other arboviral infections as well, except for CHIKV.[39, 40]
Despite the almost global distribution of CHIKV infection, data for the impact of acute CHIKV infections during pregnancy are sparse and uncertainties remain on several important clinical questions. The best studied maternal-fetal cohort for CHIKV infections during gestation is from La Reunion CHIKV outbreak in the Indian Ocean in 2005–2006[27]. Prior reviews on this specific question were not exhaustive on their searches and focused only on very few studies [41, 42]. The maternal-fetal data from all available cohorts and for all important fetal and neonatal clinical outcomes has not been previously systematically evaluated. We set up to perform a systematic review and when meaningful, synthesize also the data by meta-analysis, to address the following questions: a) What is the overall risk of Mother-To-Child-Transmission (MTCT) from maternal CHIKV infections during gestation. b) What is the risk for antepartum fetal deaths (APFDs) from maternal CHIKV-infections during gestation. c) How often maternal CHIKV-infections during gestation lead to symptomatic neonatal disease and d) whether the reported risk-differences (for MTCT risk and risk of symptomatic disease) from maternal infections during the intrapartum period vs antepartum or peripartum period are consistent across diverse cohorts. Moreover, we wanted to record the spectrum of clinical manifestations reported in the scientific literature for neonatal CHIKV infections from maternal acute CHIKV-infections during gestation.
We performed a systematic review to address the above questions and when meaningful, we also synthesized the data by meta-analysis. We searched PubMed and CINAHL databases (last search 3/2017) using the following terms: “chikungunya” and a pregnancy-related term that included any of the following terms: pregnan*, neonat*, perinat*, infant, mother, congenital, vertical transmission, miscarriage, abortion; limiting search results to human studies. Eligible for inclusion at the initial screening at title/abstract level were articles that studied CHIKV-infection in pregnant women, reported outcomes for the fetuses and/or newborns and had an English abstract. Potentially eligible articles were further screened in full text. The reference lists of key pertinent articles were also screened. Review article without original data were excluded. Article screening was done by two independent investigators (SNL and CC) and full texts of potentially eligible articles were also screened by a third investigator (DCI) and consensus was reached. Data extraction from the eligible articles was done by two independent investigators (SNL and CC) and confirmed by a third investigator (DCI).
For the quantitative data synthesis, we included cohort studies, case series or case control studies that provided data on maternal CHIKV-infections during gestation and the CHIKV-infection status of their fetuses and/or newborns to allow for the calculation of the risk for: a) mother-to-child transmission (MTCT), b) antepartum fetal deaths (APFD), c) symptomatic neonatal disease, and d) neonatal deaths from maternal CHIKV-infections during gestation. We also calculated the overall combined fetal/neonatal disease impact of maternal CHIKV infection during gestation for a composite outcome of symptomatic neonatal disease plus the APFDs.
For the qualitative data synthesis, we considered studies of any study design, including case reports, that reported clinical manifestations of neonates exposed to maternal-CHIKV-infections during gestation. Reports of postnatally acquired CHIKV-infections from mosquito exposure were excluded.
For endpoints with ≥5 events in more than one study, we also synthesized data by random-effect- model meta-analyses.[43]
From each eligible study for the quantitative data synthesis we extracted the following information: authors, year, locations, year of study, period of recent regional CHIKV-infection outbreak, duration of study, any possible overlap with prior published reports from the same cohort, number of pregnant women infected during gestation, number of neonatal infections from maternal infections during gestation, number of neonatal infections from intrapartum (-2 ds prior-to-delivery to +2 ds post-delivery), peripartum (-7 ds to -3 ds prior-to-delivery) and antepartum (>7 ds prior-to-delivery) maternal infections, number of antepartum fetal deaths, number of CHIKV-confirmed APFDs, number of symptomatic neonatal infections, number of symptomatic neonatal infections from intrapartum, peripartum and antepartum maternal infections, number of neonatal deaths, methods for ascertainment of maternal and neonatal CHIKV-infections. For the qualitative data synthesis, we extracted information on the clinical manifestations of neonatal infections documented to have occurred from suspected or confirmed maternal infections during gestation.
Data were synthesized across cohorts by simple pooling. For each outcome of interest (MTCT-risk, APFD-risk, CHIKV-confirmed-APFD-risk, Symptomatic neonatal disease-risk; Neonatal death-risk) we calculated -for each cohort and across all analyzed cohorts- the pooled risk (and 95% confidence intervals thereof) among total CHIKV maternal infections during gestation (N of fetuses/neonates with the outcome of interest/ total N of CHIKV maternal infections during gestation). For the neonatal mortality outcome, we also calculated the risk of neonatal deaths among total neonatal infections.
As for the majority of the analyzed cohorts, data were reported only for symptomatic neonatal infections rather than for total neonatal infections (symptomatic plus asymptomatic), in our overall data synthesis for the MTCT-risk, we included also studies reporting only symptomatic-neonatal-disease-risk and considered that the MTCT-risk for those studies was at least equal to the risk for symptomatic neonatal disease. We also used random effect models (REMs)[43] for the calculation of the above risks to account for the between study variance. For outcomes with events <5 we used only simple pooling as REMs in such cases give unreliable results. We used the i2 test for the calculation of the between study heterogeneity. All proportion meta-analyses were done in STATASE 15.0 (Stata, College Station, TX, USA). When there were multiple publications from the same cohorts we considered in our overall data synthesis only the report with the maximum number of events for the outcome(s) of interest, per total number of analyzed maternal infections reported from the whole cohort.
In our systematic review, we followed the PRISMA (Preferred Reporting Items for Systematic reviews and Meta-Analyses) guidelines of reporting (S1 Table). [44]
Of the 563 identified articles, 13 [27, 45–56] (from 8 cohorts) with pertinent data were included in the quantitative data synthesis (Fig 1).
These pertained to data from outbreaks in La Reunion Island (n = 7), Mayotte Island (n = 1), Sri Lanka (n = 1), Thailand (n = 2) and Latin America (n = 2) (S2 Table). Furthermore, 33 articles [27, 45–51, 53–55, 57–77] were included in the qualitative data synthesis of neonatal clinical manifestations from maternal infections during gestation. (Fig 1).
In the majority of articles in the quantitative data synthesis, maternal CHIKV-infections were ascertained by maternal serology (IgM and IgG) and /or blood RT-PCR and/or maternal symptoms typical of CHIKV-infections. Only in the recent outbreak from Santo Domingo the diagnosis of maternal CHIKV-infections was based only on clinical criteria (S2 Table).
In most of the analyzed maternal/neonatal cohorts only symptomatic neonatal cases were reported among maternal CHIKV infections during gestation. By extrapolation also of these data [27, 51, 55], the overall pooled-MTCT-risk across cohorts was at least 15.5% (95% CIs: 13.57%-17.53%; 206/1331) (Table 1; S3 and S4 Tables) and the risk among maternal infections during the intrapartum period was at least 50.0% (95% CIs: 34.90%-65.10%; 23/46) vs 0% (0/712) among antepartum/peripartum maternal infections. The timing of maternal infections was analyzed only in three cohorts [27, 53, 56]; 5% of all analyzed maternal infections in these three cohorts occurred during the intrapartum period. Results by REM synthesis of data were similar (MTCT-overall risk: at least 12.6% [95% CIs 4.47%-20.77%]; MTCT-risk-intrapartum infections: at least 50.3% [3.75%-96.93%]). (Fig 2, Table 1)
The pooled-risk of APFDs was 1.7% (95% CIs: 1.02%-2.56%; 20/1203) among maternal infections (Table 1; S3 and S4 Tables). APFDs occurred with maternal infections in all trimester, including during early gestation. Ascertainment of the CHIKV-infection status of APFDs was very rarely performed and was confirmed in only three cases from La Reunion outbreak, after maternal infections at 12.5 weeks, 15 weeks and 15.5 weeks of gestation respectively.[57] The pooled-risk of CHIKV-confirmed APFD cases was 0.3% (95% CIs: 0.05%-0.73%; 3/1203).
Overall, the pooled-risk of symptomatic neonatal infections was 15.3% (95% CIs: 13.36%-17.30%; 203/1331) among maternal infections during gestation (Table 1; S3 and S4 Tables). However, this risk was 50.0% (95% CIs: 34.90%-65.10%; 23/46) among intrapartum maternal infections vs 0% (0/758) among antepartum/peripartum maternal infections. Only three maternal-fetal cohorts (La Reunion[27], ShriLanka [53] and Colombian cohort[56]) analyzed their data according to the timing of maternal infection and the reported risks for symptomatic neonatal disease from intrapartum paternal infections across these three cohorts were 48.7% (19/39) [27], 100% (4/4) [53] and 0% (0/3) [56] respectively. The reported cases of symptomatic neonatal disease were almost exclusively from intrapartum maternal infections. The majority of the cohorts did not provide information on the percentage of pregnant women with infections during the intrapartum period. Results by REM synthesis of data were similar (Symptomatic Neonatal disease-overall risk: 11.9% [95% CIs: 3.89%-19.95%]; Symptomatic Neonatal Diseases Risk-intrapartum infections: 50.3% [95% CIs: 3.75%-96.93%]) (Table 1; Fig 3).
Recording of long-term neurodevelopmental outcomes was very limited and was available only from La Reunion cohort, which showed neurodevelopmental delays at ~2 years of age in 50% of symptomatic neonatal infections (12 with CHIKV-encephalopathy and 22 with mild/moderate prostration) (S3 Table).
The pooled-risk for neonatal death was 0.6% (95% CIs: 0.20%-1.40%; 5/832) among all maternal infections and 2.8% (95% CIs: 0.90%-6.29%; 5/182) among neonatal infections. (Table 1)
The pooled combined disease impact to the fetus and newborn (MTCT and APFD) was 17.0% (95% CIs: 15.00%-19.11%; 226/1331) among maternal infections during gestation, considering both the neonatal infections and the APFDs. (Table 1). Limited data were available on the number of premature births from maternal CHIKV infections during gestation to allow for a meaningful data synthesis; however, the reported rates for premature births were low (3–8%).[47, 54, 56] (S5 Table)
In this systematic review of published data of the MTCT-risk and risk of symptomatic neonatal infection among maternal CHIKV infections during gestation, the number of identified cohorts, with pertinent data for such analyses, was very small compared to the number of recently reported CHIKV-infection outbreaks and the global distribution of CHIKV.[78] Most cohorts that reported neonatal infections had reported only symptomatic cases. By extrapolation of data also from symptomatic disease cases, the overall pooled-risk of MTCT across the 8 analyzed maternal-fetal cohorts was at least 15.5%. The risk of APFDs and CHIKV-confirmed APFDs was small (<2% and <0.5% respectively). APFDs and CHIKV-confirmed-APFDs occurred from maternal infections in all trimesters, including also during early gestation. Reporting of data on APFDs was limited across the analyzed cohorts and ascertainment of CHIKV infection status of APFD was reported for only 3 cases from La Reunion outbreak. The overall MTCT risk in our study might have been underestimated as the majority of the analyzed cohorts reported only symptomatic neonatal infections rather than on all neonatal infections. In resource poor settings, where most of the CHIKV outbreaks occurred, asymptomatic neonatal infections might have remained undiagnosed, leading to a possible selection bias among the cases studied. Selective follow-up of the sickest babies may also have skewed the results of several papers. Moreover, many had significant losses to follow-up or relied on neonatal disease incidence to estimate actual neonatal infection rates.
The overall pooled risk of symptomatic neonatal disease was 15.5% among maternal infections during gestation. However, the risk was 50.0% among intrapartum maternal infections vs 0% among antepartum/peripartum maternal infections. Data on the percentage of pregnant women infected during the intrapartum period were limited across the analyzed cohorts. Symptomatic neonatal disease occurred almost exclusively from intrapartum maternal infections. The pooled-risk for neonatal death was 0.6% among all maternal infections and 2.8% among neonatal infections. Long-term global neurodevelopmental delays have also been reported to occur in 50% of symptomatic neonatal infections during gestation, however this was based on a limited number of 33 such neonatal infections.[79]
In our qualitative data synthesis, we generated a compilation list of clinical manifestations reported in CHIKV-infected infants from maternal infections during gestation. Such infants presented with a wide spectrum of clinical manifestations ranging from asymptomatic to severely symptomatic. Symptomatic infected newborns, from maternal infections during gestation usually developed symptoms during their first week of life, but not at the time of birth. Commonly reported symptoms included fever, polyarthralgias, diffuse limb edema, irritability, poor feeding, painful syndrome and rashes; occasionally, also sepsis-like syndrome with multiple organ involvement, meningoencephalitis with brain MRI abnormalities and can also cause long term neurodevelopmental delays and devastating neurologic outcomes such as cerebral palsy.
There are anecdotal data for the use of interventions like tocolysis for the prolongation of transplacental transfer of protective maternal antibodies, for maternal infections acquired in the intrapartum period.[56] The average interval of ~6.3 +/- 1.4 days from the onset of maternal symptoms to delivery might have been enough time for the passive transfer of maternal antibodies to prevent MTCT and symptomatic disease in the newborn.[56] Tocolysis (as long as there are no obstetric contraindications), has been used also in other arboviral maternal infections, such as in dengue virus maternal infections to reduce the risk of vertical transmission [56] and in maternal varicella-zoster-virus infections during the peripartum period.[80] The safety and clinical effectiveness of tocolysis as a preventive measure in such intrapartum maternal infections requires additional systematic evaluation. Moreover, the number of pregnant women infected during the intrapartum period should be reported in maternal-fetal cohorts. In La Reunion,[27] the Sri Lanka[53] and the Colombia cohort[56] only 5% of maternal infections were acquired during the intrapartum period. The role of delivery via caesarean section (C/S) was analyzed only in La Reunion cohort [27] and appeared to have no influence on the MTCT risk. In La Reunion cohort[27] this observation may support the notion of transplacental transmission of CHIKV-infection from the mother to the fetus, rather than from exposure in the birth canal. The C/S rate among the 61 pregnant women in this cohort with peripartum/intrapartum infections was elevated compared to the baseline rate (43% vs 17%); the majority of those C/S was done due to fetal distress. [27] However, this was not seen in the Thailand cohort were the majority of the infants were born via vaginal delivery. [54]
For the interpretation of neonatal serologic test results, pediatricians and neonataologists should be aware that the absence of positive neonatal CHIKV IgG and IgM antibodies at birth in infants born to mothers with acute CHIKV-infections in the peripartum/intrapartum period does not exclude CHIKV neonatal infection. Infected newborns from such late maternal infections may have a delayed development of CHIKV IgG and IgM antibodies, within the first 3–4 weeks of life.[81] Serial serologic monitoring of these infants might be indicated as infected infants, particularly so symptomatic infants, might be at risk for poor long term neurodevelopmental outcomes.
Understanding the true impact of acute maternal CHIKV infections in the fetus and newborn requires systematic consideration also of fetal and neonatal mortality as well as ascertainment of long term neurodevelopmental outcomes in addition to the neonatal morbidity. Retrospectively extracted information about clinical signs and symptoms suggestive of acute maternal CHIKV infection during gestation likely underestimates the true incidence of maternal infections, due to recollection bias and non-capturing of mild or asymptomatic maternal infections. Moreover, standardized outcome collection and reporting across maternal-fetal cohorts is mandatory, to allow for prompt identification of the accurate risks to fetuses and newborns from maternal infections during gestation. Focus should be given during study design phase and outcome reporting for congenital/perinatal infections on all of the following: a) estimated time of maternal infection during gestation, with accurate reporting of the number of intrapartum maternal infections; b) consideration of APFDs in the overall combined fetal/neonatal disease impact from congenital CHIKV-infections; c) ascertainment of CHIKV-infections-status of APFDs; d) ascertainment of CHIKV infection status in all newborns exposed to suspected or confirmed maternal CHIKV-infections during gestation; and prompt documentation of losses to follow-up; e) serial screening of newborns exposed to late gestation maternal infections for the first month of life, even if they are seronegative at birth, given the likely delayed neonatal IgM and IgG production after late gestation maternal infections and f) ascertainment of long term neurodevelopmental outcomes for at least the symptomatic neonatal infections.
We observed significant variation in the reported rates of MTCT and symptomatic neonatal disease across cohorts. Referral selection bias and confounding by differences in the gestational age during maternal infections across cohorts might have explained the reported differences in the risks of symptomatic neonatal disease across cohorts, as symptomatic neonatal disease occurred almost exclusively from intrapartum maternal infections.[27, 45, 46] We were not able to make robust conclusions on the possible role of the implicated CHIKV strain in the observed variation in the MTCT-risks and risks of symptomatic disease across cohorts, given the limited number of cases. There is preliminary evidence that the different CHIKV-strains (Asian vs Central-East-South Africa [CESA] vs West Africa strain)[82, 83] might have different pathogenicity.[84] In outbreaks caused by the CESA CHIKV-strain[83], such as the La Reunion[50] and Mayotte[51] outbreaks the overall risk of symptomatic neonatal disease among all maternal infections was 6.26% (37/591) and 5.5% (9/163) respectively. In outbreaks where the Asian CHIKV-strains were implicated, the reported rates of symptomatic disease varied even more, with 0% MTCT rates from the Thailand cohorts[52, 54] and a small Colombian cohort[56]; versus 8% (4/50) for severely symptomatic neonatal disease from the Shri Lanka cohort,[53] 27.7% (53/191) from the El Salvador cohort[55] and 48% from the Santo Domingo cohort.[55] Nevertheless, there are recent data indicating that in South America CHIKV outbreaks, the African CESA CHIKV strains might also be implicated.[85, 86] Moreover, differences across cohorts were also noted in the reported neonatal case fatality rates, with 0% for La Reunion cohort [27] versus 5.1% for the Santo Domingo cohort.[55]
The number of neonatal CHIKV infections could be significantly underestimated using neonatal CHIKV IgG and IgM antibodies at birth. Ramful et al[49] showed that CHIKV infected infants (even symptomatic ones) from late-gestation maternal infections during the peripartum/intrapartum period can be seronegative at birth and might have delayed production of CHIKV antibodies; up to 3 weeks after birth for the development of IgMs and up to 4 weeks after birth for the development of positive IgGs. This is known to occur also in other congenital infections (e.g. congenital Toxoplasmosis[87]) when maternal infections occur very late in gestation and this might have underestimated the overall rate of mother-to-child transmission of CHIKV-infections in some of the analyzed reports.
Continued monitoring of the clinical implications of CHIKV infections during pregnancy is needed as CHIKV outbreaks can reemerge in regions where the virus had already previously circulating or emerge in new regions, where it had not been previously detected. Recently in 2017 a CHIKV outbreak was noted again in Italy, in the Anzio west-coast recreational region close to Rome [88–90], caused by an CESA strain. This strain was genetically slightly different from the strain implicated in the large 2007 outbreak in the Emilia-Romagna region in North-Eastern Italy.[23]
Furthermore, the potential benefit from tocolysis for intrapartum maternal infections is an intervention that needs systematic investigation, and if confirmed in larger scale studies to be effective, routine implementation in pregnant women with intrapartum maternal infection might have important public health implications. This might provide further support for the need for prenatal screening of pregnant women with suspected CHIKV infections during the peripartum/ intrapartum period. Future validation of the diagnostic performance of point-of-care tests for the serologic diagnosis of CHIKV maternal and/or neonatal infection and/or other arboviral infections is urgently needed. Moreover, preventive measures targeting avoidance of mosquito bites in pregnant women close to the expected time of delivery, might be cost-saving and effective strategies, given the high neonatal morbidity associated with intrapartum maternal infections. Furthermore, neonatologist, need to become aware that CHIKV-infected newborns from maternal infections late in gestation would need close clinical and laboratory monitoring of their hematologic parameters during their first week of life, even if they appear asymptomatic at birth, as symptomatic neonatal infections usually develop within 3–7 days after birth. Moreover, transplacentally transferred CHIKV-IgG antibodies on average disappear by 8 months of age in uninfected neonates.[49] However, the time to neonatal seroconversion is inversely related to the time of maternal infection during gestation; with >75% of non-infected neonates being still IgG positive by 12 months of age if maternal infection was in the first trimester vs 30% and <1% if maternal infection was in the second and third trimester respectively.[49] Moreover, it may take as long as 24 months for complete neonatal seroconversion to IgG negative status among uninfected neonates.[49]
Some study limitations should be acknowledged: First, in all analyzed cohorts (even those with serologic and/or molecular confirmation of CHIKV maternal infections), it was the presence of maternal symptoms the first indicator that led to subsequent testing of those women for CHIKV-infections. This may have led to overestimation of the MTCT risk and the risk of symptomatic neonatal disease from maternal CHIKV infections, if symptomatic maternal infections have an incrementally higher risk of MTCT, independently of the time of maternal infection. The effect of asymptomatic CHIKV maternal infections during pregnancy remains largely unknown. The majority of the CHIKV infections were originally thought to be symptomatic [91], nevertheless, recent reports indicate that the number of asymptomatic CHIKV infections might have been higher than it was originally thought. A report from the 2008 Thailand outbreak[92] showed that 50% of all cases were asymptomatic and a more recent report from that region showed that 87.5% of pregnant women infected during gestation were asymptomatic.[54] Nevertheless, conflicting results from the same region were also reported, with only 9% reported asymptomatic cases.[93] Additional surveillance studies from recent CHIKV outbreaks in Nicaragua also showed that 65% of cases were asymptomatic, with a symptomatic to asymptomatic ratio of 1:1.91.[94] Recollection bias might also explain some of the observed differences in the reported rates of symptomatic CHIKV-infections. Second, in the majority of the analyzed maternal/neonatal cohorts, only symptomatic neonatal infections were reported which might have underestimated the true MTCT-risk. Third, we could not identify published cohorts with an English abstract with pertinent data for our quantitative data synthesis from the majority of countries with recent CHIKV outbreaks; such as outbreaks in African[5] [12] and Asian countries[14, 95–97], including large outbreaks in India[98–104] after the 2005 reemergence of CHIKV in India, outbreaks in the Carribeans [32, 105, 106], Pacific Island[78] and Saudi Arabia.[107] Language-bias is also a possible reason for this phenomenon. Moreover, the published cohort studies with pertinent data from the outbreak in Central and South America since 2013 were very limited[55, 56] compared to the scale of CHIKV transmission across more than 45 countries throughout the Americas and with >1.7 million suspected CHIKV-infection cases reported to the Pan American Health Organization (PAHO).[78] For the majority of these outbreaks, only isolated case reports and small case series were identified, which we included in the qualitative data synthesis on the list of reported clinical manifestations from neonatal CHIKV- infections from maternal infections during gestation. Language-bias is also a possible reason for this phenomenon. It is possible that publications from several of these developing countries where such outbreaks occur are published only in grey literature [108] or in local non-English journals and indexed only in local journal databases, but not in PubMed. Moreover, the lack of financial resources and availability of accurate diagnostic tests[109] in several of the settings where such CHIKV outbreaks occur, contribute to the phenomenon of over-estimation of the true incidence and severity of the disease.
In conclusion, CHIKV is an emerging arbovirus with a global distribution that can cause significant morbidity and also death in infected fetuses and newborns after maternal infections during gestation. Neonatal morbidity likely occurs predominantly from intrapartum maternal infections. Improvement is needed in the reporting of clinical important outcomes for congenitally/perinatally acquired fetal and neonatal infections. Data should be collected and reported in a standardized way across maternal-fetal cohorts for all clinically important endpoints to allow for informative meta-analyses and individual patient-level meta-analyses in this field of congenital infections. With increasing climate instability and human migration, additional CKIKV outbreaks may be expected and non-immune pregnant women in developing as well as developed countries are at risk. Additional systematic studies of the impact of the CHIKV maternal infections during gestation to the fetuses and newborns are needed.
|
10.1371/journal.pmed.1002715 | Comparative efficacy and acceptability of psychosocial interventions for individuals with cocaine and amphetamine addiction: A systematic review and network meta-analysis | Clinical guidelines recommend psychosocial interventions for cocaine and/or amphetamine addiction as first-line treatment, but it is still unclear which intervention, if any, should be offered first. We aimed to estimate the comparative effectiveness of all available psychosocial interventions (alone or in combination) for the short- and long-term treatment of people with cocaine and/or amphetamine addiction.
We searched published and unpublished randomised controlled trials (RCTs) comparing any structured psychosocial intervention against an active control or treatment as usual (TAU) for the treatment of cocaine and/or amphetamine addiction in adults. Primary outcome measures were efficacy (proportion of patients in abstinence, assessed by urinalysis) and acceptability (proportion of patients who dropped out due to any cause) at the end of treatment, but we also measured the acute (12 weeks) and long-term (longest duration of study follow-up) effects of the interventions and the longest duration of abstinence. Odds ratios (ORs) and standardised mean differences were estimated using pairwise and network meta-analysis with random effects. The risk of bias of the included studies was assessed with the Cochrane tool, and the strength of evidence with the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach. We followed the PRISMA for Network Meta-Analyses (PRISMA-NMA) guidelines, and the protocol was registered in PROSPERO (CRD 42017042900). We included 50 RCTs evaluating 12 psychosocial interventions or TAU in 6,942 participants. The strength of evidence ranged from high to very low. Compared to TAU, contingency management (CM) plus community reinforcement approach was the only intervention that increased the number of abstinent patients at the end of treatment (OR 2.84, 95% CI 1.24–6.51, P = 0.013), and also at 12 weeks (OR 7.60, 95% CI 2.03–28.37, P = 0.002) and at longest follow-up (OR 3.08, 95% CI 1.33–7.17, P = 0.008). At the end of treatment, CM plus community reinforcement approach had the highest number of statistically significant results in head-to-head comparisons, being more efficacious than cognitive behavioural therapy (CBT) (OR 2.44, 95% CI 1.02–5.88, P = 0.045), non-contingent rewards (OR 3.31, 95% CI 1.32–8.28, P = 0.010), and 12-step programme plus non-contingent rewards (OR 4.07, 95% CI 1.13–14.69, P = 0.031). CM plus community reinforcement approach was also associated with fewer dropouts than TAU, both at 12 weeks and the end of treatment (OR 3.92, P < 0.001, and 3.63, P < 0.001, respectively). At the longest follow-up, community reinforcement approach was more effective than non-contingent rewards, supportive-expressive psychodynamic therapy, TAU, and 12-step programme (OR ranging between 2.71, P = 0.026, and 4.58, P = 0.001), but the combination of community reinforcement approach with CM was superior also to CBT alone, CM alone, CM plus CBT, and 12-step programme plus non-contingent rewards (ORs between 2.50, P = 0.039, and 5.22, P < 0.001). The main limitations of our study were the quality of included studies and the lack of blinding, which may have increased the risk of performance bias. However, our analyses were based on objective outcomes, which are less likely to be biased.
To our knowledge, this network meta-analysis is the most comprehensive synthesis of data for psychosocial interventions in individuals with cocaine and/or amphetamine addiction. Our findings provide the best evidence base currently available to guide decision-making about psychosocial interventions for individuals with cocaine and/or amphetamine addiction and should inform patients, clinicians, and policy-makers.
| Cocaine and amphetamines are the most commonly abused stimulants in people aged 15–64 years, and they are linked to significant physical and mental illness as well as a substantial burden for society.
Currently, clinical guidelines recommend the use of psychosocial interventions as first-line treatment for people with cocaine and/or amphetamine addiction.
However, it is still unclear which psychosocial intervention, if any, is the most effective treatment for such patients.
We used network meta-analysis to analyse 50 clinical studies (6,943 participants) on 12 different psychosocial interventions for cocaine and/or amphetamine addiction.
We found that the combination of 2 different psychosocial interventions, namely contingency management and community reinforcement approach, was the most efficacious and most acceptable treatment both in the short and long term.
To our knowledge, this is the best evidence base to guide decision-making about psychosocial interventions for cocaine and/or amphetamine addiction. Clinical guidelines should be updated to reflect these results, and policy-makers are encouraged to invest accordingly.
Future trials should use contingency management plus community reinforcement approach as the reference treatment.
| Drug use disorders are the 15th leading cause of disability-adjusted life years in high-income countries [1]. Cocaine and amphetamines are the most commonly abused stimulants in people aged 15–64 years, with an annual prevalence of misuse of 0.38% and 1.20%, respectively [2]. Patients addicted to stimulants experience a range of psychological and physical sequelae including psychosis and other mental illnesses, neurological disorders and cognitive deficits, cardiovascular dysfunctions, sexually transmitted diseases, and blood-borne viral infections such as HIV and hepatitis B and C [3], and are at increased risk of all-cause mortality [4]. Moreover, the social burden of stimulant abuse is worsened by its association with crime, violence, and sexual abuse [5].
Currently, international clinical guidelines recommend the use of psychosocial interventions for cocaine and/or amphetamine addiction as first-line treatment, and there is little evidence supporting pharmacotherapy or brain stimulation treatments [6–8]. In the absence of approved pharmacotherapies, several structured psychosocial and self-help approaches are available, such as contingency management (CM) (a behavioural approach that consists in providing stimulant users with rewards upon drug-free urine samples), community reinforcement approach (a multi-layered intervention involving functional analysis, coping-skills training, and social, familial, recreational, and vocational reinforcements), and 12-step programme (a set of guiding principles outlining a course of action for self-help recovery from addiction) [8]. International guidelines are unclear on whether any specific intervention should be considered first [9,10]; for example, the National Institute for Health and Care Excellence (NICE) recommends CM alone, cognitive behavioural therapy (CBT) alone, or self-help groups based on 12-step programme alone for the treatment of individuals with stimulant use disorders [8]. However, a recent systematic review showed that CM and CBT were well accepted and moderately efficacious at the end of treatment, but not at follow-up after treatment completion [11].
Previous pairwise meta-analyses relied on a limited number of studies with direct comparisons between different interventions [12,13]. From a clinical perspective, it is important to assess whether psychosocial interventions are effective and acceptable in both the short and long term, and also whether the combination of 2 approaches can produce a significant benefit. We therefore performed a network meta-analysis to compare and rank the efficacy and acceptability of individual and combined psychosocial interventions for the treatment of cocaine and/or amphetamine addiction at different time-points.
This network meta-analysis was conducted following a pre-established protocol registered on PROSPERO (CRD 42017042900), and is reported according to the PRISMA for Network Meta-Analyses (PRISMA-NMA) guidelines [14]. We searched the Cochrane Drugs and Alcohol Group Specialised Register, PubMed, Embase, CINAHL, ISI Web of Science, and PsycINFO from the date of database inception to 8 April 2018. We also screened international registers, hand-searched the reference lists of retrieved articles, and looked at key conference proceedings (for the full search strategy, see S1 Text). When needed, we contacted the investigators and relevant trial authors to obtain information about unpublished or incomplete trials. All searches included non-English language literature.
We included randomised controlled trials (RCTs) comparing any structured psychosocial intervention against a control intervention—another psychosocial intervention or treatment as usual (TAU)—for the treatment of individuals with cocaine and/or amphetamine addiction, according to the Diagnostic and Statistical Manual of Mental Disorders (DSM) III, III-R, IV, IV-TR, or V or the International Classification of Diseases–10th revision (ICD-10) criteria. CBT, CM, community reinforcement approach, meditation-based therapies, non-contingent rewards, supportive-expressive psychodynamic therapy, 12-step programme, and their combinations were all identified as structured psychosocial interventions. We excluded studies on occasional users not actively seeking treatment and RCTs with study duration less than 4 weeks. We did not exclude studies on individuals with a comorbid substance use disorder (including opioid, alcohol, or cannabis use) or with a comorbid psychiatric disorder.
Four authors (FDC, GLD, MC, RDG) independently screened the references retrieved by the search, selected the studies, and extracted the data, using a predefined data-extraction sheet. The same reviewers discussed any uncertainty regarding study eligibility and data extraction until consensus was reached; conflicts of opinion were resolved with another member of the review team (AC). 2 authors (GLD, MC) independently assessed the risk of bias of the included studies with the Cochrane tool [15], Three authors (FDC, CDG, AC) used the Grading of Recommendations Assessment, Development and Evaluation (GRADE) approach [16], through the Confidence in Network Meta-Analysis Software (CINeMA) [17], to evaluate the strength of evidence for results at the end of treatment from the network meta-analysis (S2 Text).
We considered as primary outcomes the efficacy and the acceptability of the interventions at the end of treatment [18]. Efficacy was measured as the proportion of individuals abstinent (assessed by urinalysis), and acceptability as the proportion of individuals who dropped out from the study due to any cause. As secondary outcomes, we also measured efficacy and acceptability at 12 weeks from the start of treatment and at the longest follow-up (with follow-up starting at the end of treatment, independent of the duration of the intervention). If 12-week data were not available, we used data ranging between 4 and 20 weeks (giving preference to the time-point closest to 12 weeks). Other secondary outcomes were the longest duration of abstinence measured both at 12 weeks and at the end of treatment.
We performed first pairwise meta-analyses using a random-effects model to estimate pooled odds ratios (ORs) for dichotomous outcomes and standardised mean differences (SMDs) for continuous outcomes with their 95% confidence intervals (CIs) using STATA [19]. We assessed statistical heterogeneity in each pairwise comparison with τ2, I2 statistic, and P value [15]. If at least 10 studies were available, we used the funnel plot and Egger’s test to detect publication bias [15].
We incorporated indirect comparisons with direct comparisons using random-effects network meta-analyses within a frequentist framework using STATA (network package), and results are presented with the network graphs package [20]. We report the results of network meta-analyses for both primary and secondary outcomes in league tables with effect sizes (OR or SMD) and their 95% CIs. When dichotomous outcome data were missing, we assumed that patients who dropped out after randomisation had a negative outcome. Missing continuous outcome data were analysed on an endpoint basis, including only participants with a final assessment, as reported by the original study authors. We also calculated the number needed to treat (NNT), which is the number of patients that need to be treated in order for 1 to benefit from the intervention compared with TAU.
We assessed incoherence between direct and indirect sources of evidence using local and global approaches. Coherence (or consistency) is an important assumption to check in network meta-analyses because it is the manifestation of transitivity in the data from a network of interventions: coherence exists when treatment effects from direct and indirect evidence are in agreement (subject to the usual variation due to heterogeneity in the direct evidence) [21]. Local incoherence was measured by using a loop-specific approach (which identified inconsistent loops of evidence) [22] and a side-splitting approach (which separated evidence on a particular comparison into direct and indirect evidence) [23]. Global incoherence was measured with the between-studies standard deviation (SD) (heterogeneity parameter) by using both a coherence and incoherence model and by measuring the chi-squared incoherence, with its P value. We estimated the presence of publication bias by plotting comparison-adjusted funnel plots for the network meta-analyses with a linear regression line [24]. We also estimated the ranking probabilities for all treatments, i.e., their probability of being at each possible rank for each intervention. We report the treatment hierarchy as the surface under the cumulative ranking curve (SUCRA) and as the mean rank (S2 Text) [24].
To determine whether the results were affected by study characteristics, we performed subgroup network meta-analyses for abstinence and dropout at the end of treatment according to the following variables: year of publication, sex ratio, mean age group, intensity of the treatment, type of stimulant, risk of bias, opioid therapy, sample size, and comorbid alcohol misuse. Additionally, we performed sensitivity network meta-analyses for the primary outcomes by considering (a) only trials on individuals addicted to cocaine and no other stimulant and (b) only trials on individuals addicted to stimulants and on opioid substitution therapy.
From the initially identified 7,261 citations, we retrieved 160 potentially eligible articles in full text (Fig 1). We excluded 88 reports, but then included 4 additional studies (3 from trial registers and 1 from screening the references), resulting in 76 publications (S3 Text) describing 50 RCTs (6,942 participants), published between 1993 and 2016 (Fig 2; Table 1), comparing 12 psychosocial interventions or TAU (listed and defined in S4 Text). Overall, 5,158 participants were randomly assigned to psychosocial treatments, and 1,784 to TAU. Full clinical and demographic characteristics are reported in Table 1. The mean study sample size was 139 participants, ranging between 19 and 487 participants. The median duration of treatment was 12 weeks (range 6–36). Dropout rates varied between 15.1% (CM + CBT) and 60.2% (meditation-based therapies) (S1 Table). A total of 37 studies were followed up after study completion, for a mean duration of 41.4 weeks (range 16–96). A total of 42 (84%) trials recruited patients from North America, 6 from Europe, 1 from Latin America, and 1 from Oceania. About a third of the population was women (35.9%), and the mean age was 36.8 years. A total of 76% of trials (38 of 50) enrolled participants with cocaine addiction, 8% of trials (4 of 50) with amphetamine addiction, and 16% (8 of 50) with both. About one-third of trials (18 of 50) enrolled participants on methadone maintenance. The mean addiction severity was moderate/high (S2 Table). In terms of risk of bias, 22 (44%) trials were rated low risk, 13 (26%) moderate, and 15 (30%) high (S3 Table; S1 Fig).
We present all the networks for specific outcomes in S2 Fig. Eight psychosocial interventions had at least 1 trial versus TAU, and all of them were directly compared with at least another psychosocial intervention. We obtained unpublished or supplementary information for 5 of the included studies [45,49,59–61].
The pairwise meta-analyses are presented in S4 Table, while data on heterogeneity are presented in S5 Table. The pairwise meta-analyses showed some statistically significant results in terms of abstinence and dropout. CM plus CBT, CM, and 12-step programme were superior to TAU in terms of abstinence at the end of treatment, while CBT, CM, and the combination of CM plus community reinforcement approach were superior to TAU in terms of dropout at the end of treatment (S4 Table).
The results of the network meta-analysis are presented in Fig 3. In terms of abstinence at the end of treatment, the combination of CM plus community reinforcement approach, the combination of CM plus CBT, and CM alone were superior to non-contingent rewards (OR ranging between 2.59 [95% CI 1.70–3.93], P < 0.001, and 3.31 [95% CI 1.32–8.28], P = 0.010) and to TAU (OR ranging between 2.22 [95% CI 1.59–3.10], P < 0.001, and 2.84 [95% CI 1.24–6.51], P = 0.013). Moreover, the combination of CM plus community reinforcement approach was also superior to the combination 12-step programme plus non-contingent rewards and to CBT (OR 4.07 [95% CI 1.13–14.69], P = 0.031, and 2.43 [95% CI 1.02–5.88], P = 0.045, respectively), while CM alone and the combination of CM plus CBT were superior to CBT (OR 1.88 [95% CI 1.52–2.85], P = 0.003, and 2.08 [95% CI 1.28–3.33], P = 0.002, respectively). In terms of dropouts at the end of treatment, the combination of CM plus community reinforcement approach, community reinforcement approach alone, non-contingent rewards, CM alone, and CBT were better accepted than TAU (OR ranging between 1.41 [95% CI 1.10–1.82], P = 0.007, and 3.63 [95% CI 2.01–6.55], P < 0.001). Moreover, the combination of CM plus community reinforcement approach was better accepted than CBT, CM alone, CM plus CBT, community reinforcement approach plus non-contingent rewards, meditation-based therapies, non-contingent rewards, supportive-expressive psychodynamic therapy, 12-step programme alone, and 12-step programme plus non-contingent rewards (OR ranging between 2.06 [95% CI 1.04–4.08], P = 0.037, and 4.61 [95% CI 1.92–11.06], P < 0.001).
Fewer studies reported results for abstinence measured at 12 weeks of treatment (S3 Fig) and at the longest follow-up after treatment completion (S4 Fig), but findings were in line with the outcome data at the end of treatment. Comparative abstinence and dropout at different time-points for each psychosocial intervention versus TAU are presented in Fig 4.
In terms of the longest duration of abstinence measured at 12 weeks (S5 Fig), we found that the combination of CM plus CBT and CM alone were superior to TAU (SMD 0.75 [95% CI 0.31–1.19] and 0.62 [95% CI 0.43–0.80], respectively). Likewise, for the longest duration of abstinence measured at the end of treatment (S6 Fig), we found that the combination of CM plus CBT and CM alone were superior to TAU (SMD 0.74 [95% CI 0.43–1.06] and 0.60 [95% CI 0.43–0.76], respectively).
The common heterogeneity SD for the coherence model was 0.46 and 0.21 for abstinence and dropout at the end of treatment, respectively (it was 0.47 and 0.19 for abstinence and dropout at 12 weeks, respectively). The global incoherence was not significant for all the outcomes considered (S6 Table). Tests of local incoherence did not show any inconsistent loops for abstinence and dropout at the end of treatment, although in some cases the ratio of the odds ratios (RoR) from direct and indirect evidence was large (i.e., RoR > 2), and we cannot definitely exclude the presence of incoherence [22]. We found only 1 inconsistent loop for abstinence measured at 12 weeks and no other inconsistent loops for the other outcomes considered at 12 weeks (S7 Table; S7 Fig). The test of incoherence from the side-splitting model did not show significant differences for abstinence at the end of treatment but found some differences between some comparisons for dropout at the end of treatment (S8 Table). The comparison-adjusted funnel plots of the network meta-analysis for abstinence and dropout at the end of treatment were not suggestive for significant publication bias (S8 Fig).
The ranking of psychosocial interventions based on cumulative probability plots and SUCRAs is presented in S9 Table and S9 Fig. We also performed subgroup analyses for abstinence and dropout at the end of treatment to study the effect of several potential moderator variables, the findings of which did not substantially differ from those of the primary analysis for most of the comparisons (S10 Table). Pre-planned sensitivity analysis on individuals addicted to cocaine only did not affect the main results (S10 Fig), while pre-planned sensitivity analysis on individuals on opioid substitution therapy showed a superiority of CM alone and CM plus CBT over TAU and non-contingent rewards, and a superiority of CM alone over CBT (S11 Fig). Predictivity intervals of mixed estimates are presented in S11 Table, while the overall limitations per comparison are presented in S12 and S13 Figs.
We found that 4 patients needed to be treated with CM plus community reinforcement approach to have 1 additional patient abstinent at the end of treatment compared toTAU (NNT 4.07, 95% CI 2.29–21.95), with consistent results at longest follow-up after treatment completion (NNT 3.68, 95% CI 2.36–14.24). Similarly, 3 patients needed to be treated with CM plus community reinforcement approach to have 1 fewer patient dropping out at the end of treatment compared to TAU (NNT 3.25, 95% CI 2.42–5.79). For abstinence at the end of treatment, the NNT for the combination of CM plus CBT and for CM alone versus TAU was 4.80 (95% CI 2.99–12.12) and 5.44 (95% CI 3.74–9.75), respectively. For dropout at the end of treatment, the NNT ranged from 4.02 (95% CI 2.58–12.62) for CRA to 7.15 (95% CI 4.15–27.66) for non-contingent rewards, 10.52 (95% CI 5.83–53.65) for CBT, and 11.82 (95% CI 6.74–43.26) for CM.
The overall strength of evidence according to GRADE is summarised in S12 Table for abstinence and for dropout at the end of treatment. For CM plus community reinforcement approach versus TAU, the strength of evidence was rated as “moderate” for abstinence and as “high” for dropout due to any cause. For CM plus CBT and CM alone versus TAU, the strength of evidence for abstinence was rated as “moderate” for both, while the strength of evidence for dropout due to any cause was rated as “very low” and “moderate”, respectively.
This network meta-analysis is based on 50 studies including 6,942 individuals randomly assigned to 12 different psychosocial interventions or TAU. To our knowledge, it is the most comprehensive synthesis of data for all available psychosocial interventions in individuals with cocaine and/or amphetamine addiction.
We found that CM alone or in combination with either community reinforcement approach or CBT had superior efficacy and acceptability compared to TAU at 12 weeks and at the end of treatment. This effect was not significantly influenced by clinical modifiers in the subgroup analyses and remained significant in the sensitivity analyses. Moreover, CM in combination with community reinforcement approach and community reinforcement approach alone were more effective than TAU at the longest follow-up after treatment completion. The clinical relevance of this finding is key, because achieving long-term abstinence is the main treatment goal for individuals with cocaine and/or amphetamine addiction [18].
Several major guidelines recommend the use of either CBT or CM alone for the treatment of cocaine and/or amphetamine addiction [6,8,10]. Self-help groups following the 12-step programme are also recommended [8]. Our results do not support these recommendations. We found that CBT alone was more acceptable than TAU (NNT 10.5, 95% CI 5.8–53.6), but it was not superior for abstinence on any dichotomous or continuous outcome measured and was less effective than CM alone. CM alone showed greater efficacy (NNT 5.2, 95% CI 3.6–9.3) and acceptability (NNT 12.5, 95% CI 7.0–48.8) than TAU at 12 weeks of treatment and at the end of treatment (NNT 5.4, 95% CI 3.8–9.8, and 11.9, 95% CI 6.8–43.3, respectively), but the effect was not sustained at the longest follow-up after treatment completion. Both CBT and CM were inferior to CM in combination with community reinforcement approach at the longest follow-up after treatment completion. Community reinforcement approach alone was not different from TAU for abstinence at 12 weeks of treatment or at the end of treatment, but showed increased abstinence at the longest follow-up after treatment completion (NNT 4.1, 95% CI 2.4–36.2). CM in combination with community reinforcement approach was superior to TAU for abstinence at 12 weeks of treatment (NNT 2.1, 95% CI 1.6–6.2), at the end of treatment (NNT 4.1, 95% CI 2.3–21.9), and at the longest follow-up after treatment completion (NNT 3.7, 95% CI 2.4–14.2), as well as for acceptability at 12 weeks of treatment (NNT 3.1, 95% CI 2.2–6.1) and at the end of treatment (NNT 3.3, 95% CI 2.3–6.3), as shown in Fig 4. CM plus community reinforcement approach was also superior to 12-step programme for abstinence and dropout at 12 weeks of treatment, for dropout at the end of treatment, and for abstinence at the longest follow-up after study completion.
Behavioural interventions have proved effective for the treatment of other forms of addiction [75–77]. There is growing evidence that reducing punishment—such as incarceration—and adopting positive reinforcement for people with substance use improves their access to services, their reintegration into society, and, ultimately, public safety [78–80]. Recent experimental data emphasise the potential of interventions that focus on improving goal-directed behaviour and positive reinforcement rather than punishment in people with cocaine addiction [81]. The efficacy of a purely behavioural intervention—such as CM alone—shows that financial rewards can compete with biological rewards mediated by cocaine and amphetamine cues and incentives [82]. This seems to be true only if rewards are contingent upon the provision of drug-free urine samples, as non-contingent rewards were not shown to be effective (Fig 4). Indeed, CM strategies help individuals to overcome apathy or resistance to the recovery process. In this study we found that, although CM was efficacious at the end of treatment, the effect of CM alone was not sustained at longest follow-up after treatment completion (Fig 4). Cocaine and amphetamine addiction is conceptualised as a chronic and recurrent brain disease, which entails behavioural and psychological abnormalities (primarily reward-processing deficits) following the learned or conditioned pairing of situational and social cues with the reinforcing effects of drug use [83,84]. It seems unlikely that a behavioural strategy alone could address in the long term the whole complexity of biological, psychological, and behavioural factors that underlie addiction. The addition of community reinforcement approach to CM potentiates an otherwise purely behavioural intervention with psychological and social components that may enhance its effect. Notably, community reinforcement approach alone performs no differently from TAU in the short term, but its effect is more sustained at longest follow-up. The combined intervention, CM plus community reinforcement approach, overall achieves the best outcomes.
Cocaine and amphetamine addiction is highly prevalent in the world and is incredibly costly economically. In 2015, illicit drugs cost tens of millions of disability-adjusted life years, with Europeans proportionately experiencing more, but with the greatest mortality rate in low- and middle-income countries [85]. We did not do a formal cost-effectiveness analysis. Indeed, recent cost-effectiveness analyses on psychosocial interventions for substance use are encouraging [8,86], but without a full economic model our recommendation cannot be made unequivocally.
This study has some limitations. Some comparisons were appraised as having low or very low quality, potentially restricting the validity of those results. All RCTs of psychosocial interventions for cocaine and/or amphetamine addiction are not blinded, which increases the risk of performance bias for self-reported outcomes. For this reason, we only reported data based on objective outcomes (abstinence on urinalysis and data on attrition), which are less likely to be influenced by the lack of blinding. The risk of selective study reporting was minimised as we contacted study authors to retrieve unpublished data, but we cannot exclude that some unpublished studies remain missing or that published reports overestimated the efficacy of treatments. Finally, some interventions are designed to last more than 12 weeks, namely, CM in combination with community reinforcement approach (24 weeks) [40], community reinforcement approach alone (time unlimited, but in the studies included the intervention lasted always 24 weeks), and supportive-expressive psychodynamic therapy (which could be either time-limited or unlimited and lasted 36 weeks in the only study included) [87]. For these interventions, the evaluation at 12 weeks was extracted before the end of treatment, which was a disadvantage over other interventions requiring shorter duration.
The results of this network meta-analysis support the use of combined CM plus community reinforcement approach as the most effective and acceptable intervention for both short- and long-term treatment of individuals with cocaine and/or amphetamine addiction. The provision of evidence-based psychosocial treatments for stimulant use disorders is all the more important because of the lack of validated pharmacological or brain-stimulation-based treatment for cocaine and/or amphetamine addiction. These findings may influence clinical guidelines; however, further studies are warranted to confirm these results and evaluate cost-effectiveness.
|
10.1371/journal.ppat.1004855 | Mechanisms of Stage-Transcending Protection Following Immunization of Mice with Late Liver Stage-Arresting Genetically Attenuated Malaria Parasites | Malaria, caused by Plasmodium parasite infection, continues to be one of the leading causes of worldwide morbidity and mortality. Development of an effective vaccine has been encumbered by the complex life cycle of the parasite that has distinct pre-erythrocytic and erythrocytic stages of infection in the mammalian host. Historically, malaria vaccine development efforts have targeted each stage in isolation. An ideal vaccine, however, would target multiple life cycle stages with multiple arms of the immune system and be capable of eliminating initial infection in the liver, the subsequent blood stage infection, and would prevent further parasite transmission. We have previously shown that immunization of mice with Plasmodium yoelii genetically attenuated parasites (GAP) that arrest late in liver stage development elicits stage-transcending protection against both a sporozoite challenge and a direct blood stage challenge. Here, we show that this immunization strategy engenders both T- and B-cell responses that are essential for stage-transcending protection, but the relative importance of each is determined by the host genetic background. Furthermore, potent anti-blood stage antibodies elicited after GAP immunization rely heavily on FC-mediated functions including complement fixation and FC receptor binding. These protective antibodies recognize the merozoite surface but do not appear to recognize the immunodominant merozoite surface protein-1. The antigen(s) targeted by stage-transcending immunity are present in both the late liver stages and blood stage parasites. The data clearly show that GAP-engendered protective immune responses can target shared antigens of pre-erythrocytic and erythrocytic parasite life cycle stages. As such, this model constitutes a powerful tool to identify novel, protective and stage-transcending T and B cell targets for incorporation into a multi-stage subunit vaccine.
| Malaria is arguably one of the deadliest infectious diseases in human history. Today, it infects nearly 300 million people each year and kills up to 1 million of those—mostly women and children under the age of 5—and no effective malaria vaccine has been developed. Traditional subunit vaccines for pathogens work by training the immune system to recognize a single pathogen target. Attempts at developing a subunit malaria vaccine have, however, been stymied by the complexity of the parasite genome which encodes a complex life cycle with specific stages in the mosquito, as well as in the liver and blood of the mammalian host. Only the blood stage parasites cause malaria symptoms and mortality. Previously, it was assumed that immunity to malaria is stage-specific, either targeting parasites in the liver or in blood, but not both. The herein described vaccination approach uses genetically engineered, attenuated rodent malaria parasites that are able to infect the mouse liver and replicate, but die shortly before red blood-infectious parasite stages are formed and released. Immunization with these attenuated parasites induces the immune system to build defenses against both parasite stages in the liver and blood. Protection is mediated by multiple arms of the immune system. The antibody arm recognizes parasite targets shared between liver stages and blood stages. This not only demonstrates the optimal potency of this live-attenuated vaccination strategy, but also provides a potential source of new malaria subunit vaccine targets.
| Unlike other infectious diseases, malaria parasites continue to defy the development of a protective vaccine. One main difference between pathogens currently amenable to vaccination and malaria parasites is the degree of complexity of the parasites causing malaria, Plasmodium spp. These eukaryotic parasites have complex genomes that control elaborate life cycles. They progress through multiple, antigenically distinct stages of replication and infection within mammalian hosts and mosquito vectors—making it difficult to target with traditional vaccination methods[1]. Infection is initiated when a parasitized Anopheles mosquito injects tens to hundreds of sporozoites into the dermis of the host. Sporozoites traverse through multiple host cell types in the dermis for minutes to hours until they traverse the vascular endothelium and into the circulation. The sporozoites are then carried into the sinusoids of the liver where they again traverse multiple cell types to reach and infect hepatocytes. This begins the clinically silent liver stage development of infection, during which each parasite undergoes many rounds of replication in a single hepatocyte and eventually forms tens of thousands of red blood cell-infectious exoerythrocytic merozoites. They are released in to the circulation and begin the asexual blood stage (BS) cycle whereby cyclic infection, replication within and lytic release from red blood cells (RBCs) occurs. This rapidly propagates the parasite and causes all malaria-associated morbidity and mortality as parasite numbers expand into the billions. A fraction of parasites terminally develop into gametocytes, which can be transmitted back to a mosquito during blood meal acquisition. To date, malaria vaccination strategies have largely focused on either the sporozoite and liver stages (“pre-erythrocytic”, PE) or BS of infection by targeting parasite antigens specific to each stage[2]. However, success has been limited with these stage-specific approaches, raising the question as to whether there should be a greater emphasis on multi-stage vaccination approaches.
PE vaccines have the advantage of targeting a bottleneck in the parasite population with only tens to a few hundred sporozoites injected in the skin and even fewer successfully infecting the liver. In addition, PE infection is clinically silent and completely eliminating PE parasites (termed “sterile protection”) would prevent BS infection and thus both disease and transmission. Both humoral and cellular immune defenses can contribute to PE immunity. Antibodies against sporozoites can act in the skin to immobilize the parasite and can bind to sporozoites in circulation to prevent hepatocyte infection[3–7]. Once parasites are within hepatocytes, CD8 T cells can target the infected hepatocyte and kill it[8]. However, successful infection of the liver by even a single parasite can lead to fulminant BS infection. Indeed, the stringent requirement for both antibodies and T cells to eliminate 100% of PE parasites has contributed to the limited success of PE subunit vaccine candidates in clinical trials. The first malaria vaccine candidate to reach phase III clinical trials is RTS,S, which targets only the circumsporozoite protein (CSP). RTS,s is capable of significantly reducing the cases of severe disease, but the long-term efficacy of RTS,S in eliciting protection is limited[9,10]. An alternative strategy for PE vaccination is immunization with live-attenuated sporozoites. Radiation-attenuated sporozoites (RAS) and sporozoites administered under chloroquine cover (known as “infection treatment immunization”, ITI) can confer 100% sterilizing PE protection in humans[11–13] Similarly, immunization with genetically attenuated parasites (GAP) has been shown to elicit complete sterile protection against PE infection in mice[14–16]. CD8 T cells are required for live-attenuated sporozoite protection in mice and also correlate with protection in non-human primates[17–22]. However, immunization with RAS requires the development of very large numbers of sporozoite-specific CD8 T cells for complete protection which in animal models needs to account for ~1% of the total CD8 repertoire[23]. While antibodies elicited by attenuated whole parasites are also able to strongly reduce liver infection, they have not been shown to be essential for protection[4,5,17,19,21,24,25].
The parasite BS have been the other major focus of malaria vaccine development efforts. However, attempts to create BS subunit vaccines have been stymied by suboptimal clinical performance perhaps due to the large degree of antigenic variation and polymorphisms within BS proteins and the high parasite burden as compared to the PE stages[26],[27].
In contrast to the stage-specific approaches to malaria vaccine development, targeting PE and BS life cycle stages simultaneously may be more fruitful as PE immunity can reduce the number of developing liver stages, which in turn reduces the number of merozoites released from the liver—thus potentially making BS infection easier to control and eliminate by an immune response. However, there is scant evidence that such stage-transcending protection (STP) is possible and the antigens and immunological mechanisms potentially capable of mediating STP remain undefined. Thus, establishing an STP model and understanding the mechanisms required for potent immunity against multiple parasite stages could be critical in developing fully protective, multi-stage subunit malaria vaccines.
Our previous work has indicated that late liver stage-arresting GAP confer STP[28]. Herein, we build upon this evidence to show that STP can be mediated by both T cells and by antibodies. Furthermore, protective antibodies predominantly rely on FC-mediated effector mechanisms and recognize potentially novel protective antigens shared between the late liver stages and BS parasites. Not only do our findings provide a rationale for the development of a late liver stage-arresting GAP as a vaccine candidate, but they also offer a platform to identify novel antigens and investigate the immune mechanisms mediating robust STP against PE stages and BS of Plasmodium.
P. yoelii (Py) fabb/f—parasites that are deficient in endogenous fatty acid biosynthesis undergo substantial liver stage growth, develop into late stage exoerythrocytic schizonts but fail to complete differentiation into exoerythrocytic merozoites[29] As a consequence, mice immunized with Pyfabb/f—parasites only experience PE infection and are not exposed to BS parasites. As such, they constitute late liver stage-arresting GAP (LAGAP). Mice immunized with LAGAP are not only protected against sporozoite challenge but are also protected against direct intravenous challenge with Py-infected RBCs (iRBCs)[28]. However, the immune mechanisms elicited by PE LAGAP-immunization that control and eliminate BS infection remain to be elucidated.
To assess the relative importance of antibodies and T cells in protection, BALB/cJ and C57BL/6 mice were immunized with LAGAP sporozoites, isolated from mosquito salivary glands, and given an intravenous (iv) challenge of 10,000 Py lethal strain iRBCs 25 days after the final immunization. While mock-immunized mice (mice injected with uninfected mosquito salivary gland debris) succumbed to hyperparasitemia within a week after challenge, LAGAP-immunized mice of both strains controlled parasitemia and cleared infection (Fig 1A). Interestingly, C57BL/6 immunized mice controlled the BS infection more robustly than BALB/cJ mice, displaying a lower peak parasitemia (∼2% vs. ∼13% at day 8 after challenge Fig 1A). To examine the respective roles of antibodies (Ab) and T cells in protection, we depleted T cells using monoclonal Ab (mAb) specific for CD4 and CD8 24 hours prior to lethal iRBC challenge (S1A and S1B Fig). Strikingly, only LAGAP-immunized C57BL/6 mice were able to control BS infection in the absence of T cells whereas BALB/cJ mice succumbed to hyperparasitemia similar to mock-immunized control mice (Fig 1C). Thus, the immune mechanisms of STP depend on the mouse genetic background. Whereas humoral immunity is sufficient to protect C57BL/6 mice from a lethal BS challenge, T cells are required to protect LAGAP-immunized BALB/cJ mice. Given the stage-transcending immunity observed, we wanted to ensure that there was no self-limiting blood stage infection caused by breakthrough during LAGAP PE stage immunization, which could be inducing blood stage immunity. To do this, 250μL of pooled blood from C57BL/6 mice immunized with LAGAP sporozoites three days prior was injected into naïve C57BL/6 mice. None of the 5 recipient mice became blood stage patent. This contrasts with transfer of just 2 LAGAP blood stage parasites, which can result in patency of 100% of mice by day 7[29]. Thus, LAGAP PE immunization did not induce a submicroscopic blood stage infection that could be causing the observed blood stage immunity.
To further examine if antibodies elicited by LAGAP immunization are required for protection against BS infection, we immunized C57BL/6 AID-/- mice, which possess B cells that are incapable of producing class-switched antibodies [30,31]. These mice developed a robust CD4+ and CD8+ T cell response to immunization as measured by markers of antigen-experienced cells in the peripheral blood[32] (S2A and S2B Fig) but failed to control a lethal BS challenge (Fig 1C). Furthermore, passive transfer of wildtype C57BL/6 immune sera to BALB/cJ mice conferred protection against a lethal BS challenge in all mice (Fig 1D). This confirms the ability of antibodies raised in C57BL/6 LAGAP-immunized mice to control BS infection and demonstrates that the difference in protection between the two strains is not explained by a higher susceptibility to infection in BALB/cJ mice. Thus, these data indicate that the antibodies elicited by LAGAP immunization of C57BL/6 mice are potent and essential for STP against BS infection.
To further investigate the host strain-specific differences in STP, we characterized the titers of IgG antibodies against sporozoites and BSs in sera obtained from LAGAP-immunized mice of both strains. Both C57BL/6 and BALB/cJ had similar IgG responses to CSP after immunization (Fig 2A), indicating that humoral immune responses to the major sporozoite surface protein are similar in both strains. Although both strains showed an increase in total IgG against BS parasites following LAGAP immunization, C57BL/6 mice had significantly higher titers than BALB/cJ mice (Fig 2B).
Taken together, these data show that both cell-mediated and humoral immunity protects against BS infection following LAGAP immunization. Furthermore, antibodies from LAGAP immunized C57BL/6 but not BALB/cJ mice are both sufficient and essential for STP. The observation that immunization of C57BL/6 mice produces higher antibody titers against BS proteins when compared to immunization of BALB/cJ mice might contribute to the superior protection against a lethal BS challenge observed in the former.
Antibodies can function independent or dependent of the FC portion of antibodies. FC-independent mechanisms include interference with pathogen activities by steric hindrance or blocking of target proteins (e.g. pathogen ligands for host cell infection). The FC-dependent mechanisms include complement-mediated lysis of the target pathogen and opsonization of the pathogen or pathogen-infected cell, flagging it for phagocytosis or destruction by FC-receptor (FCR)-bearing cells. To determine which mechanisms were playing a role in the antibody-mediated STP observed, we immunized C57BL/6 mice with LAGAP as before, depleted them of T cells and then additionally depleted complement via injection of cobra venom factor (CVF). CVF is a C3 convertase, which rapidly and efficiently depletes complement within hours of administration for 3–5 days[33] (S3A Fig). We injected 30 μg of CVF 6 hours prior to challenge and 4 days after challenge to ensure complement depletion throughout BS challenge. When LAGAP-immunized C57BL/6 mice were depleted of complement and T cells, only 40% survived a lethal blood stage challenge (Fig 3A). In contrast, 100% of immunized C57BL/6 mice lacking T cells but not depleted of complement survived the same challenge (Fig 1B). This indicates a strong role for complement-mediated destruction of opsonized parasites and/or iRBCs in the elimination of a BS infection in immunized mice. We also performed similar immunizations with C57BL/6 FCRγ-/- mice—which lack the γ-chain subunit of the FcγRI, FcγRIII and FcεRI receptors—to determine the role of FC-receptor binding in protection. These mice developed antibody titers against sporozoites and BS parasites that were comparable to wild type C57BL/6 mice (S3B and S3C Fig). Yet, lack of FCR functions also resulted in a reduction of mouse survival after lethal BS challenge from 100% (Fig 1B) to 40% (Fig 3B), implicating this effector pathway in LAGAP-elicited antibody-mediated protection. Elimination of all FC-dependent effector functions by CVF administration in immunized FCRγ-/- mice further reduced survival to 20% (Fig 3C and 3D). This again indicates a strong role for FC-dependent antibody effector mechanisms in LAGAP-immunized mice. The survival of a small proportion of mice suggests that FC-independent basic neutralization of parasites by LAGAP-elicited antibodies is also contributing to protection, although this was minor when compared to FC-dependent protection.
LAGAP elicit STP and antibodies play a pronounced role in this protective immunity. It has been shown previously that RAS and early liver stage-arresting GAP (EAGAP) do not elicit STP[14,34]. Thus, we predicted that the antibodies mediating STP are elicited by antigens expressed in late liver stage parasites and that these antigens are shared with BS parasites. To analyze the targets of STP, we investigated the stages of the parasite that are recognized by LAGAP-elicited antibodies using immunofluorescence assay (IFA). As a control, we used serum collected from C57BL/6 mice immunized with the EAGAP, Pysap1-, which efficiently invades hepatocytes but is completely attenuated by 6h post infection[35]. Antibodies from both EAGAP and LAGAP-immunized mice recognized sporozoites with a circumferential surface-staining pattern, likely indicative of CSP recognition (Fig 4A). Staining of liver stage parasites 24h post-infection with the same immune sera also showed a circumferential pattern for both sera (Fig 4B). Interestingly, both immune sera showed little/no reactivity against 33h-old liver stage parasites (Fig 4B). However, we observed pronounced differences in reactivity against 48h late liver stage parasites, a time when exoerythrocytic merozoites begin to differentiate. While there was little/no detectable reactivity with EAGAP immune serum (Fig 4B), LAGAP immune serum showed robust reactivity that localized to the exoerythrocytic merozoite surface and to the parasitophorous vacuole membrane (Fig 4B). We next performed IFA with immune sera on BS parasites to determine if the antibodies cross-reacted with these stages. While EAGAP immune serum had no detectable reactivity, LAGAP immune serum displayed an intense circumferential staining on merozoites that co-localized with MSP1 (Fig 4B). Interestingly however, we did not detect antibodies against either the 19 or 42kD fragment of merozoite surface protein 1 (MSP1) in LAGAP-immunized mice (S4 Fig). This provides an unprecedented demonstration that LAGAP immunization elicits antibodies against the late liver stages and BSs, which are mostly reactive with merozoite surface determinants. This could constitute one major mechanism by which STP is achieved.
Both LAGAP immunized BALB/cJ and C57BL/6 immunized mice produce antibodies that can recognize BS proteins by ELISA with C57BL/6 producing slightly higher titers (Fig 2B). This quantitative difference, however, cannot explain the inferior protection afforded by antibodies in BALB/cJ as passive transfer of C57BL/6 immune serum to naive BALB/cJ mice results in antibody titers as low as actively immunized BALB/cJ mice (S5 Fig), yet these passively immunized BALB/cJ mice are still protected against a BS challenge (Fig 1D). Thus, the differential protection could be due to different BS antigens being recognized by antibodies from the two strains or, given the demonstrated role of FC-mediated functions, by differences in the isotype distribution of the antibodies. To determine if the antibodies produced by the two strains of mice differ qualitatively by either specificity or isotype, we performed IFAs on iRBCs using serum from both LAGAP-immunized BALB/c and C57BL/6 immunized mice and secondary antibodies representing different IgG isotypes. Immunized C57BL/6 mice produced IgG of both IgG1 and 2b isotypes which co-localized with MSP1 at the surface of exoerythrocytic merozoites (Fig 5). In contrast, IFAs using serum from LAGAP-immunized BALB/cJ mice showed antibodies that are primarily of the IgG2b isotype and recognized the parasite interior (Fig 5). Quantification of immune serum staining patterns in 65 iRBCs confirmed the dichotomy of C57BL/6 serum recognizing the periphery of schizonts, whereas BALB/c immune serum recognized the parasite interior (Table 1). Western blots probing BS lysates with immune sera also demonstrated a distinct set of proteins recognized by sera from C57BL/6 immunized mice that were not apparent in serum from BALB/cJ immunized mice (S6 Fig). Thus, although immunized BALB/cJ mice produce anti-BS antibodies that were detectable by ELISA, IFA and Western blot, these antibodies offered no protection against a lethal BS challenge (Fig 1C). Therefore, whereas LAGAP-immunized BALB/cJ mice produce non-protective anti-BS antibodies, LAGAP-immunized C57BL/6 mice produce antibodies that recognize a unique set of BS antigens capable of potent STP.
LAGAP are unique amongst all current malaria immunization strategies in that they are designed to arrest the immunizing parasites late in liver stage development, cause no exposure to BS parasites and yet protect against PE parasite- and BS parasite challenge [28]. Here, we show that STP is mediated by T cells and antibodies, with that the latter recognizing antigens shared between the late liver stage and BS parasites. Immunization strategies focusing on single stages of infection must either be 100% effective in preventing PE infection of the liver or they must overcome the significant antigenic diversity, immune evasion mechanisms and high parasite burden present during BS infection in order to prevent disease and transmission. Other whole sporozoite vaccination types such as RAS or EAGAP provide potent, antigenically diverse PE immunity but they require complete prevention of development of even a single liver stage parasite. Otherwise they would fail to confer protection. In contrast, LAGAPs that elicit STP can maintain efficacy in the face of potentially leaky PE protection and breakthrough BS infection if one or a few parasites escapes PE immunity. Here, we demonstrate for the first time that in addition to PE immunity, immunization with LAGAP invokes both protective cellular and humoral BS immune responses. This not only provides a platform for investigation of novel cross-protective antigens and immune mechanisms, but together with the robust PE immunity observed after LAGAP immunization[24] provides further rationale for development of LAGAP for potential use in human immunization.
It has been previously shown that immunization with LAGAP elicits both robust cellular and humoral PE immunity[24,28]. Thus, it was reasonable to hypothesize that the observed STP against a BS challenge in LAGAP-immunized mice could be mediated by antibodies, CD4+ or CD8+ cells, as all have also been implicated in controlling BS infection[36–41]. Our data demonstrate that immunization with LAGAP elicits functional T cell responses to BS parasites that are essential for protection in BALB/cJ mice. Further studies using antibody-deficient mice on the BALB/cJ background would be required to determine which cell types are involved and if these cells are sufficient for protection in the absence of antibodies. Conversely, immunized AID-/- mice on the C57BL/6 background were unable to control a lethal BS challenge, pointing to antibodies as critical for protection. However, their parasitemia was curtailed and their time to death longer than WT controls, indicating a role for effector T cell immunity in this strain as well. CD8 T cells are widely recognized as essential effectors in eliminating liver stage parasites[8], and their role in BS protection is becoming more evident[38,39,41]. Although data demonstrating a clear role for CD8 T cells in mediating blood stage immunity in humans is lacking, identification of the antigens recognized by CD8 T cells in LAGAP-immunized BALB/cJ mice might be useful as these antigens are potentially present in multiple stages, are protective targets and thus could be prime candidates for a cross-stage protective T cell subunit vaccine.
In contrast to the increasingly appreciated role of T cells in BS immunity, antibodies have long been considered the main mechanism of protection against BS parasitemia and disease. This is based on early studies showing that passive transfer of convalescent serum from malaria-experienced individuals to unprotected individuals resulted in protection against BS disease [42,43] and high antibody titers against BS antigens correlate with reduction of morbidity and mortality in endemic areas [43,44]. However, whether or not this is mediated simply by antibody binding and impairment of merozoite activities, such as invasion, or mediated also by FC-dependent effector mechanisms still remains unclear. One study in mice using passive transfer of Py hyperimmune sera or an anti-MSP1 mAb to wildtype and FCRγ-/- mice suggested that FC-mediated mechanisms are dispensable[45]. However, additional studies in animal models and naturally immune individuals highlighted the importance of “cytophilic” antibodies (IgG1 and IgG3 in humans, IgG2a/b in mice) acting through FC-dependent functions for control of BS parasitemia[46–54]. In our current study, the antibodies engendered by LAGAP immunization are strongly dependent on FCR binding as the majority of immunized FCRγ-/- mice lost protection despite high levels of antibodies. FC-mediated complement fixation and destruction of antibody-bound iRBCs, merozoites or parasite proteins in immune complexes (the “classical complement pathway”) has been poorly defined. Only a few in vitro studies[55–57] have implicated complement fixation in the destruction of parasites while the one in vivo study conducted in non-human primates concluded that complement depletion via CVF had no impact on natural control of parasitemia[58,59]. In contrast, opsonized-iRBC phagocytosis by macrophages has been well documented and has been correlated with protection in naturally immune individuals[52,60]. Our data suggest a strong role for the classical complement pathway as immunized mice lacking complement showed a severe defect in controlling BS infection in the presence of LAGAP-induced antibodies. Importantly, localization data indicate that the antibodies do not preferentially bind the surface of iRBC but strongly react with the merozoite surface, suggesting that these antibodies bind and fix complement directly on the merozoite. This is in line with a previous study demonstrating that antibodies recognizing the merozoite protein SERA have enhanced inhibitory capacity in the presence of complement in vitro[57]. Complement fixation by opsonized merozoites could enhance parasite clearance by a number of mechanisms including direct killing via the membrane attack complex, recruitment of leukocytes via generation of anaphylatoxins (C3a and C5a) or by opsonization and subsequent phagocytosis of complement-bound parasites. Regardless of the mechanism, our data provide the first in vivo evidence of the functional importance of the classical complement pathway playing a major role in the control of blood stage parasitemia.
We have previously speculated that the STP resulting from LAGAP immunization is targeting protective antigens that are shared between the late liver stages and BS parasites[28]. This arises from the observation that parasites arresting development early in the liver, such as RAS or EAGAP, do not afford STP[15,34]. Here, we provide direct evidence that it is indeed antigens shared between the late liver stages and BSs that are the targets of protection. IFAs using serum from mice immunized with an EAGAP (Pysap1-,) and the LAGAP show that immunization with both parasites elicits antibodies against sporozoites and liver stages up to 24 h post infection. This is consistent with the presence of CSP on the sporozoite- and liver stage surface and the abundant anti-CSP antibody titers in the immune sera. In contrast, only LAGAP-immune serum recognized late liver stages, exoerythrocytic merozoites and BS merozoites. Combined with our data showing that LAGAP-induced antibodies alone provide protection from iRBC challenge, this confirms that there are indeed yet to be identified antigens in late liver stage parasites and BS parasites capable of eliciting STP. Targeting these antigens by both T cells and antibodies (i.e. with viral vectors) could allow for multiple opportunities to eliminate the parasite in both the liver and blood if it is indeed the same antigens providing both PE and blood stage protection. Importantly, this immunity does not appear to target MSP119 or MSP142, which although capable of conferring protection in mice, has failed to protect in clinical studies[26,27].
How exactly LAGAP antigens are acquired and presented by antigen presenting cells (APC) remains to be elucidated. Numerous types of cells in the liver are capable of antigen uptake and presentation including Kupffer cells (KC, liver-resident macrophages), multiple types of dendritic cells (DC), liver sinusoidal endothelial cells (LSEC) and even hepatocytes[61,62]. Hepatocytes only possess MHCI and can present parasite antigens[63] but lack MHCII and thus this cannot explain the antibody responses to late liver stages we observed. Even though LSEC are very efficient at presentation of exogenous antigen[62], their ability to generate the type of mature, class-switched IgG response seen in LAGAP immunization has not been established[61]. A likely scenario is that as the LAGAP parasite dies late in liver stage development, the hepatocyte undergoes apoptosis[64,65] and releases parasite antigens to liver-resident DC or KC[66], which prime responses to late liver stage/blood stage antigens. Priming against early PE antigens likely occurs against extracellular sporozoites and liver stage parasites that die early as a part of normal parasite infection[29,64,65]. This speculation is bolstered by the fact that we see serum reactivity to early (12 and 24h) and late (48h) liver stages but not to 33h as fewer parasites are dying at this mid liver stage. These liver-resident APCs can then migrate to draining lymph nodes where they prime productive adaptive immune responses. In support of this, a recent report by Lau et al. demonstrates that following immunization with RAS by iv injection, substantial parasite-specific T cell activation occurs in the liver-draining lymph nodes at a rate that is surpassed only by the spleen[67]. Another study showed accumulation of CD8α+ DCs in the liver of RAS-immunized mice and that these DCs were capable of directly activating T cells in vitro[68]. Several reports identify peripheral lymphoid organs such as the spleen as instrumental in immune priming following sporozoite immunization[67,69], but direct evidence of peripherally primed T cells being the mediators of protection after whole sporozoite immunization by iv injection is lacking. As the antigens mediating STP in our model are present in late liver stages, it is more likely that liver-resident APCs are responsible for priming the immune response to late liver antigens in the liver-draining lymph nodes.
Antibodies in LAGAP immune sera overwhelmingly recognize the surface of the developing and mature merozoite and co-localize with MSP1 in late liver stage and BS parasites. MSP1 is the most abundant protein on the merozoite surface and antibodies against this protein have been shown to be protective against BS infection in mice [70–73]. However, we were unable to detect antibodies against either protective MSP1 regions in serum from LAGAP-immunized mice[74,75]. Thus, the protective antigen(s) are as yet unidentified merozoite surface protein(s). Our data from IFAs and Western blot indicate that the protective antibodies in LAGAP-immunized C57BL/6 mice recognize different antigens than non-protective antibodies from BALB/c immunized mice. Accordingly, by identifying the antigens uniquely recognized by C57BL/6 immune serum, it is conceivable that a subset of these antigens could be incorporated into a multi subunit vaccine that could induce STP.
The potent STP observed in our studies also lends support to using LAGAP as a whole parasite vaccination strategy. The only other example of true STP with live parasites has been observed in mice immunized with iRBC under chloroquine cover[76]. This immunization strategy in mice controlled liver infection via CD4+ and CD8+ T cells, conferred partial BS immunity and has been demonstrated as protective against sporozoite challenge in humans[77]. Multi-stage protection has also been demonstrated in animal models of ITI (infection with wild type sporozoites under drug cover) where there is both antibody-dependent and independent protection against both a sporozoite and BS infection[25,76,78]. However, this is not truly stage transcending protection as the development of BS immunity requires exposure to low levels of BS parasitemia during immunization[25,78].
Trials using cryopreserved RAS in humans have demonstrated that administration of live, attenuated sporozoites is effective, safe and well-tolerated[79]. Significant hurdles remain in manufacturing and delivering a live, attenuated sporozoite vaccine, but the success of RAS in humans has provided the impetus for creative solutions to these barriers. Yet, no gene knockout in the human-infective P. falciparum species has been created that is phenotypically similar to the LAGAP described here and knockout of the orthologous P. falciparum gene results in a parasite incapable of forming sporozoites[80]. Given the promise of superior and stage-transcending immunity, development of a late liver-arresting P. falciparum GAP that is free from breakthrough during immunization is of high priority and should be under intense investigation.
In summary, we have shown that immunization with LAGAP can elicit both T cell and antibody-mediated immunity to BS parasites via recognition of antigens shared between the late liver stage- and BS parasites. Furthermore, antibodies act through complement and FCR binding to control and eliminate BS parasitemia. Since these T cells and antibodies are both highly efficacious and directed against potentially novel antigens, mechanistic studies using this model can critically inform the development of the next generation of subunit vaccines. These should be designed to elicit T cells as well as antibodies of the correct isotype, each directed against critical antigens and effective in eliminating liver stages and blood stage parasites.
6–8 week old female BALB/cJ and C57BL/6 mice were purchased from the Jackson Laboratory. Age-matched female FCRγ-/- mice on the C57BL/6 (B6.129P2-Fcer1gtm1Rav N12, model 583) background were purchased from Taconic Biosciences, Inc. All mice were maintained in a pathogen-free facility accredited by the Association for Assessment and Accreditation of Laboratory Animal Care at the Seattle Biomedical Research Institute. All experiments were conducted in accordance with animal protocols approved by the Institutional Animal Care and Use Committee.
Six-to-eight week old female SW mice were injected with blood from Py knockout (fabb/f- or sap1-)-infected mice to begin the growth cycle. The infected mice were used to feed female Anopheles stephensi mosquitoes after gametocyte exflagellation was observed. On days 14–17 post infectious blood meal, salivary gland sporozoites were isolated from the mosquitoes for experimentation.
Mice were immunized by injecting 50,000 sporozoites intravenously via the tail vein two weeks apart. As a control, equivalent amounts of salivary gland debris from uninfected mosquitos were used.
Frozen blood stocks of PyXL or PyYM-infected blood containing 1% iRBCs was ip-injected into BALB/c or C57BL/6 mice and allowed to develop for 2–4 days until parasitemia reached a maximum of 1% as determined by Giemsa-stained thin smear. These mice were terminally bled via cardiac puncture and the blood diluted in PBS to contain 10,000 iRBCs/200μL. iRBCs were then iv-injected at a volume of 200μL/mouse into congenic recipient mice. Parasitemia was monitored by Giemsa-stained thin smears beginning on day 3 post-infection. Mice were euthanized when parasitemia reached 60% or became moribund.
Serum from mice immunized with three doses of Pyfabb/f- sporozoites or uninfected salivary gland debris (mock) was collected on day 7 and day 14 after the final immunization and pooled. Naïve mice were intravenously injected with 300 μl of pooled serum on day 0, 3 and 5 after a challenge with 10,000 lethal PyXL or PyYM iRBCs injected intravenously.
CD8+ and CD4+ T cells were depleted in mice as previously described [22]. Briefly, 0.5 mg of anti-CD8 mAb 2.43 (BioXCell) and 0.35 mg of anti-CD4 mAb 1.5 (BioXCell), or 0.85 mg of isotype control rat IgG2b (BioXCell) was iv-injected into mice 24 hours prior to parasite challenge. T cell depletion was confirmed before each challenge by collecting 50–100 μl of peripheral blood via the retro-orbital plexus from each mouse and assaying peripheral blood lymphocytes by flow cytometry staining for CD19, CD3, CD4 and CD8.
Cobra venom factor (CVF) is a complement activating C3b analog that when administered rapidly depletes complement (Vandenberg 1991). For complement depletion, mice were administered 30 μg of CVF intraperitoneally 6h prior and 4 days after iRBC challenge. Depletion of complement was confirmed prior to challenge by C3 sandwich ELISA (Genway Biotech).
Serum was isolated from peripheral blood at day 25 post-immunization, immediately prior to challenge as previously described[24]. ELISA plates (Corning, Inc.) were coated with full length PyCSP protein at a concentration of 0.1 μg/mL or with 2 μg/mL of either Py sporozoite or BS lysate in calcium bicarbonate/sodium carbonate coating buffer overnight at 4°C. For MSP1 ELISAs, plates were coated at 0.1μg/mL of either the 19 or 42kD fragment (generously provided by Dr. James Burns) as above. Plates were washed prior to addition of a 1:800 (for PyCSP), 1:20 (for sporozoite and BS lysate) or 1:2000 (for MSP1 19 and 42kD) dilution of serum in duplicate followed by incubation at 37°C for two hours. After washing, anti-mouse IgG conjugated to HRP (SouthernBiotech) was added at a 1:2000 dilution for an additional 2h at 37°C. Plates were again washed and 100 μL of SigmaFast OPD (Sigma-Aldrich) substrate was added for 2–10 minutes prior to colorimetric detection of antibodies by measuring absorbance at 450 nm.
Sporozoites, infected hepatocytes and infected red blood cells were harvested, fixed and stained as previously described [81,82]. Briefly, fixed cells were stained with a 1:200 dilution of serum collected from Pysap1- and Pyfabb/f- immunized C57BL/6 mice. Rabbit antibodies against BiP and MSP-1 were used as control antibodies for early liver stages and late liver stages/BSs, respectively. Fluorescently labeled secondary antibodies (Alexa Fluor 488 or Alexa Fluor 594) from Life Technologies were used to detect mouse IgG (catalog # A-11059), IgG1 (catalog # A-21125) and IgG2b (catalog #A-21145). Images were acquired using Olympus 1 x 70 Delta Vision deconvolution microscopy. For quantification of staining pattern, slides were prepared as above and blinded to the microscopist. Infected red blood cells containing schizonts were identified by MSP1 staining and anti-mouse IgG staining was classified as “interior”, “exterior” or “both” based on the MSP1 border.
Calculations and statistical tests indicated in the figure legends were performed using GraphPad Prism. A p value < 0.05 was considered significant.
All animal procedures were conducted in accordance with and approved by the Seattle BioMed Institutional Animal Care and Use Committee (IACUC) under protocol SK-09. The Seattle Biomed IACUC adheres to the NIH Office of Laboratory Animal Welfare standards (OLAW welfare assurance # A3640-01).
|
10.1371/journal.pgen.1005858 | Expression Quantitative Trait Locus Mapping Studies in Mid-secretory Phase Endometrial Cells Identifies HLA-F and TAP2 as Fecundability-Associated Genes | Fertility traits in humans are heritable, however, little is known about the genes that influence reproductive outcomes or the genetic variants that contribute to differences in these traits between individuals, particularly women. To address this gap in knowledge, we performed an unbiased genome-wide expression quantitative trait locus (eQTL) mapping study to identify common regulatory (expression) single nucleotide polymorphisms (eSNPs) in mid-secretory endometrium. We identified 423 cis-eQTLs for 132 genes that were significant at a false discovery rate (FDR) of 1%. After pruning for strong LD (r2 >0.95), we tested for associations between eSNPs and fecundability (the ability to get pregnant), measured as the length of the interval to pregnancy, in 117 women. Two eSNPs were associated with fecundability at a FDR of 5%; both were in the HLA region and were eQTLs for the TAP2 gene (P = 1.3x10-4) and the HLA-F gene (P = 4.0x10-4), respectively. The effects of these SNPs on fecundability were replicated in an independent sample. The two eSNPs reside within or near regulatory elements in decidualized human endometrial stromal cells. Our study integrating eQTL mapping in a primary tissue with association studies of a related phenotype revealed novel genes and associated alleles with independent effects on fecundability, and identified a central role for two HLA region genes in human implantation success.
| Little is known about the genetics of female fertility. In this study, we addressed this gap in knowledge by first searching for genetic variants that regulate gene expression in uterine endometrial cells, and then testing those functional variants for associations with the length of time to pregnancy in fertile women. Two functional genetic variants were associated with time to pregnancy in women after correcting for multiple testing. Those variants were each associated with the expression of genes in the HLA region, HLA-F and TAP2, which are have not previously been implicated female fertility. The association between HLA-F and TAP2 genotypes on the length of time to pregnancy was replicated in an independent cohort of women. Because HLA-F and TAP2 are involved in immune processes, these results suggest their role in specific immune regulation in the endometrium during implantation. Future studies will characterize these molecules in the implantation process and their potential as drug targets for treatment of conditions related to implantation failure.
| Natural variation in fertility traits is heritable in humans [1], yet identifying genes contributing to these traits remains challenging. Although genome-wide association studies (GWAS) have identified variants associated with many other complex phenotypes, its application to fertility traits is challenging. In particular, widespread contraceptive use among fertile couples and significant clinical heterogeneity among infertile couples makes it difficult to sample large numbers of fertile subjects with unprotected intercourse or infertile subjects whose inability to conceive results from the same underlying biological processes. Although a few GWAS of male fertility [2] or infertility [3, 4] traits have identified promising candidate genes, to date there have been no such studies in women.
To address these limitations, we have focused our genetic studies of fertility on members of a founder population, the Hutterites [1, 2, 5–8]. This communal living group of European ancestry has limited contraceptive use, a uniform desire for large families, a prohibition of smoking, and fertility rates that are among the highest ever reported [9, 10]. For example, only 2% of Hutterite couples are childless [9] compared to 10–15% of the general population [11]. Whereas miscarriages of clinically recognized pregnancies among Hutterite couples is 15.6% [8], nearly identical to estimates of clinically recognized miscarriage rates in outbred populations [12], recurrent miscarriages in childless couples are rare (0 of 525 interviewed Hutterite women [1]) compared to 5% in the general population [13]. Moreover, their communal lifestyle ensures that sociocultural factors influencing fertility are relatively uniform among Hutterite couples [1, 14]. We have proposed that their naturally high fertility rates, their reduced variance in environmental and lifestyle factors, and their limited gene pool due to the founder effect make the Hutterites an ideal population in which to dissect the genetic architecture of reproductive traits [1, 2, 15]. Here, we use an integrated strategy that first identifies a set of candidate regulatory single nucleotide polymorphisms (SNPs) in mid-secretory phase endometrium by expression quantitative trait locus (eQTL) mapping in non-Hutterite women with two or more previous miscarriages. We then tested for associations between those putatively functional expression (e)SNPs and fecundability in Hutterite women who are participants in a prospective study of pregnancy outcome [7, 8], and replicated the significant findings in an independent sample of women [16]. We report here the discovery of independent associations between SNPs that are eQTLs for the HLA-F and TAP2 genes in mid-secretory phase endometrium and fecundability (the probability of achieving pregnancy), thereby implicating maternal HLA region genes for the first time in implantation processes.
We performed eQTL mapping in the mid-secretory phase endometrium, corresponding to the luteal phase of the ovarian cycle, from 53 women with two or more early pregnancy losses, using 378,362 common (≥10%) SNPs that were within 200kb of one or more of the 10,191 genes detected as expressed in these tissues (i.e., cis-eQTLs) (see Methods). We observed 423 cis-eQTLs (416 unique SNPs) for 132 genes at a false discovery rate (FDR) of 1% (S1 Dataset).
We next looked for gene ontology enrichments in the genes associated with eQTLs at a FDR of 1% using DAVID [17, 18] and GREAT [19]. We found an enrichment of the GO Biological Process of “antigen processing and presentation” (DAVID, FDR 2.5x10-5), the GO Molecular Function of “MHC class 1 receptor activity” (DAVID, FDR 5.30x10-4), and GO Cellular Component “MHC class 1 protein complex” (GREAT, FDR 2.02x10-6). Many of the DAVID and GREAT enrichments overlapped, and both highlighted the importance of immune related genes among those with eQTLs in mid-secretory phase endometrium.
To assess the clinical relevance of these eQTLs on female fertility traits, we first pruned the 416 SNPs for strong LD (r2 ≥ 0.95 in the Hutterites) and then carried forward 245 expression (e)SNPs for association studies in the prospective study participants.
We previously genotyped 208 of the 327 Hutterite women in a prospective study of pregnancy outcomes [6, 7, 15, 20] with the Affymetrix 500k or 6.0 genotyping chips. Using these genotypes as framework markers, we imputed all variants present in the whole genome sequences of 98 Hutterites to these women using PRIMAL, an imputation program that utilizes both pedigree- and LD-based imputation to provide on average of 87% coverage and >99% accuracy in the Hutterite pedigree [21]. From among the 245 (LD-pruned) SNPs that were eQTLs at a FDR 1%, genotypes for 189 (associated with the expression of 108 genes) were known for at least 85% of Hutterite women in the study of fecundability.
We compared the length of the intervals from the resumption of menses after a prior pregnancy or miscarriage or following the discontinuation of birth control use (referred to as time0) to a positive pregnancy test in women who were not nursing at time0 of each included interval. For first pregnancies, we considered the length of the interval from the first menses after marriage to a positive pregnancy test (see Methods). If pregnancy occurred prior to the resumption of menses (or before the first period after marriage in first pregnancies), we considered the interval to be 14 days. Data were available for 191 intervals in 117 women (see Methods); 178 of the observed intervals resulted in a pregnancy at the time of the last follow-up. We used life-table analysis to compare intervals between genotype classes, adjusting for two significant covariates: maternal age and number of prior births (classified as 0–1, 2–3, or ≥4) (see Methods). Among the 189 eSNPs that we examined, genotypes for 21 were associated with the length of the interval to pregnancy at a P < 0.05 (Table 1). Two of these eSNPs were significant at a FDR of 5% and after Bonferroni correction for 189 tests. The most significant association was with rs2071473, an eSNP associated with expression of the TAP2 gene in the HLA class II region; the C allele at this SNP was associated with longer intervals to pregnancy and higher expression of TAP2 gene in mid-secretory phase endometrium (Fig 1). The median interval lengths to pregnancy were 2.0 (lower, upper quartile 1.2, 4.7), 3.1 (1.9, 6.2), and 4.0 (2.0, 7.6) months among women with the TT, CT, and CC genotypes, respectively, at this eSNP (P = 1.3x10-4). The second association was with rs2523393, an eSNP associated with expression of the HLA-F gene in the HLA class I region; the G allele at this SNP was associated with longer intervals to pregnancy and lower expression of HLA-F in mid-secretory phase endometrium (Fig 2). The median interval lengths to pregnancy were 2.3 (1.8, 4.5), 2.6 (1.4, 4.8), and 4.9 (2.0, 11.7) months among women with the AA, AG, and GG genotypes, respectively, at this eSNP (P = 4.0x10-4). For both eSNPs, intervals were longer and genotype differences more pronounced among women at lower parity (S1 and S2 Figs). Among the 21 eSNPs with P <0.05, nine (43%) were associated with expression of HLA region genes: one with TAP1, three with HLA-F, three with HLA-G, and two with MICA, consistent with the gene ontology analysis identifying enrichments for genes with antigen processing and presentation functions among those with eQTLs in mid-secretory phase endometrium. The results for all 189 eSNPs and their associated genes are shown in S2 Dataset.
Previous studies of HLA and fertility in the Hutterites have shown that HLA matching between partners for alleles at the class II locus HLA-DRB1 is associated with reduced fecundability, presumably due to the higher risk for class II compatible embryos among these couples [7]. To rule out that maternal-fetal compatibility at the TAP2 or HLA-F locus accounts for the observed effects in this study we repeated the fecundability analysis, first stratifying couples based on husband’s genotype (rather than wife’s genotype) and then stratifying couples based on the wife’s genotype (as above) but now including the husband’s genotype as a covariate. We reasoned that if longer intervals are due to maternal-fetal compatibility and not maternal genotypes per se, then results of analyses stratifying by husband’s genotype should yield results similar to analyses stratified by wife’s genotype, and analyses including both husband’s and wife’s genotypes should be more significant than analyses considering either one individually. Neither eSNP was significant in the analysis considering the husband’s genotype as a main effect on length of intervals to pregnancy (HLA-F rs2523393 P = 0.94; TAP2 rs2071473 P = 0.56). When husband’s genotype was included as a covariate in the model, the P-values were reduced from 4.0x10-4 to 0.0014 for rs2523393 and from 1.3x10-4 to 0.0015 for rs2071473, but the effect size associated with the risk alleles remained largely unchanged (β coefficients changed from 0.39 to 0.34 for rs2523393 and -0.44 to -0.36 rs2071473 when husbands’ genotypes were included as a covariate). These data indicate that maternal genotype at these two eSNPs is driving the association with time to pregnancy; there is no evidence for paternal or fetal genotype effects at these eSNPs contributing to interval lengths.
Although these two eSNPs reside at opposite ends of the HLA region and are separated by ~3Mb, there are moderate levels of LD between them in the Hutterites (r2 = 0.19). To determine the statistical independence of the associations with fecundability, we repeated the time to pregnancy analysis but included the genotype at the other eSNP as a covariate. In the analysis of rs2071473 (TAP2) that included genotype at rs2523393 (HLA-F) as a covariate, the effect size and P-value changed from β = 0.39 (P = 1.3x10-4) to β = 0.23 (P = 0.0064); in the analysis of the rs2523393 (HLA-F) that included genotype at rs2071473 (TAP2) as a covariate, the effect size and P-value changed from β = -0.44 (P = 4.0x10-4) to -0.34 (P = 0.047). Thus, while the magnitude of each association is reduced in the analyses conditioning on the alternate eSNP, both retain independent effects on fecundability. The observed reduction in β values and significance is likely due to the LD between the SNPs. To further examine this, we stratified the women into three groups based on being homozygous at both, one, or neither of the high risk (longer interval) alleles at each eSNP (CC at rs2071473 [TAP2] and GG at rs2523393 [HLA-F]) (Fig 3). If the effects at these two loci were independent, then women who are homozygous for the high risk allele at both eSNPs should have longer intervals than women who are homozygous at only one or neither high risk allele. Indeed, intervals to pregnancy were longest among women homozygous for both rs2071473-CC and rs2523393-GG (median interval 5.2 months [1.9, 11.0 months]), intermediate among homozygous for only one of the high risk alleles (median interval 4.0 months [2.1, 9.3 months], and shortest among women who were not homozygous for either high risk allele (median interval 2.4 months [1.4, 4.8 months]) (P = 2.9x10-4). Moreover, women homozygous for the risk alleles at both the TAP2 and HLA-F eSNPs had significantly longer intervals compared to women who were homozygous for a risk allele at only one of the two eSNPs (P = 1.6x10-5). Taken together these analyses indicate that the TAP2 and HLA-F associations are independent and have additive effects on fecundability.
Using the same approach as that used in the Hutterites, we first examined the genotype effects of each SNP on fecundability and then the joint effects of the combined genotypes at each locus. At rs2071473, the TAP2 eSNP, the median interval lengths to pregnancy were 5.0 (4.0, 7.0), 6.0 (4.0, 8.2), and 6.0 (4.0, 9.0) months among women with the TT, CT, and CC genotypes, respectively (P = 0.083; Fig 4A). At rs2523393, the HLA-F eSNP, the median interval lengths were 5.0 (4.0, 8.5), 6.0 (4.0, 8.0), and 6.0 (5.0, 10.0) months among women with the AA, AG, and GG genotypes, respectively (P = 0.155; Fig 4B). Although these results did reach nominal significance in the RFTS cohort, the 95% confidence intervals of the ORs in the Hutterites and RFTS cohort overlap (S1 Table). In the combined analysis, intervals to pregnancy were longest among women homozygous for both rs2071473-CC and rs2523393-GG (median interval 8.0 months [5.5, 12.5 months]), intermediate among homozygous for only one of the high risk alleles (median interval 6.0 months [4.0, 9.0 months], and shortest among women who were not homozygous for either high risk allele (median interval 5.0 months [4.0, 7.0 months]) (P = 0.033; Fig 4C), as we observed in the Hutterites.
Because there are many SNPs in strong LD with our lead eSNPs in the Hutterites, it cannot be inferred from association studies which of these linked SNPs are the true causal variants. To address this question, we used in silico analyses to determine which of the fecundability-associated eQTLs are in or near regulatory elements in decidualized human endometrial stromal cells [22, 23], ENCODE-annotated functional sites in the endometrial cell lines ECC-1 and Ishikawa [24], as well as the complete ENCODE regulatory element dataset. To interrogate more completely the variation in these regions, we used whole genome sequence data in the Hutterites to survey all variation in the 500kb windows flanking each of the two eSNPs that were associated with fecundability. After filtering our variants with minor allele frequencies <0.10 and call rates <85%, 4,442 variants remained in the TAP2 region and 2,675 variants remained in the HLA-F region. We then filtered these variants based on their LD with each lead eSNP and retained the 70 variants with LD r2 ≥ 0.7 with rs2071473 and the 62 variants with LD r2 ≥ 0.7 with rs2523393. The variants that had LD r2 ≥ 0.7 with the lead eSNP in each region defined an approximately 6kb window around the lead SNP. We repeated the association studies with all variants within each 6kb window and fecundability. Because many of these variants were not included in the eQTL study in the 53 women with recurrent early pregnancy loss, we also imputed the missing genotypes using whole genome sequences from 100 European American individuals [25] (see Methods) and performed eQTL mapping in the mid-secretory phase endometrial RNA using these variants, as described above.
The lead eSNP at the TAP2 locus, rs2071473, is located within an intron of the HLA-DOB gene and is near (~600bp) an NF2R2 transcription factor (TF) binding site in hESC; EP300, FOXM1, ATF2, and RUNX3 binding sites in a B-lymphoblastoid cell line (GM12878); and a DNase-I hypersensitivity site in 40 ENCODE cell lines (Fig 5A). This SNP is also within ~800bp of a FAIRE peak in hESC. Another SNP, rs2856995, that is 732bp upstream of and in perfect LD (r2 = 1) with rs2071473 resides within the NF2R2 binding site and 92bp upstream of the FAIRE site in hESC. Multiple other variants are located adjacent to FAIRE sites and among the 36 of these variants with eQTLs, 34 were eQTLs only for TAP2 (FDR <15%), and two were eQTLs for TAP1 (FDR = 14%) (S3 Dataset).
The lead eSNP at the HLA-F locus, rs2523393, is located within an intron of HLA-F-AS1, an antisense transcript that was not expressed in mid-secretory endometrium. The eSNP is about 10kb downstream of HLA-F, and in perfect LD (r2 = 1) with a cluster of SNPs ~360bp away that reside within multiple ENCODE-annotated functional sites in endometrial cell lines [26] (Fig 5B). One such SNP, rs2523389, is in a DNaseI hypersensitivity site in 120 ENCODE cell lines including the endometrial derived cell lines ECC-1 and Ishikawa treated with 10nM estradiol. This variant is also in a CTCF binding site that is present in 97 ENCODE cell lines, including the endometrial cell line ECC-1, and in a c-Myc binding site in a leukemia cell line. Another variant, rs2523391, is in a STAT3 binding site in a mammary gland cell line and 6bp from a CEBPB TF ChIP-seq binding site in HeLa and HepG2 cell lines, in addition to the same functional sites as rs2523389. Although the c-Myc, STAT3, and CEBPB binding sites were not present in the hESC or ENCODE endometrial cell lines, these transcription factors are essential for decidualization of endometrial stromal cells and the successful establishment of pregnancy [27–29]. Among the variants in LD with rs2523393 at r2 ≥ 0.7 and with eQTL results, all were most strongly associated with expression of HLA-F (S4 Dataset). One eSNP in the HLA-F promoter (rs1362126; r2 = 0.78 to rs2523393) was also an eQTL for HLA-G (FDR <1%), although to a lesser degree than for HLA-F (HLA-F eQTL P = 2.18x10-6, HLA-G eQTL P = 2.06x10-3).
Overall, these data indicate that our lead eSNPs and/or a small number of variants in perfect LD with those eSNPs are plausible causal candidates for the observed associations with fecundability in each region.
The mechanisms that allow the fetal allograft to avoid maternal immunologic rejection and survive over relatively long gestational periods in placental mammals are still incompletely known, although our understanding of these processes have advanced considerably since Medawar proposed this paradox over 60 years ago [30]. In particular, it has become clear that major histocompatibility complex (MHC) antigens, which play a central role in the rejection of non-self tissues, also contribute to maternal tolerance of the fetus, which is maintained in normal pregnancies. For example, our group previously demonstrated that matching of HLA antigens (the human MHC loci) between Hutterite couples is associated with longer intervals from marriage to each birth compared to couples not matching for HLA [31], and that longer intervals resulted from both higher miscarriage rates among couples matching for class I HLA-B antigens [8] and longer intervals to pregnancy among couples matching for class II HLA-DR antigens [7]. More recently, we reported associations between maternal HLA-G genotypes and miscarriage [6]. We and others have also shown associations between maternal or fetal HLA-G genotypes with recurrent pregnancy loss and preeclampsia [32–42]. Finally, recent studies have elegantly demonstrated that two HLA that are expressed by fetal extravillous cytotrophoblast (EV-CTB) cells at the maternal-fetal interface, HLA-G and HLA-C, are ligands for inhibitory receptors ILT2/IL4 and KIR2DL3, respectively, on immune cells [43–45]. Collectively these data indicate that multiple HLA molecules play important independent roles at the maternal-fetal interface in human pregnancy and that their effects can influence pregnancy outcome throughout gestation.
In this study, we hypothesized that perturbations of genes expressed in mid-secretory phase endometrial cells could affect implantation and be visible as delayed time to pregnancy in otherwise fertile couples. Although our study was unbiased with respect to genome location because we interrogated variation that was first identified through a genome-wide eQTL study, the results of both the eQTL study and the subsequent study of fecundability highlighted the importance of HLA region genes in achieving pregnancy. Among the eQTLs taken forward to studies of fecundability in the Hutterites, nine (43%) of the eSNPs with association P-values <0.05 were eQTLs for HLA region genes compared to 15% of all 189 eSNPs tested. Two of the associations with fecundability were significant at a FDR of 5%, remained significant after correction for multiple testing, and were replicated in an independent sample of fertile women: one SNP was with an eQTL for TAP2 and one for HLA-F. To our knowledge neither of these two HLA region genes has previously been directly implicated in pregnancy processes. We further demonstrated that eQTLs for these two HLA loci in mid-secretory phase endometrium are independently associated with fecundability in fertile women and that neither paternal nor fetal genotype at these loci contributed to these effects. These findings may be particularly relevant to women with primary infertility of unknown etiology, with recurrent implantation failure following in vitro fertilization (IVF), or possibly even with recurrent early pregnancy loss.
Both genes are intriguing candidates for fecundability genes. The TAP2 gene in the class II HLA region encodes the antigen peptide transporter 2 protein. TAP2 forms a heterodimer with TAP1 (encoded by the TAP1 gene, located 7kb away) in order to transport peptides from the cytoplasm to the endoplasmic reticulum, where they are loaded into assembling class I HLA molecules prior to their transport to the cell surface. The association of TAP complex with HLA class I molecules, including HLA-F [46], is critical for their expression on the cell surface [47]. HLA-F, which is located ~3Mb telomeric to TAP2, encodes a class I HLA protein that is considered “non-classical” because it has limited coding polymorphisms and restricted tissue distribution [48], and functions that are still poorly characterized but likely distinct from the classical class I HLA (HLA-A, HLA-B, HLA-C). In fact, recent studies have shown that HLA-F physically interacts with the KIR3DL2 and KIR2DS4 receptors on natural killer (NK) cells [49], an abundant and critical cell in the maternal uterus that proliferates during the secretory phase and then throughout pregnancy [50]. Later in pregnancy, HLA-F is expressed in EV-CTB [51–54], although its function in placental cells is not well characterized [52]. Our combined results suggest that perturbations in expression of either gene in endometrial cells in the mid-secretory phase influences implantation success, with overexpression of TAP2 and underexpression of HLA-F resulting in delayed time to pregnancy.
We found multiple variants in perfect LD with both eSNPs that reside in transcription factor binding sites and other regulatory elements in endometrial cell lines. The TAP2-associated variants are located within a NR2F2 (COUP-TFII) binding site. Multiple studies have shown that female mice deficient in NR2F2 have implantation failure, with impairments of both embryo attachment and uterine decidualization [55–57], and NR2F2 knock downs a human endometrial stromal cell line significantly reduces TAP2 expression [22]. These combined data suggest a potential mechanism for the association we observed with expression of TAP2 and fecundability in women. At the HLA-F locus, variants associated with gene expression level and fecundability are in DNaseI hypersensitivity sites, a marker of open chromatin and transcriptional activity, in multiple human endometrial cell lines [26], suggesting that one or more of these variants may indeed be causally associated with both gene expression and fecundability.
Although genome-wide association studies can be powerful approaches for identifying susceptibility loci for common diseases and complex phenotypes, they require very large sample sizes that may be infeasible to acquire for many important phenotypes. We used an alternative approach for mapping fecundability genes by first reducing the search space to SNPs that were associated with gene expression in a relevant tissue and then taking this smaller set of regulatory SNPs forward to an association study in carefully phenotyped subjects. This approach revealed two novel associations with fecundability and immediate intuition regarding the genes underlying each association, the relationship between gene expression and fecundability, and potential mechanisms for these associations. Future studies will be required to characterize the role of these molecules in the implantation process and to evaluate their potential as drug targets for treatment of conditions related to suboptimal implantation.
Fifty-eight women underwent endometrial biopsies as part of their clinical evaluation for recurrent pregnancy loss at the University of Chicago, after obtaining informed consent. These women were between the ages of 26 and 43 years and had at least two previous pregnancy losses before10 weeks gestation. Fifty-two (90%) were of European ancestry, two (3%) were of Asian ancestry, and four (7%) were of African ancestry. Medical records and individual diagnoses were not available to us for this study, and all women were included in the expression studies. Because recurrent miscarriage can result from many potential causes, and nearly half remain unexplained after evaluation, we reasoned that gene expression in the endometrium from these women would maximize variation in gene expression and increase our power to detect eQTLs. We were unable to obtain samples from women without a history of pregnancy loss for this study.
Endometrial biopsies were performed during the mid-secretory phase (9–11 days after endogenous luteinizing hormone [LH] surge, detected by each woman testing her daily urine) and immediately frozen on dry ice; samples were stored at -80°C until RNA was extracted, as previously described [58]. Histological examination of the biopsies confirmed endometrial tissue from the fundus of the uterus; endometrial glands and epithelium were present.
RNA was extracted from the endometrial biopsies using a phenol-chloroform phase separation with TRIzol per the manufacturer’s directions (Life Technologies Corp., Carlsbad, CA, USA) and RNeasy RNA extraction kit (Qiagen, Venlo, Netherlands; per manufacturer’s directions). RNA quality was assessed using the Agilent 2100 Bioanalyzer (Agilent Technologies, Santa Clara, CA). The average RNA integrity number (RIN) score [59] was 8.06 (range 6.7–9.3). We were unable to obtain RIN scores for two samples due to extremely low concentration. However, these two samples passed all gene expression QC and were therefore included in the eQTL mapping studies. Gene expression was measured in RNA from 58 individuals. We included triplicate samples for three women and duplicate samples for 29. Gene expression was interrogated using the Human HT12v4 Expression BeadChip (Illumina, Inc., San Diego, CA, USA), which contains 47,231 probes that target 11,121 unique RefSeq genes. cDNA synthesis, hybridization, scanning and image processing and returned probe intensity measurements were performed at the University of Chicago Functional Genomics Core. Intensity estimates were log-transformed and quantile normalized using the ‘lumi’ package in R [60] (see S3 Fig). To remove probes for targets that were likely not expressed, all probes that did not have a detection P-value <0.05 in at least 70% of the samples were removed (leaving 17,208 probes). We further removed probes that did not map uniquely to the HG19 genome using Burrows-Wheeler Aligner (BWA), and probes that contained CEU HapMap SNPs with the QC+ designation (4,825 probes total excluded). After quality control (QC), 12,383 probes remained and were included in our eQTL studies. Probe averages were taken for replicate samples. Of the 11,121 unique genes targeted on the array, 2,192 (20%) had multiple probes [61]. For those genes with multiple probes, we chose the most 3’ probe in the gene to estimate expression. Samples from two women were excluded prior to QC because there were too few probes detected after hybridization (13,189 and 15,863 probes, respectively, compared to a median count of 21,341 out of 47,231 probes on the array); and three women were excluded prior to analysis because there was no DNA available for one and low genotyping call rates in two (see below). The remaining 53 women (49 European ancestry, 2 Asian ancestry, 2 African American ancestry) had both high quality expression and genotype data. In those 53 samples, 10,191 genes were detected as expressed. Processing batch and array were significantly associated with the variance in gene expression based on principle component (PC) analysis and their effects were regressed out using a linear model (see S4 Fig). Other covariates that were considered (age, BMI, race, and season of biopsy collection) were not significant in these samples (see S2 Table). An overview of the sample inclusion pipeline is shown in S5A Fig.
DNA from women in the gene expression studies were genotyped with the Affymetrix Axiom Genome-Wide CEU 1 Array at the UCSF Genomics Core Facility. We performed QC checks using PLINK [62], and removed 4,922 SNPs with <95% genotype call rates, 503 with Hardy-Weinberg P-values ≤0.001, 336 non-autosomal SNPs, and 252,872 with minor allele frequencies <0.10. There were 370,008 SNPs remaining. After excluding two women with low call rates, the remaining 53 subjects had SNP call rates >97%.
Linear regression was used to test for associations between the expression levels of 10,191 genes and genotypes at the 378,362 SNPs with a minor allele frequency greater than 10%, using the R package Matrix eQTL [63]. Genotypes were recoded as 0, 1, or 2 to reflect an additive model. To maximize the power in our sample, we tested for associations only with SNPs within ~200kb of the transcription start site of a gene, a distance that would include nearly all cis regulatory SNPs [64–66]. The Matrix eQTL package assigns both a p-value and a false discovery rate (FDR) to each SNP-gene association. The FDR was calculated using the Benjamini and Hochberg [67] procedure (see [63] for detailed methods). eQTL mapping was additionally performed including only the 49 women of European ancestry. As expected, p-values were generally less significant in the smaller sample, however the HLA-F and TAP2 eQTLs remained significant at a FDR <1% (S5 Dataset).
We used the Database for Annotation, Visualization and Integrated Discovery (DAVID) v6.7 [17, 68] to interrogate pathway and gene enrichment for eQTLs at a FDR 1% compared to all gene-SNP combinations in our analysis (background). An enrichment score was calculated using Fisher's Exact test (modified as EASE score) on gene count for eQTLs at a FDR of 1% compared to all genes tested. We used the high classification stringency for our analysis.
We also used the Genomic Regions Enrichment of Annotations Tool (GREAT) to analyze the significance of SNPs which are eQTLs at a FDR of 1% [19]. GREAT first associates genomic regions with nearby genes and then applies the functional annotations for those genes to the regions. We used the basal plus extension definition of a gene regulatory domain, in which a gene’s defined regulatory domain expands until it reaches the nearest gene’s basal domain or maximally 5kb upstream and 1kb downstream. Using this definition, SNPs located between two genes may include both gene regulatory domains. GREAT uses a hypergeometric test over these defined genomic domains to assess enrichment between foreground (FDR 1% eQTLs) and background (all SNPs for which there is a gene-SNP pair tested in the eQTL analysis).
The data included in our analyses of fecundability were derived from a prospective study of pregnancy outcome in South Dakota Hutterites that was initiated in 1986, as previously described [6–8, 69]. The women in this study are provided with calendar diaries and EPT pregnancy test kits (Warner-Lambert Co.). They record in the diary dates of menses, changes in nursing patterns, illnesses or travel for the husband and wife, and dates of miscarriages or deliveries. In addition, they are instructed to test for pregnancy if they do not start menses exactly one month after the first day of their previous period, and to record the results of all pregnancy tests in the diaries. They are also asked to start testing for pregnancy on a monthly basis starting 6 months after delivery until menses resumes. Results of all pregnancy tests and outcomes of each pregnancy are recorded in the diaries, which are collected yearly either in person or through the mail. The results reported here include data collected through 2013.
Among the 325 Hutterite women who have participated in the prospective study, 156 women had at least one interval during which they were not nursing at time0. Of these 156 women, genotype data were available for 117. The following studies were performed in these 117 women, who provided information on 191 intervals. An overview of the sample inclusion pipeline is shown in S5B Fig.
The distribution of the interval lengths until a positive pregnancy test was compared between genotype groups using non-parametric life-table analyses [70]. The product-moment method was used to compute the time-to-pregnancy curves. These curves were compared with the Mantel-Haenszel log-rank. Women were stratified based on their genotype. Potential confounding effects of maternal age at the beginning of each interval, number of prior miscarriages, number of births (parity), maternal birth year, maternal inbreeding, and the kinship coefficient of the couple were assess by Cox regression analysis, as previously described [7]. Only maternal age and number of prior births (parity) were significant covariates and included in subsequent analyses. Because the Cox model assumes that the hazard ratios for continuous variables are log linear, we classified women (at each observation) by the number of prior births in three categories: 0–1, 2–3, or ≥4 prior births.
RFTS is a pregnancy cohort that enrolled study participants from the community between 2001 and 2012. Participants were recruited from Galveston, Texas; Memphis, Nashville, Knoxville, and Chattanooga, Tennessee; and the Research Triangle region (Raleigh, Durham, and Chapel Hill) in North Carolina. Detailed descriptions of direct marketing and recruitment strategies have been previously described [16].
Participants completed a baseline interview at enrollment and a computer assisted telephone interview at the end of the first trimester. All pregnancies were confirmed by pregnancy tests performed either by their provider (with confirmation), the study staff, or by the participant with BFP Early Pregnancy Test Strips (Fairhaven Health) provided by the study staff. The baseline and first trimester interviews provided information on reproductive history and potential confounders. Information about fecundability was collected on the baseline interview. Women were asked to report the number of cycles or months that it took them to conceive (if pregnant) or how long they have been trying to become pregnant (if not pregnant). Time to pregnancy was censored after 11 months; women in the study did not use any contraception during this time. We considered only one interval (pregnancy) per woman. DNA samples were obtained either in person or by mail during follow-up using Oragene saliva DNA kits (DNA Genotek Inc., Ontario, Canada).
These analyses included 314 RFTS participants who were 18 years or older, non-Hispanic white, and had self-reported time-to-pregnancy. In these women, we genotyped the two SNPs associated with fecundability in the Hutterites (rs2071473 and rs2523392) using TaqMan assays. We then excluded from subsequent analyses 22 women who did not know or declined to answer questions about their contraceptive practices, eight women who could not provide information on their cycle length, 22 women whose cycle lengths were <21 days or >35 days, one woman with cycle lengths >35 days and did not respond to questions about her contraceptive practices, and one woman with missing genotype data. These exclusions resulted in 260 women that were included in subsequent analyses. An overview of the sample inclusion pipeline is shown in S5C Fig.
The genotypic effects on time to pregnancy were estimated using a discrete time hazard model, a discrete time analog to the Cox proportional hazards model. We considered as covariates maternal age, education, marital status, income, smoking, alcohol use, caffeine consumption, body mass index, number of previous pregnancies, number of previous elective pregnancy terminations, and whether they were pregnant at the start of the study using a forward selection method. Number of previous elective terminations (range 0–2; median = 0) was inversely associated with interval lengths (P = 0.028) and being pregnant at the start of the study was associated with shorter interval lengths (P = 0.0016). These two covariates were included in the analysis of genotype effects. Although age and number of previous pregnancies (parity groups: 0–1, 2–3, ≥4) were not significant predictors of interval lengths in the RFTS study (maternal age, P = 0.91; parity groups compared to 0–1, P = 0.26 and 0.74, respectively), we included them in the model to be consistent with the analysis in the Hutterites.
Variants that were present in the Hutterites in each of the two associated regions but were not genotyped in the women with recurrent pregnancy loss who were included in the eQTL study were imputed in those women for a second stage eQTL mapping. We used whole genome sequences from 100 European American individuals as the reference genotypes for imputation [25]. Before imputation, variants with minor allele frequencies <1% or genotype call rates <95% were removed and variants on the reverse stand were flipped to the forward strand using PLINK [62]. To decrease computation time, we pre-phased the haplotypes in the reference genomes using Mach [71]. We then imputed genotypes in the women in the eQTL study using Minimac [72]. To avoid using SNPs with low imputation quality, we removed SNPs with an estimated R2 less than 0.5 before performing the eQTL mapping. We were able to impute or directly genotype 117 of the 7,062 variants in the two associated regions.
The eQTL mapping study in women with recurrent pregnancy loss was approved by the University of Chicago IRB (protocol number 14599B); all women gave written consent. The prospective study in the Hutterites was approved by the University of Chicago IRB (protocol number 5444); all participants gave written consent. The RFTS study was approved by Vanderbilt Human Research Protection Program (VHRPP) and Vanderbilt IRB (protocol numbers 070037 and 100396); all participants gave both verbal and written consent.
|
10.1371/journal.pgen.1005697 | Top2 and Sgs1-Top3 Act Redundantly to Ensure rDNA Replication Termination | Faithful DNA replication with correct termination is essential for genome stability and transmission of genetic information. Here we have investigated the potential roles of Topoisomerase II (Top2) and the RecQ helicase Sgs1 during late stages of replication. We find that cells lacking Top2 and Sgs1 (or Top3) display two different characteristics during late S/G2 phase, checkpoint activation and accumulation of asymmetric X-structures, which are both independent of homologous recombination. Our data demonstrate that checkpoint activation is caused by a DNA structure formed at the strongest rDNA replication fork barrier (RFB) during replication termination, and consistently, checkpoint activation is dependent on the RFB binding protein, Fob1. In contrast, asymmetric X-structures are formed independent of Fob1 at less strong rDNA replication fork barriers. However, both checkpoint activation and formation of asymmetric X-structures are sensitive to conditions, which facilitate fork merging and progression of replication forks through replication fork barriers. Our data are consistent with a redundant role of Top2 and Sgs1 together with Top3 (Sgs1-Top3) in replication fork merging at rDNA barriers. At RFB either Top2 or Sgs1-Top3 is essential to prevent formation of a checkpoint activating DNA structure during termination, but at less strong rDNA barriers absence of the enzymes merely delays replication fork merging, causing an accumulation of asymmetric termination structures, which are solved over time.
| Replication termination is the final step of the replication process, where the two replication forks converge and finally merge to form fully replicated sister chromatids. During this process topological strain in the form of DNA overwinding is generated between forks, and if not removed this strain will inhibit replication of the remaining DNA and thus faithful termination. In this study, we demonstrate that the cell has two redundant pathways to overcome topological problems during rDNA replication termination, one involving Top2 and the other involving the RecQ helicase Sgs1, in concert with Top3. In the absence of both pathways a checkpoint is activated in late S/G2 phase due to faulty replication termination at the strongest rDNA replication fork barrier (RFB). At less strong barriers termination is merely delayed under these conditions resulting in an accumulation of termination X-structures, which are solved over time.
| During DNA replication termination two replication forks coming from opposite directions merge to form two fully replicated sister chromatids. The process is essential for correct transmission of the genetic information to the next generation, but it has so far attracted little attention in eukaryotes. In E. coli, replication terminates in a region defined by two sets of Ter sites bound by the polar terminator protein, Tus. This protein stops replication forks from one direction, but allows free passage of forks from the opposite direction. The Tus-Ter sites are organized so that they form a trap for replication forks, thereby ensuring termination in a region opposite oriC in the circular E. coli genome [1]. Polar replication fork barriers with a function in replication termination have also been identified in yeast. In S. cerevisiae the rDNA locus holds the Replication Fork Barrier sequence (RFB). This barrier binds the Fob1 protein, which mediates polar fork stalling at RFB, resulting in replication termination in this region [2]. In S. pombe polar replication fork barriers are found both at the rDNA and mating type loci, where replication fork arrest occurs at the termination sites TER1-3 and RFP4 [3] and at the Replication Termination Sequence 1 (RTS1) [4], respectively. At yeast barriers members belonging to the Pif1 family helicases, Rrm3 and Pif1 in S. cerevisiae and Pfh1 in S. pombe, have been demonstrated to play profound roles for fork stalling and fork merging [5,6,7].
71 termination regions (TERs) have been identified in the S. cerevisiae genome outside the rDNA locus in one of the first large-scale studies performed on this subject in eukaryotes [8]. A common theme to the sequences at the identified TERs was that they contained fork pausing elements and that the Rrm3 protein assisted fork progression through these zones. Furthermore, DNA topoisomerase II located to the TERs during the S and G2/M phases and prevented DNA breaks and genome rearrangements, suggesting that topoisomerase II plays a role to ensure proper replication termination.
Over the years several studies have implicated topoisomerase II in the final steps of replication. Early studies of the circular SV40 genome and the yeast-borne 2μ plasmid reported incomplete replication with nascent strands containing smaller or larger gaps upon inhibition of topoisomerase II activity [9,10,11,12]. Based on these studies a model was presented, suggesting that positive supercoiling accumulates between converging forks, leading to a rotation of the replisomes and formation of precatenanes behind the forks. As a consequence, genuine catenanes form following termination in the absence of topoisomerase II activity, since precatenanes are exclusively substrates for this enzyme [13,14].
The STR complex, which in S. cerevisiae consists of the Sgs1 RecQ helicase, topoisomerase III (Top3) and the Rmi1 protein, has mainly been studied in relation to its role downstream of homologous recombination (HR) [15,16,17]. Studies have provided evidence that the complex is involved in dissolution of double Holliday Junctions (dHJ) in a non-crossover process [18]. In this process Sgs1 is thought to disrupt local annealing between parental and nascent strands, thereby forming hemicatenanes, which can be decatenated by Top3 [16,18]. However, based on early results demonstrating an interaction between Sgs1 and topoisomerase II (Top2), as well as a chromosomal missegregation phenotype of sgs1Δ cells, components of the STR complex have also been proposed to play a role during late stages of replication [19]. In support of this, Marians and co-workers demonstrated that RecQ and topoisomerase III from E. coli in collaboration with the single-stranded DNA-binding protein SSB performed the resolution of a synthetic termination substrate in vitro [20]. The parental duplex DNA region between two stalled replication forks was here unwound by the RecQ helicase, while topoisomerase III simultaneously decatenated the resulting catenated strands, leading to gapped, but untangled termination regions. In line with this, Hickson and co-workers reported that the BLM-TOP3-RMI1-RMI2 complex localized specifically to ultrafine DNA structures, the so-called anaphase bridges, during M phase in human cells. It has been speculated that these bridges represent late replication intermediates that are resolved by the BLM-TOP3-RMI1-RMI2 complex [21,22,23]. Similarly, Sgs1 and Top3 were found to localize to anaphase bridges in budding yeast [24]. Together, these data indicate that RecQ helicases in concert with topoisomerase III play a role in late stages of replication besides their well-established role in the resolution of recombination structures.
Our aim with the present study has been to investigate if S. cerevisiae Top2 and Sgs1-Top3 act redundantly in vivo to ensure faithful replication termination and resolution of replicating chromatids. Our findings demonstrate that they do, and their redundant function in this process is restricted to the rDNA locus.
Replicative stress and perturbations activate the S phase checkpoint pathway, which coordinates replication, repair, and cell cycling [25]. The Rad53 kinase is essential to this pathway and is phosphorylated, when the pathway is activated. If Top2 and the STR complex play redundant functions during final stages of replication, we expect that the absence of Top2 and Sgs1 will cause problems in late S/G2, which may activate Rad53. To test this hypothesis we monitored Rad53 phosphorylation in sgs1Δtop2ts cells using the In Situ Autophosphorylation (ISA) assay, which takes advantage of the autophosphorylation activity of Rad53, when it has been primed by upstream kinases [26]. As illustrated in the experimental setup presented in Fig 1A, yeast cells were grown at 25°C, synchronized in the G1 phase of the cell cycle with α-factor and released into the S phase, where samples were withdrawn at different time points and processed for checkpoint analyses. Release was at 34°C, the restrictive temperature for top2ts. In the top2ts mutant, checkpoint activation was seen 120 minutes after release from α-factor (Fig 1B) in accordance with the previously identified function of Top2 in chromosome segregation [27,28]. Consistent with this, no checkpoint activation was seen in top2ts cells treated with nocodazole, which prevents the cells from entering mitosis by inhibiting microtubule polymerization (Fig 1C). Interestingly, the sgs1Δtop2ts double mutant showed robust checkpoint activation already 60 minutes after release into S phase (Fig 1B). To analyze whether this checkpoint was connected to failure in chromosomal segregation due to lack of Top2, sgs1Δtop2ts cells were treated with nocodazole. However, the robust checkpoint activation persisted in the presence of nocodazole and was thus independent of chromosome segregation (Fig 1C). As revealed by FACS analyses the observed checkpoint occurred in late S/G2 after bulk DNA synthesis had taken place.
Most known functions of Sgs1 are mediated through the STR-complex, where Sgs1 acts in concert with Top3 and Rmi1. To investigate if this was the case here, we tested a top2tstop3ts strain for checkpoint activation. The top2tstop3ts cells showed robust checkpoint activation 60 minutes after release into S phase (Fig 1D), illustrating that Sgs1 and Top3 are equally important in cells lacking Top2. Taken together, the data demonstrate that Top2 and components of the STR-complex function in late S/G2 to avoid the accumulation of checkpoint activating structures.
Pulsed-field gel electrophoresis (PFGE) has earlier been used to demonstrate replication abnormalities in yeast, because DNA structures associated with incompletely replicated chromosomes prevent gel entrance [29,30]. We therefore applied this technique to investigate if the checkpoint activating structures observed in sgs1Δtop2ts cells would affect the fate of the individual chromosomes during replication. EtBr stainings of the PFGs revealed a decrease in the intensity of the individual chromosomal bands 40 minutes after α-factor release (Fig 2A, upper panels), which correlated with active replication for all strains as revealed from the FACS profiles. However, after 60 minutes the intensity of the bands had increased, demonstrating completion of replication and that none of the strains experienced any overall defect in replication. However, in the sgs1Δtop2ts cells, chromosome XII (chr. XII), which holds the rDNA locus, never re-entered the gel after replication, as confirmed by Southern blotting with an rDNA specific probe (Fig 2A, middle panels). In contrast, chr. II, which was used as a control for the other chromosomes, re-entered the gel (Fig 2A, lower panels). A quantification of the amount of chr. XII relative to chr. II re-entering the gel is shown in Fig 2B. Thus, although bulk DNA synthesis seems to occur without major problems in sgs1Δtop2ts cells, either Sgs1 or Top2 is required to complete replication of chr. XII.
A major part of budding yeast chr. XII is made up of the rDNA locus, consisting of 100–200 repeats of a 9.1 kb unit holding the 35S and 5S rRNA genes. Each repeat holds a replication origin (ARS) and a replication fork barrier (RFB) sequence (Fig 2C), where the latter forms a unidirectional replication fork block, when bound by the Fob1 protein. This block specifically stalls replication forks coming from the direction of the 5S rRNA gene and inhibits head-on collision between replication and transcription of the 35S rRNA gene [31]. Stalling of the leftward moving fork at RFB also ensures that replication terminates within this region. The rDNA locus differentiates chr. XII from the remaining chromosomes and could thus be an obvious cause to the problems observed with this chromosome in sgs1Δtop2ts cells. To address if this was the case and if the underlying cause of the checkpoint activation observed in sgs1Δtop2ts cells was connected to Fob1-mediated unidirectional replication, we investigated the checkpoint response in a sgs1Δtop2tsfob1Δ triple mutant. Interestingly, the checkpoint signal was reduced to background levels in the triple mutant (Fig 2D). Thus, checkpoint activation in sgs1Δtop2ts cells is Fob1-dependent and therefore takes place as a result of events occurring at the rDNA locus.
It is well established that excessive amounts of ssDNA coated with the ssDNA binding protein RPA is a signal for checkpoint activation through the Mec1 kinase [32,33]. To investigate if the Fob1-dependent checkpoint activation observed in sgs1Δtop2ts cells was related to the formation of DNA structures containing ssDNA, foci analysis were performed with cells having the large subunit of RPA (Rfa1) tagged with CFP and Nop1 (a marker for the nucleolus [34]) tagged with RFP (S1 Fig). The results demonstrated that sgs1Δtop2ts cells experienced significantly more RPA foci and thus more ssDNA 60–100 minutes after α-factor release relative to wt cells and single mutants, and most of the foci had a perinucleolar localization (S1A Fig Furthermore, formation of the ssDNA was Fob1-dependent (S1B Fig). Thus, the Fob1-dependent checkpoint activation observed in sgs1Δtop2ts cells is associated with a Fob1-dependent formation of ssDNA at the rDNA locus. The data indicate that the checkpoint activating DNA-structures observed in sgs1Δtop2ts cells contain regions of ssDNA.
To further analyze the nature of the checkpoint activating structures formed in the rDNA in sgs1Δtop2ts cells, we performed Neutral-Neutral two-dimensional (2D) gel electrophoresis with genomic DNA from sgs1Δtop2ts and control strains (Fig 3A). The BglII restriction sites used for generation of a DNA fragment with the RFB site centrally located (BglIIB) is shown in Fig 2C, and the migration of the replication structures obtained with this fragment in 2D gels is shown in Fig 3B. 40 minutes after release from α-factor, active rDNA replication occurred in all strains as demonstrated by formation of single and double Y-structures as well as structures generated due to replication fork blockage and convergence at RFB (Fig 3A). 80 minutes after α-factor release replication termination had occurred at most rDNA repeats in wt and single mutants as reflected by the disappearance of the majority of replication intermediates. However, a remarkable accumulation of DNA structures giving rise to a significant X-spike had taken place in sgs1Δtop2ts cells already 60 minutes after α-factor release, which coincided with the timing of checkpoint activation. Notably, the X-spike included the dot from symmetric X-structures representing forks converging at RFB as well as asymmetric X-structures extending the X-spike half way towards the 2N dot (indicated by the stippled area in Fig 3B). Quantification of the X-spike signal relative to the signal from Y-structures demonstrated an increase in the relative amounts of X-spike over time (Fig 3C), although the total amount of replication structures, including the X-structures, had decreased significantly after 100 minutes (Fig 3A). In contrast, sgs1Δtop2ts cells did not show an increase in RFB stalling relative to wt cells as revealed from a quantification of the RFB signal relative to the signal from all Y-structures (Fig 3D).
To investigate if X-structure formation was restricted to the area around RFB we analyzed the migration of replication structures obtained in the BglIIA fragment (see Fig 2C) covering most of the 35S transcription unit. Asymmetric X-structures were also formed in this fragment with the same timing and Xs/Ys ratio as in the BglIIB fragment (Fig 3E). In contrast, we did not see an accumulation of X-structures in sgs1Δtop2ts cells, when replication structures were analyzed in a fragment outside the rDNA, containing the TER102 site on chr. I [8] (S2 Fig). Thus, lack of Sgs1 and Top2 causes an accumulation of X-structures in late S/G2, which seems to be restricted to the rDNA locus.
Our data demonstrate that sgs1Δtop2ts cells show two strong characteristics, checkpoint activation and formation of X-structures, which are both connected to the rDNA locus. To investigate if the X-structures were the cause of checkpoint activation we took advantage of the Fob1-dependency of checkpoint activation and analyzed replication structures formed in the sgs1Δtop2tsfob1Δ triple mutant by 2D gel electrophoresis (Fig 4). Interestingly, the X-spike was still present in both BglII fragments. Thus, in contrast to the checkpoint activation (Fig 2D), all X-structures (except the symmetric X-structure formed due to forks converging at RFB) are formed independent of Fob1. Based on this we conclude that asymmetric X-structures are not responsible for checkpoint activation.
An investigation of chromosome migration in the sgs1Δtop2tsfob1Δ triple mutant by PFGE furthermore demonstrated that lack of Fob1 in the sgs1Δtop2ts strain was unable to suppress the migration defect of chr. XII (Fig 4), strongly indicating that the presence of X-structures in the rDNA is responsible for the inability of chr. XII to re-enter the gel after replication.
Taken together, our data are most consistent with a formation of two different DNA structures in sgs1Δtop2ts cells, a Fob1-dependent structure, which causes checkpoint activation, and a Fob1-independent structure causing the formation of the X-spike and the migration defect of chr. XII.
That checkpoint activation and X-spike formation are caused by different DNA structures was further supported by results obtained from experiments, where we investigated the sensitivity of the different structures to topoisomerase activity. In these experiments we reactivated Top2 in sgs1Δtop2ts cells (S3B and S3C Fig) and Top2 and Top3 in top2tstop3ts cells (S3D and S3E Fig) either before or after checkpoint activation and X-spike formation (25 or 60 minutes after release, respectively). We found that checkpoint activation and X-spike formation were inhibited upon early topoisomerase reactivation in both strains. In contrast, checkpoint activation was fully resistant to late reactivation in both strains, whereas X-spike formation was sensitive, showing a small reduction in the Xs/Ys ratio upon Top2 reactivation in sgs1Δtop2ts cells and a significant reduction upon reactivation of both topoisomerases in top2tstop3ts cells.
The rDNA locus has previously been demonstrated to be highly recombinogenic with recombination hot spots located close to the RFB [35]. Furthermore, increased recombination activity has been observed within this locus both in sgs1Δ cells and top2ts mutants kept at semi-permissive conditions [29,36]. Since HR is visualized as X-structures in 2D gels [37] and since one of the important functions of Sgs1-Top3 is to dissolve dHJs downstream of Rad52-mediated HR [38], we wanted to investigate the relationship between HR and checkpoint activation as well as formation of X-structures in the sgs1Δtop2ts cells. We therefore deleted RAD52 or RAD51 in the sgs1Δtop2ts strain to investigate if lack of HR would suppress the checkpoint phenotype of sgs1Δtop2ts cells (Fig 5A and 5B). Although single and double mutants with rad52Δ (Fig 5A and S4 Fig) or rad51Δ (Fig 5B) showed a slight increase in basal checkpoint activation as expected, robust checkpoint activation was observed in the two triple mutants. This demonstrates that the checkpoint activation observed in sgs1Δtop2ts cells occurs independently of HR. Thus, checkpoint activation does not arise due to a HR structure left unresolved in the absence of the Sgs1-Top3 pathway.
X-spike generating DNA structures have been observed in several studies both at the rDNA locus [39,40,41,42] and in other chromosomal regions [8,42]. To investigate if the X-spike observed in sgs1Δtop2ts cells represents HR structures we investigated replication structures generated in the sgs1Δtop2tsrad52Δ (Fig 5C) and sgs1Δtop2tsrad51Δ (Fig 5D) triple mutants by 2D gel electrophoresis. In both mutants the X-spike was still present, and it persisted 100 minutes after α-factor release with a relative proportion of Xs to Ys similar to the one obtained in sgs1Δtop2ts cells. The same was true for the X-spike formed in the BglIIA fragment covering most of the 35S transcription unit (S5 Fig). Thus, like checkpoint activation, X-spike formation in sgs1Δtop2ts cells is not caused by unresolved recombination structures. This was further supported by the migration of the structures in 2D gels. Due to branch migration of junctions within HR structures these are expected to form X-spikes that extend with equal intensity over the entire spike, when experiments are performed in the absence of crosslinking agents, which is the case here [8]. In contrast, the X-structures in sgs1Δtop2ts cells only gave rise to signals in the upper half of the spike and often with a punctuate nature of the spike (see e.g. Figs 4 and 5C). By the same token it is unlikely that the X-structures in sgs1Δtop2ts cells represent hemicatenanes, which also form X-spikes in 2D gels [42].
Besides HR structures and hemicatenanes, forks converging during replication termination form X-structures. The X-spike in sgs1Δtop2ts cells could therefore represent termination structures and be indicative of a failure during replication termination. If this is the case converging forks at rDNA replication fork barriers other than RFB should be responsible for the asymmetric X-structures constituting the major part of the X-spike (indicated by the stippled area in Fig 3B) and faulty termination at RFB should cause checkpoint activation. Barriers other than RFB have been demonstrated in the rDNA repeat both at ARS and the 5S transcription unit in the BglIIB fragment as well as in the 35S transcription unit in the BglIIA fragment [6]. Dots representing forks stalled at some of these positions were visible both in wt and sgs1Δtop2ts cells (indicated by arrowheads in Fig 3A–3E). If asymmetric X-structures represent forks converging at these barriers and checkpoint activation is a result of faulty termination at RFB we speculated that a general weakening of all barriers would inhibit the accumulation of X-structures and diminish or abolish checkpoint activation. Rrm3 facilitates replication past replication fork barriers and has also been suggested to be involved directly in fork merging [8]. However, Pif1 has been demonstrated to counteract Rrm3 [6]. Therefore, if the DNA structures formed in sgs1Δtop2ts cells would represent termination structures formed at different rDNA barriers we would expect that a deletion of PIF1 should reduce the formation of these structures either by directly facilitating fork merging at the barriers or by reducing fork stalling and thereby termination at these positions. To investigate this we deleted PIF1 and analyzed replication structures generated in the sgs1Δtop2tspif1Δ triple mutant by PFGE (S6A Fig) and 2D gel electrophoresis (Fig 6A and S6B Fig) to see the implications of a PIF1 deletion for chr. XII migration and formation of asymmetric X-structures, respectively. Interestingly, chr. XII re-entered the PFG after replication in the sgs1Δtop2tspif1Δ triple mutant and only wt levels of X-structures were observed in 2D gels for the triple mutant in both BglII fragments, consistent with X-structures representing termination structures. An alternative explanation for the Pif1 dependency of the asymmetric X-structures could be that sgs1Δtop2ts cells in the presence of Pif1 experience increased fork stalling at the different rDNA barriers, where the stalled forks are processed into X-structures in the mutant. However, we believe this is highly unlikely for several reasons. First, we do not observe increased stalling at RFB in sgs1Δtop2ts cells as expected if the cells in general show increased stalling (Fig 3D). Furthermore, processing of stalled forks into X-structures is expected to require HR, which is not involved (Fig 5). Finally, we observed an increased Xs/Ys ratio in sgs1Δtop2ts cells (Fig 3C), demonstrating an accumulation of X-structures rather than stalled forks, which indicates that processing of X-structures and not processing of Y-structures becomes the time limiting step in sgs1Δtop2ts cells. Based on this, it seems unlikely that forks stall more often in sgs1Δtop2ts cells than in wt cells. Rather the data suggest that when forks stalled at the different rDNA barriers are met by a fork coming from the opposite direction, fork merging becomes the time limiting step in sgs1Δtop2ts cells, thus resulting in an accumulation of termination X-structures.
To investigate, if a PIF1 deletion affected checkpoint activation at RFB as expected if checkpoint activation is a result of faulty termination at this position, we investigated whether or not checkpoint activation occurred in sgs1Δtop2tspif1Δ cells. As seen in Fig 6B checkpoint activation was fully abolished in the triple mutant. Thus, a deletion of either PIF1 or FOB1 inhibits checkpoint activation at RFB, whereas only a deletion of PIF1 eliminates X-spike formation, consistent with a role of Pif1 at all rDNA replication fork barriers.
Taken together, the data suggest that lack of Top2 and Sgs1 becomes detrimental during replication termination at the rDNA locus in sgs1Δtop2ts cells. Thus, replication termination at RFB causes checkpoint activation in these cells, which can only be abolished if either Fob1 is fully removed or the barrier is “loosened” as expected in the absence of Pif1. In contrast, termination at less strong rDNA barriers is merely delayed in sgs1Δtop2ts cells and therefore accompanied by an accumulation of asymmetric termination X-structures, which decrease in amount over time and are sensitive to Top2/Top3 reactivation.
Neutral-Alkaline (N-A) 2D gels have earlier been used to verify the presence of unsolved termination structures [7]. With this method, X-shaped molecules generated due to termination at RFB are separated from replication forks stalled at RFB in the first dimension due to differences in their molecular weight. After migration in the second dimension, where denaturation will separate DNA strands, both termination structures at RFB and forks stalled at RFB will consist of full length template strands as well as nascent strands of approximately half the size (Fig 7A). The template strands will thus form two dots located on the same horizontal line and the nascent strands will form two dots on a line below. Termination structures generated at positions other than RFB will form asymmetric X-structures. The template strands from these molecules will locate on the upper horizontal line, extending from the dot representing structures terminating at RFB towards the 2N dot. Besides full length parental strands, each asymmetric termination structure contains nascent strands of two sizes, which together make up the size of the parental strand. These strands will therefore in the second dimension form a “<” with legs emanating from the dot representing nascent strands from termination at RFB (Fig 7A, indicated by thick grey lines). Hemicatenanes and HR structures also form asymmetric Xs, but in contrast to termination structures both parental and nascent strands are full length and will locate on the upper horizontal line.
When DNA from sgs1Δtop2ts cells was analyzed with this method, spots representing nascent strands from replication forks blocked at RFB (black arrowhead) as well as nascent strands from X-shaped structures representing forks converging at RFB (open arrowhead) were revealed (Fig 7B, second row). The emergence of these spots 40 minutes after α-factor release correlated with the appearance of corresponding spots in wt cells (Fig 7B, first row) in agreement with the observations from Neutral-Neutral 2D gels (Fig 3). However, whereas the spots representing forks converging at RFB had disappeared to background levels in wt cells 80 minutes after release, they remained at a higher level in sgs1Δtop2ts cells (Fig 7C), although with a more diffuse appearance, strongly suggesting that termination at RFB was faulty. Furthermore, a smear extending as a “<” from the termination dot at RFB was present in wt and sgs1Δtop2ts cells 60 minutes after release, which remained in the mutant but disappeared in wt (Fig 7D). The presence of DNA fragments causing the “<”-smear shows that termination structures are formed at positions other than RFB in the rDNA and they remain for a prolonged time in sgs1Δtop2ts cells. In agreement with this, analysis of DNA isolated from the sgs1Δtop2tspif1Δ and sgs1Δtop2tsfob1Δ triple mutants with this method demonstrated that the “<”-smear was absent in sgs1Δtop2tspif1Δ cells as expected, but still present in sgs1Δtop2tsfob1Δ cells (except for the dots representing fork merging and stalling at RFB) (Fig 7B, third and fourth row and Fig 7D).
In this paper we demonstrate that cells lacking Top2 and Sgs1-Top3 show two strong characteristics, checkpoint activation and X-spike formation. Both are connected to the rDNA locus, appear during late stages of replication, and are independent of HR. Interestingly checkpoint activation is Fob1-dependent, whereas X-spike formation is not. In contrast, structures responsible for X-spike formation are sensitive to Top2/Top3 reactivation, whereas those responsible for checkpoint activation are not. Thus, two different structures are formed in sgs1Δtop2ts cells. This could either be due to a redundant function of Top2 and Sgs1-Top3 in two different processes or in a single process, having two different structural outcomes, when the enzymes are absent. Our results strongly suggest that it is the latter situation that is occurring, and that the process in which Top2 and Sgs1-Top3 are involved is replication termination. Thus, checkpoint activation occurs due to lack of Top2 and Sgs1-Top3 during replication termination at RFB, whereas X-spike formation is caused by a delay in replication termination at rDNA barriers other than RFB.
The results raise several questions. First, why do sgs1Δtop2ts cells have problems during replication termination and why do these cause checkpoint activation at RFB and only a delay in termination at other fork barriers? Furthermore, why do sgs1Δtop2ts cells show termination outside the normal RFB termination zone?
The finding that checkpoint activation occurs during termination at the strongest rDNA barrier whereas termination is merely delayed at other rDNA barriers when cells lack both Top2 and Sgs1-Top3, suggests that the problem experienced in the cells during termination is related to the nature of the barrier as well as to DNA topology. When a moving fork approaches a fork stalled at a barrier the topological tension between the forks increases and eventually leads to the formation of precatenanes behind the fork [13,14]. Several results have strongly suggested that the individual rDNA repeats are anchored to the nuclear membrane at RFB due to Fob1 interactions [43,44,45]. This fixation has been suggested to impose mobility constraints to the rDNA, and thus further increases the topological tension generated when forks converge at RFB. The structural consequence of the increased topological tension at RFB is speculative, but a checkpoint activating structure is finally formed, where either Top2 or Sgs1-Top3 can prevent formation of this structure as well as a deletion of either FOB1 or PIF1. In this topologically tense region we propose that Top2’s role is to continuously decatenate precatenanes formed in the termination zone behind the replication forks. Top2-mediated decatenation will directly influence replication fork progression. However, the Fob1- and Pif1-dependency of checkpoint activation suggests that Top2 activity furthermore facilitates Rrm3-mediated Fob1 removal/fork merging as has been suggested earlier [8]. Our data demonstrate that Sgs1-Top3 work redundantly with Top2 in this process. In support of this, Rrm3 has been demonstrated to be synthetic lethal with Sgs1 and Top3 [46,47]. An obvious role of Sgs1 and Top3 would be to unwind and decatenate, respectively, the DNA between the two converging forks. In support of this the E. coli homologs of Sgs1 and Top3 have been demonstrated to perform this reaction in vitro, when acting on a DNA substrate holding two closely located forks, thus mimicking a late replication structure [20].
Cozzarelli’s lab has earlier demonstrated that the excess topological tension generated between replication forks promotes the formation of chickenfoot structures [48]. An equilibrium may thus exist between formation of precatenanes and chickenfoot structures. Another function of Sgs1-Top3 could therefore be to constantly revert chickenfoot structures to ensure replication fork progression and facilitate Rrm3-mediated Fob1 removal/fork merging together with Top2. In sgs1Δtop2ts cells the equilibrium may be shifted towards the formation of chickenfoot structures. We observed that ssDNA was generated at the rDNA locus in a Fob1-dependent manner with the same timing as checkpoint activation. If chickenfoot structures are the cause of checkpoint activation they could be subject to DNA end resection, generating a ssDNA overhang, which recruits RPA and mediates checkpoint activation through Mec1. We would not be able to discern these resected strands in N-A 2D gels due to the smear produced by replication termination at rDNA barriers other than RFB. The two suggested roles for Sgs1/Top3 are not mutually exclusive.
At the less strong barriers we expect that the topological tension generated when forks converge in the absence of Top2 and Sgs1-Top3 is allowed to slowly dissipate to more remote areas. This may be possible either because no anchorage is present to inhibit dissipation at these barriers or because the barriers are of a more transient nature. If the topological tension is lower than at RFB, chickenfoot structures may not be generated to an extent, where the amount of ssDNA exceeds the threshold required to trigger checkpoint activation. Rather, replication fork merging is merely delayed, causing an accumulation of asymmetric X-structures, which await dispersal of topological tension for final termination to take place. In correlation with this, the X-structures decreased in amount over time and were sensitive to late reactivation of Top2/Top3. Furthermore, X-structures were not visible in pif1Δ cells, indicating that the topological tension is easier to deal with when the barrier is more “loose”. This observation furthermore supports that the function of Top2 and Sgs1-Top3 also at the less strong rDNA barriers is to facilitate Rrm3-mediated barrier removal/fork merging.
If asymmetric X-structures in sgs1Δtop2ts cells represent termination structures, this means that termination to a great extent occurs outside the general RFB termination zone and thus that some replication forks pass RFB and are trapped at other barriers. Barriers other than RFB have been demonstrated in the rDNA including the 35S and 5S transcription units and the ARS element [6], where the nature of these barriers is unclear. Fork arrest was observed at some of these positions in wt as well as in the single and double mutants in the present study (Fig 3, arrowheads). Based on the migration of the asymmetric termination X-structures in 2D gels it seems as if mainly the transcription units act as fork barriers in sgs1Δtop2ts cells besides RFB. At these positions the barrier effect may be caused by collision of the fork with the transcription apparatus or with topological tension generated in excess in these regions due to lack of Top2 [49]. Besides the demonstration that barriers other than RFB exist in the rDNA, it has also been demonstrated that not all leftward moving forks are arrested at RFB despite the general unidirectional replication mode at the rDNA locus [50,51]. Thus, relative to wt cells rrm3Δ cells show increased fork arrest both at RFB and at the other rDNA barriers, demonstrating that passage through RFB to some extent occurs in wt cells. RFB escape has also been demonstrated in pif1Δ cells, where the fraction of DNA in leftward moving forks was increased 2.5-fold relative to wt cells [6]. Very interestingly, when these observations are taken into account, the increased termination we see in sgs1Δtop2ts cells at barriers other than RFB indicate that termination in general occurs outside RFB in wt cells, but only in sgs1Δtop2ts cells is termination at these positions delayed, resulting in the accumulation of termination X-structures. In support of this, the observation that the relative level of X-structures to Y-structures was very high and remained high in sgs1Δtop2ts cells suggests that termination rather than fork stalling becomes the rate limiting step in these cells. Furthermore, sgs1Δtop2ts cells showed no sign of increased fork stalling.
Our data strongly suggests that Sgs1-Top3 becomes essential for replication termination when Top2 is absent, but that this redundant action of Top2 and Sgs1-Top3 is restricted to the rDNA locus. In sgs1Δtop2tsfob1Δ cells, where the anchoring to the nuclear membrane of Fob1 bound rDNA repeats as well as unidirectional replication are abolished the situation is expected to mimic the situation outside the rDNA locus. However, under these conditions asymmetric termination X-structures were still observed in the rDNA, whereas we saw no accumulation of X-structures in sgs1Δtop2ts cells, when analyzing replication structures at TER102 (S4 Fig) in correlation with earlier observations, where Top2, but not Top3, was found at termination sites outside the rDNA locus [8]. One explanation for this difference could be the high transcriptional activity at the rDNA locus. Transcription will increase the topological tension at the rDNA barriers which could create a need for Sgs1-Top3 besides Top2 and Rrm3 for efficient barrier dispersal/fork merging. Furthermore, it may well be that multiple copies are required to induce the robust cellular response observed in the rDNA. Thus, Top2 and Sgs1-Top3 may still play a role during termination outside the rDNA, but lack of Top2 and Sgs1-Top3 could cause less pronounced effects, which would be able to escape the detection limits of the assays employed in the present study.
The employed strains were constructed using standard genetic techniques and are listed in S2 Table. All strains are derivatives of the original W303-1a.
Unless otherwise stated, cells were grown to logarithmic phase in YPD media. Synchronization in G1 was achieved by transferring cells to YPD (pH 5.0) containing α-factor (2 μg/ml, Lipal Biochem) followed by incubation at 25°C for 150 min. Additional α-factor (1 μg/ml) was added after 1 hour of incubation to avoid escape from G1. To release the cells from arrest, they were washed once in water and transferred to fresh, pre-warmed (34°C) YPD medium. G2 arrest was achieved by adding nocodazole (Calbiochem) to a final concentration of 15 μg/ml.
Pulsed-field gel electrophoresis was performed as described in [52]. Cell cultures were grown to 3 x 107 cells/ml, and approximately 1.8 x 107 cells were cast into each plug to be run on the pulsed-field gel. The standard yeast genome size marker (Bio-Rad) was included on all gels. Gels were stained with ethidium bromide and transferred to Hybond XL membrane (GE Healthcare). Southern blotting was carried out using probes for chr. XII (Probe 1) and chr. II. Probes were amplified from purified yeast genomic DNA using 5’-CGCTTACCGAATTCTGCTTC and 5’-CTAGCATTCAAGGTCCCATT as forward and reverse primers, respectively, for chr. XII (Probe 1) and 5’-TCTCCGTCTTTAGTTGTTGC and 5’-GCCCTAGCAGTATTGCTTTG as forward and reverse primers, respectively, for chr. II. Experiments were performed 3–4 times with similar results.
Samples were taken for FACS analysis during the various experiments and processed as described in [53]. Samples were analyzed in a BD FACSCalibur.
All steps of the ISA were as described in [26], except that 5 μCi/ml [γ-32P] ATP was used. In short, protein extracts were generated from TCA-treated cells. For every sample, protein concentration was determined by Coomassie blue to allow loading of equal amounts of proteins on 10% SDS-polyacrylamide gels along with 5μl of a standard containing a known amount of MMS activated Rad53 (“+ control”). After gel electrophoresis proteins were transferred to PVDF filters (Immobilon-P, Millipore membranes). Filters were subjected to a denaturation/renaturation protocol before the autophosphorylation reaction was performed by incubating membranes in kinase buffer in the presence of [γ-32P] ATP. Dried filters were exposed on a Typhoon Trio+. After exposure, filters were re-probed with goat anti-Mcm2 (Santa Cruz) to check loading and allow comparison among different gels and mutants. Experiments were performed 2–3 times with similar results. MMS control (“+ control”): An Ay-120 culture (wt) with a density of 0.4 x 107 cells/ml was treated with 0.1% MMS for ~60 minutes and harvested.
Yeast genomic DNA was isolated from 1 x 109 cells using Genomic-tip 20/G (QIAGEN) as described in [54]. After digestion with the BglII restriction enzyme (New England Biolabs) half of the purified DNA was subjected to Neutral-Neutral two-dimensional gel analysis as described in [55]. Southern blotting was carried out with the probes shown in Fig 2C, which were generated by PCR using genomic DNA as template. Probe 1 was generated as described above. For probe 2, recognizing the BglIIA fragment 5’-GTTTCTTTTCCTCCGCTT-3’ and 5’-ATCTCTTGGTTCTCGCAT-3’ were used as forward and reverse primers, respectively. For the probe near TER102 on chr. I 5’-GAAGGTTCAACATCAATTGATTGATTCTGCCGCCATGATC-3’ and 5’- GCTTCCCTAGAACCTTCTTATGTTTTACATGCGCTGGGTA-3’ were used as forward and reverse primers, respectively.
For Neutral-Alkaline two-dimensional gel electrophoresis BglII digested DNA was run on a Neutral gel in the first dimension. In the second dimension the excised DNA was run on a 1.5% agarose gel in 50 mM NaOH plus 1 mM EDTA at 4°C [7]. Probe 3 (Fig 2C) used for southern blotting was made by PCR with 5’-CAGCCATAAGACCCCATC-3’ and 5’-GCAGTTGGACGTGGGTTA-3’ as forward and reverse primers, respectively, and genomic DNA as template.
The intensity of DNA structures was measured using QuantityOne software. For Neutral-Neutral 2D gels the relationship between either X-structures and Y-structures or the RFB dot and Y-structures was calculated for each time point. Unless otherwise stated, the ratio at the different time points was related to the ratio at the 0 minute time point to allow comparison between strains. For the Neutral-Alkaline 2D gels the signal of the RFB dot or the “<”-smear at the different time points was related to the signal at the 0 minute time point. For the PFG’s the ratio between chr. XII and chr. II re-entering the gel was calculated for each time point.
All strains harbored the pWJ1321 plasmid encoding Nop1-RFP and were therefore grown in synthetic complete media without histidine [56]. Cells were synchronized in G1 by treatment with α-factor for 150 minutes and released into SC-his medium supplemented with 100 μg/ml adenine at 37°C. Cell samples were collected, centrifuged at 2,000g and prepared for fluorescence microscopy as described in [57]. Fluorophores were visualized using band-pass CFP (31044) and RFP (41002c) filter sets from Chroma. Fluorescence images were acquired and processed using Volocity software (PerkinElmer). Statistical probabilities were calculated using Fisher’s exact test (two-tailed). See S1 Table for morphology of cells included in the study.
|
10.1371/journal.pcbi.1006403 | Seizure pathways: A model-based investigation | We present the results of a model inversion algorithm for electrocorticography (ECoG) data recorded during epileptic seizures. The states and parameters of neural mass models were tracked during a total of over 3000 seizures from twelve patients with focal epilepsy. These models provide an estimate of the effective connectivity within intracortical circuits over the time course of seizures. Observing the dynamics of effective connectivity provides insight into mechanisms of seizures. Estimation of patients seizure dynamics revealed: 1) a highly stereotyped pattern of evolution for each patient, 2) distinct sub-groups of onset mechanisms amongst patients, and 3) different offset mechanisms for long and short seizures. Stereotypical dynamics suggest that, once initiated, seizures follow a deterministic path through the parameter space of a neural model. Furthermore, distinct sub-populations of patients were identified based on characteristic motifs in the dynamics at seizure onset. There were also distinct patterns between long and short duration seizures that were related to seizure offset. Understanding how these different patterns of seizure evolution arise may provide new insights into brain function and guide treatment for epilepsy, since specific therapies may have preferential effects on the various parameters that could potentially be individualized. Methods that unite computational models with data provide a powerful means to generate testable hypotheses for further experimental research. This work provides a demonstration that the hidden connectivity parameters of a neural mass model can be dynamically inferred from data. Our results underscore the power of theoretical models to inform epilepsy management. It is our hope that this work guides further efforts to apply computational models to clinical data.
| A fundamental question in clinical neuroscience is how and why the brain generates epileptic seizures. To address this problem it is important to unify theoretical models of seizure mechanisms with clinical data. This study investigated a large database of human epileptic seizure recordings. Model inversion was used to track seizure dynamics through the lens of a mathematical model for cortical regions. These models can reveal the relative activity and coupling between excitatory, inhibitory and pyramidal neural populations that cannot be directly measured. Measuring cortical dynamics during seizures can provide insight into epilepsy, and facilitate new treatment strategies. Our analysis of connection strengths revealed important aspects of seizure onset and seizure termination. Our findings have implications for understanding seizure mechanisms and treating epilepsy.
| Understanding how and why the brain generates spontaneous seizures is an unsolved problem in neuroscience. The medical implications of seizures are profound, with over 50 million people affected by epilepsy, and at least 30% not adequately controlled by available therapies [1]. Surgical treatment does not provide complete seizure freedom for all patients [2], and novel drugs have not greatly improved on the level of seizure freedom that can be achieved [3]. On the other hand, data-driven, computational techniques have shown early promise in obtaining a more individualized picture of a patient’s seizures, which may shed new light on mechanisms of seizures and lead to targeted treatment strategies [4, 5].
Patient-specific, computational models can provide unique insight into seizure mechanisms, and are well accepted in the study of epilepsy [6]. In particular, lumped parameter neural mass models [7, 8] have been extensively used to investigate cortical activity during epileptic seizures [9–11]. These models describe seizures as state transitions in the brain [12] that arise from endogenous noise perturbations or ‘pathways through the parameter space’ of a neural model [13]. Clinically, it is recognized that electrographic (EEG) recordings of seizures show stereotypical changes in the signal morphology that are regarded as state changes (i.e. between interictal, peri-ictal, and ictal states) [14]. Despite the ubiquity of neural mass models to study seizure transitions, the translation of these theoretical insights into clinical practice has not been widely realized.
The validation of neural network models to aid clinical decision making has made some advances in diagnosis [15], and surgical planning [16–19]. Ideally model-based techniques can also improve outcomes at earlier stage interventions, such as drug selection. Another area models can aid treatment may be in seizure forecasting [20, 21], or the design of electrical counter-stimulation (using model predictive control) [22–24]. A fundamental hurdle to overcome is validating theoretical models of brain dynamics in a clinical setting. This hurdle largely exists due to the difficulty of obtaining in vivo neural recordings from humans. Whilst simulation has proven valuable to generate new hypotheses regarding the mechanisms of seizures, a complete validation must unite empirical data with theoretical models and demonstrate that models have predictive value, as well as being descriptive of the data [4, 5].
Model inversion is a powerful approach to combine patient-specific recordings with accepted principles of brain structure and function encapsulated by the parameters of a neural model [25]. Previous work has outlined a generalizable framework to estimate the most likely states and parameters of a neural model given observed data [25]. For many years this problem was intractable for non-linear neural models [26]. Previously, model inversion has relied upon simplifying assumptions, such as linearization, or sampling techniques [27]. Another approach is to re-frame the problem, so that the objective is to find the most likely model to generate the observed data. In this way, the estimation is conditioned on the model space, which is generally explored via some heuristic model selection criteria [28, 29]. Alternatively, the inversion can be conditioned on the data, where the most likely model is identified using an assumed density (Kalman) filter [30]. This approach has been validated for investigation of seizure dynamics [31–33]. Recent advances have also incorporated a fast, semi-analytic solution to handle the propagation of estimates through the non-linear neural mass equations [25, 34, 35]. Model inversion techniques that enable time-varying estimates of key parameters provide a powerful means of inferring cortical mechanisms from functional neuroimaging data. This is particularly true for EEG/ECoG data, which has high temporal resolution. The ability to update estimates with each new data point can lead to insights into ictal dynamics that evolve over fast time scales.
Statistical observations from data are also important to validate models of seizure transitions. Some studies have investigated the distributions of times spent in different seizure states [36]. Models that are predictive of higher-order statistics derived from many seizures are more convincing than models which are only descriptive of (or fitted to) individual seizures [4]. An important observation is that the distribution of times spent in the ictal state has a patient-specific peak [37], rather than a uniform distribution. These peaks indicate that patients have a characteristic seizure duration, or trajectory length. Intriguingly, a subset of patients showed a distinctly bimodal distribution of seizure durations, indicating two populations of seizures (long and short) [37]. We hypothesize that these distributions reveal a crucial aspect of seizure dynamics, which should not be neglected in computational modeling.
This work presents a large-scale, model-based investigation to address the question of how multiple (long and short) seizure trajectories arise in the brain. Model inversion was performed for the largest database of human seizures recorded in individual patients [38]. Using this database, we have previously demonstrated that there are two populations of seizure duration [37]. The current study investigated different seizure pathways and mechanisms through the lens of a neural mass model (using the formulation of Jansen and Rit, 1995 [8]). The following sections outline the data, model and estimation techniques. Further detail is provided in S1 Appendix, and code is available online (https://github.com/pkaroly/Data-Driven-Estimation).
Seizure mechanisms were investigated for continuously recorded ECoG from 12 patients with focal epilepsy monitored during a previous clinical trial [38]. All subjects were implanted with intracranial electrode arrays with a total of 16 platinum iridium contacts around the seizure onset zone. The ECoG was sampled at 400 Hz and wirelessly relayed to an external, portable personal advisory device. Seizure detection was automated and reviewed by expert clinicians. This study used data from 3010 clinical seizures (average 250 per patient). Seizures were either associated with confirmed clinical symptoms or were electrographically similar to clinical seizures. Other epileptiform discharges without clinical symptoms were excluded. All seizures had onset and offset labelled by expert epileptologists. For further details on the data collection procedures the reader is referred to Cook et al. (2013) [38].
A similar procedure to that outlined by Cook et al. (2016) [37] was used to identify patients with bimodal seizure durations. Both k-means clustering and Gaussian mixture model fitting were used to test for bimodality. Clusters were assigned for one, two and three seizure populations (based on the logarithm of seizure duration in seconds). The optimal number of clusters was determined using gap criteria [39].
The current study used patients who had at least 20 seizures that had a lead time of one hour. Recordings were used from five minutes before seizure onset, until one minute after seizure offset. Seizures with telemetry dropouts were excluded from analysis. Data were bandpass filtered (second-order, zero-phase Butterworth filter) from 1 Hz to 180 Hz with a notch filter at 50 Hz (second-order, zero-phase Butterworth filter). The energy of the signal was computed for a 1s sliding window (50% overlap) as energy = ∑ n = 1 N x 2.
The states (mean membrane potentials) and parameters (synaptic connectivity strength) of neural mass models were fitted to data recorded during epileptic seizures. The formulation of the neural mass model in the following section is derived from the model introduced by Jansen and Rit (1995) [8]), and has also been outlined in our previous work [33, 34]. The neural mass model is suitable to model ECoG measured at this scale (electrodes approximately 5mm in diameter with spacing on the order of centimeters), in line with similar neural models used to describe EEG/MEG activity [10, 11, 40]. A single, independent neural model was fitted to each ECoG channel (16 models in total). Neural models were not coupled between channels; hence, estimates primarily captured local connection strengths within a single cortical region. The input parameter, u described non-local inputs to the pyramidal population.
The Jansen and Rit model consists of three neural populations (excitatory, inhibitory and pyramidal). Neural populations were described by their time varying mean membrane potential, vn, which is the sum of contributing mean post-synaptic potentials, vmn (post-synaptic and pre-synaptic neural populations are indexed by n and m, respectively). For the current model, the index n (post-synaptic) represents either pyramidal (p), excitatory (e) or inhibitory (i) populations, as shown in Fig 1.
The post-synaptic potential, vmn arises from the convolution of the input firing rate, ϕ(vn), with the post-synaptic response kernel,
v m n ( t ) = α m n ∫ - ∞ t h m n ( t - t ′ ) ϕ ( v n ( t ′ ) ) d t ′ , (1)
where αmn, which are the estimation parameters, represent lumped connectivities that incorporate average synaptic gain, number of connections, and average maximum firing rate of the presynaptic populations. ϕ(vn) is the sigmoid function
ϕ ( v ) = 1 2 ( erf ( v - v 0 ς ) + 1 ) (2)
where v0 = 6mV, and ς = 0.0030 (as defined by Freestone et. al. (2014) [25]).
The convolution in Eq 1 can be written as two coupled, first-order, ordinary differential equations,
d v m n d t = z m n d z m n d t = α m n τ m n ϕ m n - 2 τ m n z m n - 1 τ m n 2 v m n . (3)
where τmn is a lumped time constant. The values of τep, τpe, and τpi were fixed to 10ms and the value of τip to 20ms, as defined by Jansen & Rit (1995) [8].
External (non-local) inputs to the pyramidal population are modeled as an additive term affecting the pyramidal membrane potential,
v p ( t ) = v p e ( t ) - v p i ( t ) + u ( t ) . (4)
The recorded ECoG for each channel, i, is derived from the average pyramidal membrane potential of each independent neural mass model (resulting 16 disconnected models in the estimation),
y i ( t ) = v p i ( t ) (5)
The neural model can be expressed in matrix notation
x ˙ ( t ) = A x ( t ) + B ϕ → ( C x ( t ) ) , (6)
where x ∈ R N x is a state vector representing the postsynaptic membrane potentials generated by each population synapse and their time derivatives. There are two states per synapse and Nx = 2Ns is the total number of states, where, for Ns synaptic connections in the models, the state vector is of the form,
x = [ v 1 z 1 … v N s z N s ] ⊤ .
The definitions of A, B, and C are provided in S1 Appendix.
The observation equation is of the form
y ( t ) = H x ( t ) + v ( t ) , (7)
where H ∈ R N x × N y is the observation matrix, v ( t ) ∼ N ( 0 , R ) ∈ R N y is the observation noise, and Ny is the number of observations (here Ny = 1 as each neural mass model describes a single ECoG channel). As our measurement function is linear, H is simply an index vector of zeros and ones that defines the average pyramidal membrane potential given by Eq 4.
A joint state (membrane potentials) and parameter (external input and connectivity strengths) estimation algorithm was implemented for every sample of the recorded ECoG. To obtain estimates it was necessary to augment the state-space representation of the neural model. To define the augmented model, we first define a vector of parameters as
θ = [ u α p e α p i α i p α e p ] ⊤ .
The dynamics for the parameter are modeled as a random walk
θ ˙ = 0 . (8)
The state vector x and the parameter vector θ are concatenated to form the augmented state vector,
ξ = [ x T θ T ] ⊤ . (9)
Our augmented state-space model is
ξ t = A θ ξ t - 1 + B θ ϕ ( C θ ξ t - 1 ) + w t - 1 , (10)
where w t ∼ N ( 0 , Q ). The state vector ξ ∈ R N ξ × 1 and matrices Aθ, Bθ, and Cθ are ∈ R N ξ × N ξ and have the form
A θ = [ A 0 0 I ] , B θ = [ B 0 0 0 ] , C θ = [ C 0 0 0 ] . (11)
For simplicity we will drop the subscript θ on the system matrices, as the remainder of the equations refer to the augmented model.
The estimation scheme uses an assumed density filter. This filter provides the minimum mean squared error estimates for the states and parameters, under the assumption that the underlying probability distribution is Gaussian (the assumed density). Formally stated, the aim of estimation is to compute the most likely posterior distribution conditioned on previous measurements,
ξ ^ t + =E [ ξ t | y 1 , y 2 , … , y t ] (12)
P ^ t + =E [ ( ξ t - ξ ^ t + ) ( ξ t - ξ ^ t + ) ⊤ ] , (13)
The estimator proceeds in two stages; prediction and update. In prediction, the prior distribution (obtained from the previous estimate) is propagated though the neural mass equations. This step provides the so called a priori estimate, which is a Gaussian distribution with mean and covariance,
ξ ^ t - = E [ ξ t - 1 | y 1 , y 2 , … , y t - 1 ] (14) P ^ t - = E [ ( ξ t - 1 - ξ ^ t - 1 + ) ( ξ t - 1 - ξ ^ t - 1 + ) ⊤ ] . (15)
In the second stage, a Bayesian update is performed to shift the estimated posterior based on the observed data, giving the a posteriori distribution,
ξ ^ t + = ξ ^ t − + K t ( y t − H ξ ^ t − ) ︸ ECoG prediction error . P ^ t + = ( I − K t H ) P ^ t − , (16)
where K is the Kalman gain (readers are referred to [27] for a detailed description of the Kalman filter). After each time step, the a posteriori estimate becomes the prior distribution for the next time step, and the filter proceeds.
In general, the Kalman filter equations do not have a solution for nonlinear model or measurement functions. Previous efforts to use Kalman filtering on the nonlinear neural mass model have relied on simplifying assumptions (either linearization of the model, or sampling to estimate the posterior distribution). This work applied an exact, semi-analytic solution for the mean and covariance of a multivariate Gaussian distribution transformed by the nonlinear neural mass model. This solution provides the a priori estimate of the mean and covariance (see S1 Appendix for details).
As the observation function is linear, the updated (a posteriori) mean and covariance are obtained trivially using Eq 16.
The Kalman filter requires ξ ^ 0 + and P ^ 0 + to be initialized to provide the a posteriori state estimate and state estimate covariance for time t = 0. The other parameters that must be initialized are the model and measurement noise, Q and R, respectively. Further details of filter initialization are given in the S1 Appendix.
Fig 1A shows an overview of the estimation scheme. The following sections present data from twelve patients with focal epilepsy. The data consist of over 3000 clinical seizures (average 250 seizures per patient). Of these twelve patients, three showed bimodal distributions of seizure durations (Patient 3, 8 and 11). Note that all patients showed that either one or two clusters were optimal for seizure durations, and for all patients, k-means and Gaussian mixture modelling aligned on the same optimal number of clusters.
An assumed density filter was used to track the time-varying states and parameters of neural mass models during every seizure (as seen in Fig 1). This estimation technique finds the most likely model given the observed ECoG data. Importantly, the model is updated at every time step, so there is no loss of temporal resolution. Estimated states were mean membrane potentials, and parameters (alpha parameters in Fig 1B and 1C), which were the external input and average synaptic strengths between pyramidal, non-pyramidal excitatory, and inhibitory neural populations. In this way, the neural models provided an estimate of the average activity and effective connectivity within intracortical circuits [25]. We found that slow changes in the synaptic connectivity parameters led to seizure transitions in the neural models. As seen in Fig 1C, a deterministic forward simulation of neural models using time-varying connectivity estimates reproduced the beginning and end of seizures.
It is important to also quantify estimation accuracy before proceeding. A full summary of the model and estimation errors is given in the Supplementary Material. However, it is worth noting briefly that errors between the estimated signal and real data were small (see S1 Fig). The mean squared error ranged from 0.2-0.9 mV when averaged across all seizures (note that the mean amplitude of the measured ECoG signal ranges from approximately 25-100mV). The uncertainty (covariance) of state variables and parameters was also small (see S2 and S3 Figs), suggesting that key seizure activity was well described by the model, rather than by the residuals. Across patients, the mean covariance ranged from 2-16% for state variables, and from 0.1-10% for connectivity parameters (expressed as a percentage of the estimated value). Numerical instability of the filter was occasionally observed (for all patients, instability occurred for less than 1% of the data). Estimates that became unstable were removed from further analysis.
Fig 2 shows the average energy of recorded ECoG during every seizure (averaged across 16 electrode channels). Patients’ seizures showed strikingly consistent patterns of signal energy between seizures. These patterns were generally time locked to seizure onset, as demonstrated by the vertical alignment of energy changes (note that Patients 2, and 4 did not show a vertically aligned onset pattern). Long and short seizures began similarly, but evolved differently (see Patients 3, 8, and 11). Long seizures entered a secondary phase where energy increased. Short seizures and the early phases of long seizures were characterized by an energy reduction (note the darker vertical band following seizure onset). Some patients’ seizures only showed the “long” stereotypical pattern with a high-energy phase (see Patients 2, 4, 6, 9, and 15). Patient 13 had only low energy, stereotypical “short” seizures. Patient 7 had a large majority of short seizures, with a small number evolving to have increased energy.
These two patterns of seizure energy suggest that long and short seizures reflect distinct event types, each with a characteristic electrographic evolution. We hypothesized that these stereotypical signal patterns represent two alternative seizure trajectories, which could be differentiated by their onset and/or offset mechanisms. Note that although the average energy (averaged across electrode channels) was presented, the observed patterns were consistent across different channels. Full plots are provided in the Supplementary Material (S8 to S19 Figs).
Fig 3 presents the dynamic estimation results for the five connectivity parameters of a neural model during every seizure. Seizures followed a remarkably consistent trajectory through the parameter space of the neural mass models, showing similar patterns across all events for an individual. This indicates that seizure transitions follow a stereotypical pathway. Note that transitions during the seizure are locked to the onset time (as demonstrated by the vertical banding in parameter changes). A higher contrast (normalized) version of these patterns is provided in S4 Fig, to more clearly expose connectivity patterns.
For all patients, the strongest ictal changes in connectivity strength occurred for in-going connections to the pyramidal neurons (the first three columns of Fig 3). Conversely, outgoing pyramidal connections (to inhibitory and excitatory neurons) were more stable over the durations of the seizures, demonstrated by values which were closer to zero (reflecting no change from baseline), and less vertical patterning (reflecting no stereotypical transitions during seizures). Patient 6 was one possible exception, showing some decrease in outgoing pyramidal connections (6D and 6E).
Note that although neural models were fitted independently to all 16 electrode channels, Fig 3 shows results for a single example channel per patient. The data associated with every channel generates 60 full page figures, which are provided in an online repository (https://github.com/pkaroly/Data-Driven-Estimation). The consistency of stereotypical patterns between channels is investigated in the following sections.
Fig 4 shows the mean change in connectivity strength during seizures. A consistent motif was a decrease followed by an increase in ingoing connections to the pyramidal population (see columns A and C for Patients 1, 3, 6, 8, 9, and 15). Patient 11 showed the same motif, but with connectivity strength always above baseline. Patients 7, 10, and 13 showed only decreases in ingoing pyramidal connections. Patients 2 and 4 demonstrated only strengthened connections.
There were three classes of ictal parameter transitions: decrease, increase, and decrease-then-increase, where connections into the pyramidal populations were on average weaker, stronger, or weakened then strengthened (compared to a pre-ictal baseline), respectively. These classes aligned well with the stereotypical seizure evolution patterns that were identified based on signal energy (Fig 2). For instance, “decrease-then-increase” patients (1, 3, 6, 8, 9, and 15) showed long seizures that began with lower energy and evolved into a higher energy state. The “increase” patients (2 and 4) showed primarily high energy seizures without an obvious alignment to seizure onset. The “decrease” patients (7, 10, and 13) showed short, low-energy seizures.
Outgoing connections from pyramidal cells to excitatory/inhibitory populations showed little to no change. For some patients (2, 4, 9 and 10), a slight increase in pyramidal to inhibitory strength was observed.
Fig 5 shows the average seizure trajectory for all channels. Trajectories were qualitatively similar across channels. Most subjects showed focal patterns in which a subset of channels demonstrated connectivity changes during seizures while other channels did not have significantly increased or decreased connectivity during seizures (compared to a pre-ictal baseline period). Such patterns were not surprising, given that all subjects had focal seizures, which typically appear first on a subset of EEG channels before spreading. Apart from focal connectivity changes, there was some inter-channel variability at the ends of seizures. Subject 6 showed some channels with increased connection strengths and others with decreased strengths. Subject 9 showed increased inhibitory connections across most channels but decreased inhibition on a subset of channels (Fig 5, subpanel 9B), that occurred toward the ends of seizures. Overall, significant changes in connectivity (above or below baseline) followed the same stereotypical, patient-specific pattern across all channels. Exceptions to this consistency were observed for a few subjects in the later stage of seizures (significant changes are shown in Supplementary Material S6 Fig). In other words, there were no channels with markedly different trajectories; changes in connectivity were either in the same direction or showed no significant change from baseline. This consistency supports the finding of characteristic pathways of epileptic seizures, although these pathways were only observed on a subset of (possibly focal) channels, while other channels did not show altered connectivity patterns during seizures.
Overall, there was no difference in the average connection strength trajectories for long compared with short seizures (when connections were averaged across seizures in the long and short populations, respectively; see S5 Fig for details). Therefore, we hypothesized that short and long seizures were primarily differentiated by termination (i.e. both types follow a similar path from onset, with short seizures terminating earlier). This hypothesis was tested by measuring the correlation between connection strength (now averaged across 16 electrode channels) and seizure duration before onset and offset (correlation results were qualitatively similar when evaluated for individual electrode channels, and are provided in S7 Fig).
Fig 6 shows that almost no patients showed significant correlation between seizure duration and onset dynamics. In other words, there was no relationship between average connection strength and seizure duration at the outset (measured over a 5s window prior to seizure onset). However, at 5s before seizure offset, there were strong correlations for all connections. In general, longer seizures were associated with increased excitatory inputs and decreased inhibition to the pyramidal cells. Bimodal patients (3, 8, 10, and 11) all showed a similar relationship between connectivity strength and seizure duration. Four other patients also showed significant correlations; therefore, correlations do not arise purely because of the two duration populations.
Analysis of seizure energy (Fig 2) suggests there are two broad categories of focal seizures: short (low energy) and long (high energy) seizures. Estimation of patients’ seizure trajectories through the parameter space of a neural model revealed characteristic mechanisms underlying these different energy states (Fig 3). During the low energy phase of seizures, estimation showed decreased connectivity strength of ingoing connections to pyramidal cells. Seizures with high energy showed increased connection strength. This pattern was maintained as seizures evolved through time, with several patients showing a motif of decreased then increased connection strengths, corresponding to low energy seizures that evolved into a high-energy state.
Based on their characteristic seizure durations three patient subtypes were defined; those with exclusively short, exclusively long, or bimodal populations of seizure types. Understanding how these different patterns of seizure evolution arise may provide new insights into brain function, and guide treatment for epilepsy, as specific therapies may have preferential effects on the various parameters that could potentially be individualized. This study showed that long and short seizures reflect different underlying mechanisms in a neural model. Mechanistic differences arose almost exclusively before seizure offset, and were not evident prior to onset (Fig 6). Therefore, we conclude that seizures follow the same trajectory until termination. Apart from a bimodal distinction, connectivity patterns were strikingly similar during the evolution of each patient’s seizures; although highly patient-specific. This suggests that, once initiated, seizures follow an individualized and deterministic path through the parameter space of a neural model. It is remarkable to see these parametric pathways maintained across hundreds of seizures (see Fig 3), and over many years recording duration.
Patients were classified into three groups of connectivity patterns during seizures (seen in Fig 4): increased, decreased, and decreased-then-increased strength of ingoing connections to pyramidal cells. These parameter shifts may relate to distinct mechanisms of seizure onset. For the decrease, and decrease-then-increase patterns we speculate that seizures arise from either under-regulation or disinhibition of pyramidal neurons. The corresponding rebound of connection strength (in the decrease-then-increase group) may be linked to a regulatory mechanism that was not triggered for patients with shorter seizures (in the decrease group). Previous work using the neural mass model confirms that inhibitory populations are likely to play a role in generating epileptiform activity, with the time scale of inhibitory dynamics also highly relevant [10]. There were also two patients who showed only increased connection strength to pyramidal cells (in Fig 4, Patient 2 showed all connections were increased and Patient 4 showed an increase of excitatory inputs). For these patients, seizures may have been driven by over-excitation of pyramidal neurons.
There is a lack of consensus as to whether noisy fluctuations (multi-stability) or deterministic parameter changes (bifurcations) drive seizure onset/offset [4]. Other mechanisms, such as intermittency, may also be involved in seizure transitions [41, 42]. This study demonstrated that the transitions of connectivity parameters were locked to the onset of seizures, and not the offset (i.e. the patterns in Figs 3 and 2 arise when the seizures are aligned by start time, rather than end time). This finding suggests that there is a deterministic process conditioned on the start time of the seizure, whereas the lead up to seizure offset showed more stochasticity. Based on these results we speculate that seizure onset is more likely to occur through a deterministic process (as in a bifurcation), where the brain state is driven across some ‘point of no return’. Offset is more likely to result from noisy fluctuations. Other studies have hypothesized that seizures terminate as the result of a bifurcation [43, 44]. However, the brain’s state during a seizure may merely approach a critical transition, without crossing over [45]. Therefore, it is possible to observe signs of critical slowing (as in (Kramer et al., 2012)) yet still have seizure termination driven by noise [4, 46, 47].
The presence of characteristic seizure durations should inform theoretical approaches to modeling seizure transitions. For instance, in a bistable regime, where noisy fluctuations drive the transition between a fixed point and oscillatory (‘seizure-like’) state, characteristic dwell times can emerge for the different states [4, 46, 47]. Dwell times provide one candidate mechanism for characteristic seizure durations. Bimodal populations in some patients suggest that the brain can support two distinct seizure trajectories (short and long). It has been shown experimentally that different durations of seizures may arise as the result of distinct onset stimuli [48]. Explanations for multiple seizure types can also be derived from computational models. For instance, different background stability properties in a cortical model can result in two distinct types of seizures [49]. Multiple seizure trajectories can also arise from different onset bifurcations [43]. Similarly, multiple offset bifurcations could terminate seizures earlier or later, giving two populations of duration. The results of this work suggest that long and short seizures arise from distinct mechanisms of seizure termination. This hypothesis is supported by a recent study from Payne et al. (2018), which found that long and short seizures were associated with different durations of post-ictal suppression [50].
Knowledge of parameter transitions within neural models can increase the information extracted from EEG, informing new hypotheses of seizure mechanisms and guiding clinical practice. There is some evidence to suggest that the clinical classification of a seizure is predictable soon after its onset [51], in other words, the evolution of a seizure may be somewhat predetermined. Our results support the existence of predictable seizure types, and provide additional metrics (based on the parameters of a neural model) that may extend our understanding of traditional seizure types. The consistency of neural model parameters over many seizures suggests that, for some patients, seizure trajectories are established via repetition. The notion of ‘learned epilepsy’ [52], is an interesting interpretation of epileptogenesis whereby the abnormal process is learning and spontaneously repeating a pathological sequence, rather than the sequence itself (all brains can support seizures). For some patients, successful treatment strategies may involve disrupting or even reversing memorization of the seizure, rather than addressing an underlying cause [52]. On the other hand, the current results (Fig 3) also showed that seizure pathways were highly patient-specific and not all subjects’ trajectories were conserved over time.
Neural mass models have the potential to highlight the relative contributions of excitatory versus inhibitory connections during seizures. This information can guide whether GABAergic or glutamatergic drugs are required. Previous studies using neural mass models have demonstrated alterations in the balance of excitation and inhibition estimated from data recorded during seizures [40, 53, 54]. The current estimation technique enables previous efforts to be extended to investigate a large number of seizures. Some patients showed decreased inhibition at seizure onset, whereas others demonstrated increased excitation (Fig 4), potentially warranting different therapies. Furthermore, patients with two duration populations may require different strategies to terminate their seizures. Knowing in advance when two adjunct therapies are needed is an important clinical insight that can provide crucial benefits to patients with drug refractory epilepsy. This study found that long seizures were correlated with lower inhibition and higher excitation (in one patient, the reverse was the case), which can guide electrical stimulation designed to precipitate early termination of seizures.
The presented model inversion technique and results have wide-ranging applications. The parameter estimates were consistent across many seizures. Until now, it has not been possible to show consistency of models of seizure transition in ECoG due to the limited availability of long-term seizure recordings. Results also generalized across patients. Although the cohort of 12 patients was not large, prior studies have restricted model inversion of seizures to only one or two patients [13, 55–58]. Another important aspect is that the techniques can be generalized across models. The estimation filter is not specially formulated for the Jansen and Rit model used in this study but can be generalized to any model that uses the basic matrix representation provided in the derivation (see Supplementary Materiall S1 Appendix). That is, the approach can be applied to any combination of coupled neural populations or indeed any network model that can be represented by a linear component and a non-linear sigmoidal (error function) coupling term.
Data driven modeling may provide the opportunity to identify which drug could be helpful for different classes of seizure, as different mechanisms of anti-epileptic drug action may preferentially effect the various connectivity parameters, though further validation of model predictions is needed to translate estimation results to clinical practice. Levels of AEDs have been related to features of the EEG signal [59, 60]. Therefore, it may be possible to extend this relationship to predicting the mode of action of an AED from an individual’s EEG. The time scale of connectivity changes may also be highly relevant to suppressing epileptic activity [10]. Future work should extend estimation to include time constants and investigate the utility of the outlined neural parameters to detect and predict drug action.
This study provided estimates of independent neural circuits for each channel of ECoG. Previous studies using coupled neural mass models have highlighted the importance of inter-channel interactions, particularly for seizure propagation [61]. However, this work considered local coupling as potentially more relevant to capture the onset of focal seizures. Non-local effects were described by the lumped input parameter, u, rather than explicitly by inter-model connections. It is possible that long and short seizures could be differentiated earlier based on inter-channel connectivity patterns. Future work will focus on extending the estimation algorithm for non-locally connected neural regions. An inverse solution to the time-varying, multi-scale network problem is not trivial and is likely to require additional constraints. For example, structural MRI data may inform prior probabilities of connection strength [58]. Individual neural models can also be coupled within a larger scale network [15]. The approach taken by Schmidt et. al. (2016) can be adapted to set prior probabilities, or otherwise constrain the propagation of an assumed density (Kalman) filter.
The challenge of large-scale model inversion is relatively well understood [4, 26]. A more recent problem in EEG analysis is the challenge of dealing with very high dimensional data. This study involved separate dimensions for model parameters, seizures, patients, electrode channels, and time. Distilling insights from such a large dataset is computationally intensive. To provide some insight into this problem, the estimation results presented were 1.5TB in size. Generation of each figure can take up to a week to complete for all patients. The use of “big data” techniques for EEG are becoming more relevant to the study of epilepsy [62]. It is important that tools for large scale analysis of EEG are made clinically available. The model inversion technique presented in this work is generalizable and freely available (https://github.com/pkaroly/Data-Driven-Estimation).
It is important to note that the presented results are only valid insofar as the connectivity parameters of a neural model capture the relevant dynamics underlying seizure transitions. The use of neural mass model to investigate seizures has gained wide acceptance among epilepsy researchers [11, 63–65]. Tracking excitatory and inhibitory strengths within a network is considered highly relevant to understanding and treating seizures [66]. The ability to infer directional connections (differentiate between in-going and outgoing pyramidal connections) is also an important feature of model inversion compared with alternative graph inference measures. The estimation method was previously validated on simulated data [25]. Nevertheless, it is highly challenging to quantify the accuracy of the model reconstructions from real data, where there is no ground truth. The results showed that the difference between reconstructed and actual ECoG was small (Supplementary S1, S2 and S3 Figs). The consistency of results across many seizures provides evidence that the estimation can give overarching insight into mechanisms of patients’ seizures. It is our hope that this study provides a stepping stone towards a fully validated model inversion framework to guide the clinical management of epilepsy. Future experimental work should investigate whether modulating connectivity strengths in a stereotypical fashion does lead to different energy and/or duration of seizures, as predicted by the current analysis.
This work provided a demonstration that the hidden local connectivity parameters of a neural mass model can be dynamically inferred from ECoG. Our results showed that seizures follow stereotypical pathways through parameter space. It is apparent that once a seizure has begun, a predefined sequence of states must be traversed before termination. For a subset of patients, there were two routes (short and long) to seizure termination. Short and long seizures began the same way but showed distinct offset mechanisms. Finally, the connectivity patterns at seizure onset showed common motifs across patients. These distinct sub-groups of onset mechanisms may suggest targeted treatment.
Techniques that unify neural mass models with data provide the means to address some of the unanswered hypotheses pertaining to epileptic dynamics. For example, theoretical studies have hypothesized that seizure trajectories are “innate”, or “repeatable” [13, 52]. The current results confirm that seizure pathways are indeed patient-specific and highly stereotyped. It has also been suggested that there are limited classes of onset mechanisms for seizures [43]. The current results show that there does appear to be a limited number of seizure onset “motifs” among patients. Finally, our group had previously hypothesized that long and short seizures reflect distinct cortical mechanisms [37]. The current results demonstrate that long and short seizures follow the same pathways but have different termination mechanisms. These results underscore the power of theoretical models to shed light on seizure mechanisms. It is our hope that these insights guide further modeling studies and may even prove to be directly translatable into clinical practice.
|
10.1371/journal.ppat.1003187 | Induction of p16INK4a Is the Major Barrier to Proliferation when Epstein-Barr Virus (EBV) Transforms Primary B Cells into Lymphoblastoid Cell Lines | To explore the role of p16INK4a as an intrinsic barrier to B cell transformation by EBV, we transformed primary B cells from an individual homozygous for a deletion in the CDKN2A locus encoding p16INK4a and p14ARF. Using recombinant EBV-BAC viruses expressing conditional EBNA3C (3CHT), we developed a system that allows inactivation of EBNA3C in lymphoblastoid cell lines (LCLs) lacking active p16INK4a protein but expressing a functional 14ARF-fusion protein (p14/p16). The INK4a locus is epigenetically repressed by EBNA3C – in cooperation with EBNA3A – despite the absence of functional p16INK4a. Although inactivation of EBNA3C in LCLs from normal B cells leads to an increase in p16INK4a and growth arrest, EBNA3C inactivation in the p16INK4a-null LCLs has no impact on the rate of proliferation, establishing that the repression of INK4a is a major function of EBNA3C in EBV-driven LCL proliferation. This conditional LCL system allowed us to use microarray analysis to identify and confirm genes regulated specifically by EBNA3C, independently of proliferation changes modulated by the p16INK4a-Rb-E2F axis. Infections of normal primary B cells with recombinant EBV-BAC virus from which EBNA3C is deleted or with 3CHT EBV in the absence of activating ligand 4-hydroxytamoxifen, revealed that EBNA3C is necessary to overcome an EBV-driven increase in p16INK4a expression and concomitant block to proliferation 2–4 weeks post-infection. If cells are p16INK4a-null, functional EBNA3C is dispensable for the outgrowth of LCLs.
| Epstein-Barr virus (EBV) is a causative agent of several types of B cell lymphoma. In human B cells, EBV reduces protein levels of at least two tumour suppressors that would otherwise be activated in response to over-expressed oncogenes. These proteins are BIM, which induces cell death and p16INK4a, which prevents cell proliferation. Repression of both is via epigenetic methylation of histones and is dependent on expression of both EBNA3A and EBNA3C – two EBV proteins required for the transformation of normal B cells into lymphoblastoid cell lines (LCLs). In this report we have used EBV with a conditionally active EBNA3C – active only in the presence of 4-hydroxytamoxifen – together with B cells from an individual carrying a homozygous deletion of p16INK4a to confirm that regulation of p16INK4a expression is a major function of EBNA3C and demonstrate that if B cells lack p16INK4a, then EBNA3C is no longer required for EBV-driven proliferation of LCLs. Furthermore we show that early after the infection of normal B cells, EBV induces p16INK4a accumulation that – if unchecked by EBNA3C (and EBNA3A) – prevents LCL outgrowth. Formal proof that p16INK4a is the main target of EBNA3C comes with the production of p16-null LCLs that have never expressed functional EBNA3C.
| The CDKN2A locus at human chromosome 9p21 encodes two important tumour suppressors, p16INK4a and p14ARF (equivalent to p19ARF in mice); these proteins are both critical regulators of cell proliferation. The cyclin-dependent kinase (CDK) inhibitor p16INK4a acts upstream of the cyclin D-dependent kinases (CDK4 and CDK6) and governs their phosphorylation of the retinoblastoma protein (Rb). By binding CDKs and blocking Rb phosphorylation, increased p16INK4a expression leads to a G1 cell cycle arrest (reviewed in [1], [2]). In contrast, the p14ARF and 19ARF proteins regulate p53 stability via inactivation of MDM2, the p53-degrading ubiquitin ligase. Stabilization and activation of p53 leads to G1 and G2 arrest by inducing the CDK regulator p21WAF1 or apoptosis by inducing pro-apoptotic factors such as BAX ([1], [2]).
The products of CDKN2A are responsible for senescence or apoptosis in cells receiving unscheduled proliferative signals from mutant or deregulated oncogenes (this is sometimes called ‘oncogenic stress’) [3], [4]. As a consequence p16INK4a and p14ARF/19ARF can potentially act as barriers to immortalization of cells placed in culture and the development of cancers in vivo. Both genes are progressively up-regulated with tissue aging and they probably contribute to the process of aging in vivo by reducing reservoirs of self-renewing stem cells [3], [4]. It is now generally accepted p19ARF plays a dominant role in these processes in mice, whereas p16INK4a is the dominant player in human cells. Unsurprisingly, in a wide variety of human cancers INK4a is inactivated by gene deletion, mutation or promoter DNA methylation [1], [3]. The CDKN2A locus is regulated epigenetically by polycomb complex-generated histone modifications [1], [5] and recently it has been demonstrated that the products of the locus act as a major barrier to the reprogramming of differentiated cells into induced pluripotent stem cells. As in the other biological contexts described above, p16INK4a dominates over p14ARF as the critical polycomb-regulated barrier to de-differentiation in human cells [6].
Epigenetic (ie heritable in the absence of changes to DNA sequence) silencing of genes is most commonly associated with methylation of cytosine at CpG dinucleotides in gene promoter regions (DNA methylation). However, heritable repression mechanisms involving covalent modifications of histones can precede DNA methylation at gene promoters. The best characterized of these modifications involves the polycomb system and repression of transcription by the polycomb repressive complexes 1 and 2 (PRC1 & 2). PRC2 mediates repression through the histone methyltransferase activity of the component EZH2 that catalyses the trimethylation of histone H3 lysine 27 (H3K27me3). The non-catalytic core subunits of PRC2 are SUZ12, EED and RbAp46/48, but PRC2 has no obvious DNA sequence-binding capacity so in most cases it is unclear how the complex is targeted to specific promoters in mammalian cells. H3K27me3 can result in the binding of a second complex, PRC1, which – together with PRC2 – greatly increases the chances of the more stable DNA methylation mark being deposited in the development of cancer (reviewed in [7], [8]).
Epstein-Barr virus (EBV) is a gamma-herpes virus that latently infects B cells, persistently infects >90% of adult humans, causes infectious mononucleosis in some adolescents and is etiologically linked to B cell malignancies such as Burkitt lymphoma (BL), Hodgkin lymphoma (HL) and diffuse large B cell lymphoma (DLBCL). Infection with EBV can also induce the continuous proliferation (known as ‘transformation’ or ‘immortalization’) of primary human B cells in vitro. The lymphoblastoid cell lines (LCLs) produced in this way carry the viral genome as extra-chromosomal episomes and express only nine latency-associated proteins [six nuclear (EBNAs 1, 2, 3A, 3B, 3C & LP) and three membrane-associated (LMP1, LMP2A & 2B)] along with several RNA species (reviewed in [9], [10]). The cells are described as being latently infected because the replication of the virus and production of new virions are restricted. The latency-associated gene products are responsible for activating quiescent B cells into the cell cycle, inducing and sustaining their proliferation and maintaining the extrachromosomal episome in these latently infected cells.
EBNA3A, EBNA3B and EBNA3C are considered to comprise a family of non-redundant EBV genes which probably arose during primate gamma-herpesvirus evolution by a series of gene duplication events since they have the same gene structure (ie a short 5′ coding exon and a long 3′ coding exon), are arranged in tandem in the EBV genome and share very limited amino acid sequence homology. The EBNA3 proteins (3A, 3B and 3C) show no significant similarities to known cell or viral factors, and although none appears to bind DNA directly, they all bind the cellular DNA-binding factor CBF-1 (aka RBP-JK). All three EBNA3s can also interact with cellular factors associated with the covalent modification of histones, the repression of transcription and gene silencing. For example, EBNA3A and EBNA3C associate with histone deacetylases (HDACs), BMI1, RING1 and CtBP, ([9]–[12] and our unpublished data). EBNA3A, EBNA3B and EBNA3C are all robust repressors of transcription when targeted directly to DNA in transient assays, and EBNA3A and EBNA3C – but not EBNA3B – are necessary to establish LCLs efficiently (reviewed in [10]). EBNA3A and EBNA3C also cooperate with oncogenic Ha-Ras in the transformation/immortalization of primary rodent cells. Since this assay demonstrates a capacity to override oncogene-induced premature senescence, this indicates that EBNAs 3A and 3C may play a similar role in B cells [11]–[13].
Several reports suggest that EBNA3C can directly associate with multiple factors involved in the regulation of cell cycle progression, particularly through the G1-S transition. These proteins include the tumour suppressor pRb, the ubiquitin ligase SCFSKP2, cyclin D1, cyclin A, c-MYC, MDM2, p53, CHK2 and E2F1 [14]–[22]. It remains to be determined which of these interactions occurs in latently infected B cells and whether they are functionally significant.
Recent analysis suggests that together the EBNA3s can regulate >1000 host genes in B cells, and that EBNA3A and EBNA3C act as oncoproteins, whereas EBNA3B behaves as a tumour suppressor. The regulation of many of these genes appears to utilize the host polycomb system of epigenetic repression and this probably contributes to B lymphomagenesis induced by EBV [23]–[31].
Genes repressed by the combined action of EBNA3A and EBNA3C include the pro-apoptotic BH3-only protein BIM and the cyclin-dependent kinase inhibitor p16INK4a. EBNAs 3A and 3C are together necessary to trigger the recruitment of PRC2 core subunits and the deposition of the H3K27me3 epigenetic mark on the promoters of both of these tumour suppressor genes [23]–[25], [28], [32]. The activation mark H3K4me3 is largely unaltered at these loci irrespective of H3K27me3 status, suggesting the establishment of so-called ‘bivalent’ chromatin domains [7], [8], [33]. B cells carrying EBV encoding a conditional EBNA3C-estrogen receptor (ER)-fusion protein (EBNA3C-ER) revealed that this polycomb-mediated epigenetic repression of BIM and p16INK4a expression is reversible, but the precise kinetics and molecular details of this have yet to be investigated [25], [28]
In order to determine the relative importance of the epigenetic repression of INK4a in B cell transformation and LCL outgrowth, we made use of primary B cells from an individual with a genetic lesion that prevents expression of functional p16INK4a, together with the conditional EBNA3C-expressing virus. We have also explored the early events following infection and outgrowth of LCLs from normal p16INK4a-expressing B cells.
To establish LCLs from primary B cells that are incapable of expressing functional p16INK4a, we made use of peripheral blood leukocytes (PBL) from an individual who is homozygous for a 19-base pair germ-line deletion in the second exon of the CDKN2A gene ([34]–[36]; Figure 1A). Since exon 2 is shared by both p16INK4a and p14ARF, neither of the wild type proteins is expressed. However two chimeric polypeptides originate from the disrupted locus: chimera p16/X consists of a fusion of the first 74 residues of p16INK4a with 64 amino acids that are specified by the +1 reading frame of exon 2, whereas the p14/p16 chimera has the amino terminal 88 residues of p14ARF fused in frame to the last 76 residues of p16INK4a. The p14/16 fusion retains all known functions of p14ARF, but neither protein exhibits any of the known functions of p16INK4a ([35], [36]; Figure 1B).
PBL carrying this deletion (sometimes called the Leiden deletion) were infected with recombinant BAC-based B95.8 EBV expressing the EBNA3C-modified estrogen receptor fusion (3CHT) described previously [25]. After infection, cells were cultured in the presence of the activating ligand 4-hydroxytamoxifen (4HT) to ensure the outgrowth of infected B cells expressing the active EBNA3C fusion. Four LCLs were established using each of two independently derived 3CHT viruses (LCL 3CHT-A1, -A2, -C1, -C2). LCLs produced with Leiden B cells and carrying a conditional EBNA3C were called LCL 3CHT p16-null. LCLs produced using the same viruses and normal B cells from a healthy blood donor were called LCL 3CHT p16-competent. Once LCLs were established (generally 2–3 months post-infection), aliquots of cells were taken and diluted in fresh medium; one was then continuously grown with 4HT (+4HT) and the other grown without (−4HT). After at least 3 weeks, proteins extracted from the LCLs were separated by SDS-PAGE and analyzed by western blotting to show EBV proteins EBNA3A, EBNA3B and EBNA3C or p16INK4a-related polypeptides (see for example Figure 1C and D). Irrespective of whether the cells were p16-competent or p16-null, in the absence of 4HT the EBNA3C protein is inactivated by sequestration from the nucleus and degradation, hence less protein was detected (Figure 1C). As has been described previously, no consistent change in the expression of other EBV latency-associated proteins was detected ([25], [37]; data not shown and see below). The pattern of p16INK4a related fusion proteins expressed was similar to that described previously in Leiden fibroblasts and LCLs established by infecting Leiden B cells with the B95.8 strain of EBV [35], [36]. In the LCL 3CHT expression of both p16INK4a and p14/p16 was induced by removal of 4HT and the non-functional p16/X fusion, although difficult to detect, appeared as a polypeptide doublet (Figure 1D and also [35]).
As outlined in the Introduction, EBNA3C – by cooperating with EBNA3A – functions to repress transcription of p16INK4a in infected B cells, by facilitating the recruitment of polycomb complexes to the CDKN2A locus and the deposition of the epigenetic repressive mark H3K27me3 upstream of the INK4a transcription start site [25]. Moreover, EBV does not appear to significantly alter the amount of H3K4me3 at the locus, suggesting the promoter is bivalent in B cells as has been described in various other types of cell including fibroblasts and embryonic stem cells [33].
Chromatin immunoprecipitations, coupled with quantitative PCR (qPCR) assays directed across the CDKN2A locus, were performed on the LCL 3CHT p16-null cells cultured plus or minus 4HT in order to establish whether EBNA3C was functional and regulated by 4HT. Although the p16INK4a fusion proteins retain no p16INK4a function, EBNA3C in the p16-null LCLs promotes similar chromatin modifications to the CDKN2A locus as previously shown for p16-competent cells, that is the deposition of H3K27me3 mark across the p16INK4a promoter region while it remains marked with H3K4me3 (compare Figure 2 A and B with [25]). This confirms that the chromatin modifications induced at this locus are not dependent on p16INK4a functioning as an inhibitor of cyclin-dependent kinases and that the transcriptional reprogramming activity of EBNA3C can be switched on and off in these cells (Figure 2C).
The fact that EBNA3C still regulates expression of the INK4a locus, when proliferative advantage of the EBNA3C-expressing cells is unlikely to be a contributory factor, provides further evidence that EBNA3C (and EBNA3A) might directly target the CDKN2A locus rather than the epigenetic changes being a secondary consequence of selection. In order to test this, we made use of recombinant EBV-BAC expressing epitope-tagged EBNA3C [28]. LCLs were established with this tagged virus and with un-tagged wild-type EBV-BAC using cells from the same donor. Anti-FLAG antibodies were used to perform ChIP assays from these LCLs and qPCR primer sets corresponding to sites across the INK4b/ARF/INK4a locus were used to analyse the precipitated DNA (Figure S1). The histograms show that EBNA3C is targeted to regions proximal to not only the ARF and INK4a transcription start sites (TSS), but also the INK4b TSS. These data are consistent with EBNA3C (alongside EBNA3A) associating with chromatin at several sites and regulating the polycomb-mediated repression of the whole INK4b/ARF/INK4a locus. Similar coordinated polycomb-mediated repression of this locus has been reported in other systems (reviewed in [1], [5]). The basis of EBNA3C targeting and the nature of the coordination are presently unknown, but these ChIP results add considerable weight to the hypothesis that EBNA3C (with EBNA3A) acts directly on chromatin at 9p21, rather than indirectly through a signaling pathway.
Having established that in the presence of the activating ligand 4HT the conditional EBNA3C remains capable of acting as an epigenetic modifier, we explored its role in sustaining LCL proliferation. Initially p16-competent and p16-null LCL 3CHT lines that had been established with 4HT were split, half were cultured without 4HT and half with 4HT for the next 22 days. Aliquots were taken and cultured with the thymidine analogue BrdU for 1 hour, harvested, stained with anti-BrdU-FITC and propidium iodide and analyzed by flow cytometry to reveal the percentage of cells undergoing DNA synthesis – and therefore proliferating – during the 1 hour BrdU-pulse. A representative experiment is shown in Figure 3A. It is clear that when EBNA3C is inactive, the LCL with wild type p16INK4a incorporates substantially less BrdU, consistent with the cell cycle arrest described in previous reports [25], [37]. In contrast the p16-null cells continue to proliferate at a similar rate in the absence of 4HT as when 4HT is retained in the medium and EBNA3C is functional. A comparison of all four p16-null lines with two p16-competent lines cultured for up to 30 days without 4HT and analyzed by the same BrdU-pulse strategy confirmed and extended these results (Figure 3B).
Consistent with the DNA synthesis data, in the p16-competent LCLs inactivation of EBNA3C resulted in a predictable dephosphorylation of Rb and down-regulation of p107 (Figure 4A and [25]). In contrast no changes in the phosphorylation status or quantity of these ‘pocket’ proteins were detected in the p16-null LCLs when 4HT was removed and EBNA3C inactivated (Figure 4A). Moreover, in the p16-competent LCLs the lack of functional EBNA3C caused a reduction in the expression of an E2F1-target gene EZH2 [38], but in the p16-null cells EBNA3C had no effect on EZH2 expression (Figure 4B). Although this result is consistent with the status of Rb in these cells (ie whether or not it is activated by dephosphorylation) it was perhaps surprising, since EBNA3C has recently been reported to directly alter E2F1 activity by suppressing its expression and inhibiting its capacity to transactivate target genes [22].
In order to determine whether EBNA3C alters the expression of other known E2F1-regulated genes in p16-null cells, a genome-wide microarray approach was adopted. This had the added value of identifying genes regulated by EBNA3C in non-tumour B cells – independently of p16INK4a, the status of the pRb-E2F1 axis and cell proliferation.
To investigate the effect of EBNA3C on the transcriptome of EBV-infected p16-null cells, 4HT was withdrawn from and re-added to three 3CHT cell lines established in the background of the Leiden p16INK4a mutation (as described in Materials and Methods and Figure S2). Gene level analysis identified 429 differentially regulated genes (p value<0.001) of which 252 have a fold change >1.4, and 82 have changed more than two-fold (Table S1). This analysis identifies some previously reported [29] EBNA3C-repressed genes (for example ADAMDEC1 and ADAM28) among those exhibiting the greatest degree of de-repression upon inactivation of EBNA3C, while another – encoding the tumour suppressor BIM (BCL2L11) – is altered in expression (p = 0.0005) but only to a modest degree (1.58-fold). CDKN2A, despite the observations made in this paper, is not on this list (p = 0.0012; fold change = 1.40 just fails the cut-off used). The raw data indicate that this low statistical confidence is because one of the three cell lines appears to have lost the ability to repress the mutated CDKN2A locus (Figure S3). EBNA3C up-regulates 39 genes at least two-fold, including AICDA – the gene encoding the cytidine deaminase that is responsible for mutation of immunoglobulin variable regions in germinal center B cells [39].
Since EBNA3C has been reported to target and repress E2F1 transcriptional activity [22], we specifically reviewed E2F1 and E2F-target genes. In keeping with our observations that the dependence of cell proliferation on EBNA3C requires the action of p16INK4a, analysis of E2F1 and a set of forty-five E2F1-target genes (based on those described in [40]) failed to show any differential expression attributable to inhibition of E2F1, between cells with active or inactive EBNA3C (Figure 4C). The only differentially regulated gene (p<0.001; fold change >1.4) from this list was TP73 (indicated in Figure 4C), which was modestly (1.7-fold) but consistently down-regulated upon EBNA3C inactivation (p = 7×10−5). This is the opposite direction to that expected from a mechanism involving the inhibition of E2F1 by EBNA3C [22]. Taken together, these data suggest that in the absence of p16INK4a, EBNA3C has no effect on the regulation of E2F1 or E2F1-target genes.
The p14/p16 chimeric protein expressed in p16-null Leiden fibroblasts has been extensively characterized and shown to retain all known functions of p14ARF [35], [36]. This chimera was modestly, but consistently up-regulated in p16-null LCL 3CHT cells upon inactivation of EBNA3C by removal of 4HT (Figure 1D). In order to determine whether there were any downstream consequences of this increase, cell extracts were analyzed by western blotting for p53 – the main functional target of p14ARF, and for p21WAF1 – a transcriptional target of p53 [1], [3]. Although the levels of p53 and p21WAF1 protein varied slightly between cell lines and experiments, there was no consistent accumulation of either in response to EBNA3C inactivation (see for example Figure 5A). Moreover there was no significant increase in p21WAF1 RNA detected by quantitative RT-PCR (Figure 5B). These data are consistent with the cells still proliferating in the absence of 4HT and suggest that p14/16 has no significant anti-proliferative activity in this context. Similar results were obtained when the p14ARF-p53-p21WAF1 pathway was analysed in p16INK4a-competent LCL 3CHT, ie there was evidence of slight de-repression of the ARF locus, and a marginal increase in p14ARF was apparent when EBNA3C was inactivated (Lenka Skalska PhD thesis, Imperial College London).
It has been suggested that EBNA3C might target p53 directly and indirectly – via MDM2, Gemin3 and ING4/5 – to impair its function [18], [19], [21], [41]. Since in p16-null LCLs there was an increase in the p14/p16 fusion protein after inactivation of EBNA3C, but no consistent activation of the p53-p21WAF1 pathway or reduction in proliferation, we asked whether the pathway remained intact in these cells. This was achieved by determining whether EBNA3C had any effect on the response of these LCLs to DNA damage induced by gamma-irradiation (γ-IR), since this is a well established activator of the pathway ([42]; reviewed in [43]). The results show clearly that both p16-competent and p16-null LCLs exposed to γ-IR respond with the phosphorylation of the CHK1/2 target on p53 (serine 20, [44]), p53 stabilization and activation, the accumulation of p21WAF1 and concomitant cell cycle arrest (Figure 6A and B). This was irrespective of whether EBNA3C is functional. Similar results were obtained with a second independent pair of 3CHT LCLs (data not shown). These data suggest that in the context of latent infection of B cells, EBNA3C has no detectable effect on p53 activation, its ability to transactivate p21WAF1 and induce cell cycle arrest. The data presented here are consistent with the behavior of ‘normal’ LCLs established with the B95.8 strain of EBV and also mitogen activated primary B cells [42].
Since the p53-p21WAF1 pathway appeared to be intact, and the p16INK4a-null cells were capable of G1 and G2 cell cycle arrest following DNA damage, we wanted to show in an independent assay that Rb could be dephosphorylated and the cells could arrest in G1. When ‘normal’ p16-competent LCLs are grown to high density, the majority of the cells arrest in G1 and Rb is present only in its hypophosphorylated, activated form [45]. The two p16-null 3CHT LCLs analysed for p53/p21 function were therefore grown to saturation density (5 days in culture without a change of medium) to establish whether Rb could be activated by dephosphorylation in these cells. The results (see for example Figure S4) showed that at high density Rb is dephosphorylated and cells accumulate in G1, irrespective of either the p16INK4a or EBNA3C status of the LCLs. These data demonstrate unequivocally that Rb – the operational target of p16INK4a – can be activated in the Leiden LCLs and that EBNA3C does not override a p16INK4a-independent G1 arrest.
All the data described above are consistent with p16INK4a being the major barrier to proliferation of established LCLs; EBNA3C (cooperating with EBNA3A) epigenetically represses transcription from CDKN2A in order to prevent the cell cycle arrest induced by p16INK4a. However, soon after the infection of resting B cells, when they are activated to become B-blast-like and driven into the cell cycle the nuclear environment is very different to that in the transformed, continuously proliferating LCL cells. Furthermore these LCL populations may have already become oligoclonal or even a single clone that has been selected by the in vitro culture conditions [46]. It was therefore important to establish how p16INK4a was regulated in newly infected primary B cells.
In order to investigate the quantity of p16INK4a-encoding transcripts expressed in normal primary B cells during the first weeks after infection with EBV, two RT-PCR assays were utilized – one based on a qPCR detecting an amplicon in INK4a exon 1 (Figure 7A) and a second that targets the exon 2/3 boundary shared by p16INK4a and p14ARF (Figure 7B). Both assays showed that INK4a/CDKN2A is silent or has very low activity in purified CD19-positive, resting B cells at the time they are infected. Nevertheless, irrespective of whether or not the infecting virus expresses functional EBNA3C (WT-BAC, EBNA3C-revertant or 3CHT plus 4HT) or is EBNA3C-deficient (EBNA3CKO or 3CHT minus 4HT), during the first few days post-infection there is a substantial increase in transcripts corresponding to INK4a/CDKN2A. This is the period that corresponds to EBNA2 and EBNA-LP expression, blast activation, synthesis of cyclin D2, synthesis of c-MYC and rapid cell proliferation [47]–[50]. However, 4–7 days post-infection the effect of EBNA3C was seen. In cells infected with wild type (WT-BAC) or EBNA3C-revertant virus, expression of INK4a/CDKN2A no longer increases and remains relatively constant – presumably below a critical threshold – up to the end of the experiment at 28 days. In contrast, in cells infected with EBNA3CKO or 3CHT without 4HT, INK4a/CDKN2A transcript levels rise progressively until at least 18 days. At, and beyond this time, there was substantial cell death in the EBNA3C-deficient cell populations (see below) such that beyond 18 days, it was not possible to isolate sufficient RNA for analysis. Consequently no samples from EBNA3C-deficient cell lines were assayed for INK4a/CDKN2A transcription after day 18.
When B cells were infected with the 3CHT virus and then cultured in 4HT, the amount of p16INK4a/CDKN2A RNA reached a steady state, with similar kinetics, but at a slightly higher level than in wild type and revertant-infected cells (Figure 7A and B). These data are consistent with our original observations that the EBNA3C-estrogen receptor fusion protein is a little less efficient than unmodified EBNA3C at regulating the level of p16INK4a protein in established LCLs, but enables proliferation [25].
Taken together, the results suggest that soon after the infection of primary B cells, EBV induces p16INK4a transcription and EBNA3C (with EBNA3A) is necessary to prevent the accumulation of p16INK4a above a threshold that would block proliferation and LCL outgrowth.
The populations of CD19-positive B cells that were infected with EBNA3C-expressing or EBNA3C-deficient viruses and sampled for transcript analysis were also sampled at the times indicated to assess DNA synthesis (Figure 8). Samples were taken and pulsed for 16 hours with the alkyne- containing thymidine analogue EdU, labeled with AF488 dye and analysed by flow cytometry to reveal the percentage of cells incorporating EdU and therefore proliferating. A histogram (Figure 8A) and representative plots (Figure S5) show that after 7 days, 20–30% of the cells are proliferating, irrespective of whether or not EBNA3C is functional (as was shown previously [50]). By 13 days, only in the populations of cells carrying EBNA3C-expressing virus does the percentage of EdU-positive cells significantly increase and after 20 days nearly 50% of this population is EdU-positive and therefore proliferating. In contrast the proportion of EdU-positive cells in the EBNA3C-deficient populations fails to increase after 13 days and is declining by 20 days. Visual inspection of these cultures indicated substantial numbers of cells were dying. This was confirmed by flow cytometry after vital staining (Figure 8B). By 27 days more than 90% of the EBNA3C-deficient populations were dead, such that there were insufficient surviving cells to permit reliable analysis of proliferation at this time point.
In our experience purified CD19-positive B cells are less robust during outgrowth of LCLs than if one starts with PBL. This is probably because the latter contain a sub-population of non-infected viable cells such as macrophages that act as a paracrine source of B cells survival factors. The experiment described above was therefore repeated with PBL (independently with cells from two blood donors – D11 and D13) and EdU incorporation and viability were assessed at the times after infection indicated (Figure 8C and D; Figure S6). The percentages of cells proliferating followed similar trends to the CD19-positive B cells, with little difference in proliferation rates at 10 days, and the effects of EBNA3C-deficiency becoming more apparent as the infection proceeded, such that proliferation had dropped to 10% by 30 days in the EBNA3C-deficient infections (Figure 8C), with over 50% of detectable cells being dead (Figure 8D). We note that this assay does not distinguish between the remaining live EBV-infected B cells and other surviving non-B cell lineages, which might contribute to the cell culture at this time – however the proportion will be similar in each different infection. In both PBL and CD19-purified B cell experiments, no EBNA3C-deficient cell lines grew out as LCLs. These data are therefore consistent with the hypothesis that accumulating INK4a/CDKN2A transcripts (Figure 7) are translated into p16INK4a protein that then activates pRb and induces cell cycle arrest (and/or cell death) when EBNA3C is unavailable.
If p16INK4a is the major target of EBNA3C and barrier to B cell transformation, a prediction is that functional EBNA3C never has to be expressed to produce p16INK4a-null LCLs. Leiden PBL were therefore infected with 3CHT EBV and cultured in the presence (3CHT with 4HT) and absence (3CHT never 4HT) of 4HT. Additional infections were performed with 3CKO and 3C-revertant viruses. The relatively small number of viable cells in the Leiden PBL aliquot used for this experiment prevented meaningful proliferation analysis of the cell lines during the early stages of outgrowth.
Although outgrowth of EBNA3C-deficient lines (3CKO and 3CHT-never 4HT) was slower than EBNA3C-expressing ones, all the cell lines continued proliferating beyond 30 days – the time by which normal p16INK4a-competent B cells infected with EBNA3C-deficient viruses stopped proliferating and/or died (see above). By about 2.5 months cell lines were considered established and the culture conditions were reversed, ie 4HT was added to 3CHT-never 4HT cells, and withdrawn from 3CHT with 4HT cells. The removal/addition of 4HT did not consistently alter the proportion of cells synthesizing DNA at either 25 days (Figure 9A) or at 10 days (not shown). Western blotting of protein extracted from the cell lines as early after infection as was practical (about 1.5 months for most lines and about 3 months post-infection for 3CKO LCL) confirmed the EBNA3C status of the individual lines, and that the other EBV latency-associated proteins are generally expressed at similar levels in all the lines. The only exception is EBNA-LP, the expression of which is notoriously variable between LCLs (Figure 9B).
While we observed some slight differences in proliferation between cells expressing or lacking EBNA3C in the p16INK4a-null background, particularly early during the establishment of LCLs, here we have formally shown that the failure of EBNA3C-deficient EBVs to repress p16INK4a expression is the central reason that these viruses normally fail to transform B cells into continuously proliferating cell lines.
Two EBV latency-associated nuclear proteins – EBNA3A and EBNA3C – cooperate to harness the polycomb system for the repression of two host tumour suppressor genes. BCL2L11, the first of these genes to be characterized, encodes BIM a pro-apoptotic BCL2-family member that lowers the apoptotic threshold of cells, particularly in cells of the immune system ([23], [24], [28]; reviewed in [51]). The second target gene, INK4a, encodes the cyclin dependent kinase inhibitor p16INK4a that prevents the phosphorylation of the tumour suppressor Rb and the resulting hypophosphorylated Rb blocks entry of cells into S phase of the cell cycle ([25], [37]; reviewed in [1], [2]).
In this study we have explored further the role of p16INK4a in B cell transformation and shown that its expression is the major barrier to the initial outgrowth and subsequent proliferation of LCLs produced by the infection of primary B cells with EBV. This was made possible using Leiden B cells carrying a homozygous genomic deletion that specifically ablates production of functional p16INK4a. These cells were infected with recombinant EBVs that express either a conditional EBNA3C or no EBNA3C. A comparison of p16-null LCLs with LCLs established from normal B cells showed unequivocally that, if p16INK4a is not functional, then EBNA3C is unnecessary to sustain cell proliferation. Consistent with this, it was possible to transform p16-null B cells into LCLs with EBV, but without any functional EBNA3C ever having been expressed. A possible explanation for why EBV has evolved a mechanism for suppressing p16INK4a expression became apparent from examining the outcome of attempted transformations of normal B cells with EBNA3C-deficient EBV. These experiments revealed that EBV infection induced p16INK4a transcription in the first few days after infection – when EBNA2 transactivates directly (eg c-MYC) or indirectly (eg cyclin D2) inducers of cell cycle progression and hyperproliferation [49], [50]. It is likely that unscheduled entry into S-phase is interpreted as oncogene de-regulation, and activation of p16INK4a transcription is a consequence. When the infecting virus expressed functional EBNA3C (and EBNA3A) there was a halt to the increase of p16INK4a expression from about day 7 onwards. However, if functional EBNA3C was not expressed, transcription from INK4a continued uncontrolled and the level of mRNA progressively increased over the next 2–3 weeks, until finally most of the cells arrested and/or died. EBNA3A/3C-mediated inhibition of INK4a transcription is therefore a critical countermeasure for the virus to bypass an intrinsic host cell defense against oncogenic transformation. This ensures expansion of the infected B cell population and the initiation of long-term persistence (see below for further discussion). Strictly speaking, here EBNA3C and EBNA3A do not actually repress INK4a transcription, but rather block its activation. We assume this involves the recruitment of polycomb complexes to the CDKN2A locus, leading to H3K27me3 modifications on chromatin around the INK4a transcription start site as we see in established LCLs.
Although the Leiden B cells fail to make functional p16INK4a, they do express the fusion protein p14/p16 that – because it includes the critical, conserved amino-terminal 25 amino acids of p14ARF – has all the known functions of p14ARF [35], [36]. Furthermore expression of p14/16 was elevated when EBNA3C is inactivated and there was a concomitant reduction of H3K27me3 at the ARF locus (data not shown). It was therefore surprising to discover that in the Leiden cells, increased p14/p16 expression had no apparent effect on proliferation or LCL outgrowth. This prompted us to investigate the pathway that is normally activated downstream of p14ARF. Since neither p53 nor p21WAF1 were activated and we could detect no impairment of p53 or p21WAF1 function, irrespective of the EBNA3C status of cells, we conclude that either the amount of p14/16 does not exceed a critical threshold sufficient to trigger the p53 pathway, or that in these EBV-infected B cells another factor – perhaps EBV-encoded – inhibits p14ARF function. A good candidate for this is EBNA-LP that, in transient assays, can associate with and interfere with the capacity of p14ARF to inhibit proliferation [52]. Genetic analysis of the role of EBNA-LP during the early stages of B cell transformation, when EBNA-LP is normally expressed at high levels, should help resolve this issue. A third possibility that cannot be excluded is that p14ARF does not play a significant role in the response of any human cells to oncogenic stress – as has been reported for human fibroblasts and epithelial cells [1], [3].
Additional analysis of the Rb-E2F1 axis produced no evidence that EBNA3C compromises the action of these host proteins. Microarray analysis showed that in continuously proliferating B cells EBNA3C has no effect on the transcription of (n = 45) E2F1-target genes, with the exception of that encoding the tumour suppressor p73. The data suggest that EBNA3C expression is associated with very modest activation of TP73 in cycling B cells. These data have not been validated at the protein level, but the changes in expression clearly have no significant effect on LCL proliferation. Furthermore, by growing LCLs to saturation density it was possible to show hypophosphorylation of Rb in the p16-null cells and that a p16INK4a-independent G1 arrest is unaffected by EBNA3C in the context of normal viral latency-associated gene expression. All the results are consistent with p16INK4a being the major barrier to maintaining EBV-induced proliferation and the principal requirement of EBNA3C for LCL outgrowth is to restrain transcription of INK4a.
It was reported recently that EBNA3C helps attenuate a DDR that is particularly active in the first week after EBV infection of normal B cells in vitro – when the cells are beginning to cycle very rapidly [50]. It is striking that during the same period p16INK4a appears to be actively transcribed. Although the increase in p16INK4a is probably primarily associated with oncogene activation, one cannot exclude the possibility that the increase is partly a response to DNA damage. Such an increase in p16INK4a has been described following treatment of several types of cell with various DNA-damaging stimuli (reviewed in [3]). One can also speculate that if the increase in p16INK4a activates Rb, which then leads to the repression of E2F-regulated genes while c-MYC is constitutively active, then some cells might enter S phase with sub-optimal amounts of DNA precursors and/or replication enzymes and this could lead to stalled DNA replication that is ‘read’ as DNA damage and triggers phosphorylation of histone H2AX. So, although EBNA3C may have a direct effect on the ATM/CHK2 pathway that triggers the DDR as has been suggested [50], its absence could also exacerbate the response because of the accumulating p16INK4a. Indeed, when EBNA3C is inactivated by removal of 4HT in continuously cycling p16-competent LCLs, substantially more H2AX is phosphorylated than in p16-null LCLs treated in the same way – this strongly implicates p16INK4a in the DDR of LCLs (Figure S7). It will require systematic genetic analysis of EBNA3C and its role during the first 2–3 weeks post-infection to disentangle the DDR and the p16INK4a senescence response and their precise contributions to the inhibition of B cell transformation. Nevertheless, the data presented here clearly demonstrate that in the absence of active EBNA3C and p16INK4a, even if the DDR is not attenuated by EBNA3C, it is insufficient to prevent LCL outgrowth. In contrast, if p16INK4a is present when EBNA3C is absent, proliferation and LCL outgrowth are completely blocked – suggesting a dominant role for p16INK4a in the restriction of B cell transformation.
When p16-competent cells were infected with EBNA3C-deficient EBVs, from about day 14 post-infection there was increasing evidence of viable cells arresting and escalating levels of cell death in the population. Since this increase in cell death does not appear to occur in the p16-null B cells infected with similar viruses, it suggests there may be some crosstalk between p16INK4a and the apoptotic machinery. This would be consistent with the evidence that in lymphocytes the default pathway triggered by p16INK4a is apoptosis rather than prolonged cell cycle arrest [53], [54]. However one must remember that EBNA3A and EBNA3C cooperate to repress not only INK4a, but also BCL2L11, encoding BIM. Since in the absence of EBNA3C there is a parallel increase in BIM RNA expression between days 7 and 15 (Figure S8), we suggest it is highly likely that in cells accumulating p16INK4a, the simultaneous increase in BIM will substantially reduce the threshold to apoptosis. Cells with high levels of BIM will die because of its capacity to bind and inactivate all anti-apoptotic BCL2-family members [51]. At this stage we cannot rule out the possibility of p14/p16 playing a role in the p53-independent induction of cell death in the p16-null background.
Most of the data on EBV persistence in humans are consistent with the viral genome residing long-term in a population of memory B cells (MBCs). It appears that to establish persistence EBV initially infects resting (naïve) B cells and drives these to proliferate as activated B-blasts [9], [55]. We assume that – by analogy to what occurs in culture – repression of INK4a is essential at this stage to ensure the transient proliferation of the infected population. This expanding population of activated B cells is then thought to either migrate into or initiate a germinal center, where the cells differentiate to centroblasts, centrocytes and finally MBCs. This process includes the regulated shutdown of latent EBV expression [55]. Since the repression of INK4a involves polycomb-mediated covalent histone modifications, it is possible the epigenetic memory of this will be passed to progeny cells and carried through into the MBC population, even if the initiators, EBNA3A and EBNA3C, are no longer expressed.
It is well established that polycomb-mediated gene repression is often a precursor to promoter DNA methylation (reviewed in [7], [56]), so it is reasonable to hypothesize that the B cells reprogrammed in vivo by EBV will be particularly prone to aberrant DNA methylation at the CDKN2A and BCL2L11 loci during tumorigenesis. This is consistent with reports describing promoter methylation of these genes in EBV-positive B lymphomas and derived cell lines [24], [57]–[59]
In summary, we have formally demonstrated that the epigenetic repression of p16INK4a expression by EBV is central to the virus's ability to infect resting B cells and establish latency – and probably persist life-long in vivo. By targeting INK4a (and BCL2L11) EBV has evolved an effective countermeasure to oncogenic stress triggered by the early stages of infection. To our knowledge this particular strategy is unique among tumour viruses. Finally by making use of Leiden p16-null B cells, it has been possible to generate LCLs in which multiple EBV-regulated polycomb-targeted genes – including CDKN2A – can be switched on and off without altering the rate of cell proliferation or stage of cell differentiation.
This study was conducted according to the principles expressed in the Declaration of Helsinki. Patient peripheral blood leukocyte (PBL) samples were obtained from archival stocks of Leiden University Medical Center. These were collected as part of an ongoing international collaborative study of CDKN2A mutations and malignant melanoma (project RUL 99-1932). Written, informed consent was obtained from participants at the time of collection.
LCL under continuous culture were grown in RPMI-1640 medium supplemented with Penicillin/Streptomycin and L-Glutamine and 10% FCS. For initial outgrowth of LCLs and infection experiments, peripheral blood leukocytes (PBLs) were isolated from buffy-coat residues (UK blood transfusion service) by centrifugation over ficoll. Where indicated, B cells were further purified from PBLs by binding to CD19 microbeads (Miltenyi Biotec) and magnetically separating using positive selection on an autoMACS separator (Miltenyi Biotec). Isolated cells were kept at 1–2×106 cells/ml in RPMI-1640 supplemented with L-glutamine, 500 ng/ml Cyclosporine A (Sigma) and 15% FCS (which we had batch-tested for LCL outgrowth efficiency – PAA Laboratories) in a 37°C incubator (5% CO2) until infection (which was within 24 hours of isolation). Gamma-irradiation of cells was performed as described previously [60].
The recombinant viruses used in this study were wild-type B95-8 BAC [61], EBNA3C knockout and revertants [23] and the EBNA3C-estrogen receptor fusion 3CHT [25]. Infection and outgrowth of LCLs was generally as described previously ([25], [26]. For cells to be analysed in the first month post-infection, PBLs were seeded at 2×106 cells per well in 24 well plates and infected with 1.2–1.5×105 Raji infectious units of virus. Medium was refreshed (approximately 60% of medium removed and replaced with fresh) after 24 hours and twice weekly thereafter. Cyclosporine A was retained in the medium for the first two weeks of outgrowth. As cell cultures saturated, approximately 75% of the culture was discarded for each split. CD19-purified B cells (mixed from four donors) were seeded for infection in 5 ml at 2×106 cells/ml in 25 cm2 flasks stood upright. Approximately 5×105 infectious units of virus (1.5×105 units for 3CHT virus infections) were added to the cultures and medium was changed as for PBLs. These cell cultures were expanded to larger volumes and flasks as their numbers increased.
For the first infection of Leiden PBLs, cells were recovered from liquid nitrogen and 106 cells per well were seeded in a 24 well plate. Four infections (two with each of two independently generated 3CHT recombinant EBVs – [25] were performed with 100 µl of virus supernatant (approximately 104 infectious particles). The four cell lines were grown out in the presence of 4HT. The second Leiden PBL infection was hampered by the very low viability of the cells after thawing, resulting in fewer infected cells and longer times to establish the cell lines. Separate infections were undertaken with 3CKO and 3Crev and with 3CHT virus either in the presence or absence 4HT (using approximately 5×104 infectious particles per well).
Cell proliferation was assessed by measuring the incorporation into DNA of nucleotide analogues (BrdU or EdU). BrdU incorporation was assessed as described previously [25]. EdU (Life Technologies) incorporation was assessed as follows. For time courses of newly infected cells, infected cells were cultured in 1.5 ml in wells of a 24 well plate. Five hundred microlitres of this culture (or 300 µl where cells were dense enough to change the colour of the medium) was transferred to a new well in a final volume of 1 ml. After 24 hours, EdU (15 µM) was added in a volume of 500 µl to give a final concentration of 5 µM. Cells were harvested after a sixteen hours in the presence of EdU. For established cell lines, cells were seeded at 3×105 cells/ml in 4 ml in a well of a 6 well plates. After 20 hours, culture was supplemented with 1 ml of warm medium containing EdU for a final concentration of 10 µM. Cells were harvested after 1 hour in the presence of EdU. Cells were washed once in cold PBS, then re-suspended in 500 µl PBS containing Violet Fixable Dead Cell Stain (Life Technologies) at 1 µl/ml. Cells were kept on ice in the dark for 30 minutes, then washed once with PBS, once with PBS/1% BSA, re-suspended in 100 µl PBS, and fixed by addition of 400 µl 100% ethanol. After >1 hour on ice, cells were rehydrated in 500 µl PBS/BSA for 15 minutes, then pelleted and re-suspended in 100 µl PBS/BSA. EdU was labeled by Click chemistry using an azide-derivative of AF488 dye according to the manufacturer's instructions (Life Technologies), before washing cells twice with PBS/BSA and re-suspending in 500 µl PBS/BSA supplemented with either DRAQ5 (Biostatus Ltd - 2 µl/ml) or FxCycle Far Red (1 µl/ml) to stain for DNA content.
Cell fluorescence was measured on either an LSR II (Becton Dickinson) or iCyt ec800 (Sony) flow cytometer. Single cells were gated based on Propdium Iodide/DRAQ5/FxCycle Far Red fluorescence (comparing fluorescence area to width or height at 633/690). Fluorescence measured by 405/450 filters indicated live/dead status, and only live cells were included for assessing proliferation by EdU (488/530).
Western blotting was performed as described previously [23]. Anti-EBV antibodies used are as described previously [26] with the addition of the anti-EBNA-LP antibody clone 4D3 [62]. Antibodies and western blotting for p53, p53-phospho-ser20, p21WAF1, Rb, phospho-Rb and p107 are as described in [25], [60]. The EZH2 antibody used was from Active Motif (39934), anti-p16INK4a JC8 was a gift from Gordon Peters (CRUK), anti-p16INK4a DCS50.1 from Abcam. Throughout gamma-tubulin was probed with mAB GTU-88 (Sigma).
For qRT-PCR RNA was isolated using RNeasy mini kit (Qiagen) with DNase digestion according to the manufacturer's protocol. RNA was reverse transcribed into cDNA using the SuperScript III First-Strand Synthesis Supermix for qRT-PCR (Life Technologies), using a 55°C incubation to facilitate melting of the GC-rich first exon of p16INK4a transcripts. Quantitive (q)PCR was generally undertaken on the cDNA and analysed as described previously [25], using platinum SYBR green qPCR SuperMix UDG kit (Invitrogen) for SYBR green-based qPCR assays, and using FAST BLUE qPCR MasterMix Plus dTTP (Eurogentec) for Taqman probe-based assays. Primer sets for the exon 1 INK4a assay and the RPLP0, GNB2L1 and ALAS1 endogenous control genes are described in [25]. Total CDKN2A transcript quantity was measured using the Hs00923894_m1 Taqman assay (Life Technologies) spanning the junction between exons 2 and 3. BIM (BCL2L11) assay is as described in [28]. Primers used for p21WAF1 were: p21-fwd ctggagactctcagggtcgaa and p21-rev gcggattagggcttcctctt.
ChIP assays and qPCR analysis were performed essentially as described previously: H3K27me3 and H3K4me3 [25] and EBNA3C-TAP [28]. The primer pairs for the transcriptional start sites for p15INK4b and p14ARF are described in [63].
To generate a balanced set of RNA samples either expressing or lacking functional EBNA3C, cells were cultured and RNA extracted as described below and illustrated in Figure S2: 4HT was washed out of cell lines that had been established in the presence of 4HT (day zero). These cells were cultured in the absence of 4HT, alongside the same cell line maintained in the presence of 4HT. After thirty-two days, the cells growing in the presence of 4HT were again split into two and one culture washed and subsequently grown in the absence of 4HT. Those cells grown in the absence of 4HT were also split into two cultures. To one 4HT was re-added. At day 36, cells were seeded at 3×105 cells/ml and RNA harvested the following day, producing samples whose conditions had been altered either 5 or 37 days previously (Figure S2). The cultures whose conditions had been changed on day 32 were grown a further 28 days (ie 33 days total culture in their new conditions) and RNA was isolated as above. Thus, each cell line yielded six samples: two from cells grown in the absence of 4HT for 33 or 37 days, one always grown in the presence of 4HT, one grown with 4HT for 33 days after a period of 32 days without HT, and samples five days after 4HT withdrawal or re-addition.
RNA was reverse transcribed and amplified (using the Applause WT-Amp Plus ST kit – NuGen) and hybridized to Affymetrix Human Exon 1.0ST microarrays by UCL Genomics. Gene level analysis was performed using the MMBGX algorithm [64] to generate gene-level data according ENSEMBL genome annotation version 64, as mapped to the Exon microarray by AnnMap (formerly X:Map), broadly as described previously [65], [66].
Statistical analysis of microarray data was performed using Partek Genomic Suite v6.5 (Partek Inc). The differential expression of genes was assessed by a 3-way ANOVA model using Method of Moments [67]. The model accounts for the cell line of origin (C), and the factorial combination of whether the cells are grown with or without 4HT (H) and the time of treatment (T) - classed as 5 days or >30 days - according to the formula: Yijkl = μ+Ci+H*Tjk+εijkl where Yijkl represents the lth observation on the ith cell line jth 4HT status kth Treatment term; μ is the common effect for the whole experiment. εijkl represents the random error present in the lth observation on the ith cell line ID jth HT status kth Treatment term. The errors εijkl are assumed to be normally and independently distributed with mean 0 and standard deviation δ for all measurements. Samples having different 4HT treatments for >30 days were compared using the contrast method (Partek Genomics Suite). For visualisation, variations in gene expression between cell lines were removed using the Remove Batch Effect tool in Partek Genomics Suite, based on the same ANOVA model used to identify the differentially regulated genes.
|
10.1371/journal.ppat.1000191 | Repetitive N-WASP–Binding Elements of the Enterohemorrhagic Escherichia coli Effector EspFU Synergistically Activate Actin Assembly | Enterohemorrhagic Escherichia coli (EHEC) generate F-actin–rich adhesion pedestals by delivering effector proteins into mammalian cells. These effectors include the translocated receptor Tir, along with EspFU, a protein that associates indirectly with Tir and contains multiple peptide repeats that stimulate actin polymerization. In vitro, the EspFU repeat region is capable of binding and activating recombinant derivatives of N-WASP, a host actin nucleation-promoting factor. In spite of the identification of these important bacterial and host factors, the underlying mechanisms of how EHEC so potently exploits the native actin assembly machinery have not been clearly defined. Here we show that Tir and EspFU are sufficient for actin pedestal formation in cultured cells. Experimental clustering of Tir-EspFU fusion proteins indicates that the central role of the cytoplasmic portion of Tir is to promote clustering of the repeat region of EspFU. Whereas clustering of a single EspFU repeat is sufficient to bind N-WASP and generate pedestals on cultured cells, multi-repeat EspFU derivatives promote actin assembly more efficiently. Moreover, the EspFU repeats activate a protein complex containing N-WASP and the actin-binding protein WIP in a synergistic fashion in vitro, further suggesting that the repeats cooperate to stimulate actin polymerization in vivo. One explanation for repeat synergy is that simultaneous engagement of multiple N-WASP molecules can enhance its ability to interact with the actin nucleating Arp2/3 complex. These findings define the minimal set of bacterial effectors required for pedestal formation and the elements within those effectors that contribute to actin assembly via N-WASP-Arp2/3–mediated signaling pathways.
| Enterohemorrhagic Escherichia coli (EHEC) O157:H7 is a food-borne pathogen that causes diarrhea and life-threatening systemic illnesses. EHEC colonizes the intestine by adhering tightly to host cells and injecting bacterial molecules that trigger the formation of a “pedestal” below bound bacteria. These pedestals are generated by reorganizing the actin cytoskeleton into densely packed filaments beneath the plasma membrane. Pedestal formation is therefore not only important for EHEC disease, it provides a means to study how mammalian cells control their shape. We show here that two EHEC proteins, Tir and EspFU, are sufficient to trigger pedestal formation. Tir localizes to the mammalian plasma membrane, and its central function is to promote clustering of EspFU. EspFU contains multiple repeat sequences that stimulate actin polymerization by binding N-WASP, a host protein that initiates actin assembly. Although a single repeat of EspFU can generate pedestals, multi-repeat variants promote actin assembly cooperatively. One explanation for this synergy is that tandem repeats can potently trigger the formation of a complex of mammalian proteins that modulate the actin cytoskeleton. These findings define the minimal set of EHEC effectors required for pedestal formation and the elements within those effectors that confer their ability to alter cell shape.
| Enterohemorrhagic Escherichia coli (EHEC) O157:H7 colonize the intestinal tract of cattle and other reservoir hosts without inducing disease, but cause severe diarrheal illness in humans that ingest contaminated materials [reviewed in 1]–[3]. The mode of epithelial colonization by EHEC reflects its membership in the attaching and effacing (AE) family of pathogens. These bacteria, which include enteropathogenic E. coli (EPEC) and Citrobacter rodentium, attach tightly to the intestinal epithelium, efface microvilli, and generate filamentous (F-)actin pedestals beneath sites of adherence. The formation of AE lesions is critical for pathogenesis, because mutations that abolish their biogenesis severely impair colonization [4]–[7]. Moreover, an EHEC mutant that forms AE lesions but possesses a diminished capacity to stimulate actin assembly is defective at expanding the initial infectious niche [8].
During infection, EHEC expresses a type III secretion system capable of translocating more than 30 effector proteins from the bacterium into the mammalian cell [9]. This delivery system is encoded by the locus of enterocyte effacement (LEE), which also contains several of the substrates for injection [reviewed in 10]–[11]. Among these LEE-encoded effectors is the translocated intimin receptor (Tir), which is essential for AE lesion formation. Tir is delivered into the host cell, where it localizes to the plasma membrane in a hairpin conformation that includes a central extracellular region that binds to intimin, a LEE-encoded adhesin. Intimin-Tir interaction promotes intimate attachment to the host cell, and also results in clustering of the N- and C-terminal cytoplasmic domains of Tir [12], which are capable of interacting with host proteins.
The Tir molecules from EHEC and EPEC both trigger actin assembly pathways that involve N-WASP, an actin nucleation-promoting factor [13]–[15]. N-WASP utilizes a C-terminal WH2/verprolin-connector-acidic (VCA) segment to activate the Arp2/3 complex, a major actin nucleator in cells [reviewed in 16]–[17]. Normally, N-WASP adopts an autoinhibited conformation in which its VCA domain is sequestered by an intramolecular interaction with a central GTPase binding domain (GBD). It can be activated by several stimuli, including Nck, an adaptor protein that binds to its proline-rich domain (PRD), and Cdc42, a small GTPase that binds the Cdc42-Rac binding (CRIB) sequence within the GBD [18]–[19]. When assayed using purified proteins in vitro, either Cdc42 or Nck is sufficient to stimulate N-WASP-Arp2/3–mediated actin assembly. However, under physiological conditions, N-WASP regulation is significantly more complex, since several proteins including WIP (WASP-interacting protein), bind to its N-terminal WH1 domain and influence its activation [20]. In fact, Cdc42 is insufficient to stimulate the native N-WASP/WIP complex [21].
EPEC pedestal formation involves activation of N-WASP by signaling initiated from the C-terminal cytoplasmic domain of its Tir protein, which is phosphorylated by host tyrosine kinases [22]–[24]. In fact, Tir is the only EPEC effector required for pedestal formation, since clustering of its ectopically expressed C-terminus in mammalian cells is sufficient to generate pedestals with high efficiency [25]. The dominant pathway for N-WASP stimulation by EPEC involves recruitment of Nck to one site of Tir tyrosine phosphorylation [reviewed in 26]–[27].
In contrast to EPEC, EHEC requires a second effector to trigger pedestal formation. This protein, EspFU (also known as TccP), localizes beneath bound bacteria after delivery into host cells, co-precipitates with Tir, and promotes phosphotyrosine- and Nck-independent actin assembly [28]–[29]. Residues 456 to 458 in the cytoplasmic C-terminus of Tir are required for EspFU recruitment and efficient pedestal formation [30]–[31], but because direct interactions between Tir and EspFU have not been detected, additional factors are assumed to mediate their association.
EspFU contains an N-terminal secretion signal followed by a C-terminus consisting of multiple nearly-identical 47-residue proline-rich peptide repeats [32]. EspFU derivatives containing these repeats bind to a segment of N-WASP encompassing the GBD [27]–[28],[32]. Mutants containing the EspFU N-terminus and as few as two repeats were shown to be capable of stimulating actin assembly using purified N-WASP and Arp2/3 in vitro and also could promote some degree of pedestal formation during infection of cultured cells, whereas derivatives containing only a single repeat did not [32].
These observations provide a framework for understanding the mechanisms by which EHEC triggers actin pedestal formation: Tir and EspFU are central to actin assembly, the Tir C-terminus is critical for recruitment of EspFU and other putative factors that may contribute to actin nucleation, and multiple proline-rich repeats of EspFU are required for maximal signaling. However, important questions remain unanswered. For example, the potential roles for EHEC effectors other than Tir and EspFU during actin assembly have not been defined, nor have the precise roles of the Tir C-terminus and/or the putative factors that mediate Tir-EspFU interactions. In addition, whereas EspFU can bind and activate purified recombinant N-WASP derivatives in vitro, it is unknown how accurately these assays might reflect N-WASP-stimulating activities in the context of the complex intracellular milieu, where other N-WASP binding partners like WIP likely modulate its autoinhibited state.
In the current study, we show that Tir and EspFU are sufficient to trigger pedestal formation in the absence of any other EHEC factors. By analyzing EspFU derivatives for the ability to bind native and recombinant N-WASP and stimulate N-WASP/WIP-mediated actin assembly in vitro and pedestal formation in cells, we arrive at a model in which the critical function of Tir is to promote clustering of the C-terminal repeats of EspFU. These repeats, in turn, activate N-WASP synergistically and lead to the formation of an Arp2/3-containing multi-protein complex that promotes robust actin polymerization.
During infection, EHEC is capable of translocating more than 30 effector proteins into the host cell [9]. At least two of these effectors, EspFU and EspF, are known to directly activate N-WASP [28]–[29],[33], while additional proteins stimulate signaling pathways that may also lead to N-WASP-mediated actin polymerization [34]. However, only two effectors, Tir and EspFU, have been shown to be crucial for EHEC pedestal formation in genetic deletion studies. Moreover, KC12, an EPEC strain in which EPEC tir has been replaced by EHEC tir, is defective at actin pedestal formation, whereas expression of EspFU by KC12 allows pedestal formation at high efficiency in manner indistinguishable from that of EHEC [28]. These results are consistent with the possibility that Tir and EspFU are the only effectors essential for EHEC-mediated actin pedestal formation. To definitively test whether EspFU is the only effector in addition to Tir that is required for actin pedestal assembly, mammalian cells were transfected with plasmids encoding derivatives of EHEC Tir and EspFU in the absence of all other bacterial factors. Tir was N-terminally tagged with an HA epitope, and its first transmembrane domain replaced with the transmembrane segment of the Newcastle Disease Virus HN surface protein to promote efficient plasma membrane localization [31]. EspFU was fused to GFP at its N-terminus and a 5myc tag at its C-terminus to allow detection by fluorescence microscopy and immunoblotting, respectively (Figure 1A). In addition to full-length EspFU, we also generated GFP-fusions carrying only its N-terminal 88 residues (EspFU-N) or its C-terminal repeats (EspFU-R1-6) (Figure 1B). To cluster Tir in the plasma membrane, transfected cells were treated with a non-pathogenic strain of E. coli that was engineered to express intimin (Figure 1A), and binds selectively to Tir-expressing cells [31].
To evaluate actin pedestal formation on cells additionally co-expressing EspFU, only those cells exhibiting GFP fluorescence were examined. Adherent bacteria were identified by DAPI-staining and F-actin was visualized using fluorescent phalloidin. Bacteria that bound to cells co-expressing Tir and full-length EspFU were associated with robust localized actin assembly, indicating that clustering of Tir in the presence of EspFU is sufficient to trigger pedestal formation (Figure 1C). In addition, pedestals were formed on cells expressing the C-terminus of EspFU, but not cells expressing the N-terminus, indicating that the activity of EspFU in pedestal formation resides entirely within the repeat region.
While deletion of the N-terminal cytoplasmic domain of Tir has a modest effect on actin assembly [31], the C-terminus of EHEC Tir contains a tripeptide sequence that is critical for both recruitment of EspFU and pedestal formation [30]. Given that Tir and EspFU do not appear to bind one another directly [28]–[32], and that EspFU can activate N-WASP, the simplest model for pedestal formation is that the C-terminus of Tir serves to recruit a host factor that mediates the Tir-EspFU interaction, thereby indirectly clustering EspFU beneath the plasma membrane. If the major role of such a host protein is to act as an adaptor between Tir and EspFU, then artificial clustering of EspFU at the plasma membrane should bypass the requirement for the C-terminus of Tir during pedestal formation. To test this possibility, we replaced the C-terminal cytoplasmic domain of HN-Tir with the C-terminal repeats of EspFU, expressed this Tir-EspFU fusion in mammalian cells, and clustered it at the plasma membrane using antibodies directed against the extracellular region of Tir and formalin-fixed Staphylococcus aureus particles to engage Tir-bound antibodies (Figure 2A). We found that clustering of HN-Tir-EspFU-R1-6, which encompasses all 6 repeats, resulted in high levels of actin pedestal formation. Measurement of the fraction of cells harboring five or more pedestals yielded a pedestal “index” of nearly 90% (Figure 2B and 2C). As expected, the control HN-Tir-EspFU-N fusion did not trigger significant actin assembly.
The number of repeats present in the C-terminus of EspFU from different EHEC isolates varies from two to eight, and an espFU mutant of the prototype strain EDL933 could not be complemented for pedestal formation by a plasmid encoding a truncation with only one repeat [32],[35]. To determine whether the quantity of repeats affects the efficiency of pedestal formation, we assessed actin assembly upon clustering of Tir-EspFU chimeras containing variable numbers of EspFu repeats. All of the derivatives, including HN-Tir-EspFU-R1, a fusion containing only a single repeat, were capable of stimulating actin assembly (Figure 2B and 2C). Quantitation of the pedestal index revealed that HN-Tir-EspFU-R1 was roughly half as efficient at generating pedestals as the fusions containing multiple repeats (Figure 2C). These results suggest a remarkably simple model in which the central function of Tir is to promote clustering of the peptide repeats of EspFU beneath the plasma membrane. In fact, clustering of a single repeat is sufficient to trigger actin pedestal formation.
We next sought to further investigate the activity of EspFU derivatives harboring varying numbers of repeats in experimental systems involving full-length Tir instead of Tir-EspFU chimeras. To avoid potential variability in the efficiency of type III translocation of different EspFU derivatives, we expressed GFP-EspFU truncations in mammalian cells, and tested their ability to complement KC12, the equivalent of an EHECΔespFU mutant [28], for actin pedestal formation. As expected [36], KC12 generated pedestals on cells expressing GFP-EspFU-R1-6, which carries the entire repeat region (Figure 3A). Moreover, all of the EspFU derivatives that contained at least two repeats were also capable of complementing KC12 for pedestal formation. However, KC12 did not generate pedestals efficiently on GFP-EspFU-R1-expressing cells. Only 1–4 pedestals were occasionally observed on cells transfected with this variant (data not shown). This discrepancy in actin assembly cannot be attributed to differences in protein levels, because immunoblotting revealed that GFP-EspFU-R1 was well expressed in mammalian cells (data not shown).
To independently assess pedestal formation in the absence of other E. coli effectors that might compete with Tir or EspFU for host target proteins, we also utilized the co-transfection approach described in Figure 1. HN-Tir along with GFP-EspFU fusions carrying different numbers of repeats were expressed in mammalian cells and clustered by treating cells with intimin-expressing bacteria. When Tir was co-expressed with EspFU constructs harboring multiple repeats, intensely staining F-actin pedestals were formed in high numbers (Figure 3B). On cells expressing HN-Tir and GFP-EspFU-R1, few bacteria were associated with intense pedestals, whereas most were associated with weakly staining F-actin (arrowheads). These results indicate that although EspFU derivatives containing a single repeat can function for pedestal formation after clustering of full-length Tir in mammalian cells, the presence of 2–6 repeats is required for robust actin assembly.
Since actin pedestal formation varied depending on the number of EspFU repeats, we next examined whether the efficiency of pedestal formation correlates with the ability of the different EspFU truncations to interact with N-WASP. To characterize the interaction between N-WASP and derivatives of EspFU harboring different numbers of repeats in a complex environment reflective of the mammalian cytosol, we generated protein extracts from pig brains, which are rich in actin cytoskeleton-associated factors, including N-WASP, WIP, and the Arp2/3 complex (data not shown). We also purified recombinant N-terminally His10- and C-terminally 5myc-tagged EspFU derivatives consisting of 1 to 6 repeats, or as a negative control, an N-terminal fragment of EspFU (Figure 4A). Magnetic beads saturated with these EspFU truncations were then incubated with brain extracts, and N-WASP that bound to the beads was detected by immunoblotting. EspFU-R1-6, the derivative containing the entire C-terminus of EspFU, pulled down N-WASP, whereas the EspFU-N control fragment did not (Figure 4B). The efficiency of N-WASP pulldown appeared to be slightly lower for variants containing fewer than four repeats, whereas mutants containing only a single repeat bound N-WASP at substantially decreased levels (R1-6 = R1-4>R1-3 = R1-2≫R1 = R6).
Since it is possible that saturating beads with a single EspFU repeat may artificially mimic derivatives with multiple repeats due to the density of protein clustered on the beads, we also assessed the interaction between EspFU and native N-WASP under conditions where EspFU concentrations could be more stringently controlled. Therefore, His-EspFU derivatives were first added to extract at a defined concentration (1 µM), and then collected using cobalt-chelated magnetic particles. Immunoblotting revealed that N-WASP interacted efficiently with EspFU derivatives containing 6 or 4 repeats (Figure 4C). Slightly lower levels of N-WASP bound to EspFU proteins harboring 3 or 2 repeats, and constructs containing a single repeat unit pulled-down N-WASP at barely-detectable levels (R1-6 = R1-4>R1-3>R1-2>>>R1 = R6). Interestingly, increasing the concentration of EspFU truncations had little effect on the quantity of N-WASP pulled-down by those proteins. Thus, although one repeat is capable of binding N-WASP in brain extracts, stepwise increases in binding are associated with the presence of additional EspFU repeats within the same molecule, with an apparent plateau at 4–6 repeats.
EspFU derivatives containing the N-terminal secretion domain and two or more repeats have been shown to stimulate N-WASP-Arp2/3–mediated actin assembly using purified proteins in vitro, while a truncation containing a single repeat barely triggered assembly [32]. To similarly assess the ability of EspFU to promote actin polymerization, but in the presence of native components, we examined the ability of recombinant EspFU variants to stimulate actin polymerization in brain extract. Purified EspFU derivatives were added to extract supplemented with pyrene-labeled actin, and polymerization kinetics were measured fluorometrically [37]. Consistent with analyses of pedestal formation in cells, EspFU-R1-6 accelerated actin assembly in extract, while EspFU-N did not (Figure 5A). The degree of stimulation by EspFU-R1-6 was dose dependent, and peaked at 20 nM (Figure 5B), although polymerization rates were slower than those typically observed in a completely purified experimental system, presumably due to the presence of inhibitory factors in the extract. Concentrations above 20 nM did not increase the rate of assembly, and levels greater than 50 nM actually resulted in slower kinetics (not shown), implying that polymerization in extract is refractory in the presence of high levels of EspFU.
To characterize the relationship between the number of EspFU repeats and the ability to stimulate actin assembly, pyrene-actin polymerization was also measured when the recombinant EspFU truncations were added to extract. When tested at 20 nM, the optimal concentration for actin assembly mediated by the EspFU derivative containing all 6 repeats, EspFU variants containing 2, 3, or 4 repeats accelerated actin assembly, and the rate of polymerization positively correlated with the number of repeats that were present (Figure 5C). In contrast, neither EspFU-R1 nor EspFU-R6, the truncations containing a single repeat, caused a significant acceleration of actin assembly in this assay system. Notably, the kinetics of actin assembly in these studies correlated with the ability of the EspFU constructs to interact with N-WASP in the extract (Figure 4). Consistent with such results, increasing the concentration of EspFU variants harboring less than 3 repeats did not enhance actin assembly (data not shown). Collectively, these data indicate that the presence of multiple repeats within the same polypeptide is important for triggering actin polymerization in an extract system designed to mimic a complex native environment.
Under physiological conditions, N-WASP stably associates with WIP or it homologs CR16 and WIRE/WICH [21],[38]. In addition, WIP is capable of inhibiting Cdc42-mediated activation of N-WASP [20] and, unlike recombinant N-WASP, the native N-WASP/WIP complex is insensitive to Cdc42 treatment [21], suggesting that regulation of N-WASP in isolation may not accurately reflect physiological N-WASP regulation in vivo. We therefore examined whether EspFU can trigger Arp2/3-mediated actin assembly in the presence of an N-WASP/WIP complex. We purified a recombinant complex containing N-WASP and WIP at 1∶1 stoichiometry and also contained a trace amount of insect cell actin (see Materials and Methods; Figure 6A). We also purified recombinant Arp2/3 complex [39], and examined the basal activity of the N-WASP/WIP complex using the pyrene-actin polymerization assay. Consistent with the predicted autoregulation of N-WASP, the recombinant N-WASP/WIP complex caused only a small increase in actin assembly kinetics when compared to Arp2/3 alone (Figure 6B).
To determine if EspFU could trigger actin assembly in this purified system, pyrene-actin polymerization was measured in the presence of recombinant EspFU-R1-6, N-WASP/WIP, and Arp2/3 complex. Consistent with analyses of pedestal formation in cells and actin assembly in extracts, EspFU-R1-6 accelerated actin assembly in this reconstituted system (Figure 6B). In the presence of 20 nM N-WASP/WIP and 20 nM Arp2/3, activation appeared to saturate at approximately 35 nM EspFU-R1-6.
We next examined the abilities of the EspFU truncations to stimulate actin assembly. By identifying the length of time each reaction took to reach half of the maximal F-actin concentration and measuring the rates of actin polymerization at these times, we directly compared the activity of each construct (Figure 6C). These experiments revealed that EspFU-R1-6 and EspFU-R1-4 were nearly indistinguishable at stimulating actin assembly, as the concentrations of these proteins that were required for reaching half of the maximal polymerization rate were approximately 4.0 nM and 5.7 nM, respectively. EspFU-R1-3 was less active than either the 6- or 4-repeat constructs at all concentrations tested, and it stimulated half-maximal assembly at a concentration of 19.4 nM. EspFU-R1-2 was substantially less active than the 3-repeat derivative (half-maximal at 78 nM), while the single repeat constructs EspFU-R1 and EspFU-R6 were even less active (half-maximal at 99–117 nM). Even when present at very high concentrations (>500 nM), the EspFU derivatives harboring 1–2 repeats could not accelerate polymerization to rates comparable to those elicited by 35 nM EspFU-R1-6 (data not shown). Thus, in the presence of purified N-WASP/WIP and Arp2/3, the number of EspFU repeats positively correlates with the rate of actin assembly.
To compare the activity of the EspFU derivatives on the basis of repeat numbers, we normalized protein concentrations to the quantity of repeat units within each polypeptide. For example, 4.17 nM of the 6-repeat construct was considered equivalent to 8.33 nM of a 3-repeat truncation and 25 nM of a single repeat. Scaling of the data in this manner revealed that EspFU derivatives containing greater numbers of repeats within the same protein had substantially higher activity than smaller constructs (Figure 6D). For example, when protein concentrations were normalized to 25 nM of repeats, the 6-repeat protein was actually twice as active as the 3-repeat protein, which was more than twice as active as the 2-repeat protein, which was roughly twice as active a 1-repeat protein. A similar trend was apparent at higher concentrations (50–200 nM), as longer EspFU constructs always had greater levels of activity than smaller derivatives containing equimolar amounts of repeats. Hence, EspFU repeats cooperate when present within the same protein to promote synergistic activation of N-WASP/WIP and Arp2/3-mediated actin assembly.
Insight into the ability of multi-repeat EspFU derivatives to pulldown N-WASP from brain extracts and stimulate N-WASP/WIP-mediated actin assembly in vitro would be enhanced by a better understanding of EspFU binding by N-WASP, which is mediated by the GBD [28]–[29]. To further test the functional significance of this interaction during pedestal formation, we transfected mammalian cells with constructs expressing Flag-tagged N-WASP variants, each encompassing a different combination of N-WASP domains (Figure 7A), and then examined pedestal formation upon infection with EHEC. Consistent with previous observations showing that N-WASP residues 226–274 within the GBD mediate recruitment of N-WASP to sites of EHEC attachment [14], all tagged N-WASP derivatives containing the GBD (residues 151–273) were recruited to sites of bacterial adherence (Figure 7B, top row). In addition, quantitation of pedestal formation on transfected cells revealed that neither the N-terminal WH1 domain nor full-length N-WASP had a substantial effect on actin polymerization, whereas overexpression of each of the 4 GBD-containing derivatives that lacked the VCA domain, which is necessary for Arp2/3 activation, effectively abolished pedestal formation (Figure 7B, bottom row). This inhibition was specific, because overexpression of the GBD did not affect actin assembly triggered by an EPEC strain that generates pedestals independently of EspFU (Figure 7C). As predicted by mapping of requirements for N-WASP recruitment to EHEC [14], an H208D point mutation that abrogates N-WASP binding by the GTPase Cdc42 had no effect on the inhibitory activity of the GBD. These results indicate that the interaction between EspFU and the C-terminal portion of the N-WASP GBD is important for pedestal formation in mammalian cells.
Next, to begin to assess whether the positive correlation between the number of repeats within an EspFU derivative and its ability to activate actin assembly is reflected in binding to the N-WASP GBD, we examined this interaction in yeast two-hybrid assays. As demonstrated previously [28], the N-WASP GBD interacted with the C-terminus of EspFU that contains the repeat sequences, but not the EspFU N-terminus (Figure 7D). In addition, all of the C-terminal subfragments of EspFU, even those containing only a single repeat, interacted with the GBD in these assays. However, the degree of interaction did not positively correlate with increasing numbers of repeats. Rather, the number of repeats in EspFU inversely correlated with the degree of expression of the lacZ reporter. Although reporter activity in the yeast two-hybrid assay reflects a number of parameters, including levels of expression and nuclear import of the binding partners, these results provide no evidence that the more efficient activation by multi-repeat EspFU derivatives is a consequence of cooperativity in GBD binding.
To further assess potential cooperativity in GBD binding by EspFU, we examined the interactions of EspFU derivatives with the GBD of WASP, the hematopoetic-specific homologue of N-WASP that has been shown to also promote actin pedestal formation in cells [14]. Recombinant EspFU proteins containing 5, 2, or 1 repeats (R′1-5, R′4-5, and R′5, respectively) were generated with N- and C-terminal repeat boundaries based on recent structural studies that defined the GBD-binding sequences of EspFU [40]–[42]. The affinity of the WASP GBD for these EspFU constructs was determined using isothermal titration calorimetry and yielded predicted differences in stoichiometry (i.e., 5.7, 1.8, and 1.0 for the 5, 2, and 1 repeat derivatives, respectively) but similar dissociation constants (i.e., 83 nM, 95 nM, and 89 nM, respectively) (Figure 8A). Thus, variation in the number of EspFU repeats is not associated with differences in affinity for the GBD.
We next tested whether oligomerization of N-WASP by adjacent EspFU repeats instead might promote binding of the Arp2/3 complex. N-WASPC, a C-terminal fragment of N-WASP that contains the GBD, PRD, and VCA domain, is a fragment previously shown to function in pedestal formation [14]. To determine whether EspFU derivatives differing in repeat number also differed in their ability to form a complex containing N-WASPC and Arp2/3, EspFU variants containing 1 or 2 repeats, fluorescently labeled at their N-termini with AlexaFluor647, were incubated with equivalent molar concentrations of N-WASPC in the absence or presence of the Arp2/3 complex. The relative size of EspFU-containing complexes, detected by absorbance at 650 nM, was then determined by gel filtration chromatography. The single EspFU repeat bound N-WASP at essentially stoichiometric levels when mixed in equimolar (2 µM) quantities, as determined by an increase in the apparent size (i.e., earlier elution profile) of EspFU (Figure 8B left, red vs. blue peaks). The addition of 2 µM (i.e., a two-fold higher molar concentration) Arp2/3 caused ∼60% of the EspFU to shift to an even more rapidly eluting fraction indicative of an EspFU-N-WASP-Arp2/3 complex (Figure 8B, yellow profile). Very little of this complex was detected using 1 µM Arp2/3 (Figure 8B left, small shoulder in gray profile).
When 1 µM N-WASPC was incubated with 0.5 µM of the 2-repeat EspFU derivative (i.e., containing a normalized repeat concentration), the two proteins bound one another at essentially stoichiometric levels (Figure 8B, right, compare light blue and green profiles). However, in contrast to the requirement for 2 µM Arp2/3 to observe appreciable complex formation with a single repeat, the addition of as little as 0.5 µM Arp2/3 to N-WASPC and the two repeat derivative resulted in the formation of an Arp2/3-containing complex (Figure 8B right, blue or pink profiles, respectively). The greater complex formation even at lower Arp2/3 concentration indicates that the ability of tandem EspFU repeats to assemble N-WASP dimers (or by implication higher order multimers) may facilitate binding of the Arp2/3 complex, providing a likely source of inter-repeat cooperativity.
Since the assay demonstrating inter-repeat cooperation of EspFU for the formation of an Arp2/3-containing complex described above involved the addition of only EspFU, N-WASPC, and Arp2/3, N-WASP-associated proteins such as WIP should not be required for cooperativity between the EspFU repeats in actin assembly. Furthermore, domains of WASP (or N-WASP) other than the GBD (which binds EspFU) and VCA (which binds Arp2/3) should also be dispensable. We therefore tested whether the EspFU repeats were capable of synergistically activating a minimized (and normally autoinhibited [40]) derivative of WASP containing only the GBD and VCA domain. Similar to results using EspFU truncations and an N-WASP/WIP complex, pyrene actin assays using WASP GBD-VCA revealed that the number of EspFU repeats correlated with increased actin polymerization rates, even when protein concentrations were normalized to repeat units (Figure 8C). These results indicate that synergy in actin assembly require no WASP sequences other than the GBD and VCA domain.
In order to better understand EHEC pedestal formation, we first sought to define the minimal set of bacterial effectors essential to this process. We found that EspFU is likely the only EHEC effector besides Tir that is required for pedestal formation, because pedestals can be induced on mammalian cells that express these two factors in the absence of any other bacterial proteins. In addition, the cytoplasmic C-terminus of Tir could be functionally replaced with the C-terminus of EspFU, indicating that the only critical function of this Tir domain is to recruit the EspFU repeats to sites of EHEC attachment. The C-terminus of the prototype EspFU protein consists of six 47-residue proline-rich repeats, and a fusion containing a single repeat unit retained significant ability to generate pedestals when clustered at the plasma membrane. This is consistent with a recent report describing actin recruitment to the plasma membrane after artificially clustering a single repeat with antibodies [42]. That study, as well as the recent report of the structure of a single EspFU repeat bound to the WASP GBD [41], revealed that EspFU binds to the autoinhibitory region within the GBD that normally interacts with the VCA domain, a finding consistent with previous mapping studies [14],[27],[32]. Therefore, the bare minimum components required for pedestal formation are the domains of Tir that facilitate its membrane localization and clustering by intimin, and a single GBD-binding EspFU peptide repeat.
Within the Tir C-terminal domain, a tripeptide sequence (NPY458) is required for recruitment of EspFU and actin pedestal formation [30], but a direct interaction between Tir and EspFU has not been detected, suggesting that they interact indirectly. The delineation of Tir and EspFU as the only bacterial effectors required for pedestal formation indicates that the putative adaptor that mediates this interaction is of host origin. Efficient association of EspFU with the putative adaptor linking it to Tir may require multiple EspFU repeats, because when Tir and EspFU were expressed separately rather than as a single fusion protein, at least two repeats were required for robust EspFU function. Consistent with these results, a translocated 2-repeat EspFU derivative, but not a 1-repeat derivative, localized to sites of bacterial attachment [32].
Aside from promoting Tir-EspFU interactions, this host factor may itself promote some level of actin assembly, because residual pedestals can form at low levels in the absence of EspFU [28]. Cortactin, which has the ability to stimulate Arp2/3 and contributes to pedestal formation via an unknown mechanism, can interact with both Tir and EspFU, but binds the Tir N-terminus [36], a domain that is largely dispensable for actin assembly [31]. Regardless of the identity of the adaptor, the finding that a hybrid protein containing Tir fused directly to the EspFU repeats is fully functional for pedestal formation indicates that neither the adaptor nor the C-terminus of Tir play any essential role in actin assembly other than to recruit EspFU. Thus, EspFU is the primary effector that signals to the actin cytoskeleton during pedestal formation.
In addition to promoting more efficient recruitment to Tir, the presence of multiple repeat units in EspFU provides more robust signaling function, because a Tir-EspFU fusion carrying two or more repeats triggered pedestal formation at levels 2-fold higher than a fusion carrying a single repeat. Recently, a full-length repeat region was shown to recruit GFP-actin to the plasma membrane somewhat (∼20%) more frequently than a single repeat [42]. That report and others have also shown that multi-repeat EspFU derivatives promote more efficient actin assembly in vitro than a single-repeat derivative [32],[42]. Such studies have relied upon analyses of recombinant N-WASP or small N-WASP fragments containing the minimal autoinhibitory module, a GBD-VCA fusion. However, in recent years it has become apparent that in the cell cytoplasm the intrinsic activity of N-WASP may be significantly modulated through interactions with other factors [20]–[21]. We therefore examined EspFU-induced actin polymerization in brain extracts, and found that in this complex environment, the number of repeats correlated both with the ability of EspFU derivatives to interact with native N-WASP and to stimulate actin assembly. Single-repeat constructs did not accelerate polymerization in these assays, possibly due to competition with endogenous N-WASP activators or the presence of inhibitory factors in the extract.
Under physiological conditions, N-WASP is stably associated with WIP, a protein that binds to its N-terminal WH1 domain and influences its activation [20]–[21]. We found that EspFU constructs are capable of potently activating Arp2/3-mediated actin assembly in the presence of a recombinant N-WASP/WIP complex, and as predicted based upon N-WASP pulldown assays and measurements of actin polymerization in extracts, EspFU derivatives containing greater numbers of repeats stimulate actin assembly more rapidly. Interestingly, when we normalized the EspFU derivatives to the concentration of repeats and quantitated actin assembly rates, we found that increasing the number of repeats within individual proteins does not simply result in additive increases in actin polymerization. Instead, the presence of multiple repeats in the same derivative enhances N-WASP/WIP-mediated actin assembly synergistically.
It is possible that this enhancement phenotype is due to cooperativity among the repeats in EspFU-N-WASP binding. Indeed, an earlier study reported that the dissociation constants (Kd) for the EspFU-N-WASP interaction varied depending on the numbers of repeats (e.g., 3.6 nM for 6 repeats, 6.4 nM for 4 repeats, and 11 nM for 2 repeats, and no measurable binding for 1 repeat) [32]. However, these values were calculated on the basis of in vitro actin assembly assays that involve the formation of an N-WASP-Arp2/3 complex, and may not simply reflect the interaction between EspFU and N-WASP. Similarly, the more efficient pulldown of N-WASP by multi-repeat EspFU derivatives in brain extracts that we observed could be influenced by other endogenous factors. Although a previous study found that 2 repeats is the minimal N-WASP-binding module within EspFU in gel overlay assays [32], we found that a single repeat bound to the GBD with high efficiency in yeast two-hybrid assays. Furthermore, the Kd for binding of purified 1-, 2-, or 5-repeat EspFU derivatives to the WASP GBD were all similar to one another (i.e., 83–95 nM), and within an order of magnitude of that measured for the N-WASP GBD (18 nM; [42]). These results provide convincing evidence that the EspFU repeats do not bind to the GBD in a cooperative fashion.
An alternative explanation for inter-repeat cooperativity during actin polymerization is that an EspFU-N-WASP-Arp2/3 complex can be formed more efficiently. The major threshold for pedestal-forming function apparently resides within 2 EspFU repeats, and we were unable to detect substantial differences in the frequency or intensity of actin pedestal formation promoted by EspFU proteins harboring 2, 3, 4, or 6 repeats (although in vitro actin assembly assays indicate that the number of repeats present in an EspFU derivative correlates positively with function). Similarly, analysis of a diverse collection of espFU-containing E. coli strains showed that all EspFU proteins contain at least 2 repeats [35]. Therefore, we compared the ability of EspFU derivatives containing 1 versus 2 repeats to enter into a complex with recombinant N-WASP and Arp2/3 complex. As predicted from the affinities of EspFU fragments for the GBD, the single and two repeat derivatives bound to N-WASP indistinguishably. However, the 2-repeat protein formed a complex with both N-WASP and Arp2/3 with much greater efficiency. These experiments suggest that N-WASP multimers assembled by tandem EspFU repeats may have higher affinity for Arp2/3 complex. As predicted from such a model, the inter-repeat cooperativity was revealed during in vitro actin assembly assays using a WASP derivative containing only the EspFU-binding GBD and the Arp2/3-binding VCA domain. Further study will be required to illuminate the molecular basis of this enhanced interaction.
Interestingly, the multivalency that is clearly a critical property of EspFU is likely an important feature of other Arp2/3-mediated processes that are triggered by microbial pathogens. Pedestal formation by EPEC, which occurs in an EspFU-independent manner, requires clustering of Tir, a step that may mimic multivalent binding. The critical EPEC Tir phosphopeptide has no activity when added in soluble form in standard pyrene actin assays [S. Rankin, unpub. obs.], but potently stimulates actin assembly when clustered on a bead [25]. EspF, an EHEC effector that is 35% similar to EspFU but not involved in pedestal formation [28], also uses peptide repeats to activate N-WASP in vitro [33]. Finally, the Shigella IcsA (VirG) protein uses repetitive sequences to bind N-WASP and promote actin-based intracellular motility [43]. Future investigation into how the multivalent nature of EspFU promotes actin assembly is likely to provide insight into related phenomena central to multiple pathogenic processes.
The EHEC dam mutant used in this study was derived from TUV93-0, a Shiga toxin-deficient version of the prototype O157:H7 strain EDL933 [44]. The EPEC bacterium was the prototype O127:H6 strain, JPN15/pMAR7. EPEC KC12, which contains the EPEC tir-cesT-eae operon replaced with the corresponding EHEC operon, has also been described [45]. A non-pathogenic strain of E. coli (MC1061) expressing EHEC intimin was also described elsewhere [31]. For yeast two-hybrid analyses, espFU derivatives were generated by PCR from EDL933 genomic DNA and cloned into the EcoRI and BamHI sites of pGAD424 [28] to create fusions between an N-terminal GAL4AD and C-terminal 5myc tag. For transfections, espFU fragments were subcloned into the KpnI and BamHI sites of pKC425 [46] to generate fusions between an N-terminal GFP-tag and C-terminal 5myc tag. For protein expression and purification, most espFU derivatives were subcloned into the NdeI and XbaI sites of the vector pET16b (Novagen) to create fusions between an N-terminal His10-tag and C-terminal 5myc tag. Some EspFU constructs contained only an N-terminal His-tag (Figure 8) [41]. For Tir-EspFU hybrids, espFU-myc derivatives were subcloned into the KpnI and XbaI sites of the HN-Tir-related vector pKC689 [31]. EspFU truncations contained amino acids 1-88 (N), 80-384 (R1-6), 80-275 (R1-4), 80-228 (R1-3), 80-181 (R1-2), 80-134 (R1), 229-384 (R4-6), 276-384 (R5-6), 323-384 (R6), 80-314 (R′1-5), 221-314 (R′4-5), and 268-314 (R′5). For dominant negative transfections, rat N-WASP derivatives were generated by PCR with an N-terminal Flag-tag using domain boundaries described previously [28] and cloned into the KpnI-EcoRI sites of the vector pCDNA3 (Invitrogen). N-terminally GFP-tagged derivatives of the GBD were generated by subcloning into pKC425. Plasmids for expression of membrane-targeted HN-TirFL in mammalian cells and expression of the N-WASP GBD fused to the LexA DBD in pBTM116 have been described previously [28],[31]. Plasmids for expression of His-Flag-N-WASP and His-Myc-WIP in insect cells are described elsewhere (ADS and MDW, submitted). N-terminally His- or GST-tagged WASP GBD (residues 242–310), N-WASPC (residues 193-501) and WASP GBD-VCA (residues 230–310 and 420–502 with a GGSGGS linker) constructs are described elsewhere [41]. For routine passage, all E. coli strains were grown in LB media at 37°C. Prior to infections, EHEC was cultured in DMEM+100 mM HEPES pH 7.4 in 5% CO2 to enhance type III secretion. HeLa, Cos7, and murine fibroblast-like cells (FLCs) [47] were used interchangeably and cultured in DMEM+10% FBS at 37°C in 5% CO2.
All transfections were performed as described previously [31]. Infections for 3 h with non-pathogenic E. coli expressing intimin [31] and EHEC or EPEC strains [44]–[45] have also been described. To cluster Tir, cells were treated with antibodies that recognize its extracellular domain prior to the addition of formalin-fixed S. aureus Pansorbin particles (Calbiochem) [25].
Infected cells were fixed in 2.5% paraformaldehyde for 35 minutes and permeabilized with 0.1% Triton-X-100 in PBS as described previously [45]. Bacteria were identified with 1 µg/ml DAPI (Sigma), and F-actin was detected using 4 U/ml Alexa568-phalloidin (Molecular Probes). To visualize N-WASP derivatives, cells were treated with an anti-Flag M5 antibody (Sigma) and Alexa488 goat anti-mouse antibodies (Molecular Probes). To visualize HN-Tir derivatives, cells were treated with an HA.11 antibody (Covance) and Alexa488 or Alexa350 goat anti-mouse antibodies. To quantify the pedestal formation index in cells expressing high levels of N-WASP derivatives, which were identified by bright anti-Flag or GFP fluorescence, the percentage of cells harboring at least 10 adherent bacteria and 5 actin pedestals was measured. To quantify the pedestal index in cells infected with KC12, the percentage of cells harboring at least 10 adherent bacteria and 5 actin pedestals was measured. To quantify the pedestal index in cells treated with pansorbin particles, the percentage of HA-fluorescing cells harboring at least 10 adherent particles (the latter identified by virtue of their ability to bind fluorescently labeled secondary antibodies) and 5 actin pedestals was measured. At least 50 cells were examined per sample per experiment. Our previous work suggests that scoring of a pedestal index accurately reflects similar quantification methods that measure the fraction of bound bacteria that generate pedestals [25]. Cells expressing extremely high fluorescence levels of EspFU were refractory to pedestal formation and were not included in these analyses. All scalebars are 1 µm in length.
For yeast two-hybrid assays. EspFU variants consisting of different combinations of the 6 repeats were fused to the Gal4 transcriptional activation domain, whereas the N-WASP GBD was fused to the LexA DNA-binding domain. Pairwise combinations of these fusion proteins were tested for interactions by measuring activation of a lacZ reporter, as described [28]. For some pulldown assays (Figure 4B), cobalt-chelate conjugated magnetic Talon particles (Invitrogen) were saturated with His-EspFU-myc derivatives by incubating them with 75 µg/ml of each recombinant fragment for 1 h. After removal of unbound EspFU, beads were incubated with brain extract for 1 h in 50 mM NaPO4 pH 7.4, 150 mM NaCl, and 0.015% Triton X-100. Bound proteins were eluted by boiling in SDS-PAGE sample buffer. For other pulldown assays (Figure 4C), His-EspFU-myc derivatives were incubated in brain extract at specific concentrations for 0.5 h and collected using an excess of Talon particles. Bound proteins were again eluted by boiling in SDS-PAGE sample buffer.
To prepare mammalian cell lysates, transfected cells were collected in PBS+2 mM EDTA and lysed in 50 mM Tris-HCl, pH 7.6, 50 mM NaCl, 1% Triton X-100, 1 mM Na3VO4, 1 mM PMSF, and 10 µg/ml each of aprotinin, leupeptin, pepstatin, and chymostatin (Sigma)), prior to mixing with SDS-PAGE sample buffer. Protein samples were boiled for 10 minutes, centrifuged, and analyzed by 10% SDS-PAGE prior to staining with Coomassie blue or transferring to nitrocellulose membranes and staining with Ponceau S. Membranes were blocked in PBS+5% milk (PBSM) before probing with N-WASP or GFP antibodies, as described previously [28]. Following washes, membranes were treated with secondary antibodies conjugated to alkaline phosphatase or horseradish peroxidase and developed using BCIP/NBT [45] or enhanced chemiluminescence [48].
His-EspFU-myc fusion proteins were expressed in E. coli BL21-Rosetta (Novagen) at 37°C in the presence of 0.1 mM IPTG for 3 h. Bacteria were lysed in 10 mM Tris pH 7.4, 150 mM NaCl, and protease inhibitor cocktail (Roche) using a cell disruptor (30kpsi, Constant Cell Systems). EspFU was purified first using His-tag affinity for Ni-NTA-agarose beads (Qiagen) and eluted in lysis buffer containing 400 mM imidazole. To remove EspFU degradation products, an anion exchange purification step, facilitated by the negatively charged 5myc-tag, was performed. EspFU was bound to a HiTrap Q column (GE Healthcare) and eluted in 10 mM Tris pH 9.5 containing 1 M NaCl. Other EspFU constructs (R′5, R′4-5, R1-5; Figure 8) with just an N-terminal His-tag were expressed in E. coli BL21-DE3 at 20°C in the presence of 1 mM IPTG for 16 h, and after Ni-NTA affinity purification the tag was cleaved with thrombin. Cleaved proteins were subjected to ion exchange and gel filtration chromatography to remove the tag and other impurities. Protein concentrations were estimated by Bradford assay (Bio-Rad).
Brains obtained from freshly slaughtered pigs (Dalehead Foods Ltd, Linton, UK) were cleaned, sectioned, and resuspended in extraction buffer (0.1 M MES pH 6.8, 1 mM EGTA, 0.5 mM MgCl2, 0.1 mM EDTA, 1 mM DTT, protease inhibitor cocktail), prior to homogenization using a Waring blender (2×15 s, 4°C). After clarification (7,000 g, 20 min, 4°C), extracts were filtered through cheesecloth and further clarified (11,000 g, 40 min, 4°C) prior to storage (−80°C). Extract to be subfractionated (60 ml) was dialysed against 2×5 L 20 mM Tris-Cl pH8, 2 mM MgCl2, 5 mM EGTA, 1 mM EDTA, 0.5 mM DTT, 0.2 mM ATP, 2.5% glycerol at 4°C, prior to loading onto tandem HiTrap Heparin columns (2×5 ml), pre-equilibrated with dialysis buffer. Bound proteins were eluted using a 0–0.5 M KCl gradient in dialysis buffer (2 ml/min) and 20 3 ml fractions collected. Fractions 11–17 contained both Arp2/3 complex and N-WASP upon immunoblotting, and were combined, dialysed (as above), concentrated 5-fold (Centricon), and stored at −80°C.
His-tagged human N-WASP and WIP were expressed using recombinant baculovirus-infected High5 insect cells, as described in detail elsewhere (ADS and MDW, submitted). A recombinant N-WASP/WIP complex was purified by Nickel affinity chromatography followed by gel filtration chromatography to separate the complex from N-WASP or WIP alone. To maintain an autoinhibited N-WASP/WIP complex, freeze-thaw cycles and prolonged periods of storage on ice were kept to a minimum. Recombinant human Arp2/3 complex and native bovine Arp2/3 complex were purified as described previously [39],[41].
Assays utilizing brain extract contained samples supplemented with 2.5 µM skeletal muscle actin (10% pyrene-labeled), and polymerization was measured as described previously [49]. Assays using N-WASP/WIP complex contained 2.0 µM actin (7% pyrene-labeled) and 20 nM recombinant Arp2/3 complex, and polymerization was measured as described previously [48]. Assays using the N-WASP GBD-VCA contained 4.0 µM actin (5% pyrene-labeled) and 10 nM bovine Arp2/3 complex, and polymerization was measured as described elsewhere [41].
Isothermal titration calorimetry was performed as described elsewhere [41]. Briefly, the WASP GBD was titrated into R′5, R′4-5, or R′1-5 in KMEI buffer (50 mM KCl, 1 mM MgCl2, 1 mM EGTA, and 10 mM imidazole pH 7.0) plus 5 mM β-mercaptoethanol. Prior to gel filtration analyses, the N-termini of EspFU fragments R′5 and R′4-5 were fluorescently labeled by dialysis against 100 mM sodium bicarbonate pH 8.3 and treatment with a five-fold molar excess of AlexaFluor647-carboxylic acid, succinimidyl ester (Invitrogen) at 4°C for 16 hours. Extra dye was removed by desalting chromatography and subsequent dialysis. Conjugation was confirmed by mass spectrometry. Labeling efficiency was estimated as >97% by the ratio of absorbance at 280 and 650 nm. Interactions between Arp2/3 complex and N-WASPC in complex with EspFU were examined using a Superdex 200 10/300 GL column (GE Healthcare) equilibrated in KMEI plus 1 mM dithiothreitol buffer.
|
10.1371/journal.pgen.1006269 | HAPLESS13-Mediated Trafficking of STRUBBELIG Is Critical for Ovule Development in Arabidopsis | Planar morphogenesis, a distinct feature of multicellular organisms, is crucial for the development of ovule, progenitor of seeds. Both receptor-like kinases (RLKs) such as STRUBBELIG (SUB) and auxin gradient mediated by PIN-FORMED1 (PIN1) play instructive roles in this process. Fine-tuned intercellular communications between different cell layers during ovule development demands dynamic membrane distribution of these cell-surface proteins, presumably through vesicle-mediated sorting. However, the way it’s achieved and the trafficking routes involved are obscure. We report that HAPLESS13 (HAP13)-mediated trafficking of SUB is critical for ovule development. HAP13 encodes the μ subunit of adaptor protein 1 (AP1) that mediates protein sorting at the trans-Golgi network/early endosome (TGN/EE). The HAP13 mutant, hap13-1, is defective in outer integument growth, resulting in exposed nucellus accompanied with impaired pollen tube guidance and reception. SUB is mis-targeted in hap13-1. However, unlike that of PIN2, the distribution of PIN1 is independent of HAP13. Genetic interference of exocytic trafficking at the TGN/EE by specifically downregulating HAP13 phenocopied the defects of hap13-1 in SUB targeting and ovule development, supporting a key role of sporophytically expressed SUB in instructing female gametogenesis.
| Ovules, being the progenitors of seeds, hold an important role in crop production. Ovule is a complex organ connecting two generations, i.e. diploid sporophytic cells and haploid female gametophytes (FG). Fertilization of female gametes leads to the development of embryo and endosperm while the sporophytic cells of ovules become seed coats. Because of its importance for plant reproduction, the development of ovules is tightly controlled. The integument cells are divided in a polar way so that integument cell layers fully enclose a mature FG, leaving an opening only for pollen tube entrance. The integuments also instruct FG development, synchronizing the two generations. Receptor-like kinases, especially STRUBBELIG (SUB), are crucial both for the establishment of ovular polarity and the intercellular communication between sporophytic and gametophytic cells. However, how SUB is post-translationally regulated, as would be expected for cell-surface proteins, is unclear. We report here that the dynamic trafficking of SUB via a subpopulation of endosomes is crucial for its functionality during ovule development. The membrane association of SUB was impaired when the recycling route from the trans-Golgi network/early endosome to the plasma membrane was disrupted in hap13-1, a mutant of HAPLESS13. We further demonstrate that defects of SUB targeting in the outer integuments compromised FG development, suggesting an intriguing intercellular communication. Our results highlight the importance of vesicular trafficking in the establishment of ovular polarity and in the communications between spatially separated cells.
| Tissue or planar polarity that distinguishes multicellular organisms from unicellular ones requires coordinated cell behavior within a tissue or the plane of a tissue layer through intercellular communications and intracellular asymmetry [1,2]. Ovule development in angiosperms manifests the complexity by establishing and maintaining polarity in the proximal-distal axis, along which three morphologically distinct units are observed: the nucellus at the distal end harbors the megaspore mother cell (MMC) that undergoes meiosis to form an embryo sac, i.e. the female gametophyte (FG); from the chalaza at the central region, outer and inner integuments initiate growth, eventually enveloping the nucellus; the proximal funiculus connects the ovule to the placenta [1,3–6]. Fine-tuned intercellular communications are critical for primordium initiation, pattern formation, planar morphogenesis and female gametogenesis, collectively leading to ovule development [1,3,7].
Intercellular communication in plant development is mainly controlled by the phytohormone auxin as well as receptor-like kinases (RLKs) [2,8,9]. Auxin has been attributed as an instructive factor in the establishment and maintenance of tissue or planar polarity, through its asymmetric gradient [2,8,10,11]. Auxin gradient or maximum is a result of local biosynthesis, metabolism, and most importantly, transport facilitated by auxin influx and efflux carriers [8,10,11]. PIN-FORMED (PIN), the auxin efflux carriers, is encoded by 8 genes in Arabidopsis (PIN1-PIN8), whose dynamic distribution, often asymmetric, is essential for auxin gradient [8,10,11]. Among the eight PINs, PIN1 was reported to be critical for ovule development [8,12,13]. Tissue or planar polarity is also regulated by cellular perception and responses of neighboring cells through RLKs, whose extracellular domains enable signal perception while cytoplasmic domains allow intracellular signal relay [9]. Mutations at a few RLK-coding genes, such as STRUBBELIG (SUB), ARABIDOPSIS CRINKLY4 (ACR4), and ERECTA-family genes (ER and ERLs), resulted in defective ovular development [14–16].
The membrane distribution of PINs and RLKs is dynamically regulated by vesicle trafficking-mediated protein sorting [11,17–19]. In the past decade, extensive studies revealed the key role of vesicle trafficking on the dynamic asymmetry of PINs [11,17,19]. RLKs are another dominant class of transmembrane (TM) proteins whose dynamic distribution through vesicles has been under close scrutiny [18]. The internalization of RLKs could be constitutive or induced whereas the internalized RLKs may recycle back to the PM non-selectively or selectively, as well as go to vacuoles for degradation [18]. As a consequence of their differential distributions, intracellular signaling is enhanced, reduced, or switched [18]. Therefore, vesicle trafficking-mediated protein sorting is fundamental to development. However, little is known on vesicle trafficking during ovule development, the trafficking routes involved, the key proteins vesicle-sorted, as well as the regulators of protein sorting.
An exocytic or anterograde trafficking route and an endocytic or retrograde trafficking route operate in plant cells to mediate protein sorting within the endomembrane system [20–22]. The two routes converge on the trans-Golgi network/early endosome (TGN/EE) [23], a likely heterogeneous group of endosomes equal to the combination of metazoan sorting endosomes, early endosomes, and recycling endosomes [21,22]. Protein sorting within the endomembrane system requires adaptor protein (AP) complexes, i.e. tetrameric protein complexes for cargo selection and coat recruitment [20,22,24]. Among the five AP complexes in eukaryotes, AP1 is associated with the TGN/EE and thus plays essential roles in protein sorting in plant cells [25–27]. AP1 consists of two large subunits (γ, β), one medium subunit (μ), and a small subunit (σ) [20,22,24]. We and others previously reported the identification of Arabidopsis AP1μ, HAPLESS13 (HAP13)/AP1MU2 [25–27]. Mutations of HAP13 severely reduced male gametophytic transmission [25,27,28] whereas its homozygous mutants showed growth retardation [27]. Key TM proteins such as PIN2 [25,27], BRI1 [27], and KNOLLE [26] were mis-localized in the mutant of HAP13, hap13-1, demonstrating the key role of HAP13 in TGN/EE-centered protein sorting.
We report here that HAP13-mediated protein sorting at the TGN/EE is critical for ovule development, specifically the asymmetric growth of outer integuments. By scanning electron microscopy (SEM), plastic section, and confocal laser scanning microscopy (CLSM), we show that hap13-1 is defective in outer integument growth, leading to exposed or protruding nucellus. The defect in outer integuments severely affected female gametogenesis such that a large portion of hap13-1 ovules did not contain a well-patterned embryo sac. Consequently, pollen tube guidance and reception were impaired, resulting in reduced female fertility. We further show that the key RLK for ovule development, SUB, was mis-targeted in hap13-1, correlating with the phenotypic similarity in ovules between hap13-1 and SUB mutants. However, unlike PIN2 whose membrane distribution relies on HAP13 [25,27], the asymmetric targeting of PIN1 at the nucellus was intact in hap13-1, suggesting distinct sorting mechanisms for different PINs. Downregulating HAP13 expression by RNA interference (RNAi) specifically in outer integuments phenocopied the ovular defects in hap13-1. Results presented here demonstrate the importance of TGN/EE-centered protein sorting in outer integument growth and provide clues for a deep understanding of molecular mechanisms underlying the establishment and maintenance of ovule polarity.
HAP13 was expressed in various tissues and developmental stages [27]. To determine whether HAP13 was expressed during ovule development, we generated a nuclear-localized YFP reporter line for HAP13 (ProHAP13:NLS-YFP). Confocal laser scanning microscopy (CLSM) revealed that HAP13 was expressed in all cell layers during ovule development (Fig 1A and 1B). Finally, the expression of HAP13 in ovules during development was confirmed by examining the HAP13g-GFP hap13-1 transgenic plants (Fig 1C–1F). The expression of HAP13 in ovules from primordium formation till maturation hinted at the possibility of its function in ovule development.
To determine whether HAP13 was required for ovule development, we examined hap13-1 pistils by SEM and CLSM analyses. Ovules are connected with the gynoecium through the funiculus (Fig 2A and 2B). At the distal end of the funiculus is a cleft, i.e. micropyle, formed by inner and outer integument cells enveloping the seven-celled embryo sac in wild type (Fig 2A–2D). In comparison, around 80% of hap13-1 ovules (Fig 2H) were abnormal such that the surface cell layers were broken and thus only partially wrapping the inner structure (Fig 2B and 2D). At maturity, wild-type ovules contained well-patterned FGs with a central cell, an egg cell and two synergid cells (Fig 2E). By contrast, FGs were abnormal in over 80% of hap13-1 ovules (Fig 2F and 2G). In addition, due to incomplete growth of outer integuments, embryo sacs in hap13-1 ovules were exposed or protruding (Fig 2B and 2D).
Pollination of emasculated hap13-1 pistils with wild-type pollen only yields a few seeds although female transmission of hap13-1 was normal [27]. To determine the cause of the reduced female fertility, we performed aniline blue staining of emasculated wild-type or hap13-1 pistils hand-pollinated with wild-type pollen at different time points. At 4 hours after pollination (HAP), pollen tubes already entered the transmitting track in wild-type pistils but only exited the style in hap13-1 pistils (Fig 3A). It suggested that hap13-1 pistils were compromised in the support of pollen tube growth. At 9 HAP, pollen tubes reached the bottom of both wild-type and hap13-1 pistils (Fig 3A). However, hap13-1 pistils were much shorter than those of wild type due to a reduced number of ovules (Fig 3A). Therefore, pollen tube growth inside hap13-1 pistils was substantially slow. At 48 HAP when wild-type pistils contained enlarged ovules indicative of successful fertilization (Fig 3A and 3B), only few ovules were enlarged in hap13-1 pistils (Fig 3A and 3D). Most hap13-1 ovules were not targeted by pollen tubes (163 out of 213, Fig 3C), indicating compromised pollen tube guidance. Another portion of hap13-1 ovules did attract pollen tubes (26 out of 213, Fig 3E). However, these ovules failed to instruct the cessation of pollen tube growth, resulting in overgrown pollen tubes inside embryo sacs and failed fertilization (Fig 3E).
To determine the reason of reduced pollen tube guidance in hap13-1, we performed immunofluorescence staining on wild-type and hap13-1 ovules using an anti-LURE antibody [29]. Because LURE is a key peptide secreted by synergid cells to attract pollen tubes [30,31], we wanted to test whether hap13-1 ovules were compromised in the production of LURE. In mature wild-type ovules, strong signals were detected in the micropylar region (Fig 3F), indicative of LURE secretion. By contrast, hap13-1 ovules (45 out of 57 examined) showed a substantial reduction of fluorescence signals (Fig 3G). This result explained the reduced pollen tube guidance of hap13-1 ovules. Next, we introduced an egg cell-specific reporter ProDD45:GUS [32] into hap13-1. By histochemical analysis of the homozygous transgene ProDD45:GUS in the heterozygous and homozygous hap13-1, we found that egg cells were present in almost all ovules of the heterozygous hap13-1 pistils, comparable to those of wild type (S1 Fig). However, only a few ovules of the homozygous hap13-1 pistils showed GUS signals (S1 Fig). These results demonstrated that defective patterning of hap13-1 FGs was caused by defects in surrounding sporophytic tissues.
To determine at which stage hap13-1 was defective during ovule development, we performed semi-thin plastic sections and CLSM on ovules at different developmental stages. The phenotype of hap13-1 in ovule development was variable, ranging from severely affected to normal and fertile ovules. Seventy percent of ovules from hap13-1 mutants displayed developmental irregularities (N = 256 of hap13-1 ovules for percentage calculation). Developmental defects of hap13-1 started at stage 2-III (Fig 4B), immediately after outer integuments were initiated [5,6]. Upon initiation at an abaxial position, outer integuments in hap13-1 sometimes failed to spread around the circumference of ovules, resulting in incomplete outer integuments (Fig 4B and 4F), unlike those in wild type where outer integuments fully covered inner integuments (Fig 4A and 4E). Cell expansion in outer integuments of hap13-1 was impaired (Fig 4C and 4D). This irregular cell growth and possibly division in hap13-1 resulted in ovules with exposed inner integuments, enveloped by fragmented outer integuments (Fig 4B and 4F). The two-cell-layered organization of outer integuments was also disrupted (Fig 4B and 4F), likely caused by mis-oriented division planes (Fig 4B and 4F). About 80% ovules in hap13-1 contained abnormal FGs, which always correlated with abnormal growth of outer integuments (Fig 4F).
The phenotypic defects of hap13-1 ovules resembled that of sub mutants to a great extent [14] and an earlier study showed that SUB could be internalized upon treating root cells with Brefeldin A (BFA) [33], a fungal toxin that inhibits post-Golgi secretion [34]. We thus hypothesized that membrane distribution of SUB relied on HAP13-mediated protein sorting at the TGN/EE. To test this hypothesis, we analyzed the distribution of SUB in hap13-1 by introducing a YFP-translational fusion of SUB genomic fragment driven by its endogenous promoter, SUBg-YFP [35]. A post-transcriptional regulation was proposed earlier based on the differential distribution patterns of a full SUB genomic fusion and a fusion with the coding sequence of SUB [33,36,37]. Nevertheless, both were able to fully restore ovular defects of sub mutants [33,36,37].
In ovules of the SUBg-YFP transgenic lines, YFP was barely detectable in the nucellus but present at both inner and outer integuments during early stages of ovule development as well as in all cell types in mature ovules (Fig 5A–5E), as reported [33,36,37]. Immediately after ovular primordia formation, YFP was distributed evenly at the plasma membrane (PM) of epidermal cell layer and also in a few vesicles (Fig 5A). With the development of ovules, YFP signals became asymmetric, i.e. excluded from the outward PM of the epidermal cell layer (Fig 5B and 5C). Punctate vesicles were still detectable (Fig 5B–5D). At later stages when outer integuments began rapid asymmetric elongation and finally wrapped embryo sacs, punctate vesicles labeled by YFP could hardly be detected (Fig 5E). Also YFP was more evenly distributed at the PM of the epidermal cell layer (Fig 5D and 5E). By contrast, YFP signals were mostly diffused in the cytoplasm rather than at the PM of hap13-1 ovules (Fig 5F–5J). Punctate vesicles, much larger than those in wild type, were visible despite the cytosolic signals (Fig 5F–5I). The asymmetry of YFP signals observed in wild type was often compromised such that it labeled the outward PM of the epidermal cell layer (Fig 5I). These results indicated that SUB is dynamically trafficked during ovule development and this process relies on HAP13. Interestingly, YFP signals were substantially reduced in hap13-1 despite that the same transgene was analyzed (Figs 5H and S2). Transcript analysis confirmed that the mRNA abundance of SUB was comparable between wild type and hap13-1 (S2 Fig), suggesting a post-transcriptional regulation of SUB. Indeed, a post-translational regulation was proposed earlier because treatment of the 26S proteasome inhibitor MG132 reduced the protein level of SUB [33].
To further verify that the subcellular targeting of SUB depends on HAP13, we analyzed SUB dynamic targeting in roots by using the fluorescence lipophilic dye FM4-64 together with pharmacological treatments. FM4-64 enters cells via endocytosis, labeling various endomembrane compartments over time and finally reaching vacuolar membrane, i.e. the tonoplast [38]. As shown in ovules, SUB was localized at the PM of wild-type root cells, overlapping with FM4-64 immediately after pulse labeling (S3 Fig). Treatment of roots with BFA at the presence of cycloheximide (CHX) resulted in the accumulation of SUB signals into BFA compartments co-labeled by FM4-64 (S3 Fig). Because BFA compartments formed at the presence of CHX contain only proteins internalized from the PM [27], the result indicated that SUB was constitutively internalized from the PM. The internalization of SUB upon BFA treatment was comparable between wild type and hap13-1 (S3 Fig), which is consistent with the intact endocytosis in hap13-1 [27]. Washout of BFA with CHX caused complete re-distribution of SUB from the BFA compartments to the PM in wild type, when FM4-64 labeled the tonoplast via vacuolar trafficking (S3 Fig). By contrast, the internalized SUB was insensitive to BFA washout in hap13-1 (S3 Fig). Instead, it remained sequestered at the BFA compartments together with FM4-64 (S3 Fig). These results suggested that the recycling of SUB from the TGN/EE to the PM requires HAP13.
Except for SUB, PIN1 is another key component for ovule development by channeling auxin flux to the primordia tip [8,13]. We and others previously showed that the dynamic targeting of PIN2 relies on HAP13 whose mutations resulted in defective gravitropism due to mis-targeting of PIN2 [25,27]. In addition, both PIN1 and PIN2 belong to the same subgroup of PINs with a long hydrophilic loop [39]. We thus hypothesized that PIN1 dynamic targeting also relied on HAP13. To test this hypothesis, we introduced the PIN1 genomic-GFP translation fusion, PIN1::GFP [8], into hap13-1 and followed its membrane distribution at different developmental stages in ovules. PIN1 was detected at the outer cell layer of the nucellus, pointing toward the distal tip (Fig 6A and 6B). Surprisingly, the asymmetric membrane localization of PIN1 was comparable between wild type and hap13-1 despite the abnormal integument growth of hap13-1 (Fig 6C and 6D). Consistent with the intact PIN1 distribution in the nucellus, auxin responses at the nucellus (Fig 6E and 6F) as indicated by the activity of DR5:GFP [40] also did not show abnormality in hap13-1 (Fig 6G and 6H).
To determine whether the HAP13-independent PIN1 targeting also applied to other cells, we examined the localization of PIN1 in the roots of hap13-1. No substantial difference of PIN1 distribution was found between wild type and hap13-1 in root cells (S4 Fig), confirming the HAP13-independency of PIN1 targeting.
We previously showed that a full length HAP13 genomic fragment fully restored developmental and cellular defects in hap13-1 [27], confirming the mutant identity of hap13-1. Indeed, HAP13g-GFP completely restored ovule development of hap13-1 (Fig 1C–1F). However, HAP13 is constitutively expressed and its mutations results in a full spectrum of developmental defects [27]. Thus, the defects in outer integument growth of hap13-1 might be non-cell-autonomous. To exclude the possibility, we used an RNAi approach. We generated ProUBQ10:HAP13-RNAi and analyzed the transcript level of HAP13 in seedlings as well as examined the effect of HAP13-RNAi on vesicle trafficking. Six out of twenty ProUBQ10:HAP13-RNAi transgenic lines were randomly selected. Real-time quantitative PCRs (qPCRs) verified a significant reduction of HAP13 transcripts by ProUBQ10:HAP13-RNAi (S5 Fig). In addition, hap13-1 was insensitive to BFA washout [27], which was confirmed in root epidermal cells of ProUBQ10:HAP13-RNAi transgenic lines by an uptake experiment using the lipophilic dye FM4-64 (S5 Fig). These results verified the efficient down-regulation of HAP13 by RNAi.
To downregulate the expression of HAP13 specifically in outer integuments, we generated a construct containing HAP13-RNAi driven by the promoter of INNER NO OUTER (ProINO). ProINO was reported previously for its specific expression in ovules based on RNA in situ hybridization [41]. To test whether ProINO was specific for outer integuments where severe growth defects were observed in hap13-1, we performed histochemical GUS staining on ProINO:GUS transgenic ovules at different stages. GUS signals were first observed at the stage 2-I in cells to be differentiated into outer integuments (S6 Fig). The asymmetrically distributed GUS signals, correlating with the asymmetric growth of outer integuments, were getting stronger during development. GUS signals were nearly undetectable when embryo sacs were fully formed (S6 Fig). Although ProINO showed a slightly wider spatial activity than the transcripts of INO based on in situ hybridization [42], the histochemical GUS analysis nevertheless demonstrated that ProINO was able to driven gene expression specific in outer integuments during ovule development.
Over 20 independent ProINO:HAP13-RNAi transgenic lines were generated, all of which showed reduced fertility to different extents as compared to wild type (Fig 7B). Two lines with strong or medium defects of ovule development were analyzed by qPCRs using maturing ovules as the materials (floral stage 11–12) (Fig 7A). These two ProINO:HAP13-RNAi lines representing medium or strong reduction of HAP13 expression (RNAi-1 and RNAi-6) were used for further analysis by CLSM and SEM. SEM of developing ovules revealed that HAP13-RNAi resulted in severe reduction of outer integument growth (Fig 7E–7F), similar to and sometimes severer than that in hap13-1, which is a weak rather than null allele of HAP13 [27,28]. At stage 3-I, the mild knockdown line showed regular two-cell-layer inner integuments symmetrically surrounding the nucellus (Fig 7G). By contrast, the growth of outer integuments, although initiated, was severely reduced (Fig 7G). For the severe knockdown line, the growth of outer integuments was arrested immediately after initiation, i.e. the enlarged cells of initiating outer integuments failed to proliferate or grow (Fig 7I). As a result, ovules at maturation stage resembled stubby tubules with only inner integuments enveloping the nucellus (Fig 7J). Downregulating HAP13 specifically in outer integuments phenocopied hap13-1, indicating that defects in ovule development in hap13-1 were not a secondary effect of HAP13 through constitutive expression elsewhere.
Specifically downregulating HAP13 in outer integuments resulted in defects of embryo sac patterning such that most ovules of the ProINO:HAP13-RNAi transgenic plants contained extruding nucellus without the embryo sac at maturation (Fig 7), similar to that shown in hap13-1 (Fig 4). However, it was also possible that genetic interference of HAP13 affected ovule development and thus altered the expression specificity of ProINO, leading to the defects in embryo sac patterning. To exclude the possibility, we generated a binary vector containing two independent expression cassettes, ProINO:HAP13-RNAi and ProINO:NLS-YFP. We generated transgenic plants and applied CLSM to verify the spatial specificity of ProINO in the transgenic lines. In severely affected lines where outer integuments failed to extend, hardly any NLS-YFP signals could be observed in either cell layers within the ovules (S7 Fig). In mildly affected lines where outer integuments showed incomplete extension and wrapped inner integuments partially, NLS-YFP signals were specifically present in outer integuments (S7 Fig). These results confirmed the specific expression of ProINO even when the growth of outer integuments was compromised by downregulating HAP13.
We show here that the hap13-1 mutant was compromised in the asymmetric growth of outer integuments, as a result of both division and elongation defects (Figs 2 and 4). Although HAP13 is expressed in all cells during ovule development (Fig 1), its function does not seem to be crucial for inner integuments because transgenic plants expressing HAP13-RNAi in outer integuments mimicked the hap13-1 mutant (Fig 7). Despite the fact that hap13-1 contained a large portion of ovules with disrupted FGs (Fig 2) and was compromised in pollen tube guidance and reception (Fig 3), HAP13 is also not crucial for female gametogenesis because the heterozygous hap13-1 plants has normal female gametogenesis and female transmission [27] comparable to those in wild type (S1 Fig).
As a subunit of the AP1 complex, HAP13 is critical for vesicle-mediated protein sorting from the TGN/EE to the PM and to vacuoles [25,27]. SUB is the most extensively studied RLK regulating ovule development [14,33,37], in addition to its roles in epidermal patterning and cell fate determination in roots [36,43]. It was reported that SUB accumulates into BFA compartments [33]. However, it was unclear whether SUB within the BFA compartments was trapped during its secretion to the PM or from the PM through constitutive recycling, like ACR4 [44], another RLK regulating epidermal patterning [15]. By using BFA treatment at the presence of CHX, we demonstrated that SUB is constitutively internalized from the PM (S3 Fig). BFA washout resulted in the disappearance of SUB from the BFA compartments in wild type but not in hap13-1 (S3 Fig), suggesting that SUB recycles from the TGN/EE to the PM and this process depends on HAP13. The PM association of SUB was substantially affected in hap13-1 ovules (Fig 5), suggesting that HAP13-mediated SUB recycling plays a role in its PM association in ovules. Intriguingly, in root epidermal cells, no substantial difference was observed on the PM association of SUB between wild type and hap13-1 (S3 Fig), implying that the recycling pathway might be cell- or tissue-specific.
Although it is tempting to propose that HAP13-mediated sorting of SUB at the TGN/EE is critical for pattern formation during ovule development, we can not exclude the possibility that the mis-targeting of other transmembrane proteins also contributes to the observed ovule defects of hap13-1. SUB is the best studied and likely most important RLK regulating integument growth [14]. However, other RLKs such as ACR4 and ER family members are also involved in this process [15,16]. ACR4 was present both at the PM and at endosomes via vesicle trafficking [44] and thus is likely sorted via AP1. Whether ER and ERLs are dynamically trafficked via vesicles is currently unknown. However, the hap13-1 mutant showed compact inflorescence architecture as contrast to that of the ecotype Ws in which the mutant was generated [27]. Because mutations at ER and ER family members impaired cell proliferation in aerial organs and led to similar inflorescence architecture [45], it implies HAP13-dependency of ER function. Indeed, hap13-1 showed abnormal cell division during ovule development (Fig 4), which was less severe in the sub mutants [14]. A key protein involved in cytokinesis, KNOLLE, was reported to rely on HAP13 for its dynamic targeting during cell division [26]. Similar regulators during ovule development may rely on HAP13 for cell division to be properly executed.
Because PIN2 was mis-targeted in hap13-1 [25,27], while PIN1 is another long-hydrophilic looped PIN similar to PIN2 [39], it turned out to be a surprise that the asymmetric localization of PIN1 is not affected in hap13-1, either during ovule development or in roots (Figs 6 and S4). Consistently, auxin response, which plays an instructive role in ovule polarity [8,13], is not compromised in hap13-1 ovules (Fig 6), unlike the case in roots during gravitropic response [27]. The differential requirement of HAP13 for the dynamic targeting of PIN1 and PIN2 suggested distinct trafficking routes involved. Indeed, a previous report showed that Endosidin1, a chemical inducing endocytosis, defines a compartment involved in the endocytosis of PIN2 but not that of PIN1 [19]. The differential requirement of PIN1 and PIN2 on HAP13 indicates that the recycling of PIN1 and PIN2 to the PM also adopts different routes, probably through distinct regulators.
Another interesting point to note is sporophytic control of female gametogenesis. The defective pollen tube guidance and reception shown by hap13-1 suggested a compromised female gametophytic function [46,47]. However, female transmission of the hap13-1 heterozygous plants was comparable to that of wild type [27], suggesting that the female gametophytic defect of the homozygous hap13-1 plants was sporophytic. Indeed, only in the homozygous hap13-1 mutant did we observed defective FG, consistent with the result obtained by using the egg cell-specific marker ProDD45:GUS (S1 Fig). Indeed, it’s known that gametophytic development relies on surrounding sporophytic tissues [3,7,48,49]. Mutations at genes controlling integument growth, such as SUB, often interfered with the formation of embryo sacs [14]. Our results seem to suggest that outer integuments play a key role in the sporophytic controlled female gametogenesis because manipulation of vesicle trafficking routes by ProINO:HAP13-RNAi was sufficient to disrupt the patterning of FGs, suggesting a non-cell-autonomous mechanism. Interestingly, in this case, inner integuments were morphologically intact. How signals from outer integuments transmit through inner integuments to instruct the development of FG is unknown. In addition, inner integuments were suggested sufficient for female gametogenesis [50]. Whether specific signaling pathways are involved in the communication between two generations in hap13-1 will certainly be a topic for future investigations.
Arabidopsis mutants and transgenic lines, including hap13-1 [27], SUBg-YFP [35], PIN1::GFP [8], and DR5:GFP [40] were described. Arabidopsis Ws ecotype was used as the wild type. Arabidopsis plants were grown as described [51]. Stable transgenic plants were selected on half-strength MS supplemented with 30 μg/ml Basta salts, 25 μg/ml Hygromycin B, or 50 μg/ml Kanamycin.
Total RNAs for detecting HAP13 transcripts in ProINO:HAP13-RNAi lines were extracted from mature unfertilized ovules using a Qiagen RNeasy plant mini kit according to manufacture’s instructions. Oligo(dT)-primed cDNAs were synthesized using Superscript III reverse transcriptase with on-column DNase II digestion (Invitrogen). Real time quantitative PCRs were performed as described [51]. Primers are listed in S1 Table.
All constructs were generated using the Gateway technology (Invitrogen) except the RNAi construct. Entry vectors were generated using pENTR/D/TOPO (Invitrogen). Coding sequences or promoter sequences were cloned with the following primer pairs: ZP1781/ZP1782 for ProINO (2012 bp before the start codon of INO); ZP96/ZP97 for ProDD45 (1003 bp before the start codon of DD45). The destination vector used to generate ProDD45:GUS and ProINO:GUS were described previously [52]. Expression vectors were obtained by combining respective entry vectors and destination vectors in LR reactions using LR Clonase II (Invitrogen). HAP13-RNAi fragment (401 bp to 700 bp of its coding sequence) was amplified with the primer pair ZP1650/ZP1651. The resultant PCR products were sub-cloned into the RNAi vector pTCK303 [53] to obtain the ProUBQ10:HAP13-RNAi construct. Later, ProUBQ10 was replaced with ProINO to generate ProINO:HAP13-RNAi. A ProINO:NLS-YFP expression cassette was amplified with primers ZP3872/ZP3873 containing the restriction enzyme site SacI and EcoRI, respectively. The PCR fragment was inserted into the expression vector to generate the ProINO:HAP13-RNAi ProINO:NLS-YFP construct. Primers are listed in S1 Table.
Histochemical GUS staining, ovule whole-mount clearing, semi-thin sections, and SEMs were performed as described [51]. Aniline blue staining of pollen tubes growing inside pistils was performed as described [54].
Whole-mount immuno-fluorescence staining of ovules using an anti-LURE antibody [29] was performed as described [55]. Briefly, unfertilized mature pistils were placed in a fixative solution (3.7% paraformaldehyde, 1 mM CaCl2, 1 mM MgSO4, 50 mM HEPES (pH 7.4), 5% sucrose, pH8.0 with NaOH) for 10 min with vacuum desiccation at room temperature (RT). The fixative solution was then removed and pistils were washed three times for 10 min each with the PME buffer (50 mM PIPES, 1 mM MgCl2, 5 mM EGTA, pH6.8 with NaOH). The fixed pistils were treated further for permeation sequentially with a PME buffer, cellulose solution, and 3% IGEPAL CA-630. After permeation, pistils were sequentially incubated with the primary antibody (anti-LURE antibody, 1:300) and second antibody (FITC-labeled goat anti mouse, 1:100), followed by tissue mounting and CLSM examination. Uptake of FM4-64, BFA treatment and CLSM were performed as described [27]. Optical sectioning of ovules was performed as described [7,56].
Sequence data from this article can be found in the GenBank databases under the following accession numbers: At1g60780 for HAP13, At2g21750 for DD45, At1g23420 for INO, At5g43510 for LURE1.2.
|
10.1371/journal.pgen.1006451 | A Novel Rrm3 Function in Restricting DNA Replication via an Orc5-Binding Domain Is Genetically Separable from Rrm3 Function as an ATPase/Helicase in Facilitating Fork Progression | In response to replication stress cells activate the intra-S checkpoint, induce DNA repair pathways, increase nucleotide levels, and inhibit origin firing. Here, we report that Rrm3 associates with a subset of replication origins and controls DNA synthesis during replication stress. The N-terminal domain required for control of DNA synthesis maps to residues 186–212 that are also critical for binding Orc5 of the origin recognition complex. Deletion of this domain is lethal to cells lacking the replication checkpoint mediator Mrc1 and leads to mutations upon exposure to the replication stressor hydroxyurea. This novel Rrm3 function is independent of its established role as an ATPase/helicase in facilitating replication fork progression through polymerase blocking obstacles. Using quantitative mass spectrometry and genetic analyses, we find that the homologous recombination factor Rdh54 and Rad5-dependent error-free DNA damage bypass act as independent mechanisms on DNA lesions that arise when Rrm3 catalytic activity is disrupted whereas these mechanisms are dispensable for DNA damage tolerance when the replication function is disrupted, indicating that the DNA lesions generated by the loss of each Rrm3 function are distinct. Although both lesion types activate the DNA-damage checkpoint, we find that the resultant increase in nucleotide levels is not sufficient for continued DNA synthesis under replication stress. Together, our findings suggest a role of Rrm3, via its Orc5-binding domain, in restricting DNA synthesis that is genetically and physically separable from its established catalytic role in facilitating fork progression through replication blocks.
| When cells duplicate their genome, the replication machinery is constantly at risk of encountering obstacles, including unusual DNA structures, bound proteins, or transcribing polymerases and transcripts. Cells possess DNA helicases that facilitate movement of the replication fork through such obstacles. Here, we report the discovery that one of these DNA helicases, Rrm3, is also required for restricting DNA synthesis under replication stress. We find that the site in Rrm3 critical for this new replication function is also required for binding a subunit of the replication origin recognition complex, which raises the possibility that Rrm3 controls replication by affecting initiation. This is supported by our finding that Rrm3 associates with a subset of replication origins. Rrm3’s ability to restrict replication does not require its helicase activity or the phosphorylation site that regulates this activity. Notably, cells need error-free bypass pathways and homologous recombination to deal with DNA lesions that arise when the helicase function of Rrm3 is disrupted, but not when its replication function is disrupted. This indicates that the DNA lesions that form in the absence of the two distinct Rrm3 function are different, although both activate the DNA-damage checkpoint and are toxic to cells that lack the mediator of the replication checkpoint Mrc1.
| The replication machinery is constantly at risk of encountering obstacles such as protein-DNA complexes, DNA secondary structures, transcribing RNA polymerases, RNA-DNA hybrids, and DNA damage, all of which can block fork progression. If these structures cannot immediately be resolved the paused fork may eventually collapse as replisome components become irretrievably inactivated.
The 5’ to 3’ DNA helicase Rrm3 is a member of the Pif1 family, which is conserved from yeast to humans [1,2]. Saccharomyces cerevisiae RRM3 was first discovered as a suppressor of recombination between tandem arrays and ribosomal DNA (rDNA) repeats [3]. Without Rrm3, extrachromosomal rDNA circles accumulate, suggesting a role in maintaining rDNA repeat stability, and cells accumulate recombination intermediates at stalled replication forks, which has lead to the suggestion that Rrm3 facilitates DNA unwinding and the removal of protein blocks to help fork convergence during replication termination [4–7]. Additionally, replication fork pausing has been observed in the absence of Rrm3 at centromeres, telomeres, tRNA genes, the mating type loci, inactive origins of replication, and RNA polymerase II-transcribed genes [3,5,6].
The mechanism by which Rrm3 aids fork progression is poorly understood, but it is thought that the ATPase/helicase activity of Rrm3 facilitates replication through protein blocks and may also be able to remove RNA transcripts [5,8]. Within each rRNA coding region are two intergenic spacers that contain termination sites that are bound by the replication terminator protein Fob1 to promote fork arrest and to prevent unscheduled transcription [9–11]. Termination site function also requires the intra-S phase checkpoint proteins Tof1 and Csm3, which form a complex with the replisome and antagonize Rrm3 function [12,13]. It is thought that Rrm3 removes Fob1 and other non-histone proteins from DNA before the replication fork encounters them. This ability of Rrm3 to promote replication fork progression is dependent on its catalytic activity [7]. Further supporting a role of Rrm3 in fork progression are synthetic fitness defects or lethality between rrm3Δ and mutations that disrupt genes involved in maintaining the integrity of stalled forks, including rad53Δ, mec1Δ, srs2Δ, sgs1Δ, mrc1Δ, and rtt101Δ [5,14–16].
Rrm3 possesses an N-terminal PCNA-interacting peptide (PIP) box, associates with the replication fork in vivo and is hyperphosphorylated by Rad53 under replication stress [1,17,18]. The replication damage that arises in the absence of Rrm3 causes constitutive, Mec3/Mec1/Rad9-dependent activation of the checkpoint kinase Rad53 [5,17,19,20]. As a result, Dun1 kinase is activated, leading to degradation of the ribonucleotide reductase (RNR) inhibitor Sml1 and an increase in the dNTP pool [21,22]. This increased dNTP pool has been associated with enhanced DNA synthesis in hydroxyurea (HU) in chromosome instability mutants [22].
Here we show that cells lacking Rrm3 fail to inhibit DNA replication in the presence of HU-induced replication stress and that this failure is not caused by the increased dNTP pool resulting from constitutive DNA-damage checkpoint activation. This novel replication function of Rrm3 is independent of its ATPase/helicase activity and, thus, distinct from Rrm3’s established catalytic role in facilitating fork progression through replication blocks. Instead, we have identified dependency on a novel functional domain in the Rrm3 N-terminus that we find is involved in binding the Orc5 subunit of the origin recognition complex (ORC). Together with our finding that Rrm3 associates with a subset of replication origins, this suggests that Rrm3 may control DNA synthesis by controlling origin activity. Quantitative mass spectrometry and genetic analyses further implicate Rad5-dependent error-free DNA damage bypass and Rdh54 translocase as novel repair mechanisms for DNA lesions that result from inactivating the catalytic activity of Rrm3, whereas these DNA repair factors are dispensable when the Orc5-binding domain is disrupted, leading us to conclude that the types of DNA lesions that result from the inactivation of the two independent Rrm3 functions are distinct.
In the absence of Rrm3 cells accumulate replication pause sites at the rDNA locus, in tRNA genes and at centromeric regions, as well as many other sites throughout the genome [5,6,15]. To identify DNA metabolic pathways that deal with stalled forks, we sought to identify proteins whose association with chromatin changed in the absence of Rrm3 using stable isotope labeling by amino acids in cell culture (SILAC)-based quantitative mass spectrometry [23,24]. We extracted the chromatin fraction from nuclei purified from a mixture of wildtype and rrm3Δ cells grown in the presence of heavy or light isotopes of arginine and lysine, respectively (Fig 1A and 1B). Across chromatin fractions from three biological replicates we identified 490 peptides from 137 different proteins, with the abundance of 11 proteins changing significantly in at least two of the three replicates (Fig 1C). The largest change in chromatin association was a 5.1-fold increase (p<0.001) of Rad5, which belongs to the SWI/SNF family of ATPases and defines an error-free pathway for bypassing replication-blocking DNA lesions [25–28]. The increase in Rad5 was followed by smaller, but significant, increases for Top2 (1.9-fold, p<0.01), a type II topoisomerase that is important for the decatenation of replication intermediates, and Rdh54 (1.8-fold, p<0.01), a chromatin remodeler with a role in homologous recombination that is still largely unclear. Like Rad5, Rdh54 is a member of the SWI/SNF family of ATPases; it possesses translocase activity on double-stranded (ds) DNA and has been shown to be capable of modifying DNA topology, especially in chromatinized DNA [29–31]. We observed significant decreases in chromatin association for the Rsc1 subunit of the chromatin-structure-remodeling (RSC) complex (2-fold, p<0.01), the Mcm4 subunit of the minichromosome maintenance (MCM) replicative DNA helicase (1.9-fold, p<0.01), and the catalytic subunit Hda1 of the histone deacetylase (HDAC) complex (1.7-fold, p<0.01).
Upon treatment with HU, which induces replication stress by reducing the nucleotide pool [21], Rdh54 abundance in the chromatin fraction of the rrm3Δ mutant increased the most (2.6-fold, p<0.01) whereas the histone deacetylase Set3 and the Rsc9 subunit of the RSC chromatin remodeling complex saw the largest decreases (2.7-fold, p<0.05) (Fig 1D and 1E). The complete list of proteins that underwent significant changes in the HU-treated or untreated rrm3Δ mutant, including the FANCM-related Mph1 helicase, the recombination factor Mgm101, and the cohesin components Smc1, Smc3 and Scc3, is provided in S1 Table.
Rrm3 helicase is required to prevent excessive replication fork pausing at nonhistone-protein-bound sites, possibly by acting as a protein displacement helicase [6]. The role of Rdh54 as a dsDNA translocase that can act on chromatinized DNA [32,33] and the fork reversal activity of Rad5 suggest that they are recruited to chromatin to recover forks that are blocked due to the lack of Rrm3 or to substitute for Rrm3 in preventing fork pausing. We therefore examined the effect of deleting RAD5 and RDH54 in the rrm3Δ mutant on genome stability and sensitivity to DNA damage caused by methyl methanesulfonate (MMS) and to replication stress caused by HU. We found synergistic increases in sensitivity to HU and MMS in the rrm3Δ rad5Δ and rrm3Δ rdh54Δ mutants (Fig 2A). The negative genetic interaction between rrm3Δ and rad5Δ was particularly strong; both single mutants were no more sensitive to HU than wildtype, but the double mutant failed to form colonies on 100 mM HU and grew very poorly even on 20 mM HU. In contrast to HU, the rad5Δ mutant was extremely sensitive to MMS, and deleting RRM3 led to a further synergistic increase in MMS sensitivity. Inactivation of the ATPase activity of Rrm3 (rrm3-K260A/D) caused the same hypersensitivity in the rad5Δ mutant as an RRM3 deletion (Fig 2B). We also identified a negative genetic interaction between rrm3Δ and rdh54Δ, which was especially strong on MMS. The increased sensitivity of rdh54Δ cells to HU and MMS upon deletion of RRM3 extended to diploid cells (Fig 2G), suggesting that the lesions generated in the absence of Rrm3 are also substrates for recombination between homologous chromosomes that is controlled by Rdh54. Even though the rrm3Δ rad5Δ mutant was hypersensitive to MMS and HU, deletion of RDH54 caused further synergistic increases in sensitivity to both chemicals, indicating that Rad5 and Rdh54 define important pathways for dealing with DNA lesions that arise in the absence of Rrm3, and that they perform (at least some) independent roles. In addition to structure-specific helicase activity, Rad5 also possesses a RING motif associated with ubiquitin ligase activity that plays a role in polyubiquitination of PCNA [34–37]. Two mutations in Rad5, Q1106D and C914A/C917A, were recently described to disrupt its helicase and ubiquitin-ligase activity, respectively [27,38]. Disrupting either of these Rad5 activities in the rrm3Δ mutant significantly increased sensitivity to MMS and to HU, indicating that both activities make important contributions to the repair of DNA lesions that arise in the absence of Rrm3 (Fig 2B).
Although Rad5 and Rdh54 chromatin association increased most in the absence of Rrrm3 (Fig 1E), gross-chromosomal rearrangements (GCRs) did not accumulate at higher rates in the rrm3Δ rad5Δ or rrm3Δ rdh54Δ mutants compared to the single mutants, even after exposure to HU and MMS (Table 1). However, disruption of both, RAD5 and RDH54, in the rrm3Δ mutant caused a significant, albeit small, increase in chromosome instability compared to disruption of the single genes, especially upon exposure to HU or MMS, supporting independent contributions of Rdh54 and Rad5-mediated repair mechanisms to genome stability and DNA damage tolerance in the absence of Rrm3. That the effect on the GCR accumulation was small despite a synergistic effect on hypersensitivity to HU and MMS (Fig 2A) could indicate that the DNA lesions are not substrates for GCRs or that chromosomal rearrangements that form in these mutants are not viable.
Whereas rdh54Δ and rad5Δ cells moved through an undisturbed cell cycle with similar kinetics as wildtype cells, rrm3Δ cells were delayed in progressing to G2/M, consistent with previous observations [6,14]. We find that this delay was enhanced when RDH54 or RAD5 were deleted (Fig 2C). To examine progression of rrm3Δ rad5Δ and rrm3Δ rdh54Δ cells through S phase under replication stress, we released α-factor arrested cells from G1 phase in the presence of HU and trapped them in G2/M with nocodazole (Fig 2D and 2E). After 140 minutes, virtually all wildtype cells had reached 2C DNA content, whereas rrm3Δ and rdh54Δ showed a marked delay (Fig 2D, 120 minute time point). When we combined rrm3Δ and rdh54Δ mutations, this slowdown was so severe that most cells still had near 1C DNA content 100 minutes after release from G1 arrest, consistent with the synergistic increase in HU hypersensitivity of the rrm3Δ rdh54Δ mutant. Similarly, rad5Δ rrm3Δ cells were delayed in reaching 2C DNA content in HU (Fig 2E). However, all mutants were able to recover from a 2-hour arrest in 100 mM HU and resume the cell cycle normally (S1 Fig). When we examined the ability of nocodazole-arrested cells in G2/M to complete mitosis and reach G1 phase, we found that most wildtype cells were in G1 after 10 minutes, whereas rrm3Δ, rad5Δ, rdh54 cells and the double mutants showed 2C DNA content 20 minutes after release (Fig 2F). Together, these findings indicate that Rad5 and Rdh54 facilitate the progression of rrm3Δ cells through S phase, both in the presence and in the absence of HU, and that in the absence of Rrm3, cells accumulate DNA damage that impairs exit from mitosis.
In addition to Rad5 and Rdh54, which exhibited the most significant increases in chromatin association in the absence of Rrm3 (Fig 1E), we tested DNA damage sensitivity of cells that lacked Rrm3 in combination with other nonessential factors revealed in the proteome screen (Fig 1C and 1D), including Mgm101, Hda1, Set3, and Mph1. Whereas deletions of MGM101, HDA1 or SET3 had no effect on sensitivity of wildtype or rrm3Δ cells to HU or MMS (S2 Fig), deletion of MPH1 caused a synergistic increase in HU and MMS sensitivity of the rrm3Δ mutant (S3 Fig), consistent with our previous finding [39]. In the absence of Mph1, rrm3Δ cells progressed very slowly through an undisturbed cell cycle and accumulated in G2/M when they were released from a 2-hour incubation in 100 mM HU (S3B and S3C Fig). When cells were released from HU arrest into media with 40 mM HU and α-factor, virtually all wildtype cells and the single mutants were trapped in G1 phase after 60 minutes (with a slight S phase delay in the mph1Δ mutant), whereas the majority of rrm3Δ mph1Δ cells accumulated in S phase, never forming a majority peak at 1C DNA content in the 120-minute time course (S3D Fig). Deleting MPH1 in the rrm3Δ rad5Δ mutant led to a slight increase in sensitivity to HU compared to the double mutants, but not to MMS (S3E Fig). These findings implicate Mph1 as another crucial factor for overcoming spontaneous and DNA-damage-induced replication-blocking lesions when Rrm3 is absent and suggest pathways for error-free bypass of DNA polymerase blocking lesions, implicated by Rad5 and Mph1, and homologous recombination, implicated by Rdh54, as independent mechanisms that are recruited to chromatin to act on blocked replication forks.
All functions of the Rrm3 helicase known to date are dependent on its ATPase/helicase activity. During our analysis of cell cycle progression, however, we observed that cells with a deletion of RRM3 continue to replicate DNA in the presence of HU, similar to a rad53Δ checkpoint mutant, whereas the helicase-defective rrm3-K260A and rrm3-K260D mutants maintained near 1C DNA content after 2 hours in HU, similar to wildtype (Fig 3B and 3C). This observation suggested the presence of a previously unknown, ATPase/helicase-independent function of Rrm3 in DNA replication. Since this replication defect was independent of the ATPase/helicase activity located in the ordered C-terminal domain of Rrm3 (residues 250–723), we explored a possible involvement of the 230-residue, disordered N-terminal tail (Fig 3A, S4A Fig). The only motifs previously identified in this tail are a putative PCNA-interacting peptide (PIP) box between residues 35–42 [18] and a cluster of phosphorylated residues between S85 and S92 [17]. Deletion or mutation of the PIP-box (rrm3-ΔN54, rrm3-FFAA) had no effect on DNA replication in HU, whereas deletion of the entire N-terminal tail (rrm3-ΔN230) caused the same replication defect as deleting RRM3 (rrm3Δ) (Fig 3C). By constructing a series of N-terminal truncations (Fig 3A and 3C) we determined that a deletion of up to 186 residues, which also eliminates the PIP-box and the phosphorylation site, was able to maintain the wildtype replication phenotype in HU, whereas deletions of 212 or 230 residues caused the same inability to restrict DNA synthesis in HU as rrm3Δ (Fig 3D–3G), thus narrowing down the critical functional site for control of DNA replication to the 26 residues between residues 186–212. This defect was not due to changes in protein stability or levels of expression of the rrm3 mutant alleles (Fig 3H), which had been observed for other Rrm3 truncations before [20]. The importance of residues 186–212 for controlling DNA replication was limited to HU, and not observed when cells were exposed to the alkylating agent MMS (S4B Fig).
Deletion of RRM3 or inactivation of its ATPase/helicase activity was recently reported to partially suppress the HU hypersensitivity of the rad53Δ mutant [17]. We obtained the same findings, and observed that the new rrm3-ΔN212 allele does not act as a suppressor (Fig 3I), indicating that the rrm3-ΔN212 allele codes for a functional ATPase/helicase. To verify this, we disrupted the Walker A motif in the rrm3-ΔN212 mutant (rrm3-Δ212-K260A) and, as expected, it now suppressed the HU hypersensitivity of the rad53Δ mutant to the same extent as the ATPase-defective rrm3-K260A allele (Fig 3I).
The Rad53 checkpoint kinase was constitutively activated in the rrm3-Δ212 mutant just like in the ATPase/helicase-defective rrm3-K260A/D mutants, and Rad53 activation in both mutants was dependent on the mediator of the DNA damage checkpoint Rad9 (Fig 3J and 3K). Through degradation of the ribonucleotide reductase (RNR) inhibitor Sml1, the nucleotide pool increases upon Rad53 activation, and this correlates with enhanced fork progression [22]. We found that the rrm3 mutants that continued DNA replication in HU (rrm3Δ, rrm3ΔN212) as well as the rrm3 mutant that maintained a peak at 1C DNA content (rrm3-K260D) had constitutively increased nucleotide pools upon entrance into S phase (Fig 3L), indicating that the continued DNA replication in HU seen in the rrm3-ΔN212 mutant could not be explained by a larger nucleotide reservoir prior to its depletion by HU addition. In fact, based on quantification of the cell cycle profiles obtained by flow cytometry and by visual analysis of morphology the vast majority of rrm3Δ, rrm3-ΔN212 and rrm3-ΔN230 cells entered S phase in HU and continued to progress, whereas the majority of wildtype cells and the other rrm3 mutants did not enter S phase during the 2-hour incubation in HU, with the peaks of DNA content remaining at 1C (Fig 3B–3G, S4C Fig).
Together, these findings suggest a new function of Rrm3 in restricting DNA replication in the presence of HU and prevention of S phase damage, which maps to residues 186–212 of the N-terminal tail and does not require Rrm3’s established activity as an ATPase/DNA helicase.
Long disordered tails, such as the N-terminal 230 residues of Rrm3 that extend from a structured catalytic core, typically serve as sites for protein binding and posttranslational modification [40]. The phenotype of the rrm3-ΔN212 allele in the rad53Δ mutant indicates that it encodes a proficient ATPase/helicase, suggesting that the replication defect of this allele is caused by loss of a protein-binding site in the disordered tail. Because deletion of the putative PIP-box and the recently identified phosphorylation site did not impair the ability of Rrm3 to control DNA replication, we explored the possibility that Orc5, an ATP-binding subunit of the origin recognition complex (ORC), binds to the N-terminal tail of Rrm3. An interaction between the two full-length proteins had previously been identified in a yeast-two-hybrid screen [41]. When we combined ORC5 with the various rrm3 truncation alleles in a yeast two-hybrid assay, we found that deletion of 186 residues did not diminish Orc5 binding to Rrm3, in the presence or absence of MMS or HU, whereas deletion of 212 or 230 residues eliminated binding (Fig 4A). To verify the importance of the N-terminal region of Rrm3 for Orc5 binding in vivo we tested the ability of Orc5 to co-immunoprecipitate myc-epitope-tagged Rrm3, rrm3-ΔN186 and rrm3-ΔN212. Consistent with the yeast-two hybrid assay, wildtype Rrm3 and rrm3-ΔN186 bound efficiently to Orc5 whereas binding of rrm3-ΔN212 was impaired (Fig 4B). These findings show that the same site of Rrm3 that restricts DNA replication in HU is required for a physical interaction with Orc5 in vivo and raise the possibility that Rrm3 may control DNA replication by affecting replication origins.
To test the hypothesis that Rrm3 acts on replication origins we tested if Rrm3 and rrm3-ΔN212 associate with origins and if this association is affected by the presence of HU. Since progression of rrm3Δ and rrm3-ΔN212 mutants into S phase in the presence of HU is not as pronounced as in the absence of the Rad53 checkpoint kinase, which modulates the timing of origin firing and S phase progression upon exposure to HU, we considered that Rrm3 might act only on a subset of origins. We therefore selected a variety of replication origins for analysis by chromatin immunoprecipitation (ChIP), ranging from early to late-initiating origins and including two origins near telomeres (ARS319, ARS501). We performed ChIP on asynchronous cultures, cultures synchronized in G1 with α-factor, as well as cultures that were released from G1 into S phase for 45 minutes in the presence or absence of HU. We observed that Rrm3 and rrm3-ΔN212 associated with ARS305, ARS601, ARS603, ARS607, and ARS1414 in asynchronous cultures, in G1 and in S phase, but not with ARS1411 (Fig 4C). In fact, the lack of a PCR product for ARS1411 in the asynchronous culture indicates that Rrm3 does not associate with this replication origin for any extended period during the cell cycle; Rrm3 was also not at ARS306, ARS319, ARS416, ARS522, ARS606, and ARS609 (S4D Fig). This suggests that Rrm3 associates with a subset of origins of replication in unperturbed G1 and S phase independently of its Orc5-binding site. However, when we released cells into S phase in the presence of 200 mM HU, rrm3-ΔN212 lost its association with ARS602, ARS603, ARS607, and ARS1414 whereas wildtype Rrm3 remained bound (Fig 4C), suggesting that the failure of the rrm3-ΔN212 mutant to halt DNA synthesis and progression into S phase in the presence of HU might be due to a failure of rrm3-ΔN212 to act on a subset of replication origins (40% of origins tested in this study).
To investigate the link between Rrm3 functions and DNA replication, we examined the replication checkpoint. Replication mutants exhibit strong genetic interactions with Mrc1/Claspin, which acts as a mediator of the replication stress checkpoint–a Rad9-independent pathway of the intra-S-phase checkpoint [42–46]. Mrc1 is also a component of normal replication forks, which is loaded at origins of replication and stays associated with the replisome [42,44,47,48]. Mrc1, like Rrm3, is required for efficient replication [49]. The function of Mrc1 in DNA replication is essential for the viability of cells lacking Rrm3 [47] whereas Mrc1 phosphorylation on SQ and TQ sites linked to its checkpoint function is dispensable [19]. However, the role of this functional interaction between Rrm3 and Mrc1 in DNA replication has remained unclear. We therefore tested if the ability of Rrm3 to control DNA replication was required for the viability of the mrc1Δ mutant. For this purpose, we transformed diploids heterozygous for the mrc1Δ and rrm3Δ mutations with plasmids expressing N-terminal truncations of Rrm3 and analyzed the viability of meiotic products. Fig 5A shows that the rrm3-ΔN186 allele restored viability to the rrm3Δ mrc1Δ mutant as effectively as the wildtype RRM3 allele, whereas the helicase-dead alleles and the rrm3-ΔN212 allele were as ineffective as the null allele (empty plasmid). Thus the helicase activity of Rrm3 is not sufficient for viability of the mrc1Δ mutant; Rrm3’s new Orc5-binding domain for controlling DNA replication is also required.
In addition to Mrc1, Tof1 promotes normal progression of the replication fork; however, in contrast to Mrc1, its requirement for fork progression appears more limited, assisting primarily replication through non-histone protein complexes with DNA [50]. TOF1 deletion was not lethal in the rrm3Δ mutant and neither single mutant was hypersensitive to HU or MMS. The combined loss of Rrm3 and Tof1, however, caused a synergistic increase in DNA-damage sensitivity (Fig 5B). Identical to the functional requirements in the absence of Mrc1 both, the ATPase/helicase activity of Rrm3 and the Orc5 binding domain, were required for growth in the presence of DNA damage and replication stress in the absence of Tof1.
In contrast to mrc1Δ and tof1Δ mutants, we found that only the ATPase/helicase activity of Rrm3 was required for the suppression of HU and MMS hypersensitivity of the rdh54Δ mutant (Fig 5C). The N-terminal tail, including its function in controlling DNA replication, was dispensable, with the rrm3-ΔN212 allele exhibiting a wildtype phenotype in the rdh54Δ mutant. When we extended this analysis to the Rdh54 homolog Rad54, and the HR factor Rad51, which both Rdh54 and Rad54 interact with [29,51], we observed that the rrm3 alleles caused the same phenotypes in the rad51Δ mutant as in the rdh54Δ mutant, but had no effect on the rad54Δ mutant (Fig 5D and 5E). Like rdh54Δ and rad51Δ mutants, rad5Δ and mph1Δ mutants exhibited increased sensitivity to HU and MMS only if the ATPase activity of Rrm3 was disrupted, whereas the rrm3-ΔN212 allele caused the same phenotype as the RRM3 wildtype allele (Fig 5F and 5G).
Together, these findings suggest two separable functions of Rrm3 in DNA replication. First, an ATPase/helicase-dependent function that facilitates fork progression through protein-DNA complexes, which if disrupted (rrm3-K260A/D) causes aberrant replication intermediates that can be rescued by Rad5, Rdh54/Rad51 or Mph1 mechanisms. Second, an N-terminal function that restricts DNA replication in the presence of HU, mediated by Rrm3 association with replication origins, which if disrupted (rrm3-ΔN212) requires the replication checkpoint factors Mrc1 and Tof1 for viability and DNA damage survival. This differential requirement of factors involved in DNA repair and DNA damage tolerance pathways in the rrm3-ΔN212 and rrm3-K260A/D mutants also suggests that the types of DNA lesions that accumulate upon inactivation of the two Rrm3 functions are different, but both lead to dependence on Mrc1 for survival and both are sufficient for constitutive activation of the DNA-damage checkpoint.
If Rrm3 is important for the response to replication stress induced by HU, cells lacking the catalytic activity of Rrm3 or its Orc5-binding domain might be prone to accumulating mutations at higher rates than wildtype cells. To test this, we measured forward mutation rates at the CAN1 locus and the accumulation of GCRs on chromosome V in the presence and absence of HU or MMS (Table 2).
Two-fold (ung1Δ) to 50-fold (rad27Δ) increases in CAN1 forward mutation rates compared to wildtype have previously been reported for numerous DNA metabolism mutants [52]. Deletion of RRM3 or disruption of its ATPase/helicase activity caused a significant increase in spontaneous CAN1 mutations (Table 2). Of the truncation alleles, which encode catalytically active rrm3 mutants (Fig 3I), rrm3Δ-N186 was indistinguishable from wildtype whereas rrm3Δ-N212 caused a small, but significant, increase in the CAN1 mutation rate in untreated cells and upon exposure to HU. In contrast, expression of the rrm3Δ-N212 allele had no effect on the CAN1 mutation rate if cells were treated with MMS, consistent with our observation that the rrm3-ΔN212 mutant exhibits a defect in controlling replication in HU, but not MMS. GCRs accumulated at increased rates in the rrm3Δ and rrm3-K260A/D mutants in the absence and presence of HU or MMS, but accumulated at wildtype levels in cells expressing N-terminal truncations under all conditions. These mutator phenotypes, albeit mild, reveal that Rrm3’s ATPase/helicase activity helps to suppress all tested mutation types induced by either HU or MMS, or in their absence, whereas the N-terminal plays a role specifically in the suppression of spontaneous and HU-induced mutations, but not for the suppression of MMS-induced mutations, or GCRs under any conditions.
By quantifying changes in chromatin composition we have identified Rad5 and Rdh54 as novel factors that respond to increased replication fork stalling induced by the absence of Rrm3, and affirmed the importance of Mph1. These factors suggest that error-free post-replicative repair (PRR), implicated by Rad5 and Mph1, and HR, implicated by Rdh54, act on DNA polymerase blocking sites that arise throughout the genome in the absence of Rrm3. The N-terminal unstructured tail, including the Orc5-binding site identified in this study, is dispensable for this ATPase/helicase-dependent role of Rrm3 in facilitating fork progression. Instead, we have discovered that the N-terminal tail encodes a new function of Rrm3 –to control DNA replication in the presence of HU. This function of Rrm3 is distinct from its established role as an ATPase/helicase, is not regulated by the previously identified phosphorylation cluster [17] or the PIP-box [18] and, in contrast to the ATPase/helicase activity of Rrm3, does not contribute to the HU hypersensitivity of the rad53Δ mutant.
Based on changes in DNA content as measured by flow cytometry, we observed that wildtype cells maintained near 1C DNA content for 180 minutes after release from G1 phase into HU, whereas rad53Δ, rrm3Δ and rrm3-ΔN212 did so for only 60 minutes (Fig 3, S4C Fig). The extent of continuing DNA replication in the presence of HU, however, was not as pronounced in the rrm3 mutants as in the rad53Δ mutant. Although the DNA-damage checkpoint is chronically activated in the rrm3-ΔN212 mutant and, as a consequence, nucleotide levels are increased as cells are about to enter S phase, these increased nucleotide levels do not appear to be not sufficient for ability of the rrm3-ΔN212 mutant to continue DNA replication upon HU exposure because the rrm3-K260A/D mutants showed the same nucleotide level increase and DNA-damage checkpoint activation, but maintained a peak at 1C DNA content in HU with only a small percentage of cells entering S phase (Fig 3G).
Therefore, considering Rrm3’s known function as an accessory ATPase/helicase that facilitates progression of the replication fork, and its new function in controlling DNA synthesis reported here, we propose a model (Fig 6) where Rrm3 performs two genetically and physically separable functions to deal with challenges during genome duplication: (1) the N-terminal tail of Rrm3 plays a structural role in preventing untimely replication in the presence of replication stress (HU) and in normal S phase, and (2) the C-terminal ATPase/helicase domain plays a catalytic role in preventing fork pausing. The site between residues 186 to 212, which is in a segment of the N-terminal tail not previously assigned a function, is not only involved in restricting DNA synthesis in HU, but also for the association of Rrm3 with Orc5 and a subset of origins of replication. That rrm3-ΔN212 fails to restrict DNA synthesis and S phase progression in the presence of HU suggests that the association of Rrm3 with origins of replication is inhibitory and that this inhibition is realized through binding Orc5, the ATP-binding subunit of ORC. By binding Orc5, Rrm3 could act as an inhibitor of ORC ATPase activity, which is required for loading of minichromosome maintenance (MCM) proteins and for initiation of DNA replication [53,54], or Rrm3 could block the recruitment of another replication protein. Although ORC is associated with origins throughout the cell cycle, Orc5 does not appear to play a role in the completion of S phase, or the remainder of the cell cycle [55,56].
Besides the association of Rrm3 with origins in HU, which depends on the N-terminal tail, our ChIP data from the rrm3-ΔN212 mutant also show association with origins in normal G1 and in unperturbed S phase in a manner that does not require the N-terminal tail, invoking the presence of another protein binding site in the rrm3-ΔN212 polypeptide, which is made up almost entirely of the catalytic domain. Indeed, the phenotype of rrm3-ΔN212 mutants deficient in HR (rdh54Δ, rad51Δ) or PRR (rad5Δ, mph1Δ) suggests that the association of Rrm3 with the replisome [1], appears to occur outside of the N-terminal tail. While the N-terminal tail is required for binding Orc5, it is not required for association with origins of replication, unless cells are exposed to replication stress. Together with continued DNA synthesis and progression into S phase in the presence of HU our findings, thus, raise the possibility that Rrm3 performs a replication checkpoint-like function in response to HU.
Instead of a global role in controlling origin activity, the wildtype level of HU sensitivity of rrm3Δ cells, the less pronounced S phase progression in HU than that of the rad53Δ mutant, and the importance of Rrm3 for replicating through certain nonhistone-protein-bound regions suggest that Rrm3 may play a role at origins in specific loci, such as those in highly transcribed regions and regions with converging transcription, which are often late-firing [57], rRNA and tRNA coding loci, or highly transcribed metabolic genes, where ORC has been found to be bound to the open reading frames, possibly to coordinate the timing of replication with transcription [58]. Indeed, our analysis so far has revealed that Rrm3 appears to associate only with a subset of replication origins. The group of origins not associated with Rrm3 includes early and late replicating origins as well as two origins near telomeres. It is currently unclear what distinguishes the origins that associate with Rrm3 from those that do not. For example, there are no consistent differences in how their activity in HU is affected by mutations in the DNA-damage checkpoint or the DNA replication checkpoint [59], the time they are activated [60,61], or any obvious chromosomal features. Association with some but not other origins in unperturbed G1 and S phase does indicate, however, that Rrm3 interacts with a factor specific to some origins rather than a replication protein common to all pre-RCs or replisomes.
The role of Rrm3 at certain origins in HU is likely to be the cause of enhanced DNA synthesis and S phase progression in the rrm3Δ and rrm3-ΔN212 mutants. However, DNA-damage checkpoint activation in rrm3-ΔN212, but not rrm3-ΔN186, and synthetic lethality between mrc1Δ and rrm3-ΔN212, but not rrm3-ΔN186, suggest that there are conditions of replication stress other than exposure to HU that require the integrity of the Orc5-binding domain of Rrm3.
The role of Rrm3 in controlling DNA replication is not affected by inactivation of the ATPase/helicase activity (Fig 6C). Instead, it impairs Rrm3’s established function in facilitating fork progression through replication blocks, leading to the accumulation of DNA lesions that activate the DNA-damage checkpoint and can give rise to mutations. By identifying changes in chromatin composition combined with genetic assays we have identified Rad5 and Rdh54 as novel factors that contribute to the maintenance of genome stability in the absence of Rrm3’s ATPase/helicase activity. Rad5 defines an error-free pathway for the bypass of DNA polymerase blocking lesions [26–28,62–64]. As a structure-specific DNA helicase, Rad5 is capable of regressing replication forks in vitro [25]. Such a regressed fork is thought to provide an alternative template for DNA synthesis, generating enough nascent DNA to eventually bypass the replication block. The ATPase activity of Rad5 and the RING motif involved in polyubiquitination of PCNA [34–36,65] contribute to DNA damage tolerance in the absence of Rrm3. Evidence for a role of the ATPase activity of Rad5 in remodeling blocked replication forks has been obtained in vitro [25] whereas a role of Rad5-dependent polyubiquitination of PCNA in activating HR-dependent template switching has more recently been suggested [66]. Evidence that these two Rad5 activities can function independently, as we determined here in the rrm3Δ mutant, was also observed for bypass of MMS-induced lesions by sister-chromatid recombination [66].
Besides fork regression, Rad5 has also been implicated in DNA damage bypass by HR-dependent template switching between sister-chromatids [66] and the major HR factors Rad51, Rad52 and Rad54 as well as Sgs1 have been implicated in error-free DNA lesion bypass [67]. It was therefore surprising that Rdh54, a dsDNA translocase that is known to play a major role in meiotic, but not mitotic, HR [68–70], is recruited to chromatin when Rrm3 is absent—both in the presence and absence of HU. Rdh54 was only required in the absence of the ATPase/helicase activity of Rrm3, but not in the absence of the Orc5-binding domain, implicating Rdh54 in repair of DNA lesions that arise when Rrm3 cannot facilitate fork progression through replication blocks. Even though Rdh54 does not affect gene conversion repair of a DSB, a role specifically in repair that involves template switches was recently reported [71], and could be related to its increased chromatin association and the DNA-damage hypersensitivity of the rrm3Δ rdh54Δ and rrm3-K260A rdh54Δ mutants. That Rad51 was required for this repair, but Rad54 was not, suggests a Rad51-dependent HR pathway in mitotic cells where Rdh54 takes the place of Rad54.
Although it is unknown how Rdh54 acts in template switching, its activities in vitro seem compatible with those that may be required to rescue a paused fork. Like Rad5 and the human Rad5 ortholog, HTLF, Rdh54 is a dsDNA translocase of the SWI/SNF family [29,30,72]. In vitro, it can dislodge Rad51 from dsDNA and introduces negative supercoiling into dsDNA that can cause strand separation [29–31]. These Rdh54 activities could help to regulate restart at fork pause sites in Rad5-mediated pathways, such as fork regression/reversal or template switching, and in HR-mediated events. Whereas Rdh54 can remove proteins from dsDNA and remodel chromatinized DNA, an ability to remove bound proteins from DNA has not yet been shown for Rad5, and RecQ-like helicases are only capable of acting on forked DNA structures that are protein-free [32,33]. The synergistic interactions between rad5Δ and rdh54Δ in the absence of RRM3 clearly identify a requirement of Rdh54 outside of a Rad5 mechanism. In addition to facilitating template switching HR when error-free PRR is inactivated, Rdh54 could act in the avoidance of replication fork pausing in a manner similar to Rrm3 by removing certain proteins from dsDNA, such as shown for Rad51, which appears to have a tendency to associate with nonrecombinogenic dsDNA [29,31,70].
That the ATPase activity of Rrm3 is required in the absence of Rad5, Rdh54, Rad51 or Mph1, whereas the role of Rrm3 in controlling DNA replication is dispensable strongly suggests that the types of DNA damage checkpoint activating DNA lesions in the rrm3-K260A and rrm3-ΔN212 mutants are different, and that Rad5, Rdh54, Rad51 and Mph1 act on DNA lesions that form when replications forks are unable to move through obstacles, but not on DNA lesions that form during untimely DNA replication (rrm3-ΔN212 mutant). In contrast, Mrc1 and Tof1 were required for viability and DNA damage tolerance when either of the two Rrm3 activities was disrupted. Mrc1, the mediator of the replication stress checkpoint, mediates Rad53 phosphorylation specifically in response to replication fork pausing, leading to intra-S checkpoint activation and inhibition of late-origin firing [44,59]. That synthetic lethality between rrm3Δ and mrc1 is limited to those mrc1 alleles that cause DNA damage accumulation during S phase [49], whereas the checkpoint function of Mrc1 is dispensable [19,49] suggests that the additive accumulation of S phase damage due to lack of both, Mrc1 and Rrm3, is lethal and suggests that dysregulated replication in the rrm3-ΔN212 mutant (Fig 6B) also leads to S phase damage, consistent with our observation of Rad9-dependent activation of Rad53.
Finally, the new N-terminal Rrm3 function in controlling DNA replication is separated from Rrm3’s established C-terminal function as an ATPase/helicase in facilitating fork progression not only by the differential requirement for Rad5, Rdh54, Rad51 and Mph1, but also by different spontaneous and DNA-damage induced mutation spectra. This supports that the N-terminal tail is neither involved in the recruitment of Rrm3 to active replication forks nor in facilitating fork progression through protein-bound sites, and that a separate replisome binding site is likely to be located in the ATPase/helicase domain. The accumulation of GCRs and point mutations in the ATPase /helicase mutant, spontaneously or induced by HU or MMS, could be indicative of DNA break formation as a result of replication fork stalling. In contrast, the Orc5-binding domain mutant did not accumulate GCRs under any conditions, suggesting wildtype levels of DNA breaks, including in HU and MMS, but increasingly formed point mutations. That these point mutations formed specifically in response to HU, but not MMS, suggests that they arise during the enhanced DNA synthesis that occurs in this mutant in HU.
In summary, this study has revealed a 26-residue region in Rrm3 that is critical for a novel, ATPase-independent function of Rrm3 in preventing untimely DNA replication and for binding Orc5, which appear to be mechanistically linked. We also identified association of Rrm3 with a subset of replication origins and the dependence of this association on the N-terminal tail under replication stress, but not in unperturbed cells. Genome-wide ChIP and quantification of DNA synthesis in cells expressing the new rrm3 alleles will help to reveal the regions undergoing untimely DNA replication and provide further insight into the mechanism at replication origins underlying continued DNA synthesis under replication stress and lethality with mrc1Δ. That yeast has two DNA helicases (Rrm3, Pif1) that belong to the Pif1 family, whereas multicellular eukaryotes where the Pif 1 helicase family is conserved [2] typically have one (e.g. Pif1 in humans), might be an indication that Rrm3’s role in DNA replication is highly specialized to control replication and facilitate fork progression in genomic regions that are distinctively organized in yeast and to deal with the high gene density imposed on its small genome that requires tight coordination between replication initiation and ongoing transcription.
For double isotope labeling of lysine and arginine, yeast strain KHSY5144 (lys2Δ arg4Δ) was grown at 30°C with and vigorous shaking for at least ten generations in “heavy” medium (6.9 g/l yeast nitrogen base without amino acids [Formedium], 1.85 g/l amino acid dropout mixture without arginine and lysine [Kaiser formulation, Formedium], 2% glucose, 15 mg/l [13C6] L-arginine and 30 mg/l [13C6] or [13C6, 15N2] L-lysine). KHSY5143 (lys2Δ arg4Δ rrm3Δ) was grown in “light” medium, containing 15 mg/l L-arginine and 30 mg/l L-lysine at 30°C with and vigorous shaking.
Chromatin was isolated using a method adapted from [73]. Approximately 4 x 109 cells were resuspended in 10 ml of 100 mM PIPES/KOH, pH 9.4, 10 mM dithio- treitol (DTT), 0.1% sodium azide, then incubated for 10 min at room temperature, followed by incubation in 10 ml of 50 mM KH2PO4/K2HPO4, pH 7.4, 0.6 M sorbitol, 10 mM DTT, containing 200 mg/ml Zymolyase-100T and 5% Glusulase at 37°C for 30 min with occasional mixing. Spheroplasts were washed with 5 ml of ice-cold wash buffer (20 mM KH2PO4/K2HPO4, pH 6.5, 0.6 M sorbitol, 1 mM MgCl2, 1 mM DTT, 20 mM beta-glycerophosphate, 1 mM phenyl-methylsulfonyl fluoride (PMSF), Protease inhibitor tablets (EDTA free, Roche) and resuspended in 5 ml of ice-cold wash buffer. The suspension was overlaid onto 5 ml of 7.5% Ficoll-Sorbitol cushion buffer (7.5% Ficoll, 20 mM KH2PO4/K2HPO4, pH 6.5, 0.6 M sorbitol, 1 mM MgCl2, 1 mM DTT, 20 mM beta-glycerophosphate, 1 mM PMSF, Protease inhibitor tablets) and spheroplasts were spun through the cushion buffer at 5000 rpm for 5 min to remove proteases derived from Zymolyase-100T. Pelleted spheroplasts were resuspended in 200 ml of ice-cold wash buffer and dropped into 18% Ficoll, 20 mM KH2PO4/K2HPO4, pH 6.5, 1 mM MgCl2, 1 mM DTT, 20 mM beta-glycerophosphate, 1 mM PMSF, Protease inhibitor tablets, 0.01% Nonidet P-40, with stirring. After homogenization, unbroken cells were removed by two spins (5000 x g for 5 min at 4°C). Nuclei were pelleted by centrifugation at 16,100 x g for 20 min and the cytoplasmic fraction removed. After washing in ice-cold wash buffer, nuclei were resuspended in 200 ml of EB buffer (50 mM HEPES/KOH, pH 7.5, 100 mM KCl, 2.5 mM MgCl2, 0.1 mM ZnSO4, 2 mM NaF, 0.5 mM spermidine, 1 mM DTT, 20 mM beta-glycerophosphate, 1 mM PMSF, protease inhibitor tablets) and lysed by addition of Triton X-100 to 0.25%, followed by incubation on ice for 10 min. Lysate was overlaid on 500 ml of EB buffer, 30% sucrose, 0.25% Triton X-100, and spun at 12,000 rpm for 10 min at 4°C. The top layer was removed and the chromatin pellet was washed in 1 ml of EB buffer, 0.25% Triton X-100 and spun at 10,000 rpm for 2 min at 4°C.
The chromatin pellet was resuspended in 40 μl of 1.5x Tris-Glycine SDS sample buffer and incubated for 2 min at 85°C. DTT was added to a final concentration of 5 mM and incubated for 25 min at 56°C. Iodoacetamide was added to 14 mM final concentration and incubated for 30 min at room temperature in the dark. DTT was added to a final concentration of 5 mM and incubated for 15 minutes at room temperature in the dark. The protein mixture was diluted 1:5 in 25 mM Tris-HCl, pH 8.2; CaCl2 was added at a final concentration of 1 mM, and trypsin was added at 4–5 ng/μl, followed by incubation at 37°C overnight. Trifluoroacetic acid was added to 0.4% final concentration and centrifuged at 2,500 x g for 10 min at room temperature. Peptides in the supernatant were desalted using reverse-phase tC18 SepPak solid-phase extraction cartridges (Waters). The sample was eluted with 5 ml of 50% acetonitrile and lyophilized. The lyophilized product was resuspended in 0.1% formic acid prior to tandem mass spectrometric analysis on an LTQ Orbitrap XL (Thermo). Scans on the Orbitrap were obtained at a mass resolving power of 60000 at m/z 400 and top 10 abundant ions were selected for fragmentation in the LTQ ion trap. Further processing of the RAW files was done in MaxQuant version 1.5.3.30 [74] against the Saccharomyces genome database (SGD). A database of known contaminants in MaxQuant was used as well as constant modification of cysteine by carbamidomethylation and variable modification of methionine oxidation. The first search tolerance was set at 20 ppm, then 8 ppm tolerance for the main search. Fragment ion mass tolerance was set to 0.5 Da and the minimum peptide length was 6 amino acids. Unique and razor peptides were used for identification and the false discovery rate was set to 1% for peptides and proteins [74,75]. Statistical analysis of the data was carried out with Perseus software using an approach by Benjamini and Hochberg [76].
Yeast strains used in this study are derived from S288C background. Strains for SILAC labeling and chromatin fractionation were derived from KHSY5036 (MAT ɑ, ura3-52, trp1Δ63, his3Δ200). Yeast strains used in DNA-damage sensitivity and mutation assays were derived from KHSY802 (MAT ɑ, ura3-52, trp1Δ63, his3Δ200, leu2Δ1, lys2Bgl, hom3-10, ade2Δ1, ade8, hxt13::URA3). Gene deletions were carried out as described using HR-mediated integration of a selectable cassette [77]. Mutants containing more than one gene deletion or mutation were obtained by random spore isolation from diploids heterozygous for the desired mutations. Point mutations were introduced in plasmids by site-directed mutagenesis (QuickChange, Stratagene) and verified by sequencing. RRM3 truncations were created in plasmid pKHS137 and plasmid pJG4-5* using HR-mediated integration in KHSY2331 (lig4Δ) and verified by sequencing. Yeast was grown at 30°C in yeast extract (10g/l), peptone (20g/l), dextrose (20g/l), media (YPD) with or without Bacto agar, or in synthetic complete (SC) media (yeast nitrogen base 6.7g/l, dextrose 20g/l) supplemented with the appropriate amino acid mix. All yeast strains and plasmids used in this study are listed in S2 and S3 Tables, respectively.
Gross-chromosomal rearrangement (GCR) rates were determined by fluctuation analysis by taking the median rate of at least 15 cultures from at least two isolates and are shown with 95% confidence intervals [78–80]. Cells with GCRs were identified by their resistance to canavanine and 5-fluoro-orotic acid (Canr 5-FOAr), which is indicative of simultaneous inactivation of CAN1 and URA3 on chromosome V, on selective media as previously described [80]. To observe GCRs after exposure to DNA-damaging agents HU and MMS, cells were grown to an OD600 = 0.5, released into medium containing the drug and were cultured for 2 hours at 30°C. Cells were then released into fresh YPD medium and grown for 24 hours before being plated. Forward mutation rates were determined by fluctuation analysis by method of the median as previously described and are shown with 95% confidence intervals [78,81,82]. Briefly, fifteen cultures from at least two different isolates were grown overnight at 30°C in 3 ml of YPD media. Dilutions were plated on YPD agar to determine the viable cell count, and 500 μl was plated on synthetic media supplemented with 60 μg/ml canavanine, but lacking arginine to select for can1 mutants. To obtain forward mutation rates and GCR rates after exposure to MMS and HU, cells were grown to OD600 = 0.5, released into medium containing the drug and cultured for 2 hours at 30°C. Cells were then released into fresh YPD medium and grown for 24 hours before being subjected to fluctuation analysis.
Cell cultures were grown as indicated either in YPD or selective media to maintain plasmids (SC-Leu) to OD600 = 0.5, and 10-fold serial dilutions were spotted on YPD or SC-Leu supplemented with methyl methanesulfonate (Sigma Aldrich) or hydroxyurea (US Biological) at the indicated concentrations. Colony growth was recorded after 2 to 3 days of incubation at 30°C.
Cells were prepared for DNA content analysis as previously described [83]. Briefly, cells were washed and fixed in 70% ethanol for 1 hour at room temperature, sonicated in 50 mM sodium citrate (pH 7), washed in 50 mM sodium citrate (pH7), and RNAse A was added to a final concentration of 250 μg/ml. After overnight incubation at 37°C, cells were washed in 50 mM sodium citrate and stained with Sytox green (Molecular Probes) at a final concentration of 1 μM in the dark at room temperature for 1 hour immediately prior to fluorescence-activated cell sorting (FACS) on a BLD LSR II analyzer. The distribution of cells throughout the cell cycle phases was quantified with the FlowJo v8.3.3 software.
Cells were grown to OD600 = 0.5 in YPD at 30°C, synchronized in G1 with α-factor (15 μg/ml), released into fresh, pre-warmed YPD. Cells were harvested after 30 min, immediately put on ice and adjusted for cell number. Whole cell extracts were prepared with 20% trichloroacetic acid as previously described [84] and separated by 10% SDS-PAGE for Western blot analysis. Phospho-specific Rad53 antibody EL7 was a gift from A. Pellicioli (FIRC Institute of Molecular Oncology Foundation, Milan, Italy). Antibody sc-6680 (SCBT) was used for Mcm2, ab34680 (Abcam) for Adh1, ab46765 (Abcam) for histone H3, AS07214 (Agrisera) for Rfa1, and MMS-150P (Covance) for the myc-epitope.
Cells expressing Orc5-V5-6xHis and either Rrm3-myc, rrm3-ΔN186-myc, or rrm3-ΔN212-myc were grown to an OD600 ~ 1.0 and cells were harvested by centrifugation. Cells were re-suspended in cell lysis buffer (50 mM HEPES pH 7.5, 10% v/v glycerol, 140 mM NaCl, 1 mM EDTA, 0.5% Igepal, 1 mM PMSF, EDTA-free Protease inhibitors (Pierce)) and vortexed in a cell disruptor with glass beads for 5 minutes. Cell lysates were cleared by centrifugation at 4°C and lysates were split in half. 1 mM PMSF and 20 mM MgCl2 was added to all samples, and 300 U of Benzonase (Sigma) was added to one half of the sample whereas the other half was not treated. Cell lysates were placed on ice for 30 minutes, added magnetic beads coated with to C-MYC antibody and incubated with mixing for 2 h at 4°C. Beads were washed thoroughly 10 times in 1 ml of cell lysis buffer, re-suspended in protein sample buffer, and boiled for 5 min. Samples were separated on a 7% SDS-PAGE gel and presence of Rrm3.myc, rrm3-ΔN186.myc, rrm3-ΔN212.myc, Orc5.V5.6xHIS and GAPDH was determined by western blotting using monoclonal antibodies against myc (Covance), V5 (Sigma), and GAPDH (Pierce).
Fifty millilters of cells corresponding to 2.0 x 107 cells/ml were incubated for 15 min at room temperature with or without 1% formaldehyde and harvested. Cell pellets were washed twice with PBS, re-suspended in 600 μl of cell lysis buffer (50 mM HEPES pH 7.5, 140 mM NaCl, 1 mM EDTA, 1% Triton X-100, 0.1% sodium deoxycholate, 1 mM PMSF, 1 mM benzamidine, 1 μg/ml aprotinin, 1 μg/ml leupeptine, and 1 ug/ml pepstatin), lysed in a cell disruptor with glass beads at 4°C, and sonicated four times for 20 s each at 4°C. The lysate was clarified by centrifugation and whole cell extract was added to magnetic beads coated with c-myc antibody or V5 antibody, followed by incubation for 60 min at 4°C. Beads were washed four times in cell lysis buffer, three times in TE buffer (10 mM Tris-Cl, pH 7.5, 1 mM EDTA), and re-suspended in TE/1% SDS buffer. Sample was incubated at 65°C for 10 min and processed for DNA purification as previously described [85]. Sequences of primers used for PCR are available upon request.
Cells grown to stationary phase were transferred to acidic media (pH 3.5) and grown to logarithmic phase. Cells were synchronized in G1 phase over two hours with the addition of 2 μg/ml alpha-factor (Genscript) every hour. Cells were washed twice with sterile water. 2.5×108 yeast cells were pelleted, resuspended in 1 ml 60% methanol, and disrupted by 5 consecutive freeze and thaw cycles using liquid nitrogen and warm water, followed by incubation at -20°C for 90 minutes and boiling at 100°C for 3 minutes. The lysate was centrifuged at 19000×g for 15 minutes and the supernatant frozen in liquid nitrogen. Methanol was evaporated in a SpeedVac (Thermo Scientific) and the residue was rehydrated in 100 μl Ultra-pure H2O (Invitrogen, GIBCO). Determination of cellular dNTP concentration was performed as earlier described [86]. Each extraction was performed at least in triplicate.
Yeast strain EGY48 containing reporter plasmid pSH18-34 was co-transformed with a lexA-fusion bait vector (pEG202) and a B42-tagged prey vector (pJG4-5*), and transformants were selected on synthetic complete (SC) medium plates lacking histidine and tryptophan (SC-Trp-His). Single transformants were purified and resuspended in liquid SC-His-Trp medium containing 2% galactose and 1% raffinose, and grown overnight at 30°C. Cultures were diluted to OD600 = 0.2, grown to OD600 = 0.8, and 2-fold serial dilutions were spotted onto SC-His-Trp-Leu containing either 2% glucose or 2% galactose with or without hydroxyurea (HU) or methyl methanesulfonate (MMS). Colony growth was recorded after 72 hours.
Diploids heterozygous for the desired mutant alleles (rrm3::TRP1, mrc1::HIS3) transformed with plasmid-borne alleles of RRM3 (linked to LEU2) were sporulated by nitrogen starvation in 0.1% potassium acetate for 5 days at 30°C with vigorous shaking. Asci were incubated in the presence of 500 μg/ml of zymolase in 1M sorbitol for 20 min at 30°C, enriched for meiotic products as previously described [87] and plated on nonselective media (YPD). After incubation for 2 days at 30°C, colonies were spotted on SC-Leu media and 100 leu+ colonies genotyped further by spotting on SC-Leu-Trp, SC-Leu-His, and SC-Leu-Trp-His to identify rrm3Δ, mrc1Δ and rrm3Δ mrc1Δ mutants, respectively, all harboring various plasmid borne RRM3 alleles.
Cells were grown to early log phase and incubated with 100 mM hydroxyurea for 2 hours at 30°C with shaking. Cells were fixed with 3.7% formaldehyde, permeablized with ethanol, and mounted in Vectashield medium with DAPI (Vectorlabs). Cells were examined using an EVOS fluorescence microscope and grouped into unbudded (G1 phase), small-budded (S phase, bud is up to one-fourth of the diameter of mother cell), or large-budded (G2/M, bud is equal to or greater than one-fourth of the diameter of the mother cell). 200 cells for each yeast strain were scored.
|
10.1371/journal.ppat.1000531 | Differential Tolerance to Direct and Indirect Density-Dependent Costs of Viral Infection in Arabidopsis thaliana | Population density and costs of parasite infection may condition the capacity of organisms to grow, survive and reproduce, i.e. their competitive ability. In host–parasite systems there are different competitive interactions: among uninfected hosts, among infected hosts, and between uninfected and infected hosts. Consequently, parasite infection results in a direct cost, due to parasitism itself, and in an indirect cost, due to modification of the competitive ability of the infected host. Theory predicts that host fitness reduction will be higher under the combined effects of costs of parasitism and competition than under each factor separately. However, experimental support for this prediction is scarce, and derives mostly from animal–parasite systems. We have analysed the interaction between parasite infection and plant density using the plant-parasite system of Arabidopsis thaliana and the generalist virus Cucumber mosaic virus (CMV). Plants of three wild genotypes grown at different densities were infected by CMV at various prevalences, and the effects of infection on plant growth and reproduction were quantified. Results demonstrate that the combined effects of host density and parasite infection may result either in a reduction or in an increase of the competitive ability of the host. The two genotypes investing a higher proportion of resources to reproduction showed tolerance to the direct cost of infection, while the genotype investing a higher proportion of resources to growth showed tolerance to the indirect cost of infection. Our findings show that the outcome of the interaction between host density and parasitism depends on the host genotype, which determines the plasticity of life-history traits and consequently, the host capacity to develop different tolerance mechanisms to the direct or indirect costs of parasitism. These results indicate the high relevance of host density and parasitism in determining the competitive ability of a plant, and stress the need to simultaneously consider both factors to understand the selective pressures that drive host–parasite co-evolution.
| Parasites are a potent selective force, as they reduce the fitness of their hosts through a direct cost of infection, due to parasitism itself, and an indirect cost, due to modification of the competitive ability of the infected host. Theory predicts that fitness reduction will be higher under the combined effects of costs of parasitism and host population density than under each factor separately, but experimental support for this prediction is scarce, and derives mostly from animal–parasite systems. We have analysed the combined effects of host density and costs of infection using the plant virus Cucumber mosaic virus and its host plant Arabidopsis thaliana. The interaction between these factors may result in a reduction or an increase of plant competitive ability, depending on host genotype, which determines the plasticity of life-history traits and consequently, different tolerance mechanisms to the combined effects of plant density and direct or indirect cost of parasitism. These tolerance mechanisms are associated with resource reallocation. Our results stress the relevance of the interaction between host and parasite traits in determining the outcome of infection contributing to understand the selective pressures that drive host–parasite co-evolution.
| Competition is one of the major selection factors in nature, acting at all phases of development [1]. The key role of competition in shaping evolution is one of the bases of the Darwinian Theory, underlying the colonization success, expansion and suppression of genotypes and species. The relevance of competition for ecology and evolutionary biology has lead to a large body of work based on life-history theory, which states that competitive ability depends on trade-offs between the capacity of an organism to grow, survive and reproduce [2],[3]. The optimal amount of resources allocated to each of these components may be modified depending on environmental conditions in order to maximize the organism's fitness [4]. Experimental analyses have shown that competition due to increased population density may induce severe alterations in life history traits that are components of competitive ability such as mortality rate [5], time span to maturity [6], adult size [7] or fecundity [8].
The impact of population density on the effects of predation and herbivory have been widely investigated [9]–[11]. In contrast, density dependent effects on host-parasite systems, have received much less attention [12]. In host-parasite systems there are different competitive interactions: intra-class competition among uninfected hosts or among infected hosts, and inter-class competition between uninfected and infected hosts. Each interaction may have different effects on host life-history traits, resulting in a direct cost of infection, due to parasitism itself, and in an indirect cost, due to modification of the competitive ability of the infected host, both being modulated by host population density [13]. Theory predicts that fitness reduction will be higher under the combined effects of host population density and parasitism than under each factor separately [14]–[17]. The outcome of this interaction may also depend on the prevalence of infection, the effect on the host competitive ability being less severe as prevalence increases [13]. Experimental analyses of these predictions derive mostly from animal-parasite systems [18]–[22], and few similar ones have been carried out with plants [23]–[26].
Furthermore, hosts have evolved defences against parasites, including tolerance mechanisms. Here, we define tolerance as the host ability to reduce the effect of infection on its fitness [27]–[29]. Theoretical and experimental analyses support that tolerance involves modification of life-history traits in order to maximize progeny production through resource reallocation from growth to reproductive structures [4],[27],[30]. In plants, tolerance has been shown to act in response to the combined effects of increasing population density and herbivory [10], but its role under the combined effects of host population density and parasitism has not been previously analysed. Thus, the interaction between parasitism and host density remains largely an unexplored aspect of the evolutionary ecology of parasites that requires further experimentation with a larger array of systems.
We have addressed this question in a plant-virus system, using the widespread virus Cucumber mosaic virus (CMV-Bromoviridae), a generalist parasite and an important plant pathogen [31], and its host plant Arabidopsis thaliana L. (Heynh.) (Brassicaceae) (from here on, Arabidopsis), a model organism for molecular plant genetics and, more recently, for plant-parasite co-evolution [30],[32],[33]. The capacity of Arabidopsis genotypes to modify life-history traits depends on the allometry between vegetative and reproductive organs [30],[33]. Hence, three Arabidopsis wild genotypes (referred to as accessions) with different allometry were analysed to estimate costs of CMV infection on life-history traits related to different components of competitive ability at different plant densities and virus prevalence. Our results indicate that the interaction between plant density and costs of infection may result in a reduction or an increase of plant competitive ability. It is shown that host genotype determines the plasticity of life-history traits and consequently, different tolerance mechanisms to the direct or indirect cost of parasitism, which also depend on plant density.
Since the capacity of Arabidopsis genotypes to modify life-history traits depends on the allometry between vegetative and reproductive organs [30],[33], three Arabidopsis accessions with different allometry were selected for experimentation. Accessions Cen-1 and Ler, with a short life cycle and a higher proportion of resources dedicated to reproduction than to growth, and accession Boa-0, with a long life cycle and a higher proportion of resources invested to growth than to reproduction [30]. Individuals of each accession were grown at three different plant densities chosen based on a previous experiment (Fig. S1), which cover from no resource competition to crowded conditions: 1, 2 and 4 plants per pot, arranged in all possible combinations of infected and mock-inoculated plants to simulate different CMV prevalences (Fig. 1). Plants of each accession were inoculated with the LS strain of CMV. The costs of CMV infection on life-history traits related to different components of competitive ability were quantified: rosette weight (RW), as a measure of growth effort; inflorescence weight, including seeds, (IW) as a measure of total reproductive effort; and seed weight (SW), as a measure of progeny production (See Table S1).
The impacts of host population density and viral infection on the competitive ability of Arabidopsis were analysed separately. The impact of host population density was measured as the differential performance of RW, IW and SW of plants on monocultures of CMV-infected and on monocultures of mock-inoculated plants at the three plant densities, i.e., in the intra-class treatments I, I/I, I/I/I/I, and M, M/M, M/M/M/M, for 1, 2 and 4 plants per pot, respectively. Two-way ANOVA using plant density and accession as factors showed that in both infected and mock-inoculated plants, the three traits depended on plant density and accession (F2,126≥13.2, P≤1×10−5), but only RW and IW depended on the interaction between both factors (Table S2). Therefore, each accession was analysed separately. In the three accessions, each studied life-history trait was significantly reduced as plant density increased, both for CMV-infected and mock-inoculated plants (F2,44≥7.6, P≤0.002) (Fig. 2 and Table S3). Hence, competition for resources occurred when more than one plant, either mock-inoculated or infected, grew in the same pot, the higher the density, the stronger the decrease of plant growth and progeny production.
The direct cost of CMV infection was determined as the impact of parasitism on intra-class competitive ability comparing plant performance of RW, IW and SW between monocultures of CMV-infected and mock-inoculated plants. All traits differed among plant conditions (infected and mock-inoculated), plant densities and accessions (F1,272≥11.73, P≤7×10−4; F2,272≥63.06, P≤1×10−5; F2,272≥41.10, P≤1×10−5, respectively), interactions being significant for RW and SW, but not for IW (Table S4). Thus, we analysed the direct cost of infection in each plant density and accession separately. For the three accessions, a general reduction of RW, IW and SW was observed at each density for infected plants compared with mock-inoculated ones (F1,29≥4.37, P≤0.046), with the exception of IW at 1 and 2 plants per pot, and SW at 1 plant per pot in accession Boa-0 (F1,29≤0.30, P≥0.619) (Fig. 2 and Table S5). Thus, accessions Cen-1 and Ler suffered a direct cost of infection on all traits, both in the absence or presence of competition for resources. In contrast, CMV infection affected the growth (RW) of Boa-0 at all densities, but reproductive traits (IW and SW) of this accession were affected only under severe resource competition.
The influence of host population density on the direct cost of infection was analysed comparing the effect of infection, defined as the ratio between the value of each trait in infected and in mock-inoculated plants (Traiti/Traitm, i and m denote infected and mock-inoculated plants, respectively), among plant densities (Fig. 3). The effect of infection significantly differed with plant density for all traits (F2,126≥4.02, P≤0.031), but only SWi/SWm differed significantly between accessions (F2,126 = 4.13, P = 0.026) (Table S6). In Boa-0 plants, the direct cost of infection on RW and IW did not significantly differ as plant density increased (F2,44≤2.33, P≥0.108), but it increased for SW (F2,44 = 6.89, P = 0.002). In Cen-1 and Ler, direct costs of CMV infection decreased as plant density increased in all traits (F2,44≥4.88, P≤0.027), with the exception of IW in Cen-1 (F2,44 = 1.08, P = 0.348) (Table S7). Defining tolerance as stated in the introduction, these results indicate that Cen-1 and Ler increase their tolerance to the direct cost of CMV infection as plant density increases, while the tolerance of Boa-0 decreases when competition occurs.
The indirect cost of infection is determined by the difference between intra- and inter-class competitive ability of infected and non-infected plants [13]. Hence, the performance of infected and mock-inoculated plants was compared between both environments (Fig. 2). RW, IW and SW differed between plant densities and accessions in mock-inoculated and infected plants (F1,303≥11.42, P≤8×10−4; F2,303≥25.65, P≤1×10−5, respectively), but significant differences between classes of competition were found only for infected plants (F1,303≥11.08, P≤0.001) (Table S8). Hence, analyses were done for each plant density and each accession separately. In the three accessions, the value of RW, IW and SW was similar for mock-inoculated plants from M/I and M/M treatments (F1,29≤2.51, P≥0.124) and was lower for infected plants from M/I than from I/I treatments (F1,29≥4.64, P≤0.047), with the exception of SW in Boa-0 (F1,29 = 0.02, P = 0.889). At 4 plants per pot, trait values of Boa-0 did not differ between intra- and inter-class treatments of infected plants (F4,74≤1.50, P≥0.210); in mock-inoculated Boa-0 plants RW and IW values did not differ between classes (F4,74≤1.17, P≥0.329), while SW values were higher for M/M/M/M than for the inter-class treatments (F4,74 = 2.99, P = 0.024). Mock-inoculated plants of Cen-1 and Ler showed lower values of all traits for M/M/M/M than for M/M/I/I, M/I/M/I and M/I/I/I, but not than M/M/M/I (F4,74≥2.80, P≤0.033) (Fig. 2). For infected individuals of these accessions, RW did not differ between intra and inter-class treatments (F4,74≤1.18, P≥0.327); IW and SW were higher for I/I/I/I than for M/M/M/I (F4,74≥2.80, P≤0.033), but no differences were found with the other inter-class treatments (Table S1 and Table S9). These results show that there is an indirect cost of CMV infection that depends on host density, CMV prevalence and accession.
Indirect costs of CMV infection were further analysed by comparing the effect of infection (Traiti/Traitm) between intra- and inter-class treatments (Fig. 3A). The effect of infection on RW, IW and SW significantly differed according to class of competition and accession (F1,303≥7.97, P≤0.005; F2,303≥3.33, P≤0.037, respectively), but only RWi/RWm depended on plant density (F1,303 = 13.30, P≤3×10−4). In addition, a significant interaction between these three factors was detected on all traits (F2,303≥2.93, P≤0.049) (Table S10). Therefore, the effect of infection was analysed for each accession at each plant density separately. The effect of infection on Boa-0 plants differed depending on the trait: for RW it was higher for inter- than for intra-class treatments at both plant densities (F1,29 = 3.63, P = 0.043; F1,74 = 10.72, P = 0.002, for 2 and 4 plants per pot, respectively); for IW it was higher for inter- than for intra-class treatments at 2 plants per pot (F1,29 = 4.88, P = 0.037), but not at 4 plants per pot (F1,74 = 0.25, P = 0.619, respectively); and for SW no difference was found at 2 plants per pot (F1,29 = 0.01, P = 0.931), but the effect of infection was higher on intra- than on inter-class treatments at 4 plants per pot (F1,74 = 9.52, P = 0.004) (Fig. 3A). The effect of infection on Cen-1 and Ler plants was higher on inter- than on intra-class treatments for all traits at both plant densities (F1,29≥6.36, P≤0.023; F1,74≥9.67, P≤0.003 for 2 and 4 plants per pot, respectively) (Fig. 3A and Table S11). Thus, in accessions Cen-1 and Ler there is an indirect cost of CMV infection on both growth and reproductive traits. In contrast, accession Boa-0 did not show costs on reproductive traits at high plant density, which indicates an increased tolerance to the indirect cost of infection.
Host density dependence of the indirect cost of infection was analysed by quantifying the ratio between the effect of infection on inter-class competition and on intra-class competition treatments [(Traiti/Traitm)Inter-class/(Traiti/Traitm)Intra-class] (Fig. 3A). For all traits, this ratio significantly depended on the accession (F2,219≥4.63, P≤0.011), but not on the plant density (F1,219≤3.38, P≥0.067). However, the interaction between both factors was significant (F2,219≥5.17, P≤0.006), indicating that the effect of plant density in the indirect cost of infection differed between accessions (Table S12). Hence, the ratio between the effect of infection on inter-class competition and intra-class competition treatments was analysed for each accession separately. In Boa-0, the indirect cost of infection on RW did not differ between plant densities (F1,74 = 0.78, P = 0.379). However, for IW it was higher at 2 than at 4 plants, while for SW it was higher at 2 than at 4 plants per pot (F1,74≥4.33, P≤0.041). In Cen-1 and Ler, the indirect cost of infection on IW did not differ between plant densities (F1,74≤2.39, P≥0.127), while on RW it was higher at 4 plants per pot in Cen-1 (F1,74 = 4.18, P = 0.040), and at 2 plants per pot in Ler (F1,74 = 55.54, P = 1×10−5). The indirect cost on SW was higher at 4 plants per pot for both accessions (F1,74≥4.57, P≤0.039) (Fig. 3A and Table S13). Similar results were obtained when treatments with the same prevalence at 2 and 4 plants per pot were compared (M/I, and M/M/I/I or M/I/M/I). Hence, plant density affects the indirect cost of infection, and may increase or decrease such cost depending on host genotype.
The effect of parasite prevalence on the indirect cost of infection was analysed by comparing the effect of infection (Traiti/Traitm) between the intra-class treatment and the various inter-class treatments at 4 plants per pot (Fig. 3B). Indirect cost of infection depended on the CMV prevalence and the accession for RW and IW (F4,210≥3.14, P≤0.016; F2,210≥8.94, P≤2×10−4, for prevalence and accession, respectively), but not for SW (F4,210≤1.57, P≥0.184; F2,219≤1.36, P≥0.259, respectively), the interaction being significant for all traits (F8,219≥2.12, P≤0.045; Table S14). Hence, we analysed the effect of parasite prevalence for each accession separately. In Boa-0, the indirect cost of infection did not differ between interclass treatments for RW and IW (F4,74≤2.17, P≥0.113), and it was higher in M/I/I/I than in the rest of inter-class treatments for SW (F4,74 = 4.17, P = 0.004). In Cen-1 and Ler, indirect costs on RW increased as prevalence increased (F4,74≥7.63, P≤1×10−5), and costs in SW did not differ among inter-class treatments (F4,74≥2.43, P≤0.056). In Cen-1 the indirect cost of infection on IW was lowest at the lowest prevalence, while in Ler it was highest at the lowest prevalence (F4,74≥9.24, P≤1×10−5). Hence, in all accessions prevalence differentially affected the indirect cost of infection on each trait, seed production only being affected in Boa-0 plants, where high prevalence of infection reduced tolerance (Table S1 and Table S15).
Tolerance to CMV infection in Arabidopsis under non-competitive conditions is associated with changes in resource allocation patterns [28]. To analyse whether this mechanism is also involved in tolerance to virus infection at increased population density, the relationship between SW and RW was compared between infected and mock-inoculated plants for each treatment (Fig. 4). The SW/RW ratio varied according to plant condition (infected or mock-inoculated), plant density and accession (F1,702≥3.92, P≤0.045; F2,702≥8.71, P≤2×10−4; F2,702≥196.46, P≤1×10−5) (Table S16). Thus, the SW/RW ratio was compared between infected and mock-inoculated plants for each accession at each plant density (Table S17). In intra-class treatments, the SW/RW value of Boa-0 was higher on infected than on mock-inoculated plants at 1 plant per pot (F1,29 = 12.87, P≤1×10−5), but increasingly lower at 2 and 4 plants per pot (F1,29 = 0.16, P = 0.696; F2,29 = 18.01, P≤1×10−5,, for 2 and 4 plants per pot, respectively). In Cen-1 and Ler, SW/RW was higher on mock-inoculated than on infected plants at 1 plant per pot (F1,29≥6.96, P≤1×10−3), it was similar at 2 plants per pot (F1,29≤1.32, P≥0.260) and it was higher on infected plants at 4 plants per pot in Ler (F1,29 = 6.10, P = 0.019), but not in Cen-1 (F1,29 = 0.80, P = 0.498), (Fig. 4). In inter-class treatments, Boa-0 SW/RW value was higher for infected than for mock-inoculated plants for all treatments (F1,29≥4.62, P≤0.035), except for M/I/I/I treatment (F1,29 = 0.06, P = 0.802). However, in Cen-1 and Ler at 2 plants per pot, SW/RW was higher on infected than on mock-inoculated plants (F1,29≥5.97, P≤0.021), while at 4 plants per pot this ratio was higher in mock-inoculated than in infected plants for M/M/M/I treatment (F1,29≥5.88, P≤0.024), no differences were observed for M/M/I/I and M/I/M/I (F1,29≥0.67, P≤0.420), and it was higher for infected plants for M/I/I/I treatment (F1,29≥4.21, P≤0.042). Therefore, tolerance to virus infection appears associated with increased resource allocation to seed production, with and without host competition. The degree of this reallocation depended on Arabidopsis accession.
Our results demonstrate that the interaction of host population density and parasite infection is highly relevant in determining the competitive ability of a plant. The outcome of this interaction varies for different components of competitive ability, and depends on host genotype and infection prevalence. Thus, in Arabidopsis accessions Cen-1 and Ler that dedicate a higher proportion of resources to reproduction than to growth (Fig. 2), the direct cost of CMV infection decreased as plant density increased, indicating a density-dependent tolerance to CMV infection (Fig. 3). In contrast, these accessions showed a lower level of tolerance to CMV in the absence of competition (Fig. 2, and [30]). The opposite behaviour was observed in accession Boa-0 that invests a higher proportion of resources to growth than to reproduction, since the high tolerance to CMV in the absence of competition (Fig. 2, and [30]) decreased when resource limitation occurred. Boa-0 plants have significantly higher biomass and need more resources to complete their life cycle than Cen-1 or Ler plants (Fig. 2). Resource availability may limit plant plasticity [34], and the lower amount of resources available under competition may explain the reduced fraction of resources invested in reproduction in infected Boa-0 plants and their reduced tolerance when host density increases. In Cen-1 and Ler, competition for resources is more intense as population density increases in monocultures of mock-inoculated than of infected plants (Fig. 2). Thus, at higher population densities a larger proportion of resources is available for each infected plant compared with mock-inoculated ones, which may explain their higher plasticity in resource allocation (Fig. 4) and consequently, their tolerance to the direct cost of infection under competition than under unlimited resources. Thus, modification of the resource allocation pattern may partly determine genotype-specific tolerance to the combined effects of plant density and the direct cost of CMV infection, which results in increased competitive ability of Cen-1 and Ler infected plants, but not of Boa-0. However, factors other than resource allocation between rosette growth and seed production may contribute also to tolerance, as suggested by the observed SW/RW differences between Cen-1 and Ler.
Most experimental studies of host-parasite interactions quantify only the direct cost of infection [35]. Our results indicate that CMV infection has also an indirect cost on Arabidopsis, which has evolved genotype-specific tolerance to it. This tolerance suggests that indirect costs are relevant in determining the total costs of parasitism on host fitness, and should be considered for obtaining a realistic evaluation of parasitism. Theory predicts that the indirect cost of infection will depend on the intensity of competition, which is determined by host density and parasite prevalence. It is estimated that the higher the competition intensity, the higher the indirect costs [14]–[16]. In agreement with this prediction, the indirect cost of infection on Cen-1 and Ler plants increased with plant density, as in most reports on the interactions of plants with parasites or herbivores (e.g. [10],[24],[25],[36]). In contrast, in Boa-0 plants, indirect costs on progeny production disappeared or were overcompensated (infected plants showed a higher inter- than intra-class competitive ability) as plant density increased. As for Cen-1 and Ler tolerance to the direct cost of infection, tolerance of Boa-0 plants to the indirect cost of CMV infection appears associated with resource reallocation from growth to reproduction (Fig. 4).
The indirect cost of CMV infection was also affected by infection prevalence, but varied depending on the competitive ability component and the plant genotype. In Boa-0, infected plants show a lower intra- than inter-class competitive ability, resulting in a decrease of tolerance as prevalence increases, at odds with theoretical predictions [13]. In contrast, in Cen-1 and Ler plants, the higher the CMV prevalence, the higher the cost of infection on growth, but no effect of prevalence was observed on seed production. Most reported experimental analyses of competitive ability have focussed on the ability to harvest resources, i.e., growth [9],[37]. In this work, survival and reproduction were also measured, showing their different contribution to competitive success [9],[37]. The differential costs of infection on each life-history trait as prevalence increases indicates that analyses of the costs of parasitism that only consider one trait (e.g., growth) may result in biased conclusions, what underlines the relevance of considering different components of competitive ability to obtain a realistic view of the selection pressures exerted by parasites on their hosts.
In conclusion, plant density and costs of infection shape the competitive ability of plants. The outcome of the interaction between these factors depends on the plant genotype, which determines the plasticity of life-history traits and, hence, tolerance to the combined effects of both factors. Resource reallocation-based tolerance plays a key role in the competitive ability of Arabidopsis, which has evolved different strategies to maximize competitiveness in each genotype. Therefore, future analyses should consider not one but all these factors to understand the selective pressures that drive host-parasite co-evolution.
Strain LS-CMV, belonging to subgroup II of CMV isolates, was derived from biologically active cDNA clones [38] by in vitro transcription with T7 RNA polymerase (New England Biolabs, Ipswich MA, USA). Transcripts were used to infect tobacco plants for virus multiplication. CMV virions were purified from infected tobacco leaves as described in [39] and viral RNA was extracted by virion disruption with phenol and sodium dodecyl sulphate.
Three accessions of Arabidopsis thaliana were used: Boa-0 (Boadilla, Spain), Cen-1 (Centenera, Spain) and Ler (Landsberg, Poland). Boa-0 invests a higher proportion of resources to growth than to reproduction, and presents a longer life cycle than Cen-1 and Ler, which dedicate a higher proportion of resources to reproduction than to growth [30],[33]. The three accessions were multiplied simultaneously in the same greenhouse to obtain the seeds used for the experiments described in this work. Hence, maternal effects were not considered.
Costs of infection were analysed at 1, 2 and 4 plants per pot using monocultures of infected (I, I/I and I/I/I/I) and mock-inoculated (M, M/M and M/M/M/M) plants, as well as all possible combinations of mixed cultures of infected and mock-inoculated plants, simulating different CMV prevalences (Fig. 1). The following mixed cultures were used: M/I; M/M/M/I; M/M/I/I and M/I/I/I (infected and mock-inoculated plants next to each other); M/I/M/I (infected and mock-inoculated plants opposite to each other). Fifteen replicated pots per treatment were analysed. For plant growth, seeds of each accession were sown on filter paper soaked with water in a single plastic Petri dishes, and stratified in darkness at 4°C for 3 days before transferring for germination to a growth chamber (22°C, 14 h light and 70% relative humidity). Five day-old seedlings were planted in soil containing pots (10.5 cm of diameter and 0.43 l volume) for all plant densities. Plants were grown in a greenhouse (20–25°C day/night, 16 h light) in a completely randomised design. Three rosette leaves per plant were mechanically inoculated with purified CMV RNA (100 ng/µl) in 0.1 M Na2HPO4 when rosettes presented 4–5 leaves (stages 1.04–1.05 in [40]).
Plants were harvested at complete senescence stage, and dry weight was determined after plants were maintained at 65°C until constant weight. The weights of rosettes (rosette weight, RW), inflorescence structures including seeds (inflorescence weight, IW) and seeds (seed weight, SW) were measured separately. Rosette weight was used as an estimate of growth effort, inflorescence weight was taken as an estimate of total reproductive effort (reproductive structures plus seed output). Seed weight was quantified after threshing as a proxy to the number of viable seeds, since CMV infection does not affect either the weight per seed or seed viability in these accessions [31]. Thus, seed weight was used as an estimator of progeny production. To quantify the effect of CMV infection on life history traits under competition (here referred to as competitive ability), the mean value of the infected plants in each pot was divided by the mean value of the mock-inoculated plants of the same treatment (Traiti/Traitm, i and m denote infected and mock-inoculated plants, respectively).
RW, IW and SW and their various transformations, were homocedastic and were analysed using analysis of variance (ANOVA). All the analyses were done using pot as the unit of replication, that is, considering the mean value of each trait for plants of each condition (infected or mock-inoculated) within each pot. All traits were compared among conditions (infected or mock-inoculated), treatments, classes of competition (intra or interclass) or densities by one-way ANOVA. To determine interactions between these factors, complete two-way or three-way ANOVA models were used. Significance of differences among classes within each factor was determined by Least Significant Difference (LSD) analyses. All comparisons were done for the raw untransformed data, and for ratios between values of infected and mock-inoculated plants. All statistical analyses were done using the statistical software package SPSS 13.0 (SPSS Inc., Chicago, USA).
|
10.1371/journal.pntd.0004361 | Treatment Failure after Multiple Courses of Triclabendazole among Patients with Fascioliasis in Cusco, Peru: A Case Series | Triclabendazole is reported to be highly effective in treatment of human fascioliasis. We present 7 of 19 selected cases of human fascioliasis referred to our center in the Cusco region of Peru that failed to respond to triclabendazole. These were mostly symptomatic adults of both sexes that continued passing Fasciola eggs in the stool despite multiple treatments with 2 doses of triclabendazole at 10 mg/kg per dose. We documented the presence of eggs by rapid sedimentation and Kato Katz tests after each treatment course. We found that repeated triclabendazole courses were not effective against fascioliasis in this group of people. These findings suggest that resistance to triclabendazole may be an emerging problem in the Andes.
| Fascioliasis is a zoonotic food borne trematode infection with a wide distribution. The complex epidemiology of this infection makes control efforts difficult. The paucity of drugs available for treatment may further hinder their success. Triclabendazole, the only first line drug for Fasciola, has been used for many years in the livestock industry. Resistant livestock Fasciola infections have emerged in developing and developed countries. However, most human trials report triclabendazole efficacies close to 100% after a few doses. Only a few cases of triclabendazole treatment failure have been published. We document 7 patients infected with Fasciola hepatica in Cusco–Peru that failed several treatment courses with triclabendazole. This raises concerns regarding preparedness to address resistant parasite infections and calls for more research to find new medications and tools to evaluate resistance.
| Fascioliasis is a worldwide zoonotic infection caused by the trematode parasite Fasciola hepatica. Livestock infection causes large economic losses in developed and developing countries.[1] Even in some wealthy countries, up to 50% of the dairy and meat herds may be infected; but data from resource-poor countries are limited.[2–4] Heavily infected cattle have significantly decreased milk (≥ 1.5 L daily) and meat (≥ 3 kg) production.[5,6] Human infection has been reported in more than 70 countries, but the highest burden occurs in the Andes and parts of the Middle East.[7] School-age children have the highest prevalence of fascioliasis and bear most of its severe consequences. Lopez et al. described a threefold increase in anemia risk among children with fascioliasis compared with children without infection.[8] Significant weight loss during the acute and chronic infections has been described in other studies.[9,10] Thus, the long term effects of fascioliasis complications have motivated significant efforts to tackle livestock and human infection.
Triclabendazole is the most effective drug for fascioliasis based on safety and cure rates reported in small mostly uncontrolled studies.[11] Mass treatment with triclabendazole has been proposed as a strategy to control fascioliasis in livestock and humans. In developed countries cattle and sheep herds are treated with triclabendazole under professional supervision. However, in resource-poor countries, mass livestock treatment is often inconsistent.[12] Mass treatment and inconsistent dosing of triclabendazole may select resistant parasites.[13] The emergence of triclabendazole resistance has been described among sheep and cattle herds in Scotland, Northern Ireland, and Australia and has been associated with decreased beef and dairy production.[12,14] Increasing resistance has also been reported in Cajamarca, Peru, where only 31% of cattle treated with 12 mg/kg of triclabendazole were cured after 14 days.[15] Triclabendazole resistance in humans has only rarely been noted.[16,17] Reports of resistance are of concern given that triclabendazole is the only highly effective treatment available. In this report, we describe 7 patients with fascioliasis with persistent infection despite multiple treatment courses with triclabendazole.
The Cusco region in the south highlands of Peru is an endemic area for fascioliasis. In rural areas of this region the prevalence of Fasciola hepatica infection among children is 11%.[8] The Universidad Peruana Cayetano Heredia and University of Texas Medical Branch Collaborative Research Center in Cusco is a referral center for research and management of Fasciola infection. Patients referred to us with diagnosed or suspected fascioliasis are evaluated with up to three Lumbreras rapid sedimentation and Kato Katz stool tests. Subjects with negative stool tests and significant eosinophilia are evaluated with Fas2 ELISA for serum antibodies against Fasciola hepatica. Except when noted, treatment courses for patients with stool or serologic evidence of fascioliasis consisted of 2 doses of triclabendazole at 10 mg/kg every 12 hours preceded by a meal rich in fat. All subjects received counseling on avoidance of vegetables that might put them at risk for reinfection. Treatment response was assessed with Lumbreras rapid sedimentation and Kato Katz stool tests between 1 and 3 months after treatment.[18,19] Those found to remain infected received repeated courses of triclabendazole.
Between January 2014 and April 2015, 7 out of 19 patients with Fasciola hepatica infection referred to our center for evaluation and treatment failed to respond to multiple courses of triclabendazole (Egaten 250 mg tablets, Novartis Pharma AG, Basel, Switzerland, expiration date December 2015) 2 doses at 10 mg/d every 12 h with a fatty meal. Three of these were males, 2 were younger than 18 years old, and all but one were born in Cusco City. Four patients had acute presentations with delayed diagnosis, severe symptoms requiring prolonged hospital admissions for hypereosinophilia. All patients had a history of eating fresh watercress and other green leafy vegetables and self-medicated at least once with triclabendazole for veterinary use without response. In an attempt to improve their clinical response, all were prescribed triclabendazole and nitazoxanide (Colufase, Roemmers SA, Lima, Peru) 500mg PO every 12 hours for 7 days in combination after failing courses of treatment with triclabendazole monotherapy. Only one patient had consecutive negative stool tests and was deemed cured. The characteristics of the patients are shown in Table 1. The cases and their clinical course are briefly described below. Fig 1 shows the egg counts 1 to 3 months after each treatment course with triclabendazole.
Case 1 was a 51 years old female farmer with 16 kg weight loss, right upper quadrant abdominal pain, night sweats, anorexia, malaise, fever, and hypereosinophilia. She was first treated with a single dose of 10 mg/kg triclabendazole with improvement of symptoms. However, after 8 months her symptoms returned and she was again noted to be shedding Fasciola eggs. The patient was treated 2 additional times with triclabendazole (each with 2 doses of 10 mg/kg every 12 hours) and then with 2 doses of triclabendazole 10 mg/kg every 12 hours followed by nitaxozanide 500 mg every 12 hours for 7 days with improvement of symptoms. However, continued to pass Fasciola eggs on stool tests.
Case 2 was a 36 years old female with right upper quadrant abdominal pain, jaundice, severe joint pain, fatigue, 8 kg weight loss, and hypereosinophilia. Her stool tests and Fas-2 ELISA were positive for Fasciola. After failing a treatment course with triclabendazole (2 doses of 10 mg/kg every 12 hours), she was referred to our center. Over 10 months she received 1 additional course of triclabendazole treatment (2 doses of 10 mg/kg every 12 hours), a course of 3 doses of triclabendazole 10 mg/kg every 12 hours, and a course of 2 doses of triclabendazole 10 mg/kg every 12 hours followed by nitazoxanide (500 mg every 12 hours for 7 days) with marked improvement of symptoms but persistence of Fasciola eggs in the stools.
Case 3 was a 43 years old female who was asymptomatic. She was tested for Fasciola hepatica ova after her husband (case 4) was diagnosed with the infection. Both Fas2 ELISA and stool tests were positive for Fasciola infection. She failed a course of triclabendazole (2 doses of 10 mg/kg every 12 hours) and was prescribed 2 doses of triclabendazole 10 mg/kg every 12 hours followed by nitaxozanide 500 mg every 12 hours for 7 days after which she developed intrahepatic bile obstruction with removal of 3 adult Fasciola by endoscopic retrograde cholangiopancreatography. Also a migratory subcutaneous nodule due to Fasciola developed despite self-medication with a veterinary formulation of triclabendazole in 2 occasions. She was prescribed triclabendazole (2 doses of 10 mg/kg every 12 hours) with negative stool tests for Fasciola at 5–6 weeks follow up.
Case 4 is a 42 years old male born in the jungle of Cusco state married with case 3. He was admitted to the hospital with fever, diffuse abdominal pain, cough, severe fatigue, 5 kg weight loss, rash in the lower extremities and buttocks, and hypereosinophilia (eosinophil count > 30,000/dL). Lumbreras rapid sedimentation and Fas2 ELISA tests were positive for Fasciola hepatica infection. A few days after receiving triclabendazole treatment (2 doses of 10 mg/kg every 12 hours) his symptoms disappeared, but continue passing Fasciola ova. He was prescribed 2 additional treatment courses with triclabendazole (2 doses of 10 mg/kg every 12 hours) followed by a course of 2 doses of triclabendazole 10 mg/kg every 12 hours combined with nitaxozanide 500 mg every 12 hours for 7 days and 2 courses of triclabendazole veterinary formulation turning his Kato Katz tests negative. However, the Lumbreras rapid sedimentation test has remained positive.
Case 5 was a 15 years old male who presented with epigastric pain, fever, shortness of breath, chest pain, 5 kg weight loss, hypereosinophilia, ascites, and pleural effusion. The stool tests for Fasciola were initially negative, but the Fas2 ELISA was positive. He was treated with a single dose of veterinary formulation triclabendazole with partial improvement of symptoms. His follow up stool tests were positive and remained positive since then despite 3 courses of triclabendazole (2 doses of 10 mg/kg every 12 hours each), and a course of 2 doses of triclabendazole 10 mg/kg every 12 hours in combination with nitaxozanide 500 mg every 12 hours for 7 days.
Case 6 was a 12 years old male brother of case 5 diagnosed with asymptomatic F. hepatica infection. He failed the initial treatment with a single dose of triclabendazole. He was prescribed and failed 2 additional triclabendazole courses (2 doses of 10 mg/kg every 12 hours). He was subsequently treated with 2 doses of triclabendazole 10 mg/kg every 12 hours followed by nitaxozanide 500 mg every 12 hours for 7 days but his stool tests have remained positive for Fasciola eggs.
Case 7 was a 36 years old woman mother of case 5 diagnosed with chronic F. hepatica infection. She received 3 courses of triclabendazole (2 doses of 10 mg/kg every 12 hours) followed by a course of 2 doses of triclabendazole 10 mg/kg every 12 hours combined with nitaxozanide 500 mg every 12 hours for 7 days with persistently positive stool tests.
The study was reviewed by the Institutional Ethics Committee from Universidad Peruana Cayetano Heredia in Lima, Peru. Written informed consent was obtained from the subjects.
Although triclabendazole resistance in veterinary medicine is well known, resistant human infections have only rarely been reported. In this manuscript, we report 7 cases of Fasciola hepatica infection that failed to respond to multiple treatment courses with recommended doses of triclabendazole. Two other case reports of failure of triclabendazole treatment for fascioliasis have been published. In 2012, Winkelhagen et al. from the Netherlands reported a single case of multiple treatment failures with triclabendazole and nitazoxanide.[16] In 2014, Gil et al. reported 4 cases of triclabendazole failure in Chile.[17] However, in 3 of those cases the timeline between symptoms, treatment, and evaluation of response suggest reinfection rather than treatment failures. Of note, none of the reported cases had quantitative tests to evaluate the response of egg burden to treatment. Most of our patients had low egg burdens and the egg counts did not showed significant reductions after treatment.
The microscopic detection of Fasciola eggs in human stool after the ingestion of metacercaria takes approximately 12 weeks. Our cases were followed up and tested for cure between 1 and 3 months after treatment with triclabendazole. This approach to monitoring was chosen to distinguish the treatment response from reinfection. Failure of anthelmintic treatment may be due to a number of factors. Quality control of the medications is essential. In these cases, all treatment failures received Egaten 250 mg tablets (Novartis Pharma AG, Basel, Switzerland) stored according to the manufacturer recommendations with an expiration date in December 2015 (well after treatment). Treatment failure may also result from inadequate drug absorption. Food has significant effects on the absorption of triclabendazole. It is recommended that patients with fascioliasis ingest a fatty meal before each triclabendazole dose to increase medication absorption in the intestine, as was done in all of our cases. The impact in cure rates of not following this recommendation has not been studied. However, uncontrolled clinical trials with single dose triclabendazole report efficacies around 90% despite absence of fat ingestion before treatment.[11] Reduced triclabendazole conversion to triclabendazole sulfoxide and triclabendazole sulfone in the presence of severe liver impairment have been proposed as a cause of treatment failure.[20] All our cases had relatively low Fasciola egg counts suggesting mild infections and probably minimal liver damage. Of note, most of our patients presented with symptoms of acute Fasciola infection. Whether this was associated with the initial treatment failure cannot be ascertained. In early stages of infection with Schistosoma sp. the parasite has reduced susceptibility to praziquantel as demonstrated in vitro and in vivo.[20,21] Some in vitro studies with Fasciola hepatica infection suggest reduced triclabendazole susceptibility among juvenile parasites compared to adults.[22] This has not been rigorously documented in case series of patients with acute infections.[23,24] Marcos et al. reported the resolution of eosinophilia after a single dose of triclabendazole in 10 patients with acute Fasciola infection. However, the authors failed to report the absence of eggs in the stools during the follow up.[25] Thus, the clinical evidence gathered suggests the presence of triclabendazole resistant Fasciola hepatica infection in our cases.
The mechanisms by which the fluke can become resistant to triclabendazole remain to be elucidated. The β-tubulin gene mutations that cause benzimidazole resistance in nematode parasites does not seem to explain triclabendazole resistance in Fasciola hepatica.[26] Changes in drug uptake and parasite drug metabolism seem to play a bigger role. The uptake of the drug by the fluke is influenced by a P-glycoprotein linked efflux pump. Experiments have shown that inhibition of this pump leads to potentiation of triclabendazole activity.[27] In addition, triclabendazole resistant flukes have been shown to metabolize triclabendazole sulfoxide to sulfone to a greater extent than susceptible flukes.[28]. Thus, the combined effect of reduced drug uptake and more active drug metabolism could reduce the effective concentrations of triclabendazole.
Triclabendazole has been used in livestock to treat Fasciola for many years. Inconsistent dosing and schedules have led to widespread resistance in cattle rearing areas in the last decade. Human infections with triclabendazole resistant Fasciola in areas with zoonotic transmission is a potential problem. In contrast to veterinary medicine in which other treatment options for Fasciola exist, in humans triclabendazole is the only first line medication with reported high efficacy. Thus, the emergence of triclabendazole resistance in fascioliasis among humans is an important clinical and public health concern as no alternative drugs are available to treat the infection. In our case series, subjects were treated with nitaxozanide (500 mg twice daily for 7 days) after 2 doses of triclabendazole. This approach was based on a double blind placebo controlled trial of nitazoxanide for the treatment of fascioliasis. Although the trials showed limited efficacy in children (40%), the efficacy was slightly higher in adults (60%).[29,30] The development of biliary colic in some of the cases could have suggested response to the medication, but none were cured by combination treatment.
This study has some limitations that made difficult the assessment of resistance. Most of the information was collected retrospectively and the number of cases was small. We were not able to recover live flukes from the subjects after treatment or generate metacercariae from the eggs collected for susceptibility testing in vitro. Nevertheless, our clinical observations suggest the presence of triclabendazole resistant Fasciola infections in a selected group of patients from Cusco. Resistant infection in livestock has already been reported in the northern highlands of Peru. Although, this report does not reflect in any way the community prevalence of triclabendazole resistance among humans in Cusco, triclabendazole resistance appears to be an emerging problem deserving attention in Peru and probably other highly endemic areas. Research on new drugs and methods to evaluate drug resistance is urgently needed to control Fasciola.
|
10.1371/journal.pgen.0040025 | The Evolution of Spinnable Cotton Fiber Entailed Prolonged Development and a Novel Metabolism | A central question in evolutionary biology concerns the developmental processes by which new phenotypes arise. An exceptional example of evolutionary innovation is the single-celled seed trichome in Gossypium (“cotton fiber”). We have used fiber development in Gossypium as a system to understand how morphology can rapidly evolve. Fiber has undergone considerable morphological changes between the short, tightly adherent fibers of G. longicalyx and the derived long, spinnable fibers of its closest relative, G. herbaceum, which facilitated cotton domestication. We conducted comparative gene expression profiling across a developmental time-course of fibers from G. longicalyx and G. herbaceum using microarrays with ∼22,000 genes. Expression changes between stages were temporally protracted in G. herbaceum relative to G. longicalyx, reflecting a prolongation of the ancestral developmental program. Gene expression and GO analyses showed that many genes involved with stress responses were upregulated early in G. longicalyx fiber development. Several candidate genes upregulated in G. herbaceum have been implicated in regulating redox levels and cell elongation processes. Three genes previously shown to modulate hydrogen peroxide levels were consistently expressed in domesticated and wild cotton species with long fibers, but expression was not detected by quantitative real time-PCR in wild species with short fibers. Hydrogen peroxide is important for cell elongation, but at high concentrations it becomes toxic, activating stress processes that may lead to early onset of secondary cell wall synthesis and the end of cell elongation. These observations suggest that the evolution of long spinnable fibers in cotton was accompanied by novel expression of genes assisting in the regulation of reactive oxygen species levels. Our data suggest a model for the evolutionary origin of a novel morphology through differential gene regulation causing prolongation of an ancestral developmental program.
| Human domestication of plants has resulted in dramatic changes in mature structures, often over relatively short time frames. The availability of both wild and domesticated forms of domesticated species provides an opportunity to understand the genetic and developmental steps involved in domestication, thereby providing a model of how the evolutionary process shapes phenotypes. Here we use a comparative approach to explore the evolutionary innovations leading to modern cotton fiber, which represent some of the more remarkable single-celled hairs in the plant kingdom. We used microarrays assaying approximately 22,000 genes to elucidate expression differences across a developmental time-course of fibers from G. longicalyx, representing wild cotton, and G. herbaceum, a cultivated species. Expression changes between stages were temporally elongated in G. herbaceum relative to G. longicalyx, showing that domestication involved a prolongation of an ancestral developmental program. These data and quantitative real time-PCR experiments showed that long, spinnable fiber is associated with a number of genes implicated in regulating redox levels and cell elongation processes, suggesting that the evolution of spinnable cotton fiber entailed a novel metabolic regulatory program
| One of the central questions in evolutionary biology concerns the developmental and genetic processes by which new phenotypes arise. The recent merger of genomic technologies with phylogenetics has generated important insights into the evolution of developmental transformations in maize [1,2], rice [3,4] and other taxa, e.g., [5,6]. These studies demonstrate that morphological change in complex organs can often be initiated by relatively few mutations, most often in regulatory regions [7], although the genetic underpinnings of most evolutionary change remains unknown. An exceptional example of evolutionary innovation involving a single-celled structure is the cotton seed trichome, present in all 50 species in the genus Gossypium and colloquially termed “cotton fiber” in the domesticated species. On the day of anthesis (flower opening), approximately one in four cells of the ovular epidermis has already been fated to become a trichome, initially appearing as a spherical protrusion and subsequently elongating through stages of primary wall synthesis, secondary wall synthesis, maturation and cell death. Representing one of most distinct single cell types in the plant kingdom, cotton fibers may attain a final length of 6 cm in some cultivars, with a length/width ratio of more than 2000 [8]. A single cotton ovary contains ∼500,000 elongating cells representing a single cell type.
The long, strong and fine fibers of modern cotton cultivars were wrought through a long history of both natural and human-mediated selection [9–11]. Following its origin about 10 MYA [12,13], Gossypium diversified into approximately 50 species in the warmer, arid to semi-arid regions of both hemispheres. This radiation was accompanied by cytogenetic differentiation, which now is reflected in the recognition of eight, monophyletic “genome groups” (A to G and K) (Figure 1A). Remarkably, four wild Gossypium species were independently domesticated by aboriginal domesticators ∼5000 years ago, or more, and transformed into fiber and seed-oil plants [10,14]. Two of these (G. arboreum and G. herbaceum) are A-genome diploids from the Old World, while the other two (G. hirsutum and G. barbadense) are AD-genome allotetraploids from the New World.
In contrast to the cultivated diploids and tetraploids, wild diploid species have short (mostly <5mm), coarse and tightly adherent trichomes that would not be recognized as “cotton” by a casual observer. The only exception is the wild form of G. herbaceum (G. herbaceum subsp. africanum), which has sparse but elongated and spinnable fiber. Both wild and cultivated cottons produce fiber on the seed coat, but there are striking morphological and structural differences between these fibers, the most obvious of which is their size. Some species (e.g., G.thurberi, G. trilobum, G. davidsonii, and G. klotzschianum) do not possess obvious seed hairs, but they actually are present as developmentally repressed structures not visible to the unaided eye [15]. The duration of the elongation phase and the timing of onset of secondary wall synthesis appear to be key determinants of the final length of the fiber in both wild and cultivated plants [15–18]. This trait of prolonged elongation was passed on to the allopolyploids, which in turn was a key component of their eventual domestication [15].
When viewed in a phylogenetic context (Figure 1A), the origin of spinnable fibers is diagnosed to have occurred once in the history of Gossypium, following divergence of the A-genome and F-genome clades. This “pre-adapted” A-genome ancestor later contributed this genome and its propensity for the development of elongated cotton fibers to the allotetraploid cottons that colonized and diversified in the New World, ultimately giving rise to the modern, annualized forms of G. hirsutum that account for >90% of contemporary world cotton commerce.
To gain insight into the genetic factors that led to the evolution of long, spinnable cotton fibers, we performed global gene expression analysis, comparing the A-genome taxon G. herbaceum to its closest wild relative, G. longicalyx. The latter species was described relatively recently, following its discovery in eastern Africa [19], and is cladistically resolved as the sister taxon to the A-genome in molecular phylogenetic analyses of both plastid and nuclear gene sequences (13) (Figure 1A). We used a newly designed long oligonucleotide microarray and ovular trichome isolation procedures to analyze gene expression over a time-course of trichome development in both species. This analysis revealed major differences in genes related to stress responses and cell elongation, as well as a prolonged developmental profile in wild G. herbaceum. We suggest that the evolution of spinnable fiber was accompanied by prolongation of an ancestral developmental program, mediated through avoidance of stress-like processes in the developing fiber cells. This suggests that the evolutionary origin of a novel cell phenotype was facilitated by hypermorphosis.
We used an experimental loop-design system to compare mRNA expression levels in developing fiber-cells derived from A- and F-genome cotton species (Figure 1B). RNAs isolated from four developmental time-points, 5, 10, 20 and 25 days post-anthesis (DPA), were amplified and hybridized to cotton oligonucleotide microarrays containing 22,827 probes derived from deep EST sampling of diverse tissues and organs [20]. A summary of the number of genes differentially expressed between adjacent time-points during fiber development in both species is presented in Figure 2. Within each genome, many genes were found to be differentially expressed during fiber development (FDR < 0.05), but the distribution of the number of genes differentially expressed between adjacent time points was different for the two species. In G. herbaceum, the distribution was relatively even throughout the developmental stages studied (notice that the interval 10–20 DPA is twice the duration of 5–10 and 20–25 DPA), whereas in the F-genome species, G. longicalyx, ∼80% of significant expression changes occurred between the two adjacent time points 10 and 20 DPA. Also, in the transition from 20 to 25 DPA only 4 genes were differentially expressed in G. longicalyx, in comparison to 493 genes that were differentially expressed during this same interval in the A-genome species G. herbaceum. Because longer, spinnable fiber is phylogenetically derived [15], these results indicate that the fiber-cell developmental program in the A-genome has become prolonged during its evolution.
To better appreciate changes in global expression during fiber development and evolution, we tracked differences in expression between A- and F-genomes for all genes at all developmental stages. For each comparison upregulated and downregulated genes were tabulated, from which we derived categories of statistically overrepresented biological processes (Table S1). As expected from an inter-specific comparison, hundreds of genes were found that were differentially expressed between fibers from A- and F-genome plants. Inspection of the gene lists revealed that, among genes which previously had been described as regulating fiber elongation [21], some did indeed show significant expression differences, while others did not. An example of the developmental expression patterns for 5 differentially expressed and 5 non-differentially expressed, well-described genes, between F- and A- genomes, is provided in Figure S1.
A major difference in the developmental programs of fiber cells in the two taxa was revealed by GO family representation analyses (Table S1). Most noteworthy is the observation that at the beginning of fiber development in F-genome fibers, many genes involved with stress-response processes were highly upregulated. Comparison of statistically overrepresented biological processes at 5 DPA, for example, showed that in A-genome fibers, processes important for elongation, such as “respiration”, “energy” and “ribosome biogenesis” are overrepresented (Table 1). At the same time-point, however, genes upregulated in F-genome fibers belong to biological processes such as “response to stress”, “transcription regulatory activity” and “flavonoid biosynthesis”. Moreover, analyzing the 60 most-differentially expressed genes at 5 DPA, representing the top 2% of the upregulated genes in the F-genome in comparison to the A-genome, showed that more than a third were related to “response to stress” processes (Table S2). The expression pattern of some of these highly upregulated “stress response” genes in developing fiber-cells of the F-genome in comparison to the A-genome are presented in Figure S2.
A possibility that emerged from differences in gene expression between A-genome and F-genome fibers was that high levels of H2O2 and other reactive oxygen species (ROS) may be responsible for the stress-like processes evident in F-genome fibers early in fiber development. Hydrogen peroxide has previously been shown to be important for cell elongation, as it is required for cell wall loosening and expansion [22–25]. At high concentration, however, H2O2 becomes toxic, leading to stress processes that may lead to early onset of secondary cell wall synthesis and the end of cell elongation [22,26]. These observations suggest the hypothesis that the evolution of long spinnable A-genome fibers was accompanied by novel expression of genes that assist in regulating H2O2 and other ROS levels.
To evaluate this possibility, we examined the microarray data for genes that may control ROS and cell stress in elongating cells (Table 2). Three genes shown in other systems to regulate H2O2 levels by functional or regulatory means were investigated further. GAST1-like is a member of the gibberellin-induced, cysteine-rich protein family previously shown to be induced by H2O2. In transgenic Petunia, suppression of GAST1-like homologs inhibits elongation, whereas upregulation stimulates H2O2 scavenging, perhaps via redox-active cysteines, and cell elongation [27]. GAST1-like was previously shown to be expressed mainly in cotton fibers [21] (Figure S3). Cop1/BONZAI is part of a calcium-dependent, membrane-binding protein family, shown to promote growth and development in addition to repression of cell death by inactivation of stress-promoting R-genes [28]. Pex1 is a gene that encodes a protein important for the biogenesis of the peroxisome, an organelle that rids the cell of toxic substances such as H2O2 [29]. Both Cop1/BONZAI and Pex1 genes are only detected in fiber-specific EST libraries, out of more than 30 EST libraries that exist for cotton from a diverse set of tissues and organs (http://cottonevolution.info/).
For each of these three genes, we conducted real-time PCR using elongating and non-elongating fibers from several additional cotton accessions and species with either short or long fiber. As shown in Figure 3, GAST1-like, Cop1/BONZAI and Pex1 were highly upregulated in A-genome relative to F-genome fibers. Paralleling this result, all three genes were expressed in additional taxa having long fibers (G. arboreum, and both cultivated and wild forms of G. hirsutum), at the beginning of fiber-cell development, but were either undetected or were only weakly expressed in species with short fibers (G. raimondii (D clade), in addition to the F-genome species G. longicalyx). This consistent association between fiber development and gene expression across divergent clades suggests a functional association.
Transcription profiling of cotton using microarrays has been the subject of several recent studies, either using ovular tissue with fibers attached, or the fibers themselves [21,30–33]. These studies simultaneously evaluate mRNA expression levels for thousands of genes, providing a powerful tool for analyzing biological processes important to cotton fiber differentiation and development. One conclusion is that the transcriptome of cotton fibers is extraordinarily complex [21], involving thousands of genes that vary in expression levels through the stages of cellular initiation, primary wall synthesis, secondary wall deposition, maturation, and death. Here we provide the first comparative evolutionary analysis of fiber cell development, focusing on the initial phylogenetic steps implicated in the natural transformation of epidermal seed trichomes into long, spinnable fibers prior to and during human domestication. In this regard, the accession we used in this study is a domesticated form of G. herbaceum. Thus, our study reflects the evolution of either species-level differences, human selection (domestication), or both. The above-described qPCR analysis, however, was performed using other domesticated and wild species, showing that for this set of genes the changes occurred prior to domestication. In addition to its relevance to the evolutionary transitions in morphology and to cotton crop improvement, to our knowledge this study is the first to characterize the evolutionary, developmental genomics of a single cell type in any eukaryotic organism.
Our proposed model describing the developmental and evolutionary processes that led to the formation of spinnable fiber is presented in Figure 4. Previous ultrastructural characterization of various developing cotton fibers, including those of F- and A- genome species, demonstrate that the earliest stages of fiber initiation and development are phenotypically similar for species with either short or long fibers [15]. At 2 DPA, for example, fiber cells appear as the same spherical protrusion both in F- and A- genomes. This stage is followed by a period of rapid cell elongation, a process known to involve cell-wall relaxation, which itself has been shown to require non-enzymatic reactions mediated by H2O2 and other ROS that cleave polysaccharides [22–25]. Therefore, H2O2, produced enzymatically by oxidation reactions, is a necessary molecule for cell elongation. Higher levels of H2O2, however, may halt elongation through apparent stimulation of cell wall stiffening [22,26], and can even promote programmed cell death or necrosis. Accordingly, Cosgrove [34] has suggested that fine regulation of steady-state levels of ROS are essential for proper cell elongation.
Our model suggests that the curtailed developmental duration of the F-genome, relative to the A-genome, is caused by an insufficient control of cellular H2O2 and other ROS levels, eventually arresting elongation and leading to an induction of secondary cell wall synthesis. In cotton, H2O2 has been suggested to function as a developmental signal in the differentiation of secondary walls in cultivated G. hirsutum fibers, evidenced by the fact that inhibition of H2O2 production or scavenging existing H2O2 from the system prevents cell wall differentiation [35]. Similarly, exogenous addition of H2O2 prematurely promotes secondary cell wall formation in young fibers [35]. Our data are consistent with this interpretation and with earlier secondary cell wall formation in the F-genome, as indicated by expression differences between F- and A- genomes for the cellulose synthase A1 (CeSA1) gene. CeSA1 is a well-characterized gene involved in fiber secondary cell wall synthesis, and CeSA1 RNA expression levels have been suggested as a marker for secondary wall cell synthesis [30]. In our study, CeSA1 increases its expression earlier in the F-genome than in the A-genome (Figure S1–S1J).
Similar to the F-genome, early elongating fibers of the A-genome are exposed to increasing levels of H2O2. A-genome fibers, however, did not show increased RNA levels of stress-related genes, suggesting that this lineage evolved a metabolism to modulate ROS levels by either functional or regulatory means. At the functional level, genes controlling antioxidant levels, including ascorbate peroxidase, glutathione peroxidase and lipolic acid synthase, are all upregulated in A-genome fibers relative to those of the F-genome. These proteins scavenge and detoxify H2O2 and other ROS. Recently, comparative proteomic analysis between regular and mutant cotton fibers showed the involvement of a cotton ascorbate peroxidase in H2O2 homeostasis during cell development [36].
An additional example of possible functional modulation of H2O2 levels is offered by the protein GAST1, which we studied further. GAST1 is a cysteine-rich, gibberellin-induced gene, initially identified in tomato, which is suppressed in the GA-deficient mutant gib-1 and for which expression coincides with stem elongation [37]. It belongs to a protein family, identified in many plant species, that is suggested to play a role in many biological processes, including cell division, cell elongation (promotion and inhibition), transition to flowering, root development, fruit development and defense (summarized in [27]). RNAi suppression of expression in one member of this family, GIP2, was shown to inhibit stem elongation under low-temperature conditions in transgenic Petunia [38]. Wigoda et al. [27] have shown that GIP2 overexpression promotes stem and corolla elongation. In the same study they showed that GAST2 is expressed in the cell-wall and suggested that its putative redox-active cysteines may act as antioxidants that control H2O2 levels at this site. The fact that our cotton GAST1 has the same conserved 12 cysteine residues as other GAST-like proteins (data not shown), is expressed mainly in the fiber [21] (as shown in Figure S3), and is not expressed in the fuzz-like F-genome fibers and the D-genome fibers make it a promising gene for further investigation.
At the regulatory level, we explored two candidate genes further using quantitative RT-PCR and a broader sampling of species and accessions having either short or long fibers. The first is the Cop1/BONZAI (Cop1) gene. Cop1 is a calcium-dependent, membrane-binding protein isolated from a mutant with a temperature-dependent growth defect and an enhanced disease resistance phenotype in Arabidopsis [39,40]. Further study revealed that Cop1 acts as a repressor of a disease resistance (R) gene and as such it prevented programmed cell death (PCD) processes (i.e. hypersensitive response) [28]. It is unclear if the F-genome fiber is undergoing early “classical” PCD processes, indicated by the fact that “cell death processes” was not a statistically overrepresented biological process in our study. Cop1, however, was ranked in our microarray analysis as the single most upregulated gene in the A-genome, relative to the F-genome, among thousands of differentially expressed genes, making it as a strong candidate for controlling stress-like processes. Further quantitative RT-PCR analyses using a broader sampling of species with short and long fibers yielded comparable results (Figure 3), lending additional support to the hypothesis of a role for Cop1in fiber elongation.
A second regulatory gene studied further is Pex1. This gene is one of a cascade of peroxisome biogenesis genes that previously have been shown to be induced by H2O2 in both plant and animal cells, and have been suggested to assist in restoration of cellular redox balance in response to wounding and infection with an avirulent pathogen [29]. Pex1 encodes a member of the AAA (ATPases associated with diverse cellular activities) superfamily of ATPases that have been suggested to mediate lipid and/or membrane addition to peroxisome membranes, facilitating peroxisome growth [41,42]. As shown in Figure 3, Pex1 mRNA levels are strongly correlated with fiber length.
To verify the hypothesized roles of GAST1, Cop1, and Pex1 genes, additional functional analyses are needed in growing fiber-cells, including in vivo measurements of H2O2 levels and elongation rates in developing fibers derived from F- and A-genome species. These studies comprise one focus of our ongoing efforts. In addition, hundreds of other genes were differentially expressed between A- and F-genome fibers, including many known to be involved in fiber development, suggesting that in addition to cellular redox balance, other biological processes may be involved in the evolution of spinnable fibers. Thus, additional work is necessary to reveal the nature and relative contributions of these additional processes to fiber transformation during evolution.
The results presented here add perspective to results from our previous comparative study of fiber development in wild Gossypium species [15], in which wild A- and F-genome species exhibited continued fiber elongation up until approximately 21 DPA. One possible reconciliation between this observation and the present study is that the timing of the period of maximum fiber elongation is a key developmental difference. Some species with short fibers showed a nearly linear rate of elongation over most of the growth period, whereas long-linted species exhibited more complex growth curves. Another possibility is that the most important factor in determining final fiber length is the absolute fiber elongation rate and not the relative percentage of mature length (as [15]), suggesting that the effects of H2O2 and other ROS are not qualitative, but instead are quantitative, operating via effects on amount of elongation.
The present study implicates several molecular mechanisms as being involved in the evolution of elongated epidermal seed trichomes, providing the foundation for later human domestication of an important crop plant. Why elongated epidermal seed hairs evolved is a matter of speculation. Perhaps fibers evolved to aid in bird dispersal, as suggested [43]. This hypothesis gains credibility from observations by J. Stewart (unpubl.) of a bird's nest in NW Puerto Rico that contained numerous seeds of feral G. hirsutum, as well as a collection of G. darwinii from a finch's nest in the Galapagos Islands (J. Wendel, unpubl.). One might also speculate that fibers serve to inhibit germination unless there is sufficient moisture to saturate the fibers; should germination occur following a light rain, there might not be sufficient water for subsequent survival of the seedling. A related possibility is that seed hairs function as “biological incubators” to facilitate germination only when ecological conditions are appropriate, by recruiting particular microbial communities under appropriate moisture regimes. Irrespective of the veracity of these speculations, one outcome is that these processes set the stage for human domestication millions of years later.
We show that a major heterochronic change included prolongation of an ancestral developmental program, and coincidently, this change pre-adapted the derivative cell type for human domestication. At least in part it appears that avoidance or delay of stress-like processes may underlie the increased elongation in A-genome fiber development compared to F-genome fiber, in conjunction with an increased ability to modulate cellular redox balance in the growing cell. These evolutionary processes, occurring as they did perhaps several million years ago [9,10], may be interpreted as the key events responsible for the domestication of one of the world's most important crop plants.
Plants from Gossypium herbaceum (A1), G. longicalyx (F), G. hirsutum var TM1 (AD1), G. hirsutum var yucatanense (wild AD1), G. arboreum (A2) and G. raimondii (D5) were grown in three separate replicates of 4–8 plants in the Horticulture Greenhouses at Iowa State University. For each replicate, ovules were excised, immediately frozen in liquid nitrogen, and stored in −80 °C. At each developmental time point, fibers were isolated from ovules using a liquid nitrogen/glass bead shearing approach [21]. Initially, ovules were visually inspected for cell damage and the fibers were inspected for contaminating tissue. Subsequent RNA extractions were performed using a hot borate method [44]. RNA quality was confirmed on a BioAnalyzer (Agilent, Palo Alto, CA).
For microarray analyses, an indirect labeling procedure of amplified aminoallyl a-RNA was used as described [21]. Two dyes, Cy3 and Cy5, were coupled to 8 μg aliquots of aRNA using the Post-Labelling Aminoallyl-aRNA CyDye reactive dyes (Amersham Biosciences). A newly designed cotton long-oligonucleotide microarray containing over 22,827 probes derived from deep EST sampling of diverse tissues and organs [20] was used. All hybridizations, slide scanning and normalizations were performed as described previously [21].
A balanced developmental loop design for microarray analysis was performed (Figure 1B). For G. herbaceum (A1) and G. longicalyx (F), four fiber developmental time-points, 5, 10, 20 and 25 DPA were sampled. Within each species, hybridizations were performed between each pair of consecutive developmental stages by labeling one with Cy5 and the other with Cy3, and by closing the loop with a comparison of 25 and 5 DPA. In addition, 2 hybridizations were done between species at each time-point, using a dye swap for each pair. With three biological replications and 16 slides each, we generated gene expression data from a total of 48 microarrays. Statistical analyses were performed using R and SAS statistical software (code available upon request).
Log transformed, median-normalized values of the 22,827 genes were examined for expression differences between each fiber developmental stage within and between species. We considered a standard mixed linear model for the data for each gene as:
where yijklm denotes the normalized log-scale signal intensity for genotype i, time-point j, biological replication k, dye l and slide m; μ denotes an intercept parameter; δi denotes the fixed effect of genotype i; τj denotes the fixed effect of time-point j; sk denotes the random effect of replication k; rl denotes the fixed effect of dye l; δτij denotes the fixed effect of the interaction between genotype i and time-point j; δτsijk denotes the random effect of the interaction between genotype i, time-point j and replication k; s(v)mk denotes the effect of the slide m inside replication k; eijkl denotes a random error term intended to capture all other sources of variability. Contrasts for differential expression between genotypes, time points and the interaction genotype x time-points were conducted using this model. For each gene, differences were calculated using the following pair-wise contrasts:
where letters denote genotype (F- or A-genome) and numbers denote developmental time-point (5, 10, 20, or 25 DPA).
The 22,827 p-values from each comparison were converted to q-values using the method of Storey and Tibshirani [45]. These q-values were used to identify the number of differentially expressed genes for a given comparison when controlling the false discovery rate (FDR) at various levels.
Blast2GO (http://www.blast2go.de/) was used to identify biochemical pathways involved in a given comparison and to calculate the statistical significance of each pathway. Blast2GO includes the Gossip package [46] for statistical assessment of annotation differences between two sets of sequences, using Fisher's exact test for each GO term. FDR controlled p-values (FDR < 0.05) were used for the assessment of differentially significant metabolic pathways.
Amplified aRNA was used as a template for first strand cDNA synthesis with the Super-script II pre-amplification system reverse transcriptase kit (Gibco BRL Life Technologies) at 42 °C according to the supplier's instructions. Specific primers with amplicons for quantitative PCR were designed based on the sequence derived from the EST sequence corresponding to the candidate and reference genes (Table S3). We used the RNA helicase gene (Q9ZS12) as the reference gene. RNA helicase gene was found to be equally expressed in different developing fibers as well in other plant tissues [21]. cDNA was used as the template for quantitative PCR amplification using the GeneAmp 5700 Sequence Detection System (PE Biosystems) with SYBR Green Master Mix containing AmpliTaq Gold, according to the manufacturer's instructions (PE Biosystems). Standards containing logarithmically increasing known levels of cDNA were run with each set of primers for normalization. All real-time PCR products were confirmed by sequencing. |
10.1371/journal.pgen.1003730 | Vitellogenin Underwent Subfunctionalization to Acquire Caste and Behavioral Specific Expression in the Harvester Ant Pogonomyrmex barbatus | The reproductive ground plan hypothesis (RGPH) proposes that the physiological pathways regulating reproduction were co-opted to regulate worker division of labor. Support for this hypothesis in honeybees is provided by studies demonstrating that the reproductive potential of workers, assessed by the levels of vitellogenin (Vg), is linked to task performance. Interestingly, contrary to honeybees that have a single Vg ortholog and potentially fertile nurses, the genome of the harvester ant Pogonomyrmex barbatus harbors two Vg genes (Pb_Vg1 and Pb_Vg2) and nurses produce infertile trophic eggs. P. barbatus, thus, provides a unique model to investigate whether Vg duplication in ants was followed by subfunctionalization to acquire reproductive and non-reproductive functions and whether Vg reproductive function was co-opted to regulate behavior in sterile workers. To investigate these questions, we compared the expression patterns of P. barbatus Vg genes and analyzed the phylogenetic relationships and molecular evolution of Vg genes in ants. qRT-PCRs revealed that Pb_Vg1 is more highly expressed in queens compared to workers and in nurses compared to foragers. By contrast, the level of expression of Pb_Vg2 was higher in foragers than in nurses and queens. Phylogenetic analyses show that a first duplication of the ancestral Vg gene occurred after the divergence between the poneroid and formicoid clades and subsequent duplications occurred in the lineages leading to Solenopsis invicta, Linepithema humile and Acromyrmex echinatior. The initial duplication resulted in two Vg gene subfamilies preferentially expressed in queens and nurses (subfamily A) or in foraging workers (subfamily B). Finally, molecular evolution analyses show that the subfamily A experienced positive selection, while the subfamily B showed overall relaxation of purifying selection. Our results suggest that in P. barbatus the Vg gene underwent subfunctionalization after duplication to acquire caste- and behavior- specific expression associated with reproductive and non-reproductive functions, supporting the validity of the RGPH in ants.
| One of the main features of social insects is the division of labor, whereby queens monopolize reproduction while sterile workers perform all of the tasks related to colony maintenance. The workers usually do so in an age-dependent sequence: young workers tend to nurse the brood inside the nest and older workers are more likely to forage for food. Previous studies revealed that vitellogenin, a yolk protein typically involved in the regulation of reproduction in solitary insects, has been co-opted to regulate division of labor in the honeybee. In this study, we investigate such a role of vitellogenin in another group of social insects: the ants. We first use phylogenetic analyses to reveal the existence of multiple vitellogenin genes in most of the sequenced ant genomes. Then we compare the expression of the two vitellogenin genes (Pb_Vg1 and Pb_Vg2) among queens, nurses and foragers in the seed-harvester ant Pogonomyrmex barbatus. Our results suggest that, after the initial duplication in ants, the vitellogenin genes acquired caste and behavioral specific expression associated with reproductive and non-reproductive nutritionally related functions. This study also shows that ants and bees, despite having evolved sociality independently, have conserved similar mechanisms to regulate division of labor.
| Division of labor is the cornerstone of insect societies and implies the coexistence of individuals that differ in morphology, reproduction and behavior [1], [2]. There are usually two levels of division of labor among individuals in social insect colonies. The first relates to a reproductive division of labor, whereby reproduction is monopolized by one or several queens while sterile workers perform all the tasks related to colony maintenance. The second relates to a division of labor among the worker force, whereby workers perform different tasks in an age-dependent sequence: young workers usually perform tasks inside the colony (e.g. brood care) while old workers forage outside the nest [3], [4]. The colony organization of advanced eusocial insects evolved independently in ants, bees, and wasps [5], [6]. While the ecological constraints favoring social evolution are well studied [7], it remains largely unknown whether the genetic mechanisms regulating behavior are conserved among species [8]–[12].
The ovarian ground plan hypothesis (OGPH) is a theoretical framework that seeks to explain the evolution of reproductive division of labor in social insects [13], [14]. The OGPH proposes that the physiological pathways regulating the reproductive and behavioral cycles of solitary ancestors have been co-opted and selected to evolve into the queen and worker castes of existing eusocial insects. This hypothesis is based on the observations that the ovarian cycle of alternate development and depletion phases of solitary wasps is associated with reproductive and non-reproductive behavioral traits that resemble the queen and worker castes of eusocial insects: females with developed ovaries lay eggs while females with undeveloped ovaries forage for food and defend the nest [13]. The same was likely true in the solitary ancestors of ants and bees. The reproductive ground plan hypothesis (RGPH) extends this concept to explain the evolution of worker division of labor in honeybees [15]–[17]. Indeed, honeybee worker subcastes have two distinctive phases of ovarian activity, with nurses having large ovaries and high titers of the yolk protein vitellogenin (Vg), and foragers small ovaries and low titers of Vg [15], [18], [19]. The RGPH suggests that the mechanisms controlling ovarian activity influence the behavioral development and the mechanisms of food collection in worker honeybees. Support for this hypothesis is provided from studies demonstrating that workers with larger ovaries and higher titers of Vg are more likely to forage at younger ages and show a pollen foraging bias compared to workers with smaller ovaries and lower titers of Vg, which are more likely to forage at older ages and show a nectar foraging bias [15], [16]. These variations in reproductive traits have a genetically inherited component as strains with different ovarian activity and foraging bias have been selected from wild type populations [15], [20].
Although it has been established that the pleiotropic mechanisms connecting reproduction and division of labor have a genetic component, three lines of evidence suggest that the two processes are linked by an additional nutritional factor. First, in honeybees, reproductive queens and potentially reproductive nurses (with large and medium-sized ovaries, respectively) [18], [21], [22] consume diets with higher protein content [23] compared to sterile foragers with smaller and presumably non-functional ovaries [18]. Second, pollen consumption in nurses [23] is associated with higher Vg protein levels [19], compared to foragers that only consume honey [23]. Finally, there is a causal relationship between nutrition, Vg levels and behavior as pollen consumption is required to induce Vg expression [24], [25] and experimental reduction of Vg expression results in precocious foraging [26], [27].
To determine whether the co-option of reproductive pathways plays a major role in social evolution would require to investigate the link between reproductive physiology and behavior in other social insects, preferentially in those, such as ants, that evolved sociality independently from bees [5], [6], [28]. Ants have two additional characteristics that make them a particularly interesting model to study the predictions of the RPGH. First, in contrast to the honeybee genome that contains a single Vg gene, ant genomes can harbor multiple Vg genes. Indeed, the genome of the fire ant Solenopsis invicta harbors four Vg genes, two of them preferentially expressed in queens (Si_Vg2 and Si_Vg3) and the two others in foraging workers (Si_Vg1 and Si_Vg4) [29]. Vg duplication and subsequent subfunctionalization to acquire caste-specific expression provides a unique opportunity to test whether the genes associated with reproduction were co-opted to regulate worker behavior. Second, also in contrast with bees and wasps, where workers are facultatively sterile, workers are fully sterile in a significant number of ant species, including P. barbatus and S. invicta [30]–[32], which allows one to test the hypothesis that Vg reproductive function was co-opted to regulate behavior in species with fully sterile workers.
The first aim of this study was to determine the number of Vg genes in P. barbatus and other ants and investigate their phylogenetic relationships. This analysis is expected to determine whether the occurrence of multiple Vg genes is a phenomenon specific to S. invicta [29] or shared with other ant species as well as provide information on the origin and evolution of Vg genes in ants. Our second objective was to test whether Vg genes in P. barbatus display caste-specific expression profiles similar to that observed in S. invicta, which will address the question whether Vg gene duplication and subfunctionalization to acquire caste-specific functions is a common feature in ant species. Finally, the third objective of this study was to investigate whether the expression of Vg genes in P. barbatus is associated with task performance as predicted by the RGPH. We carried out these objectives by annotating Vg genes, building a phylogenetic tree, measuring mRNA levels of P. barbatus Vg genes in queens, nurses and foragers and performing molecular evolution analyses.
We identified two adjacent copies of Vg genes (Pb_Vg1 and Pb_Vg2) in the genome of Pogonomyrmex barbatus [33] with predicted lengths of 1742 and 1654 amino acids, respectively (Table 1). The two genes are separated by a putative mariner-like transposon, a DNA transposable element that has been involved in duplication events [34]. The two Vg genes have an identical number of exons (Figure 1A) and share the three structural domains typical of vitellogenins: the lipoprotein N-terminal domain (LPD-N), the domain of unknown function 143 (DUF1943) and the von Willebrand factor type D domain (VWD) [35], [36] (Figure 1B). However, the coding sequence of Pb_Vg2 is truncated compared to Pb_Vg1 because of an earlier stop codon in the last exon of Pb_Vg2.
To determine whether the presence of multiple Vg genes is a general feature of ants, we searched for Vg genes in the five additional recently published ant genomes. These are divided up into four different subfamilies: Myrmicinae (Atta cephalotes and Acromyrmex echinatior) [37], [38], Formicinae (Camponotus floridanus) [39] Dolichoderinae (Linepithema humile) [40] and Ponerinae (Harpegnathos saltator) [39]. We found that numbers of Vg genes vary between one and five per species (Table 1), and that when a genome contains multiple Vg genes, they are always adjacent. To determine the evolutionary history of these genes, we subsequently constructed a phylogenetic tree using known hymenopteran Vg sequences.
The phylogenetic analysis (Figure 2) revealed that the first duplication of the ancestral Vg gene in ants resulted in two gene subfamilies with different predicted amino acid length (Table 1). The phylogenetic analysis also showed that additional duplications occurred in the lineages leading to Acromyrmex echinatior, Solenopsis invicta and Linepithema humile. Interestingly, the two Pogonomyrmex barbatus genes (Pb_Vg1 and Pb_Vg2) respectively cluster with the S. invicta genes preferentially expressed in queens (Si_Vg2 and Si_Vg3) and foraging workers (Si_Vg1 and Si_Vg4) [29].
To test the prediction that P. barbatus Vg genes display caste-specific expression profiles similar to their closest S. invicta orthologs, we performed quantitative RT-PCR analysis of Vg genes in P. barbatus queens and workers in five independent colonies (Figure S1). On average, Pb_Vg1 was 4.7 times more highly expressed in queens than in nurses (pMCMC <0.0001) and 908 times more highly expressed in queens than in foragers (pMCMC <0.0001). The expression of Pb_Vg2 did not differ between queens and nurses (pMCMC = 0.98), but it was on average 5.7 times more highly expressed in foragers than in queens (pMCMC = 0.0028) (Figure 3).
Furthermore, we tested whether the expression of Vg genes in P. barbatus is associated to task performance as predicted by the RGPH. We found that Pb_Vg1 was significantly more highly expressed in nurses than in foragers in 5 out of 5 colonies (Wilcoxon tests; col1: W = 56, p = 0.001; col2: W = 70, p = 0.0001; col3: W = 56, p = 0.0003; col4: W = 49, p = 0.0006; col5: W = 48, p = 0.0007), while Pb_Vg2 was significantly more highly expressed in foragers than in nurses in 4 out of 5 colonies (Wilcoxon tests; col1: W = 28, p = 0.06; col2: W = 16, p = 0.005; col3: W = 0, p = 0.0003; col4: W = 5, p = 0.02; col5: W = 0, p = 0.0007). On average, Pb_Vg1 was 190 times more highly expressed in nurses than in foragers (pMCMC<0.0001) and Pb_Vg2 6.5 times more highly expressed in foragers than in nurses (pMCMC <0.0001) (Figure 3). Pb_Vg1 is the predominant transcript in workers as it is expressed in strikingly high levels in nurses compared to Pb_Vg2 in foragers (Figure 3); a pattern similar to that observed for the single Vg gene in honeybees.
Finally, we performed molecular evolution analyses to determine the relative contributions of selection for novel biochemical functions (i.e. positive selection), selection for the maintenance of existing biochemical functions (i.e. purifying selection) and neutral evolution in the evolution of ant Vg genes. In particular, we estimated selective pressures on two basal branches of Vg respectively leading to primitive ants and modern (Formicoid) ants, and on the two branches that followed the duplication of Vg in the ancestor of modern ants (Figure 2). Analyses of the branch common to the ancestor of all ants (Formicidae) and the branch common to all modern ants yielded identical results: 60.5% of codon sites evolved under purifying selection (dN/dS = 0.24), 39.5% neutrally (dN/dS = 1), and none had evidence for positive selection. The two branches that followed the duplication of Vg are interesting because one branch includes all Vg genes known to be more highly expressed in queens than in workers (hereafter referred to as subfamily A, which includes Pb_Vg1, Si_Vg2 and Si_Vg3), while the other branch includes all Vg genes more highly expressed in foraging workers than in queens (hereafter referred to as subfamily B, which includes Pb_Vg2, Si_Vg1 and Si_Vg4) (Figure 2 and Table 1). The branch leading to subfamily B shows overall relaxation of purifying selection and no significantly positively selected sites. However, the branch leading to subfamily A shows strong evidence for positive selection (p = 0.008), with a total of 7.1% of sites under positive selection. The three sites with the highest posterior probabilities of being under positive selection in this branch (S44, E382 and N456; pBEB>95%) are part of the main vitellogenin N-terminal lipoprotein domain (LPD-N) that is likely implicated in the uptake of vitellogenin to the ovary [41], providing further support that these changes affect the biochemical properties of the protein.
This study reveals that the genome of P. barbatus harbors two Vg paralogs and that Vg underwent one or several rounds of duplication in other species, demonstrating that the existence of multiple Vg genes is a common phenomenon in ants. The phylogenetic analysis clarifies how Vg genes evolve in ants. First, it shows that the first duplication of the ancestral Vg gene occurred after the divergence between the poneroid and formicoid clades. The poneroid clade includes primitive ants of the subfamily Ponerinae while the formicoid clade includes the three main subfamilies of modern ants: Myrmicinae, Dolichoderinae and Formicinae [28], [42]. Because the divergence between primitive and modern ants apparently coincided with the duplication of Vg genes, it is tempting to speculate that this molecular event could have contributed to the evolution of modern ants. Second, the initial pair of Vg paralogs did not experience further duplication or losses in the lineages leading to Pogonomyrmex barbatus and Atta cephalotes but the ancestor of Pb_Vg2 appears to be lost in Camponotus floridanus. Third, several rounds of duplications occurred independently in the lineages leading to Solenopsis invicta, Linepithema humile and Acromyrmex echinatior, giving rise to multiple Vg genes in each of these species. Intriguingly, the S. invicta Vg genes more highly expressed in queens than in workers (Si_Vg2 and Si_Vg3) cluster with Pb_Vg1 on one branch of the tree, while the genes preferentially expressed in foraging workers (Si_Vg1 and Si_Vg4) cluster with Pb_Vg2 on the other branch of the tree, suggesting that Vg genes in P. barbatus might share a caste-specific expression pattern similar to their closest S. invicta orthologs.
This prediction was confirmed by our analyses showing that Pb_Vg1 is preferentially expressed in queens and Pb_Vg2 in forager. This suggests that Vg gene duplication and subfunctionalization to acquire caste-specific expression related to reproductive and non-reproductive functions may be a general feature of ant species with multiple Vg genes. Expression and functional analyses in additional species will need to be performed to determine the extent to which this is the case.
Three lines of evidence suggest that the first round of duplication of Vg genes facilitated the evolution of queen-worker specialization. First, the duplication occurred in the common ancestor of formicoids (modern ants). A key feature of these ants is a marked morphological, physiological and behavioral differentiation between queens and workers. This contrasts with non-modern ants that exhibit few or no differences between castes. Second, the gene expression differences we identified suggest that the two subfamilies of Vg paralogs evolved different functions, with genes from subfamily A predominantly playing roles in queens and subfamily B predominantly playing roles in workers. Finally, our molecular evolution analyses suggest that these two subfamilies are evolving differently since the duplication. Positive selection on the paralogs in subfamily A may stem from a process of neofunctionalization associated with the evolution of a dramatically higher reproductive output of queens in modern ant species. Furthermore, relaxation of purifying selection on the subfamily B is consistent with the loss of the reproductive constraints and evolution of new functions of Vg (i.e. subfunctionalization) in workers. These results are thus consistent with the proposal that gene duplication followed by caste-specific expression can circumvent the constraints of antagonistic pleiotropy, thus facilitating the evolution of new caste-specific functions in ants [43], [44]. Subfunctionalization of duplicated genes has been described in other contexts. For example, the duplication of a single copy gene in a basal vertebrate gave rise to oxytocin and vasopressin neurotransmitter genes. These two genes have distinct physiological and behavioral roles in vertebrates while the single homologous gene in invertebrates has both vasopressin-like and oxytocin-like functions [45]. Similarly, duplication of estrogen receptor ER-β occurred in the lineage leading to teleost fish. These two copies are expressed in alternate parts of the brain suggesting that subfunctionalization occurred and affects behavior [46].
Interestingly, there is apparently a single copy of Vg in Apis mellifera and other bees. A comparative analysis of the evolution of Vg and seven other genes in A. mellifera and several closely related species showed evidence of positive selection for Vg [47]. In particular replacement polymorphisms were significantly enriched in parts of the protein involved in binding lipid, suggesting a link between the structure of the gene, its function, and its effects on fitness [47]. These data have been interpreted as social pleiotropy leading to only limited constraints on adaptive protein evolution [47], [48]. This raises the question of why multiple duplications occurred in ants but not in bees. A possible reason lies in the much greater phenotypic differences between queens and workers in ants compared to bees. In many modern ants, queens and workers greatly differ in size and other morphological, physiological and behavioral traits [1], [2]. For example ant workers have lost the ability to fly and in some species they are completely sterile. This higher differentiation in ants than bees may lead to greater selection for different roles of Vg in queens and workers, and thus greater potential benefits for Vg duplication and subfunctionalization in ants. Interestingly, the association between the strength of female caste differentiation and the presence of multiple Vg genes seems to hold in social wasps and termites. Although no genome has been sequenced so far in these two groups of social insects, the available data suggest that the social wasps Vespula vulgaris and Polistes metricus (low differentiation between castes) only have a single Vg gene [49], [50] while the termite Reticulitermes flavipes (high differentiation between castes) have two Vg genes [51].
The pattern of expression of Pb_Vg1 (high in queens, medium in nurses and low in foragers) is similar to that observed for Vg in the honeybee [25], [52], suggesting an association between ovarian activity and the expression of this gene. In contrast to honeybee nurses which can produce fertile eggs in queenless conditions and sometimes even in the queen presence, workers are sterile in most Pogonomyrmex species [32]. However, Pogonomyrmex workers can produce trophic eggs which are thought to be the main method of nutrient redistribution because trophallaxis – mouth-to-mouth food transfer - has not been observed in this genus [32]. Interestingly, a recent study in Pogonomyrmex californicus showed that nurses had significantly increased ovarian activity compared to foragers [53], suggesting that trophic eggs are specifically produced by nurses. This pattern of nutrient sharing differs from the honeybee, where the proteins and lipids provided to the larvae, queen and foragers come from the hypopharyngeal and mandibular gland secretions of the nurses [53]–[56]. These results, together with our finding that Pb_Vg1 is predominantly expressed in nurses, suggest that the primary role of this protein as a source of amino acids and lipids for the developing embryo has been co-opted to a novel nutritionally related role associated with the production of trophic eggs. In honeybees Vg expression and ovarian activity are linked with the genetic pathways associated with the regulation of behavior [15], [16]. It remains to be tested if such relation is conserved in sterile ants with functional ovaries that produce trophic eggs.
Our results revealed that the level of expression of Pb_Vg2 was much lower than that of Pb_Vg1 in queens. By contrast Pb_Vg2 was expressed at higher level than Pb_Vg1 in sterile foragers suggesting that the function of this gene is not related to ovarian activity but likely has a different function in foragers. Our molecular evolution analysis showing overall relaxation of purifying selection and no positively selected sites on the paralogs in subfamily B suggests a new role of these proteins neither associated to its existing biochemical function nor novel functions derived of positively selected domains. Instead, the release of functional constrains on the LPD-N structural domain, implicated in lipid transport and Vg receptor binding [41], suggests that these proteins are not imported to the ovary and after having lost their lipid-binding capabilities, they may be primarily used as source of amino acids in foragers. In honeybees, earlier findings that Vg protein was present not only in fertile queens but also in functionally sterile workers [19], [57] and drones [58] led to the proposal that this yolk protein could have both reproductive and non-reproductive functions [59]. Our results indicate that similar functional division of Vg took place in P. barbatus and other modern ant species via gene duplication and subfunctionalization.
The preferential expression of Pb_Vg1 in nurses and of Pb_Vg2 in foragers follow a general gene expression pattern observed in honeybees, in which egg-laying workers show up-regulation of genes associated with reproduction while non-reproductive workers overexpress genes involved in foraging-related functions [60]. This is consistent with the RGPH prediction that reproductive pathways were co-opted to regulate worker behavior and supports the evolutionary theory prediction that potentially reproductive individuals are selected to carry out low-risk tasks, in order to not compromise their reproductive future [61], [62]. In ants, this ancestral mechanism regulating division of labor has been conserved, even in species with workers having lost their reproductive potential.
The results of this study suggest that Vg has been co-opted to regulate worker behavior in the ant P. barbatus as in the honeybee. Support for RGPH in groups of insects that evolved sociality independently, demonstrates that the co-option of reproductive pathways to regulate behavior is a major director in the evolution of sociality in insects. On the other hand, the expression of one Vg paralogs in sterile workers reveals that Vg adaptation to regulate worker behavior is not necessarily linked to reproduction but maybe linked exclusively to nutritional functions. Our result suggests that, after the initial duplication in ants, Vg genes underwent neo- or subfunctionalization to acquire caste- and behavioral-specific functions. Overall, our results suggest that even though ants and bees evolved sociality independently, they have conserved similar mechanisms to regulate division of labor.
To determine gene models, we first ran TBLASTN using known hymenopteran Vg sequences against ant genome sequences downloaded from the fourmidable database [63]. Subsequently, ruby/bioruby scripts [64] were used to extract relevant subsets of each genome. Gene predictions were generated on each subset using MAKER2 [65] s65mand subsequently manually refined using Apollo [66]. Conflicts gene predictions were resolved by using EST data when available, splice prediction algorithms (http://www.fruitfly.org/seq_tools/splice.html) and manual verification of splicing consensus sequences.)
Inaccurate sequence alignment or phylogeny leads to misleading or incorrect results in molecular evolution analyses. We used an approach centered on the use of phylogenetically aware codon-level aligner PRANK, which is likely to minimize the risks of introducing errors [67], [68]. This required several steps. We preliminarily aligned the 26 Vg protein sequences with MAFFT linsi [69] and removed ambiguous sections of the alignment using trimAl (option -gappyout) [70]. A first tree was built with RAxML (model GTRGAMMAI) [71] and rooted with “nw_root” (Newick Utilities package [72]). This tree was used as a guide tree in PRANK [73] to obtain a high-quality codon-level alignment of the 26 Vg coding sequences. Ambiguous sections of the alignment were removed using trimAl (option -gappyout) and a final tree was built with RAxML (GTRGAMMAI model); 10,000 bootstraps were generated to assess its confidence. Selective pressures (dN/dS) on different parts of the phylogenetic tree were estimated using the branch-site codon-substitution model from CodeML (PAML 4.6) [74]. Such dN/dS ratios are obtained by computing the number of non-synonymous changes (dN) over synonymous changes (dS) (see Table 2 for more details). Vg site coordinates (S44, E382, N456) are given as in Apis mellifera Vg (Uniprot identifier VIT_APIME).
Pogonomyrmex barbatus founding queens were collected during nuptial flights on July 15th, 2008 in Bowie, Arizona, USA (N32°18′54″//W109°29′03″). Colonies were then kept in laboratory conditions (30°C, 60% humidity and 12 h/12 h light∶dark cycle) in 15×13×5 cm plastic boxes with water tubes, and were fed once a week with grass seeds and a mixture of eggs, honey and smashed mealworms. Thirty months later, five of these colonies were used to collect samples on December 16th, 2010. Task performance in workers is age related, thus nurses tend to be younger than foragers [2]. Young ants interacting with the brood in the nest tube were considered as nurses. To collect foragers, each colony was connected with a cardboard-made bridge to a foraging area composed of a plastic box containing grass seeds. Any ant handling a food item in the foraging area was considered as forager. Ant samples were flash-frozen in liquid nitrogen and kept at −80°C for further RNA extraction.
Whole body worker samples were used to measure the expression of Pb_Vg1 and Pb_Vg2 genes. RNA extractions were performed using a modified protocol including the use of Trizol (Invitrogen) for the initial homogenization step, RNeasy extraction kit and DNAse I (Qiagen) treatment to remove genomic DNA traces. For each individual worker, cDNAs were synthesized using 500 ng of total RNA, random hexamers and Applied Biosystems reagents. Levels of mRNA were quantified by quantitative real-time polymerase chain reaction (qRT-PCR) using ABI Prism 7900 sequence detector and SYBR green. All qPCR assays were performed in triplicates and subject to the heat-dissociation protocol following the final cycle of the qPCR in order to check for amplification specificity. qRT-PCR values of each gene were normalized by using an internal control gene (RP49). Paralog-specific primers (Table S1) were designed using sequence alignment [75] and primer analysis [76] programs. Primer sequences overlapped coding regions split by introns, allowing the specific amplification of cDNA levels over eventual genomic DNA contaminations. Transcript quantification calculations were performed by using the ΔΔCT method [77].
All data were analyzed using R (http://www.r-project.org/) and the R packages lme4 [78] and language R [79]. The effect of caste on gene expression relative values was analyzed using linear mixed effects models. To avoid pseudoreplication, the colony was included as a random effect. We checked for normality and homogeneity by visual inspections of plots of residuals against fitted values. To assess the validity of the mixed effects analyses, we performed likelihood ratio tests to confirm that the models with fixed effects differed significantly from the null models with only the random effects. Throughout the paper, we present MCMC (Markov-chain Monte Carlo) estimated p-values that are considered significant below the 0.05 threshold (all significant results remained significant after Bonferroni correction).
|
10.1371/journal.pgen.1001031 | A Collection of Target Mimics for Comprehensive Analysis of MicroRNA Function in Arabidopsis thaliana | Many targets of plant microRNAs (miRNAs) are thought to play important roles in plant physiology and development. However, because plant miRNAs are typically encoded by medium-size gene families, it has often been difficult to assess their precise function. We report the generation of a large-scale collection of knockdowns for Arabidopsis thaliana miRNA families; this has been achieved using artificial miRNA target mimics, a recently developed technique fashioned on an endogenous mechanism of miRNA regulation. Morphological defects in the aerial part were observed for ∼20% of analyzed families, all of which are deeply conserved in land plants. In addition, we find that non-cleavable mimic sites can confer translational regulation in cis. Phenotypes of plants expressing target mimics directed against miRNAs involved in development were in several cases consistent with previous reports on plants expressing miRNA–resistant forms of individual target genes, indicating that a limited number of targets mediates most effects of these miRNAs. That less conserved miRNAs rarely had obvious effects on plant morphology suggests that most of them do not affect fundamental aspects of development. In addition to insight into modes of miRNA action, this study provides an important resource for the study of miRNA function in plants.
| MiRNAs are small RNA molecules that play an important role in regulating gene function, both in animals and in plants. In plants, miRNA target mimicry is an endogenous mechanism used to negatively regulate the activity of a specific miRNA family, through the production of a false target transcript that cannot be cleaved. This mechanism can be engineered to target different miRNA families. Using this technique, we have generated artificial target mimics predicted to reduce the activity of most of the miRNA families in Arabidopsis thaliana and have observed their effects on plant development. We found that deeply conserved miRNAs tend to have a strong impact on plant growth, while more recently evolved ones had generally less obvious effects, suggesting either that they primarily affect processes other than development, or else that they have more subtle or conditional functions or are even dispensable. In several cases, the effects on plant development that we observed closely resembled those seen in plants expressing miRNA–resistant versions of the major predicted targets, indicating that a limited number of targets mediates most effects of these miRNAs. Analyses of mimic expressing plants also support that plant miRNAs affect both transcript stability and protein accumulation. The artificial target mimic collection will be a useful resource to further investigate the function of individual miRNA families.
| MicroRNAs (miRNAs) are a class of small RNA (sRNA) molecules that has recently emerged as a key regulator of gene activity. In plants, miRNAs are released from larger precursors (pri-miRNAs) in the nucleus mainly, by DICER-LIKE1 (DCL1) [1]. The resulting sRNA duplex is methylated and translocated to the cytoplasm where it can be loaded into an RNA-induced silencing complex (RISC) that includes a member of the ARGONAUTE (AGO) family as catalytic component. The RISC can then recognize mRNAs containing sequences complementary to the loaded miRNA [2]. In plants, cleavage of the target mRNA is an important mechanism for plant miRNA action, but there are also direct effects on protein accumulation, as reported for many animal miRNAs [3]–[11].
The spatio-temporal expression pattern of miRNA genes is regulated to a large extent at the transcriptional level, and different members of a miRNA family can have distinct, specialized expression domains [12]–[17]. An additional layer of regulation in miRNA action has been reported by Franco-Zorrilla and colleagues [18]. IPS1 (INDUCED BY PHOSPHATE STARVATION 1) encodes a non-coding RNA with a short motif that is highly complementary to the sequence of miR399, which like IPS1 is involved in the response to phosphate starvation [19]–[23]. In contrast to regular miRNA target sites, the IPS1 sequence contains a three-nucleotide insertion in the center, corresponding to the position where normally miRNA-guided cleavage takes place, and this bulge in the miRNA/target pair prevents endonucleolytic cleavage of IPS1 transcripts. This results in sequestration of RISCmiR399, leading to a reduction of miR399 activity. A similar phenomenon, negative regulation of small RNA activity by a partially complementary mRNA, has been recently described in bacteria as well [24], [25].
MiRNA target mimicry can be exploited to study the effects of reducing the function of entire miRNA families [18]. Simultaneous inactivation of all miRNA family members by constructing multiply mutant lines has so far been achieved for only two relatively small families [16], [26]. Plant target mimics are conceptually similar to miRNA sponges, used to reduce miRNA activity in animals. MiRNA sponges are transcripts containing multiple miRNA binding sites that compete with endogenous target mRNAs, thereby reducing the efficiency of the corresponding miRNA [27]. Although in animals perfect-match miRNA binding sites seems sufficient to sequester miRNAs [28], such optimal sites would be generally cleaved in plants, and they would not succeed in sequestering the miRNA-loaded RISC. Consistent with this, plants overexpressing non-modified versions of miR156 and miR319 target genes show much milder phenotypes than plants expressing the corresponding target mimics [18], [29], [30]. Modifications of the miRNA binding site that prevent cleavage but still allow miRNA binding are therefore required to reduce miRNA activity in plants.
Here, we present a collection of transgenic plants expressing artificial target mimics designed to knockdown the majority of Arabidopsis thaliana miRNA families. One fifth of these lines have obvious morphological defects, which is in the same range as the approximately 10% of miRNA knockouts that caused phenotypic abnormalities or lethality in Caenorhabditis elegans [31]. We found a clear correlation between the evolutionary conservation of plant miRNA families and their effect on aerial plant morphology.
We generated artificial target mimics for 73 different families or subfamilies of miRNAs and expressed them in Arabidopsis thaliana plants under the control of the constitutive 35S CaMV promoter. As described [18], we modified the 23 nucleotide, miR399-complementary motif in IPS1. The different constructs, and the corresponding transgenic lines, are named “MIM”, followed by the numeric identifier of the targeted miRNA family or subfamily. We targeted all miRNA families reported in miRBase (http://microrna.sanger.ac.uk/sequences/index.shtml) and ASRP (http://asrp.cgrb.oregonstate.edu) [32] at the beginning of 2007, plus some of the miRNAs described subsequently [33]. The majority of the analyzed families have only been described in Arabidopsis thaliana and Arabidopsis lyrata [34], [35]. The remaining families are shared with other angiosperms, and less than a quarter has been detected in non-flowering plants, including gymnosperms, ferns or mosses [32], [33], [36], [37]. A complete list of MIM constructs, and the primer pairs used to generate them, can be found in Table S1. For miRNA target predictions, see [8], [33], unless stated otherwise.
A single artificial target mimic could be designed for most miRNA families. The mature miRNAs produced by members of the miR169 and miR171 families differ slightly, and different target mimics were designed for these subfamilies. Two target mimics were also designed for the miR161 family, which produce two mature miRNAs that have only partially overlapping sequences, and that target similar subsets of the PPR gene family [38]. Conversely, some miRNA families have very similar sequences and overlapping in vivo targets (e.g., miR159/319, miR156/157 and miR170/171a), and artificial target mimics might not be able to unambiguously discriminate between different miRNAs.
In some cases, the sequence of the bulge in the miRNA/target mimic pair had to be modified. For example, maintaining the original central sequence of IPS1 in MIM172 could have reconstituted a cleavage site for miR172. Consistent with such modifications being important, plants expressing the appropriately modified version of MIM172 showed an altered phenotype (see below), whereas plants expressing an initial version of MIM172 in which a putative miR172 cleavage site was present (MIM172cs) did not. Moreover, plants expressing a MIM172 version with only a single-nucleotide mismatch corresponding to position 11 of the mature miRNA (MIM172sn) did not show any abnormal phenotype either, suggesting that the three-nucleotide bulge is required for target mimic activity (Figure 1).
We generated at least 20 independent transformants for each of 75 separate constructs. Of these, 15, targeting 14 different families, caused reproducible phenotypes in the shoot system of the plants, which are described below. Phenotypic alterations were consistent across most, if not all, independent transformants examined for each construct. An example of the phenotypic variation among primary transformants is shown in the histograms in Figure 1. An overview of all lines with morphological defects is given in Table 1, together with the main target genes of the corresponding miRNA family and a list of other taxa in which they can be found. The phenotypes of MIM156 and MIM319 plants have been briefly described before [18], [39]. All miRNA families whose inactivation resulted in visible phenotypical alterations are conserved among the angiosperms, and most of them are also found in non-flowering plants.
MIM156 and MIM157 plants (Figure 2) had reduced leaf initiation rates, such that they flowered at about the same time as wild type, but with only two or three true leaves. This phenotype is similar to what is seen in plants carrying non-targetable versions of SPL9 or SPL10, two of the miR156/157 targets, and opposite of plants overexpressing miR156b or spl9 spl15 double mutants [10], [40]–[42]. In addition, these plants had bent, spoon-shaped cotyledons. The few rosette leaves were characterized by serrated margins, indicating adult leaf identity, consistent with a role of miR156 and its targets in controlling phase change [30].
MIM159 plants had extensive pleiotropic defects, and similar phenotypes were observed in most MIM319 lines. These plants had reduced stature, with rounder, upward curled leaves (Figure 2), shorter stem internodes, and smaller flowers with short sepals, reduced petals and anthers that did not develop completely. More severe MIM319 lines were progressively smaller, had warped leaves and lacked well-developed petals (Figure 3A). Stem elongation was often completely suppressed (Figure 3B). Most plants had reduced fertility, and this phenotype was particularly severe in MIM319 plants, for which only a few viable seeds could be recovered after they were grown for prolonged periods at 16°C in long days. Both vegetative and floral phenotypes reminiscent of MIM159 defects have been reported for plants that express non-targetable forms of miR159 target genes [29], and in plants doubly mutant for miR159a and miR159b [26]. In particular, upward curled leaves have been observed in plant expressing non-targetable forms of MYB33, which can be targeted both by miR159 and miR319 [43]. Milder MIM319 lines showed different leaf defects, with leaves curled downward (Figure 2). This is consistent with what has been reported for plants that express non-targetable forms of TCP2 and TCP4, which are both exclusive miR319 targets [29], suggesting that target mimics can at least partially discriminate between these two miRNA families.
Serrated and hyponastic leaves were seen in MIM160 plants (Figure 2), in agreement with the phenotype of plants that express non-targetable versions of ARF10 or ARF17, two of the three miR160 targets [44], [45]. In addition, MIM160 plants were smaller than wild type. Compared to other constructs, fewer transformants were recovered, consistent with the known requirement of miR160 for seed viability or germination [44].
A different type of leaf serration was caused by MIM164 (Figure 2), similar to what has been reported for plants expressing a non-targetable version of CUC2, one of the miR164 targets, and for plants lacking one of the miR164 isoforms, miR164a [13]. While expression of MIM160 affected the entire leaf, with the serrations being regular and jagged, MIM164 caused mainly serration of the basal part of the leaf, with more irregular and rounded sinuses and teeth (Figure 3C). Although carpel fusion defects have been described for plants lacking miR164c [12], the carpel defects in MIM164 plants seemed to be different, with ectopic growths forming at the valve margins (Figure 3D), resembling those seen in the cuc2-1D mutant, in which a point mutation affects the miR164 complementary motif in CUC2 [46]. In some cases, this tissue could develop into adventitious pistil-like structures (Figure 3E).
Rounder leaves with an irregular surface, which appeared to be hollowed out between the main veins, were caused by MIM165/166. Younger leaves tended also to be cup-shaped (Figure 2). Targets of miR165/166, including the transcription factor-encoding genes PHAVOLUTA and PHABULOSA, control leaf polarity, and dominant mutations that disrupt the miRNA target site in these genes cause severe alterations in leaf morphology [47]–[49].
A substantial delay in flowering was observed in MIM167 plants, which flowered with 20.8±4.2 (mean ± standard deviation; n = 30) leaves in long days, compared to 13.0±0.9 rosette leaves in wild-type plants (Figure S1A and Figure S2). These plants had in addition twisted leaves (Figure 2), as well as defects in the maturation of anthers (Figure 3F) and in the development and shattering of seeds, which often remained attached to the dehiscent siliques (Figure 3G), resulting in reduced seed production and germination (not shown). This is consistent with what has been observed in plants that express a non-targetable form of the miR167 target ARF6 or ARF8. Such plants have smaller leaves and are often sterile due to defects both in ovule and anther development [17]. Effects on flowering time have not been previously associated with miR167 [17], [50], and the late-flowering phenotype of MIM167 plants reveals a new role for this miRNA family.
Two constructs were used to downregulate different subfamilies of miR169 family, whose main targets are HAP transcription factors. MIM169 was designed for miR169a, b, c, h, i, j, k, l, m and n, and MIM169defg for miR169d, e, f and g. Both target mimics reduced the size of transgenic plants (Figure 2).
MiR170 and miR171 target a group of SCARECROW-like transcription factor genes [9], and both MIM170 and MIM171A plants had round, pale leaves (Figure 2), as well as defective flowers, with sepals that did not separate properly, resulting in reduced fertility (Figure 3H and 3I). Expression of target mimics against the b and c members of the miR171 family did not confer any phenotype, suggesting less important roles for these two miRNAs.
MIM172 plants were also late flowering, with 20.0±3.5 (n = 30) rosette leaves in long days (Figure S1B), consistent with the flowering time phenotype of plants that have increased expression of miR172 targets [4], [6], [51]. In addition, leaves of MIM172 plants appeared to be somewhat narrower than those of wild type, and mildly curled downward, and severe MIM172 lines presented reduced apical dominance (not shown). In contrast to plants that express a non-targetable version of AP2 [52], flowers of MIM172 plants were normal. These differential effects could be due to the particularly high levels of miR172 levels during early flower development [6].
MiR393 targets a small group of auxin receptor genes. MIM393 plants had mild defects in leaf morphology, with narrow leaves that were curled downward (Figure 2). Leaf epinasty is often associated with high auxin levels [53], and is consistent with an increase of auxin signaling caused by downregulation of miR393 activity.
Finally, epinastic leaves were observed also in MIM394 plants (Figure 2). MiR394 is predicted to target a gene encoding an F-box protein.
Artificial target mimics are thought to sequester their target miRNAs, presumably by stably binding to miRNA-loaded RISCs. To obtain additional evidence for such interactions, we embedded a functional MIM159 site in the 3′-UTR of a triple- Enhanced Yellow Fluorescent Protein (EYFP) reporter; stable recruitment of RISCmiR399 to the mimic site could be expected to interfere with EYFP translation. In 80% of MIM159 expressing T1 plants, as in control plants, the EYFP transgene was completely silenced. In the remaining 20%, we detected EYFP signal that was strongly reduced in the region where MIR159 genes are known to be expressed (Figure 4A) [26]. In addition, these plants presented the typical phenotypic defects of MIM159 plants, confirming that the EYFP:MIM159 construct functions properly as a target mimic.
RISCmiRNA sequestration in turn should relieve target genes from miRNA-dependent regulation, resulting in increased levels of the encoded protein. In agreement with such a scenario, activity levels of a genomic MYB33:GUS reporter were markedly increased in MIM159 plants (Figure 4A). In analogy with EYFP:MIM159, reporter activity was increased in the tissues expressing MIR159 genes [26], as expected.
Sequestration of RISCmiR399 by the natural target mimic IPS1 prevents miR399-guided cleavage of PHO2 mRNA, thus increasing PHO2 mRNA levels [18]. To assess the effects of artificial target mimics on the levels of mRNA of miRNA target genes, we tested them by reverse transcription followed by quantitative PCR (qRT-PCR) in a subset of MIM lines. We preferentially analyzed organs in which miRNA abundance was high according to the ASRP database [32], [54], or organs with major phenotypic alterations in MIM lines. Two independent lines were tested for each construct. Among the miRNA targets, we chose ones known to induce phenotypic defects when expressed as non-targetable forms [44], [45], [47] and ones that show altered expression in miRNA biogenesis mutants [32], [54], [55]. PCR products spanned the miRNA target sequence, allowing quantification of the attenuation in slicing activity by the corresponding miRNA. Surprisingly, in most cases there were no major changes in target transcript levels (Figure 4B and Figure S3).
For comparison, we examined the expression of the same miRNA target genes in seedlings of several mutants impaired in small RNA biogenesis and function, including dcl1-100, se-1, hyl1-2 and ago1-27, and in plants overexpressing viral silencing suppressors that are known to counteract the action of the small RNA machinery, including P1/HC-Pro, P0, P19 and p21 [56]–[60]. In most cases, the changes seen in MIM lines correlated with those seen in miRNA biogenesis mutants. Stronger effects were observed only in dcl1-100 plants (Figure 4C). These results are consistent with what has been observed in microarray studies of miRNA biogenesis mutants, including other dcl1 alleles, se and hyl1 [55], .
As in animals, inhibition of translation is an important component of miRNA function in plants [4], [6], [11]. To test whether artificial mimics impact miRNA effects independent of changes in target transcript accumulation, we monitored the protein levels produced by CIP4, a gene that is regulated by miR834 through translational inhibition [5], [62]. In MIM834 lines, CIP4 levels were appreciably increased, while CIP4 mRNA levels were unchanged (Figure 4D). Direct effects on protein translation could explain the absence of a clear correlation between target mRNA levels and plant phenotype in plants expressing artificial target mimics.
Finally, we investigated the levels of mature miRNAs in plants expressing artificial target mimics. In all MIM lines we examined, levels of the targeted miRNA were decreased, suggesting that unproductive interaction of RISCmiRNA with a decoy affects miRNA stability (Figure 4E). Although such an effect has not been observed in case of the endogenous IPS1-miR399 interaction [18], a similar reduction in small RNA levels triggered by a target mimic has been reported in bacteria [24], [25].
We have generated a collection of transgenic plants expressing artificial target mimics designed to reduce activity for most of the known miRNA families in Arabidopsis thaliana. Inhibiting the function of 14 out of 71 miRNA families with target mimics led to morphological abnormalities. All of these families belong to the more abundant and widely conserved miRNA families, which were the first ones to be discovered (Table 1). This agrees with results from experiments in which miRNAs were overexpressed, miRNA target genes were mutated, or miRNA genes were inactivated by conventional knockouts [reviewed in 63]. Together, these findings are consistent with the scenario of frequent birth and death of miRNA genes, with only a few becoming fixed early on during evolution because they acquired a relevant function in plant development [33], [36]. More recently evolved, species-specific miRNAs could instead play a role in adaptation to certain abiotic or biotic challenges, or have no discernable function at all. Some miRNAs are known to regulate physiological traits, and they do not cause morphological abnormalities under standard benign conditions [20], [21], [64]. Such conditional effects would have escaped our screen, as would have defects in the root system of the plant. Moreover, compared to expression of non-targetable forms of miRNA target genes, or miRNA loss-of-function mutants, the defects of MIM plants were often weaker. Examples are the absence of an altered floral phenotype in MIM172 plants, which is seen in plants that express a non-targetable version of AP2 under the control of normal regulatory sequences [52], or the extra-petals phenotype seen in mir164c mutants, but not in MIM164 plants [12]. Another caveat is that some miRNAs might be required for embryonic development, in which case only lines with relatively weak expression of the artificial target mimic might have survived. Such limitations could be overcome by tissue-specific or inducible expression of target mimics. On the other hand, while artificial mimics increase levels of individual miRNA target genes less strongly than what can be achieved by expression of miRNA-resistant forms, mimics have the advantage that they affect all targets simultaneously [18]. Apart from translational regulation [3]–[7], another possibility for the absence of a clear correlation between phenotypic severity and change in mRNA levels of miRNA targets could be that many miRNAs affect their targets only in a small set of cells. In these cases, assaying expression in whole organs would obscure the effects of miRNA downregulation on mRNA levels.
It has recently been suggested that plant miRNAs could also repress the translation of target mRNAs that have only limited sequence complementarity, as often happens in animals [5]. Support for the existence of miRNA binding sites with reduced complementarity in plants comes from an analysis of miR398, which regulates COPPER SUPEROXIDE DISMUTASE (CSD) genes. Certain mutations in the miR398 complementary motif site reduced the effects of miR398 on CSD mRNA, but not on protein levels [3]. We have shown that mimic-like sites, when introduced into the 3′-UTR of a protein-coding gene, not only are active in sequestering the targeted miRNA, but can also reduce protein levels produced by the mRNA linked in cis. This reduction likely occurs at the translational level, since mimic sites are not subject to miRNA-dependent slicing [18]. This observation opens an intriguing scenario in which mRNAs containing mimic-like sites, or possibly other sites with reduced complementarity to miRNAs, are regulated by miRNAs exclusively through translational inhibition. A further level of complexity is added by such sites reducing the effects of an miRNA on other mRNA with a sliceable miRNA targeting motif, similarly to what has been recently proposed in animal systems [65].
Nevertheless, as pointed out before [43], miRNA overexpression and knockout of major target genes normally produce very similar phenotypes, and these are generally the opposite of what is seen in plants with reduced activity of the miRNA. These observations are supported by our finding of extensive similarities between phenotypes caused by target mimics and by expressing resistant forms of individual targets. We conclude that, at least for the instances in which developmental defects could be observed, target genes with extensive complementarity likely account for the majority of miRNA effects, but that in certain cases targets regulated solely through translational inhibition via diverged target sites might be important as well.
Plants were grown on soil in long days (16 h light/8 hours dark) under a mixture of cool and warm white fluorescent light at 23°C and 65% humidity. The se-1, ago1-27, and hyl1-2 and dcl1-100 mutants have been described [66]–[69]. MIM834 plants were grown on MS media plates supplemented with 1% sucrose for 14 days in long days at 23°C. Plants overexpressing viral proteins Hc-Pro, P0, P19 and P21 were a kind gift from the Carrington lab.
Artificial target mimics were generated by modifying the sequence of the IPS1 gene [18]. All target mimics constructs were placed behind the constitutive CaMV 35S promoter in the pGREEN vector conferring resistance to BASTA [70]. For the MYB33:GUS reporter, a MYB33 genomic fragment was PCR amplified, cloned into the TOPO-PCR8 Gateway vector (Invitrogen), and recombined through LR clonase reaction into pGWB433 [71] to generate a GUS translational fusion. The MIM159 construct was introduced into three independent MYB33-GUS T2 lines. A MIM159 site was placed in the 3′-UTR of a triple-EYFP sequence linked to a fragment encoding a nuclear localization signal (NLS) and driven by a CaMV 35S promoter. Constructs were introduced into A. thaliana (accession Col-0) plants by Agrobacterium tumefaciens-mediated transformation [72].
Nine-day-old seedlings from three independent T2 lines for all the GUS reporter backgrounds were fixed in acetone 90%. GUS activity was assayed as described [73].
Total RNA was extracted from 11-day old seedlings and 30-day old inflorescences (47 days for the MIM172 lines), using TRIzol Reagent (Invitrogen). For dcl-100, 13-day old seedlings were collected, to obtain a similar developmental stage compared to the other plants. For real time RT-PCR, two biological replicates with tissue pooled from 8 to 10 plants were assayed from two independent MIM lines per miRNA family or subfamily. Complementary DNA was produced with the RevertAid First Strand cDNA Synthesis Kit (Fermentas), using as starting material 4 µg of total RNA that had been treated with DNase I (Fermentas). PCR was carried out in presence of SYBR Green (Invitrogen) and monitored in real time with the Opticon Continuous Fluorescence Detection System (MJR). Oligonucleotide primers are given in Table S2. Small RNA blots were performed on the same RNA used as template for real time RT-PCR, with DNA oligonucleotides as probes.
Proteins were extracted from four MIM834 lines using a Tris buffer (50 mM Tris pH 7,5; 150 mM NaCl; 1 mM EDTA; 10% [v/v] Glycerol; 1 mM DTT; 1 mM Pefablock and 1 complete protease inhibitor cocktail [Roche]). Protein concentration was measured using a commercial Bradford assay (BioRad). 50 µg of raw protein extract per sample were resolved on an 8% acrylamide gel. Blotting and antibody incubation were performed as described [5], except that the secondary antibody was incubated for 8 hours at 4°C. Two biological replicates from 4 independent lines were analyzed. Band intensity was measured using the ImageJ software (http://rsbweb.nih.gov/ij/).
|
10.1371/journal.pcbi.1000847 | Promoter Complexity and Tissue-Specific Expression of Stress Response Components in Mytilus galloprovincialis, a Sessile Marine Invertebrate Species | The mechanisms of stress tolerance in sessile animals, such as molluscs, can offer fundamental insights into the adaptation of organisms for a wide range of environmental challenges. One of the best studied processes at the molecular level relevant to stress tolerance is the heat shock response in the genus Mytilus. We focus on the upstream region of Mytilus galloprovincialis Hsp90 genes and their structural and functional associations, using comparative genomics and network inference. Sequence comparison of this region provides novel evidence that the transcription of Hsp90 is regulated via a dense region of transcription factor binding sites, also containing a region with similarity to the Gamera family of LINE-like repetitive sequences and a genus-specific element of unknown function. Furthermore, we infer a set of gene networks from tissue-specific expression data, and specifically extract an Hsp class-associated network, with 174 genes and 2,226 associations, exhibiting a complex pattern of expression across multiple tissue types. Our results (i) suggest that the heat shock response in the genus Mytilus is regulated by an unexpectedly complex upstream region, and (ii) provide new directions for the use of the heat shock process as a biosensor system for environmental monitoring.
| Adaptation of sessile animals, such as molluscs, to stress is achieved by a number of molecular mechanisms, few of which are clearly understood. Insights from this research can provide clues about stress tolerance both for sessile and mobile organisms. The Mediterranean mussel, of the genus Mytilus, is a model organism for the study of stress at the molecular level, with sufficient gene structure and function data available. We have thus investigated a key stress response gene, Hsp90, and in particular its upstream region, using a combination of sequence and expression analysis approaches. We demonstrate that this region, responsible for the regulation of heat shock-associated gene expression, exhibits an unparalleled structural and functional complexity compared to other model organisms, as well as subtle gene expression patterns across multiple tissues. These results form the basis upon which the heat shock response can be used as a molecular biosensor for environmental monitoring in the future.
| The majority of molluscan species go through two principal developmental phases, a larval embryo (motile phase) followed by a clumping structure (sessile phase), when they are permanently attached to an underwater substrate. This lifecycle, common amongst marine invertebrates, poses challenges for adaptation and tolerance for a wide range of conditions at the littoral zone, including steep salinity or temperature gradients. Key model organisms for molluscan biology include species from the genus Mytilus, in particular M. edulis, M. galloprovincialis and M. californianus. Crucially, the latter species is a target organism for a genome sequencing project, whose results are eagerly expected by the community (http://www.jgi.doe.gov/sequencing/why/3090.html).
The Mytilus species group provides an ideal model both for fundamental questions of animal adaptation to stress response, as well as biotechnological applications, primarily as a pollution biosensor [1]. Its use extends into biomimetics [2], in particular protein-based medical adhesives [3], with potential applications in fields such as dentistry [4]. Moreover, its relatively complex developmental structure and higher taxonomic status as an invertebrate, combined with the fact that it can suffer from mussel haemic neoplasia, renders this organism a potential model for human leukemia and an ideal biomarker for pollution-induced disease [5]. In this context, it is important to understand the mechanisms by which mussels tolerate and cope with environmental stress, given that their behavioral options are highly restricted, due to the sessile phase of their lifecycle.
In the past, comparisons between motility and sessility for higher organisms have been primarily confined to animals versus plants [6], with follow-up studies focusing on comparisons between large animals, e.g. humans, versus large plants, e.g. trees, and the trade-offs for the tree body plan [7]. Less attention has been paid to adaptations by sessile animals, in particular intertidal invertebrates (or, “plant” equivalents) [8]–[9], and the molecular mechanisms through which they achieve tolerance to stress. One exception is represented by heat shock response, a key factor for temperature adaptation that has been studied in this context to a certain extent [8], and specifically in Mytilus with regard to the Hsp70 [10] and Hsp90 [11] genes.
Transcriptional regulation can be achieved either by an extensive repertoire of paralogs and transcription factors (‘gene content strategy’) or a complex structure of promoters (‘gene structure strategy’). Analysis of comprehensive datasets has clearly demonstrated that transcription factors (TFs) and transcription-associated proteins (TAPs) are not universally distributed but highly taxon-specific and that relative TF gene content increases with the taxonomic scale [12]–[13]. Such comparisons have been later extended by follow-up studies that analyzed TAP complements and their expansion rates in plants [14]–[15]. Thus, it is now known that one way by which plants, sessile organisms par excellence, achieve a finer degree of regulation is by the expansion of TF/TAP complements and a ‘gene content strategy’. Yet, it is unclear whether similar trends are followed in sessile animals, since entire genome sequences for those are lacking so far, limiting the range of comparative genome-wide studies that can be performed.
As far as paralogs are concerned, recent studies that have focused on the heat shock response in plants, and in particular Arabidopsis thaliana, have revealed that the process involves up to 21 known TFs and four heat shock protein (Hsp) families (Hsp20/70/90/100) [16]–[17]. Despite a cursory resemblance to mammals, in Drosophila thermal sensing is achieved by a unique repertoire of genes [18], including thermostat systems not exclusively involving heat shock proteins [19]. In other words, and probably for different reasons, a gene content strategy might prevail in both model organisms for plants (A. thaliana) and motile invertebrates (Drosophila). Thus, it is worth examining what are the mechanisms through which stress response is regulated in sessile marine invertebrates in general, and the Mytilus genus in particular, and which strategy dominates gene expression.
We focus on the Hsp90 family as a case study for stress response in sessile animals and examine the structure and function of the Hsp90 upstream region in M. galloprovincialis. Previously, two distinct Hsp90 genes with the same genomic organization have been isolated from M. galloprovincialis [11], herein called Mghsp90 genes. Detailed sequence analysis revealed that the two genes contain nine exons and exhibit great similarities in both the 5′ non-coding and the coding regions but differ in their 3′ non-coding regions, as well as in three introns, due to the presence of repeated sequences [11]. The 5′ non-coding region of both genes contains a non-translated exon and multiple binding sites for various transcription factors, highly suggestive of potential interactions of these factors with the Hsp90 promoter and subtle patterns of gene regulation [11].
A comparative analysis of Hsp90 gene content across all taxa with available sequence data has clearly shown that invertebrate genomes contain a relatively small number of Hsp90 genes (3–4 genes), compared to those of vertebrates (>5 genes) [20]. Thus, it appears that the Mytilus genome might contain a relatively small number of TFs (e.g. heat shock factors or HSFs – no such factors can be detected in the Mytilus californianus EST collection, not shown) and/or Hsp90 genes, raising the question how the expression of Hsp90 and other heat shock genes is regulated in sessile invertebrates.
In the present work, we perform a detailed analysis of the Mghsp90 upstream region in terms of structure and expression, and reveal the presence of previously undetected sequence elements of unknown function. Based on tissue-specific expression data, we also delineate the potential associations of Mghsp90 with another 174 genes that are involved in a complex pattern of expression across tissues. These two discoveries are discussed within the context of existing knowledge and are expected to contribute towards a deeper understanding of the heat shock response in sessile organisms.
The comparison of the 5′ upstream region of Mghsp90 genes to their homologs in two model organisms for which there is extensive genomic evidence and humans reveals an increase of complexity in TF binding sites including heat shock elements (HSEs, binding sites for HSFs – see Methods). The M. galloprovincialis Hsp90 region exhibits a peculiar degree of unexpected complexity with regard to its phylogenetic context, not only in terms of quantity of predicted elements but also in fine structure of the promoter (Figure 1). The Mytilus region contains more regulatory sites than the D. melanogaster region (namely, 14 sites vs. 8), a total count similar to that of the human Hsp90 beta gene (17 sites – Figure 1). Moreover, it is host to two newly identified elements (Gamera and a genus-specific sequence), both of unknown function (represented by blue bars, Figure 1), followed by a HSE-rich region with a CAAT binding site and a putative p53 binding site (see also below, and Figure 1 in Protocol S1).
Curiously, upstream of the first exon (1158 nucleotides) of Mghsp90, there exists a 201-base pair (bp) sequence element with a putative GAGA factor binding site (Figure 1), 69% identical over 181 nucleotides to the medaka fish Oryzias curvinotus LINE-like repetitive sequence Gamera [21]. The similarity extends over positions 1907–2085 of the O. curvinotus 4493-bp sequence entry (Genbank accession number AB081572, GI:19570857) and more specifically over the ‘open reading frame’ b (defined at positions 1353–3052) [21]. Thus, this region of approximately 200 nucleotides is only a fraction of the putative ORF b and, to our knowledge, it is the first time this segment is reported outside the Oryzias genus and its closest relatives [21] (Figure 2). Moreover, multiple copies of this region can also be identified in the genome of the blood fluke Schistosoma mansoni [22] (Figure 2, Figure 2.1 in Protocol S1). Fragments of this sequence are also present in (i) the Expressed Sequence Tag (EST) database, more specifically in the neural transcriptome and thus genome of the gastropod Aplysia californica [23], the termite Hodotermopsis sjoestedti [24], the African cichlid fish Oreochromis niloticus (Lee et al., unpublished, GI: 253867024), the mollusc Lymnaea stagnalis [25] and the sea anemone Nematostella vectensis [26] (in that order of sequence similarity – Figure 2.2 in Protocol S1); (ii) the unfinished high-throughput genomic sequence database (Figure 2.3 in Protocol S1), in the genome of sea urchin Strongylocentrotus purpuratus [27] and (iii) the Whole-Genome-Shotgun Sequence database (Figure 2.4 in Protocol S1) in the genome of the hemichordate Saccoglossus kowalevskii (unpublished).
The functional significance of this element is not clear, yet given that the region can be identified in at least ten – highly unrelated and primarily aquatic – species, the presence of a transposable element of a highly mobile nature (or its evolutionary relic) is indicated (Figure 2). In M. galloprovincialis, it has also been shown that mobile elements reside within introns of the Hsp70 genes [10], however there is no detectable sequence similarity between those elements and the Mghsp90 Gamera-like sequence presented here.
Another feature of the Mghsp90 upstream region is a genus-specific sequence, approximately 100-bp long, located 787 positions before the first exon of Mghsp90 genes (Figure 1). This region is much more phylogenetically restricted than the Gamera element, found only in the genus Mytilus, namely the M. galloprovincialis mytilin B precursor gene [28]–[29] – (accession number: AF177540.1, positions 777–815 antisense strand, non-coding region), a lysozyme gene (AF334662.1, positions 1016–1050 sense strand, second intron) [30] and a cDNA (AM878017.1) both from M. edulis, and a cDNA sequence from M. californianus (GE753693.1) (Figure 3). This genus-specific sequence does not contain any transcription factor binding sites (Figure 1), thus its functional significance is not known at present.
It is worth noting that similarly to the mytilin B gene, another antimicrobial peptide gene, the M. galloprovincialis defensin 2 (MGD2) gene, contains a 160-bp long element with similarities to the M. edulis lysozyme gene (fourth intron), two glycosidase gene introns (endo-1,4-beta-D-glucanase – AJ308548.1, 2nd intron; endo-1,4-mannanase – AJ271365.2, 5th intron), the 3′-UTR of the M. galloprovincialis Hsp70-1 gene, all being similar to an ISSR sequence (AJ938114), indicating the presence of a transposable element [31]. The above mentioned genes all have catabolic roles and might indeed be connected to defense mechanisms, broadly associated with stress. Further study is required in order to understand the role of these genus-specific sequences in the molecular physiology of the above mentioned loci.
A putative binding site for p53 is located between two HSEs in the 5′ regulatory region of the Mghsp90 genes [11] (Figure 1), being identical to the consensus binding site of human p53 to retinoblastoma susceptibility gene [32]. This binding site is evidently absent from other species, including C. elegans and D. melanogaster, but present in the human Hsp90 beta gene [33] (Figure 1). The p53 proteins from two Mytilus species exhibit very high similarity to their human homologs, and especially in the DNA binding domain, the transcriptional activation domain (TAD) and the nuclear localization signal. In addition, residues mutated in various human cancers are also conserved in the Mytilus p53 proteins [34]. It should be noted that p53 is phylogenetically restricted to animals while the molluscan versions (Decapodiformes, Bivalvia and Haliotis sp.) exhibit a very high similarity to the vertebrate sequences (not shown). The prediction of the p53 binding site in Mytilus is based on the known association of p53 with the upstream region of the human Hsp90 beta gene [33], the conservation of the Mytilus p53 genes [34] and the observation that an identical site is present in human Hsp90 (an Mghsp90 homolog) [11].
In order to further establish the validity of the predicted p53 binding site in a phylogenetic context, we have searched the non-redundant nucleotide database with the Mghsp90 genes as queries (see Methods). We subsequently identified 215 homologous target regions, with the closest sequence-similar entries carefully selected to exclude cDNA clones or partial coding sequences, across a wide taxonomic spectrum (Figure 1 in Protocol S1). These sequences were scanned for putative p53 binding sites (732 matches in total, see Methods), conditioned on the p53 phylogenetic distribution mentioned above; in other words, sites found in organisms known to encode for p53 were considered as positive cases (727 in total), while the remainder were treated as negative cases (5 in total). Despite well-understood limitations, e.g. the under-representation of certain species in terms of comparable Hsp90 sequence data and the over-representation of others in terms of redundant sequences, it is evident that p53-containing species exhibit a high number of predicted p53 binding sites (primarily chordates), while other organisms (such as fungi or plants), present a sporadic pattern of false positive hits, as expected. The exception in this otherwise consistent picture is the molluscs (Bivalvia and Haliotis sp.), having a small number of predicted p53 binding sites (Figure 1 in Protocol S1). The shortage of sequence information for molluscs, coupled with a possibly non-canonical sequence motif, leaves the question open for the unambiguous detection and experimental confirmation of the elusive molluscan p53 binding site.
The presence of a putative p53 binding site in the promoter region of the Mytilus Hsp90 genes raises questions about the possible implication of Hsp90 proteins in molluscan leukemia. Very recent studies on the association of p53 with heat shock response [35], the differential expression of p53 in mussel haemic neoplasia [5], and the impact of pollutants on p53 expression [36] underline the potential involvement of p53 in both heat shock response and neoplasia and its irregular similarity to vertebrate homologs [37], as well as its potential use as a marker for environmental monitoring [34]. In other species, namely soft-shell clams, certain results also indicate that environmentally induced alterations in p53 might contribute to leukemia [38]–[39].
Indeed, expression studies have established that Hsp genes and a p53-like gene are abundant in M. galloprovincialis [40], especially in pollutant exposed mussels [41], now searchable through the Mytibase resource [42]. Moreover, there is evidence from proteomics studies that Hsp proteins are expressed in stress conditions and can potentially be used as pollution biomarkers [43]–[44] or temperature biosensor [45].
In order to investigate co-expression patterns for Mghsp90 genes, we have extracted tissue-specific gene expression data available in Mytibase, encompassing 3840 cDNA sequences [42]. Following normalization (see Methods), we detected 547 genes (14% of total, in the ‘original’ network) that are differentially expressed across all four tissue types under investigation (namely gills, gonads, foot and digestive gland – Figure 4).
A two-way clustering across genes and tissues confirms that the four tissue types can be accurately detected (Figure 4A). This step also suggests that the 547 differentially expressed genes can be clustered into four distinct classes corresponding to the four tissues, with relatively low overlap (Figure 4A). A Principal Component Analysis of the original network further confirms the inter-replicate reproducibility and tissue specificity, indicating the high quality and consistency of the initial gene expression data (Figure 4B).
To infer gene associations via co-expression profiles, PCCs (see Methods) were computed for all possible pair-wise gene permutations of the original network. High PCC values correspond to a large similarity in expression profiles across four tissue types. Only those gene pairs with PCC>0.90 were further considered. This step yielded a global co-expression network, defined as the ‘inferred’ network, containing 3692 nodes and 57697 edges (Figure 5). The inferred network represents 96% of all cDNA clones in the original network. Such high coverage may be explained by the limited number of experimental replicates provided in the dataset.
To ensure that only significant associations are considered, MCL clustering (see Methods) was performed to produce a ‘clustered’ network with 1719 nodes and 43286 associations (Table 1). The clustered network represents a subset of the inferred network enriched with the most highly connected genes with the strongest co-expression values (Figure 5). Interestingly, of the 547 differentially expressed genes obtained initially, 271 (∼50%) are present in the clustered network, signifying a sufficient coverage of differential expression. This enriched network thus maintains 75% (43286/57697) of network edges, from which more reliable associations can then be extracted.
To delineate the involvement of the wider Hsp class of genes in normal M. galloprovincialis tissue, 8 cDNA sequences corresponding to 4 distinct Mytilus Hsp genes, labeled as Hspa5 (Grp78 homolog), Hsp70, Hspa90 (Mghsp90), and Ankrd45 (similar to heat shock 70 KD protein C precursor) were identified in the clustered network (Figure 5). The “Ankrd45”-like sequence (e.g. XP_290882.1) warrants description: its N-terminal part contains ankyrin repeats most similar to the ankyrin repeat domain of the human p53 binding protein [46], while its C-terminal part is similar to Grp78, a homolog of Hsp70 (Figure 3 in Protocol S1). Structural evidence indicates that the ankyrin repeats of p53 binding proteins (53BP2) bind to the L2 loop of p53 [47], implicating a configuration of ankyrin repeats such as the one found in Ankrd45, in a potentially mediated p53-Hsp70 domain interaction.
In fact, since the initial discovery that the Hsp70 promoter is regulated by p53 [48], there is mounting evidence that these two proteins are involved in various processes, including oral dysplasia [49], endometrial carcinomas [50], gastric cancers [51], ischemia [52] and wound healing [53]. These interactions have been reviewed elsewhere [54]–[55]. Similarly, it has been demonstrated that p53 requires the activity of Hsp90s [56] and the structural [57] and biochemical [58] basis of this interaction has been deciphered. In fact, it appears that p53, Hsp70 and Hsp90 are involved in a complex interplay during carcinogenesis [59].
To examine Hsp-related associations in greater detail, the nearest-neighbor members of 8 Hsp cDNA clones were selected, defined as the Hsp network (Figure 5). This network contained 174 genes and 2226 associations, accounting for 4.5% of genes in the original network (Tables 1 and 2 in Protocol S1 – node labels refer to MyArray1.0 identifiers, see Methods). The Hsp network contains clones with similarity to perlucin (a biomineralization-associated protein) [60] and the M. edulis polyphenolic adhesive protein [61], among others (Table 1 in Protocol S1); it is curious that in this set, there is also a clone highly similar to the M. edulis gene for endo-1,4-mannanase, discussed above.
Remarkably, 30/547 (5.5%) of differentially expressed genes are found to be co-expressed with the Ankrd45 clone. This suggests that members of the Hsp class are involved in complex transcription patterns across multiple tissue types rather than a single one. Indeed, the closest co-expression neighbors of Mghsp90 are two cDNAs for calreticulin – a calcium-binding chaperone (AJ624756/AJ625361) known to be associated with Hsp proteins [62] (Figure 6). Given the high-quality, yet limited data, the gene expression analysis outlined here strongly indicates that the known Hsp-associated genes in Mytilus are involved in intricate ways with each other, are possibly controlled by a small number of TFs over a number of tissues and conditions. It is thus possible that a mechanism for heat response might involve a ‘gene structure’ strategy, with few genes involved in a multitude of gene expression pathways.
In this study, we have dissected computationally the upstream region of the Mghsp90 genes to investigate its structure and function. The structural complexity of this region strongly suggests that the transcription of Hsp90 stress response is tightly regulated via a dense region of heat shock elements and other regions of varying phylogenetic dispersion (Figure 1). Compared to other model organisms, such as C. elegans and D. melanogaster, this regulation appears to be achieved through a ‘gene structure’ strategy, i.e. a complex gene structure. In addition, expression analysis of the heat shock response indicates that a handful of key molecules belonging to the heat-shock class, exhibit a differential tissue-specific expression profile, possibly in gills and the digestive gland, while at the same time maintaining a multitude of associations through a complex co-expression network (Figure 5). Our results are consistent with current knowledge about chaperone function both within molecular [63]–[64] and ecological contexts [65]–[66], and demonstrate the efficacy of both comparative genomics and systems biology for the elucidation of complex relationships between genotype, environment and phenotype.
The nature of sessile animals, with the Mytilus genus as a model organism, can shed light into their metabolic capabilities [67], stress responses [68] and resilience of evolutionary extinction [69]. The stress response for sessile animals is of particular interest, especially in cases where different ecological niches can be compared for close relatives, e.g. different growth potential in varying hydrostatic pressure or temperature [70]. Heat shock proteins in particular are used as indicators of thermal stress [68]; for instance, in the case of marine snails (Tegula genus), the time course and magnitude of the heat shock response was measured in a field study by monitoring the synthesis of heat shock proteins [71]. In another field study on the Oregon coast, M. californianus and its predator Pisaster ochraceus were examined for production of the Hsp70 heat shock proteins; it was found that while mussels (a sessile species) have an increased production of Hsp70, its starfish predators (a mobile species) do not, potentially exhibiting decreased heat shock adaptation compared to their prey [72]. Sessile marine invertebrates have been studied in the context of rising sea temperatures, including M. edulis [73] and Rhopaloeides odorabile, a common Great Barrier Reef sponge [74].
In the future, the thermal ecology of stress response can potentially inform policy decisions for environmental management in the context of climate change [75] – including the analysis of biogeographical range shifts [76], particularly important for sessile animals, the understanding of complex prey-predator interactions e.g. the above mentioned pair of P. ochraceus and M. californianus [77], and instigate a more integrated approach that will eventually include both weather records and niche-level measurements [78]. Currently, more established approaches for the use of Mytilus relate to its use as a biosensor system for the environmental monitoring of coastal water pollution [79], heavy metals or organic pollutants [80] – including manufactured substances such as fiberglass [81]. In conclusion, this work forms a basis upon which the stress response in Mytilus will be better understood at the molecular level.
The M. galloprovincialis Hsp90 sequence was analyzed as previously [11]. The Hsp90 upstream regions from three other representative animal species, namely Caenorhabditis elegans, Drosophila melanogaster and Homo sapiens, were analyzed in a similar fashion. Previously published data concerning Hsp90 genes from these species were also taken into consideration for annotation purposes [82]–[88]. Sequence database searches were performed by BLAST (for nucleotide sequences, blastn, version 2.2.22) [89]. Sequence alignments were computed using ClustalW [90] and visualized by JalView [91].
Regulatory elements, in the 5′ non-coding regions of the Mghsp90 genes were identified with Alibaba2 [92], P-MATCH [93] and the Transcription Element Search System (TESS) [94]. An extensive comparative analysis for p53 binding sites was performed using the Matrix Search analysis tool of the TRED database [95], scanning query sequences against the p53-specific sequence matrix (cut-off score 2) from the JASPAR collection [96].
Data were obtained from the Gene Expression Omnibus (GEO) database, representing one spotted cDNA array (Accession number: GSE2176) for gene expression in normal mussels (M. galloprovincialis), using the MyArray 1.0 platform targeting 1712 clones with a total of 3840 cDNA sequences [42]. This dataset encompasses the total RNA isolated from gills (n = 2), gonads (n = 2), foot (n = 2), and digestive gland (n = 2) [42]. Data normalization was performed by taking the binary logarithm (log2) of normalized intensities (defined as test signal/reference signal). Normalized data (‘original’ network) was subsequently subjected to statistical validation.
Gene associations were identified by computing the Pearson Correlation Coefficient (PCC) for all gene pairs in the raw network. Gene pairs with positive correlations indicated by a PCC>0.90 were considered to be co-expressed. Co-expression patterns were represented as networks where each node corresponds to a unique gene and each edge represents a co-expression association. The final network (‘inferred’ network) was clustered using the Markov Clustering Algorithm (MCL) in order to both filter noisy associations and identify biologically meaningful clusters (‘clustered’ network), as previously described [97]. The inflation parameter for MCL was set to 3.0. Only clusters with >10 genes were further analyzed for biologically meaningful associations.
Differential expression analysis was performed by applying Analysis of Variance (ANOVA) to all genes across four distinct tissue types. Only those genes with the overall p-value bellow 0.05 were considered as differentially expressed. Two-way unsupervised hierarchical clustering of differentially expressed gene signals was performed using Euclidean distance as a similarity measure. Principal component analysis (PCA) was also performed to confirm the validity of the analysis for the four tissue-specific datasets. All statistical analyses were performed with MATLAB (The MathWorks, Natick, MA – www.mathworks.com).
|
Subsets and Splits
No community queries yet
The top public SQL queries from the community will appear here once available.